Load Frequency Control of Multi Area

International Review of
Electrical Engineering
(IREE)
Contents
Phase Drift Phenomenon in Dual Active Bridge Converter – Analysis and Compensation 1
by Vesa Väisänen, Jani Hiltunen, Raimo Juntunen
Replacing Copper with New Carbon Nanomaterials in Electrical Machine Windings 12
by J. Pyrhönen, J. Montonen, P. Lindh, J. J. Vauterin, Marcin J. Otto
Computation of Electromagnetic Losses in Double-Rotor Vernier PM Motors 22
with Three Topologies Using TS-FEM
by Huijuan Liu, Yue Hao, Shuangxia Niu, Weinong Fu
Copyright © 2015 Praise Worthy Prize S.r.l. - All rights reserved
Optimum Design of a Surface Mounted Fractional Slots Permanent Magnet Motor 28

by A. Mansouri, M. Hejra, H. Trabelsi
Real Time Implementation of Sliding Mode Control for Induction Motor Drives Using dSPACE 36
by A. Essalmi, H. Mahmoudi, A. Bennassar, A. Abbou, Y. Zahraoui, M. Akherraz
Performance Improvement of Subspace-Based Direction-Finding Algorithms 42
Using Higher-Order Statistics
by Supawat Supakwong
An Up-to-Date Technologies Review and Evaluation of Wave Energy Converters 52
by Hosna Titah-Benbouzid, Mohamed Benbouzid
Planning of Low-Voltage Distribution Networks Using Hierarchical Methods 62
by Amir Softic, Marinko Stojkov, Admir Jahic
Single Phase Fault Location in Distribution Networks Based on Charge Transients 71
by Šeila Gruhonjić Ferhatbegović
Model for Aggregation of Plug-in Electric Vehicles Power Demand 79
for Distribution Systems Planning
by Mario A. Ríos, Camilo A. Rincón
The Effect of Distributed Generation Type and Location Constraints on the Solution 88
of the Allocation Algorithm
by R. Prenc, N. Bogunović, A. Cuculić
Future-Proof MV Distribution Network Short-Circuit Protection Scheme 98
by H. Laaksonen
Voltage Instability Analysis on PV and QV Curves for Radial-Type 109
and Mesh-Type Electrical Power Networks
by Marizan Sulaiman, Ahmad Fateh Mohamad Nor, Noor Ropidah Bujal
P&O and Incremental Conductance MPPT Implementation 116
by Sadik Farhat, Rachid Alaoui, Abdelilah Kahaji, Lahoussine Bouhouch, Ahmed Ihlal
Optimization FOU of Interval Type-2 Fuzzy Inference System Using Big Bang – Big Crunch 123
Algorithm for Short Term Load Forecasting on National Holiday Case Study: South and Central
Kalimantan-Indonesia
by Akhmad Ramadhani, Agus Dharma, Imam Robandi
(continued)
Photovoltaic System Equipped with a Reliable and Efficient Regulator/MPPT 131
and Energy Flow Controller
by K. Hirech, M. Melhaoui, E. Baghaz, R. Malek, K. Kassmi
Modeling and Control of a Hybrid Microgrid by Multi-Agent System 145
by Lahoussine Elmahni, Lahoussine Bouhouch, Rachid Alaoui, Ali Moudden
Load Frequency Control of Multi Area SSSC and CES Based System Under Deregulation 154
Using Particle Swarm Optimization
by P. Marimuthu, B. Basavaraja, Subhransu Sekhar Dash
Power Line Inspections and Dedicated Flight Envelopes 163
by Sandra C. R. Antunes, K. Bousson, João G. Mota
International Review of Electrical Engineering (I.R.E.E.), Vol. 10, N. 1
ISSN 1827- 6660 January – February 2015
Phase Drift Phenomenon in Dual Active Bridge Converter –

Analysis and Compensation
Vesa Väisänen, Jani Hiltunen, Raimo Juntunen
Abstract – In this paper the mechanisms causing phase drift in the dual active bridge topology
are presented and two compensation methods are proposed to overcome their effects. The phase
drift causes discrepancy between the predicted phase shift and the actual phase shift measured
from the transformer primary and secondary voltages. The error in the phase shift causes the input
power to be larger or smaller than predicted and it can also interfere with the operation of several
modulation methods. By using the phase drift compensation methods presented in this paper the
converter input power can be predicted accurately by using the classical power equation. The
effects of dead-time on the phase drift are also discussed and a method to calculate a suitable
dead-time with respect to the converter operating point is proposed. The value of the phase drift
compensation is presented along with a variable frequency modulation method, which enables
zero voltage switching at a wide load range with a minimal circulating current. Copyright © 2015
Praise Worthy Prize S.r.l. - All rights reserved.
Keywords: DC-DC Power Converters, Dual Active Bridge, Error Correction, Modeling,
Modulation, Power MOSFET, Switched-Mode Power Supply
Nomenclature t Time
tdead Dead-time between the switching leg
Co(tr) Time related output capacitance transistors
Coss Transistor output capacitance tdelay Voltage rise time after a switching instant
DAB Dual Active Bridge tdrift Time difference between the primary and
DSP Digital Signal Processor secondary switching delays
FB Full-bridge V1 Voltage across the leakage inductor,
fsw Switching frequency primary side
HB Half-bridge V2 Output voltage referred to the transformer
hpri Switching configuration parameter, primary
primary side VDS Drain-source voltage
hsec Switching configuration parameter, VFM Variable Frequency Modulation
secondary side Vin Input voltage
ID Drain current Vout Output voltage
Iin Input current ZCS Zero Current Switching
Ipeak Peak current ZVS Zero Voltage Switching
Isw Current at transistor switching instant Phase shift as a percentage of the
Izvs Minimum current required for zero switching period
voltage switching ctrl Phase shift from the feedback controller
Llk Leakage inductance The difference between the actual and
drift
MOSFET Metal-Oxide-Semiconductor Field-Effect desired phase shift as a percentage of the
Transistor switching period
n Transformer turns ratio Phase shift in radians
rad
Npri Transformer primary turns Sum of the feedback controller and
sum
Nsec Transformer secondary turns feedforward phase shifts
P Power
Pctrl Power given by a PI-controller controlling
the transferred power I. Introduction
Pref Power reference
Qtot Total charge in the switching leg transistor Dual Active Bridge converter (DAB) is a galvanically
output capacitances isolated DC/DC converter topology, which has gained a
SiC Silicon Carbide lot of research interest during the last decades due to its
1
low passive component count, soft-switching capabilities Vin2 Vout   

and bidirectional characteristics [1]-[4]. The topology has P rad 1  rad  (1)
2 f sw Llk nVin   
also been used in fuel cell power conversion, which is
also the motivating application in this paper [5], [6].
It was mentioned in [1], [7] and [8] that the ideal The parameter fsw is the switching frequency. The
power equation may introduce significant calculation phase shift rad is expressed in radians. Eq. (1) can be
errors in practice due to a phenomenon called phase drift. derived into a more general form (2) to simplify its usage
Because of the phase drift the actual phase shift with different primary and secondary configurations:
between the transformer primary and secondary voltages
is different than what is requested by the converter V1V2
P  1  2   (2)
control. This can cause power flow also at zero phase Llk f sw
shift where, according to (2), the transferred power
should be zero. It was also mentioned in and that due to The phase shift is now expressed as a percentage of
phase drift there is a certain phase shift range where the switching period (-1 ≤ ≤ 1) and the voltages V1 and
power cannot be transferred and this was mentioned to be V2 can be expressed as follows:
related to the dead-time between the phase leg
transistors. The phase drift phenomenon will also disturb
V1  h priVin
modulation methods that relay on calculations based on 
the assumption of an ideal converter. The ideal converter  Vout (3)
can produce exactly the desired phase shift and therefore V2  hsec
 n
also the desired waveforms providing that the other terms
of the power equation are accurate [9], [10]. where hpri and hsec are 0.5 for a half bridge (HB) and 1 for
In this paper it is shown that the phase drift does not a full bridge (FB) configuration. The transformer turns
result solely from the dead-time as stated in [8], but ratio n is Nsec/Npri, where Npri and Nsec are the number of
rather from the switching delays and the charge/discharge primary and secondary turns, respectively. By using
times of the nonlinear transistor capacitances. these notations the power equation can be used with all
The effect of dead-time is relevant only if the dead- kinds of primary and secondary configurations, not only
time is too short or too long with respect to the converter with FB-FB configuration.
operating conditions and these limits will be presented in In Figs. 2 the current Izvs denotes the minimum current
this paper. The phase drift phenomenon is emphasized that is required to achieve zero-voltage switching in
when using HB-FB or FB-HB configurations, since the primary and secondary bridges. If there is no error in the
voltages and currents in the primary and secondary phase shift and if the switches are ideal, we can
bridge transistors can be different and thus also the accurately calculate the lengths of the time intervals t1-t3
voltage dependent nonlinear capacitances and the and the magnitudes of currents Izvs and Ipeak based on the
switching delays can differ considerably. simplified DAB model. In the simplified model the
Two different methods to compensate the phase drift converter is modeled as two voltage sources connected
are introduced: the first method is based on the with an inductor.
calculation of transistor output capacitance However, since the transistors can have large
charge/discharge times and the second one is based on a nonlinear output capacitances which need to be charged
feedback controller. It is shown that by compensating the and discharged at each switching instant, the transistor
phase drift with these methods the power equation can be voltages and therefore the transformer primary and
used to predict the power flow even with very small secondary voltages cannot change instantly.
phase shift values. As a result, the actual switching instant differs from
the intended switching instant and the desired current
II. Causes for Phase Drift (such as Izvs or zero crossing) at turn-off may not be
achieved.
in DAB Converter
In this section the reasons for phase drift are analyzed Iin
in both zero-voltage switched and hard switched modes. PA+ SA+ C1
PB+
Llk 1:n
II.1. Phase Drift When Operating in Zero-Voltage V2
Vin V1 Ilk Vout
Switching Mode
Fig. 1 shows a circuit diagram for a DAB converter PA- PB- SA- C2
and Figs. 2 illustrate the idealized voltage and current
waveforms for this converter. The power flow of the
converter is controlled by adjusting the phase shift
Fig. 1. A circuit diagram of a Dual Active Bridge converter using a FB-
between the primary and secondary bridges according to FB configuration. The primary transistors are marked with prefix P and
the following equation [11]: the secondary transistors with prefix S
Copyright © 2015 Praise Worthy Prize S.r.l. - All rights reserved International Review of Electrical Engineering, Vol. 10, N. 1
2
Vin = 100 V, Vout = 250 V

2
Delay
Delay
1
Izvs = 2 A
0 Ipeak = 11 A
Vpri Vpri
Vsec V
-1 sec
Ilk I
lk
-2
0.5 1 1.5 2 2.5 3 3.5
Time [µs]
Vin = 130 V, Vout = 250 V

2
Delay Ipeak = 7 A Delay
-1
Izvs = 6.5 A
-2
0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Time [µs]
Vin = 130 V, Vout = 210 V

2
Delay Delay
0
Izvs = 2 A
-1
Figs. 2. Idealized waveforms for the DAB converter in two cases:
a) V1<V2 and b) V1>V2. The power flow direction is from V1 to V2. Ipeak = 11 A
-2
0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8
If the delays are exactly the same on the primary and Time [µs]
secondary side, they will effectively cancel each other
Fig. 3. Measured phase drifts for three cases: V1<V2 (top), V1 ≈ V2
and there is no phase drift. However, Fig. 3 illustrates (middle) and V1>V2 (bottom). The input current Iin = 6 A in all cases.
that this may not be the case in practice. The references for primary and secondary voltages are marked with
In Fig. 3 (top) the primary switching leg transistor dashed lines. The converter is using a FB-HB configuration and the
output capacitances are charged and discharged by converter is in zero-voltage switching in all cases. When the delays are
known, the phase drift can be calculated from (7)
current Izvs and the secondary transistor capacitances by
current Ipeak. Since Ipeak is considerably larger than Izvs,
the transistor drain-source voltages can change much The difference between the primary and secondary
faster on the secondary side than on the primary side. delays is not as large as in the case where V1<V2, since
This will cause the effective phase shift to be shorter the primary side Coss charges are larger than the
than intended. In Fig. 3 (middle) the converter is secondary side charges with the voltages used in the
operated near a 1:1 point, where V1 ≈ V2. In this case the measurement.
leakage inductance current is square shaped and both the According to our measurements the dead-time does
primary and secondary sides are switched with same not have an appreciable impact to the phase drift
current. Therefore, the overall phase drift is ideally zero, providing that the dead-time is not too short or too long.
since the switching delays between the primary and Fig. 4 illustrates the effects of a dead-time selection in a
secondary bridges are identical. The identical switching case where Vin = 100 V, Vout = 250 V and Iin = 4 A.
currents do not ensure identical delays in cases where If the dead-time is too short and the transistor is
there are differences between the charges in the primary switched on during the drain-source voltage resonance
and secondary transistor output capacitances. period, the output capacitances are charged and
When the input voltage is higher than the reflected discharged abruptly and this is seen as an overshoot in
output voltage, the primary side transistors are switching the drain current. The effect on the phase drift is that the
at Ipeak and the secondary side transistors at Izvs as transformer voltage transitions become faster and thus
illustrated in Figs. 2 and Fig. 3. The primary side is the delays become smaller. The drawback of a too short
switching faster than the secondary side and the actual dead-time is degraded efficiency, since the transistors are
phase shift is now longer than intended. partially hard switched.
3

 C gs
C gd
Rg


Fig. 5. Zoom of the drain-source voltage (red dashed line) and drain
current (blue solid line) behavior of transistor PA- with 200 ns dead-
time. Since the drain voltage is not changing during t3-t4, there is no
distinct Miller plateau in the gate-source voltage
At time instant t1 transistor PA+ is turned off. The

energy stored in the leakage inductance begins to charge
the output capacitance of transistor PA+ and discharge the
output capacitance of PA. The time t1-t3 that is required
for the discharging can be approximated based on the
total charge in the switching leg transistor output
capacitances (the determination of this charge will be
discussed later) and the leakage inductance current at
instant t1 as:
Qtot
Fig. 4. Measured effects of dead-time on the voltage and current t1  t3  (4)
behavior of transistor PA-. The instant where the gate-source voltage I lk  t  t1 
has risen to the conduction threshold level is marked with Vgs(th)
The current Ilk is not constant during t1-t3 or during the
The optimal instant to turn on the transistor is when delay periods in Fig. 3, but is changing nonlinearly.
the transistor output capacitance is discharged and the However, in most cases it is sufficient to calculate (4)
body diode (or an external freewheeling diode) starts with Ilk(t = t1) being Izvs or Ipeak depending on the
conducting. Since the MOSFET channel is typically a operating mode. In cases where the stored energy in the
less lossy route for the current than the body diode, the leakage inductance is very close to the minimum energy
conduction losses can be minimized by allowing the required for zero-voltage switching, a better
current to commutate to the channel as soon as possible. approximation can be obtained by using the average
If the dead-time is so long that the transistor drain leakage inductance current value during time period t3 in
current changes direction from negative to positive Figs. 2. The length of time period t3 in Figs. 2
before the transistor channel is opened (Fig. 4, bottom) corresponds to the time interval t3-t4 in Fig. 5 and the
the transistor output capacitance begins to charge and the upper limit for its length can be approximated from:
drain-source voltage increases.
If the dead-time is long enough, the transistor voltage I sw Llk
will reach its off state value. This voltage overshoot can t3  t4  (5)
be seen as a voltage sag in the complementary transistor V1  V2
drain-source voltage in the same switching leg. The large
oscillations in the drain-source voltage are likely to In order to ensure zero-voltage switching, the
degrade the efficiency more than the partial hard- transistor PA- gate-source voltage should not be applied
switching during the short dead-time operation. In Fig. 4 before instant t2. After the gate-source voltage has been
the efficiency was 93.4% in the 0 ns case, 93.7% in the applied, the gate-source voltage begins to rise with a time
200 ns case and 92% in the 700 ns case. constant τ, which depends on the gate resistor and on the
The dead-time effects such as voltage sag and gate-drain and gate-source capacitances. If the gate-
overshoot were briefly presented also in [12], but no source voltage rises to the threshold voltage between
guidelines have been given on how to select a proper time instants t3-t4, the transistor is able to conduct the
dead-time with respect to the operating conditions. positive drain current after t4 and there will be no voltage
4
overshoots and voltage sags. The dead-time which causes If operating in case where V1<V2 the primary side is
the smallest interference to the current commutation can hard switched while the secondary side switches at peak
now be expressed as: current. In case where V1>V2 the primary side switches at
peak current and the secondary side is hard switched. In
Qtot VDS  I sw Llk both cases the delay differences between the primary and
 tdead  (6) secondary voltages are much smaller than in the soft-
I lk  t  t1  V1  V2
switching mode (Fig. 3) and thus the phase drift is also
smaller.
where Isw is the current at transistor switching instant
When operating in hard-switching the phase drift may
(either Izvs or Ipeak).The inequality (6) applies only to ZVS
not be an issue at all providing that we are not interested
operation. In practice the leakage inductance current
in the leakage inductance current waveform, but rather
behaves nonlinearly during the switching process and
on the power to be transferred. If the converter is current
cannot change with the rate described by (5) until the
(or power) controlled, the controller will adjust the phase
resonant process during t1-t3 has completed and the
shift so that the current (power) reference is met
voltage across the leakage inductance has reached its
regardless of the phase drift.
maximum value. Therefore, a better approximation for
the maximum dead-time length can be calculated as a
sum of (4) and (5). If the converter operates under hard II.3. Phase Drift Differences Between Different
switching it is enough to ensure that the dead-time is Input-Output Configurations
adequately long to avoid switching leg cross-conduction
and short enough to avoid zero-sequences in the As was mentioned in the introduction, the phase drift
transformer current. phenomenon is emphasized when the primary and
secondary sides have a different switching configuration.
Fig. 7 presents a simulated comparison with a FB-HB
II.2. Phase Drift When Operating in and FB-FB configuration, when using PSpice and
Hard Switched Mode STWA88N65M5 MOSFETs modelled with nonlinear
capacitances.
Fig. 6 provides support for the statement that the
phase drift is smaller in hard switched mode than in zero-
PA+
voltage switching mode.
10
Leakge inductance current [A]
7.5 FB-HB
FB-HB IZVS too FB-FB
5 low
2.5
0
-2.5
-5
-7.5
-10
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Time [µs]
8
6 FB-HB hard
Drain current [A]
switching
4
0
FB-FB soft switching
-2
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Time [µs]
150
Drain-source voltage [V]
100
50
-50
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Time [µs]
Fig. 6. Measured phase drifts for two cases: V1<V2 (top) and V1>V2
(bottom). The input current Iin = 4 A in all cases. The references for Fig. 7. An example of the phase drift difference between a FB-FB and a
primary and secondary voltages are marked with dashed lines. The FB-HB converter. The turn-off of transistor PA+ initiates zero-voltage
converter is using a FB-HB configuration and the converter is in hard- turn-on process for the transistor PA-. The leakage inductance current
switching in both cases should be 2.5 A at PA+ turn-off
5
The operating parameters in the simulation are: Vin = function of the applied voltage. Fig. 8 illustrates the
100 V, Vout = 250 V, Iin = 4 A, Izvs = 2.5 A. The switching charges obtained for a STWA88N65M5 MOSFET when
frequency that should result in the desired Izvs is 96154 using a PSpice model and the Coss values from the
Hz and the corresponding phase shift is 0.1. The datasheet. For the sake of comparison the charges are
transformer turns ratio in the FB-HB is 1:1, while it is calculated also for a C2M0080120D SiC MOSFET. If a
1:2 in the FB-FB. phase drift calculation is performed for the SiC transistor
In the FB-HB configuration the current that is using the parameters of Fig. 3 (top), the total phase drift
charging and discharging the secondary side transistor is only 0.0017, which is 92% smaller than the phase drift
output capacitances is twice as large as in the FB-FB with the STWA88N65M5.
configuration. Because of this the FB-HB secondary Besides having lower absolute Coss charge values the
switches faster and there is a phase drift like in Fig. 3 SiC charge behavior as a function of drain-source voltage
(top). Due to the phase drift the current Izvs is smaller is much more linear compared to the Si transistor, which
than the specified 2.5 A at the turn-off instant of PA+ and, also decreases the amount of phase drift in various
as a consequence, transistor PA- is hard switched. In the operating points [16].
FB-FB configuration the primary and secondary It is not necessary to measure the charges by using
switching times are closer to each other and the phase only one switching leg. By measuring the delays from a
drift is small. Therefore, the desired turn-off current for complete converter setup, the additional parasitic
PA- is close to the specified value and soft-switching capacitances such the transformer winding capacitances
conditions are met. The phase drift affects the input can be included in the measurement.
current as well. In the FB-HB configuration the input The compensation algorithm can be implemented on a
RMS current was 80% of the reference current, while in DSP by creating a look-up table from the charge and
the FB-FB configuration it was 95%. voltage values, calculating the phase drift from (7) and
adding the phase drift to the phase shift given by the
modulation algorithm.
III. Phase Drift Compensation
In this section two different methods that can be used
to compensate the phase drift effects are introduced. The
phase drift compensation is important when using
modulation methods such as the ones described in [9],
[13] and [14] where the current values at certain time
intervals need to be known.
III.1. Charge Based Compensation

In Fig. 3 it was illustrated that the phase drift depends
on the differences between the primary and secondary
switching leg voltage transition times. The phase drift
can be calculated from:
Q pri VDS  Qsec VDS 

tdrift  tdelay ,pri  tdelay ,sec  
I sw,pri I sw,sec (7)
drift  tdrift f sw
The phase drift can be compensated by adding drift to

the phase shift given by the modulation algorithm such as
(2) or (8). As an example in Fig. 3 (top) the primary leg
is switching at 2 A and the switching leg charge is 834
nC. The primary delay is thus 417 ns.
The secondary leg switches at 11 A and the charge is
787 nC, which results in a delay of 72 ns. The total phase
drift in this case is (417-72) ns•62.4 kHz = 0.0215. If the
transition times between the primary and secondary are
identical, there is no phase drift. The total capacitive
Fig. 8. STWA88N65M5 capacitances from the datasheet and the
charge in a switching leg can be obtained for example via corresponding Cosscharges obtained from a simulation model built with
double pulse test [15]. nonlinear capacitances. The datasheet charge value is calculated from
The charges can also be obtained without assembling (2·Co(tr)·VDS), where Co(tr) is the given equivalent time related
actual hardware by using a circuit simulator where the capacitance. The total Coss charge for C2M0080120D SiC transistor is
provided for comparison
output capacitance can be modeled to be a nonlinear
6
The charge values can also be used when calculating a those that would be required to achieve ZVS under the
suitable dead-time from (6). A suitable average charge operating conditions. The converter error dynamics are
value could also be used instead of a look-up table similar to the converter overall dynamics and therefore
depending on the output capacitance nonlinearity and the the stability criteria of the compensation controller can be
maximum allowable error in the switching delay derived similarly as would be done for the converter
estimation. power controller.
III.2. Controller Based Compensation

The charge based compensation method has some
limitations which undermine its practical value. The
number of elements in the look-up table must be quite
high especially around the voltage regions where the
charge changes steeply in order to avoid sudden changes
in the calculated phase drift compensation.
The fluctuation of the compensation value may lead to
large oscillation in the input power depending on the
operating conditions. Also, since there is a possibility of
measurement errors and since there are differences
between the individual transistors and gate drivers, there
is most likely some error in the calculated compensation
value. In order to make the compensation independent on
the transistor capacitances, it can be implemented as a
feedback PI-controller, Fig. 9(c). The modulation
algorithm block calculates the required phase shift and
switching frequency based on the measured voltages and
the reference current. In this case the other parameters
fsw, Llk, n, hpri and hsec have fixed values. The calculated
phase shift based on the given power reference Pref is fed
to the summation block. The feedback controller
observes the actual converter input power and if there is
difference between the measured power and the power
reference the controller adjusts the phase shift
compensation value ctrl so that the phase shift sum results
in the desired input power.
The zero difference between the reference power and
the measured power means that if the other terms than Figs. 9. Various control approaches for the DAB converter. The
approach (c) includes the phase drift compensation, while approaches
in (2) are known, the phase shift between the transformer (a) and (b) are feedback control methods that are prone to phase drift
primary and secondary voltages has to be correct. effects
Although the phase shift error will be compensated, the
converter efficiency is still affecting the accuracy of the
power equation. This could be compensated by using IV. Experimental Results
efficiency estimates according to the transferred power. To provide experimental verification for the results
When using a basic phase shift modulation the phase presented in this paper, a 1 kW prototype was built (Fig.
shift could be controlled without the phase shift 10).
calculation block by feeding the modulator directly from
the power controller, Fig. 9(a). However, in modulation CONTROL
PCB
DSP-FPGA
CARD
methods where the current values at certain time intervals
need to be known, it is important that the desired phase
shift can be applied to the converter.
If all the parameters in (2) can be considered accurate,
the system can be treated as an ideal system, which MAIN PCB
OUTPUT
produces the desired voltage and current waveforms. In FILTER
any case, the inclusion of the phase shift calculation INPUT FILTER
SECONDARY
block along with the phase shift controller will speed up MOSFETS
PRIMARY
the system step response, Fig. 9(b). The problem with TRANSFORMER
MOSFETS
approach (b) is that if there is error in the system the
controller output Pctrl differs from the reference Pref and Fig. 10. Dual active bridge prototype having a full bridge primary
this causes the modulator output variables to differ from and a half bridge secondary (FB-HB).
7
The component values for the converter are given in shifts are larger than intended. In the 1:1 case the delays
Table I. Fig. 11 presents the measured phase shifts at are nearly identical on the primary and secondary sides
various input and reflected output voltage ratios, when no and therefore the measured phase shifts are close to the
compensation is used. reference. In order to evaluate the impact of phase drift to
the accuracy of the power Eq. (2), the actual phase shift
TABLE I and the input power are measured at various phase shift
COMPONENT VALUES FOR THE DAB PROTOTYPE
references and Vin/Vout values. The efficiency is also
Symbol Quantity Value
Sw Power MOSFETs 6 × STWA88N65M5
measured at each point in order to rule out its effect on
TF Transformer ETD54 N87 the power prediction.
n Turns ratio 1 The curve "calculated" corresponds to the power
n1 Primary winding 25 turns 1050 × AWG44 litz values calculated from (2) using the phase shift reference
n2 Secondary winding 25 turns 420 × AWG46 lits
L1k Leakage inductance 26.4 μH (internal)
and the curve "calculated with losses" is obtained from
Cin Input capacitor 28 × 1μF/450V ceramic cap. (2) using the measured phase shift and by adding the
Cdc1 Voltage doubler capacitor 14 × 1μF/450V ceramic cap. measured power losses to the result.
Cdc2 Voltage doubler capacitor 14 × 1μF/450V ceramic cap.
Cb1 DC-blocking capacitor 10 × 10μF/25V ceramic cap.
500 100
La1 Input choke 63μH, KoolMu 77930/125μ
450 90
Lb1 Input filter damping choke 31μH, KoolMu 77930/125μ
Rb1 Input filter damping resistor 2.2Ω 400 80
La2 Output choke 63μH, KoolMu 77930/125μ 350 70
Input power [W]
Efficiency [%]
Lb2 Output filter damping choke 31μH, KoolMu 77930/125μ 300 60
Rb2 Output filter damping resistor 2.2Ω 250 50
200 40
0.12 Measured
150 30
Reference Calculated
0.9:1 100 20
0.1 Calculated with losses
1:1 50 10
Efficiency
1.1:1
0 0
0.08 0 0.02 0.04 0.06 0.08 0.1 0.12
Phase shift
Effective phase shift
650 100
0.06
600 90
550
500 80
0.04
450 70
Input power [W]
Efficiency [%]
400 60
0.02 350
50
300
250 40
0
200 30
150
20
-0.02 100
0 0.02 0.04 0.06 0.08 0.1 0.12 10
50
Phase shift reference
0 0
0 0.02 0.04 0.06 0.08 0.1 0.12
Phase shift
Fig. 11. Measured phase shifts with respect to the reference phase shifts
650 100
at various Vin/nVout ratios. The output voltage Vout is 160 V and the input
600 90
voltages are 74.7 V (0.9:1), 83 V (1:1) and 91.4 V (1.1:1). 550
80
500
450 70
The effects described in conjunction with Fig. 3 can
Input power [W]
Efficiency [%]
400 60
also been seen in Fig. 11. With the voltage ratio of 0.9:1 350
50
the primary side delays are longer than the secondary 300
250 40
delays, which results in smaller phase shifts than 200 30
intended. The reason for the large phase shift error in the 150
20
range of 0.03-0.08 is the change in the converter 100
50 10
operating mode. 0 0
0 0.02 0.04 0.06 0.08 0.1 0.12
Below the phase shift of 0.03 the converter is Phase shift
operating under hard-switching and therefore the
transistor output capacitances are charged and discharged Fig. 12. Measured input powers at various phase shifts with different
abruptly. After the 0.03 mark the converter shifts into input voltage to reflected output voltage ratios. Please see Fig. 11 for
soft-switching, but since the leakage inductance current the voltage levels
is small the zero-voltage transition takes time and the
delays are long (we are allowing the resonance process to It can be seen that by removing the phase drift from
take place by using a long enough dead-time). At larger the calculation and by taking the power losses in to
phase shifts the leakage inductance current is larger and account the calculated power correlates well with the
therefore the phase drift calculated from (7) is smaller. measured power. There should not be large discrepancies
When the voltage ratio is 1.1:1 the secondary delays between the predicted power and the measured power
are larger than the primary delays and therefore the phase like there was in [7], if the converter is operating under
the conditions listed below:
8
• The primary and secondary transistors should be able Again, the phase drift compensation is needed to
to produce similar switching times in order to enable switching at zero current. Fig. 13 illustrates the
minimize the phase drift. The factors affecting the operation of the VFM algorithm with and without the
switching times are the transistor output capacitance phase drift compensation. The converter operating point
(the lower the better), the voltage difference between is Vin = 125 V, Vout = 300 V, Iin = 4 A and Izvs = 2.5 A.
primary and secondary sides (the output capacitance
is voltage dependent) and the switching currents.
• The efficiency should be as high as possible to
minimize the effect of power losses.
• The dead-time should be long enough to avoid hard-
switching and short enough to avoid the voltage
overshoot and voltage sag phenomena (Fig. 4).
• The converter has proper input and output filtering so
that we measure the active current instead of the
circulating reactive current [17], [18].
The value of the drift compensation is best
demonstrated by investigating the leakage inductance
current waveforms. If we know that the measured phase
drift is equal to the phase shift given by the power
equation, we also know the lengths of the time intervals
t1, t2 and t3 and the values of Izvs and Ipeak in Figs. 2. The
compensation was used along with a variable frequency
modulation method (VFM), which can be derived from
the idealized DAB model as follows [19]:

  1 
 
   I ref   sign I ref 

 4   2 I 2  2 I    2 
 ref ref  (8)

 h priV2
 f sw   1  2  
 I ref Llk
where:
1 V1  V2

   V1 (9) Fig. 13. Measured waveforms of VDS(PA-), inverted ID(PA-), Ilk and Vsec
V V1  V2 with the compensation controller disabled (top) and enabled (bottom).
 2 Without the phase drift compensation the value of Izvs is approximately
2 A instead of the desired 2.5 A, which causes the transistor PA- to go
into hard-switching. The transistor current was measured with a PEM
 V1 CWT Ultra mini with an output of 50 mV/A
 V1  V2
  V2 (10)
1 V1  V2 Although the difference in Izvs in Fig. 13 was only 0.5
 A, it was enough to make a difference between ZVS
(efficiency 93.44%) and partial hard-switching
and: (efficiency 93.09%). The importance of the drift
  sign  I ref  I zvs h pri (11) compensation increases together with the amount of
phase drift.
In (8) Iref is the reference current for the controller.
The VFM modulation method is explained in more detail
in [19].
V. Conclusion
The purpose of the modulation method is to produce In this paper it was shown that the phase drift
the desired switching current Izvs by varying the phenomenon results from the different switching delays
switching frequency and phase shift accordingly. In order between the primary and secondary bridges. When using
to ensure that the algorithm produces the desired Izvs and transistors with small output capacitances (such as wide
zero-voltage switching, the actual phase shift measured band-gap MOSFETs) and if the differences between the
from the transformer primary and secondary voltages primary and secondary switching currents are small, the
must match the phase shift given by the algorithm. phase drift can be negligible. The phase drift is of
The algorithm can also be used to obtain zero current importance if such modulation methods are used, where
switching by setting Izvs = 0. the leakage inductance current at a certain time instant
9
needs to be known. This paper presents two phase drift http://dx.doi.org/10.1109/TPEL.2013.2271511.

[9] F. Krismer and J. W. Kolar, "Closed form solution for minimum
compensation methods that can be used to correct the
conduction loss modulation of DAB converters," IEEE
discrepancies between the phase shift given by the power Transactions on Power Electronics, vol. 27, no. 1, pp. 174-188,
equation and the actual phase shift measured from the 2012.
transistor primary and secondary voltages. The method [10] F. Krismer and J. W. Kolar, "Efficiency-optimized high-current
dual active bridge converter for automotive applications," IEEE
based on output capacitance charge calculation can also Transactions on Industrial Electronics, vol. 59, no. 7, pp. 2745-
be used for selecting a suitable dead-time under zero 2760, 2012. http://dx.doi.org/10.1109/TPEL.2011.2157976.
voltage switched conditions. To compensate for the [11] M. N. Kheraluwala, R. W. Gascoigne, D. M. Divan and E.
possible errors in determining the switching leg charges, Baumann, "Performance characterization of a high-power dual
active bridge DC-to-DC converter," IEEE Transactions on
a feedback controller based compensation method was
Industry Applications, vol. 28, no. 6, pp. 1294-1301, 1992.
proposed. This method is more accurate in the phase drift http://dx.doi.org/10.1109/28.175280.
compensation, although the power losses and voltage [12] A. Kadavelugu, B. Seunghun, S. Dutta, S. Bhattacharya, M. Das,
drops still slightly reduce the accuracy of the VFM A. Agarwal and J. Scofield, "High-frequency design
considerations of dual active bridge 1200 V SiC MOSFET DC-
algorithm. The converter voltage drops can be taken into
DC converter," in Applied Power Electronics Conference and
account by calculating them based on the known circuit Exposition (APEC), 2011.
resistances and currents or by using measurements. http://dx.doi.org/10.1109/APEC.2011.5744614.
[13] J. Everts, F. Krismer, J. Van den Keybus, J. Driesen and J. W.
Kolar, "Optimal ZVS modulation of single-phase single-stage
Acknowledgements bidirectional DAB AC-DC converters," IEEE Transactions on
Power Electronics, vol. 29, no. 8, pp. 3954-3970, 2014.
The research leading to these results has received http://dx.doi.org/10.1109/TPEL.2013.2292026.
[14] L. Xiaodong and Y.-F. Li, "An optimized phase-shift modulation
funding from the European Union’s Seventh Framework for fast transient response in a dual-active-bridge converter,"
Programme (FP7/2007-2013) for the Fuel Cells and IEEE Transactions on Power Electronics, vol. 29, no. 6, pp.
Hydrogen Joint Technology Initiative under grant 2661-2665, 2014. http://dx.doi.org/10.1109/TPEL.2013.2294714.
agreement n° 621213. [15] F. Krismer, D.Sc. dissertation: Modeling and optimization of
bidirectional dual active bridge DC-DC converter topologies,
Eidgenössische Technische Hochschule ETH Zürich, 2010.
[16] P. Giammatteo, C. Buccella and C. Cecati, "Matlab/Simulink
References modeling of SiC power MOSFETs," International Review of
Electrical Engineering (IREE), vol. 9, no. 4, pp. 671-680, 2014.
[1] B. Zhao, Q. Song, W. Liu and Y. Sun, "Overview of dual-active-
http://dx.doi.org/10.15866/iree.v9i4.2909.
bridge isolated bidirectional DC-DC converter for high-
[17] Ertay, M.M., Tosun, S., Zengin, A., Simulated annealing based
frequency-link power-conversion system," IEEE Transactions on
passive power filter design for a medium voltage power system,
Power Electronics, vol. 29, no. 8, pp. 4091-4106,
(2013) International Review of Electrical Engineering (IREE), 8
2014.http://dx.doi.org/10.1109/TPEL.2013.2289913.
(1), pp. 354-361.
[2] W. Huiqing, X. Weidong and S. Bin, "Nonactive power loss
[18] Ai, Y., Zheng, J., Analysis on affecting factors of harmonic power
minimization in a bidirectional isolated DC-DC converter for
flow, (2014) International Review of Electrical Engineering
distributed power systems," IEEE Transactions on Industrial
(IREE), 9 (3), pp. 585-591.
Electronics, vol. 61, no. 12, pp. 6822-6831, 2014.
[19] J. Hiltunen, V. Väisänen, R. Juntunen and P. Silventoinen,
http://dx.doi.org/10.1109/TIE.2014.2316229.
"Variable-frequency phase shift modulation of a dual active
[3] Q. Hengsi and J. W. Kimball, "Solid-state transformer
bridge converter," IEEE Transactions on Power Electronics, vol.
architecture using AC-AC dual-active-bridge converter," IEEE
PP, no. 99, 2015. http://dx.doi.org/10.1109/TPEL.2015.2390913.
Transactions on Industrial Electronics, vol. 60, no. 9, pp. 3720-
3730, 2013. http://dx.doi.org/10.1109/TIE.2012.2204710.
[4] Z. Biao, S. Qiang, L. Wenhua and S. Yandong, "A synthetic
discrete design methodology of high-frequency isolated Authors’ information
bidirectional DC/DC converter for grid-connected battery energy
storage system using advanced components," IEEE Transactions Lappeenranta University of Technology, Skinnarilankatu 34, 53850
on Industrial Electronics, vol. 61, no. 10, pp. 5402-5410, 2014. Lappeenranta, Finland.
http://dx.doi.org/10.1109/TIE.2014.2304915.
[5] L. Cao, K. H. Loo and Y. M. Lai, "Frequency-adaptive filtering of Vesa Väisänen received the M.Sc. degree in
low-frequency harmonic current in fuel cell power conditioning electrical engineering and the D.Sc.
systems," IEEE Transactions on Power Electronics, vol. 30, no. (Technology) degree from Lappeenranta
4,pp. 1966-1978, University of Technology (LUT), Lappeenranta,
2015.http://dx.doi.org/10.1109/TPEL.2014.2323398. Finland, in 2005 and 2012, respectively. He is
[6] X. Pan and A. K. Rathore, "Novel bidirectional snubberless soft- currently a Post-Doctoral Researcher with LUT
switching naturally clamped zero current commutated current-fed and his current research interests include power
dual active bridge (CFDAB) converter for fuel cell vehicles," in converters, magnetic components and system
Energy Conversion Congress and Exposition (ECCE), 2013. integration.
http://dx.doi.org/10.1109/ECCE.2013.6646939. E-mail: vesa.vaisanen@lut.fi
[7] Y. Xie, J. Sun and J. S. Freudenberg, "Power flow
characterization of a bidirectional Galvanically isolated high- Jani Hiltunen received the M.Sc. degree in
power DC/DC converter over a wide operating range," IEEE electrical engineering from Lappeenranta
Transactions on Power Electronics, vol. 25, no. 1, pp. 54-66, University of Technology (LUT), Lappeenranta,
2010. http://dx.doi.org/10.1109/TPEL.2009.2024151. Finland in 2011, where he is currently working
[8] B. Zhao, Q. Song, W. Liu and Y. Sun, "Dead-time effect of the toward the D.Sc. (Technology) degree. His
high-frequency isolated bidirectional full-bridge DC-DC current research interests include power
converter: comprehensive theoretical analysis and experimental converters and their control.
verification," IEEE Transactions on Power Electronics, vol. 29,
no.4, pp. 1667-1680, 2014.
10
Raimo Juntunen received the M.Sc. degree in

electrical engineering from Lappeenranta
University of Technology (LUT), Lappeenranta,
Finland, in 2011. Since 2011 he has worked at
the Laboratory of Applied Electronics at
Lappeenranta University of Technology. He is
working on a dissertation on the control of high-
power grid-connected inverters.
11
Replacing Copper with New Carbon Nanomaterials

in Electrical Machine Windings
J. Pyrhönen, J. Montonen, P. Lindh, J. J. Vauterin, Marcin J. Otto
Abstract – Electrical machines have significant improvement potential. Nevertheless, the field is
characterized by incremental innovations. Admittedly, steady improvement has been achieved, but
no breakthrough development. Radical development in the field would require the introduction of
new elements, such that may change the whole electrical machine industry system. Recent
technological advancements in nanomaterials have opened up new horizons for the macroscopic
application of carbon nanotube (CNT) fibres. With values of 100 MS/m measured on individual
CNTs, CNT fibre materials hold promise for conductivities far beyond those of metals. Highly
conductive, lightweight and strong CNT yarn is finally within reach; it could replace copper as a
potentially better winding material. Although not yet providing low resistivity, the newest CNT
yarn offers attractive perspectives for accelerated efficiency improvement of electrical machines.
In this article, the potential for using new CNT materials to replace copper in machine windings is
introduced. It does so, firstly, by describing the environment for a change that could revolutionize
the industry and, secondly, by presenting the breakthrough results of a prototype construction. In
the test motor, which is to our knowledge the first in its kind, the presently most electrically
conductive carbon nanotube yarn replaces usual copper in the windings. Copyright © 2015 Praise
Worthy Prize S.r.l. - All rights reserved.
Keywords: Electrical Machine, Winding Material, Carbon Nanotube Yarn, Machine Design,
Efficiency Improvement, Motor Prototype Construction
I. Introduction permanent magnet materials which provide low-loss

excitation for machines [3]-[9].
Since 2009, with the European Commission Nevertheless, considering material physics, traditional
Regulation (EC) No 640/2009 [1], continuous materials are reaching their limits; they do not offer clear
incremental improvement in electrical machines has perspectives for disruptive development of magnetic
contributed effectively to the low-carbon economy circuits. Copper and aluminium have been used as
targets set by the European Union, especially in regard to conductor materials in rotating machines ever since their
the eco-design requirements for electric motors. introduction by electrical engineering pioneers of the
Nonetheless, electrical machines are considered a fairly nineteenth century. Substituting copper and aluminium
mature technology. This can be observed from the fact for other metals does not provide a reasonable solution.
that radical innovations, which are usually pioneered in Silver has only slightly higher conductivity than copper.
niches, seem to have a hard time to break out. Still, Moreover, it is an expensive metal.
electrical machines hold the potential for significant In principle, superconductivity could take the
improvement. Similarly as for other technologies, development of machines a giant step further while
however, a true breakout innovation for accelerated enabling also considerably higher flux densities than
improvement in electrical machines can only happen by those observed with steel-core machines. Yet so far, no
riding along with growth in emerging new technologies such materials could be developed that remain
and associated markets [2]. superconductive in temperatures where rotating electrical
Radical advances that would substantially affect the machines operate.
cost, efficiencies and performance of electrical machines
call for the introduction of new elements. In particular,
new conductive materials are needed to replace the II. Copper Ultimately Increases Losses
traditionally used metals. and Resource Dependency
Many of the significant efficiency improvements in
electrical machines have been initiated by an emergence Copper is the most commonly used conductor material
of enabling new materials technology. Such emerging in electrical machines. For comparison, the conductivity
technologies have been the development of low loss values for copper and other metals are the following. At
magnetic circuit steel materials and high energy density room temperature copper has a conductivity of 59.6
12
MS/m and its resistivity temperature coefficient is [15]. Because of the “one-dimensional” and symmetric
3.886×10−3/K [12]. For silver, the respective values are structure of CNT fibre, the charge carriers can travel
63MS/m and 3.8×10−3/K [11]. But the improvement along the nanotubes almost without “scattering”, which
found by using silver instead of copper would be minor is a phenomenon that is commonly referred to as
because the density of silver is higher than that of copper, “ballistic transportation”. The almost absence of
as a result of which the machine mass might even scattering helps CNTs to carry very high current
increase. The respective value for aluminium are 35 densities, theoretically in the order of J = 100 MA/cm2
MS/m and 3.9×10−3 /K. Aluminium has thus a lower [16]. CNTs could then offer, again in theory, a thousand
conductivity, [10] but aluminium is widely used e.g. in fold increase in current density compared to that of
the rotors of squirrel cage induction motors [11]. Normal metals such as copper. Conductivity values in the range
construction steel has a conductivity of about 4 MS/m of 100 MS/m have been measured on individual single-
and electrical steel qualities can have conductivity values walled armchair CNTs at room temperature, resistivity
in the range of 1.5 – 2 MS/m [13]. values in the range of 1.0×10-8m [11], and the
Despite the fairly high conductivity of copper, a major resistivity temperature coefficient is negative; –0.2×10−3
share of the machine losses still occur in the copper /K [17].
windings; this phenomenon is denoted as copper losses So far, the electrical properties of CNT
[13]. A typical industrial motor has a minimum rated macrostructures such as yarns and sheets are
point efficiency of 91.4 %. At the rated operating point it considerably lower than those of individual CNTs [18].
converts about 1040 W into heat losses. Of these losses, However, from a theoretical viewpoint and assuming that
ca. 45 % are created in the stator winding, 27 % in the the intrinsic high conductivity nature holds from the
rotor windings, 21 % are iron losses and the rest are microscopic level of CNTs to the macroscopic level of
friction and additional losses [13]. In an induction motor, CNT yarn, it would seem that CNT yarn does, in
the copper losses are responsible for ca. 72 % of all the principle, not have a pre-defined upper limit for
losses in the machine, which corresponds to 750 W of conductivity. This means that the electrical conductivity
heat power. properties of CNT yarn are still improving day by day as
Considering that the demand for valuable metals such a result of continuous research and development efforts
as copper is likely to increase tenfold as developing and advances in the field. It is therefore possible to
economies surge ahead and that the production of copper assume that the conductivity of CNT yarn will soon catch
requires energy-intensive extraction from its ore, copper up with and even exceed the conductivity of copper, and
is increasingly considered an energy-expensive and thereby open up enormous development potential for the
carbon-intensive material for use in rotating electrical electrical machine industry.
machinery. Moreover, the levels of cradle-to-gate energy Recently, the Japanese-Dutch company Teijin Aramid
consumption in the production of copper may be very developed a new spinning technology in collaboration
high. Copper has also a relatively high metal density with Rice University in the USA [19]. CNTs are spun to
(8960 kg/m3) which means that copper products may be form multi-fibre yarn by wet spinning technology. The
heavier than comparable other products. industrial applications of the new CNT yarn are,
Increasing the amount of copper in a machine does however, still in their infancy. The problem of spinning
hence not necessarily improve the overall energy highly conductive multifibre yarn is with the purity of the
efficiency of the machine but may negatively affect the CNTs. The Teijin Aramid wet spinning process itself
carbon footprint of the machine. There also remain the results in high-quality joints between individual CNTs
risks associated with growing import dependency on and, as such, has potential to produce high conductivity
metals. The EU, for instance, is dependent on imports to fibre as soon as the problem of CNT purity is solved
meet 50% of its copper demand [14] and, hence, also rigorously. Though the wet spinning technology is yet at
dependent on globally set high copper prices. Aiming at laboratory scale, the techniques provide promising
upgrading the performance and efficiencies of electrical outlooks, and are beginning to bridge the gap between
machines, higher-conductivity wires are searched for basic research and development for CNT-fibre yarn to
windings. meet industrial needs. This assumption is based on
observed similarities between development processes of
CNT yarn and Twaron Aramid yarns, the latter of which
III. Emerging Alternatives from has been successful in scaling up the technology to
Nanomaterials Research industrial volumes. Today, it seems that finally it is
Where metals seem to have hit a ceiling, the new within reach to improve the conductivity in the electrical
nanomaterials may offer a case for more powerful machine windings by replacing metals with new CNT
improvement of electrical machines. Theoretical fibre materials.
physicists and experimentalists have worked on carbon Finally, the most recently achieved resistivity for CNT
nanotube (CNT) materials for over two decades, but only yarn would be low enough to allow for testing yarns in
recently the use of carbon nanotubes (CNTs) in concrete different electrical apparatuses. Applications of carbon
applications and in constructions has become acceptable. conductors have been reported previously, but these
Armchair CNTs are, by structure, highly conductive experiments predominantly focus on the use of CNT
13
materials in electric devices. Examples of these are the operating temperatures than what we are used to apply in
supply of current to an incandescent lamp [19], a 400 Hz electrical machine design today. This is due to the fact
small transformer with CNT windings [20]. that there will be no increase of Joule losses as the
Nanostructured carbon materials for electrical temperature increases. As far as we use steel in the
applications have also been tested in energy storages magnetic circuit, its losses get smaller as the temperature
[10], [21] and in solar cells [12]. The results showed that increases because the eddy currents will be limited by the
nanomaterials could have significant potential as increased lamination resistivity. The eddy current losses
electrode materials e.g. as a cathode for wearable and are inversely proportional to the resistivity of the
lightweight energy storage devices [22]-[24]. laminations [9]. The higher operating temperatures will
naturally result in new challenges regarding the
insulation and permanent magnet materials. Present-day
IV. Exploiting CNT Conductivity: An insulation materials seldom can operate in temperatures
Important Step in the Development higher than 220 C [11] and NdFeB-based permanent
of Electrical Machines magnets should normally be operated below 150 C.
Let us assume that in the near future researchers will
If it will be possible to achieve higher conductivity
succeed in manufacturing 100 MS/m CNT yarn
values by substituting copper for new CNT fibre material
conductors. Copper conductivity is 59.6 MS/m at room
in the windings of an electrical machine, most probably
temperature with a resistivity temperature factor of
the efficiency of the machine will be improved or the
3.886×10−3/K; at 150 C it is only 39.6 MS/m. If eddy
dimensions and mass of the machine will be decreased,
currents and circulating currents decrease the practical
or both. Such development, consequently, will bring
conductivity further by 20 % we stay at the level of about
about the need for change in the machine properties
32 MS/m while CNT yarn could maintain its 100 MS/m
determination and design. It will involve fundamental
change in the overall design principles of electrical at 150 C. This means that, in practice, we could reduce
machines. the Joule losses of a future CNT-yarn electrical machine
In present-day electrical machines the operating by two thirds if the machine geometry is kept the same.
Already today, the properties of available CNT yarn do
winding temperature is normally in the range of 120 C.
allow important niche development, especially in the
Under load the copper DC-conductivity decreases to the
domain of turbomachinery for instance for special high-
level of 42.9 MS/m. This is 72 % of the copper
frequency high-temperature compressor applications.
conductivity at room temperature. In CNT yarn
Furthermore, CNT fibre materials also allow
conductors the resistivity should stay about constant
manufacturing of lightweight and strong electrical
when their temperature has risen to 120 C. Hence, CNT
conductors. The density (CNT  1500 kg/m3) of CNT-
yarn windings may have multiple operating-temperature
fibre yarn is about one sixth the density of copper.
conductivity values compared with the conventional
Already now, the specific conductivity of CNT-fibre
copper windings.
A further benefit of CNT yarn conductors is that they yarn (/ [Sm2/kg]) is close to the level of the specific
consist of very thin sub-conductors which should conductivity of copper. It is therefore realistic to expect
considerably limit the possibility of skin effect at that in the near term ongoing efforts to leverage the
electrical machine frequencies. The yarn inherently properties of CNTs will accelerate the development of
resembles litz wires. Moreover, if we may assume that CNT yarn such that will be able to conduct electricity
future commercial CNT yarn will be manufactured from with the same mass as copper conductors do nowadays.
multifibres which can be easily transposed, no circulating Though the significance of such development will be
currents should appear either. Both skin effect and minor in electrical machines where the space reserved for
circulating currents can be very harmful in traditional conductors is limited, nevertheless, it might greatly
high-current windings and lead into increased AC- benefit the development of other, special applications.
resistance. In the copper conductors of a high-power And last but not least, the availability of carbonic
electrical machine there is always the problem of materials is of major importance for the further
possible skin effect and circulating currents which make development of resource-efficient CNT winding
the AC-resistance of copper conductors significantly machines. Carbon materials are among the most
higher than the DC-resistance value. It is not uncommon abundant materials on the Earth. As such, the presently
that the resistance factor (kR = RAC/RDC) gets values of high price of carbon-based conductors should decrease
150 % in higher-frequency applications [11]. In high- with the growth of industrial applications and
frequency, high-temperature applications we might, commercial use. At present, the price of copper is in the
therefore, theoretically reach a situation where the AC- range of 10 €/kg and its density is about six times the
conductance of a carbon conductor is even three-times density of CNT yarn conductor material [25]. It is not
the conductance of a similar size copper conductor. This expected that CNT yarn conductors will soon beat copper
should bring about a remarkable change in electrical conductors. However, we may expect that in the short
machine design. term CNT materials will enable the creation and
It should then also be possible to design carbon- development of new applications and new devices, there
winding electrical machines that can operate at higher where CNT wires can, in niche applications, bring
14
advantage over copper in light weight, flexibility, However, in our motor application a much lower
bending fatigue resistance, resistance to corrosion, high current must be used because the present-day insulation
strength and high modulus of elasticity. materials should not be exposed to too high operating
Yet today, since the manufacturing development of temperatures. The 26 AWG yarn type was selected for
carbonic conductors is still at laboratory scale, in the test motor. This type has a cross-sectional surface of
practice, the conductivity of CNT yarn conductors is still 0.1280 mm2, an equivalent single wire diameter of 0.405
at a modest level; that is 3.4 MS/m, which is in the range mm and corresponds to CNT yarn of approximately 2000
of construction steels’ conductivities. dtex.
However, poorer conductive steel can be and is often Each motor conductor is made of ten parallel 2000
used in solid-rotor induction motors as rotor conductor dtex yarns. The motor coil conductors are 1.2 m long,
material. The presently poorer conductivity of available and their measured resistance is 0.4 , yielding only 2.4
CNT yarn should therefore not be a barrier for looking at MS/m average conductivity whilst the predicted
the use of CNT yarn for special electrical machine conductivity was 3.4 MS/m. The yarn has also a small
applications. positive temperature coefficient (0.0008/K) whilst we
With this article, we suggest that is now time to focus expected the yarn to have zero or a small negative
research efforts on the largely untapped potential for temperature coefficient. The conductivity level of the
advance in electrical machine technology offered by sample conductors is thus very low. Therefore, the yarn
already available and future CNT materials as these hold can be used only for demonstration purposes; that is to
the promise of significantly improving the conductivity show the potential of CNT yarn for application in an
in electrical machines. In the following section we electrical machine.
present the results of an experiment we have recently Regarding the test motor manufacturing, we faced the
conducted. We have designed and constructed the – to problem that Teijin Aramid CNT-yarn is not provided
our knowledge – first electrical machine in which we with insulation. We solved the problem by preparing a
have used the presently most electrically conductive tape of parallel conductors (with insulating Twaron yarns
CNT yarn to replace the conventional copper wires in the as side protection) on insulating aramid (Twaron) paper
windings. strip. The turn-to-turn insulation is thus made by using
The overall aim of the project was to launch a this kind of insulating paper. We recognized that, while
comprehensive study on the feasibility of improving having parallel conductors on a flat tape, this might
rotating electrical machine efficiencies and performance increase the risk of circulating currents in the windings.
by replacing copper with highly conductive new CNT However, as the winding material has high resistance
materials. One of the main objectives of the experiment we decided to ignore this problem for the test machine.
was to demonstrate that there is an urgent need for Fig. 1 illustrates the conductor material on an Aramid
collaboration between the fields of CNT materials paper tape. For transport purposes the wires are wound
research and engineering research. Collaboration needs on a large paper cylinder. The final conductor is spread
to reach a level where scientists and manufacturers will on a 9 mm wide Aramid paper strip so that the surfaces
work together so as to increase better understanding of of the flat conductor are insulated just on one side.
the benefits and challenges regarding the development The cross-sectional surface dimensions of the flat
and application of carbon nanomaterials to improve the conductor with single-sided insulation are 0.5  9 mm2.
energy and resource efficiency of rotating electrical
machines.
V. The Design and Construction of a

Small Prototype Machine with
CNT Yarn Conductors
Recently, Teijin Aramid BV, in collaboration with
Rice University, has developed CNT yarn with
conductivity in the range of 3.4 MS/m. At Lappeenranta
University of Technology, we have launched a research
project to investigate, understand and demonstrate the
feasibility of using CNT yarn, in rotating machine
windings. At Teijin Aramid, the CNT fibres are
processed in a novel way and the fibres are used to
assemble wire gauges, e.g. 26 and 39 AWG CNT based
yarn. 39 AWG corresponds to 0.00632 mm2 and has
resistance of ca. 50 Ohm/m. Fig. 1. Ten parallel 26 AWG conductors (black ones) with aramid
(Twaron) yellow yarns at the edges glued on aramid paper tape strips.
The maximum allowable current in this wire is 0.7 A
In the Fig. the conductors are placed on a white paper cylinder for
corresponding to 1100 A/mm2. In principle, it can smooth transporting. The ends of the conductors have been treated by
operate at several hundreds of degrees Celsius. silver solution to allow for sleeve joints to external motor cables
15
V.1. The Test Machine

The high resistance and complicated insulation system
of the winding material dictates to a large degree the
design of the machine.
The winding should have only a few turns and the
physical length of the winding must be low enough to
minimize the resistive losses in the low-conductivity
material.
The machine must also be excited by means of
permanent magnets to avoid excitation losses in the
highly resistive conductors. For this purpose we decided
to manufacture a very low voltage permanent magnet
(PM) synchronous machine (Uph = 7 V) with 15000 min-1
rotational speed. In high-speed machines, generally, only
a very few winding turns are needed.
The target was to have less than ten turns per phase
winding. To keep the winding work as simple as possible
we designed a tooth-coil permanent magnet machine
Fig. 2. A graph of the stator design with softened bending angels
with three stator teeth and two rotor poles which gives q at the kinks
= 0.5 slots per pole and phase. Table I gives the main
design data of the machine.
A graph of the stator design with softened bending
angels at the kinks is shown in Fig. 2, and Fig. 3
illustrates a single stator lamination laser-cut from SURA
NO10 and the rotor of the machine.
TABLE I
CNT-YARN PERMANENT MAGNET SYNCHRONOUS MACHINE
DESIGN PARAMETERS IN GENERATING
Parameter Value Fig. 3. Single stator lamination sheet on the left and the assembled PM
STATOR MAGNETIC CIRCUIT rotor on the right. The magnetic circuit form is designed to reserve
Stator stack length lsFe [mm] 42 rectangular spaces for the cross-sections of the aramid-strip-insulated
Stator lamination space factor kFe 0.96 conductors
Stator core material SURA NO10
Stator inner diameter Ds [mm] 25 A three-tooth two-pole machine is the simplest
Stator stack outer electromagnetic diameter Dse possible three-phase machine. It has the benefit that the
75
[mm]
Number of stator slots Qs 3 winding procedure can be simplified when insulated
Number of slots per pole and phase q 0.5 conductors are used, as shown in Fig. 1.
ROTOR The slots are large and therefore the motor could take
Rotor outer diameter Dr [mm] 23 much more conducting material in the stator. The CNT-
PM cylinder diameter, DPM [mm] 20
Rotor PM length, lrPM [mm] 50 yarn space factor is now 15 % and could theoretically be
increased to about 50 %. This should also significantly
PM material (N38UH) remanence Br @ 120 C
1.15 lower the carbon losses. However, the insulation system
[T] illustrated in Fig. 1, does not allow a better space factor
PM relative permeability r at temperature of 120
degrees Celsius
1.05 for the conductors.
PM material coercive field strength Hc [kA/m] 871 To have a high space factor for the CNT yarn, the yarn
Rotor construction: Cylindrical PM is located should be insulated somehow similarly as enamelled
inside a stainless steel tube with 1.5 mm wall copper wires nowadays are.
thickness.
The rotor of the machine has only one cylindrical
STATOR WINDING
Winding type Fractional slot, NdFeB-magnet in a stainless steel tube which allows to
concentrated, non- keep the overall design of the machine as simple as
overlapping, single possible. In a machine with such high resistance in the
layer q = 0.5
winding, PM excitation is, in practice, the only possible
Winding connection: star connected
Number of pole-pairs p 1 option.
Stator coil-turns in series per phase Ns 7 The rotor is made of four parts, two shaft parts, a
In each coil turn there are 10 strands of 0.4 stainless steel tube and a cylindrical magnet with 50 mm
mm diameter. length and 20 mm diameter.
COOLING
Cooling method Air cooling The stator stack end finger plates were manufactured
to have semi-circular cross-sectional areas which makes
the end winding bends as smooth as possible.
16
V.2. No Load and Load Simulations 5
A finite element analysis (FEA) was performed with 4

the CEDRAT Flux 2D software. A dynamic FEA with 3
electric circuit was used to obtain the induced voltages at 2
Induced voltage [V]

no-load. The generator load resistances were adjusted to
1
get sufficient output power. Fig. 4 shows the machine
0
air-gap flux density distribution.
Fig. 5 shows the no-load induced voltage and its -1
harmonic spectrum. The peak value of the induced -2

voltage per phase is 5.6 V. Induced no-load RMS phase -3
voltage EPM = 3.96 V. The machine was then simulated -4
as a generator at its rated load. We used generator -5
simulation instead of motoring because the FEA
34.5 35 35.5 36 36.5 37 37.5 38 38.5
stabilizes much more easily in generating. The phase Time [ms]
resistive loads were adjusted to c. 1.33 Ω. The voltage
and the current curves over one of the load resistors are 6
shown in Fig. 6 and Fig. 7, respectively.
The calculated iron losses of the machine at rated 5
operating point are PFe = 6.34 W. The Joule losses in the
stator – the carbon losses PC are then PC = 3 × 0.4 × 2.32
Induced voltage [V]

4
= 6.35 W. If we would have used copper instead of CNT
yarn, the resistance with the same cross-sectional area
3
would have been only RsCu = 0.022  at 120 C operating
temperature and the corresponding copper loss 0.35 W.
The additional and mechanical losses will be in the 2
range of PMech = 1 W.
1
0.5
0.4 0
0 5 10 15 20 25 30
0.3 Harmonic order
Air gap flux density [T]
0.2
0.1 Fig. 5. Induced voltage at no-load rated speed 15000 min-1. In addition
to the fundamental the voltage contains some fifth and seventh
0
harmonic voltages. The fundamental RMS no-load voltage is EPM =
-0.1 3.96 V
-0.2
5
-0.3
-0.4 4
-0.5 3
0 50 100 150 200 250 300 350 2

Electrical angle [degrees]
1
Voltage [V]
0.6
0
-1
0.5
-2
Air gap flux density [T]
0.4 -3
-4
0.3
-5
0.345 0.3455 0.346 0.3465 0.347 0.3475 0.348 0.3485 0.349 0.3495 0.35
0.2
Time [s]
Fig. 6. Phase voltage under generating to 30 W resistive load

0.1
0 The total loss of the CNT-yarn machine under rated

0 5 10 15 20 25 30
Harmonic order operating is thus about 13.7 W at generator 30 W output
power. This yields a generating efficiency of about ηCNT
Fig. 4. Air gap flux density peak value 0.51 T and RMS value 0.361 T. = 0.69. With copper conductors the loss would be 7.7 W
The deep dips in the otherwise fairly sinusoidal waveform result from and the efficiency at the same operating point in
the permeance variation caused by the slot openings
generating ηCu = 0.8.
17
TABLE II
3 SUMMARY OF THE MACHINE SIMULATED AS GENERATOR
Parameter Absolute value
2 Rated Voltage, V 3.2/5.5
Rated Current, A 2.3
1 Apparent power, VA 33.8
Rated Input Mechanical Power, W 43.7
Current [A]
Rated output power, W 30

0
Rated speed, rpm 15000
Rated torque, Nm 0.019
-1 Back-emf at 15000 rpm 3.96/6.86
Rated impedance,  1.66
-2 Rated inductance, mH 1.057
Calculated synchronous inductance, mH 0.277
-3 Rs, /pu measured 0.4/0.24
0.345 0.3455 0.346 0.3465 0.347 0.3475 0.348 0.3485 0.349 0.3495 0.35
Time [s]
Fig. 7. Phase current under generating to 30 W resistive load
V.3. Machine Design Summary and

Prototype Manufacturing
Table II gives the design summary of the machine
simulated as a generator.
The stator laminations were laser-cut and the stack
was glued to allow easy manufacturing. The stack was
inserted in a steel tube housing and finally the winding Fig. 9. The machine on the right ready to be tested as a generator.
material, illustrated in Fig. 1, was used to prepare the The CNT-yarn-winding machine is on the right and a high-speed
commutator motor grinder machine on the left connected back to back
windings for the machine.
Fig. 8 illustrates the wound stator before and after TABLE III
impregnation. We used normal electrical machine FOUR-WIRE MEASURED PHASE RESISTANCES OF THE TEST MACHINE
impregnation polyester varnish to mechanically fix the Phase, temperature 20°C 50°C 90°C
winding and to make sure that no earth faults should take U, resistance,  0.400 0.416 0.455
V, resistance,  0.393 0.396 0.437
place during the operation of the machine.
W, resistance,  0.388 0.390 0.430
The results indicate that the Teijin Aramid-

manufactured CNT wires have a slightly positive
temperature coefficient for the resistivity.
The slightly positive temperature coefficient may be
explained by a slight increase of electron-phonon
scattering in the material. The temperature coefficient for
the resistivity based on this measurement is in the range
of +0.00155–0.00196/K which is about 40 % of the
corresponding coefficient of copper.
Fig. 8. Machine stator wound with 10 parallel 0.4 mm CNTF-wires
after the winding manufacturing, left. On the right, the same stator after
impregnating. The length of each phase winding is 1.2 m and the
measured DC-resistance is ca. 0.4  at 20 C. These figures also show
V.4.1. No load Measurement
the sparse design of the machine to allow easy winding The no-load measurement yields promising voltage
manufacturing. The CNT yarn space factor is in the range of 15 %
waveforms as they are purely sinusoidal (Fig. 10) and
The whole machine was assembled back-to-back with correspond well to the values calculated with the FEA.
a commercial grinding machine to create a test bench for The voltage was measured by a Yokogawa power
the generator. The test bench is shown in Fig. 9. analyser. Fig. 11 shows that the machine behaves as
predicted and the induced voltage increases linearly with
the speed until 10 000 rpm.
V.4. Measurements At higher speed different eddy current phenomena
We first measured the DC-resistance for the stator (especially in the rotor magnet retaining stainless steel
assembly. The result of two tests show a higher stator cylinder) slightly slow down the increasing of the
resistance than what could have been expected. Table III voltage. The curve in Fig. 11 shows that small saturation
summarizes the four-wire DC-resistance measurement starts to occur as the speed gets higher. The points
results at three different temperatures. indicate the measured values.
18
5
5
4
4
3
Phase U
Voltage [V], Current [A]

3 2
Phase V
2 Phase W
Induced voltage [V]
1
1 0
0 -1
-1 -2
-2 -3
-3 -4
-4 -5
0 0 .0 05 0 .0 1 0 .0 15 0 .0 2
-5 Tim e [s]
0 0.5 1 1.5 2 2.5 3 3.5 4

Time [s] -3 Fig. 12. Measured voltage (peak value c. 4.2 V) and current (peak
x 10
value c. 3 A) of the generator operation with resistive load at the speed
of 15000 min-1
Fig. 10. Induced voltages at no load at 15000 min-1
6 8
6
5
4
Induced voltage [V]
4
2
Voltage [V]
3
0
2 -2
1 -4
-6
0
0 5000 10000 15000
Speed [rpm]
-8
0 0.005 0.01 0.015 0.02 0.025 0.03
Fig. 11. Induced voltage RMS value as a function of rotational speed Time [s]
Fig. 13. Voltage of the motor at the speed 15000 min-1

V.4.2. Load Measurement as Generator at no-load in converter supply
In this test, the machine was rotated with external
mechanical power supply and the machine was loaded as 3
a generator supplying power to the 1 Ω load resistors
which were connected to the machine phase terminals 2
and in star at the other end. Fig. 12 shows the Yokogawa
measured voltage and current over one of the load 1
Current [A]
resistors the machine being at room temperature.

0
V.4.3. No-load Measurement as Motor -1

A small brushless-DC-machine converter, type
Trackstar 60 A, was acquired to test the machine also as -2
a motor. The machine was driven in motor mode at no-
load and the currents and voltages were measured. -3
Fig. 13 presents the no-load voltage at rated speed 0 0.005 0.01 0.015 0.02 0.025 0.03
15000 min1 and Fig. 14 the no-load current, respectively. Time [s]
Fig. 14. Current of a motor phase at the speed 15000 min-1

VI. Conclusion at no-load
This article is the first in its kind to introduce a break- The article scans the environment and indicates some
through approach to the use of new carbon nanomaterials future perspectives for potential applications of carbon
to enable the development of a new generation of nanotube yarn in electrical rotating machines where
rotating electrical machinery. significant efficiency improvement can be achieved.
19
The topicality of the proposed feature article lies in CNTs. We believe that then a new era in electrical
the description of the potential for integration of new and machine development will start. We may mention that, in
greener carbon nanomaterials into electrical machine the course of writing this article, the best laboratory
development and innovation of industrial production results measured on Teijin Aramid CNT fibre yarn have
lines within the context of the transition towards a more been reported to be now in the range of 10 MS/m. Such
resource-efficient economy. In other words, it shows how an increase of conductivity value represents already a
it may now be within reach to considerably improve the significant improvement compared to the CNT yarn
energy efficiency of those millions of electrical rotating material we used in the test motor.
machines that keep industry and society on the move With the introduction of replacing copper with new
while coupling such development to the important and carbon nanomaterials in electrical machine windings we
topical issues of natural resource savings, carbon savings may have the solution to overcome the challenge of agile
as well as cost savings issues for more sustainable and development in electrical machines. CNT materials may
economic growth. The articles uses the experimental help realize the vision of building greener, better
results from prototype building and testing research the resource efficient, and better performing electrical
objective of which was to design and measure the machines. Notwithstanding the risk that further
world’s first electrical motor applying a textile material; technology development in CNT materials will be less
windings made of carbon nanotube yarn. A prototype of fast as may be expected today, the new carbon
a rotating electrical machine using CNT fibre -based nanomaterials can be of considerable significance for
winding material was designed and tested. important niche development in rotating electrical
It demonstrates that CNT yarn has the potential to machinery, there where CNT materials have advantage
replace copper in the machine windings. over copper in light weight, flexibility, high bending
Furthermore, based on the findings from the prototype fatigue resistance, resistance to corrosion, high strength
construction and associated research investigations, the and high modulus of elasticity.
results presented in this article strongly indicate that by
replacing copper with advanced conductive CNT
materials in the machine windings it may be possible to Acknowledgements
improve the efficiency and performance properties of the The authors wish to acknowledge the support of: Ron
future electrical machine. ter Waarbeek for spinning the CNT fibres; Peter Akker
Copper has continued to be used as an effective and Arjan van Haren for assembling the conducting tape;
conductor in the electrical machine windings for over a Dr. Hanneke Boerstoel for commenting on this paper,
century. However, in the search for higher energy Teijin Ltd. for manufacturing and providing the CNT
efficiency we must look for a solution that will decrease yarn; Antti Suikki and Jouni Ryhänen for arranging the
the losses of electrical machines. Copper losses are the manufacturing of the test machine mechanics; Harri
dominating losses in many types of electrical machines Loisa for constructing the winding; Manu Niukkanen and
and, hence, even a slight improvement in the Jan-Henri Montonen for arranging the control logic of
conductivity of the winding materials may revolutionize the converter; Kyösti Tikkanen for arranging the
electrical machine development and call for redesign of laboratory setup; LUT for funding the prototype design,
complete production lines in the entire industry. manufacturing and testing of the prototype machine.
The experimental results presented here show well
enough that further work is needed to pave to way for
supreme CNT materials to be integrated into production References
lines that will change the electrical machine industry. [1] Commission Regulation (EC) No 640/2009 of 22 July 2009.
Theoretically, the DC-resistivity of CNT yarn Available at
conductors in a future CNT-generation motor is, in eurlex.europa.eu/LexUriServ/LexUriServ.do?uri=OJ:L:2009:191:
practical operating temperatures, significantly lower than 0026:0034:EN:PDF
[2] F.W. Geels, “Technological transitions as evolutionary
that of copper conductors. Very thin sub-conductors also reconfiguration processes: a multi-level perspective and a case-
remove all practical skin effect and circulating current study”, Research Policy, vol. 3, 2002, pp. 1257–1274
related problems which further may increase the benefits [3] Z. Wenliang, T.A. Lipo, K. Byung-Il, “Material-Efficient
achieved from using CNT yarn conductors. A significant Permanent-Magnet Shape for Torque Pulsation Minimization in
SPM Motors for Automotive Applications”, IEEE Trans. on
reduction in the conductor resistivity should affect the Industrial Electronics, vol. 61 n. 10, 2014. pp. 5779-5787.
design of machines most significantly. It is then only a [4] A.Boglietti, A.El-Refaie, O.Drubel, A.Omekanda, N.Bianchi,
question of optimization if we shall exploit the properties “Electrical Machine Topologies: Hottest Topics in the Electrical
of CNT yarn material to design smaller and more Machine Research Community”, IEEE Trans. on Industrial
Electronics magazine, vol. 8 n. 2, 2014. pp. 18-30.
lightweight or more energy efficient machines.
[5] A.Tenconi, S.Vaschetto, A.Vigliani, “Electrical Machines for
The CNT-yarn spinning technology is ready for High-Speed Applications: Design Considerations and Tradeoffs”,
manufacturing of high-conductor materials. As soon as IEEE Trans. on Industrial Electronics, vol. 61 n. 6, 2014. pp.
theoretical physicists, materials researchers, engineer 3022-3029.
[6] G.-A. Capolino, A. Cavagnino, “New Trends in Electrical
researchers and experimentalists will closely together
Machines Technology – Part I”, IEEE Trans. on Industrial
they may succeed in selecting the best of pure metallic Electronics, vol. 61 n. 8, 2014. pp. 4281-4285.
20
[7] G. -A. Capolino, A. Cavagnino, “New Trends in Electrical Authors’ information

Machines Technology – Part II”, IEEE Trans. on Industrial
Electronics, vol. 61 n. 8, 2014. pp. 4931-4936. J. Pyrhönen, a Member of IEEE, was born in
[8] Alexandrova, Y., Semken, R.S., Pyrhonen, J., Permanent magnet 1957 in Kuusankoski, Finland, received the
synchronous generator design solution for large direct-drive wind Doctor of Science (D.Sc.) degree from
turbines, (2013) International Review of Electrical Engineering Lappeenranta University of Technology (LUT),
(IREE), 8 (6), pp. 1728-1737. Finland in 1991. He became an Associate
[9] Ismagilov, F., Khairullin, I., Vavilov, V., Electromagnetic Professor of Electrical Engineering at LUT in
processes in the rotor shroud of a High-Speed Magneto-Electric 1993 and a Professor of Electrical Machines and
generator under sudden Short-Circuit, (2014) International Drives in 1997. He is also Head of the LUT
Review of Electrical Engineering (IREE), 9 (5), pp. 913-918. Graduate School of Electrical Engineering. He is engaged in research
[10] J. Liu, R. Younesi, T. Gustafsson, K. Edström, J. Zhu, “Pt/α- and development of electric motors and electric drives. His current
MnO2 nanotube: A highly active electrocatalyst for Li–O2 interests include different synchronous machines and drives, induction
battery,” Nano Energy, Vol. 10, Nov. 2014 pp. 19-27 motors and drives and solid-rotor high-speed induction machines and
[11] K.C.Singh, “Basic Physics”, PHI Learning Private Limited, India drives.
2009.
[12] A. F. Mitul, L. Mohammad, S. Venkatesan, N. Adhikari, S. J. Montonen was born in 1988 in Joutseno,
Sigdel, Q. Wang, A. Dubey, D. Khatiwada, and Q. Qiao, ” Low Finland, received the Master of Science (M.Sc.)
temperature efficient interconnecting layer for tandem polymer degree from Lappeenranta University of
solar cells,” Nano Energy, Vol. 11, 2015 pp. 56-63. Technology (LUT), Finland in 2012 and
[13] J. Pyrhönen, T. Jokinen, V. Hrabovcova, “Design of Rotating Bachelor of Science (B.Sc.) in 2011. He is
Electrical Machines”. John Wiley et Sons 2nd edition, Finland currently doctoral student in LUT. He is doing
2008. the research concerning about electrical
[14] The European Copper Industry’s Manifesto, 2014, available at machines and drives.His research work focuses
www.copperalliance.eu/industry/regulatory-framework/voluntary- on permanent magnet motors, especially those which are used in
initiatives/manifesto (October, 2014) different kinds of traction applications.
[15] X.Wang, N.Behabtu, C.C.Young, D.E.Tsentalovich, M.Pasquali,
J.Kono, “High-Ampacity Power Cables of Tightly Packed and P. Lindh born in Helsinki in 1969, received her
Aligned Carbon nanotubes”, Adv. Functional Materials 2014. M. Sc. degree in energy technology in 1998 and
[16] B.Q. Wei, R.Vajtai, P.M. Ajayan, “Reliability and current her D. Sc. degree in electrical engineering
carrying capacity of carbon nanotubes”, Applied Physics Letter, (Technology) in 2004 from Lappeenranta
vol. 79, n.8, 2001, pp. 1172 - 1174. University of Technology (LUT), Lappeenranta,
[17] P.L.McEuen, M.S.Fuhrer,H.Park, “Single-Walled Carbon Finland. She is currently serving as an associate
Nanotube Electronics”, IEEE Trans.Nanotech. 2002, 1, pp. 78-85. professor at the School of Energy Systems,
[18] M.F.L.De Volder, S.H.Tawfick, R.H. Baughman, A.J. Hart, Lappeenranta University of Technology, where
“Carbon Nanotubes: Present and Future Commercial she is engaged in teaching and research of electric motors and electric
Applications”, Science, Vol. 339, n. 6119, 2013, pp. 535-539 drives. Her research work focuses on permanent magnet motors,
[19] N. Behabtu, C. C. Young, D. E. Tsentalovich, O. Kleinerman, especially concentrated winding machines.
X.Wang, A. W. K. Ma, E.A.Bengio, R. F. ter Waarbeek, J. J. de
Jong,R. E. Hoogerwerf, S. B. Fairchild, J. B. Ferguson, B. J. J. Vauterin, born in Kortrijk, Belgium,
Maruyama, J. Kono, Y.Talmon, Y. Cohen, M. J. Otto, M. received her Master’s degree in philosophy from
Pasquali, “Strong, light, multi-functional fibres of carbon Gent University (1985), and her Doctor’s degree
nanotubes with ultrahigh conductivity”, Science 2013, 339, 182. in philosophy from Lappeenranta University of
[20] L. Kurzepa, A. Lekawa-Raus, J. Patmore, K. Koziol, “Replacing Technology (2012). She is now working as an
Copper Wires with Carbon Nanotube Wires in Electrical associate professor at the School of Energy
Transformers”, Adv. Functional Materials, 2014, 24, pp.619-624. Systems at Lappeenranta University of
[21] C. Chae, H.-J. Noh, J. K. Lee, B. Scrosati, and Y.-K. Sun, “A Technology, where she is engaged in research
High-Energy Li-Ion Battery Using a Silicon-Based Anode and a on issues of knowledge management and societal impact surrounding
Nano-Structured Layered Composite Cathode”, Adv. Functional energy systems development.
materials, 12 Feb 2014, pp 3036–3042.
[22] B. S. Mitchell, “An Introduction to Materials Engineering and Marcin J. Otto. Business Manager and Project
Science: For Chemical and Materials Engineers”, John Wiley & Leader for Carbon Nanotube Fibers at Teijin
Sons, Inc,USA 2004. Aramid BV (The Netherlands). Master of
[23] S.L. Candelaria, Y. Shao, W. Zhou, X. Li, J. Xiao, J.G. Zhang, Y. Science in Physics. PhD. Solid State Chemistry
Wang, J. Liu, J. Li, G. Zao, “Nanostructured carbon for energy (1987). After starting his career as a scientist,
storage and conversion”, Nano Energy, Vol. 1, 2012, pp 195–220 moved quickly to marketing of high
[24] Rajan Prakash, R., Venkatesh, P., Kottaisamy, M., Edwin Xavier, performance fibers. He combines an
S.A., Silicone rubber MgO nanocomposite for high voltage entrepreneurial and customer oriented approach
outdoor insulation application, (2013) International Review of to business development with scientific knowledge to create a
Electrical Engineering (IREE), 8 (5), pp. 1641-1646. profitable, new supply chain.
[25] H.Shimotani, S.Tsuda, H.Yuan, Y.Yomogida, R.Moriya,
“Continuous Band-Filling Control and One-Dimensional
Transport inMetallic and Semiconducting Carbon Nanotube
Tangled Films”, Adv. Functional Materials, 2014.B.Q. Wei,
R.Vajtai, P.M. Ajayan, “Reliability and current carrying capacity
of carbon nanotubes”, Applied Physics Letter, vol. 79, n.8, 2001,
pp. 1172 - 1174.
21
Computation of Electromagnetic Losses in Double-Rotor Vernier

PM Motors with Three Topologies Using TS-FEM
Huijuan Liu1, Yue Hao1, Shuangxia Niu2, Weinong Fu2
Abstract – This paper focuses on the electromagnetic losses computation and comparative
analysis of the proposed three topologies vernier permanent-magnet (VPM) motors by using
time-stepping finite element method (TS-FEM). A dynamic core loss model of numerical analysis
considering two-dimensional (2-D) magnetic properties to grasp the magnetic phenomenon inside
the motors is adopted to compute the core losses of the three VPM motors. The core loss model can
provide reasonable accuracy and all necessary parameters are either directly available or
extracted from manufacturer-provided loss curves. The flux distribution, copper losses, core losses
and efficiency of the three proposed VPM motors have been investigated, and the core loss is the
important loss part of the proposed flux-modulation VPM motors. Copyright © 2015 Praise Worthy
Prize S.r.l. - All rights reserved.
Keywords: Electromagnetic Loss, Core Loss, Efficiency, Vernier Permanent Magnet (VPM)
Motor, Time-Stepping Finite Element Method (TS-FEM)
Nomenclature Kadd Coefficient of additional loss

Cadd Constant coefficient of additional loss
 Axial component of the magnetic vector potential  The solution domain
 Divergence operator r Phase resistor of the stator winding
 Reluctivity of material Ns Number of slots per phase
 Conductivity of material Nc Number of conductors per slot
0 Magnetic permeability in vacuum Sc Cross-sectional area of one conductor
mx The x component of the magnetization vector L Length of the winding in slot
(amperes/meter) in the PM  Resistivity of material
my The y component of the magnetization vector Ss Cross-sectional area of one slot
(amperes/meter) in the PM pcu Copper loss of stator windings
if Winding current I The rms current in stator windings
Df The polarity (+1 or 1) to represent the forward or Sw Cross-sectional area of all the slots
return path p Magnetic pole-pair number of the stator windings
Sf The total cross-sectional area of the region p1 Number of PM pairs
occupied by the coil group in the solution domain p2 Number of stator teeth
Nf Total conductor number of the winding br Residual induction of PM material
a The number of parallel branches in the winding hc Coerctive force of the PM material
mp The symmetry multiplier, which is defined as the
ratio of the original full cross-sectional area to the
solution area I. Introduction
phys Hysteresis loss Recently, saving energy is a most important task in the
peddy Eddy current loss world. Thus, obtaining high-efficiency electrical
padd Additional loss (or excess loss) machines and apparatus is a very important task.
Lz Axial length of the iron core Low-speed direct-drive gearless motors are more
T The period of the applied voltage meritorious than their high-speed counterparts with gear
h Magnetic field vector boxes because the former have no associated gear and no
hirr Irreversible component of h related problems such as oil maintenance, noise issue,
hx-irr The x component of the magnetic field vector hirr losses, and so on. However, conventional direct-drive
hy-irr The y component of the magnetic field vector hirr motors suffer from bulkiness and low efficiency [1]-[2].
b Flux density vector of the magnetic field Vernier permanent magnet (VPM) motors currently
bx The x component of the flux density vector b attract much attention due to their outstanding
by The y component of the flux density vector b flux-modulation feature of providing the high-torque
Keddy Coefficient of eddy current loss
22
Huijuan Liu, Yue Hao, Shaungxia Niu, Weinong Fu
low-speed operation [3]-[5]. There are many new proposed VPM motors. TS-FEM is employed to analyze
structures of VPM motor have been studied, such as the transient performance of three VPM motors and the
double rotor PM motors, double stator motors and core loss calculation method is employed to compare the
multi-rotor multi-stator PM motors [6]-[7]. core losses of the proposed motors.
For the purpose of reducing the volume and improving The proposed TS-FEM is used to study a
the torque density, three novel topologies VPM motors are two-dimensional problem defined in the plane.
presented and designed. And in order to achieve higher The Maxwell’s equations applied to the air-gaps, iron
efficiency for the proposed VPM motors, one method is to cores and PM regions is the following diffusion equation:
decrease the losses in the motors. But first of all, we must
compute the electromagnetic losses accurately, and there φ  m m y 
are two components electromagnetic losses in VPM   φ     0  x   (1)
t  y x 
motors, one is the copper loss, the other is core loss, and it
is still challenging to predict core loss under transient
conditions in the design of motors and the model of power The left term of Eq. (1) only exists in PM regions. The
loss has been continuously under study for over one magnetic field equation in the region of the stranded
hundred years. Currently, the electric machine design windings can be expressed as:
engineers depend heavily on finite element analysis
Df N f
(FEA), requiring an accurate core loss formula, in order to   φ   if  0 (2)
achieve better design of efficient machines. S f am p
Usually, machine design engineers select the loss
computation method based on two final goals: accuracy To achieve maximum energy efficiency of the three
and efficiency. Different core loss computation can be VPM motors in design is to minimize the losses including
applied to the FEA in order to achieve the design goals, the core and copper losses. In other hands, the power
such as the micromagnetic eddy current model [8]-[11], losses may be used as an indirect measure of the
the dynamic hysteresis core loss model [12], [13] and the temperature rise of the three motors as a first order of
traditional core loss model [14]. In general, only approximation.
numerical solution of micromagnetic equations based on The Maxwell’s tensor equation is used to calculate the
3-D Landau–Lifshitz–Gilbert equations, together with magnetic force on the rotor. The core losses calculation in
Maxwell’s equations, can provide appropriate and the frequency domain is based on the following equation
accurate physical interpretation of magnetic material [15]:
losses [8]. However, solution of the complicated pcore  phys  peddy  padd (3)
micromagnetic equations requires extensive
computational time, which is prohibitive for engineering
The three types of loss in the time domain can be
design [10]. Accordingly, a dynamic core loss model [15]
computed by:
of numerical analysis considering two-dimensional (2-D)
magnetic properties to grasp the magnetic phenomenon
Lz T db 
inside the motors is adopted to compute the core losses in phys  0   hirr  dt d   dt 
the three VPM motors, and the details of this dynamic T
core loss model will be presented in section II. (4)
L T  db dby  
Compared with traditional VPM motors, the three  z 0    hx irr  x  hy irr   d   dt
T  
 dt dt  
novel flux-modulation double rotor VPM motors make
more full use of the inner space and save more copper
material depending on their double rotor topology and Lz  db  2  db  2 
y
fractional slot concentrated winding [16-17]. peddy   K eddy  x      d (5)
According to the PM location, they can be referred as 2 2   dt   dt  
 
motor I (PM in rotor), motor II (PM in stator and rotor),
and motor III (PM in stator). They have the similar 3
operation principles. In this paper, the electromagnetic Lz  db 2  db  2  4
 x    y   d 
core losses, copper losses, efficiency, air-gap radial flux padd 
Cadd  K add  dt   dt  
(6)
density and flux distributions of three VPM motors are  
investigated and compared quantitatively. Three VPM
motors’ performances are analyzed by using The coefficient Keddy and Kadd can be calculated
time-stepping finite element method (TS-FEM), and their according to the core loss curves using linear regression
respective advantages and disadvantages are investigated. algorithm. According to [15], Cadd=8.763363.
In the analysis, the phase resistor of the stator winding
in the effective region (excluding the end winding) is:
II. TS-FEM and Electromagnetic
Losses Analysis 1 L 1 L
r N s Nc  N s N c2 (7)
There are two air-gaps and two rotating bodies in the 3 Sc 3 Ss
23
Then the total copper loss is: two sets of 18 pole-pair PMs are added and embedded in
the stator slots, then the PM pole-pairs p1 is 17, stator teeth
L L p2 is 18, and the pole-pair number of the stator windings p
pCu  3I 2 r  I 2 N s N c2  I 2 N s2 N c2 (8) is 1. In Fig. 1(c), the rotor of motor III is designed with
Ss Sw
salient structure, and each side of the stator surface is
mounted with 18 pole-pair PMs as motor II, then the PM
Sw is the cross-sectional area of all the slots and it will pole-pair number p1 is 18, and the number of steel
be the same in all the motors being studied for consistent segments p2 is 17 and the pole-pair number of the stator
comparison. NsNc is the total number of conductors in all windings p is 1.
the slots of each motor.
From Eq. (8) it can be seen that for the three VPM
motors, owing to the stator and the stator windings, and
the rated stator current (4A) are all the same, then the
ampere-conductors INsNc are the same and the copper loss
are the same.
III. Constructions and Operation

Principle of VPM Motor
The operation principle of VPM motors is based on the
magnetic gear (MG) effect, that is, in MGs, the low speed
multiple-pair and high-speed motor with fewer pole-pair
PMs on the two rotating parts can be magnetically coupled
with one another through their flux modulating ferrite
poles. The relative movement between the permeance and
the magnetomotive force (MMF) of PMs gives rise to the
desired flux variation.
A small movement of the low-speed PMs results in a
large change in the flux, and the flux rotation can interact
with the high-speed MMF of the PMs to rotate at the same
synchronous speed [18]. Based on this principle, the
high-speed low-torque output can be transferred through a
specific gear ratio to the multiple PMs on the rotor to give
rise to low speed and high torque.
In VPM motors, the basic principle is the same, which
means the MMF created by PMs is modulated by the
air-gap permeance and hence a desired fast flux variation
is produced. The major difference is that the high-speed
MMF is produced by the steady stator windings rather
than by the PMs. Based on such operation principles, Figs. 1. Configuration of three novel VPM motors
different types of VPM motors can be designed.
The PMs may be mounted on the rotor, the stator The main dimensions of the three proposed VPM
respectively, and the PM may be also installed on the rotor motors are listed in Table I. As shown in Table I, the outer
and the stator. For the vernier motor [3], the fundamental and inner air-gap length is 0.6 mm, and three VPM motors
rule is: are designed with the same outside radius and axial length,
p =± p2  p1 (9) the same PM materials (NdFeB, br= 1.23 T, hc= 8.9105
A/m), the same iron and copper materials. We assume the
current densities in the stator conductors are all the same,
Figs. 1 show the configurations of three novel
limited at 6A/mm2.
double-rotor VPM motors. Three VPM motors are
designed with the same stator between the inner and outer
rotor, and one pole-pair, three-phase drum windings are IV. Electromagnetic Losses Computation
placed in the open stator slots. There are two air-gaps and
18 slots on each side of the stator. For the proposed three VPM motors, the rated speed of
In Fig. 1(a), motor I is designed with 17 pole-pair PMs the motors is 300rpm. The rated current is 4A. Based on
mounted on the surface of the outer and inner rotor, then the above design data and TS-FEM, the air-gap flux
the PM pole-pairs p1 is 17, stator teeth p2 is 18, and the density and flux distribution of three VPM motors are
pole-pair number of the stator windings p is 1. computed. Since the structures and PM locations of three
In Fig. 1(b), motor II is designed with 17 PMs mounted motors are different, the air-gap flux density and flux
on the surface of the outer and inner rotor, and in the stator, distributions are also quite different.
24
TABLE I The computed core loss curves versus time at different

MAIN DIMENSIONS OF THREE VPM MOTORS
operation of the three VPM motors are shown in Figs. 2.
Motor type Motor I Motor II Motor III
Outside radius of outer rotor (mm) 136 136 136
Owing to the magnetic flux density of three VPM motors
Inside radius of outer rotor (mm) 112 112 112 at locked-rotor condition is the lowest; therefore the core
Outside radius of stator (mm) 110.8 110.8 110.8 loss of three VPM motors at locked-rotor condition is the
Inside radius of stator (mm) 66.4 66.4 66.4 lowest one. For motor I, the core loss at full-load
Outside radius of inner rotor (mm) 65.8 65.8 65.8
Inside radius of inner rotor (mm) 36 36 36
condition is larger than that of no-load condition by 9.6%;
PM thickness in the stator (mm) 0 3 3 for motor II the core loss at full-load condition is about
PM thickness in the rotor (mm) 3 3 0 equal to that of no-load condition, and for motor III,
Air-gap length (mm) 0.6 0.6 0.6 owing to the salient rotor structure, the core loss at no-load
Axial length (mm) 60 60 60
condition is larger than that of full-load condition by
Number of stator slots 18 18 18
Pole pairs of stator winding 1 1 1 54.2%.
PM numbers on stator 0 18 36 Fig. 3 shows the core losses curves versus time at
PM numbers on rotor 34 17 0 full-load operation of three VPM motors. The computed
Frequency (Hz) 85 85 85
2-D core loss distribution at full-load operation for three
VPM motors is shown in Figs. 4. From Fig. 3, it is noted
The TS-FEM is used to compute the core losses when that motor II suffers from the highest core-loss (109W)
the three VPM motors running at locked-rotor, no-load while motor III has the smallest core-loss (19W), and the
and full-load condition. core-loss of motor I is 83W.
120
100
CoreLoss (W)
80
60
40
20
0
0 5 10 15 20
Time (ms)
(a) Motor I
(a) Motor I
140
120
CoreLoss (W)
100
80
60
40
20
0
0 5 10 15 20
Time (ms)
(b) Motor II
(b) Motor II
80
CoreLoss (W)
60
40
20
0
0 5 10 15 20
Time (ms)
(c) Motor III
Figs. 2. Core-loss vesus time at different operation of three VPM motors (c) Motor III
140 Figs. 4. Core-loss distribution at full-load operation

120 of three VPM motors
CoreLoss (W)
100
80 TABLE II
60 PERFORMANCE COMPARISON OF THREE VPM MOTORS
40
Motor type Motor I Motor II Motor III
20
Core loss at locked rotor (W) 9 11 10
0
0 5 10 15 20
Core loss at no-load (W) 75 107 45
Time (ms)
Core loss at full-load (W) 83 109 19
Copper loss (W) 0.05 0.05 0.05
Fig. 3. Core-loss vesus time at full-load operation of three VPM motors Efficiency (%) 91 90 96
25
V. Conclusion May 2005, pp. 10E311(1-3).

[11] Sarrafan, K., Kashani, M., Darabi, A., The effect of the inverter
In this paper, a dynamic core loss model has been supply on core losses in Axial Flux Permanent Magnet Motor,
adopted for the computation of core losses in the proposed (1), pp. 81-88.
three VPM motors. The field distributions, copper losses, [12] V. Basso,G. Bertotti, O. Bottauscio, F. Friorillo,M. Pasquale, M.
core losses and efficiency of the three motors are Chiampi, and M. Repetto, Power losses in magnetic lamination
investigated by using time-stepping finite element method with hysteresis: Finite element modeling and experimental
validation, J. Appl. Phys., vol. 81, no. 8, Apr. 1997, pp. 5606-5608.
(TS-FEM). [13] A. Benabou, S. Clenet, and F. Poriou, Comparison of Preisach and
1) It can be noted that the three VPM motors have very Jiles-Atherton models to take into account hysteresis phenomenon
low copper losses compared to the corresponding core for finite element analysis, J. Magn. Magn. Mater., vol. 261, 2003,
losses. And the core loss is the most part of pp. 139-160.
[14] D. M. Ionel, M. Popescu, S. J. Dellinger, T. J. E. Miller, R. J.
electromagnetic loss in the proposed VPM motors. Heideman, and M. I. McGilp, On the variation with flux and
This is because of the open stator slot structure and the frequency of core loss coefficitens in electrical machine, IEEE
“drum winding” structure effectively reduces the end Trans. Ind. Appl., vol. 42, no. 3, May/Jun. 2006, pp. 658-666.
winding length and reduces the copper losses. [15] D. Lin, P. Zhou, W. N. Fu, S. Stanton, and Z. J. Cendes, A dynamic
core loss model for soft ferromagnetic and power ferrite materials
Therefore, in order to improve the efficiency of the in transient finite element analysis, IEEE Trans. Magn., vol. 40, no.
proposed VPM motors, the most effective method is to 2, Mar. 2004, pp. 1318-1321.
reduce the core loss. [16] Jiangui Li, K.T.Chau. Performance and Cost Comparison of
2) If only consider the efficiency, motor III is better than Permanent-Magnet Vernier Manchines, IEEE Transactions on
applied superconductivity, vol. 22, no. 3, 2012, pp.5202304(1-4).
motor I, and motor I is better than motor II, and the [17] Chunhua Liu, K.T.Chau, Jin Zhong, Wenlong Li, and Fuhua Li.
core loss of motor II is the largest one and that of Quantitative Comparison of Double-Stator Permanent Magnet
motor III is the lowest one. But the efficiencies of the Vernier Machines With and Without HTS Bulks, IEEE
three VPM motors are all above 90%. Transactions on applied superconductivity, vol. 22, no. 3, 2012,
pp.5202405(1-5).
The comprehensive analysis of the results indicates that [18] K. Atallah and D. Howe, A novel high-performance magnetic gear,
different topologies of VPM motors can be used in IEEE Trans. Magn., vol. 37, no. 4, Jul. 2001, pp. 2844-2846.
low-speed direct-drive applications to achieve optimum
performance.
Authors’ information
1
School of Electrical Engineering, Beijing Jiaotong University, Beijing,
Acknowledgements 100044, China.
This work was supported by the National Natural 2
Department of Electrical Engineering, The Hong Kong Polytechnic
Science Foundation of China (No.51377008). University, Hong Kong.
Huijuan Liu received Ph. D. in Electrical

References Engineering from Beijing Jiaotong University
(China) in 2009, and received the B.S. degree and
[1] Abdi, B., Bahrami, H., Ghiasi, M.I., Ghasemi, R., PM machine M.S. degree from Tianjin University (China) in
optimization in variable speed EMB application, (2012) 1989 and 1994 respectively. Since December
International Review of Electrical Engineering (IREE), 7 (3), pp. 2005, she has been an Associate Professor with
4412-4418. Beijing Jiaotong University. Her current research
[2] Li, Z., Magnetic field analysis of a novel deflection-type PM interests mainly focus on numerical methods of
multi-DOF actuator, (2013) International Review of Electrical electromagnetic field computation, optimal design and control of
Engineering (IREE), 8 (1), pp. 89-95. induction machine, doubly fed brushless machine, and permanent
[3] A. Toba and T. A. Lipo, Generic torque-maximizing design magnetic machine for wind power and other new power source
methodology of surface permanent-magnet vernier machine, IEEE development.
Trans. Ind. Appl., vol. 36, no. 6, Nov. 2000, pp. 1539-1546.
[4] S. Niu, S. L. Ho, and W. N. Fu, A novel direct-drive dual-structure Yue Hao is a master in Electrical Engineering of
permanent magnet machine, IEEE Trans. Magn., vol. 46, no. 6, Beijing Jiaotong University. Her research interest
June 2010, pp. 2036-2039. is analysis, design of new structure electrical
[5] S. Niu, S. L. Ho, W. N. Fu, and L. L. Wang, Quantitative machines, such as veriner machine,
comparison of novel vernier permanent magnet machines, IEEE flux-modulated motor for direct-drive
Trans. Magn., vol. 46, no. 6, Jun. 2010, pp. 2032-2035. applications.
[6] J. Li and K. T. Chau, A novel HTS PM vernier motor for
direct-drive propulsion, IEEE Trans. Appl. Supercond., vol. 21, no.
3, May 2011, pp. 1175-1179.
[7] S. L. Ho, S. Niu, and W. N. Fu. Design and comparison of Vernier Shuangxia Niu received the B.Sc. and M.Sc.
permanent magnet machines, IEEE Trans. Magn., vol. 47, no. 10, degrees in electrical engineering from Tianjin
2011, pp. 3280-3283. University in 2002 and 2005 respectively, and
[8] G. Bertotti, Hysteresis in Magnetism. New York: Academic, 1998. the Ph.D. degree in electrical engineering from
[9] G. Bertotti, A. Magni, I. D.Mayergoyz, and C. Serpico, The University of Hong Kong in 2009. She
Landau–Lifshitz magnetization dynamics and eddy currents joined the Department of Electrical Engineering
inmetallic thin films, J. Appl. Phys., vol. 91, no. 10, May 2002, pp. at the Hong Kong Polytechnic University as a
7559-7561. Research Assistant Professor in August 2012.
[10] G. Hrkac, M. Kirschner, F. Dorfbauer, D. Suess, O. Ertl, J. Fidler, Her research interests mainly focus on numerical methods of
and T. Schrefl, Three-dimensional micromagnetic finite element electromagnetic field computation and novel electrical machine design
simulations including eddy currents, J. Appl. Phys., vol. 97, no. 10, and driver control.
26
Weinong Fu received the Ph.D. in electrical

engineering from The Hong Kong Polytechnic
University in 1999. Since October 2007, He has
been an Associate Professor with The Hong
Kong Polytechnic University. Previously, he was
one of the key developers at Ansoft Corporation
in USA. During his seven years at Ansoft, he
focused on the development of the commercial
software Maxwell. He has published over 110 papers in journals and in
leading conferences. His current research interests mainly focus on
numerical methods of electromagnetic field computation, optimal design
based on numerical models, applied electromagnetics and novel electric
motors.
27
Optimum Design of a Surface Mounted

Fractional Slots Permanent Magnet Motor
A. Mansouri, H. Msaddek, H. Trabelsi
Abstract – In recent years, permanent magnet machines have been widely used in many
applications such as electric vehicles and hybrid applications. The present work carries out the
design of a surface mounted permanent magnet motor (SMPM) with fractional slots. At a first
stage, an analytical model of the SMPM is developed. At a second stage, based on this derived
model, an optimization procedure of the machine, using the Genetic algorithm, is applied to find
optimal machine geometry in agreement with a set of design criteria. The used objective function
is to maximize the machine efficiency.
A SMPM with 18 slots and 16 magnets is chosen. At a last stage, the obtained analytical results of
the optimal design are validated with finite element analysis (FEA). Based on this validation a
great concordance between analytical and FEA results is shown. Copyright © 2015 Praise
Keywords: Fractional Slots SMPM, Analytical Modeling, Optimization, Efficiency, Optimal

Geometry, Finite Element Analysis
Nomenclature kc Carter factor

ke Excess constant
Asl Slot area kh Hysteresis constant
Acu Copper area kopen Slot opening ratio
Acond Area of the copper conductor kw Winding factor
bts Stator tooth width lm Magnet thickness
bs0 Slot opening L Active machine axial length
bss1 Inner stator slot width m Number of phases
bss2 Outer stator slot width n Rated speed
Bm Maximum airgap flux density above the nc Number of conductors
magnets Nm Number of permanent magnets
Br Remanence flux density of the magnet p Pole pairs number
Bst Stator teeth flux density P Rated power
Bsy Stator yoke flux density Pcu Copper losses
Bry Rotor yoke flux density Pteeth Stator teeth iron losses
Bm,teeth Stator teeth peak flux density Pyoke Stator yoke iron losses
B̂ Fundamental airgap flux density Qs Number of slots
Bm,yoke Stator yoke peak flux density rcu Copper radius
dlam Lamination sheet thickness Rph Phase resistance
D0 Stator outer diameter Ŝ1 Fundamental peak current loading
Drc Rotor outer diameter U Nominal voltage
Di Rotor inner diameter Vteeth Stator teeth volume
D Stator inner diameter Vyoke Stator yoke volume
f Frequency σiron Iron lamination conductivity
fs Fill factor of the stator winding δ Airgap length
hsy Stator yoke height α Half pole angle
hsw Slot wedge height s Stator slot pitch
hss Stator slot height µr Relative permeability of the magnet
hry Rotor yoke height Γ Output torque
I Rated current loading β Machine saliency, Steinmetz constant
Iph Phase rated current ρcu(120) Copper resistivity at 120°
I Maximum value of the current τc Coil pitch
Js Stator current density
28
I. Introduction In the presented work, the genetic algorithm is used to

optimize the considered permanent magnet machine.
Due to their several advantages, such as high The present paper describes the optimization
efficiency, compact structure, small size, low weight and procedure of a 16 poles 18 slots fractional- slots-
flexibility of control, permanent magnet machines concentrated-windings permanent magnet machine. In
becomes nowadays a good choice in so many industrial section II the machine geometrical parameters and sizing
applications. High speed manufacturing [1], electric and equations are reported. Section III presents the analytical
hybrid vehicle traction [2], [3] and wind power model of the machine.
generation [4], [5] are some applications of these The optimization procedure design is the objective of
machines. the section IV. In the last section are presented the FEA
Taking into account these advantages, replacing and the optimization results.
conventional induction machines with permanent magnet
ones has recently becomes very suitable. Indeed,
permanent magnet machines tend to reach higher II. Studied Machine
efficiency compared to the induction machines. This is
basically due to the elimination of the rotor copper II.1. Machine Geometry
losses. As mentioned in the previous section, the studied
In the literature, different permanent magnet machine topology is a surface mounted radial flux inner rotor
configurations can be distinguished: radial flux, axial permanent magnet motor.
flux, transversal flux and slotted or without slots stator. The major geometrical of the machine are illustrated
In the presented work, the studied topology is a radial in Fig. 1. Figure 2 shows the slot geometry for the
flux (RF) inner rotor surfaced mounted permanent studied motor topology.
magnet motor (RFSMPM) with slotted stator. The major
advantage of the RFSMPM machine is its simplicity and
therefore its lower manufacturing cost compared to other
permanent magnet machines.
In the case of slotted machine stator, the winding can
be arranged into two ways: concentrated coils around the
stator teeth or distributed coils in the stator slots. Based
on several studies, it has been found that the concentrated
windings are more suitable for numerous advantages [6].
These advantages include, shorter end windings, lower
cogging torque, higher efficiency and higher slot fill
factor particularly applied with segmented slots [7] [8].
For these reasons, PMSMs with fractional slots
concentrated windings are nowadays extensively studied.
However, the major drawback of fractional slot
configuration is the choice of the slot/pole number.
Consequently, the poles and slot combination have to
be chosen properly. In the literature, there are several
studies dealing with the choice of the most ideal slot/pole
combination to reach desired performances ([26]-[28]).
In all industry applications, an efficient design must Fig. 1. Sizing parameters for the SMPM topology
be succeed by an optimization procedure in order to find
the optimum design suitable for the considered
application. Generally, design optimization procedures of
permanent magnet machines aim to determine the
geometrical variables as air gap length, machine
diameters, permanent magnets arrangement, slot
dimensions and so on [9]-[11]. The main objective of
such optimization may be: maximization of the
efficiency or minimization of the machine weight or the
cogging torque. However, because of the large number of
sizing variables and the complexity of the structure, some
variables are set constant in order to limit the number of
design variables. In our case, the outer rotor diameter is
set constant. In the literature, numerous optimization
techniques such as genetic algorithm, pattern search,
simulated annealing, particle swarm and interior point Fig. 2. Slot dimensions
method can be distinguished.
29
The general machine layout in which are reported the Stator yoke height, hsy:
different parts, is illustrated in Figure 3.
D0  D  2hss
hsy  (2)
2
Inner stator slot width, bss1:
D  2hsw
bss1    bts (3)
Qs
Outer stator slot width, bss2:
D  2hss
bss 2    bts (4)
Qs
Stator slot pitch, s:
D
s   (5)
Qs
Fig. 3. Machine layout
Ratio of the slot opening over the slot width, kopen:

II.2. Machine Parameters
bs 0
Generally, the design of the machine starts with the kopen  (6)
definition of the required characteristics. Table I gives bss1
the specifications and the characteristics of the studied
topology. Slot area, Asl:
TABLE I
MACHINE SPECIFICATIONS
Designation Unit Value 1
Rated power, P kW 15
Asl   bss1  bss 2  hss  hsw  (7)
2
Rated speed, n rmp 1500
Nominal line-to-line voltage, U V 203
Desired power factor, cosφ 0.8 Rotor inner diameter, Di:
Number of phases, m 3
Number of poles, 2*p 16 Di  Drc  2hry (8)
Number of slots, Qs 18
hry in Eq. (8) defines the rotor yoke height.

II.3. Materials Modeling
The materials used to define the physical properties of III. Analytical Model of the Machine
the different regions of the motor are described in this
section: III.1. Airgap Flux Density
(i) A linear model of rare earth Neodymium-Iron-Boron
In all procedure design methods, the amplitude of the
(NdFeB) permanent magnet materiel characterized by
constants values: the relative permeability μr, the fundamental airgap flux density B̂ has to be carefully
coercive force Hc and a constant value of the calculated as the magnetic design of the stator starts from
remanence flux density Br, is chosen. this parameter. Let assume a radially magnetized
(ii) For iron regions (stator and rotor core), a non-linear permanent magnets, the following equation gives the
laminated iron model (M800-65A) was used. The maximum value of the airgap flux density (Bm) above the
data required is the B(H) curve. magnets [13]:
Br kleak
II.4. Machine Sizing Bm  (9)
 k
Referring to Figs. 1 and 2, we can deduce the 1 r c
lm
following equations:
Inner stator diameter, D:
kleak is determined according to [12], kc is the Carter
factor [14]:
D  Drc  2lm  2 (1)
30
s Acu is the copper area:

kc  2
(10)
s 
 kopenbss1  Acu  f s Asl (18)
 kopenbss1   5
fs considers both slot and conductor insulations as well
Let us consider a rectangular shape of the airgap flux as gaps [13]. Typical values of the fill factor are in the
range of 0.4:0.5. The area of the copper conductor is
density, the fundamental airgap flux density B̂ can be given by the following equation:
then determined as follows [13]:
Acu
4 Acond  (19)
B̂  Bm sin  (11) nc

Once the number of conductors and the rated current
III.2. Analytical Calculation of the Flux Density are known, we evaluate then the copper losses:
in the Teeth and Yoke
Pcu  R ph I 2ph (20)
The second flux densities to determine are those of the
stator teeth and yoke created by the permanent magnets.
A special attention should be given to these parameters. The resistance per phase winding can be calculated
Because an overestimation than expected leads to the [8]:
saturation of the stator teeth and yoke and hence a high  120  nc  c  2 L  Qs
magnetic flux leakage and high iron losses. R ph  cu (21)
rcu2  m
The flux densities in the stator teeth Bst and yoke Bsy
are connected to the stator teeth width, bst, and stator
yoke height, hsy, through the following equations [15]: The coil pitch τc is calculated as [8]:
 DBm Q D
bst  (12)  c  int  s  (22)
Bst QS  Nm  Nm
 DBm The copper radius rcu is given by:

hsy  (13)
2 Bst p
Asl f s
rcu  (23)
2nc
III.3. Machine Performances
The impact of the different design parameters of the Iron losses
studied permanent magnet motor on the output torque Three iron losses components are usually
can be seen according to the following equation [16]: distinguished: magnetic hysteresis losses ph, classical
eddy currents losses pc and excess losses pe.
4ˆ ˆ 2 According to [17]-[21], in the frequency domain, and
 S1 B k w  D    L sin  (14)
 for a sinusoidal flux density, the total iron losses per
volume unit can be expressed as follows:
By means of the fundamental peak current loading
value and the maximum value of the current, the number ph  K h fBm (24)
of conductors can be deduced as follows:
2
 iron dlam
nc Iˆ  Sˆ1 s (15) pc   2 Bm2 f 2 (25)
6
 Ŝ   pe  K e Bm1.5 f 1.5 (26)
nc  round  1 s  (16)
 Î 
The total iron losses in the different machine regions
The stator current density is then calculated from the are calculated as follows:
rated current loading:
  2 d2 2 2

nc I  KhfBm,teeth   iron lam Bm,teeth f  
Js  (17) Pteeth  6  Vteeth (27)
Acu  8.67 K f 1.5 B1.5 
 e m,teeth 
31
 d2  The first ones guarantee the mechanical structure

 2 2 2
 KhfBm,yoke   iron lam Bm,yoke f   rigidity of the machine, while the seconds one were
Pyoke  6  V yoke (28) applied in order to avoid the magnetic saturation of the
 8.67 K f 1.5 B1.5  machine magnetic circuit. These constraints are
 e m,yoke 
illustrated respectively in Tables III and IV.
The rotor core losses are considered to be small and TABLE III
negligible. This because of the rotor is running at MECHANICAL STRUCTURE RIGIDITY CONSTRAINTS [22]
synchronous speed. Design variable Range
Consequently, the stator core generates the main stator teeth width bts ≥ 0.3s
portion of the total iron losses. stator yoke height hsy ≥ hss/2
Outer stator tooth width 0.15hss ≤ bss2 ≤ 0.5hss
TABLE IV
IV. Optimization Procedure MAGNETIC CONSTRAINTS
Variable Constraint
As described in section II, the machine geometry is
stator teeth flux density Bst 1.5 ≤ Bst ≤ 2
completely defined through 10 parameters. To identify stator yoke flux density Bsy 1 ≤ Bsy ≤ 1.5
these unknown parameters, an optimization procedure rotor yoke flux density Bry 1.3 ≤ Bst ≤ 1.6
must be realized. In the literature several objectives Airgap flux density B̂ 0.8 ≤ Bst ≤ 1.5
function can be focused, minimization of the permanent
magnet material weight, minimization of the total losses
Concerning the used objective function is the
or maximization of the efficiency. In our work, the
maximization of the machine efficiency η defined as
chosen objective function is the maximization of
follows:
machine efficiency. An optimization design procedure is
generally realized throughout three major steps. It begins Pout
 (29)
by identifying the unknown machine design variables. Pout  Ptot
Secondly the machine constraints and the adjective
function are defined and finally an appropriate solver is Pout is rated output power of the machine and Ptot
applied to find the optimum geometry satisfying all the defines the total losses given by:
requirements.
Ptot  Pteeth  Pyoke  Pcu (30)
IV.1. Machine Design Variables As it has been mentioned, the optimization problem is
The machine design parameters are the input variables the maximization of the efficiency, however the used
of the optimization problem from which the machine solver is the genetic algorithm, then the optimization is
geometry is totally defined. In our work, these transformed to a minimization problem as follows:
parameters are: the airgap length, the magnet height, the
pole angle, the stator exterior diameter, the stator teeth objective function = minimum 1    (31)
width, the stator slot height, the slot opening over the slot
width ratio, the slot wedge height, the active machine The flowchart illustrating the optimization procedure
length and the rotor yoke height. is presented in Fig. 4.
All these parameters with their ranges are reported in
Table II.
V. Results
IV.2. Constraints and Objective Function At first the optimization problem is solved by means
In optimization PM machine design several types of of the genetic problem. All simulations are carried out by
constraints can be used. In our case, two classes of applying the Matlab 2012 software. The obtained design
constraints were used. variables, subject of the optimization, are presented in
Table V.
TABLE II
DESIGN VARIABLE RANGES TABLE V
OPTIMUM DESIGN VARIABLE
Design variable Range
airgap length δ (mm) [1;2] Design variable Range
magnet height hm (mm) [2;4] airgap length δ (mm) 1.2
pole angle α (°) [18;21] magnet height hm (mm) 2
stator exterior diameter D0 (mm) [150;200] pole angle α (°) 20
stator teeth width bts (mm) [6;8] stator exterior diameter D0 (mm) 171
stator slot height hss (mm) [18;25] stator teeth width bts (mm) 6.6
slot opening ratio kopen [0.5;0.6] stator slot height hss (mm) 22
slot wedge height hsw (mm) [1;2] slot opening ratio kopen 0.55
active machine length L (mm) [150;200] slot wedge height hsw (mm) 1.6
active machine length L (mm) 165
32
A coupled set is generated for each selected degree of

freedom for each pair of coincident nodes. All nodes
must be checked for coincidence using a defined
command to select all the nodes.
A 2-D flux plot is illustrated in Figure 6. Referring to
this figure, one can notice the existence of certain flux
lines, that instead of crossing the airgap, it turns around
the stator or the rotor cores. These flux lines are the so
called leakage flux.
Fig. 5. Study domain and Boundary conditions
Fig. 4. Flowchart of the optimization procedure
The flux densities over the machine magnetic circuit

Fig. 6. 2-D flux lines
are reported in Table VI. It is to be noted that the
returned objective function value after this optimization Concerning the magnetic flux density values
is 0.94. performed by the machine through the magnetic circuit,
TABLE VI the obtained results are presented in Table VII.
ANALYTICAL MAGNETIC FLUX DENSITIES
Variable Constraint TABLE VII
ANALYTICAL MAGNETIC FLUX DENSITIES
stator teeth flux density Bst 1.55
stator yoke flux density Bsy 1.16 Variable FEA Value Analytical value Error (%)
rotor yoke flux density Bry 1.4 Stator teeth flux
1.6 1.55 3,12
density Bst
Airgap flux density B̂ 0.93
Stator yoke flux
1.14 1.16 1,72
density Bsy
Rotor yoke flux
In what follows, the machine calculated by means of density Bry
1.36 1.4 2,85
the optimization procedure and analytical model is Airgap flux
evaluated. This evaluation is carried out by using the 0.896 0.93 3,65
density B̂
finite element analysis method. Owing to the machine
periodicity and symmetry, the study is limited to 2-D and
to the machine half. The study domain in which are Referring to Table VII, we can clearly notice that
reported the boundary conditions is presented in Fig. 5. there is a good agreement between the analytical results
The nodes interpolation technique, shown in Fig. 5, is and those obtained by FEA. This justifies the good
used to ensure the FEA model continuity when the rotor modeling of the problem either analytically or with the
moves [24]. This technique consists in defining the FEA. The results dealing with the iron losses are
degrees of freedom coupled between coincident nodes presented in Table VIII.
(with a tolerance). According to these results we can deduce a reasonable
precision between numerical and analytical results.
33
TABLE VII pp. 107-121, 2010.

IRON LOSSES [11] P. Salminen, M. Niemelä, J. Pyrhönen, J. Mantere , High-torque
Iron loss low-torque-ripple Fractional slot PM motors, IEEE International
FEA Value Analytical value Error (%) Conference on Electric Machine and Drives, pp. 144-148, 2005.
component
Stator teeth 245 235 4 [12] F. Libert, Design, Optimization and Comparison of Permanent
Stator yoke 371 372 0.2 Magnet Motors for a Low-Speed Direct-Driven Mixer, Master
dissertation, The Royal Insitue of technology, Stockholm 2004.
[13] S. Meier, Theoretical design of surface-mounted permanent
magnet motors with field weakening capability, Master
VI. Conclusion dissertation, Electrical Machines and Power Electronics,
Department of Electrical Engineering, Royal Institute of
In the present work an analytical model, required to Technology, Stockholm, 2002.
optimize a surface mounted fractional slot permanent [14] C. Sadarangani, Design and analysis of induction and permanent
magnet motor, has been developed. magnet motors, Handbook, Royal Institute of Technology,
Stockholm, ISBN 91-7170-627-5, 2000.
In order to find optimal machine geometry, an
[15] N. Bianchi, S. Bolognani, and P. Frare, Design criteria of high
optimization procedure, based on the Genetic algorithm efficiency SPM synchronous motors, IEEE International Electric
has been presented. The purpose of this optimization Machines and Drives Conference, IEMDC'03, pp. 1042-1048,
procedure is the maximization of the machine efficiency. 2003.
[16] C. Sadarangani, Electrical Machines. Design and Analysis of
To achieve this goal, many simulations have been
Induction and Permanent Magnet Motors, Handbook, Royal
carried out to find the optimum machine design Institute of Technology, Stockholm, 2006.
variables. In addition, the FEA has been used to study the [17] G. Bertotti, General propertie of power loss in soft ferromagnetic
machine characteristics. The comparison between FEA materials, IEEE Trans. on Magnetics, vol. 27, pp. 5007-5009,
1989.
and analytical results for the optimized machine data
[18] A. Mansouri, H. Trabelsi and M. H. Gmiden, Iron Losses
shows a good agreement. Calculation of a Surface-mounted Permanent Magnet Motor in
Nevertheless, the applied iron loss model is valid only Transient Finite Element Analysis, In CD-ROM Fourth IEEE In.
in cases of sinusoidal magnetic flux densities. Con. on Signals Systems Decision and Information Technology
In SMPM motors, the variation of the flux density in (SSD’2007), 19-22 Mars 2007, Hammamet, Tunisia.
[19] Trabelsi, H., A. Mansouri, and M. H. Gmiden, On the modelling
iron core is far from being sinusoidal. For this purpose, a of a SMPM and the iron losses calculation in transient FEA,
more accurate iron loss model has to be used especially Trans. on Systems, Signals and Devices (TSSD) Issue son Power
for the hysteresis losses component in which the Electrical Systems, Vol. 3, No. 1, 2008.
hysteresis phenomena have to be considered in the [20] A. Mansouri, H. Trabelsi and M. H. Gmiden, Calculation of
Magnetic Fields and Iron Losses in a SMPM by Using Vector
magnetic field [25]. Preisach Model and Transient Finite Element Analysis,
International Multi-Conference on Systems, Signals and Devices,
2008.
References [21] A. Mansouri, H. Trabelsi, Effect of the Number Magnet-
Segments on the Output Torque and the Iron Losses of a SMPM,
[1] A. Fratta, A. Vagati, and F. Villata, On the evolution of AC International Multi-Conference on Systems, Signals and Devices,
machines for spindle drive applications, IEEE Trans. Ind. Appl., 2013.
vol. 28, no. 5, pp. 1081–1086, Sep.–Oct. 1992. [22] T. Heikkila, Permanent magnet synchronous motor for industrial
[2] Z. Q. Zhu and D. Howe, Electrical machines and drives for inverter applications analysis and design, Phd dissertation,
electric, hybrid and fuel cell vehicles, Proc. IEEE, vol. 95, no. 4, Lappeenranta University of Technology, 2002.
pp. 746–765, Apr. 2007. [23] P. Ponomarev, Tooth coil permanent magnet synchronous
[3] J. Wang, K. Atallah, Z. Q. Zhu, and D. Howe, Modular 3-phase machine design for special applications, Phd dissertation,
per-manent magnet brushless machines for in-wheel applications, Lappeenranta University of Technology, 2013.
IEEE Trans. Veh. Technol., vol. 57, no. 5, pp. 2714–2720, 2008. [24] A. Mansouri and H. Trabelsi, Performances Evaluation and Iron
[4] Y. Chen, P. Pillay, and A. Khan, PM wind generator topologies, Losses Modelling of an Inset Surface Mounted Permanent
IEEE Trans. Ind. Appl., vol. 41, no. 6, pp. 1619–1626, 2005. Magnet Motor, Accepted in Trans. on Systems, Signals and
[5] X. Sun, C. Ming, W. Hua, and L. Xu, Optimal design of double- Devices (TSSD) Issue son Power Electrical Systems, 2013.
layer permanent magnet dual mechanical port machine for wind [25] Mansouri, A., Trabelsi, H., Incorporation of vector Preisach
power ap-plication, IEEE Trans. Magn., vol. 45, no. 10, pp. hysteresis model in transient finite element analysis for a SMPM,
4613–4616, Oct. 2009. (2007) International Review of Electrical Engineering (IREE), 2
[6] F. Meier, Permanent-Magnet Synchronous Machines with Non- (3), pp. 448-454.
Overlapping Concentrated Windings for Low-Speed Direct-Drive [26] Mendaci, S., Mekideche, M.R., Allag, H., Analytical model of
Applications, Phd dissertation, The Royal Insitue of technology, radial surface permanent magnet synchronous motor dedicated to
Stockholm 2008. optimal design, (2011) International Review on Modelling and
[7] U. J. Seo, Y. D. Chun, J. H. Choi, S. U. Chung, P. W. Han and Simulations (IREMOS), 4 (4), pp. 1542-1549.
D. H. Koo, General Characteristic of Fractional Slot Double [27] Caballero, D., Zeberio, I., FontÃ¡n, L., MartÃnez-Iturralde, M.,
Layer Concentrated Winding Synchronous Machine, J. Electr. A calculation of the magnetic characteristics of surface mounted
Eng. Technol., vol. 8, no. 2, pp. 282-287, 2013. pmsms under heavy load conditions considering the cross-
[8] T.D. Strous, Design of a permanent magnet radial flux magnetizing effect, (2013) International Review of Electrical
concentrated coil generator for a range extender application, Phd Engineering (IREE), 8 (4), pp. 1181-1188.
dissertation, Delft University of Technology, 2010. [28] Madescu, G., Biriescu, M., Greconici, M., MoÅ£, M.,
[9] P. Salminen, J. Pyrhonen, H. Jussila and M. Niemelä, Performances comparison of two surface-mounted permanent
Concentrated wound permanent magnet machines, International magnet generators with fractional - slot windings, (2013)
Conference on Power Engineering, Energy and Electrical Drives, International Review of Electrical Engineering (IREE), 8 (1), pp.
pp. 514-517, 2007. 75-80.
[10] A. M. El-Refaie, Fractional-slot concentrated-windings
synchronous permanent magnet machines opportunities and
challenges, IEEE Transactions on Industrial Electronics, vol. 57,
34
Laboratory of Computer, Electronics & Smart Engineering Systems
Design Engineering School of Sfax, Tunisia.
Dr. Ali Mansouri received the B.S. degree in

electromechanical engineering, the M.S. degree
in electrical machine analysis and control and
the Ph. D. degree in electrical engineering from
Sfax Engineering School (SES), University of
Sfax, Sfax, Tunisia, in 2002, 2003 and 2009
respectively. He is working in the field of
electrical machine design at SES. He joined the
Tunisian University, Tunisia, in 2003 as an Assistant at Gafsa
Engineering Institute. Currently he is working as Associate Professor in
the Higher Institute of Applied Science and Technology of Gafsa. He is
a member of the Laboratory of Computer, Electronics & Smart
Engineering Systems Design the University of Sfax.
E-mail: mansouriali2002@yahoo.fr
Hejra Msaddek received the B.S. degree in

electrical and automatic engineering, the M.S.
degree in automatic and intelligent technic from
Gabes Engineering School (GES), University of
Gabes, Gabes, Tunisia, in 2011and 2012
respectively. She is working in the field of
electrical machine design and optimization at
Sfax Engineering School SES. She is a member
of the Laboratory of Computer, Electronics & Smart Engineering
Systems Design the University of Sfax.
E-mail: msaddek_hejra@yahoo.fr
Pr. Hafedh Trabelsi received the B.S. degree

from Sfax Engineering School (SES), University
of Sfax, Sfax, Tunisia, in 1989, the M.S. degree
in the Central School of Lyon, France, in 1990,
the Ph.D. degree from the University of Paris XI
Orsay, France, in 1994 and the "Habilitation à
Diriger des Recherches" (accreditation to lead
research) degree from Sfax Engineering School
(SES), University of Sfax, Sfax, Tunisia, in 2009 all in electrical
engineering. He is working toward the Research Management Ability
degree in the field of electrical machine design at SES. He joined the
Tunisian University, Tunisia, in 1995 as an Assistant Professor at Gafsa
Engineering Institute. He is currently a Professor of Electrical
Engineering at SES. He is a member of the Laboratory of Computer,
Electronics & Smart Engineering Systems Design the University of
Sfax.
E-mail: Hafedh.trabelsi@yahoo.fr
35
Real Time Implementation of Sliding Mode Control

for Induction Motor Drives Using dSPACE
A. Essalmi, H. Mahmoudi, A. Bennassar, A. Abbou, Y. Zahraoui, M. Akherraz
Abstract – The principal aim of this paper is devoted to real time implementation of sliding mode
control (SMC) strategy based on indirect field oriented control (IFOC) for an induction motor
(IM) drive. The proposed method is a powerful tool to use for a robust nonlinear control law to
model uncertainties and disturbances. Three controllers qualified as sliding mode controllers are
investigated. The design and implementation for a 1.5 kW induction motor drive are presented and
completely carried out using a dSPACE 1104 controller board based on digital signal processor
(DSP). Experimental results show the improvement in performances of the proposed control
algorithm. Copyright © 2015 Praise Worthy Prize S.r.l. - All rights reserved.
Keywords: Induction Motor Drive, Sliding Mode Control Strategy, dSPACE 1104 Controller
Board
Nomenclature in the presence of disturbances and variations of

parameters. These last decades, the sliding mode control
usd,q Stator voltage in the stationary d, q axis knew a big success, due to its simplicity of
isd,q Stator currents in the stationary d, q axis implementation, robustness with respect to uncertainties
rd,q Rotor flux in the stationary d, q axis of system and external disturbances [16]-[24]. This
 Rotor speed method consists to bring the state trajectory to the sliding
Te, TL Electromagnetic and load torque surface and make it evolve over with some momentum to
ωr Rotor electrical speed the balance point [1]-[3]. This paper deals with design
ωsl Slip frequency and implementation using dSPACE 1104 controller
ωs Stator frequency board of sliding mode control (SMC) of an induction
un Discontinue law motor drive. The experimental results show the
ueq Equivalent law effectiveness of the control method. The contents of this
p Number of pole pairs paper are organized as follows. In section two, the
Rs, Rr Stator and rotor resistances oriented model of induction motor is presented, the
Ls, Lr, Lm Stator, rotor and mutual inductances sliding mode control strategy is developed in third
Tr Rotor time constant section and section four presents an experimental setup
σ Leakage coefficient and results. Finally, a conclusion ends the paper.
I. Introduction II. Induction Motor Oriented Model

The low cost, robustness, performance and ease of The dynamic model of the induction motor in (d, q)
maintenance are of interest control in many industrial reference frame for field oriented control can be given as:
applications. Now the development in power and digital
electronics technology can address the variable speed  disd
control of machines in low power applications.    isd  s isq  K  r  1 vsd
 dt Tr  Ls
The researchers have developed various approaches to
 disq
control flux, torque and speed in real time for electrical
   isq  s isd  K  r r  1 vsq
machines [1]-[24]. The algorithms of the conventional  dt  Ls
control using proportional and integral (PI) controller can  d rd Lm
precisely control an undisturbed linear process which has   isd  1 r (1)
 dt Tr Tr
constant parameters. Thus, if the controlled system is
 d rq Lm
subjected to disturbances or changes in system   isq  s  r  r
parameters, automation adaptive solution by adjustment  dt Tr
of controller parameters is mandatory to keeps in  d  PLm ƒ T
   r isd    L
advance the fixed performance of the controlled system  dt JLr J J
36
where: The condition of convergence is defined by the

equation of Lyapunov [5], it make the surface attractive
Rs 1   1 and invariant:
   ; K
 Ls  Tr  Lm
S  x  S  x   0 (7)
L2 L
  1  m ; Tr  r
Ls Lr Rr The control algorithm is defined by the equation:
The stator voltage angular frequency s is determined u  ueq  un (8)

as the sum of the slip frequency sl and rotor electrical
speed r:
with u is control variable, ueq is the equivalent control
s  sl  r (2) variable and un is the discontinuous control.
The discontinuous function is given [6]:
where sl is expressed as:
u n  K x  sign  s  x   (9)
*
1 isq
sl  (3)
Tr i*sd Kx is a positive gain.
Furthermore: III.1. Sliding Mode Speed Design
r  p  (4) The synthesis of control exploits the sliding modes

technique, by using the principle of cascade adjustment
method, which requires the choice of the surfaces that
Thus the space angle of the rotor flux is given by:
assure the objectives of control.
*
In Eq. (6) we replace by 1, the surface selected for
1 isq the control of speed error is:
s  r   (5)
Tr i*sd
S     *   (10)
III. Sliding Mode Control By derivation of Eq. (10) and we replace the fifth
equation of the system (1) in (10), we obtain:
The sliding mode control is to bring the trajectory
state and to evolve it on the sliding surface with a certain
dynamic to the equilibrium point. To determine the  *  pLm  i  TL  ƒ 
S      (11)
rd sq
JLr J J
sliding surface, we take the form of general equation
given by J. J. E. Slotine [4]:
We define the control law:
n-1
d 
S  x      e  x (6) isq  isqeq  isqn (12)
 dt 
During the sliding mode and in permanent regime,
with e(x) is denote the error of the controlled greatness, λ
is a positive coefficient and n is the order of the system. S     S     0 , isqn  0 . The equivalent control
action can be defined as follow:
JLr  * TL ƒ 
isqeq    J  J  (13)
pLm rd  
During the convergence mode, the condition


S    S     0 must be verified.
Therefore, the discontinue control action can be given
as:
isqn  k   sign  S     (14)
Fig. 1. Block diagram of sliding mode control strategy

K is Positive gain for speed regulator.
37
A. Sliding Mode Currents Design  A 1.5 kW three phase squirrel cage induction
The sliding surfaces for the control of current errors machine, 2 poles, A/3.4A, 220V/400V, 50Hz and
are: 1423rpm;
S  isd   i*sd  isd (15)  A static power electronics convertor from Semikron
composed of a diode rectifier and a three leg voltage
source IGBT inverter;
 
S isq  i*sq  isq (16)  Current are measured using LEM sensors and both of
them are then transformed to be a voltage ranging
By derivation of Eq. (15) and (16) and we replace the from 0 to ±10 volts which will be the input of A/D
first and second equation of the system (1) in (15) and respectively.
(16) respectively, we obtain:  The rotor speed is measured by an incremental
encoder, which is mounted at the rotor shaft end. A
K 1 digital filter is used to remove the noise from the
S  isd   isd
*
  isd  s isq  r  vsd (17) speed signal;
Tr  Ls
 A dSPACE 1104 DSP board with ControlDesk
software used for real time control of the induction
1
S isq  isq
  *
  isq  s isd  K  r r  vsq (18) motor drive;
 Ls  A personal computer with MATLAB/Sirnulink
software to design a simulate induction machine
We define the control law: control.
The motor parameters used in this experimental
vsd  vsdeq  vsdn (19) investigation are indicated in Appendix. Figs. 2, 3 show
the experimental platform scheme used.
The results presented in the Figures 4-6 below show
vsq  vsqeq  vsqn (20) experimental results of the sliding mode control strategy
for a squirrel cage induction motor. Here, the reference
During the sliding mode, S  isd   S  isd   0 , flux is kept constant at the nominal value 1 Wb.
vsdn  0 and S isq  S isq  0 , vsqn  0 .

   
The equivalent control actions can be defined as
follow:
* K 
vsdeq   Ls  isd   isd  s isq  rd  (21)
 Tr 
vsqeq   Ls isq
*
  isq  s isd  K r  rd  (22)
During the convergence mode, the conditions

S  isd  S  isd   0 and S isq S isq  0 must be verified.
     Fig. 2. Experimental setup
Therefore, the discontinue control action can be given
as:
vsdn  K d  sign  S  isd   (23)
  
vsqn  K q  sign S isq (24)
Kd and Kq are positive gains for current regulators.
IV. Experimental Results

The control algorithm has been implemented using
MATLAB/Simulink and dSPACE DS1104 real time
controller board with TMS320F240 DSP.
The experimental testing ground is essentially Fig. 3. Experimental setup photograph
composed of:
38
Fig. 4. Functioning in step reference speed (150 rad/s)
Fig. 5. Functioning in step reference speed (150 rad/s)
Fig. 6. Functioning in step reference speed (100 rad/s)
39
The SMC strategy show good results in transient [9] A. Essalmi, H. Mahmoudi, A. Bennassar, M. Akherraz, A.
Abbou, neuro-genetic sensorless sliding mode control of a
conditions in speeds. The stator phase current in the
permanent magnet synchronous motor using Luenberger observer,
induction motor remains sinusoidal and takes appropriate Journal of Theoretical and Applied Information Technology, vol.
value. Speed closed-loop control using a sliding mode 53 n. 2, 2013, pp. 236-246.
controller is good for the different cases considered. [10] S. El M.Ardjoun, M. Abid, A.G. Aissaoui, A. Naceri, A robust
fuzzy sliding mode control applied to the double fed induction
machine, International Journal Of Circuits, Systems And Signal
Processing, vol. 5, n. 5, 2011, pp. 315-321, NAUN, USA.
V. Conclusion [11] Y. Sayouti, A. Abbou, M. Akherraz, H. Mahmoudi, Real-time
DSP implementation of DTC neural network-based for induction
This paper presents experimental results of sliding motor drive, PEMD, vol. 3, April 2010, pp. 1-5.
mode control strategy for a 1.5 kW induction motor [12] A. Abbou, T. Nasser, H. Mahmoudi, M. Akherraz, A. Essadki,
drives. The efficacy of these controllers has been verified Induction motor controls and Implementation using dSPACE,
through its hardware and MATLAB/Simulink WSEAS TRANSACTIONS on SYSTEMS and CONTROL, vol. 7, n.
1, 2012, pp. 26-35.
implementations. All experimental results confirm the [13] S. Amamra, L. Barazane, M.S. Boucherit, A. Cherifi, Inverse
good dynamic performances of the developed drive fuzzy model control for a speed control induction motor based
systems and show the validity of the proposed method. dSPACE implementation, Modern Electric Power Systems 2010,
The features of this method, applied to the non linear Wroclaw, Poland, paper P50.
[14] M. Boussak, Implementation and experimental investigation of
systems, that it ensures a fast dynamic response. sensorless speed control of permanent magnet synchronous motor
The proposed controllers are implemented in real time drive, IEEE Transactions on Power Electronics, vol. 20, n. 6,
by employing a dSPACE system with DS1104 controller 2005, pp. 1413–1422.
board based on digital processor Texas Instruments [15] A. Abbou, H. and Mahmoudi, Real time implementation of a
sensorless speed control of induction motor using DTFC strategy,
TMS320F240 DSP. Proceedings of the International Conference on Multimedia
Computing and Systems ICMCS’09, Ouarzazate, Morocco, April
2-4, 2009, pp.327-333.
Appendix [16] Moutchou, M., Abbou, A., Mahmoudi, H., Induction machine
speed and flux control, using vector-sliding mode control, with
TABLE A1 rotor resistance adaptation, (2012) International Review of
INDUCTION MOTOR PARAMETERS Automatic Control (IREACO), 5 (6), pp. 804-814.
Rated power 1.5 kW [17] Khiari, B., Sellami, A., Andoulsi, R., Mami, A., A novel strategy
Voltage 400V Y control of photovoltaic solar pumping system based on sliding
Frequency 50 Hz mode control, (2012) International Review of Automatic Control
Pair pole 2 (IREACO), 5 (2), pp. 118-125.
Rated speed 1440 rpm [18] Houda, B.M., Khadija, D., Said, N.A., New discrete sliding mode
Stator resistance 4.5 Ω controller with predictive sliding function, (2013) International
Rotor resistance 4.54 Ω Review of Automatic Control (IREACO), 6 (4), pp. 529-536.
Inductance stator 0.754 H [19] Abdellah, A., Abdelhafid, A., Mostafa, R., Combining sliding
Inductance rotor 0.7536 H mode and linear quadratic regulator to control the inverted
Mutual inductance 0.7421 H pendulum, (2013) International Review of Automatic Control
Moment of Inertia 0.015 kg m2 (IREACO), 6 (1), pp. 69-76.
Viscous friction 0.003N s/rad [20] Mezouar, A., Terras, T., Fellah, M.K., Hadjeri, S., Composite
sliding mode control of induction motors using singular
perturbation theory, (2012) International Review of Automatic
Control (IREACO), 5 (6), pp. 901-910.
References [21] Guisser, M., El-Jouni, A., Abdelmounim, E.H., Robust sliding
mode MPPT controller based on high gain observer of a
[1] V.I. Utkin, Sliding mode control design principles and
photovoltaic water pumping system, (2014) International Review
applications to electrical drives, IEEE Trans. on Industrial
of Automatic Control (IREACO), 7 (2), pp. 225-232.
Electronics, vol. 40, n. 1, February 1993.
[22] El-Jouni, A., Guisser, M., Salhi, M., A sliding mode optimization
[2] S. Müller, Doubly fed induction generator systems, IEEE Industry
of a photovoltaic pumping system, (2014) International Review
Applications Magazine, vol. 8, n. 3, May-June 2002, pp. 26-33.
on Modelling and Simulations (IREMOS), 7 (3), pp. 466-473.
[3] H. De Battista and al, Sliding mode control of wind energy
[23] Anil Kumar, T., Venkata Ramana, N., Design of sliding mode
systems with DOIG-power efficiency and torsional dynamics
observer based optimal controller for load frequency control in
optimization, IEEE Trans. Power Systems, vol. 15, n. 2, May
multi area deregulated thermal system connected with HVDC
2000, pp. 728-734.
link, (2013) International Review on Modelling and Simulations
[4] J.J.E. Slotine, W. Li, Applied nonlinear control, Prentice Hall,
(IREMOS), 6 (2), pp. 543-553.
USA, 1998.
[24] Benbouzid, M., Beltran, B., Ezzat, M., Breton, S., DFIG driven
[5] P. Lopez, A.S. Nouri, Théorie Elémentaire Et Pratique De La
wind turbine grid fault-tolerance using high-order sliding mode
Commande Par Les Régimes Glissants, Springer, 2006.
control, (2013) International Review on Modelling and
[6] Shaojing Wen and Fengxiang Wang, Sensorless Direct Torque
Simulations (IREMOS), 6 (1), pp. 29-32.
Control of High Speed PMSM Based on Variable Structure
Sliding Mode, Proceedings of Electrical Machines and Systems,
ICEMS-IEEE, International Conference on, 2008, pp. 995 – 998.
[7] F. Blaschke, The principle of field orientation as applied to the Authors’ information
new transvektor closed-loop control system for rotating-field
machines, Siemens Rev., vol. 34, 1972, pp. 217–220.
[8] Bennassar, A., Abbou, A., Akherraz, M., Barara, M., A LEEP, Department of Electrical Engineering, Mohammed V University,
comparative study of IP, fuzzy logic and sliding mode controllers Mohammadia School’s of Engineers, Rabat, Morocco.
in a speed vector control of induction motor, (2013) International E-mails: abbou@emi.ac.ma
Review on Modelling and Simulations (IREMOS), 6 (6), pp. 1865- yassin.zahraoui@gmail.com
1871. akherraz@emi.ac.ma
40
Adil Essalmi, was born in Taza, Morocco in

1981. He received DESA degree in power
electronics from Mohammadia School’s of
Engineers, Rabat, Morocco in 2007. Currently,
he is pursuing PhD degree at Mohammadia
School of Engineers, Rabat. His researcher
interests include the control strategies for
electrical motor drives.
E-mail: adilh@hotmail.fr
Hassan Mahmoudi was born in Meknes,

Morocco, on January 4, 1959. He received the
B.S degree in electrical engineering from
Mohammadia School’s of engineers, Rabat,
Morocco, in 1982, and the PhD degree in power
electronic from Montefiore Institute of electrical
engineering, Luik, Belgium, in 1990. He was an
Assistant Professor of physics, at the Faculty of
sciences, Meknes, Morocco, from 1982 to 1990. Since 1992, he has
been a Professor at the Mohammadia School’s of engineers, Rabat,
Morocco, and he was the Head of Electric Engineering Department
during four years (1999, 2000, 2006 and 2007). His research interests
include static converters, electrical motor drives, active power filters
and the compatibility electromagnetic.
E-mail: mahmoudi@emi.ac.ma
Abderrahim Bennassar, was born in

Casablanca, Morocco in 1987. He received
Master degree in treatment of information from
Hassan 2 University, Casablanca in 2011.
Currently, he is pursuing PhD degree at
Mohammadia School’s of Engineering, Rabat.
His researcher interests include the control
strategies for Induction Motor Drives.
E-mail: ab.bennassar@gmail.com
41
Performance Improvement of Subspace-Based Direction-Finding

Algorithms Using Higher-Order Statistics
Supawat Supakwong
Abstract – Traditional array signal processing techniques have been relying on the use of
received signal's second-order statistic for many years. However, it suffers with some fundamental
limitations. Studies of array processing based on higher-order statistic has been proposed aiming
to overcome these limitations. This paper is aimed to assess the array performance enhancement
when using higher-order statistic from the differential geometry perspective. Defined as the locus
of all array response vectors over a set of signal parameters, the array manifold's geometrical
shape and properties are known to be crucially important in characterizing the array
performance. In this paper, the geometry of an array manifold associated with a higher-order
statistic is investigated using of the concept of virtual sensor array. Performance analysis is
presented to examine the array performance enhancement both in terms of the Cramer Rao lower
bound and the array detection and resolution capabilities. Copyright © 2015 Praise Worthy Prize
S.r.l. - All rights reserved.
Keywords: Array Signal Processing, Subspace-Based Direction-Finding, Higher-Order Statistic,

Differential Geometry
Nomenclature search for a set of directions of arrival (DOAs) where the

array response vectors lie in the signal subspace.
 Field of real numbers In spite of exhibiting a high resolution performance
 Field of complex numbers and having asymptotically unbiased, the second-order DF
a, A Scalar algorithms suffer with some fundamental limitations
a, A Column Vector such as the limited number of statistically independent
A, Matrix sources that the methods can process, which is bounded
by the number of sensors in the array, the performance
exp  A  Element by element exponential of A degradation due to the presence of additive Gaussian
A Euclidian norm of vector A noise, as well as the inability to deal with non-Gaussian
signals that practically employed in digital
 Frobenius norm of matrix  communications.
T Array processing methods based on the use of
 · Transpose
received signal's higher-order statistic have been actively
 · H Hermitian studied in search to overcome these performance
 Kronecker product limitations. Initially, the research was focused on the
N Identity matrix of size  N  N  development of the fourth-order array processing
methods [3]-[4]. For instance, in [4], two DF methods
were proposed which use the fourth-order cumulants of
the received signal vector; the first is a MUSIC-like
I. Introductory Background
technique based on singular value decomposition of the
The advancements of array signal processing in the cumulant matrix, and the other is the asymptotically
past few decades have been relying firmly on processing minimum variance algorithm derived from a nonlinear
the received signal's second-order statistics (commonly optimization of a system's cost function.
known as the signal covariance matrix). The fourth-order statistic framework was then
In parameter estimation, subspace-based direction extended to exploit the data statistics of an arbitrary even
finding (DF) methods such as the MUltiple SIgnal order, namely 2q  order statistics, where q  1 [5]-[8].
Classification (MUSIC) [1] and the Estimation of Signal In spite of having higher variance, higher-order-
Parameters via Resolution Invariance Technique statistic (HOS) methods offer the array enhancements
(ESPRIT) [2] exploit the eigenstructure of the that the traditional second-order methods lack of. This
information contained in the second- order statistics in includes the ability to 1) process sources, where the
42
Supawat Supakwong
maximum number of sources is not limited by the formulation is first presented. This includes the array
number of sensor elements N , 2) separate non-Gaussian signal modelling, and the analysis of the received signal's
signals from additive white noise, and 3) provide higher-order statistic. Then, in Section III, the differential
robustness to modelling error. geometry framework is proposed to investigate shape and
Subsequently, a generalization of subspace-based DF properties of manifold associated with higher-order
methods such as 2q  MUSIC was proposed [7]. statistics.
The performance analysis will be presented in Section
The essential understanding come from the concept of
IV focusing initially on the choice of circular covariance
virtual-sensor array proposed in [8], where it draws the
matrix arrangement that will optimize the array
relationship between an underlying of array's higher- performance, before examining the array performance
order statistics and the second-order statistics in terms of enhancement in terms of the CRB, detection and
the virtual sensor array to show the virtual increase of the resolution thresholds for various orders of statistics.
array aperture as well as the number of array's virtual Finally, the paper is concluded in Section V.
sensors. The analysis offers important insights into the
mechanism of higher-order methods from the traditional
second-order statistic approaches. II. Problem Formulation
To assess performance of array processing methods, To proceed, the signal modelling is first presented in
either for second-order and higher-order statistics, it is this section. Then, the analysis on the use of higher-order
essential to evaluating through the characteristic/property statistic (HOS) of the received signal is examined.
of an array manifold. By definition, array manifold is the
locus of all array response vectors (manifold vectors)
over the set of signal parameters. For the case of a single II.1. Signal Modelling
parameter, the locus of these manifold vectors represents Consider an array system consisting of N sensor
a curve embedded in the complex space, while the locus
of two-parameter manifold vectors represents a surface elements located at r   r x , r y , r z    N 3 measured in
[9]. half wavelength, where r x, r y , and r z are column
The fact that the array manifold incorporates all
information regarding the array and signal environments vectors of the sensor position in a Cartesian coordinate
enables the manifold to completely characterize the system with the array centroid is located at the reference
system. Important array performance parameters such as point  0 , 0 , 0  .
the Cramer Rao lower bound (CRB), the array's detection The array is operated in the presence of M narrow-
and the array resolution capabilities (the ability of an band point sources propagated from the array far-field
array to correctly estimate the number of sources, and to with a plane wave propagation is assumed from azimuth-
distinctively obtain signal parameters) can be expressed elevation i ,i  ,i  1,...,M  with the azimuth
in terms of the manifold's intrinsic parameters.
Issues of array uncertainty analysis and array i   0,180  is measured anti-clockwise from the
ambiguity that is caused by linear dependence amongst positive x  axis, and the elevation i   0 ,90  is
source's manifold vectors [10]-[12] as well as the array
measured anti-clockwise from the x  y plane.
formation [13] are also related to the shape of array
manifold. Differential geometry has been used as a tool The baseband received signal x  t    N 1 is
to investigate the geometrical shape and characteristic of expressed as:
the manifold [14].
So far, the majority of the work has focused on the T
array performance analysis with respect to second-order x  t    x1  t  , x2  t  , , xN  t  

statistics, where system parameters are expressed in M (1)
terms of the array manifold's intrinsic parameters such as   a  i , i  mi  t   n  t 
i 1
the manifold arc length, rate-of-change of arclength, and
the manifold curvatures.
where mi  t  is the message signal of the i th source and
The framework was extended in [15] to analyze
performance enhancement when using the fourth-order T
the vector n  t    n1  t  , ,nN  t   denotes an additive
statistic. To the best of our knowledge, performance
analysis with respect to higher-order statistics has not white Gaussian noise with zero mean and variance  2 .
been investigated, and hence be an objective in this Furthermore, the vector a i ,i  denotes a manifold
paper. From the differential geometry perspective, the
vector corresponding to a plane wave impinging from the
manifold's geometrical shape shall be analyzed for
different orders of statistics to exploit the impacts of direction i ,i  and is given by:
these geometrical properties on the overall array
performance. An organization for the rest of the paper is a i ,i   exp   j r k i ,i     N 1 (2)
given as follows. In the next section, the problem
with:
43
Supawat Supakwong
k i ,i     cos i cos i ,sin i cos i ,sin i 

T where c2 q,mi denotes 2q  order circular autocumulant
of mi  t  ,  is Kronecker product and * corresponds to
denotes the wave number vector. Conventional array the complex conjugation. In addition, the vector
processing methods exploit the information contained in l l
1
a i ,i  N is defined as the Kronecker products
the second-order statistics of the received signal x  t  to
obtain parameters of interest either for detection, over l terms of the manifold vectors a i ,i  , i.e.:
parameter estimation, or reception, where the data
covariance matrix  xx is given by: a i ,i 
l
 a i ,i   a i ,i   a i ,i  (6)

 xx   x  t  x H  t   Note that the circular covariance matrix  2q,x  l  in
M (3) Eq. (5) can be arranged in different ways depending on
  c2,m a i ,i  a H i ,i    2  N
i an arbitrary integer l, 0  l  q, which has a direct impact
i 1
on the array performance. The study in [3] showed that
with c2 ,mi denoting the second-order autocumulant of the the number of signals that the array can process is
maximized when the integer l is given by:
message signal mi  t  ,  N is an N  dimensional
H
identity matrix, and  is the Hadamard operation q / 2,
l
if q even;
(7)
(conjugate-transpose). Subspace-based DF algorithms,  q  1 / 2, if q odd.
such as MUSIC, exploits the eigenstructure of this data
covariance matrix  xx in search for a set of Another advantage from processing signals using
corresponding manifold vectors that lies in the signal higher-order statistic is the robustness to the presence of
subspace (likewise, orthogonal to the noise subspace) Gaussian noise. By comparing the expression of the
with the cost function   ,  given by: 2q  order circular covariance matrix in Eq. (5) with the
expression of the second-order covariance matrix in Eq.
  ,   a H  ,   N  HN a  ,  (4) (3), one can draw a relationship between the term
a  , l  a  , *  q l   in Eq. (5) and the vector
 i i i i

where  N denotes a matrix whose columns contain the
a i ,i  in Eq. (3).
associated noise eigenvectors, obtained from the
eigendecomposition of the covariance matrix  xx . This indicates an important relationship between the
second-order and a higher-order statistic in a way that the
The use of second-order statistic in array signal
2q  order circular cumulant matrix can be thought of as
processing has several limitations. For instance, the
number of sources the methods can process cannot being the second-order covariance matrix of a virtual
exceed N  1 . In addition, the array performance is sensor array with an effective manifold vector ae i ,i 
degraded due to the presence of Gaussian noise, the given by:
modelling error, as well as its inability to deal with non-
Gaussian signals that practically employed in digital l *  q  l 
communications. ae i ,i   a i ,i   a i ,i  (8)
hence:
II.2. Higher-Order Statistics of the Received Signal
To overcome the limitations caused by the second- M
H
order statistics, array processing methods based on  2 q,x   c2q,m ae i ,i  ae i ,i 
i
(9)
2q  order statistics of the received signal, where q  1, i 1
exploit the information contained in the circular

To locate position of virtual sensors in the array, let's
covariance matrix  2q,x of the size N q  N q , whose   simplify the expression of the effective manifold vector
elements denote the 2q  order circular cumulants of the in Eq. (8).
data, given by: First, consider the Kronecker product between two
manifold vectors a i ,i  , i.e.:
M
l *  q  l  
 2 q,x  l    c2q,m a i ,i   a i ,i 

i 1
i
(5) 
a i ,i   a i ,i   exp  jrT k  exp  jrT k   
l *  q  l  
H (10)
 a i ,i 

 a i ,i    
 exp  j rT  rT  k
44
Supawat Supakwong
with the vector k  k i ,i  for the sake of simplicity eigenvalues corresponding to the signal subspace, and
second, a matrix of eigenvectors with zero eigenvalue
and the operator  denotes the Kroneckor addition.
corresponding to the noise subspace. The generalized
Consequently, the Kronecker products over l terms of 2q  MUSIC algorithm is based on a search for effective
a i ,i  can be simplified to: manifold vectors in Equation (8) that lie in the
corresponding signal subspace, or similarly orthogonal to
a i ,i 
l
 
 exp  j rT  rT  rT  k the noise subspace.

 exp  j rT  
l
k  (11) III. Differential Geometry of an
Array Manifold
T l Array manifold represents a locus of all manifold
where r   rT  rT  rT . Furthermore, the
vectors over the set of signal parameters. It can be
Kronecker product between two conjugated manifold viewed as a geometrical object embedded in a high-
vectors a* i ,i  gives: dimensional complex space with the dimensionality is
defined by the total number of sensors in an array. In this
section, differential geometry will be used as a tool to

a* i ,i   a* i ,i   exp jrT k  exp jrT k    analyze the shape and characteristic of the manifold.
Understanding the geometrical properties of an array
(12)

 exp j rT  rT  k manifold is crucially important as it is directly related to
the array processing performance in all aspects ranging
from array detection, parameter estimation, to reception.
Hence, the Kronecker products over  q  l  terms of To proceed, we will first review the basic
the conjugated manifold vectors is simplified to: characteristics of array manifold with respect to the
second-order statistics before later extend the framework
 q  l  to analyze the array manifold due to higher-order
a* i ,i  
 exp j rT  rT  rT  k statistics.
 q l 
(13)
 

 
 exp  j rT k
 III.1. Characterizing Array Manifold Curve
For the simplicity in our analysis, let's assume that an
Therefore, we can derive the expression of ae i ,i  N  element sensor array under consideration has a
in Eq. (8) to: linear configuration (i.e. r y  r z  0 ) and all sources are
impinging from the same elevation, assumed   0 .
ae i ,i   exp  j r  T l
  k  Under these assumptions, the expression of the array
response vector in Eq. (2) can now be expressed as a
  q l   single-parameter function of an azimuth direction  :

 
exp  j rT k

(14)
a    exp   j rx cos   ,   N1 (16)
 exp  j  r  k 
e
T
where the field-of-view  is restricted to:

where the matrix re can be found from:
T l T  q l 

   :   0 ,180 
re  r    r   (15)
The locus of these manifold vectors a   ,  

with the operators  represent the Kronecker
forms a manifold curve embedded in an N  dimensional
subtraction. From this finding, it shows that the 2q 
complex space, i.e.:
order statistic of the sensor array with respect to the
position r is equivalent to the second-order covariance
matrix of the virtual sensor array with the position re . 
A  a     N 1 ,  :   0 ,180   (17)
Consequently, we can generalize the standard MUSIC
algorithm to estimate directions-of-arrival for the higher- An illustration of the curve is shown in Fig. 1.
order statistics by taking an eigendecomposition of the It should be noted that these assumptions are made
cumulant matrix in Eq. (5) to find two sets of mainly to aid the simplification of our analysis since the
eigenvectors; first a matrix of eigenvectors with non-zero characteristics of a curve is easier to comprehend.
45
Supawat Supakwong
1  da
s  s 1     s 1    d (19)
1 d
with the rate-of-change of arclength s   is given by:
ds da
s     (20)
d d
where, in general, the dot operator is used to denote the

differentiation with respect to the angular parameter  .
Since this curve is embedded in an N  dimensional
complex space, an orthonormal system of coordinates
ui  s  , together with a set of curvatures  i  s  for
i  1, 2 , , is required to attach along each point s on
the curve, as shown in Fig. 1. The real skew-symmetric
Cartan matrix   s  whose entries contain different
Fig. 1. An illustration of array manifold curve embedded
in an $N$ dimensional complex space orders of curvatures 1  s  , , d 1  s  is given in the
form:
The framework can be generalized to consider an
arbitrary array configuration with sources impinging
from any direction in space. Because the locus of  0 , 1  s  , 0,  0 
 
manifold vectors in this generalized case is a function of 
 1  s  , 0 ,  2  s  ,  0 
both azimuth  and elevation  , hence the  0,  2  s  , 0,  0 
corresponding array manifold is viewed as a surface s    (21)
embedded in an N  dimensional complex space.       
 0, 0, 0,   d 1  s  
Detailed analysis of manifold surface can be found in  
[9]. Focusing on a single-parameter array manifold curve  0 , 0, 0,  0 
with respect to the locus of all manifold vectors in Eq.
(16), as shown in Fig. 1, it is simple to see that the
where the parameter  i  s  for i  1, ,d denotes the set
tangent vector with respect to the azimuth  exists at all
points on the manifold. This regularity condition ensures of non-zero curvatures. The first curvature (known as the
that this Riemannian manifold has a smooth transition principal curvature) at s is defined as
from point-to-point throughout the manifold, hence the d 2as
differential geometry is a proper tool for analysis. 1  s   a''  s   , where a ''prime'' is used to
ds 2
To characterize the shape of manifold curve, the
following parameters are considered essential (a) the denote a differentiation with respect to s. This parameter
arclength s   , (b) the rate-of-change of arclength essentially measures how fast the curve gets pulled away
from the tangent line, and is known to be the most
s   and (c) the set of curvatures that uniquely defines a important one. The other curvatures  i  s  can be found
space curve  i  s  for i  1, 2 , , which forms the iteratively in terms of the coordinate vector ui  s  and a
Cartan matrix   s  . lower-order curvature  i 1  s  .
An arc length of manifold curve represents the actual The analysis in [9] showed that the curvature is
physical length of a segment on the curve in  N . It is constant and identical at any point on the curve implying
typically defined as the segment from the origin to point that it has a hyperhelical shape, embedded in an N -
 on the manifold. That is: dimensional complex space. For further detailed analysis
of the geometrical shape of the array manifold curve such
 da as the concept of curve's inclination, please refer to [9].
s    0 d (18)
d
III.2. Differential Geometry of an Array Manifold
Consequently, the arc length separation s between associated with Higher-Order Statistics
two points on the curve for directions 1 and 1   can
Let's now extend the differential geometry framework
be computed from the difference of the two arc length, to analyze the geometrical shape of the array manifold
i.e.: curve associated with the higher-order statistics.
46
Supawat Supakwong
According to the expression of the effective manifold in the similar manner. For instance, the principle
vector corresponding to the higher-order statistics of the curvature of the effective manifold curve can be found
received signal in Eq. (8), the locus of these effective from:
manifold vectors ae   forms the corresponding curve
embedded in an NVS  dimensional complex space, i.e.: d 2 ae  s 
 i , e  s   ae ''  s   (25)
ds 2

Ae  ae     NVS 1 , :   0 ,180  (22)

Given that ae  s   exp j r x,VS s   r x,VS  v  from
where NVS represents the total number of virtual sensors [9], thus:
listed in Eq. (15). Several points can be addressed
regarding this effective manifold curve. First, we should d d 
draw the fundamental difference between the effective
 i,e  s  
ds  ds

 exp j  rx,VS s   rx,VS  v  


curve and the manifold curve corresponding to the
second order statistic.   jrx,VS    jrx,VS   exp  j  rx,VS s   rx,VS  v 
This effective manifold curve represents the locus of
 r x,VS
2
manifold vectors corresponding to the virtual sensor
array with the position listed in Eq. (15), which is
different from the actual sensor array position. Regardless of the order 2q , the curvature  i,e  s  of
The number of virtual sensors NVS is always greater
the corresponding effective manifold curve Ae remains
than the number of actual sensors N implying that the
constant. This shows that the 2q  order effective
effective manifold curve is embedded in the higher-
dimensional complex space. The dimensionality of the manifold curves all have hyperhelical shape embedded in
space depends on the array configuration as well as the an NVS  dimensional complex space.
order of statistic 2q. On the other hand, the two curves
share some relationships, which can be related using the
concept of virtual sensor array. That is, to analyze an
IV. Performance Analysis
array manifold curve corresponding to a higher-order The performance enhancement of subspace-based
statistic of the sensor array r x , it is equivalent to methods when using higher-order statistics shall be
studying the second-order manifold curve with respect to investigated in this section.
the virtual sensor array r x,VS . Consequently, the For the purpose of comparison, three performance
indicators will be used. The first is the popular Cramer
presented framework can be used to assess the Rao lower bound (CRB) that expresses a lower bound on
geometrical shape of the effective manifold curve. First, the error covariance matrix of an angular estimate ˆ ,
the effective manifold's arc length can be found from:
which can be expressed in terms of the curve's
geometrical property, in case of a single emitter [9], as:
 d ae
se    0 d
d 1
CRB    2
(26)
 d
exp   j rx,VS cos   d 2  SNR  L  s  
 0 d
 The bound is shown to be inversely proportional to the
 0 j rx,VS sin   exp   j rx,VS cos   d (23)
signal-to-noise ratio SNR, the number of snapshots L,

as well as the square of the rate-of-change of manifold
 0  rx,VS sin  d
curve s   . Two other array performance parameters
  rx,VS 1  cos   are the array detection and array resolution threshold
accordingly. An array detection is referred to the ability
with the rate-of-change of arc length is given by: of an array system to uniquely estimate the presence of
sources in an environment [9], In general, the total
da number of sources can be obtained from the estimated
se      r x,VS sin  (24) dimensionality of the noise subspace using the property
d of the eigendecomposition associated with the received
signal's second order statistic. However, due to
Furthermore, the system of coordinates u i,e  s  uncertainties surrounding the channel conditions or
measurements, the dimensionality of the noise subspace
together with the set of curvatures  i,e  s  can be derived
can be difficult to estimate precisely.
47
Supawat Supakwong
An ultimate's array detection represents a It should be noted, however, that in [8] the choice of l
performance's lower bound of the system's ability to is chosen according to the maximization of the number of
determine the number of sources under a situation when virtual sensors.
two sources are closely located. Typically, the bound is Whereas for our finding here, the choice is based on
given in the form of a product between the signal-to- optimization of the array processing performance. Next,
noise ratio SNR1 and the number of snapshots L, consider an array of 2 sensors separated by half-
expressed as [9]: wavelength, i.e.:
T
2  1 1
1  P  rx    ,  (30)
 SNR1  L det thres  2 
1 1  (27)  2 2
2s  P2 
measured in half-wavelength. This illustrates a scenario
where P1 and P2 represent the signal powers for the two for a small sensor array employed in a limited space,
sources respectively. Note also that the detection such as in an aircraft or portable devices. Using the
capability is inversely proportional to the square of the second-order statistics, the array can process no more
arc length separation s, which denotes the separation than a single source.
This showed a serious drawback especially in the
of two sources in the form of arc length. multisource environment. However, with a higher-order
On the other hand, the ultimate's array resolution statistic, the 2q  subspace-based methods can be used to
capability is referred to a performance's lower bound of
the system to distinctively obtain signal parameters for process a larger number of sources. For the purpose of
signals located close together in space, which can be analysis, let's examine the characteristics of the manifold
expressed in terms of the manifold's differential curve corresponding to different orders of statistic. The
geometry as: 4th , 6th ,8th and 16th orders of statistic are chosen in this
example to compare with the conventional second-order

4 statistic.
2 P1 
 SNR1  L res thres  1  4  (28) This corresponds to the integer q being 2 ,3, 4 and 8
 1 P2 
s 4  ˆ 12    accordingly. Using Eq. (15) to obtain the virtual sensor
 N array position, the result is shown in Table I together
with the total number of virtual sensors and the virtual
It is seen that, in addition to the arc length separation, array aperture. Observe the increase in the total number
the resolution threshold also depends on the estimated of virtual sensors and the virtual array aperture when
first curvature of the manifold curve ̂1 at s. increasing the order of statistic.
To proceed, let's first examine the issue of the circular Table II summarizes the basic characteristics of each
covariance matrix  2q,x  l  arrangement and its curve including arc length, rate-of-change of arc length,
the total curve length, and the principle curvature for
performance on the array processing. It was shown in Eq. different order of statistic.
(5) that the matrix  2q,x  l  can be arranged in a number Note that these curves are embedded in a completely
different dimensional complex space depending on the
of ways depending on an arbitrary l, where 0  l  q.
number of virtual sensors for each order.
In this example, let's consider a linear array of three Several points can be addressed regarding the
sensors located at: geometrical property of the corresponding manifold
curves. First, the curvatures of all manifold curves of all
T
r x   0.5,0.1,0.4 (29) orders are constant at all points indicating that the higher-
order manifold curves also have the hyperhelical shape.
This is an important finding which will aid the
measured in half wavelength. Suppose the 8th  order analysis for a number of array processing applications
statistic method is employed  q  4  . Hence, the integer such as the analysis of manifold ambiguities caused by
linear dependence amongst manifold vectors.
l can be ranging from 0 to 4 . The corresponding virtual
Next, if we consider the manifold curve's length and
sensor array can be obtained by Eq. (15). the rate-of-change of arc length, it is seen that as the
Using CRB as our performance indicator, the result is order of statistic increases, the lengths of curves are
shown in Fig. 2 illustrating the CRB as a function of increased from 4.44 to 8.89 ,14.05,19.87 , and 48.67
azimuth direction for different arrangement l . It is
observed that the array performance is maximized when respectively for the 4th , 6th ,8th and 16th order.
the l is equal to q / 2 (in case of q being an even And, since the curve length is longer, the
number). This result further confirms with the finding in corresponding rate-of-change of arc length se   and
[8] that the array processing is optimized when the the arc length separation  s between two directions are
integer l satisfies the condition in Eq. (7). increased as well.
48
Supawat Supakwong
Fig. 2. The CRB as a function of azimuth angle for 8th order of statistic, plotted for different arrangement l
TABLE I
VIRTUAL SENSOR ARRAYS CORRESPONDING TO DIFFERENT ORDER OF STATISTIC 2q , THE TOTAL NUMBER
OF VIRTUAL SENSOR ELEMENTS NVS , AND THE ARRAY APERTURES
Order ( 2q ) Position r x ,e N vs Aperture lm ,e

T
2  1 , 1  2 1
 2 2 
4  1, 0 ,1T 3 2
T
6  3 ,  1 , 1 , 3  4 3
 2 2 2 2 
8  2 , 1, 0 ,1, 2T 5 4
16  4 , 3, 2 , 1, 0 ,1, 2 , 3, 4 T 9 8
TABLE II
IMPORTANT PARAMETERS IN CHARACTERIZING ARRAY MANIFOLD CURVES OF DIFFERENT ORDERS OF STATISTIC 2q
Order Arclength Rate-of-change Total curve length The principle curvature
2q se    of arclength se   lm i  s 
2 0.71 1  cos   0.71 sin  4.44 1.41
4 1.41 1  cos   1.41 sin  8.89 0.71
6 2.24 1  cos   2.24 sin  14.05 0.45
8 3.16 1  cos   3.16 sin  19.87 0.32
16 7.75 1  cos   7.75 sin  48.67 0.13
On the other hand, as the order of statistics increases, when the order of statistic increases. Given that the CRB
the fundamental curvature 1  s  that essentially is inversely proportional to the square of the rate-of-
measures how fast the curve's angle is pulled away from change of manifold curve s   , hence the bound is
the tangent line decreases. lowered as a result, as shown in Fig. 3(a).
This is as expected because with the higher-order In addition, the lower bounds at the array's broadside
statistics, the curve is longer in length, hence the rate-of-   90  are lower than at the array's endfire
change that the curve gets pulled away will decrease.
Analysis of the performance enhancement using (  0 ,180 ) as already expected since it is easier to
higher-order statistics is now presented. As the order of process signal at broadside rather than at the array's
statistic increases, the lower bounds is lowered indicating endfire. Moreover, Figs. 3(b) and 3(c) show the array
the improvement on angular parameter estimation. performance enhancement in terms of the array's ultimate
Looking from the differential geometry perspective, this direction and resolution capabilities respectively.
improvement can be explained by considering the rate- The presented results are based on the analysis of the
of-change of arc length listed in Table II that is increased array detection and resolution in the presence of two
49
Supawat Supakwong
closely-located sources with an angle separation of Meanwhile, Fig. 4 shows the minimum SNR required
  5. Both sources have equi-power, i.e. P1  P2  1. in order to detect or resolve two sources with an angular
Both indicators show similar patterns as in the case of separation of  . The first source is impinging from the
CRB. This shows that, when using higher-order statistic, direction  , which is chosen to be at   45 in this
the performance in terms of detection and resolution are example, and the second source is from the direction
also enhanced. In addition, note that the detection and    . The number of snapshots is assumed to be
resolution thresholds with respect to the second-order L  10. It is seen that detecting the presence of sources is
statistics  q  1 cannot be found. This is due to the fact easier than resolving sources.
that with the conventional second-order statistics an Furthermore, detecting/resolving sources that are
sensor array of two sensors cannot process more than a closely located (small  ) requires a higher SNR as
single signal. compared to a scenario when sources are wider apart.
(a)
(b)
(c)
Figs. 3. The array performance as a function of azimuth angle, plotted for different orders of statistic (a) CRB (b) The detection's lower bound
in terms of  SNR1  L  (c) The resolution's lower bound in terms of  SNR1  L 
50
Supawat Supakwong
Fig. 4. The detection's and resolution's lower bounds in terms of SNR1 as a function of azimuth angle separation  ,
plotted for different orders of statistic. L  10 ,   45
V. Conclusion [11] Supakwong, S., Further investigations on the causes of manifold

ambiguities with applications in an array formation, (2010)
In this paper, an array performance enhancement International Review on Modelling and Simulations (IREMOS), 3
(4), pp. 461-468.
when using higher-order statistic were assessed from the [12] A. Manikas, C. Proukakis, and V. Lefkaditis, Investigative study
differential geometry perspective. of planar array ambiguities based on “hyperhelical”
Since array performance is fundamentally related to parameterization, IEEE Transactions on Signal Processing,
the geometrical shape and properties of the array 47(6):1532–1541, 1999.
[13] Supakwong, S., A novel selection criterion for distributed vector
manifold, the differential geometry was employed as a sensors to optimize the array's ultimate detection capability,
tool to analyze the properties of manifold curve (2010) International Review on Modelling and Simulations
corresponding to the higher-order statistic using the (IREMOS), 3 (2), pp. 206-211.
concept of virtual sensor array. [14] T. Willmore, An Introduction to Differential Geometry, Oxford
University Press, UK, 1959.
Simulation results showed array performance [15] N. Takai and A Manikas, Array manifold properties and
enhancements in terms of the CRB, as well as the array performance of higher-order signal subspace techniques, In
detection and resolution when higher-order statistic was Acoustics, Speech, and Signal Processing, 1993. ICASSP-93.,
employed. 1993 IEEE International Conference on, volume 5, pages 666–
669 vol.5, April 1993.
References Authors’ information

[1] R. Schmidt, Multiple emitter location and signal parameter
estimation, IEEE Transactions on Antennas and Propagation Supawat Supakwong is currently a lecturer at
[legacy, pre - 1988], 34(3):276–280, 1986. the Department of Electrical and Computer
[2] R. Roy and T. Kailath, ESPRIT-estimation of signal parameters Engineering, Thammasat University, Thailand.
via rotational invariance techniques,Acoustics, Speech and Signal He received his Bachelor (Distinction) and
Processing, IEEE Transactions on,37(7):984–995,Jul 1989. Master degrees in 2004 and 2005 respectively
[3] P. Chevalier and A Ferreol, On the virtual array concept for the from University of Virginia, USA, and his Ph.D.
fourth-order direction finding problem, Signal Processing, IEEE from Imperial College London, UK in 2009, all
Transactions on, 47(9):2592–2595, Sep 1999. in electrical engineering. At present, he is the
[4] B. Porat and Benjamin Friedlander, Direction finding algorithms director of Thammasat University’s Programmes of Innovative
based on high-order statistics, Signal Processing, IEEE Engineering (TU-PINE). His main research work is in the area of array
Transactions on, 39(9):2016–2024, Sep 1991. signal processing and communications.
[5] M.C. Dogan and J.M. Mendel, Applications of cumulants to array
processing (i) aperture extension and array calibration, Signal
Processing, IEEE Transactions on, 43(5):1200–1216, May 1995.
[6] M.C. Dogan and J.M. Mendel, Applications of cumulants to array
processing. (ii) non-Gaussian noise suppression. Signal
Processing, IEEE Transactions on, 43(7):1663–1676, Jul 1995.
[7] P. Chevalier, A Ferreol, and L. Albera, High-resolution direction
finding from higher order statistics: The 2q- MUSIC algorithm,
Signal Processing, IEEE Transactions on, 54(8):2986–2997, Aug
2006.
[8] P. Chevalier, L. Albera, A Ferreol, and P. Comon, On the virtual
array concept for higher order array processing, Signal
Processing, IEEE Transactions on, 53(4):1254–1271, April 2005.
[9] A. Manikas, Differential Geometry in Array Processing, Imperial
College Press, London, 2004.
[10] A. Manikas and C. Proukakis, Modeling and estimation of
ambiguities in linear arrays, IEEE Transactions on Signal
Processing, 46(8):2166–2179, 1998.
51
An Up-to-Date Technologies Review and Evaluation

of Wave Energy Converters
Hosna Titah-Benbouzid, Mohamed Benbouzid
Abstract – The potential of electric power generation from marine renewable energy is
enormous. Ocean waves are being recognized as a resource to be exploited for the sustainable
generation of electrical power. The high load factors resulting from the fluid properties and the
predictable resource characteristics make ocean waves particularly attractive for power
generation and advantageous when compared to other renewable energies. Regarding this
emerging and promising area of research, this paper presents a complete review of wave energy
technologies describing, analyzing and fixing many of the concepts behind wave energy
conversion. The proposed review will specifically highlights the main wave energy conversion
projects around the world at different levels (demonstration stage, in production, and
commercialized projects). In particular, mooring will be discussed, as it is a key feature behind
massive deployment of wave energy converters. Finally, a discussion will highlight challenges that
wave energy converters need to overcome to become commercially competitive in the global
energy market. Copyright © 2015 Praise Worthy Prize S.r.l. - All rights reserved.
Keywords: Marine Renewable Energy, Wave Energy Converter, Design, Challenges
Nomenclature fact what was later named as a (floating) oscillating

water column (Fig. 1(b)).
WEC Wave Energy Converter These buoys were commercialized in Japan since 1965
PTO Power Take-Off (and later in USA) [6]. Since then many different other
Pw_f Power per meter of wave front concepts have been conceived. Some of these have come
Pw_mcl Power per meter crest length no further than the drawing table, others have made it into
 Water density(approximately 1000 kg/m3) small-scale models, and a few have also moved on to
g Gravity acceleration ocean testing. The technology is still immature and would
A Wave amplitude not commercially exist if governments did not subsidize it.
T Wave period Therefore, to become a competitive market, it is crucial
for the industry to reduce the overall cost of electricity
generated from waves. There are many different WEC
I. Introduction technologies, and it is not clear which one is superior.
One of the very attractive renewable energy sources is WECs developers tend to focus on the prime-mover aspect
the ocean. Indeed, it covers around three quarters of the and use off-the-shelf electrical systems to generate
earth surface and energy can be extracted from the electrical power. These electrical systems usually include a
waves, tides, currents, temperature gradients, and salinity gearbox or a hydraulic system to interface a slow moving
gradients. Wave energy, in particular, is spatially more prime mover to a conventional high-speed rotary machine.
concentrated than both wind and solar energy; it is also The use of gearboxes or hydraulics introduces potential
more persistent and predictable than wind energy. extra-scheduled and unscheduled maintenance costs.
The global wave power resource has been estimated to Moreover, the maintenance for offshore devices is much
be at least 1 TW, with a potential annual energy more expensive than onshore equivalents and limited by
production of about 2000 TWh; this is comparable to the weather conditions, which results in increased downtime
energy production from nuclear or hydropower [1]-[4]. costs.
The history of wave power research spans over more The present review aims at giving an update of the
than two hundred years. The Frenchman Pierre-Simon most recent trends regarding main wave energy
Girard is recognized as the first holder of a wave power conversion projects around the around at different levels
patent in 1799 [5] (Fig. 1(a)). (demonstration stage, in production, and commercialized
Yoshio Masuda may be regarded as the father of projects) with respect to overviews already published in
modern WEC technology, with studies in Japan since the the past years [6]-[13]. In particular, mooring will be
1940s. He developed a navigation buoy powered by discussed, as it is a key feature behind massive
wave energy, equipped with an air turbine, which was in deployment of wave energy converters.
52
H. Titah-Benbouzid, M. E. H. Benbouzid
II. Wave Energy Background

Fig. 2 show an atlas of the global power density
distribution of the oceans. The north and south
temperature zones have the best sites for capturing wave
power. The prevailing winds in these zones blow
strongest in winter. Increased wave activity is found
between the latitudes of 30° and 60° on both
hemispheres, induced by the prevailing western winds
blowing in these regions.
A wave resource is typically described in terms of
power per meter of wave front (wave crest length) [8]:
(a) Pierre-Simon Girard WEC patent 1

Pw _ f   g 2 A 2T (1)
8
It can also be described in terms of wave power per

meter crest length (Pw_mcl).
1
Pw _ mcl   g 2 H 2T (2)
32
It should be noted that the wave height H is defined as

equal to 2A (Fig. 3).
III. Wave Energy Converters

III.1. WEC Concepts
WECs have been developed to extract energy from
shoreline out to the deeper waters offshore. These
(b) Yoshio Masuda oscillating water column devices are generally categorized by the installation
location and the Power Take-Off (PTO) system.
Figs. 1. WEC history review
Locations are shoreline, near shore and offshore (Fig. 4).
Finally, a discussion will highlight challenges that
wave energy converters need to overcome to become
commercially competitive in the global energy market.
Fig. 2. Global annual mean wave power estimation in kW/m spanning 10 years period [10]
53
In this particular huge developing context, it should be

noted a new French WEC project called EM Bilboquet
[14]-[15]. The PTO extracts the mechanical power due to
incoming waves by a system made up of a cylindrical
buoy sliding along a partially submerged structure (Fig.
7). This structure is made up of a vertical cylinder,
1m referenced in the following as spar, with a damping plate
attached at its keel. Energy resulting from the relative
motion between the two concentric bodies is harnessed
by rack-and-pinion, which drives a permanent magnet
Fig. 3. Wave dimensions synchronous generator through a gearbox [16].
In this context, most devices can be characterized as
belonging to six types: Attenuator; Point absorber; IV. Wave Energy Extraction
Oscillating wave surge converter; Oscillating water
column; Overtopping device; Submerged pressure Figure 8 summarizes the different conversion stages.
differential (Fig. 5). In particular this figure shows that there is a variety of
ways to extract power from waves: pneumatically,
hydraulically, and mechanically (PTO) [17].
III.2. WEC Main Projects This mechanical interface is used to convert the slow
Fig. 6 summarizes the main WEC projects in terms of rotational speed or reciprocating motion into high-speed
concepts and locations. It should be mentioned that this rotational motion for connection to a conventional rotary
figure tries to summarize the main and well-known WEC electrical generator. In this context, attention will be
mainly over the demonstration stage. Indeed, there is a directed at the mechanism needed to convert wave
large number and variety of WEC that vary in concept energy into electricity as most building blocks in the
and design. In addition to the fact, that there were more generation system remain nearly the same after being
than 1000 patents in 2009 [10]. In fact, all of these transformed into the electrical form [18].
projects should be considered as in early stages if
compared to other renewable technologies (i.e. wind).
Offshore Nearshore Onshore
Fig. 4. WECs locations
ATTENUATOR POINT ABSORBER PRESSURE DIFFERENTIAL
OSCILLATING OSCILLATING
OVERTOPPING
WAVE SURGE WATER COLUMN
Fig. 5. WECs concepts
54
Fig. 6. WECs main projects
efficiency [29]-[32]. In this context, there are still

mechanical engineering challenges in terms of electrical
generator offshore suitability [10], [33].
Table I summarizes the PTO systems and the
electrical generator options for the some of Fig. 6 WEC
projects.
V. Wave Energy Converter Mooring

To use wave energy for electricity generation, WECs
must be anchored to the seabed and moored by cables
(Fig. 9). Similar to other offshore structures moored on
the sea floor, a typical WEC mooring system is likely to
be composed of three parts: the mooring line, the
Fig. 7. The EM Bilboquet French wave energy converter connectors and the anchor. Chain, wire rope and
synthetic fiber rope are the three main mooring line types
Linear generators are an option on the testing stage, that are used in offshore structures and could be used for
but they are not yet currently used in most developed WECs [38]-[40]. Chains provide good catenary stiffness
WECs [19-24]. In particular, different types of linear and are abrasion resistant. However, their restraining
generators were investigated for the AWS WECs [25- stiffness may not be appropriate for some WECs.
27]. These investigations have led to the conclusion that They can hamper the oscillation motion required to
the transverse flux permanent magnet generator is a good convert energy.
candidate in terms of higher power density and Synthetic ropes are advantageous because of their
efficiency. The use of permanent magnet synchronous buoyancy property, which will reduce mooring weight
generator is an intermediate option [16], [28]. The use of influence during normal operation and are good
induction generators implies a specific mechanical PTO candidates for deep-water applications [41].
that induces additional losses affecting the WEC overall
55
Fig. 8. WEC different type of conversions
56
Mooring Anchor
Fig. 9. Wave energy converters mooring and anchor
TABLE I Single point mooring allows a WEC to weather-vane

SOME WEC PROJECTS PTOS AND GENERATORS
[43]-[44]. There are several sub-types as listed in Table II
WEC PTO Generator
PELAMIS Attenuator/Hydraulics Cage induction
and it is difficult to define which one is the best without
generator considering the WEC type, location, safety, and cost
POWERBUOY Point absorber Permanent magnet [45]-[47].
synchronous However, it seems that Catenary Anchor Leg Mooring
generator
WAVESTAR [34] Point absorber/ Hydraulics Permanent magnet
(CALM) in spread mooring, and Single Anchor Leg
synchronous Mooring (SALM) in single point mooring are more
generator popular in practical projects [48]-[49]. Figs. 10 show
SEAREV [5], [35] Point absorber Permanent magnet some commonly used mooring configurations.
synchronous
generator
TABLE II
BILBOQUET Point absorber Permanent magnet
MOORING TYPES
synchronous
generator Type Spread Single point
OYSTER Oscillating wave surge Cage induction Sub-types Catenary mooring CALM
converter generator Taut mooring (Catenary Anchor Leg Mooring)
LANGLEE [36] Oscillating wave surge Permanent magnet Turret mooring SALM
converter synchronous (Single Anchor Leg Mooring)
generator ALC
LIMPET Oscillating water column & Cage induction (Articulated Loading Column)
Wells turbine generator SPAR
OCEANLINX Oscillating water column & Cage induction (Single Point Mooring & Reservoir)
Denniss-Auld turbine generator Fixed tower mooring
PICO [8] Oscillating water column & Doubly-fed
Wells turbine induction generator
WAVE DRAGON Overtopping & Kaplan Permanent magnet V.1. Mooring Requirements
turbine synchronous
generator The mooring could not be considered as an additional
SSG [37] Overtopping Permanent magnet cost item in the overall economics of a WEC. It should
synchronous
generator
be designed as an integral element of the overall system
AWS Direct drive Linear permanent that contributes to power extraction efficiency [50]-[52].
magnet generator In this context, the following list shows the main
requirements that need to be considered for WEC
There are many kinds of connectors used on WECs moorings systems (a detailed list could be found in [41]).
and other marine structures [38], [42]. Anchors are the – Mooring stiffness is an active element in the wave
terminals that transfer the whole system forces to the energy conversion principle used. The mooring
seabed. The two major requirements for a WEC mooring system should be sufficiently stiff to:
are to withstand the environmental and other loadings  Allow berthing for inspection and maintenance;
involved in keeping the device on station, and to be  Station keeping within specified tolerances;
sufficiently cost-effective so that the overall device  Maintain clearance distances between mooring;
economics remain viable. In particular the mooring  Avoid constraints in lines and power cable in
system is subject to highly cyclic, nonlinear load every tide conditions.
conditions, mainly induced by the incident waves. – It should be sufficiently compliant to the
Mooring systems, which may be suitable for WECs, environmental loading to reduce the forces acting on
can be categorized into two main configurations: spread anchors, mooring lines and the device itself to a
mooring and single point mooring. Spread mooring minimum.
restricts a WEC motion in the horizontal plane and hence – It should be sufficient to accommodate the tidal range
will not allow it to weather-vane. This type of mooring at the installation location.
may be appropriate for non-directional energy
converters.
57
Figs. 10. Typical mooring systems for wave energy converters:

(a) Catenary line; (b) Taut line; (c) Taut line with mid-column float; (d) Taut line with weights, (e) Taut line with weights and floats
It is therefore obvious that mooring design is a critical at first, small machines where developed first, and were
part of a WEC project. subsequently scaled-up to larger sizes and powers for
The devices are generally thought to be used in areas massive deployment.
of demanding environmental loads due to waves, current In fact, optimal wave energy absorption involves some
and wind. These survivability issues are addressed in kind of resonance. This implies that WECs geometry and
existing offshore standards, such as the DNV-OSE301 size are linked to wavelength.
[53]. So, if pilot plants are to be tested in the open ocean,
they must be large structures [6]. In this specific context,
challenges that WECs should to overcome to become
VI. Challenges for Commercial Viability commercially competitive leading to massive
It has been proven that wave energy extraction is very deployment could be summarized as:
attractive as it is spatially more concentrated than both – As for offshore converters, WECs should withstand
wind and solar energy; it is also more persistent and extreme wave condition leading to difficult and costly
predictable than wind energy. maintenance operations.
On the other hand, the development, from concept to – As above discussed, mooring design is a critical part.
commercial stage, has been found to be a very slow and In addition to the demanding environmental loads due
expensive process [11], [54]. Indeed, it is difficult to to waves, current and wind, the mooring system
follow what was done in the wind turbine industry where should also withstand constraints due to the WEC
alignment for capture optimization. Given the
58
continuous environmental loading, fatigue has been Proceedings of the 2014 IEEE PEAC, Shanghai (China), pp. 338-
342, November 2014.
identified as one of the key engineering challenges
[8] M.S. Lagoun, A. Benalia and M.E.H. Benbouzid, “Ocean wave
[55]. In addition, marine growth and corrosion need converters: State of the art and current status,” in Proceedings of
to be considered [56]. the 2010 IEEE ENERGYCON, Manama (Bahraïn), pp. 636-642,
– Higher costs of construction, deployment, and December 2010.
[9] Y. Hong, R. Waters, C. Boström, M. Eriksson, J. Engström and
maintenance need to be supported with substantial M. Leijon, “Review on electrical control strategies for wave
financial support from governments. energy converting systems,” Renewable and Sustainable Energy
Regarding mooring fatigue and cost issues, it has been Reviews, vol. 31, pp. 329-342, March 2014.
recently suggested to develop WECs without mooring [10] I. Lopez, J. Andreu, S. Ceballos, I. Martínez de Alegria and I.
Kortabarria, “Review of wave energy technologies and the
[57]-[58].
necessary power-equipment,” Renewable and Sustainable Energy
The developed converter has station-keeping ability Reviews, vol. 27, pp. 413-434, November 2013.
and evasive maneuver by diving. For station-keeping [11] B. Czech and P. Bauer, “Wave energy converter concepts: Design
within a uniform bound, a wave glider has been adopted challenges and classification,” IEEE Industrial Electronics
Magazine, vol. 6, n°2, pp. 4-16, June 2012.
as propulsion system [59] (Fig. 11).
[12] B. Drew, A.R. Plummer and M.N. Sahinkaya, “A review of wave
In addition, such kind of system has the ability to energy converter technology,” Proc. IMechE, Part A: Journal
submerge to a certain depth for its safety in emergency Power and Energy, vol. 223, n°8, pp. 887-902, December 2009.
condition such as typhoon [60]. [13] J. Falness, “A review of wave-energy extraction,” Marine
Structures, vol. 20, n°4, pp. 185-201, October 2007.
[14] S. Olaya, J.M. Bourgeot and M.E.H. Benbouzid, “Hydrodynamic
coefficient computation for a partially submerged wave energy
converter,” IEEE Journal of Oceanic Engineering, vol. PP, n°99,
pp. 1-14, 2014.
[15] S. Olaya, J.M. Bourgeot and M.E.H. Benbouzid, “Optimal control
for a self-reacting point absorber: A one-body equivalent model
approach,” in Proceedings of the 2014 IEEE PEAC, Shanghai
(China), pp. 332-337, November 2014.
[16] S. Olaya, J.M. Bourgeot and M.E.H. Benbouzid, “Modelling and
preliminary studies for a self-reacting point absorber WEC,” in
Proceedings of the 2014 IEEE ICGE, Sfax (Tunisia), pp. 14-19,
March 2014.
[17] K. Rhinefrank et al., “Comparison of direct-drive power takeoff
Fig. 11. A mooring-less wave energy converter concept [57] systems for ocean wave energy applications,” IEEE Journal of
Oceanic Engineering, vol. 37, n°1, pp. 35-44, January 2012.
[18] T.K.A. Brekken, H.M. Hapke, C. Stillinger and J. Prudell,
“Machines and drives comparison for low-power renewable
VII. Conclusion energy and oscillating applications,” IEEE Trans. Energy
This paper has proposed an up-to-date review of the Conversion, vol. 25, n°4, pp. 1162-1170, December 2010.
[19] L. Cappelli, F. Marignetti, G. Mattiazzo, E. Giorcelli, G. Bracco,
most recent trends regarding main wave energy converter S. Carbone, C. Attaianese, “Linear tubular permanent-magnet
technologies with respect to overviews already published generators for the inertial sea wave energy converter,” IEEE
in the past years. In addition, mooring has been discussed Trans. Industry Applications, vol. 50, n°3, pp. 1817-1828,
and has been shown to be a key feature behind massive May/June 2014.
[20] C. Boström, B. Ekergard, R. Waters, M. Eriksson and M. Leijon,
deployment of wave energy converters. Finally it has “Linear generator connected to a resonance-rectifier circuit,”
been highlighted some challenges that needs to be IEEE Journal of Oceanic Engineering, vol. 38, n°2, pp. 255-262,
overcome to enlarge the vision of large-scale commercial April 2013.
arrays of wave energy converters. [21] T.K.A. Brekken, K. Rhinefrank, A. von Jouanne, A. Schacher, J.
Prudell and E. Hammagren, “Scaled development of a novel wave
energy converter including numerical analysis and high-resolution
tank testing,” Proc. IEEE, vol. 101, n°4, pp. 866-875, April 2013.
References [22] R. Vermaak and M.J. Kamper, “Design aspects of a novel
topology air-cored permanent magnet linear generator for direct
[1] J. Crus Ed., Ocean Wave Energy: Current Status and Future drive wave energy converters,” IEEE Trans. Industrial
Perspectives. Springer, Berlin, 2008. Electronics, vol. 59, n°5, pp. 2104-2115, May 2012.
[2] J.G. Vining and A. Muetze, “Economic factors and incentives for [23] N. Hodgins, O. Keysan, A.S. McDonald and M.A. Mueller,
ocean wave energy conversion,” IEEE Trans. Industry “Design and testing of a linear generator for wave-energy
Applications, vol. 45, n°2, pp. 547-554, March-April 2009. applications,” IEEE Trans. Industrial Electronics, vol. 59, n°5,
[3] Z. Zhou, F. Scuiller, J.F. Charpentier, M.E.H. Benbouzid and T. pp. 2094-2103, May 2012.
Tang, “An up-to-date review of large marine tidal current turbine [24] J. Zhang, H. Yu, Q. Chen, M. Hu, L. Huang and Q. Liu, “Design
technologies,” in Proceedings of the 2014 IEEE PEAC, Shanghai and experimental analysis of AC linear generator with Halbach
(China), pp. 448-484, November 2014. PM arrays for direct-drive wave energy conversion,” IEEE Trans.
[4] S. Benelghali, M.E.H. Benbouzid and J.F. Charpentier, “Marine Applied Superconductivity, vol. 24, n°3, #0502704, June 2014.
tidal current electric power generation technology: State of the art [25] R. Ekström, B. Ekergård and M. Leijon, “Electrical damping of
and current status,” in Proceedings of the 2007 IEEE IEMDC, linear generators for wave energy converters—A review,”
Antalya (Turkey), vol. 2, pp. 1407-1412, May 2007. Renewable and Sustainable Energy Reviews, vol. 42, pp. 116-128,
[5] D. Ross, Power from the Waves. Oxford University Press: 1995. February 2015.
[6] F.A.O. Falcao, “Wave energy utilization: A review of the [26] H. Polinder, B.C. Mecrow, A.G. Jack, P.G. Dickinson and M.A.
technologies,” Renewable and Sustainable Energy Reviews, vol. Mueller, “Conventional and TFPM linear generators for direct-
14, n°3, pp. 899–918, April 2010. drive wave energy conversion,” IEEE Trans. Energy Conversion,
[7] H. Titah-Benbouzid and M.E.H. Benbouzid, “Ocean wave energy vol. 20, no. 2, pp. 260-267, June 2005.
extraction: Up-to-date technologies review and evaluation,” in
59
[27] M.A. Mueller, “Electrical generators for direct drive wave energy floating WEC devices,” Ocean Engineering, vol. 34,n°14-15, pp.
converters,” IEE Proc. Generation, Transmission & Distribution, 1918-1934, 2007.
vol. 149, n°4, pp. 446-456, July 2002. [47] Z. Gao and T. Moan, “Mooring system analysis of multiple wave
[28] N. Muller, S. Kouro, J. Glaria and M. Malinowski, “Medium- energy converters in a farm configuration,” in Proceedings of the
voltage power converter interface for WaveDragon wave energy 2008 EWTEC, Uppsala (Sweden), pp. 509-518, September 2009.
conversion system,” in Proceedings of the IEEE ECCE, Denver [48] A. Pecher, A. Foglia and J.P. Kofoed, “Comparison and sensitivity
(USA), pp. 352-358, September 2013. investigations of a CALM and SALM type mooring system for
[29] M.S. Lagoun, A. Benalia, and M.E.H. Benbouzid, “A predictive wave energy converters,” Journal of Marine Science and
power control of doubly fed induction generator for wave energy Engineering, vol. 2, pp. 93-122, 2014.
Converter in irregular waves,” in Proceedings of the 2014 IEEE [49] H. Ming, and G. A. Aggidis, “Developments, expectations of
ICGE, Sfax (Tunisia), pp. 26-31, March 2014. wave energy converters and mooring anchors in the UK,” Journal
[30] M.S. Lagoun, S. Nadjem, A. Benalia and M.E.H. Benbouzid, of Ocean University of China, vol. 7, n°1, pp. 10-16, February
“Predictive power control of doubly-fed induction generator for 2008.
wave energy converters,” in Proceedings of the 2012 EFEA [50] F. Cerveira, N. Fonseca and R. Pascoal, “Mooring system
(International Symposium on Environment Friendly Energies and influence on the efficiency of wave energy converters,”
Applications), Newcastle upon Tyne (UK), pp. 312-317, June 2012. International Journal of Marine Energy, vol. 3-4, pp. 65-81,
[31] M. Alberdi, M. Amundarain, A.J. Garrido and I. Garrido, “Neural December 2013.
control for voltage dips ride-through of oscillating water column- [51] B. Zanuttigh, E. Angelelli and J.P. Kofoed, “Effects of mooring
based wave energy converter equipped with doubly-fed induction systems on the performance of a wave activated body energy
generator,” Renewable Energy, vol. 48, pp. 16-26, 2012. converter,” Renewable Energy, vol. 57, pp. 422-431, September
[32] M. Alberdi, M. Amundarain, A.J. Garrido, I. Garrido and F.J. 2013.
Maseda, “Fault-ride-through capability of oscillating-water- [52] W. Valentine and K.D. von Ellenrieder, “Model scaling of ocean
column-based wave-power-generation plants equipped with hydrokinetic renewable energy systems,” IEEE Journal of
doubly fed induction generator and airflow control,” IEEE Trans. Oceanic Engineering, vol. 1, n°1, pp. 27-36, January 2015.
Industrial Electronics, vol. 58, n°5, pp. 1501-1517, May 2011. [53] Det Norske Veritas offshore standard DNV-OS-E301, Position
[33] S. Benelghali, M.E.H. Benbouzid and J.F. Charpentier, “Generator Mooring, October 2010.
systems for marine current turbine applications: A comparative [54] G. Beaudoin, D. Robertson, R. Doherty, D. Corren, B. Staby and
study,” IEEE Journal of Oceanic Engineering, vol. 37, n°3, pp. L. Meyer, “Technological challenges to commercial-scale
554-563, July 2012. application of marine renewables,” Oceanography, vol. 23, n°2,
[34] E.V Sanchez, R.H. Hansen and M.M. Kramer, “Control pp. 32-41, June 2010.
performance assessment and design of optimal control to harvest [55] P.R. Thies, L. Johanning, V. Harnois, H.C.M. Smith and D.N.
ocean energy,” IEEE Oceanic Engineering, vol. 40, n°1, pp. 15- Parish, “Mooring line fatigue damage evaluation for floating
26, January 2015. marine energy converters: Field measurements and prediction,”
[35] J. Aubry, H. B. Ahmed, B. Multon, “Sizing optimization Renewable Energy, vol. 63, pp. 133-144, March 2014.
methodology of a surface permanent magnet machine-converter [56] B. Polagye and J. Thomson, “Screening for biofouling &
system over a torque-speed operating profile: application to a corrosion of tidal energy device materials,” Northwest National
wave energy converter,” IEEE Trans. Industrial Electronics, vol. Marine Renewable Energy Center, Oregon State University,
59, n°5, pp. 2116-2125, May 2012. Report 1, April 2010.
[36] A. Babarit, J. Hals, M.J. Muliawan, A. Kurniawan, T. Moan, J. [57] H. Joe, M. Kim, S.M. Wi, H.S. Kwon and S.C. Yu, “Development
Krokstad, “Numerical benchmarking study of a selection of wave of mooring-less robotic buoy system using wave powered
energy converters,” Renewable Energy, vol. 41, pp. 44-63, May renewable energy,” in Proceedings of the 2014 MTS/IEEE
2012. OCEANS, Newfoundland (Canada), pp. 1-6, September 2014.
[37] L. Margheritini, D. Vicinanza and P. Frigaard, “SSG wave energy [58] M. Kim, S.M. Wi, H. Joe, H.S. Kwon and S.C. Yu, “Multi-body
converter: Design, reliability and hydraulic performance of an point absorber system without a mooring,” in Proceedings of the
innovative overtopping device,” Renewable Energy, vol. 34, n°5, 2014 MTS/IEEE OCEANS, Taipei (Taiwan), pp. 1-4, April 2014.
pp. 1371-1380, May 2009. [59] R. Hine, S. Willcox, G. Hine and T. Richardson, “The Wave
[38] Vryhof Anchors: http://www.vryhof.com/ammanual.htm Glider: A Powered autonomous marine vehicle,” in Proceedings
[39] I.M.L. Ridge, S.J. Banfield and J. Mackay, “Nylon fibre rope of the 2009 MTS/IEEE OCEANS, Biloxi (USA), pp. 1-6, October
moorings for wave energy converters,” in Proceedings of the 2014.
2010 MTS/IEEE OCEANS, Seattle (USA), pp. 1-10, September [60] J.S. Lee and S.C. Yu, “Preliminary study of long endurance moor-
2010. less non-drifting buoy,” in Proceedings of the 2013 IEEE IUTS,
[40] J.D. Pasternak, M. Hersley and S. Leite, “Increased offshore Tokyo (Japan), pp. 1-6, March 2013.
MODU mooring performance through MODUline® polyester
rope,” in Proceedings of the 2010 MTS/IEEE OCEANS, Seattle
(USA), pp. 1-8, September 2010. Authors’ information
[41] R.E. Harris, L. Johanning and J. D. Wolfram, “Mooring systems
for wave energy converters: a review of design issues and University of Brest, EA 4325 LBMS, Rue de Kergoat, CS 93837, 29238
choices,” in Proceedings of the 2004 MAREC, Blyth (UK), pp. 1- Brest Cedex 03, France.
10, July 2004.
[42] Bruce Connector: http:// www.bruceanchor.co.uk/ Hosna Titah-Benbouzid was born in Annaba,
[43] P.R. Thies, L. Johanning and P. McEvoy, “A novel mooring tether Algeria, in 1973. She received the Engineer
for peak load mitigation: Initial performance and service degree in agro-alimentary engineering from the
simulation testing,” International Journal of Marine Energy, vol. University of Constantine, Constantine, in 1998,
7, pp. 43-56, September 2014. the M.Sc. degree in environment engineering
[44] M. Tavakoli, T.K.A. Brekken, B. Bosma and A.A. Schacher, from the University of Picardie, Amiens, France,
“programmable mooring controller for tank testing of scaled wave in 2006, and the PhD degree in Chemical and
energy converters,” in Proceedings of the 2013 IEEE PES Environmental engineering from the University
General Meeting, Vancouver (Canada), pp. 1-5, July 2013. of Brest, Brest, France, in 2010. Dr. Titah-Benbouzid is an affiliate
[45] L. Johanning, G.H. Smith and J. Wolfram, “Mooring design member of the LBMS_Lab (EA 4325) since 2012. Her current research
approach for wave energy converters,” Proc. Inst. Mech. Eng., interests are marine renewable energy systems interactions with marine
Part M: Journal of Engineering for the Maritime Environment, environment and biofouling.
vol. 220, n°4, pp. 159-174, December 2006. E-mail: hosna.titahbenbouzid@gmail.com
[46] L. Johanning, G.H. Smith and J. Wolfram, “Measurements of
static and dynamic mooring line damping and their importance for
60
Mohamed El Hachemi Benbouzid was born in

Batna, Algeria, in 1968. He received the B.Sc.
degree in electrical engineering from the
University of Batna, Batna, Algeria, in 1990, the
M.Sc. and Ph.D. degrees in electrical and
computer engineering from the National
Polytechnic Institute of Grenoble, Grenoble,
France, in 1991 and 1994, respectively, and the
Habilitation à Diriger des Recherches degree from the University of
Picardie “Jules Verne,” Amiens, France, in 2000. After receiving the
Ph.D. degree, he joined the Professional Institute of Amiens, University
of Picardie “Jules Verne,” where he was an Associate Professor of
electrical and computer engineering. Since September 2004, he has
been with the Institut Universitaire de Technologie of Brest, University
of Brest, Brest, France, where he is a Professor of electrical
engineering. His main research interests and experience include
analysis, design, and control of electric machines, variable-speed drives
for traction, propulsion, and renewable energy applications, and fault
diagnosis of electric machines. Prof. Benbouzid is an IEEE Senior
Member. He is the Editor-in-Chief of the International Journal on
Energy Conversion (IRECON). He is also an Associate Editor of the
IEEE Transactions on Energy Conversion, the IEEE Transactions on
Industrial Electronics, the IEEE Transactions on Sustainable Energy,
and the IEEE Transactions on Vehicular Technology. He was an
Associate Editor of the IEEE/ASME Transactions on Mechatronics
from 2006 to 2009.
E-mail: Mohamed.Benbouzid@univ-brest.fr
61
Planning of Low-Voltage Distribution Networks

Using Hierarchical Methods
Amir Softic, Marinko Stojkov, Admir Jahic
Abstract – Low-voltage distribution networks represent a component of the power system with
the most prominent problems in terms of voltage quality and losses of electrical power which is
delivering to the end customers. Overcoming of such problems is a real challenge for distribution
companies, since they are obliged to deliver electricity of standard quality to the end customers,
but at the same time taking care of their own interests. In this paper, a method based on
hierarchical algorithms applicable in the planning of low-voltage distribution networks is
presented. The planning of such networks includes determining of an optimum location of
substation, radial network configuration, as well as appropriate types of power lines, all in order
to achieve the prescribed voltage levels and minimize the power losses in electricity transmission.
Copyright © 2015 Praise Worthy Prize S.r.l. - All rights reserved.
Keywords: Distribution Networks, Hierarchical Methods, Fuzzy Logic, AHP, Optimization
Nomenclature Low-voltage distribution network is characterized by a

low customers loads, frequent use of overhead power
Partition matrix lines and generally, greater length of low-voltage
, Elements of partition matrix outages. The main problem with the planning of such
( ) k-th cluster networks is finding the radial network configuration with
f Fuzzifier the lowest overall costs, whereby it have to be taken into
Euclidean distance account the size and location of the substation, the
, Coordinates of points 1 and 2 directions of propagation and the capacity of low-voltage
City-block distance power lines that supply spatially distributed customers
Chebychev distance with certain predicted loads, thermal limitations
Center of cluster (substation and low-voltage power lines), and voltage
Weight ratios levels [2]. In the market conditions, where the quality of
, Relative importance of the criteria electricity that is delivered to the end customers has an
Matrix of weight ratios important role, the primary objective in the planning of
low-voltage distribution networks becomes decreasing of
Average value
voltage variations at each customer, according prescribed
CI Consistency index
norm [3]. After that, viewed from the angle of the
RI Random index
distribution company, the next goal is achieving the
CR Consistency ratio
lower level of power losses in the network, which can be
Priority
achieved by appropriate selection of types of power lines
Geometric mean value
and optimal spatial distribution of low-voltage
q Number of columns
distribution network.
r Resistance of power line per unit length
Of course, an important issue for the distribution
x Reactance of power line per unit length
company are the construction costs of such network, but
in the market conditions customer’s demands are
I. Introduction prioritized so these costs are not in the foreground.
Finally, an important role have the capacities or
The goal of planning of low-voltage distribution throughputs of low-voltage power lines, which limit the
network is to determine the optimal network amount of loads that have to be transmitted to the end
configuration for electricity customers, arranged on customers. This problem can be solved by selecting the
particular geographic area. appropriate types and cross sections of power lines. So,
Essentially, it consists of determining the size and this is a complex problem which at the same time, have
location of the substation, types of power lines which to solve the question of optimal location of substation
will be used, and the spatial configuration of the network and low-voltage network configuration, but with respect
[1]. to mentioned criteria [4]. Herein, new introduced method
62
to solve this technical problem can be the use of modern This distance corresponds to the length of the line
hierarchical algorithms. which connects elements 1 and 2. In the same way, the
distances between all pairs of elements can be calculated.
All these distances are usually presented in the form of
II. Background Theory of Applied Models a matrix of distances. In this matrix, nondiagonal
In the literature, applied methods can be found under elements represent the distances between the pairs, and
different names, but by their nature, they can be diagonal elements are 0 (the distance of element from
classified into hierarchical methods. In the basis of these itself). Since the distance between objects 1 and 2 is
methods lies iterative process of connecting objects into identical to the distance between the objects 2 and 1, the
the groups [5]. The proposed model of planning of low- distance matrix is symmetric. This matrix is presented in
voltage distribution networks uses two hierarchical Table I.
methods, namely: Fuzzy clustering (FC) and Analytic
TABLE I
Hierarchy Process (AHP). EUCLIDEAN MATRIX OF DISTANCES
Objects 1 2 3 4 5 6 7
1 0 , , , , , ,
II.1. Fuzzy Clustering
2 , 0 , , , , ,
Fuzzy clustering is a method by which the appropriate 3 , , 0 , , , ,
data set is divided into groups or clusters, in the way that 4 , , , 0 , , ,

5 , , , , 0 , ,
elements of the same group are as similar as it is 6 0
, , , , , ,
possible. Depending on the nature and purpose of data, 7 , , , , , , 0
different measures to assess the similarity which put data
into groups, can be used. Some examples of measures Also, there are alternative measures of distances. One
that can be used in forming the groups are distance, is, so-called, the City-block distance, which uses the sum
connectivity or intensity [6]. There are two types of of absolute differences. It is often called the Manhattan
grouping, namely: hard clustering, where the data are metric because it is close to the walking distance between
divided into separate groups and each element belongs to two points of the city such as New York, District of
only one group, and soft clustering, where the elements Manhattan, where the distance is equal to the number of
may belong to more than one group and each element is blocks in North-South and East-West direction. Using
associated with a set of corresponding levels. City-block distance to calculate the distance between
The strength of specific element and the group are customers 1 and 2 gives:
emphasized by soft clustering. Fuzzy clustering is a
process of assigning the corresponding levels, and then (1,2) = | − |+| − | (3)
assigning elements to one or more groups.
One of the most widely applied algorithms is the
Finally, when working with metric data, researchers
Fuzzy C-Means (FCM) algorithm (Bezdek 1981). FCM
often use Chebychev distance, which is the maximum of
algorithm is trying to divide the group of n elements
absolute differences:
= { , … , } into subgroups, respecting certain given
criteria. On the basis of given data, algorithm provides a
(1,2) = (| − |, | − |) (4)
list of groups centres = { , … , } and partition
matrix = , ∈ [0,1], = 1, … , , = 1, … , , where
Fig. 1 illustrates the correlation between these three
each element , identify the degree of belonging of
types of distance measures.
element to the group . A standard function is [7]:
1
( )=
( . )
(1)
∑
,
There are various measures for expressing the

similarity or dissimilarity of pairs of elements. The direct
way for determining the closeness of two elements is
drawing the straight line between them [8].
This type of distance is called the Euclidean distance
(straight-line distance) and it is the most commonly used.
For the usage of hierarchical clustering procedure this
distance has to be presented mathematically. Euclidean
distance between points 1 and 2 is calculated as:
Fig. 1. Distance measures

(1,2) = ( − ) +( − ) (2)
63
II.2. Analytic Hierarchy Process The first step in AHP procedure is making
comparisons between pairs of each criterion. Any
Analytic Hierarchy Process (AHP) was developed by
decision can be presented with number at a certain scale.
Thomas L. Saaty, university professor (University of
One commonly used scale is presented in Table II [9].
Pittsburgh), in 70's and 80's of the 20th century and often,
in the literature, can be found under the name Saaty's TABLE II
method. This method allows the user to assess the SAATY’S RATING SCALE
relative weight of multiple criteria or multiple options in Intensity
of Definition Explanation
relation to the given criterion, in an intuitive way. In case
importance
those quantitative assessments are not available, the 9 Extreme importance The evidence favouring
assessor still can recognize whether some criterion is one over the other activity
more important than another. Satty has established a is of the highest possible
consistent way of converting the comparisons (X is more order of affirmation
important than Y) in a set of numbers that represents the 7 Very strong or demonstrated An activity is favoured
relative priority of each criterion. The idea of the AHP importance very strongly over another;
its dominance
method is, at first, to establish the goal, and then to demonstrated in practice
establish the criteria which can satisfy and also criteria
5 Strong importance Experience and judgement
which has the influence on the decision of the decision strongly favour one activity
maker. Criteria can be classified into one or more levels over another
(first, second, ...) which provides a hierarchy of criteria, 3 Moderate importance Experience and judgement
whereby the criteria of the first level have the greatest slightly favour one activity
influence on the decision (Fig. 2). over another
The basic of applying AHP consists of next steps: 1 Equal importance Two activities contribute
 Structuring of decision problem and selection of equally to the objective
criteria, 2,4,6,8 Intermediate values When compromise is
 Establishing of criteria priorities by comparing of needed
pairs (weighing),
 Comparison of pairs of options for each criterion The results of comparison(for each pair) are given as
(scoring), integer values from 9 (extremely different) to 1 (equal
 Obtaining an overview of the relative scoring for each value), where a larger number means that the selected
option. factor is observed as more important, in greater degree
At first, the AHP method can be implemented in four than the second factor with which it is compared.
basic steps: Weighted ratios are determined by equation:
 Decomposition of the problem into hierarchy,
 Comparison by pairs and establishing of priorities = (5)
among the elements in the hierarchy,
 Synthesizing of decisions for getting a set of overall
decisions or weights to achieve the goal, where , are relative importances of the criteria.
 Evaluation and verification of consistency of The weighted ratios are recorded in the form of a
decisions. weighted ratios matrix, in the form:
The basic procedure consists from: ⋯
 Developing a score for each decision, alternative for
= ⋮ ⋱ ⋮ (6)
each criterion,
⋯
 Developing the weights for criteria,
 Calculation of weighted average assessments for each where q is the number of criterion.
alternative decision. Selecting the one with highest
score.
Fig. 2. Decomposition of the problem
64
The next important thing is to determine matrix or:

consistency, which means to check whether the following 1.
expression is valid: ⋮ (13)
.
= (7)
Sometimes, the following expression can be used: III. Planning Methodology

Planning methodology of low-voltage distribution
= ∗ ,∀ , , (8) networks, using mentioned hierarchical methods, can be
easily presented as it is shown on diagram given in Fig.
Since it is a matter of human judgement, this relation 3.
may not be true. The following procedure is used for the
mathematical determination of consistency [10]:
 Calculating the sum of columns,
 Scaling of matrix (each column is divided by the sum
of elements of that column),
 Calculating the sum of rows (weight) in scaled
matrix,
 Dividing the weights with priority vector of first
matrix,
 Calculating the average value of obtained values
,
 Calculating the consistency index CI:
−
= (9)
−1
where n is the degree of the matrix,
 Calculating of consistency ratio CR:
= (10)
where RI is random consistency index (Table III).
TABLE III
RANDOM CONSISTENCY INDEX
n 1 2 3 4 5 6 7 8 9 10
RI 0 0 0.58 0.90 1.12 1.24 1.32 1.41 1.45 1.49
If ≤ 0.1, matrix is consistent. Fig. 3. Planning methodology

Priority indicates how much of influence have each of
characteristics to the decision. It represents the ratio of Obtaining the optimal solution according to the given
geometric mean of one matrix row and the sum of methodology can be described in several steps:
geometric means of all rows:
STEP 1
In this step, the input data have to be defined. A
= (11)
∑ geographic coordinates of measurement points of each
customer which has to be supplied through the planned
where q is the number of columns. distribution network are used as input data. They are
After all calculations, the vector of priorities of determined from georeferenced maps.
alternatives is obtained, from which, the best alternative
can be selected. STEP 2
Vector of priorities of alternatives is obtained by After obtaining the input data, it is necessary to
multiplying the matrix of priorities by criteria with vector determine the location of the substation, and the number
of priorities of criteria shown in the following Eq. (12): of low-voltage outages. This process is performed by
using a specialized fuzzy algorithm as follows:
. 1. . .1 ⋯ . 1. .
⋮ ⋱ ⋮  At the beginning, the location of substation is
. . . .1 ⋯ . . . . determined. It has to be located in the geographic
centre of consumption (Fig. 4).
65
Fig. 6. Low-voltage network outages

Fig. 4. Location of substation
 After that, the spatial grouping of consumers is

performed. Number of groups simultaneously defines
the number of low-voltage outages of substation. By
defining of the group, its spatial centre is also
defined, as well as spatial centres of subgroups that
serve as an orientation in which direction the network
should be constructed, or which type of cables have
to be installed (Fig. 5).
Fig. 5. Low-voltage outage Fig. 7. Example of low-voltage distribution network topology
 Finally, it is obtained a spatially located substation, in STEP 4

the centre of consumption, and spatially grouped For obtaining the final, optimal solution, in this step it
customers groups. In this way, a number of low- is necessary to determine the voltage drops and power
voltage outages and guidelines for spatial network losses in the distribution network defined in the previous
propagation is also obtained (Fig. 6). step. Input data, which include substation parameters,
cross-sections and lengths of low-voltage power lines, as
STEP 3 well as individual customer’s loads, are obtained in the
Determining the location of the substation and above steps. For the implementation of this any software
grouping of customers creates the conditions for forming package for power flow calculation (Matpower,
of low-voltage distribution network topology. PowerCAD etc.)can be used.
Taking into account the spatial distribution of objects,
roads, etc. on the basis of obtained data by fuzzy STEP 5
clustering, an experienced engineer is able to draw the Based on the data from all of the previous steps, the
distribution network on georeferenced maps as it is application of the AHP method in this step gives the
presented in Fig. 7. optimum solution.
66
This method formulates alternatives and ranks certain These data we were
re necessary for calculation of the
criteria, and based on them it shows which of offered voltage drops, power losses and power lines loads. The
alternatives is the best. Since the ranking ooff criteria is data for assessment of overall costs, needed fo forr
issue of subjective assessment of someone who performs construction of the network, we were
re also obtained.
the planning
planning,, that can lead to the error in the ranking of Calculation of other rrequired
equired data was
was performed in
criteria, so this step has to be repeated until it satisfied Matlab (Matpower) [11]. For the calculation of voltagevoltage
the condition of consistency. drops, power losses and
and power lines loads are used the
When this condition is satisfied
satisfied as tthe
he final solution,
solution data of installed capacity of substation, customer’s loads
an optimal location and installed capacity of substation, and parameters of power lines
line sections ((r, x,
x, length). The
The
the optimal number of outages and topological obtained data about voltage drops, for each alternative,
configuration of low-low-voltage
voltage distribution network are are presented in Tabl
Tablee IV.
obtained.
obtained
TABLE IV
CALCULATED VALUES OF VOLTAGE DROPS
Number of Un ΔU U ΔU ΔU
IV. Results clusters (V) (p.u) (V) (V) (%)
The functionality of the proposed methodology is 5 400 0.930 372.0 -28.0
28.0 -7.0
7.0
6 400 0.934 373.6 -26.4
26.4 -6.6
6.6
tested using the example of a real low low--voltage
voltage 7 400 0.935 374.0 -26.0
26.0 -6.5
6.5
distribution network. It supplies 270 70 electricity 8 400 0.940 376.0 -24.0
24.0 -6.0
6.0
customers
customers, in the category of households
households, through
hrough this
network. The network is powered by the transformer The obtained data of power losses, for each
station 10(20)
10(20)/0..4
4 kV,
kV rated power of 400 kVA. alternative, are presented in Table V.
Customers loads w were
re calculated using method of
peak loads based on annual realized consumption of each TABLE V
customer and daily, one-hour
one hour measurements of low-
low CALCULATED VALUES
LUES OF POWER LOSSES
voltage outages loads (Fig. 8). Number of clusters Losses (kW) Losses (%)
5 7.934 3.477
6 7.114 3.118
7 7.098 3.111
8 6.581 2.844
The obtained data of average loads of power lines, for

each alternative, are presented in Table VI.
TABLE VI
AVERAGE LOADS OF LINES
INE
Number of In I average I average
clusters (A) (A) (%)
5 198 71.62 36.17
6 198 59.48 30.04
7 198 50.98 25.75
8 198 44.51 22.48
Fig. 8. Daily load curve of TS 10(20)/0
10(20)/0.4
4 kV, 400 kVA
Another data required for the application of the AHP
At ffirst
irst,, a network configura tion according tto
configuration o the
method are estimated costs. Table VII presents an
proposed
oposed methodology was made and the calculations
overview of overall costs for each alternative
alternative.
were car
were ried out, and then the results we
carried were
re compared
with the data obtained for the existing network. Thus, TABLE VII
according to the methodology, at first, the coordinates of OVERALL ESTIMATED COSTS
all measurement points were obtained from Number of clusters Costs ((€)
georeferenced maps.
maps They represent
represented the input data for 5 207395
6 238631
the application of fuzzy algorithm. Using this algorithm, 7 263265
in first step, it was defined a geograp
geographic
hic location of 8 268912
substation
substation. After that, using the same algorithm, a
number anandd spatial distribution of customers groups were After calculation all of necenecessary
ssary data, the final
determined
determined, which also represent
representeded a number of low-
low solution wa
wass obtained by applying the AHP method. Four
Four
voltage outages. Since,
ince, for obtaining a final solution wawas alternatives (networks
(networks with 5, 6, 7 and 8 lowlow-voltage
voltage
necessary to have more aalternatives,
lternatives, four alternatives outages)are
outages) are obtained by Fuzzy algorithm.
algorithm. After that, the
were determined. Alternatives we re formed oon
were n the basis criteria we
were
re determined, namely:
of the number of low low-voltage
voltage outages (5, 6, 7 and 8  Criterion of voltage drops,
outages). After that, in AutoCAD, a drawing of lowlow  Criterion of power losses,
voltage networks was donedone,, for all formed alternatives.
alternatives  Criterion of average lines loads,
This was provided the network data.  Criterion of overall costs.
Copyright © 2015 Praise Worthy Prize S.r.l. - All rights reserved International Review of Electrical Engineering, Vol. 10,, N. 1
67
Ranking of criteria was carried out at first. In first step TABLE XIII
CALCULATION OF PRIORITIES IN RELATION TO ALTERNATIVES
a pair wise comparison matrix was established, as it is
Costs Alt. 1 Alt. 2 Alt. 3 Alt. 4 Priority
presented in Table VIII. Alt. 1 1 1.1506 1.2694 1.2966 0.2917
Alt. 2 0.8691 1 1.1032 1.1269 0.2535
TABLE VIII Alt. 3 0.7878 0.9064 1 1.0215 0.2298
PAIR WISE COMPARISON MATRIX Alt. 4 0.7712 0.8874 0.9790 1 0.2250
Voltage Lines
Losses Costs
drops loads TABLE XIV
(C2) (C4)
(C1) (C3) WEIGHTED AVERAGE RATINGS FOR EACH ALTERNATIVE
Voltage drops 1 2 4 5 Alternative Weighted average ratings
Losses 0.50 1 4 5 Alternative 1 0.24154
Loads of lines 0.25 0.25 1 2 Alternative 2 0.24776
Costs 0.20 0.20 0.50 1 Alternative 3 0.24682
Alternative 4 0.26390
After determining the rows weights and priorities
values of comparison matrix there were determined the From these results, it is obvious that alternative four is
vectors of eigenvalues of the comparison matrix as it is the best solution, as it is provided the highest value of the
presented in Table IX. vector of priorities. It means that the optimal solution for
such low-voltage distribution network is construction of
TABLE IX the network with eight outages. From the table, too, it
PRIORITY VECTORS can be concluded that a larger number of low-voltage
Priority outages does not necessarily mean a better solution
C1 C2 C3 C4 Ʃ
vectors because the solution with six outages, in this case, is
C1 0.5128 0.5797 0.4211 0.3846 1.8982 0.4746
C2 0.2564 0.2899 0.4211 0.3846 1.3519 0.3380
better than solution with seven outages. This solution
C3 0.1282 0.0725 0.1053 0.1539 0.4598 0.1149 was compared with the existing distribution network in
C4 0.1026 0.0580 0.0526 0.0769 0.2901 0.0725 terms of voltage drops and power losses. The values of
voltage drops and power losses for the existing network,
After calculating = 4.08876, it was necessary to with five low-voltage outages, are calculated in
examine consistency. After calculating, the values Matpower and the results are presented in Table XV.
= 0.02959 and = 0.03287were obtained.
Since ≤ 0.1, it means that ranking of criteria was TABLE XV
VOLTAGE DROPS AND POWER LOSSES IN EXISTING NETWORK
performed well. After that, the calculation of priorities of
specific criteria in relation to alternatives was done. The Un ΔU U ΔU ΔU Losses Losses
(V) (p. u) (V) (V) (%) (kW) (%)
results are presented in Tables X – XIII.
After obtaining of these results, a multiplication of 400 0.909 363.6 -36.4 -9.1 7.226 3.167
these priorities with values of comparison matrix was
performed. The obtained results are presented in Table Comparative results with the solution obtained by
XIV. proposed methodology are presented in Table XVI.
TABLE X TABLE XVI

CALCULATION OF PRIORITIES IN RELATION TO ALTERNATIVES COMPARATION OF VOLTAGE DROPS AND POWER LOSSES
Voltage ΔU ΔU ΔU Losses Losses
Alt. 1 Alt. 2 Alt. 3 Alt. 4 Priority
drops (p. u) (V) (%) (kW) (%)
Alt. 1 1 0.9429 0.9286 0.8571 0.2323 Existing
0.909 -36.4 -9.1 7.226 3.167
Alt. 2 1.0606 1 0.9849 0.9091 0.2464 network
Alt. 3 1.0769 1.0154 1 0.9231 0.2502 Methodology
0.940 -24.0 -6.0 6.581 2.844
Alt. 4 1.1667 1.1000 1.0833 1 0.2711 network
TABLE XI It is obvious that the solution obtained by the planning

CALCULATION OF PRIORITIES IN RELATION TO ALTERNATIVES
methodology gives better results. At first view, the
Losses Alt. 1 Alt. 2 Alt. 3 Alt. 4 Priority
Alt. 1 1 0.8967 0.8947 0.8295 0.2253
improvement is small, but bearing in mind that in the
Alt. 2 1.1153 1 0.9978 0.9251 0.2513 distribution system there are a number of such networks,
Alt. 3 1.1178 1.0023 1 0.9272 0.2518 in the total amount, a significant improvements could be
Alt. 4 1.2056 1.0810 1.0786 1 0.2716 obtained. Although the power losses in the network are
not significantly reduced methodological solution gives
TABLE XII
CALCULATION OF PRIORITIES IN RELATION TO ALTERNATIVES good results in terms of voltage drops.
Lines
Alt. 1 Alt. 2 Alt. 3 Alt. 4 Priority
loads
Alt. 1 1 1.2424 1.3025 1.2644 0.2973 V. Conclusion
Alt. 2 0.8049 1 1.0483 1.0177 0.2393
Alt. 3 0.7678 0.9539 1 0.9708 0.2283 In the paper a method for planning of low-voltage
Alt. 4 0.7909 0.9827 1.0301 1 0.2351 distribution networks is presented. The method is based
on the hierarchical methods.
68
Fig. 9. Low-voltage distribution network obtained by planning methodology
It implies the planning process from selecting the type demands for the quality of supplied electricity have the
and location of the substation to the determination of the primary role. In addition, by defining the configuration of
optimal configuration of low-voltage network. distribution network, distribution companies can pre-plan
On a real example it is easy to see that methodology the necessary funds for the reconstruction or construction
gives better results in comparison with the existing of distribution networks. Therefore, this methodology
distribution network, especially in the case of voltage can be successfully applied to the reconstruction process
drops at end customers. This is particularly important and in the process of constructing a completely new low-
because in the process of market deregulation customers’ voltage distribution networks, while at the same time can
69
also be used to detect problems in the network, as well as Marinko Stojkov (1970, Croatia) graduated
BSc (1994), MSc (1998) and PhD (2002) on
to offer suggestions for their solution. By correct ranking
Faculty of Electrical Engine
Engineering,
ering, Energy
of criteria and the precise definition of the network Department, University of Zagreb. He was
topology,, methodology gives better results.
topology employed in HEP HEP-DSO
DSO (1995-2009)
(1995 2009) on
Therefore, for an experienced engineers who positions: manager at the Maintenance
Department, Head of Department for
participate in the planning process this methodology can Development and Head of Planning and
serve resented
rve as a very useful tool. Presen
Presen ted methodology,
methodology in the Investment of Elektra SlavonskiBrod. He started his academy career on
same way,
way can be used for planning the medium
medium--voltage
voltage Faculty of Electrical Engineering, University of Osijek (1998). He was
distribution networks. employed as Assistant Professor on Mechanical Engineering Faculty in
SlavonskiBrod, University of Osijek (2009) and as Associate professor
(2010) in the scientific field of technical sciences - electrical
engineering. He attended LV and MV cables and cable equipment
References course (1996) in Budapest, Hungary and "Electric Distribution
[1] Eloy Diaz-Dorado,
Diaz Dorado, Jose Cidras Pidre, Edelmiro Miguez Garcia, Garcia Management" course (2003) in Dublin, Republic of Ireland. He
Planning of Large Rural LowLow-Voltage
Voltage Networks Using Evolution published 3 CC scientific journal papers, 5 SCI SCI-Expanded
Expanded journal
Strategies, IEEE T
Strategies, Transactions
ransactions on Power Systems, vol. 18, No. 44, scientific papers, 8 papers in other scie
scientific
ntific journals, 25 international
November 2002003. scientific conferences papers, 1 invited lecture, 15 expert papers. Also
[2] Alejandro Navarro, Hugh Rudnick, Large Large-Scale
Scale Distribution he reviewed 20 scientific papers in journals and 4 international
Planning Part I: Simultaneous Network and Transformer
Planning-Part conferences papers.His research interests are: power distribution
Optimization, IEEE Transactions on Power Systems, vol. 24, No. network, power loslosses,
ses, harmonic distortion, renewable energy sources,
2,, May 2009. load forecast, electric and magnetic fields, energy efficiency and energy
[3] Zvonimir Klaic, Srete Nikolovski, Zorislav Kraus Kraus,, Voltage savings. He is a member of IEEE.
variation performance indices in distribution network
network,, Technical
Gazette, Vol. 18
18,, No. 4, 547 551, 2011.
4 pp. 547-551, Admir Jahic was born in Tuzla in Bosnia and
[4] Ian Dent, Tony Craig, Uwe Aickelin, Tom Rodden Rodden,, An Approach Herzegovina, on December 4, 1975. He received
for Assessing Clustering
Clustering of Households by Electricity Usage Usage, his gr
graduation
aduation (dipl. Eng.) and M.Sc. degree
2014.
Cornell University, September 2014 from the Faculty ofElectrical Engineering at the
[5] Sadeghi Sarcheshmah, M., Seifi, A.R., A new fuzzy power flow University of Tuzla, both in electrical
analysis based on uncertain inputs, (2009) International Review of engineering. In 2011 elected for a teaching
Electrical Engineering (IREE), 4 (1), pp. 122122-128
128. assistant in the Department of Power Systems at
[6] Paracha, Z.J., Kalam, A., F Fuzzy
uzzy clustering techniques for the Faculty of Electrical Engineering Tu Tuzla.
zla.
analysis of PQ data in electrical power distribution system, (2010) Currently holds a position of an “Engineer for the energetics and
International Review of Electrical Engineering (IREE), 5 (5), pp. investments” in “Elektrodistribucija“ Tuzla, a distributive part of a
2152 2158.
2152-2158. Public Enterprise Elektroprivreda B&H – Sarajevo, the public
[7] Marinko Stojkov, MirzaAtic, Amir Softic, Application of Fuzzy enterprise for generation, distributi
distribution
on and sale of electrical power. His
logic
ogic for reactive power compensation by synchronous motors, topics of interest are
are: analysis, optimization, planning, control, state
Technical Gazette, Vol.19, No. 4, pp. 753
753-758,
758, 2012. estimation and distribution networks automation.He is a member of
[8] Eduardo Gouveia, Manuel A. Matos Matos,, Constrained Fuzzy Power IEEE, CIGRE and CIRED.
Flow
Flow,, Power Tech, IEEE Lausanne,
Lausanne, 2007
2007.
[9] Thomas L. Satty, Decisions making with the analytic hierarchy
process, Int. J. Services Sciences, vol. 1, No. 11,, 2008
2008.
[10] Thomas L. Satty, How to make a decision: The Analytic
[10
Hierarchy Process, Journal of Operational Research
Research, vol. 24,Issue
6, December 1994.
[11 Ray D. Zimmerman, Carlos E. Murillo-Sanchez,
[11] Murillo Sanchez, MATPOWER
4.1 User's Manual, 2011.
Amir Softic
Softi was born in Tuzla (Bosnia and
Herzegovina) on 30th of January, 1964. He
Herzegovina)
received a B.S. degree in Electrical Engineering
and M.S. de degree
gree from University of Tuzla
(Bosnia and Herzegovina
Herzegovina),), in 1996 and 2009,
respectively. From 2005 he is working in PE
ElektroprivredaBiH Sarajevo, at works which
are related to distribution networks analysis and
management.His topics of interest are are: distribution networks, power
network optimization,
optimization, energy analysis and distribution network
planning, supervising and managing
managing,, electromagnetic compatibility,
renewable energy sources.
sources
70
Single Phase Fault Location in Distribution Networks

Based on Charge Transients
Šeila Gruhonjić Ferhatbegović
Abstract – In this paper the one possibility of single phase fault location in unearthed MV
distribution network is presented. This method is based on transients which appear in the case of
charge redistribution in the instance of fault occurrence. The issues of fault location in
distribution networks was emphasized through influence of inhomogeneity, distributed loads,
considerable fault resistance and numerous branches on distribution feeders. On the extracted
transients of currents and voltages of faulty phase, method based on solving differential equations
is applied.
The low frequency charge transients, with minimum sample frequency of several kHz is used. This
method was examined by computer simulations on the modeled network as well as on the real
network and fault. In the case of real fault, it was used data registered by power quality analyzer.
Copyright © 2015 Praise Worthy Prize S.r.l. - All rights reserved.
Keywords: Distribution Medium Voltage (MV) Network, Fault Location, Charge Transients
Nomenclature I. Introduction
vA, vB, vC Transient voltages of phases A, B and C [V] For this type of fault location the currents and voltages
iA, iB, iC Transient currents of phases A, B and C [A] at the sending end of distribution feeder in supplied
v0 Zero sequence voltage [V] transformer station (TS) registered during a fault are
i0 Zero sequence current [A] used. This way of determining fault place is referred as
t Time [s] on-line location, automated or fault location in real time.
x Estimated distance to the fault place [km] In transmission networks of voltage levels 110, 220
Δt Sampling time [s] and 400 kV is in almost regular usage. In medium
Rf Fault resistance [Ω] voltage (MV) distribution networks, commercial
Cz, Cm Ground and phase to phase feeder modules of fault locators occur only in recent years.
capacitances [µF] These are mainly fault locators that estimate fault
Ltr, Lv Transformer and feeder inductances [H] location on the basis of fundamental frequency currents
ω Angular frequency [rad/s] and voltages and reactance estimation to the fault place.
vk Phase voltage at fault time inception [V] If a network is branched it can be happened that
Iz Uncompensated RMS value of stationary estimated location corresponds to multiple points in a
fault current [A] network (problem of multiple fault location) or, if a
Ruz Resistance of neutral transformer grounding neutral point of network is isolated, there is no possibility
[Ω] of determining a place of single-phase short circuit.
l Total length of a feeder [km] Regarding that a single-phase short circuit is the most
Rx Faulty path resistance [Ω] common fault type (approximately 70% of faults in
Lx Faulty path inductance [H] distribution networks are single-phase faults), usability
L0’, L1’, L2’ Feeder inductances of zero, positive and of these devices is small in isolated network and in these
negative sequences per unit length [H/km] networks they can be used only for location phase to
Ik3 Short circuit current [A] phase faults. In grounded network, it is possible to
f Frequency [Hz] estimate impedance to a fault place also in the case of
ϑ Phase angle [°] single-phase short circuit. Difficulties with determining
x1 , x0 Positive and zero sequence reactances of a reliable location of single-phase short circuit in the case
feeder [Ω] of MV distribution networks with isolated neutral point,
Ceq Equivalent capacity [µF] are caused by the fact that fundamental frequency fault
Leq Equivalent inductance [H] current often small compared with phase to phase fault
I Initial amplitude of a charge transient [A] currents. This happens even if fault resistances are small,
i.e. in the case of direct single-phase short circuit (or
grounding).
71
Š. Gruhonjić Ferhatbegović
It should be borne in mind that a large portion of MV After identification of the nearest reactance value, it
network works with isolated neutral point. This applies follows determination of a distance to the fault place in
particularly to overhead network, which are often the [km] from nearest nodal point or nearest 10(20)/0,4 kV.
longest and with the most number of branches [1], [2]. In this way a location where team for repairing and
The choice of direct transformation 110/20 kV, may remedy a problem need to go out is determined. The
further highlight the need and advantages of automated problem of fault locating in distribution networks was
fault location. The rated voltage of 20 kV, according to considered in the literature [4]-[8].
the characteristics of topology, configuration, neutral
point treatment, the existence of branches and distributed
loads remains the same as in distribution feeders at 10 II. Charge Transient Frequency
kV. The difference is that 20 kV distribution feeders, Calculation
because of better transmission capacity, can supply
Transient fault location methods use currents and
customers in a wider area within allowed losses and
voltages in transient conditions that occur in a process of
voltage drops. This will cause longer feeders and
discharge and charge network capacity at the moment of
distance estimation during a fault, may be of great
single phase fault inception. In general, distribution
importance and can significantly reduce cost of field,
system is made up of resistance, inductance and capacity
value of unserved energy and periods without voltage.
which exist as concentrated parameters but are also
In relation to a fault location in transmission network,
distributed along the line, transformers, etc.. During each
in the case of fault location in MV network there are
state variation, such as operation of interrupting of short
certain characteristics that make difficult estimation of
circuit, fault condition, this system should generate a
distance to the fault place. At first these are non-
wide range of transients.
homogeneous feeders, distributed loads and many
The process of charging network capacity causes
branches. The impact of fault resistance, especially for
voltage disturbance in healthy phases. Because of voltage
single-phase faults in distribution system should be taken
drop of faulty phase to ground potential, a voltage of
into account. A load model is needed to predict load
neutral point of transformer in supply substation
behavior depending on changes in supply parameters
increases, which leads to increase in phase voltages of
(voltage and frequency).
healthy phases. Depending on fault resistance, earth
Loads distributed along distribution feeders can make
capacitance of healthy phases of all feeders in a network
significant errors during fault location. Although it is
have been charged through the inductance of transformer
tried to make a symmetry, during connection single-
to a voltage whose effective value is between phase to
phase loads at low voltage (0,4 kV), it cannot be said that
ground and phase to phase voltage. It is necessary to
loads in the distribution system are completely
analyze a transient fault that can be used for the purpose
symmetric. In the process of fault location is essential
of fault location. Typical fault transients can be classified
dynamic behavior of loads and here it is necessary to
as low-frequency and high-frequency transients. Low-
take into account the impact of the fault duration or the
frequency transients are related to charge transients and
impact of the time required to make the estimation of the
transient currents of compensation windings (in case of
distance to the fault [3]. In general, a fault time interval
compensated distribution networks). High-frequency
depends on several factors; value of a fault current,
transients include discharge transients of network
protection system in area where fault occurs, type of
capacity and traveling waves.
fault, etc. If fault location based on transient phenomena
Charge transients are relatively low frequencies, so
are used, it should be noted that electromagnetic
that the required level of the sampling frequency is a few
transients on distribution level are enough attenuated and
kHz. Travelling waves, in the case of overhead lines,
disappear for a few [ms].
spread around the speed of light. Even if a network is
Practically, electromagnetic transients never last
relatively long, it is necessary a considerable high
longer than a few cycles. Distribution feeders are
sampling frequency to identify and record traveling
comprised of laterals that have different electrical
waves. For overhead feeders, the minimum required
parameters so that appearance of discontinuities in the
sampling rate is in the order of one MHz. Charge
direction of the impedance is occurred, i.e., variations in
transients are low frequency, and in distribution
feeder impedance. It is caused by different types of
networks, have frequencies in the range of 100-800 Hz.
conductors, cross sections, geometry of stubs, different
The frequencies are near the upper limit for a faults
types of cables, switching from overhead to a cable
close the buses. In the case of method testing, transient
network, etc. For this reason, determination of a distance
spectrum of currents and voltages of faulty phase (or zero
to the fault point on the basis of estimated parameters
sequence current and voltage) was first observed in order
from supply substation to the fault point, such as
to determine the dominant frequencies [8]-[10].
reactance, inductance and impedance can be difficult. In
The spectrum of transient voltage and current was
such cases it is necessary to take data from a program for
extracted by using fast Fourier transform on signal of
network analysis where all discontinuity points are
voltage and current for the first cycle from the moment
identified and determined by values of short-circuit
of a fault inception. Fast Fourier transform (FFT) is a
currents.
complex linear transformation of a time domain to the
72
Secondary side
frequency domain and thus provides an estimation of of transformer in TS x/10(20) kV
frequency components in input signal [11], [12]. Overhead feeder
Phase A
Ltr
With established charge and discharge frequencies for Lv
Cm
Vf
a specific distribution network, it can be accessed to Phase B

Ltr
calculations of distance to the fault location. Also, for Lv
Cm Cm
Phase C
purpose of mathematical estimates of amplitude and
Ltr Lv
frequency of charge transient, the approximate relations
according to the simplified situation with neglected Iz Cz Cz Cz
Earth fault in phase C
Rf = 0 ohma
effective resistance of conductor and transformer
windings are derived. During inception of single-phase
fault in power system, voltages in two healthy phases Fig. 1. Single-phase fault of phase C in an isolated network.
Model of system for transient frequency estimation
grow rapidly, which creates transient phenomena. These
transients pass through coils of transformer x/10(20) kV.
A charging process causes transients that are visible in
zero current of all feeders in a network. Transients of all
healthy phases close across the ground through a fault
location and in opposite direction pass a faulty feeder.
Process of charging capacity of healthy feeders can be
described by the equation [8]-[10]:
t
1
v0  t   v0  t0   i0  d (1)
Cz t
0 Fig. 2. The equivalent scheme for the charge transient
frequency estimation
where is:
Cz – equivalent ground capacitance of a feeder, In the case of a fault on the buses 10(20) kV, it holds
I0 – zero current of a feeder that Lv=0, so that angular frequency of charge transient
For the estimation of charge transient frequency of has a slightly higher value:
network capacity, equivalent scheme with lumped
parameters, as indicated in Fig. 1, has been observed. 1
 (3)
In the first approach the effective resistance of 3  Ltr   C z  Cm 
conductors and transformers in the supply TS x/10(20)
kV were neglected. It was considered network with Eq. (3) is valid in the case where the inductance of
isolated neutral point and the direct earth fault, i.e. the transformer is large in comparison to the feeder
situation with zero fault resistance was assumed. inductance. In specific cases it is necessary to make a
During a fault in phase C, ground capacitance of comparison between transformer inductance and feeder
faulty phase can be neglected because it is short-circuited inductance whereas, when applying fault locator based
by fault itself. Phase to phase capacitances Cm of adjacent on charge transients, assign one particular frequency.
phases are parallel with their earth fault capacitances Cz, The assigned frequency need to be applicable for fault
as shown in the Fig. 1. The capacitance between phases in different places on a feeder. The initial amplitude of
that are not faulty (phases A and B in the analyzed case) the charge transient, without attenuation, can be
can be omitted because the ends are at the same potential calculated as:
with a transient viewpoint. The inductance of the
transformer substation in the supply TS x/10(20) kV is vk
Ltr. In accordance with the previously used labels the I  vk    Ceq (4)
  Leq
equivalent diagram was shown in the Fig. 2.
According to this situation, the charge transient
frequency of phases that are not faulty can be calculated If a fault occurs at the moment of passing through the
from the following equation: phase voltage maximum, the amplitude of a charge
transient is given in the following equation:
1
 (2) 2  Ceq  
3   Ltr  Lv    C z  Cm  I  Iz (5)
3  C z 1
where is: where is:

Ltr – transformer inductance in supply TS x/10 (20) kV; ω1 – fundamental angular frequency
Lv – feeder inductance; vk – value of phase voltage at a fault time inception
Cm – phase to phase feeder capacitance; Iz – uncompensated RMS (root mean square) value of a
Cz – phase to ground feeder capacitance. stationary fault current at zero fault resistance
73
B1
In the case of networks with compensated neutral,
charge frequency transients is approximately the same as Fault
locator
in the case of networks with isolated neutral. The L1 L2 L3
resistance that compensation coil provides toward v(t) Supply TS x/10(20) kV Main sections:
passing of charge transients has great value. L1=9 km, L2=14 km, L3=8 km
B2
For example, if it is assumed charge transients Branches:
B1=11 km, B2=7 km
frequency of 400 Hz and a compensation coil inductance
of 5 H, then the inductor resistance toward passing of a Fig. 3. Example of 10 kV distribution feeder where computer
charge transients is 2000 Ω. Therefore, the path across simulations were made
charge transients are closed is approximately the same as
in the case of an isolated network. Also, this is shown by For this model the frequency spectrum of transients
computer simulations. In a network with low-resistance was observed. During testing, it was necessary to obtain
grounding, estimated value of frequency transients, an insight into the parameters that influence to the
according to an equivalent scheme, should correspond to spectrum of current and voltage. Therefore, a fault
the following relation: resistance, fault time, initial angle of faulty phase
voltage, treatment of neutral point, dynamic of loads and
location of a fault were changed. For example, with
1  1 
Ruz  j   Ltr  Lv    modeled ground fault on the section L2, spectrum of
2    C z  Cm   current and voltage of faulty phase, obtained by Fourier

1  1  (6) transformation, was indicated in the Fig. 4. The spectrum
Ruz  j   Ltr  Lv    is calculated for the first cycle from the moment of a
2    C z  Cm  
fault occurrence.
 j  Ltr  Lv   0 Selected signal: 7.5 cycles. FFT window (in red): 1 cycles
40
where is: 20
Ruz – resistance of neutral transformer grounding in 0
supply TS. -20

0 0.05 0.1 0.15
Accordingly, it is expected that a spectrum of Time (s)
transients in low-resistance grounded network is to be
shifted in relation to a network with isolated and Fundamental (50Hz) = 24.37 , THD= 32.16%
compensated neutral. The previously equations for 4
charge transient frequencies calculation were estimated
with the assumption of direct single-phase faults, or in 3
the case when the fault resistance Rf is equal zero.

Mag
However, already from an equivalent scheme for 2
charge transient frequencies calculation, it can be seen

1
that the existence of fault resistance has an impact to
attenuation of charge transients. If, in addition to faulty 0
0 500 1000 1500 2000 2500
feeder, there are other feeders connected to the same bus Frequency (Hz)
in supply TS x/10(20) kV (as usually in distribution
Selected signal: 7.5 cycles. FFT window (in red): 1 cycles
system), then the connected feeders also have an impact
5000
on a generation of fault transients. The capacitance of
healthy phases, of other feeders to a common buses, are 0
parallel to the capacitance of healthy phases of faulty
-5000
feeder, Cz and Cm. The total equivalent feeder capacity on
0 0.05 0.1 0.15
the same bus should be added to the equivalent faulty Time (s)
feeder capacity. From the relation of charge transient
Fundamental (50Hz) = 245.7 , THD= 679.02%
angular frequency it can be seen that this leads to a 800
reduction in transient frequency and increasing value of
initial peak of fault transient current. 600
Mag
400
III. Transient Frequency Identification
by Computer Fault Simulation 200
Testing of a fault transient frequency spectrum has 0

0 500 1000 1500 2000 2500
been made on distribution feeder models by simulating Frequency (Hz)
of ground faults with different parameters. For example,

Fig. 4. The spectrum of current and voltage of faulty phase. The
a simplified distribution feeder model was given in the moment of the fault occurrence is t = 0 s. The fault resistance is Rf = 10
Fig. 3. ohms. It is observed 20 ms from the fault time occurrence
74
It can be seen that, charge transients occurring at a   ik 1  ik    vk  2  vk 1    

frequency of approximately 600 Hz, and discharge  
transients occurring at approximately frequency of 1700 t     ik  2  ik 1    vk 1  vk   
Lx    (10)
Hz. In the case of fault computer simulation, it is 2   ik 1  ik    ik  2  ik 1    
recorded the impact of fault resistance to attenuation of   i  i   i  i   
the amplitude transients.   k  2 k 1 k 1 k 
It can be mentioned that the limit of fault resistance to
detect the charge transient is approximately 250 ohms. In where is:
the case of frequency identification of charge transients it Δt – sampling time;
was observed dependence of the spectrum of the fault v, i –transient voltage and current of faulty phase;
location. k – integer variable.
It has been noted that the spectrum shifts to higher For the first order model and neglected active
frequencies for the case of faults that are closer to the resistance of conductors, the faulty path can be
buses of supply TS. represented by a serial connection of symmetrical
components of the feeder impedance. Inductance of
faulty path Lx is in proportion to a fault point distance
IV. Fault Locators based on Low according to the following equation:
Frequency Transients
The identified and isolated transient current and 1
Lx    L0 '  L1 '  L2 '   x (11)
voltage signals are input values for a given algorithm that 3
calculates distance to the fault point.
Extracting of charge transients implies that from the where is:
voltage and current signal in supply TS are allocated x – distance to the fault place
frequencies at which happens transient process in the L0’, L1’, L2’ – feeder inductances of zero, positive and
case of charging network capacity. These frequencies negative sequence per unit length, respectively
correspond to expressed peak occurring in the spectrum Based on an estimation of inductance of faulty circuit,
of current and voltage. it can be calculated distance from supply TS x/10(20) kV
As an example, in the Fig. 5 are given current and to the fault point by the Eq. (11). The equation (11)
voltage diagrams for faulty phase and estimated charge applies to the case of a homogeneous distribution feeders
transients. made by lines of the same conductor type.
This is extremely uncommon in practice, so that it
should be considered inhomogeneity distribution feeder.
IV.1. The Algorithms Based on Differential Equations After the estimation of faulty path resistance and
After separation of low frequency transients, it follows inductance according to (9) and (10) in order to estimate
solving of differential equations of faulty circuit (for the the fault location it can be calculated short circuit current
first order model), by numerical integration [12]: corresponding to the parameters Rx and Lx:
t1 t1 Vn
Ik3  (12)
 v  t dt  Rx i  t  dt  Lx i  t1   i  t0  
 (7)
t0 t0
3 R 2
x   2  L2x 
t2 t2 The calculated value of current Ik3 is compared with
v  t  dt  Rx i  t  dt  Lx i  t2   i  t1   currents in short-circuit analysis in all nodes of the
  (8)
considered feeder.
t1 t1
By identifying the nearest value of current and node
(discontinuity point) on a single line network diagram,
The previous integrals should be evaluated on the
the faulty area can be determined.
basis of samples of current and voltage. Integration of the
The difference between calculated fault current, which
Eqs. (7) and (8), for the equally spaced samples (for
correspond to estimated faulty path inductance and active
interval Δt), can be made by trapezoidal rule.
resistance, based on the charge transients and the nearest
For the samples, k + 1 and k + 2 can be made an
lower value of short circuit current, in an application for
estimation of faulty path resistance and inductance:
network analysis, was marked with ΔIk3:
  vk 1  vk    ik  2  ik 1     V1
  I k 3  (13)
    vk  2  vk 1    ik 1  ik    3 R2x   2  L2x
Rx    (9)
  ik 1  ik    ik  2  ik 1    
  i  i   i  i    Distance from the nearest point of discontinuity to the
  k  2 k 1 k 1 k  
fault point is:
75
30
V1
x  (14) 20
3  I k 3  R'2x   2  L'2x
10
faulty phase current (A)

where is: 0
R’Δx – unit value of active resistance for a faulty line -10

L’Δx – unit value of inductance for a faulty line
ω - angular frequency. -20
-30
V. Testing of Methods Based on -40
Charge Transients -50

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16
The examination of method based on charge transients time (s)
was first done on modeled distribution feeders. It was 2.5

confirmed that this approach can operate in the case of 2
different treatment of neutral point. The accuracy of the
1.5
calculation considerable depends on the accuracy of the
filtered current transients

1
given transient frequencies and fault resistance.
The boundary values of fault resistance to which it is 0.5
possible to locate the fault moves up to 250 ohms. Also, 0
the examination was done for the actual fault on the -0.5
network. The advantage of this approach is the ability to -1
use recorded transients from existing devices in the -1.5
network.
-2
It may be currents and voltages recorded by modern
-2.5
numerical protection and power quality analyzers that are 0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16
increasingly encountered in distribution networks. time
4000
V.1. Testing by Ground Fault Simulation

2000
on Modeled Network
faulty phase voltage(V)
Simulations of ground fault for the distribution system 0
shown in the Fig. 3 were made in the program MATLAB

PSB [13]. It was first observed the case of completely -2000
symmetric network, with meaning that all lines and all

loads were completely balanced. In practice this is not -4000
always the case, especially when it comes to load.

-6000
The Eqs. (9), (10) and (11) were applied at the filtered
current and voltage signals in faulty phase. As an
-8000
example, for an isolated network and the distance of the 0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16
fault from supply TS of 9 km, the results were shown in time (s)
the Fig. 6. 500

It should be noted that in the estimation of the distance 400
to the fault it is necessary to observe the time interval in
300
which fault transients exist. In previous studies it can be
filtered voltage transients
200
seen that time interval with fault transients has duration
100
of one to, at most, two cycles.
0
-100
V.2. The Identification of Ground Fault Location
-200
in the Real Network
-300
In this case at the input of fault locator, signals of -400
current and voltage for actual fault in the real network
-500
were connected. 0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16
time
It was downloaded data measured by power quality
analyzer at 10 kV feeder [14]. The sampling frequency Fig. 5. Current and voltage of faulty phase and estimated charge
was 25.6 kHz. The registered values of current and transients. The network is isolated. The moment of the fault inception
voltage were given in the Fig. 7. The ground fault is t = 0 s. The initial angle of faulty phase voltage is θ = 240˚. Fault
occurred in phase C. resistance is Rf = 10 ohms
76
150 4
x 10
1.5
vA
100
transients of current and voltage
1 vB
phase voltages vA, vB, vC (V)

vC
50
0.5
0 0
-50 -0.5
-100 -1
-1.5
-150 -0.1 0 0.1 0.2 0.3 0.4
0 0.005 0.01 0.015 0.02 0.025 0.03
time (s)
time (s)
17 300
iA
16 iB
200
iC
distance to the fault place (km)
currents iA, iB, iC (A)

15
100
14
13 0
12
-100
11
-200
10
9 -300
-0.1 0 0.1 0.2 0.3 0.4
8 time (s)
0 0.005 0.01 0.015 0.02 0.025 0.03
time (s) Fig. 7. Voltages and currents measured
in the supply TS 110/10 kV
Fig. 6. The extracted transients of current and voltage and the distance
to the fault point. The fault was simulated at distance of x = 9 km from 8000
the supply TS. The moment of the fault inception is t = 0 [s]. Fault vC
resistance is Rf = 50 Ω. Θ = 240°. The network is isolated 6000
4000
By examining the waveforms of voltage and current it
voltage vC(V)
can be concluded that, in the first 50 ms, it was occurred 2000
fault with smaller value of fault resistance. Later, the 0

fault resistance increases as can be seen in the voltage
-2000
waveforms.
The diagram shows the first 350 ms disturbance and -4000
50 ms before to the time of the disturbance. The current
-6000
and voltage signals are introduced in Matlab.
For the calculation of the distance to the fault point it -8000
0 0.02 0.04 0.06 0.08 0.1 0.12
is necessary to determine precisely the moment of the time (s)
fault inception. In the diagrams in the Fig. 8 the voltage
and current of the faulty phase, from the moment of the 250
iC
fault occurrence were shown. The extracted voltage 200
transients and estimated distance to the fault point were 150

shown in the Fig. 9. 100
It is registered that the voltage transients have the
current iC(A)
50
highest values at a frequency of 450 Hz (charge
0
transients).
-50
It can be seen that charge transients of voltage reaches
the maximum value for x = 0,02, which corresponds to -100
the estimation of the distance to the fault point of 0,98 -150
km (Fig. 9). Distribution feeder, where the transients are -200

recorded, is derived by conductors Al Fe 95 mm2, in the -250
0 0.02 0.04 0.06 0.08 0.1 0.12
first section with length of 2,76 km from the supply TS. time (s)
For positive and zero sequence reactance are taken the
next values: x1=0,336 ohm/km and x0 =1,58 ohm/km. Fig. 8. Voltage and current of faulty phase
77
80 Herzegovina (JP EP BiH) and ED Tuzla, from 2005 to 2010

[3] J.A.Jardini et.al., „Daily Load Profiles for Residential,
60 X: 0.01973
Commercial and Industrial Low Voltage Consumers“, IEEE
voltage transients (V)
Y: 68.32
40 Trans on Power Delivery, January 2000.

[4] R. Das, M.S. Sachdev, T.S. Sidhu, „A Fault Locator for Radial
20 Subtransmission and Distribution Lines“ IEEE Power
Engineering Society Summer Meeting, Seattle, Washington,
0 USA, July 2000.
[5] Mokhlis, H., Li, H.Y., Khalid, A.R., The application of voltage
-20
sags pattern to locate a faulted section in distribution network,
-40
(1), pp. 173-179.
-60 [6] Mokhlis, H., Bakar, A.H.A., Talib, D.N.A., Mohamad, H., The
improvement of voltage sags pattern approach to locate a fault in
-80 distribution network, (2010) International Review of Electrical
0 0.005 0.01 0.015 0.02 0.025
time (s)
Engineering (IREE), 5 (3), pp. 1159-1164.
[7] M. Lehtonen at. all, „Fault Management in Electrical Distribution
30
Systems“, Final report of the CIRED working group WGO3 Fault
management, Espoo, December 1998.
20
[8] Seppo Hanninen, “Single phase earth faults in high impedance
grounded networks, Characteristics, indication and location”,
10
Doctoral Dissertation, Helsinki University of Technology, Espoo,
X: 0.01988
Finland
fault location (km)
Y: 0.9921
0
[9] Š. Gruhonjić Ferhatbegović „Single phase fault location in
distribution network based on traveling waves“, Doctoral
-10
Dissertation, Faculty of Electrical Engineering and Computing,
May 2012. Zagreb, Croatia
[10] Ferhatbegovic, S.G., Marasic, A., Pavic, I., Single phase fault
-20
distance estimation in medium voltage distribution network based
on traveling waves, (2012) International Review of Electrical
-30
[11] Y.G. Paithankar, „Transmission Network Protection“, Theory and
-40
0 0.005 0.01 0.015 0.02 0.025 Practice, M.D. New York 1997.
time (s) [12] A. G. Phadke, J.S. Thorp, „Computer Relaying for Power
Systems “, John Wiley and sons, New York 1988.
Fig. 9. Voltage and current transients and estimated distance [13] Matlab – SimPowerSystems, User’s Guide, Hydro – Quebec,
to the fault point Mathworks, September 2009.
[14] Documentation of JP EP BiH: "Pilot projects for monitoring
power quality," Sarajevo 2011.
VI. Conclusion
It is possible to estimate fault location by using charge Authors’ information
transients. The estimation accuracy a lot depends on the Šeila Gruhonjić Ferhatbegović was born in
choice of transient frequencies. This means that it is 1969. in Bijeljina, Bosnia and Herzegovina. She
necessary to accurately extract the charge transients of received the B.Sc. and M.Sc. degrees from the
network capacity, which could constitute an aggravating Faculty of Electrical Engineering, University of
Tuzla, in 1994 and 2001, respectively. Doctoral
circumstance for the application of such conceived dissertation “Fault location of single phase fault
locators. Significantly attenuation of fault transients was in distribution network based on traveling
recorded for values of fault resistance greater than 200 waves” was defended at the Faculty of electrical
ohms. The limiting value of fault resistance, for reliably engineering and computing in Zagreb, in 2012.. Since 1994 she has
worked in JP EP BiH. She participated in the development of a number
fault location, ranges up to 250 ohms. of studies, reports and projects in JP EP BiH in the field of distribution
Therefore, the methods based on charge transients are systems, power analysis and power quality. As an author and co-author
not suitable for locating faults with high fault resistance she has published twenty seven papers at conferences and professional
in medium voltage distribution networks. The advantage meetings. At international conferences she has published seven papers.
of these methods is less required sampling frequency. In
these studies it was used sampling frequency of 5 kHz in
computer testing. In comparison with methods based on
traveling waves, it is possible to use data registered by
power quality analyzers, which is a significant
advantage.
References
[1] F. Božuta, „Automatski zaštitni uređaji u elektroenergetskim
postrojenjima“ (eng. “Automatic protection devices in power
plants“), Svjetlost, Sarajevo, 1989.
[2] Technical documentation and statistics of faults in the MV
network, taken from the Electric Power Industry of Bosnia and
78
Model for Aggregation of Plug-in Electric Vehicles Power Demand

for Distribution Systems Planning
Mario A. Ríos1, Camilo A. Rincón2
Abstract – A strategy to estimatethe power demand incrementfor Plug-in Electric Vehicles

(PEV) penetration in residential customers is proposed. This is based on a stochastic model of
arrivals and leavings of vehicles and has as goal to generate diversified load curves for every hour
of a day, and so on, to estimate the power demand by Electric Vehicles. So, the proposed model is
useful to be employed for planning purposes in distribution power systems. This study is divided in
two cases, one of them has a limited number of chargers to be installed at residential customers
and the second option has a free amount of charges related to number of Electric Vehicles; these
represent different types of housing. This tool provides information for planning in distribution
systems, based on specific information on mobility inthe city to study. Copyright © 2015 Praise
Keywords: Distribution Planning, Diversified Load Curves, Monte Carlo Simulation, Plug In
Electric Vehicles, Stochastic Modelling
FAOC Free amount of chargers of batteries Power demand increment by Plug-in Electric Vehicles
LAOC Limited amount of chargers of batteries (PEV) penetration is an essential issue to consider
PEV Plug-in Electric Vehicles indistribution power system planning. This is because it
SOC State of Charge of an electric vehicle battery has several impacts in the power distribution systems
Ck Electric vehicle’s battery capacity (kWh) asthe power demand growth, variations in voltage
∆Ai Expected number of vehicles arriving at the profiles, increment in harmonic pollution, reduction of
parking lot over the total number of vehicles the lifetime of feeding components, power demand
which could park for period i response, and so on [1]-[20]. However, in some cities
∆Li Expected value of percentage of electric does not exist information about the power demand
vehicles leavings parking lot increment due to PEV penetration, which means that an
∆x Distance traveled by an electric vehicle in a estimation of it is needed. Several models of PEV power
journey demand have been made [1]-[6] as a starting point for
F(Ai) Cumulative distribution fuction of the impact assessment. In general, the aspects considered by
interval I of electric vehicles arrivals to the these models are: behavior of PEV drivers, curve of
parking lot charging of PEV batteries and energy consumed by
F(Li) Cumulative distribution fuction of the vehicle due to its usage.
interval I of electric vehicles leaving the The first aspect to consider is the behavior of PEV
parking lot drivers,which defines the time of connection or
Ndays Number of days of simulation disconnection of the Electric Vehicle (EV)to the grid.
nev Number of electric vehicles associated to a However, this is a stochastic variable with high
distribution transformer dispersion because it depends upon several variables as
Nev Maximum number of electric vehicles traffic conditions of the city, labor schedules, electricity
associated to a distribution transformer tariff politics, and so on. That is whyit is assumed in [1]
Oi Parking occupancy at hour i associated to a and quantified as an hourly percentage of the total
distribution transformer amount of vehicles in residential garages. Other authors
∆Oi Variation of the parking occupancy Oiat the [2] defines this behavior with a cumulative distribution
interval of time i function of arriving time since 17:00-24:00 hh:mm in the
P(t,nev) Power demand consumption at period t by day. Another option is to assume the time of connection
nev electric vehicles associated to a of vehicles based on empirical experience on the city in
distribution transformer study, as in Malaysia’s case [3].
rdk Rate of discharge of the battery (kWh/km) The second issue to consideris the curve of charging
of PEV batteries.
79
M. A. Ríos, C. A. Rincón
This is related to the type of charger used and contains According to that, this paper proposes to estimate the
information about the power required by batteries. power demand incrementrelated to the penetration of
One option to include that information is considering electric vehicles in urban markets based on a stochastic
it with a discretized model, with different linear relations model.
between time and energy for intervals of time[4].On the Section II presents the model as the considerations
other hand, [11]-[13] present PEV battery charger models taken in count and section III y IV presents the results of
based on its circuital model. this simulation for two cases: Limited amount of
Also, information about the amount of energy chargers (LAOC) and free amount of chargers (FAOC).
consumed from the battery should be included. This is Finally, conclusions of this work are presented.
related to the discharge rate of the battery and the
distance traveled by the vehicle. The first one is defined
on the data sheet of the EV battery, and the second one is II. Hourly Diversified Load Curves Model
related to traffic, distances in the city and the behavior of The model proposed is based on hourly diversified
vehicles. In some references, statistical data of the city in load curves associated with the power demand to the
study is included[6], but in other cases a distribution penetration of electric vehicles in urban markets. It is
function of distance traveled is assumed and the traffic is because these curves are tools typically used in power
ignored[4]. distribution planning[24].
With the estimation of the power consumption of the However, statistical data of EV usage are null in cities
EVssome impact assessments could be made. One where the utilization of EV is starting. So, it is necessary
studyis about the lifetime of feed components as the to replace the statistical data with a simulation-based
transformers are also impacted by the penetration of PEV projection model that is, obviously, based on the
[7]. This impact in the lifetime could be reduced by the characteristics of electric vehicle usage and the
installation of rooftop photovoltaics [8] as well as recharging process of the batteries.
coordinated charging of PEV by the utility [9]. Also, the model proposed applies to multifamily
The PEV penetration also impacts on electricity dwellings and individual residences fed by the same
demand, prices and generation dispatch. For instance, MV/LV transformer, where the residents own a specific
some simulations and assessments had been done as the number of EVs.
one presented in [10]. According to that, the determination of the diversified
Reference [14] considers the time of use electricity load curves requires punctual information of total hourly
prices as an additional aspect to be considered in the PEV power demand for different number of electric vehicles
demand estimation,because it makes changes in the (nev) owned by the residents in a specific area. With that,
behavior of PEV drivers. the solution of the optimization problem described in (1)
Furthermore, in [4] a discrete event simulation was returns the constants (C1, C2, B) which describe the
implemented in order to study the PEV penetration diversified load curves. This optimization problem
impacts in power distribution systems, for the specific should be solved for every hour(t) in a day:
case of commercial buildings. A similar analysis is
presented in [15], but this assessment was focused in 2
unbalanced voltages caused by uneven distribution of  1 

Nev  c  t   c  t  n
PEV penetration among the phases. Besides, EPRI has  1 2 ev 
performed this kind of analysis with the purpose is to min 
C1  t  ,C2  t  ,B  t   P  ev  
t,n
(1)
nev 1
identify, define, and calculate the impact to particular   B  t   
utility distribution system architectures considering total  nev 
PEV penetration levels as well as localized concentration
[16]. Other related studies are presented in [17]-[20]. The optimization problem minimizes a quadratic error
According to the worksexposed above, the mitigation function between the hourly diversified load curve and
of the impact of PEV penetration plays an important role the average power demand by vehicle, and it has the
in power distribution planning, that is why several decision variables (C1, C2 and B).
studies in this area had been made. As an example, [21] Therefore, an estimation of the total power demand in
presents an algorithm to accommodate high penetration hour t for a numberof electric vehicles nev is required. For
of PEVs in distribution networks, which is based on this estimation, nev varies from 1 to a maximum number
allocating distributed generation units.This algorithmcan of electric vehicles Nev. Moreover, it is generated based
help the local distribution companies to better assess the on the Mote Carlo simulation as shown in Fig. 1.
PEV impacts on their systems and mitigate them. An The estimation process starts with the initialization of
additional point of view is the consideration of multi- the average hourly total power demand matrix (P),
terminal low voltage direct current for the improvement assigning a zero to every cell. This matrix contains
of PEV integration in an existing power distribution information about the aggregated power demand at hour t
network [22]. due to the presence of nevelectric vehicles, and it is an
Besides, a fuzzy coordination algorithm for charging average value of the realizations in Monte Carlo
PEV is proposed in [23]. simulation.
80
average percentage of occupied parking lot spaces of the

total amount of parking lot spaces in a residency at the
time to (e. g. if occupation O at time t is 25% and the
parking lot has 100 spaces, the eexpected
xpected amount of
occupied parking lot spaces is 25).
Fig. 2. Flow chartfor

chartfor theassignment
theassignment of SOC for EVs
This vector quantifies the flow of car drivers on the

city in study. Moreover, due to the parking lot occupancy
profile characteristics, it is ne necessary
cessary to discretize the
Fig. 1. Flow diagram for the estimation of the hourly total
power demand for vehicle penetration
time in intervals of time T, for a total of 24/T intervals.
According to the previous paragraph, the SOC also
So, the matrix P has 24 rows (by the number of hours depends on the characteristics of the batteries (capacity,
in a day) and Nevcolumns (by the maximum number of rate of discharge), as in the dynamics of charge and
electric vehicles). discharg
dischargee during the day, so that the proposed estimation
By the definition of Matrix P,, the determination of includes these.
each of its columns requires one Monte Carlo simulation. The determination of the battery characteristics
So, the nev-th
thcolumnis
columnis defined for nev electric vehicles
vehicles, depends on the size of the EV. So, the first step of this
and it requires a simulation with this number of EVs. methodology is the assignment of one of the four types
Each Monte Carlo simulation starts with the of EV to each EV. This alloc
allocation
ation depends on the market
initialization of the hourly total power demand matrix Pa share of the EV types, used as a probability distribution
as a zero column vector of 24 rows (one row by one hour function for the type of EV. As previously commented,
during the day).
day).At
At the end of the simulation, this matrix the
the definition of each EV type is based on the size of the
contains many rows as the number of realizations of the vehicle and the four types of EV are: Compact Sedan,
Monte Carlo simulation, and each row is one realization Mid Size Sedan, Mid Size Sport Utility Vehicle (SUV)
result of the total power demand at hour t due to the and Full Size SUV. The Table I presents the structure of
presence of nevelectric vehicles. Then, the i variable is the market share of the EV types, where SMk is the
initialized as 1, it defines the number of the realization. market share for the EV ttype
ype kk.
After the initialization, each realization starts with the After that, the variable j is initialized to 1, this variable
assignment of the state of charge (SOC) for each vehicle. is the number of simulated da days
ys and it varies from 1 until
The aim of this process is the determination a row the total number of days to simulate Ndays (number
vector of nevcolumns whichwhich contains the SOC of every defined based on a previous study of stabilization of the
EV battery at the beginning of a day. This assignation SOC average). Then, the vector SOC is initialized. This
process is presented in Fig. 2.. vector contains the SOC value of every EV at the
The aassignment
ssignment of SOC value at the beginning of the beginning of the day (i.e. the kk-thth column of this vector is
day depends on the probability distribution function of the SOC of tthehe k-th
k th electric vehicle, and initialization of
SOC, but it is unknown. Therefore,
Therefore, this paper proposes a this vector is the assignation of a random variab variable
le
methodology to estimate SOC value based on the uniformly distributed from 0.2 to 0.8 to each
behavior of car drivers (any car type) and the column.
column.Note
Note that the probability density function
characteristics of charge and discharge of the EV battery
battery. initially defined should not affect the final result of the
This algorithm has as a main input the parking lot methodology. The next stage is the determination of the
occupancy profile, which is a vector containing the leaving and arriving time of each EV.
leaving
Copyright © 201
2015 Praise Worthy Prize S.r.l. - All rights reserved International Review of Electrical Engineering, Vol. 10,, N. 1
81
TABLE I In (5), the variable A is the expected percentage of

PERCENTAGE OF EVERY TYPE OF VEHICLES
vehicles which not arrivedthe parking lot during the
Vehicle Type Percentage (%)
Compact Sedan SM1
day,and is calculated as follows:
Mid-Size Sedan SM2
Mid-Size SUV SM3 24 / T
Full-size SUV SM4 A  1  Oi I0,inf   Oi  (6)
i 1
This is necessary to determine the start time of the
battery charging and also to include the car driver where the indicator function I(a,b)(x) is defined as:
behavior in the SOC variations during the day.
The leaving time and the arriving time are determined 1,a  x  b
based on the parking occupancy profile as presented I  a,b   x    (7)
below. Firstly, variation ∆Oi of the occupancy Oi at the 0,otherwise
interval of time iis calculated as:
Note that (6) defines A as the complementary event of
O1  O1  O24 / T (2) the arriving event. Finally, the CDF of the interval of
arrivals is defined as:
Oi  Oi  Oi 1 ,i  2,3,..., 24 / T (3) A
FA  0   (8)
NA
This variation is positive when Oi is greater than Oi-1,
and it means an increment of the occupancy of the
Ai
parking lot, and hence an increment of the number of FA  i   FA  i  1  ; i  1, 2 ,3,..., 24 / T (9)
vehicles parked. Conversely, a negative variation NA
impliesa decrease in the number of vehicles parked. So,
time intervals with positive changes result in arrivals of As previously commented, the aim of this process is
vehicles, while time intervals with negative variations the assignment of an arrival time to every vehicle. For
translate into leavingsof vehicles. this, an interval of time i ( i>0) is defined based on FA(i)
Based on the calculation of variations, the percentages and the arrival time t is uniformly distributed in the
of arrivals ∆Ai are defined. Those percentages are the interval from (i-1)T to iT. However, if the assigned
expected number of vehicles arriving at the parking lot interval is i=0, no arrival time is assigned.
over the total number of vehicles which could park According to the definition of ∆Ai, the percentage of
there(the total number of vehicles is equal to the total leavings∆Liis the expected number of vehicles leaving
amount of spaces in the parking lot). This percentage of the parking lot over the total number of vehicles which
arrivals ∆Ai equals ∆Oi if this number is positive, and could park there. This percentage of leavings equals-∆Oi
zero otherwise. if the variation of the occupancy (∆Oi) is negative, and
In addition, an error function U(0,dV) is added to this zero otherwise. The error function U(0,dV)is also added
expected value of percentage of arrivals to avoid a zero to this expected value of percentage of leavings, thus:
in the number of arrivals in intervals where ∆Oi is
negative (note that even when occupancy decreases,  Oi  U  0,dV  , Oi  0
some cars could arrive to the parking lot). The error Li   (10)
function is defined as a random variable uniformly U  0 ,dV  , Oi  0
distributed from 0 to a maximum error value dV.
This definition is summarized as: The normalization factor NL is also needed and
defined as:
Oi  U  0,dV  , Oi  0
Ai   (4) 24 / T
U  0 ,dV  , Oi  0 NL  L   Li (11)
i 1
After these calculations, a cumulative distribution

function (CDF) for the interval of time in which a car where L is:
will arrive to the parking lot can be built.
24 / T
In order to build this CDF, it is necessary a
normalization to the percentages ∆Ai because of the L  1  Oi I  0 ,inf   Oi  (12)
i 1
errors added, therefore anormalization factor NA will
divide each ∆Ai as:
and the CDF of the interval of leavings is given by:
24 / T
NA  A   Ai (5)
FL  0  
L
(13)
i 1 NL
82
Li As in the type 4, a battery discharged is assigned at the

FL  i   FL  i  1  ; i  1, 2 ,3,..., 24 / T (14) leaving time and a battery recharge at the arriving time.
NL
The last step of one day simulation is the SOC
As in the case of the arriving times, the aim of this actualization. It is based on actions assigned to the driver
process is the assignment of the leaving time to every in Table II and an additional one. So, the SOC
vehicle. So, the interval of time i ( i>0) is defined based actualization is performed based on threefunctions:
on FL(i) and the leaving time t is uniformly distributed in Discharging the battery: The amount of energy
the interval from (i-1)T to iT. consumed from the battery depends on several variables
Also, if the intervalassigned is i=0, the vehicle does related to the use of the vehicle as well as some technical
not leave the parking lot. After that, the behavior of the features of the EV. However, for the calculation of the
EV driver is classified into 5 types. Each of them is energy consumed, this paper only includes the variables:
related to the times of arrival and leaving and it assigns Distance traveled by the vehicle (∆x in km), the rate of
an action related to the charging or discharging of the discharge of the battery (rdk in kWh/km) as well as its
battery to each EV driver. The Table II presents each of capacity (Ck in kWh). So, the relation between the
the types and the associated actions. variation of the SOC and the three variables mentioned
As seen in the Table II, the classification of driver before is given by:
behavior is based on the time of arrival and leaving. For
type 1, no time of arrival or leaving is assigned; that is, rd k
SOC   x (15)
the vehicle will not move from the parking, so the SOC Ck
is not updated in any way.
In (15), the subscript k refers to the dependence
TABLE II
CLASSIFICATION OF ELECTRIC VEHICLE DRIVERS between this variable and the k type of EV.
Leaving Arriving Recharging the battery outside the parking lot: It is
Type Recharge Discharge
Time time assumed that if the SOC of a vehicle is low (under 0,2) at
Not Not the arriving time, the driver decides to recharge the
1 No No
assigned assigned battery before arriving to its house; then, if the SOC is
Atthe At the
Not low it is set at 0,8 at the arriving time.
2 Assigned arriving beginning of
assigned Recharging the battery on the parking lot: It occurs at
time the day
Not Atthe leaving the arriving time if the parking lot space is available.
3 Assigned No
assigned time Besides, the SOC value is set at 1.
Atthe
Assigned Assigned Atthe leaving Finally, the actualized SOC in day j is the entry SOC
4 arriving
first second time of the day j+1, until the number of simulations
time
Atthe performed reach Ndays. As a result of this simulation,
Assigned Assigned Atthe leaving
5 arriving the SOC value for each vehicle is randomly assigned
second first time
time
based on the EV driver behavior, and it is the actualized
SOC in the last day simulated.
On the other hand, type 2 drivers only have assigned
According to Fig. 1 the next step is the calculation of
an arrival time, which means that those vehicles spend
thevector hourly total power demand.This column vector
the previous night outside the parking lot. For this
has 24 rows and the m-th element is one simulation result
reason, the battery is discharged at the beginning of the
of the total power demand in the parking lot transformer
day (this is due to the distance traveled before the
at the hour m, if there are nev electric vehicles.
arriving time), and also, a vehicle in this group starts the
This calculation starts withthe assignment of the
battery recharging at the arrival time.
power demanded by each EV, for this, a vector with 5
Furthermore, type 3 drivers only have assigned a
power elements and 6 time elements are generated based
leaving time, which means that they leave the parking lot
on the curves presented in [4] of the charger response to
(usually in the morning) and do not return the same
each EV type. These curves are an approximation of the
day.This type of drivers does not recharge the battery
EV battery charger consumption and has the structure
during the day, but the battery is discharged at the
presented in Fig. 3, where thepower constants dependson
leaving time (it is due to the distance traveled during the
the type of EV. These constants are presented in Table
day). Type 4 drivers have assigned one leaving time and
III. However, the time constants also depend on the SOC,
one arriving time, and the leaving event occurs before the
and these constants are added to the arriving time (when
arriving one. It is important to note that this is the most
the EV recharges the battery on the parking lot)
common type of drivers, as an example the driver leaves
determining the power demand of a single vehicle during
the parking lot at 7:00 a.m. and returns at 6:00 p.m.. For
a day.
this type of driversa battery discharge is included at the
It is important to highlight that the battery recharging
leaving time and a battery recharge at the arriving time.
time could be greater than the remaining time after the
Finally, type 5 drivers also have one leaving time and
car arrives (e.g. a car arrives at 23:30 hh: mm and the
one arriving time, but the arriving event occurs before the
time required by the battery charger to fully recharge the
leaving one.
battery is 1 hour and a half).
83
In these cases the power demand curve is divided in III. Limited Amount of Chargers (LAOC)
two, the first part is assigned from the arriving time to
the end of the day, and the second part starts at the As it was presented above, one application of this
beginning of the same day and finishes when the SOC is stochastic model has as goal to estimate a load profile
1. due to EV charging stations in multifamily dwellings.
However, since the amount of free spaces for energy
meters is very limited (one free space in some cases)it is
recommendedto measurethe energy consumption of the
battery charger with only one meter associated with the
communal account of the multifamily dwelling.
Based on this, an additional recommendation is to
install charging stations in communal parking spaces,
which are also limited. So, for the case of multifamily
dwellings, the amount of chargers is limited; and, this
restriction should be included to determine if a vehicle
could charge when it arrives in the parking.
III.1. Input Data

Occupancy profile ofresidential customersin the city
Fig. 3. Power demand of a battery charger in study is the most important entry to this model. For
this work, an empirical estimation of it was used and is
TABLE III presented (see Fig. 4).
BATTERY CHARGER CHARACTERISTICS [6]
Vehicle Modeled
P1 P2 P3 P3 P5
Type vehicle
Compact Mitsubishii-
3.20 2.60 1.82 1.17 0.96
Sedan MiEV
Mid-Size
Nissan LEAF 6.60 5.33 3.73 2.42 1.92
Sedan
Mid-Size
BYDe6 7.21 5.82 4.07 2.64 2.12
SUV
Full-Size
BYDe6 7.21 5.82 4.07 2.64 2.12
SUV
All values in kW
After the assignment of the power curves to each EV,

a power aggregation is done by adding the power
demand of all of the EVs during each interval of time T1
(in hours) which discretizes the model. With that, an Fig. 4. Percentage of occupancy profile in residential garages.
average power of the intervals is calculated by hour for
all of the 24 hours, each of these averages defines the It indicates which percentage of residential parking lot
rows of the vector hourly total power demand. So, this places are occupied by vehicles and contains information
vector is constructed. related to the behavior of vehicle drivers. Add to it, the
Then, the i-th column of the Pa matrix is defined as deviation of variation in occupancy profile is needed, to
the hourly total power demand vector of the i-th determine the value of the error presented above.
iteration, actualizing this matrix with each iteration. In this case it was assumed as 0,01. Another important
The end of each Monte Carlo realization is the input is the share market for all of the types of vehicles,
stopping criterion, in this case the simulation finishes if which is described in Table IV.
the maximum number of iterations (l) is reached or if the
coefficients of variation (CV) of the power of every hour TABLE IV
PERCENTAGE OF EVERY TYPE OF VEHICLES[5]
are less than 0.005.
Vehicle Type Percentage (%)
With the Monte Carlo simulation the Pa matrix for nev Compact Sedan 51.5
EVs is constructed, and the average value ofall of the Mid-Size Sedan 10.3
realizations (columns) in hour m (m-th row) is the final Mid-Size SUV 23.0
estimation of the aggregated power demand in this hour. Full-size SUV 15.2
So a column vector of 24 elements is constructed for
this number of EVs, and is the nev-th of the P matrix. So, Also, the characteristics of the battery and the rate of
with the simulation of nevEVs from 1 to Nev the P matrix discharge for each vehicle type are presented in Table V.
is constructed as well as the diversified load curves. On the other hand, the rate of discharge was used as
0.16kWh/km [4] for all types of EVs.
84
TABLE V
BATTERY CHARACTERISTICS
Vehicle Type Modeled vehicle Capacity (kWh)
Compact Sedan Mitsubishi i-MiEV 16
Mid-Size Sedan Nissan LEAF 24
Mid-Size SUV BYDe6 61.4
Full-Size SUV BYDe6 61.4
Finally,the ratio between number of parking slots and

number of PEVs was assumed to be 0,1 (i.e. 1 parking
slot by 10 vehicles).
III.2. Results
With the input data describedabove, diversified
demand curves were generated for each hour in a day. As Fig. 5. Hourly diversified demand curve (22:00-21:00)
an example, diversified demand curve for time interval
20:00-21:00 is presented in Fig. 5.
Stochastic results from the Monte Carlo simulation are
presented as circles for each amount of electric vehicles
between 1 and 100 and diversified demand curve as a
continuous line. It shows the expected behavior for this
type of curves, a high power average for low amount of
users (EV) and a tendency to keep constant with high
number of them.
This typical tendency is evident for some intervals of
time, including (20:00-21:00), and Fig. 6 presents its
behavior by number of EV. Moreover, those time
intervals are peak hours of demand, and interval (19:00-
20:00) is the highestduring the day, it reachesan average
of 0.623kW/EV for high amount of vehicles. Fig. 6. Hourly diversified demand curves– limited number of chargers
On the other time intervals, a constant behavior for
any number of vehicles is observed; Table VI
summarizes those constant values. IV. Free Amount of Chargers (FAOC)
As expected, time intervals of residential parking low
The second type of housing in study includes
occupancy represent low electric power demand, as is the
individual houses connected to the same MV/LV
case of intervals between 9 a.m. to 1 p.m.. On the other
transformer. In this case, it is assumed that every house
hand, intervals in the range 4-9 a.m. has low power
has the possibility to install at least EV charger per
demand because EV end to charge around 1-3 a.m.. It is
electric vehicle. So, the number of chargers is limited just
also observed that high demand intervals (7p.m.-10p.m.)
for the number of vehicles, and any vehicle could be
has the expected behavior of a diversified curve.
connected at any time.
TABLE VI The input data used in this case is the same as in the
PEAK POWER DEMAND FOR LAOC case of limited amount of chargerscase;but, the ratio
Time interval Ppeak/VE between the amount of chargers and vehicles is 1 (i.e. the
0:00-1:00 0,1332 number of vehicles is equal to the number chargers).
1:00-2:00 0,0988
2:00-3:00 0,0789
As in the other case, hourly diversified demand curves
3:00-4:00 0,0704 estimatedhad the expected behavior for some time
4:00-5:00 0,0622 intervals, and the other ones got constant values.Intervals
5:00-6:00 0,0549 withconstant values of power demand are presented in
6:00-7:00 0,0528
7:00-8:00 0,0525 Table VII, and the rest of them is presented in Fig. 7.
8:00-9:00 0,0492 It is important to note that for a free amount of
9:00-10:00 0,0440 chargers the electric power demand in intervals with
10:00-11:00 0,0355 most arriving events (7p.m.-10p.m) is higher than the
11:00-12:00 0,0317
12:00-13:00 0,0691 case of limited amount of chargers. This is presented
13:00-14:00 0,1395 because all of the vehicles which reach its parking space
14:00-15:00 0,1764 could connect to the grid; it was not limited by the
15:00-16:00 0,1700 vehicles already connected. Because the possibility of
16:00-17:00 0,2079
17:00-18:00 0,2745
free connection, vehicles have a higher average state of
18:00-19:00 0,3904 charger in stable conditions than the LAOC case. So,
23:00-24:00 0,1182 time of charging for batteries tends to be lower in FAOC
85
than in LAOC, and the power requirement is lower in the In addition, the model has not considered batteries’
latest interval in night (23:00-24:00) and first intervals managing system; that could be included adjusting the
during the day (00:00-03:00). LAOC and the FAOC models.
TABLE VII
PEAK POWER DEMAND FOR FAOC Acknowledgements
Time interval Ppeak/VE
1:00-2:00 0,0603 This work was supported in part by COLCIENCIAS,
2:00-3:00 0,0413 CODENSA, EMGESA Grant 0197-2012.
3:00-4:00 0,0318
4:00-5:00 0,0256
5:00-6:00 0,0203
6:00-7:00 0,0193
References
7:00-8:00 0,0219 [1] M.de Nigris, I.Gianinoni, S.Grillo, S.Massucco, F. Silvestro,
8:00-9:00 0,0227 Impact evaluation of plug-in electric vehicles (PEV) on electric
9:00-10:00 0,0232 distribution networks,14th International Conference on
11:00-12:00 0,0246 Harmonics and Quality of Power (ICHQP), Sept.26-29, 2010.
14:00-15:00 0,1664 [2] H. Wang, Q. Song, L. Zhang,F. Wen, J. Huang, Load
15:00-16:00 0,1551 characteristics of electric vehicles in charging and discharging
16:00-17:00 0,2005 states and impacts on distribution systems,International
18:00-19:00 0,4019 Conference onSustainable Power Generation and Supply
19:00-20:00 0,8140 (SUPERGEN 2012,) Sept. 8-9, 2012.
20:00-21:00 0,7277 [3] C.H. Tie,C.K. Gan, K.A. Ibrahim, The impact of electric vehicle
21:00-22:00 0,3391 charging on a residential low voltage distribution network in
22:00-23:00 0,1807 Malaysia, 2014 IEEE Innovative Smart Grid Technologies - Asia
(ISGT Asia), pp.272-277, May 20-23, 2014.
[4] M.A.Rios, N.M.Pena,G.A. Ramos, N. Achury, L.E. Muñoz,
Evaluation of the impact of Plug-in Electric Vehicles on the load
profile of commercial buildings: a stochastic model, 2014 IEEE
PES Transmission & Distribution Conference and Exposition -
Latin America, Sept. 10-13,2014.
[5] Malik, F.H., Lehtonen, M., SaarijÃ¤rvi, E., Safdarian, A., A
feasibility study of fast charging infrastructure for EVs on
highways, (2014) International Review of Electrical Engineering
(IREE), 9 (2), pp. 341-350.
[6] Alharbi, Abdulelah Yousef, Impact of plug in electric vehicle
battery charging on a distribution system based on real-time
digital simulator, M.Sc Dissertation, Dept. Elect. Eng., The
University of Tennessee at Chattanooga, Chattanooga, TE, 2013.
[7] Q. Gong, S. Midlam-Mohler, V. Marano, G. Rizzoni, PEV
charging impact on residential distribution transformer life,2011
IEEE Energytech, May 25-26 , 2011.
[8] T.J.Geiles, S. Islam, Impact of PEV charging and rooftop PV
penetration on distribution transformer life,2013 IEEEPower and
Energy Society General Meeting (PES), July21-25, 2013.
[9] M.A.S.Masoum,P.S. Moses,S.Hajforoosh, Distribution
Fig. 7. Hourly diversified demand curve – free number of chargers transformer stress in smart grid with coordinated charging of
Plug-In Electric Vehicles,2012 IEEE PES Innovative Smart Grid
Technologies (ISGT), Jan. 16-20, 2012.
V. Conclusion [10] J. Villar, C.A. Diaz, J. Arnau, F.A. Campos, Impact of plug-in-
electric vehicles penetration on electricity demand, prices and
A stochastic model to estimate the power demand thermal generation dispatch,2012 9th International Conference on
increment due to PEV penetration in urban markets is the European Energy Market (EEM), May 10-12,2012.
[11] P.S.Priambodo, W.Purnomo, A. Subiantoro, A. Muis, F. Yusivar,
proposed. This estimation is based on information about Transformerless high voltage and controllable current battery
the behaviors of car drivers in the city to study, the charger for e-car,2013 Joint International Conference on Rural
electric properties of the battery chargers for different Information & Communication Technology and Electric-Vehicle
sizes of vehicles andthe energy consumption from the Technology (rICT&ICeV-T), Nov. 26-28, 2013.
[12] Hu, Y., Liu, H., Damage analysis of lithium iron phosphate
battery due to the vehicle usage. batteries for pure electric vehicles, (2013) International Review of
This estimation represents a tool for distribution Electrical Engineering (IREE), 8 (2), pp. 722-728.
planners to study the impacts of the power increase due [13] M.P.Puentes, G. Ramos, Level 2 bidirectional charger for small
to the connection of plug-in electric vehicle chargers in electric vehicles: Topologies comparison,2013 Workshop on
Power Electronics and Power Quality Applications (PEPQA),
residency markets in two different cases, multifamily July 6-7, 2013.
dwellings and individual housing. [14] A.S.Bin Humayd, K. Bhattacharya, Impact of PEV penetration on
The proposed model has been developed considering distribution system planning considering time-of-use electricity
only residential customers associated to a distribution prices,2014 IEEE PES General Meeting Conference &
Exposition, July27-31, 2014.
transformer. So, the model requires further work to [15] S.K.Bunga, A.H.Eltom, N.Sisworahardjo, Impact of Plug-in
include charging EV in commercial places associated to Electric Vehicle battery charging on a distribution system, 2014
distribution transformers. IEEE Industry Applications Society Annual Meeting, Oct.5-9,
2014.
86
[16] A. Maitra, J. Taylor, D. Brooks, M. Alexander, M. Duvall, Authors’ information

Integrating plug-in- electric vehicles with the distribution system,
20th International Conference and Exhibition on Electricity 1
Department of Electrical Engineering and Electronics at the School of
Distribution CIRED 2009, June8-11, 2009. Engineering at Universidad de Los Andes, Bogotá, D.C., Colombia.
[17] C.C. Mendoza, A.M. Quintero, F. Santamaria, J.A. E-mail: mrios@uniandes.edu.co
Alarcon,Impact analysis of Plug-in Electric Vehicles on an
electric distribution network,2014 IEEE PESLatin America 2
Department of Electrical Engineering at Universidad de Los Andes,
Transmission & Distribution Conference and Exposition, Sept. Bogotá, Colombia.
10-13,2014. E-mail: ca.rincon978@uniandes.edu.co
[18] L. Pieltain, T. Gómez, R. Cossent, C. Mateo, P. Frías, Assessment
of the Impact of Plug-in Electric Vehicles on Distribution Mario Alberto Ríos, he is an electrical
Networks, IEEE Transactions onPower Systems, vol.26, n.1, Feb. engineer (1991) and M.Sc in electrical
2011, pp.206 – 213. engineering (1992) both from Universidad de
[19] Alahaivala, A., Saarijarvi, E., Lehtonen, M., Modeling electric Los Andes, Bogotá, Colombia. He has a Ph.D.
vehicle charging flexibility for the maintaining of power balance, Degree in electrical engineering from both
(2013) International Review of Electrical Engineering (IREE), 8 INPG-LEG (1998), France and Universidad de
(6), pp. 1759-1769. Los Andes (1998). Currently, he is
[20] S.V.Chakraborty,S.K. Shukla, J. Thorp, Evaluating the grid FullProfessor at the Department of Electrical
impact of plug-in electric vehicles using dynamic commuting Engineering at School of Engineering at Universidad de Los Andes
profiles,2013 IEEE Grenoble PowerTech, June 16-20,2013. (Bogotá, Colombia)
[21] M.F.Shaabar,E.F. El-Saadany, Accommodating high penetration E-mail: mrios@uniandes.edu.co
of PEV in distribution networks,2013 IEEE Electrical Power &
Energy Conference (EPEC), Aug. 21-23,2013. Camilo Rincón, He is an electrical and electronics engineer (2013)
[22] W. Pei, W. Deng, X. Zhang, H. Qu, K. Sheng, Potential of Using from Universidad de los Andes, Colombia. Currently, he is a studentof
Multi-Terminal LVDC to Improve Plug-In Electric Vehicles the M.Sc. in electrical engineering progrram at the same university. He
Integration in an Existing Distribution Network,IEEE has worked for Universidad de los Andessince 2013 as a graduate
Transactions on Industrial Electronics, for publishing, 2015. research assistant for the Department of Electrical Engineering at
[23] A. Masoum, S. Deilami, A. Abu-Siada, M. Masoum, Fuzzy School of Engineering(Bogotá, Colombia.
Approach for Online Coordination of Plug-In Electric Vehicle E-mail: ca.rincon978@uniandes.edu.co
Charging in Smart Grid,IEEE Transactions onSustainable
Energy, for publishing, 2015.
[24] H. Lee, Power Distribution Planning Reference Book (Marcel
Dekker, Inc., 2004).
87
The Effect of Distributed Generation Type and Location Constraints

on the Solution of the Allocation Algorithm
R. Prenc, N. Bogunović, A. Cuculić
Abstract – Modern planning of power distribution systems faces significant changes over the
last few decades due to the massive introduction of distributed generation units. In this paper the
authors discuss the inclusion of location constraints for optimal allocation of DG units. The goal
function will be the minimization of cumulative average daily active power losses. Four types of
DG units will be considered; a solar park, a wind farm, a power station that does not depend on
an intermittent primary energy source and a small hydro unit. Each DG unit and network load will
be modeled with its own characteristic average daily power production or consumption curve. The
network load will consist of residential and industrial consumers. The problem will be solved
using genetic algorithm and realized in Matlab programming environment. Copyright © 2015
Praise Worthy Prize S.r.l. - All rights reserved.
Keywords: DG Allocation, Location Constraints, Loss Minimization, Genetic Algorithm
I. Introduction Popović et al. have developed an algorithm that

addresses placement and sizing of DG units with the
Over the last decade there has been an accelerated addition of recloser placement for improving reliability
increase in the integration of distributed generation (DG) indices [1]. Kaluđer et al. have solved the allocation
units into power systems of many countries in the world. problem using genetic algorithm for the minimization of
Distributed generation includes intermittent renewable power losses [2].
energy sources (usually solar, wind and small hydro A multi-objective allocation problem for minimization
power plants) and power stations which do not depend on of power losses and reliability improvement was
an intermittent primary energy source. The latter can run presented in [3]. The simultaneous optimization of power
on fossil or renewable fuel (like biogas or biomass). losses, voltage profile and voltage stability was
When the feed-in tariffs were introduced to encourage demonstrated in [4]. The authors in [5] focused on the
the integration of DG sources, the private sector joined maximization of financial incentives and loss savings for
the market and greatly boosted the development of the DSO, with the integration of DG.
different distributed generation technologies. However, In the past the DG allocation problem was solved
the distribution power networks were originally built for using mostly constant values of network load and DG
unidirectional power flows, and their obsolete passive power production as input data [6]-[13]. In that way the
design cannot be momentarily abandoned. As a result, daily power consumption/production fluctuation was
the Distribution System Operator (DSO), which is neglected, seemingly making the nominal produced
responsible for developing, operating and ensuring the power coincide with peak network load. However, DG
maintenance of the distribution network in a given area, units (for example solar parks) do not produce peak
must regulate the inrush of DG units. The DSO's power in the same time as the network loads consume
planning departments face serious challenges in peak power. As a result, DG allocation will be non-
connecting DG units, because they must find balance optimal. Renewable sources like sun, wind and water
between the various requests from Independent Power flow cannot guarantee constant power output during the
Producers (IPPs) and the network existing infrastructure. 24 hour period, and their intermittent nature must be
As a response to the aforementioned challenges, a new taken into consideration while planning their integration
concept has appeared simultaneously with the in the network.
introduction of DG units and is called the distributed The preliminary measurements of wind, solar and
generation allocation problem. The allocation problem water flow potential for certain locations in relatively
refers to the optimal choice of size and location of DG long time periods must be used for allocation of those
units to be connected to the distribution network without types of DG units. Finally, all the network consumers
the need to upgrade or expand its current design. In that have specific daily power consumption curves
way the planner guarantees that the DG units will (residential, industrial customers), which greatly
minimally alter the network and that the connection determine the optimal allocation of new DG units, and
charges for IPPs will also remain minimal. thus must be also taken into consideration.
88
The Croatian Grid Code for the electric power system III. Network Model and
from 2006 defined that on medium voltage (MV) Power Flow Solution
network only DG units with output power range between
500 kW and 10 MW can be connected, and in this paper The main difference between the connection of a load
the authors will present an algorithm for individual and a DG unit is that when connecting a DG unit the
allocation of four types of DG units. feeder input current will drop and in the same time some
The DG units considered will be a solar park, a wind of the network node voltages will simultaneously rise.
farm, a power station that does not depend on an As a result the network will change its operation from
intermittent primary energy source and a small hydro passive to an active network. Although the lines closer to
unit. the HV/MV Primary Substation will now be unloaded,
the lines near the DG units (usually having lesser cross
section because of passive network design) may become
II. The Definition of the Distributed overloaded. Also, the power flows may reverse and the
Generation Allocation Problem node voltages may increase to extreme values.
An example of a DG connection to an existing
The goal in this paper will be to minimize cumulative distribution network feeder is displayed in Fig. 1.
average daily active power losses for all the lines in the
radial network in discrete 24 hour intervals, as displayed
in the following equation:
24 lines 24 lines
  plossi, j  min    3  Ilossi, j 
2
f  min  Ri (1)
j 1 i 1 j 1 i 1
In the above equation lines represents the total number

of lines in the network under study, Ri denotes series
resistance of line i, plossi,j represents active power loss of
ith line during the jth hour and Ilossi,i represents current
which flows through serial resistance Ri during the jth Fig. 1. Model of a distribution network feeder
with a connected distributed generation unit
hour.
The losses will be observed for an average day in the The load flow used in this paper is based on the
year, which relatively well describes the losses for all backward/forward method. It is specialized for solving
365 (366) days. distribution networks with very long feeders and a large
The main equation must satisfy the thermal, voltage R/X ratio, with the voltages on end nodes having much
and location constraints at all times: lower values than on the feeding MV buses of HV/MV
Primary Substations [14]. The load flow algorithm
Si  Snom,i , i  1,...,lines (2) consists of two sets of equations.
The first, called the backward sweep, calculates all the
U min,n  U n  U max,n , n  1,...,nodes (3) branch currents beginning from the last node of the
feeder and ending at the first node of the feeder. The
second sweep, called the forward sweep, calculates all
locDG ,k  1  locDG ,k  0 , k  1,...,noDG (4) the node voltages and voltage drops from the first feeder
node to the end feeder node. The backward and forward
PDG ,min,k  500kW  PDG ,k  PDG ,max,k  10MW sweep together form one iteration and the algorithm
(5) repeats until the node voltage difference between two
k  1,...,noDG
consecutive iterations is less than a predetermined
threshold. The distribution line is modeled with a
In the above equations Umin represents the minimum
detailed π equivalent model which takes into
allowed line voltage of every node in the network
consideration the parallel capacitance, as shown in Fig. 2.
(0.9∙Unom), Umax represents the maximum allowed line
voltage of every node in the network (1.1∙Unom) and Un
represents the actual node line voltage.
The total number of nodes is denoted with variable
nodes. PDG,min and PDG,max represent the minimum and
maximum installed output power of a kth DG unit and
PDG represents the DG's actual installed output power.
Variables Si and Snom,i represent the actual apparent
power through the line i and the nominal thermal
limitation of line i, and locDG,k represents the location
constraints of kth DG unit. Fig. 2. A detailed π model of a distribution line
89
The equation that determines the three phase Unlike the crossover operator used in solving optimal
distribution line active power loss is: allocation of capacitor banks [15], the crossover point
used here is fixed and positioned in the middle of a
2
plossi  3  I lossi  Ri  chromosome. That means that a new offspring will have
the DG location from one parent and the DG size from
2
 Pi cos i  Qi sin i  the other parent. The main reason why a fixed crossover
  point is used lies in the integer coding of the
 3U n   chromosomes, which allows the distinct separation of the
 Ci  U n  sin i  DG location and size.
 
 2 3  (6) Finally, the mutation used in this paper has two
 3 2
 Ri components, each one having its own predetermined
  Qi cos i  Pi sin i   probability. The first type of mutation is the location
   
3U n mutation (with its predetermined probability ploc) and the

   
  Ci  U n  cos i  second type is the size mutation (with its predetermined
   probability psize). If an individual chromosome undergoes
  2 3   both mutations, it will simultaneously change its position
and size.
In the above equation Ci denotes parallel capacitance During the algorithm each chromosome will undergo
of line i. Pi and Qi represent active/reactive power that is a process called elimination load flow on three separate
injected in node n, Un is the line voltage of the nth node occasions. The first will occur while forming the initial
and φi represents phase angle between voltage and population, the second after the crossover operator and
current. The aforementioned equation will be used to finally the third after the mutation operator has been
evaluate every chromosome while performing the genetic performed. The elimination load flow has a purpose to
algorithm for DG allocation. examine voltage and thermal constraints of a network in
a worst case scenario where the individually observed
DG unit is online and operating with full peak power
IV. Genetic Algorithm while the network load is at its minimal level. If the
In this paper the authors will use a genetic algorithm elimination load flow is unsatisfactory during
to solve the DG allocation problem. Genetic algorithm initialization, the chromosome will be discarded and a
(GA) belongs to a class of heuristic optimization new random chromosome will take its place until the
methods that imitate the process of natural evolution, initial population reaches its predetermined size. If the
called evolutionary algorithms (EA). In this paper the elimination load flow is unsatisfactory during crossover,
authors have modified the genetic algorithm for DG the child chromosome will be discarded and a new
allocation problem. The coding of the individuals in random pair of parents will be chosen to create a new
population into chromosomes was done using integer child.
values, with every chromosome containing two parts. Finally, if the elimination load flow is unsatisfactory
The first part refers to the location of a DG unit and during mutation, the mutated child chromosome will be
the second part refers to the size of a DG unit, as discarded and the original child chromosome will again
displayed in Fig. 3. undergo the mutation process. It is important to note that
the elimination load flow is a single load flow analysis
and is not used for calculation of cumulative average
daily active power losses.
The cumulative average daily active power losses are
calculated during evaluation of each chromosome in the
current population. The input parameters for the
evaluation process are the average daily power
consumption curves for all the residential and industrial
customers. While evaluating the fitness function for each
Fig. 3. The coding of an individual solution into a chromosome chromosome, every chromosome must be decoded and
the average daily power production curves must be
Furthermore, the selection chosen for this problem formed for every DG unit. The curves will be formed
was a tournament selection, which involves running using the data provided by IPPs based on their
several "tournaments" among k individuals randomly preliminary measurements or power production
chosen from the current population. It must be noted that estimation.
the algorithm used in this paper has a built-in elitism, Finally, the decoded chromosome will undergo the
which means that the best individual in a generation is optimization load flow, which must be conducted 24
automatically stored and forwarded to next generation. times during an average day. The optimization load flow
The crossover operator the authors have chosen is a calculates the losses of an entire feeder in one hour of an
single-point crossover according to a predetermined average day in the year.
crossover probability (pc).
90
The sum of losses for each hour of an average day

represents the cumulative average daily active power
losses, and they are used to evaluate each chromosome.
In the end, after the predetermined number of
generations is carried out by the algorithm, the best
solution will be presented. The best chromosome will
contain information on optimal location and size of each
DG unit with respect to minimal network cumulative
average daily active power losses. The four types of DG
units will be allocated individually in the network.
A flowchart describing the steps of a developed
algorithm is displayed in Fig. 4 and Fig. 5.
V. Case Study of a Distribution Network

The distribution network under study consists of 30 Fig. 5. A flowchart describing the calculation
nodes (MV/LV Substations), as shown in Fig. 3. of the fitness function
The total feeder net active and reactive load is 7.420
MW and 3.250 Mvar, without electrical losses and factor The network consists of industrial customers
of simultaneity taken into consideration. connected to nodes 14 and 23, and the residential
customers, connected to all the other nodes of the feeder.
As in many practical cases, each DG unit is limited to
a set of locations where it can connect to an existing
distribution network. For example, solar parks should be
located on clear sites which are exposed to direct sunlight
during the whole day and without any shade. Wind farms
should be located on clear sites which do not have any
natural or man-made barriers to slow down wind velocity
and energy. The power stations which do not depend on
an intermittent primary energy source usually do not
have strict location constraints.
However, if a power station is built as a combined
heat & power unit (whether on fossil or renewable fuel)
it should be located near heat consumers, so that the
construction of unnecessary and expensive long heat
pipes is avoided. Finally, the small hydro units should be
located near rivers or streams with enough water flow
and height difference between inlet and outlet.
In this example, the connection of a new solar park
will be limited to nodes 3–12 and 22–24. The connection
of a new wind farm will be limited to nodes 11–17.
The connection of a new power station unit that does
not depend on an intermittent primary energy source will
be limited to nodes 25–30. Finally, the connection of a
new small hydro unit will be limited to nodes 19–21.
All the location constraints are displayed in the Fig. 6.
V.1. The Average Daily Power Consumption Curves

The average daily power consumption curves for each
consuming node (MV/LV Substation) are based on the
measurements of active and reactive power during each
hour over the period of a single year. The measurements
are either estimated or obtained via smart electricity
metering devices, which record consumption of electric
energy and power and send that information to DSO's
measurement department. The information can be
gathered for all the consumers (LV and/or MV)
Fig. 4. A flowchart describing the genetic algorithm used connected to every single MV/LV Substation.
for individual DG allocation
91
350
300
Active power [kW]

250
200
150
Node 15
100 Node 28
Node 7
50
Time [h] Nodes
Fig. 7. The average daily active power consumption curves

for nodes with connected residential customers
Fig. 6. An example of a 30 node distribution network used in this paper 800

and a set of location constraints for each type of DG unit 700
Active power [kW]

600
After all the data (24 × 365 measurements) is 500
collected, a power consumption curve for an average day 400
Node 23
in the year can be constructed. 300
Node 14
When the average daily power consumption curves for 200
each MV/LV Substation are constructed, the DSO's 100
measurement department must forward the data to DSO's 0
MV network planning department. Also, the data on
minimum recorded consumed active and reactive power
during one year for every MV/LV Substation must be Time [h]
Nodes
forwarded. Planners will than use this information as
input data to the allocation algorithm. Fig. 8. The average daily active power consumption curves
It will be presumed that the average daily power for nodes with connected industrial customers
consumption curve for every node with connected
residential customers has the same shape. For example, In that way a planner can approximately estimate the
the average daily active power consumption curves for percentage of injected active power for a future solar
nodes 7, 15 and 28 are displayed in Fig. 7. Also, it is park, wind farm and a small hydro unit in each hour
presumed that the industrial customers connected to during an average day in the year.
nodes 14 and 23 have their own unique average daily After the data is collected and processed, the IPP must
power consumption curves, as displayed in Fig. 8. forward it to DSO's MV network planning department.
Planners will than use the information as input data to the
allocation algorithm. It is presumed that the data on ratio
V.2. The Average Daily Power Production Curves of measured solar potential and maximal irradiation for
Since the goal of the allocation algorithm is to find an average day in the year is displayed in Fig. 9.
optimal positions and sizes of DG units, the planners It is presumed that the data on ratio of measured
cannot specify in advance their average daily power wind potential and maximal wind speed for an average
production curves in absolute values. However, the day in the year is displayed in Fig. 10.
planners can make an estimation of the percentage of 0,60
average daily power production on the basis of
0,50
preliminary measurements for each DG type.
Paverage/Ppeak [%]
In this paper the authors will presume that the 0,40

investors of the solar park, wind farm and a small hydro
0,30
unit have made preliminary measurements of solar
(W/m2), wind (m/s) and water flow potential (m3/s) for 0,20
every hour during the period of a single year. All the data 0,10
must be collected and processed to calculate irradiation,
0,00 Solar park
wind speed and water discharge values for an average
day of the year.
However, these values for each hour of an average day Time [h]
must be divided with maximal recorded irradiation, wind
speed and water discharge during the preliminary Fig. 9. The estimation of average daily active power production curve
measurement period. for a solar park in percentage values
92
0,60
It is presumed that the data on ratio of measured water
0,50 discharge and maximal water discharge for an average
Paverage/Ppeak [%]
0,40
day in the year is displayed in Fig. 12.
0,30 0,90
0,80
0,20
0,70
Paverage/Ppeak [%]
0,10 Wind f arm 0,60
0,50
0,00
0,40
0,30
0,20 Small hydro
0,10
Time [h]
0,00
Fig. 10. The estimation of average daily active power production

curve for a wind farm in percentage values
Time [h]
It is interesting to see that, although the wind is a
highly stochastic and intermittent primary energy source, Fig. 12. The estimation of average daily active power production
curve for a small hydro unit in percentage values
the wind speed (and consequently power production) for
an average day in the year is relatively flattened.
That happens because the days with extreme wind VI. Results and Discussion
velocities superimpose on days with low wind velocities,
forming a relatively flat wind speed curve for an average The parameters of a genetic algorithm used in this
day in the year. However, a serious drawback is that the paper are displayed in Table I.
capacity factor of a wind farm will not rise above 50-
TABLE I
60%. The capacity factor is the ratio of the net electricity
THE PARAMETERS OF A GENETIC ALGORITHM USED IN THIS PAPER
generated, for the time considered, to the energy that Population size Generation number Tournament size
could have been generated at continuous full-power 20 50 3
operation during the same period.
The average daily power production percentage curve The crossover probability, mutation location and size
for a power station that does not depend on an probability used in this paper are displayed in Table II.
intermittent primary energy source can be specified in
advance without preliminary measurements. TABLE II
The reason is that its fuel (natural gas, biomass, etc) THE PARAMETERS OF GENETIC OPERATORS USED IN THIS PAPER
can be obtained on the market, unlike sun or wind. It is Crossover Mutation location Mutation size
presumed that the data on ratio of injected and nominal probability probability probability
0.7 0.7 0.7
power of a future power station for an average day in the
year is displayed in Fig. 11.
Unlike the wind farm, the output of a power station The results of individual allocation of every DG unit
for an average day in the year will be between 80-100% are displayed in the following subsections. The allocation
of peak power, because a power station owner will was done with respect to location constraints, which were
produce peak power constantly to receive maximum previously determined for each DG unit. The network
feed-in tariff incentive. The only time a power station losses before individual DG allocation were equal to
may temporarily shut down is during a fault or operation 542.48 kW.
and maintenance (O&M) procedures.
0,90 VI.1. The Allocation of a Solar Park
0,80
0,70
The calculated optimal position and size of an
Paverage/Ppeak [%]
0,60
individually allocated solar park is at node 7 with peak
0,50 power of 7 MW, as displayed in Table III.
0,40 Also, the table contains the percentage reduction of
0,30 cumulative average daily active power losses compared
0,20 Power station to the base case distribution network without DG
0,10
0,00
connection.
The optimally allocated solar park reduces cumulative
average daily active power losses up to 40.8% compared
Time [h]
to the base case without DG connection.
In Fig. 13 the dynamic daily distribution of active
Fig. 11. The estimation of average daily active power production curve power losses is displayed for a base case without DG
for a power station in percentage values connection and after optimal allocation of a solar park.
93
TABLE III 50
THE ALLOCATION RESULTS FOR A SOLAR PARK 45

Active power losses without
previously connected DG
Optimal Losses after solar Reduction 40
units
Optimal size [kW]
Active power losses [kW]

Active power losses with an
position park allocation [kW] [%] allocated wind f arm
35
7 7,000 320.93 40.8
30
50 25
45 Active power losses without
previously connected DG 20
units
40 15
allocated solar park
35 10
30 5
25 0
20 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Time [h]
15
10 Fig. 14. The comparison of active power losses for each hour
before and after the wind farm allocation
5
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Also, the table contains the percentage reduction of
Time [h] cumulative average daily active power losses compared
to the base case distribution network without DG
Fig. 13. The comparison of active power losses for each hour connection.
before and after the solar park allocation
TABLE V
It can be seen that the losses become significantly THE ALLOCATION RESULTS FOR A POWER STATION
lower during the day when a solar park is activated, Optimal OPTIMAL Losses after power Reduction
especially around midday. position SIZE [KW] station allocation [kW] [%]
25 3,000 216.35 60.1
VI.2. The Allocation of a Wind Farm The optimally allocated power station reduces
The calculated optimal position and size of an cumulative average daily active power losses up to
individually allocated wind farm is at node 11 with peak 60.1% compared to the base case without DG
power of 3.750 MW, as displayed in Table IV. Also, the connection. In Fig. 15 the dynamic daily distribution of
table contains the percentage reduction of cumulative active power losses is displayed for a base case without
average daily active power losses compared to the base DG connection and after optimal allocation of a power
case distribution network without DG connection. station that does not depend on an intermittent primary
energy source.
TABLE IV It can be seen that the losses become significantly
THE ALLOCATION RESULTS FOR A WIND FARM lower during the most part of the day, because the output
Optimal Optimal size Losses after wind farm of a power station during the average day in the year is
Reduction [%]
position [kW] allocation [kW]
11 3,750 216.82 60.0
relatively constant.
However, during the night the losses actually increase
because the load is at its minimal value.
The optimally allocated wind farm reduces cumulative
average daily active power losses up to 60% compared to 50
the base case without DG connection. 45

Active power losses without
previously connected DG units
In Fig. 14 the dynamic daily distribution of active
40
power losses is displayed for a base case without DG

allocated power station
connection and after optimal allocation of a wind farm. 35
It can be seen that the losses become significantly 30
lower during the most part of the day, because the output 25
of a wind farm during the average day in the year is
relatively flat. However, during the night the losses 20
actually increase because the load is at its minimal value. 15
10
VI.3. The Allocation of a Power Station 5
The calculated optimal position and size of an 0

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
individually allocated power station that does not depend Time [h]
on an intermittent primary energy source is at node 25
Fig. 15. The comparison of active power losses for each hour
with peak power of 3 MW, as displayed in the Table V. before and after the power station allocation
94
Although the location constraints limited the Since the lateral branch had a light load, the
connection of a power station to a lateral branch elimination load flow prevented the connection of a
(between nodes 25–30), it is obvious that the losses have hydro unit with bigger size. In this example it can be
decreased the most compared to the allocation results of observed how the location constraints have a great
other DG units. influence on network losses. For that reason the location
If the location constraints were removed and the constraints must not be neglected in the DG allocation
power station could connect on a main feeder branch algorithm.
(between nodes 1–17), the losses would be even more
reduced.
VII. Conclusion
VI.4. The Allocation of a Small Hydro Unit In this paper the authors have presented an algorithm
for the individual allocation of DG units based on
The calculated optimal position and size of an average daily power consumption and production curves.
individually allocated small hydro unit is at node 19 with The goal was the minimization of cumulative average
peak power of 2.250 MW, as displayed in the Table VI. daily active power losses. Using the proposed algorithm
Also, the table contains the percentage reduction of the obtained allocation results are much more detailed
cumulative average daily active power losses compared and precise, which in turn can have a great importance in
to the base case distribution network without DG avoiding unnecessary and often quite considerable
connection. connection costs. Also, the costs of the DSO will be
minimized, since the DSO is responsible for the
TABLE VI
generated losses in the distribution network, and must
THE ALLOCATION RESULTS FOR A SMALL HYDRO UNIT
Optimal Optimal size Losses after small Reduction
acquire energy on the electricity market to cover those
position [KW] hydro unit allocation [kW] [%] losses.
19 2,250 469.71 13.4 The next step in the authors' research is to integrate
energy storage in the distribution network (small
The optimally allocated small hydro unit reduces pumped-storage hydroelectricity, sodium-sulfur NaS
cumulative average daily active power losses up to batteries) and to incorporate more strict voltage
13.4% compared to the base case without DG constraints in the algorithm. In some Central European
connection. In Fig. 16 the dynamic daily distribution of countries a common practice is that the voltage variation
active power losses is displayed for a base case without of each feeder node must not be greater than ±3% after
DG connection and after optimal allocation of a small the connection of DG unit(s). The aforementioned topics
hydro unit. will be the subject of the next paper.
It can be seen that the losses become lower during the
most part of the day, because the output of a small hydro
unit during the average day in the year is relatively Appendix
constant. However, during the night the losses actually In Table A1 the electrical parameters of lines used in
increase because the load is at its minimal value. the network example are presented and in this paper one
The reason that the loss reduction was not so type of electric underground MV cables is used.
significant (only 13.4%) are the location constraints
which limited the connection of a small hydro unit to a TABLE A1
lateral branch (between nodes 19–21). DATA OF THE ELECTRICAL CABLES USED IN THE CASE STUDY
DISTRIBUTION NETWORK
50
Un R1 X1 B1 Inom
Type
45
Active power losses without (kV) (Ω/km) (Ω/km) (μS/km) (A)
previously connected DG units
NA2XS(F)2Y
20 0.206 0.115 82 345
40 3x(1x150) mm2

allocated small hydro unit
35
30
Table A2 contains the data regarding the maximum
measured amount of active and reactive power consumed
25
by every node in the network under study and the lengths
20 of electrical lines.
15
To calculate the minimum network load for the
elimination load flow it will be presumed that the data on
10
minimal amount of active and reactive power is known
5
for the entire feeder (Pfeeder_min=1,200 kW and
0 Qfeeder_min=500 kVAr). The data can be obtained via
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Time [h] SCADA system (System Control and Data Acquisition)
which actively monitors the power on all the feeders of a
Fig. 16. The comparison of active power losses for each hour Primary Substation and sends the data to the DSO's
before and after the small hydro unit allocation
Control Centre.
95
TABLE A2 and Technology, 2, 8(2011), pp. 832 – 837.

THE MAXIMUM MEASURED ACTIVE AND REACTIVE POWER CONSUMED [4] Raj, P.A.; Senthilkumar, S.; Palanivelu, T.G. Swarm intelligence
BY EACH NODE AND THE LENGTHS OF ELECTRICAL LINES IN THE CASE based optimization of distributed generation capacity for power
STUDY DISTRIBUTION NETWORK quality improvement. // Acta Elektrotehnica, 49, 3(2008), pp. 334-
Reactive 342.
Active [5] Harrison, G. P.; Piccolo, A.; Siano, P.; Wallace, A. R. Hybrid GA
Node No. power Branch No. Length (km)
power (kW) and OPF evaluation of network capacity for distributed generation
(kVAr)
1 200 90 1 0.7 connections. // Electric Power Systems Research, 78(2008), pp.
2 180 60 2 0.5 392–398.
3 240 120 3 0.7 [6] Ajay-D-Vimal Raj P.; Senthilkumar, S.; Raja, J.; Ravichandran,
4 120 45 4 0.8 S.; Palanivelu, T. G. Optimization of distributed generation
5 120 30 5 0.9 capacity for line loss reduction and voltage profile improvement
6 400 150 6 0.6 using PSO. // Elektrika, 10, 2(2008), pp. 41-48.
7 400 150 7 0.2 [7] Heydari, M.; Hajizadeh, A.; Banejad, M. Optimal Placement of
8 120 30 8 0.8 Distributed Generation Resources. // International Journal of
9 120 30 9 0.7 Power System Operation and Energy Management, 1, 2(2011), pp
10 90 45 10 0.3 1–5.
11 120 55 11 0.1 [8] Moravej, Z., Akhlaghi, A., A new approach for DG allocation in
12 120 55 12 0.4 distribution network with time variable loads using cuckoo search,
13 240 120 13 0.4 (2012) International Review of Electrical Engineering (IREE), 7
14 650 15 14 0.2 (2), pp. 4027-4034.
15 120 30 15 0.7 [9] Hosseini, S.A.; Karami, M.; Karimi Madahi, S. S.; Razavi, F.;
16 120 30 16 0.8 Ghandimi, A.A. Finding the Optimal Capacity and Location of
17 180 60 17 0.5 Distributed Generation Resources and Analyzing the Impact of
18 180 60 18 0.9 Different Coefficient Factors. // Journal of Basic and Applied
19 180 60 19 0.1 Scientific Research, 1, 12(2011), pp. 2578 – 2589.
20 180 60 20 0.3 [10] Yammani, C.; Maheswarapu, S.; Matam, S. Optimal Placement of
21 180 60 21 0.6 Multi DGs in Distribution System with Considering the DG Bus
22 180 75 22 0.2 Available Limits. // Energy and Power, 2, 1(2012), pp. 18 – 23.
23 840 300 23 0.3 [11] Maheswari, S.; Vijayalakshmi, C. Implementation of an
24 840 300 24 0.6 Optimization Technique for Improving Power Quality in the
25 120 40 25 0.4 Distributed System. // Indian Journal of Computer Science and
26 120 40 26 0.2 Engineering, 3, 2(2012), pp. 277 – 281.
27 120 30 27 0.1 [12] Rotaru, F.; Chicco, G.; Grigoras, G; Cartina, G. Two-stage
28 240 105 28 0.9 Distributed Generation Optimal Sizing with Clustering-based
29 400 900 29 0.4 Node Selection. // International Journal on Electrical Power and
30 300 105 30 0.3 Energy Systems (IJEPES), 40, 1(2012), pp. 120–129 .
[13] El-Zonkoly, A. M. Optimal placement of multi DG units
including different load models using PSO. // Smart Grid and
With this information the simultaneity factor for Renewable Energy, 1(2010), pp. 160-171.
minimal active and reactive load can be calculated as: [14] Prenc, R.; Škrlec, D.; Komen, V. A novel load flow algorithm for
radial distribution networks with dispersed generation. //
Technical Gazette, 20, 6(2013), pp. 969-977.
Pfeeder _ min_ SCADA [15] Barukčić, M.; Hederić, Ž.; Jović, F. Adaption of genetic algorithm
f sim _ min_ P  nodes
(1-A) for more efficient minimization of active power losses in power
 Pmax_ node network. // Technical Gazette, 15, 3(2008), pp. 11-19.
n 1
Q feeder _ min_ SCADA Authors’ information

f sim _ min_ Q  nodes
(2-A)
Rene Prenc was born in 1984 in Rijeka,
 Qmax_ node Croatia. He received his Master's Degree of
Electrical Engineering from Faculty of
n 1
Engineering, University of Rijeka, Croatia, in
2008. He has received his PhD Degree at the
when active and reactive power of each node in the Table Faculty of Electrical Engineering and
A2 is multiplied by the appropriate simultaneity factor, Computing (FER), University of Zagreb,
the correct data on minimal consumed active and reactive Croatia, in 2014. He joined the Croatian Electric
power can be obtained. Power Company (HEP) in 2008 and has been working in the field of
distribution system planning and design. In 2014 he joined the Faculty
of Maritime Studies, University of Rijeka, Croatia, and is working in
the field of naval power systems. His main areas of research are
References application of optimization techniques in distribution power system
planning, optimal integration of distributed generation and power
[1] Popović, D. H.; Greatbanks, J. A.; Begović, M.; Pregelj, A.
system analysis.
Placement of distributed generators and reclosers for distribution
network security and reliability. // International Journal of
Nikola Bogunović was born in 1981 in Rijeka,
Electrical Power and Energy Systems, 27, 5-6(2005), pp. 398–
Croatia. He received his Master's Degree of
408.
Electrical Engineering from Faculty of
[2] Kaluđer, S.; Šljivac, D.; Miletić S. The optimal placement of
Engineering, University of Rijeka, Croatia in
distributed generation. // Technical Gazette, 19, 3(2012), pp. 535-
2006. He joined the Croatian Electric Power
541.
Company (HEP) in 2007 and has been working
[3] Mohammadi, M.; Akbari Nasab, M. PSO Based Multiobjective
in the field of distribution system planning and
Approach for Optimal Sizing and Placement of Distributed
documentation. His main areas of research are
Generation. // Research Journal of Applied Sciences, Engineering
96
the implementation of Geographic Information System (GIS) in DSO's

technical system, power system planning and the optimal integration of
distributed generation.
Aleksandar Cuculić was Born in 1974 in

Rijeka, Croatia. He received his Bachelor's
degree in 1995, and his Master's Degree of
Marine Electronics and Information
Technologies in 2010 from Faculty of Maritime
Studies, University of Rijeka, Croatia. He has
twelve years (1996-2008) of experience as an
Electro technical Officer (ETO) onboard ocean
going vessels. From 2001 until 2008 he was employed at Royal
Caribbean International, holding the position of Chief Electrical
Engineer. In 2010 he joined the Faculty of Maritime Studies where he
works in the field of marine power systems. His research areas are:
marine electrical power systems, electrical propulsion and power
electronics.
97
Future-Proof MV Distribution Network Short-Circuit

Protection Scheme
H. Laaksonen
Abstract – Short-circuit protection operation time delays in medium-voltage (MV) networks have
traditionally been dependent on fault-current magnitude or measured impedance with fixed time
delays or inverse time curves. However, MV feeder protection selectivity issues with low-voltage-
ride-through (LVRT) curves of distributed generation (DG) units and possibility of intended island
operation must be considered increasingly in the future as a part of the protection scheme.
Therefore, this paper proposes a future-proof, directional short-circuit protection scheme for
Smart Grids. The proposed scheme enables the definition of LVRT curve compatible protection
operation time delays during normal, grid connected operation for protection relays in MV
distribution network having multiple protection zones without use of high-speed communication
based interlockings/blockings between relays. Copyright © 2015 Praise Worthy Prize S.r.l. - All
rights reserved.
Keywords: Protection, Smart Grids, Grid Codes, Low-Voltage-Ride-Through, Fault-Ride-

Through, Islanding, Distributed Generation
I. Introduction The operation time delay of the proposed scheme is

not dependent on fault-current magnitude or measured
In the future active network management may impedance. It is instead based on measured voltages at
simultaneously affect to protection settings if for instance that point of network taking also into account current
network topology is changed. In addition, new grid codes direction. In this way it is easier to obtain and maintain
require fault-ride-through (FRT) / LVRT capability and selectivity with DG unit LVRT requirements especially
utility grid supporting functionalities from DG units and when MV feeder is divided in multiple protections zones
can also allow intentional island operation. and high-speed communication is not available.
Therefore, protection adaptation as well as protection On the other hand, this developed protection scheme
principles and settings compatibility with new grid code could be used also as a reliable back-up protection
requirements will be increasingly required in the future scheme for high-speed communication based schemes.
Smart Grids. Related information, studies and proposed The suggested new scheme is applicable for MV
solutions can be found for example from references [1]– networks with different topologies (radial or ring). The
[4]. MV network short-circuit protection operation time proposed scheme takes also into account the possibility
delays have traditionally been dependent on fault-current of transition to intended island operation. The forward
magnitude or measured impedance with fixed time and reverse settings of the directional protection are
delays or inverse time curves. In the future, MV dependent on the active and reactive power unbalance
networks will be increasingly divided into multiple (Punb and Qunb) at connection point of the potential
protection zones in order to improve supply reliability. intended island and also on the capacity of rapidly
Therefore, short-circuit protection operate times may controllable active and reactive power inside the intended
become too long if high-speed communication based island. After islanding detection, the protection scheme
interlocking/blocking schemes are not utilized. However, should adapt to the changed topology as described for
the communication may fail or is not available example in [3], [4]. In addition to transfer trip-based
everywhere for example, due to costs. islanding detection schemes, also different new passive
This paper proposes a LVRT curve compatible, future- islanding detection algorithms and schemes has been
proof, directional short-circuit protection scheme for recently proposed and compared in [5]–[8].
Smart Grids. The proposed scheme is generally applicable with
The proposed grid code compatible scheme enables different LVRT curves defined in various grid codes.
the definition of protection operation time delays during ENTSO-E NC RfG [9] is taken as an example from
normal, grid connected operation for intelligent Europe. The proposed scheme is also applicable without
electronic devices (IEDs) in MV distribution network DG units, which means that no adaptation of the short-
having multiple protection zones without the need for circuit protection pick-up/start value or the voltage
high-speed communication. dependent operation time delays is required due to DG
98
H. Laaksonen
unit connection or disconnection. In such case where between LVRT curve of DG units (defined by grid
both inverter-based and synchronous generator based DG codes) and required time differences between CB2 and
can be simultaneously connected to the same MV feeder, CB3 in forward direction may be hard to achieve.
operation time delay curve of the proposed protection This naturally depends from the number of
scheme must be based on the most stringent DG unit consecutive protection zones and the allowed time
LVRT curve. difference between operation time delays of CB2 and
Effect of other recent and forthcoming grid code CB3.
requirements, stated for example in [9]–[12], has been
also studied in this paper. Reason for this was to confirm
that other requirements, such as converter based DG
units’ voltage support by reactive current injection, active
power/frequency (P/f)-or reactive power/voltage (Q/U) -
control of DG units will not affect to the proposed
protection scheme negatively. Previously in [13] some of
these new grid code requirements have been briefly
summarized. Also in [14] the risk of islanding situations
stabilization due to P/f- and Q/U-control requirements
has been studied.
In following, first possible time selectivity issues with
DG unit LVRT curves and traditional protections
schemes are shortly highlighted when MV feeder
consists from several consecutive protection zones. After
that, the proposed grid code compatible protection
scheme is presented. Finally, example simulation results
are presented followed by discussion and conclusions.
II. Possible Time Selectivity Issues with

Traditional Protection Schemes
In Fig. 1 one example case is used to present the
protection needs (functions, time selectivity) when
protection coordination with DG unit LVRT curve and
successful transition to island operation is considered.
This case in Fig. 1 is presented without utilization of
high-speed communication for protection interlockings /
blockings. In general, the use of high-speed Fig. 1. Possible intended islands 1.-4.and protection time selectivity
issues, setting principles without utilization of high-speed
communication would enable rapid, reliable and selective communication when a) islanding is not allowed and b) islanding is
protection coordination with DG unit LVRT curves in possible [15]
future distribution networks with multiple consecutive
protection zones and the possibility for intended island Transition to intentional island operation is only
operation. possible (Fig. 1) if active and reactive power unbalance
In Fig. 1 the idea is that the possible operation (Punb and Qunb) at CB1, CB2, CB3 or CB4 is small
principles of directional short-circuit protection in enough (or enough, rapidly controllable active and
forward direction can be 1) directional overcurrent reactive power units exist in the possible island) before
protection with fixed time delay (and high-stage / low- protection start/operation of CB1-CB4 in reverse
stage settings) or 2) distance protection with fixed time direction.
delay (in forward direction). Similarly in Fig. 1 the If this is not the case transition to island operation
possible operation principles of directional protection in should be not allowed. As also stated for example in
reverse direction (for intentional islanding) can be I) [14]the new grid codes like [9] could enable / support
undervoltage with fixed time delay (and high-stage / low- transition to intentional island operation because of P/f-
stage settings) AND current direction detection in reverse droop control requirements of DG units during over-
direction, where function pick-up/start is only based on frequency situations (under-frequency based load
undervoltage (i.e. not in overcurrent, because fault shedding schemes could have similar kind of effect) and
current levels of inverter-based DG units can be fairly possibly also due to voltage control (Q/U-control)
low) or II) distance protection with fixed time delay (in requirements.
reverse direction). In above and in Fig. 1 only short-circuit protection has
From Fig. 1 it can be seen that selectivity problems been considered, but naturally also earth-fault protection
can be possible if communication based principles and settings have to be proper for during both
interlockings/lockings are not used, because coordination normal and island operation.
99
H. Laaksonen
III. Proposed Grid Code Compatible infeed of multiple DG units in that zone (both MV and
Protection Scheme LV network connected DG units) is taken into account
and discrimination between 3-phase short-circuit faults in
In Section II (Fig. 1) it was presented how time LV busbars or LV feeders at MV/LV substations could
selectivity with DG unit LVRT curves may be be still done.
challenging to achieve, if fixed time delays are used, MV
feeder consists of many consecutive protection zones and Reverse (rev) Direction
high-speed communication utilization is not possible. Short-circuit protection fault detection (pick-up/start)
In Section I some of the main features of the proposed in reverse (rev) direction is based on exceeding
grid code compatible protection scheme were shortly undervoltage pick-up/start value (Fig. 2). Overcurrent-
stated. In Fig. 2 and in the following a more detailed based pick-up is not applicable, because fault current
description about this scheme will be given. levels of inverter-based DG units can be quite low. In
addition, final detection in reverse direction requires
correct current direction determination. In comparison to
detection in forward direction, the detection in reverse
direction is by default not allowed to give operate signal
to corresponding circuit-breaker (Fig. 2).
However, depending from the islanding possibility
(i.e. power balance through CB before fault) and
allowance to utilize intended island operation, operate
signal can be allowed (Fig. 2). In other words this means
that transition to intentional island operation is only
possible if active and reactive power unbalance (Punb and
Qunb) through connection point CB is small enough (or
enough, rapidly controllable active and reactive power
units exist in the possible island) before protection
start/operation in reverse direction. If this is not the case
transition to island operation should be not allowed.
During meshed or ring operation of MV feeders pick-
up/start criteria is similar in forward and reverse
directions.
Direction Determination
Based on PSCAD simulations, in addition to
traditional methods for fault current direction
determination, Upos_seq_angle- Ipos_seq_angle(Upos_seq_angleand
Ipos_seq_angleare voltage and current positive sequence
angles respectively) based direction determination is
possible during both 3- and 2-phase short-circuit faults as
well as during 2-phase earth-faults. Also U0_angle - I0_angle
Fig. 2. Proposed directional short-circuit protection scheme during
normal, grid-connected, radial operation of MV network (U0_angle and I0_angle are voltage and current zero sequence
angles respectively) based angle determination could be
possible in 2-phase-earth-faults in MV networks with
III.1. Directional Protection Pick-up/Start and isolated neutral treatment.
Direction Determination
Forward (fwd) Direction III.2. Voltage Dependent Time Delay
With short-circuit protection in distribution networks
having inverter or synchronous generator-based DG Time Delays with Even Time Steps (tstep) – Low Number
units, the detection of fault i.e. pick-up/start in forward of Protection Zones
(fwd) direction is based on exceeding over-current pick- After fault detection in forward or reverse direction
up/start value AND/OR undervoltage pick-up/start value (Fig. 2) voltage dependent time delay of the proposed
(Fig. 2). Also 2-phase earth-faults are included in protection scheme is determined. The main idea is to use
proposed short-circuit protection scheme (Fig. 2). In similar LVRT curves for calculation of MV feeder IED
addition, for final detection in forward direction also protection time delays (i.e. voltage dependent time
correct current direction determination is required. delays) than are required from the DG units by the grid
Also impedance or admittance based pick-up/start codes. In this way it is easier to achieve selectivity with
value could be possible with this scheme and then DG unit LVRT curves. Time delay of the proposed
impedance start value should be at least 105-110% of scheme is dependent from voltage dip magnitude i.e.
protected zone (MV line length) so that also possible distance from the fault so that the protection operation
100
H. Laaksonen
time delay is shorter than the operation time delay of DG

unit LVRT curve (Figs. 3 and 4).
The voltage dependent time delay (Figs. 2–4) is
determined in the following way:
1) The voltage dependent (e.g. lowest phase-to-phase
voltage Uph_ph_min_pu) operation time delay is
calculated 30 ms after pick-up of protection (i.e. 30
ms + CB operate time is the minimum operate time)
o Time delay difference between consecutive
protection zones for example in forward direction
tdelay_fwdshould be at least CB operate time +10 ms
(i.e. ≥ 60 ms)
2) When CB with shortest operate time opens, time
delays of the other IEDs needs to be re-calculated to
ensure selective protection
o Thereby only faulted section is isolated
o Therefore tdelay_fwdand tdelay_rev are re-calculated 30
ms + CB operate time after protection pick-up
and/or after relatively large change in magnitude
of positive sequence voltage or in magnitude of
lowest phase-to-phase voltage Uph_ph_min_puafter
pick-up and before the end of initially calculated
operate times 30 ms after pick-up
3) If fault condition is changed (e.g. fault current
direction is changed) OR removed (pick-up/start
criteria is not fulfilled anymore) before the end of
calculated tdelay_fwdor tdelay_rev then function is reset.
In Figs. 3 the example voltage dependent operation
time delay curves of the proposed scheme are shown
from case where islanding is not possible and
respectively in Figs. 4 from case where islanding is
possible. Forthcoming ENTSO-E NC RfG DG unit
LVRT requirements [9]are used as an example in Figs. 3
and 4. The calculation of the voltage dependent time
delay is based on the fact that voltage drop of lowest
phase-to-phase voltage Uph_ph_min_pu or Upos_mag_pu-
Uneg_mag_pu, (Upos_mag_puand Uneg_mag_puare voltage positive
and negative components respectively) due to short- Figs. 3. Voltage dependent time delay curve in forward direction for
short-circuit protection when islanding is NOT possible a) with
circuit faults is not equal in every point of the MV inverter-based DG units and b) with synchronous generator based DG
distribution network. For example time selectivity units (see also Eq. (1))
between MV feeder IED at HV/MV substation and IED
at MV recloser / switching station further away from the This is due to the fact that voltage dip is almost equal
HV/MV substation can be maintained if fault is at the in different points of the islanded distribution network
end of MV feeder. Additionally, for example in 2-phase after fault during island operation.
faults better selectivity is achieved by usage of lowest When islanding is not possible time delay in forward
phase-to-phase voltage Uph_ph_min_pu or Upos_mag_pu- direction (tfwd) in ms is determined as shown in Figs. 3
Uneg_mag_pu instead of Upos_mag_pu, which is often used in and (1):
the grid codes for DG unit LVRT curves.
In addition, in order to ensure selectivity also with
= − −
short MV feeders, it is preferred to use lowest phase-to- (1)
phase voltage Uph_ph_min_pu or Upos_mag_pu-Uneg_mag_pu. ∗ +− _ _
Based on the PSCAD simulations minimum protection
zone length is 2–2.5 km with the proposed scheme where:
without usage of high-speed communication. This
depends also to some extent from grid code LVRT −( + )
_ _
requirement. In general, it should be noted that based on _ = [ ]
PSCAD simulations voltage dip magnitude dependency _ _
of Uph_ph_min_pu from distance to fault can be utilized only
during normal operation, not during island operation. and nfwd_zones_total is the total number of protection zones
in forward direction, tmax is maximum time delay, tmin is
101
H. Laaksonen
minimum time delay, nfwd_zone_number is the forward zone = − +

number (number 1 zone is the furthest protection zone + (2)
from the HV/MV primary substation in forward − _ ∗
− _ _
direction) at that specific measurement point and
tdip_sensitivity_fwd is adjustable (minimum operation time
could be adjusted) and dependent from the customers where:
sensitivity to voltage dips, number of protection zones,
−( +
LVRT curve (e.g. between 0 to 40 ms + tmin).
Minimum time delay tmin could be also voltage + _ _ )
_ = [ ]
dependent instead of constant value (see Figs. 3) so that _ _
it increases linearly or in steps when voltage is higher.
and nrev_zones_total is the total number of protection zones in
reverse direction, nrev_zone_numberis the reverse zone number
(number 1 zone is the furthest protection zone from the
HV/MV primary substation in reverse direction) at that
specific measurement point and tdip_sensitivity_revis
adjustable and dependent from the customers sensitivity
to voltage dips (e.g. between 0 to 150 ms + tmin).
Also tP&Q_unbalance factor could be added in Eq. (2) so
that trev could be in certain limits (in which islanding is
allowed) dependent from power unbalance (Punb, Qunb) at
that point.
This would meand that trev is smaller when Punb and
Qunb is larger (but still within the required limits) to
ensure successful transition to island operation. The
minimum time delay tmin could be also voltage dependent
instead of constant value so that it increases linearly or in
steps when voltage is higher (Figs. 4). During ring
operation trev is equal to tfwd and it is not dependent from
the islanding possibility.
Time Delays with Even or Uneven Time Steps (tstep) –

High Number of Protection Zones
When number of protection zones is high and
blockings are not used at dual-breaker switching/recloser
stations, selectivity problems in dual-breaker schemes
with nearby fault are possible, if these are not considered
in the voltage dependent time delays determination. Time
delay difference between dual-breakers in ‘busbar zones’
should be always at e.g. in tfwd should be at least CB
operate time/switch +10 ms (i.e. ≥ 60 ms).
Therefore, time delays of other zones must be adapted
which may result in uneven time steps between
protection zones if time delays of ‘busbar zones’ are
fixed.
On the other hand, the number of protection zones
(from operate time point of view) could be reduced by
using blockings in possible dual-breaker
switching/recloser stations at MV network. In following
examples only the time delays with even time steps (tstep)
are considered.
Figs. 4. Voltage dependent time delay curve in forward and reverse
directions for short-circuit protection when islanding IS possible a) with
inverter-based DG units and b) with synchronous generator based DG III.3. Case Examples
units (see also Eq. (1) and (2))
In following (Figs. 5–8) few case examples with
When islanding is possible time delay in forward different network topologies and amount of CBs are
direction (tfwd) is determined as shown in Figs. 4 and (1). presented to highlight the need to change the
Time delay in reverse direction (trev) is defined in ms determination of forward and reverse current directions
as presented in (2) and Figs. 4: according to topology changes.
102
H. Laaksonen
Case 1 – Radially Operated MV Feeders Case 3 (Radial) – Effect of Change from Dual-Breaker to
Single-Breaker Scheme
Fig. 7. Case 3 with radially operated MV feeder, effect of change from

dual-breaker to single-breaker scheme (See Figs. 3, 4 and 6)
Fig. 5. Case 1 with radially operated MV feeders (See Figs. 3 and 4)

Case 1 – Ring Operated MV Feeders
Case 2 (Radial) – Effect of Topology Change
Fig. 6. Case 2 with radially operated MV feeder, Fig. 8. Case 1 with ring operated MV feeders
effect of topology change (See Figs. 3, 4 and 5) (See Figs. 3, 4 and 5)
103
H. Laaksonen
IV. Simulation Results TABLE I

FAULT TYPES IN CASE 1 WITH RADIALLY OPERATED MV FEEDERS
In following the proposed scheme to determine the (FIG. 9)
Fault Fault Type
voltage dependent time delays is presented by few
F1 3-phase short-circuit (fault res. Rf = 0.01 Ω) in MV grid
example cases. F2 2-phase short-circuit (fault res. Rf = 0.01 Ω) in MV grid
Forthcoming ENTSO-E NC RfG [9] LVRT F3 3-phase short-circuit (fault res. Rf = 0.001 Ω) in LV grid
requirements for DG units (Figs. 3 and 4) are also used in F4 3-phase short-circuit (fault res. Rf = 0.01 Ω) in MV grid
these example cases. F5 2-phase short-circuit (fault res. Rf = 0.01 Ω) in MV grid
F6 3-phase short-circuit (fault res. Rf = 0.01 Ω) in MV grid
It should be noted that if there exists both inverter- F7 2-phase short-circuit (fault res. Rf= 0.01 Ω) in MV grid
based and synchronous generator based DG units in the F8 3-phase short-circuit (fault res. Rf = 0.01 Ω) in MV grid
MV network, then the protection settings (i.e. voltage F9 2-phase short-circuit (fault res. Rf = 0.01 Ω) in MV grid
dependent time delays in this case) of MV feeder IEDs
should be selective and coordinated with the most Case 1 (Radial) – No DG Units
demanding DG unit LVRT curve (i.e. LVRT curve of In Table II, based on the PSCAD simulations, the
synchronous generator based DG unit). Then the scheme operation time delays in forward (tfwd) direction for
would also naturally support the possible topology different CBs (controlled by IEDs) are presented during
changes due to active network management schemes different faults (Table I and Fig. 9) from case 1 without
without the need of adaptation of protection settings. DG units i.e. DG units have been disconnected.
In following for the sake of clarity, only simulation In this case it should be noted that ENTSO-E RfG [9]
results from simulations with inverter-based DG units are based LVRT curve for inverter-based DG units (Fig.
reported. In addition, in all simulations with radial 3(a)) has been used in determination of protection
network topology tdip_sensitivity_fwd is chosen to be 0 ms in operation time delays (Table II), because it was assumed
the following (see Eq. (1)) and tdip_sensitivity_rev is chosen to that the disconnected DG units were inverter-based.
be 60 ms in the following (see Eq. (2)). Based on the simulations over-current pick-up/start
value in forward direction could be 160 A (in order to
detect all MV faults, but not to operate in LV side faults
IV.1. Case 1 (Radially Operated MV Feeders)
i.e. when back-up protection for LV side protection is not
In Fig. 9 the location of the simulated faults in case 1 needed).
with radially operated MV feeders is shown.
TABLE II
In Table I the fault types for case 1 with radially
OPERATION TIME DELAYS TFWD FOR DIFFERENT CBS IN CASE 1
operated MV feeders (Fig. 9) are presented. WITHOUT DG UNITS (FIG. 9 AND TABLE I AND EQ. (1) AND (2))
Operation time delays tfwd for different CBs in Fig. 9
A1 A2a A2b B1 B2
tfwd tfwd tfwd tfwd tfwd
Fault
[ms] [ms] [ms] [ms] [ms]
F4 661 - - - -
F5 953 107 - - -
F6 928 107*) 35*) - -
F7 1987 1008 568 - -
F8 - - - 178 -
F9 - - - 2230 781
*)
Here in dual-breaker scheme also blockings between A2a and A2b could be
beneficial to ensure selectivity
Case 1 (Radial) – Inverter-based DG Units

In Table III, the operation time delays in forward (tfwd)
and reverse (trev)direction for different CBs are presented
during different faults (Table I and Fig. 9) from case 1
with inverter-based DG units 1 and 2 (Fig. 9) according
to the proposed protection scheme (Figs. 2–4 and Eq. (1)
and (2)). DG units 1 and 2 are induction generators with
full-converter interface (Sn_inv = 1.65 MVA). Otherwise,
the used simulation model is to a large extent similar to
those used in [5]–[7]. Also in this case the over-current
pick-up/start value in forward direction was160 A.
From Table III it can be seen, when compared to
Table II, thattfwd operation time delays are quite closely
matched between cases without and with inverter-based
DG units. For example in F5 fault (Fig. 9),tfwdof CB A1
is a bit longer without DG units (Table II) and in F6 fault
Fig. 9. Case 1 with radially operated MV feeders tfwd is slightly shorter without DG units (Table II) than
and fault locations F1–F9 with inverter-based DG units (Table III).
104
H. Laaksonen
Operation time in reverse direction trev is in brackets in control with 40 ms time delay wouldn’t ever have an
Table III, and also on other following Tables, if islanding effect on the performance of the proposed scheme.
is not possible or feasible from point of view of power On the other hand, many of these ‘equal or less than
unbalance (Fig. 2). 40 ms frequency transients’ can be a result from nearby
fault. Therefore, another possibility to prevent
TABLE III unnecessary operation of P/f -droop control with very
OPERATION TIME DELAYS TFWDAND TREV FOR DIFFERENT CBS IN CASE 1 short operation time delay (less than 40 ms) could be
WITH INVERTER-BASED DG UNITS 1 AND 2 (FIG. 9 AND TABLE I AND
EQ. (1) AND (2)) simultaneous use of additional clause like Uph-ph_min<
Operation time delays tfwd and trev for different CBs 0.2*Upos_pu, where Uph-ph_min is the minimum phase-to-
in Fig. 9 phase voltage and Upos_puis the positive sequence voltage
A1 A2a A2b B1 B2 in per units. This additional clause would then prevent
tfwd /trev tfwd /trev tfwd tfwd /trev tfwd /trev unnecessary operation of P/f -droop control of LV
Fault
F1 o)
95X) o)
(173)X) - o)
95X) o)
(173)X) network connected DG units during nearby 3-phase
F2 o)
95X) o)
(173)X) - o)
95X) o)
(173)X) short-circuits. In following this kind of under-voltage
o)
F4 671 173X) - o)
214 o)
(571) blocking has been used to prevent P/f -droop control
o) o)
F5 945 107 - 283 (805) operation in large voltage dips.
F6 947 107*) 35*) o)
285 o)
(818)
F7 2013 1025 582 o)
1090 o)
(1820) In Table IV, the operation time delays in forward (tfwd)
F8 o)
95X) o)
(173)X) - 178 o)
173X) and reverse (trev) direction for different CBs are presented
o) o)
F9 1339 (2011) - 2211 768 during different faults (Table I and Fig. 9) from case 1
*)
Here in dual-breaker scheme also blockings between A2a and A2b with inverter-based DG units 1 and 2 (Fig. 9) when the
could be beneficial to ensure selectivity, X) DG unit 1 (Fig. 9) could get effect of P/f -droop control of LV network connected DG
disconnected and transition to island operation could fail due to
ENTSO-E NC RfG[9]LVRT requirements, o) Small power unbalance units with 0 ms operation time delay and under-voltage
(Punb, Qunb) => Islanding allowed i.e. operation in reverse direction if blocking is studied.
enough rapidly controllable DG, energy storage or controllable loads
connected to intended island part of the network (largest possible island TABLE IV
could be prioritized) OPERATION TIME DELAYS TFWDAND TREV FOR DIFFERENT CBS IN CASE 1
(P/F -CONTROL) WITH INVERTER-BASED DG UNITS 1 AND 2 (FIG. 9 AND
TABLE I AND EQ. (1) AND (2))
IV.2. Case 1 (Radial) – Effect of P/f -droop Control Operation time delays tfwd and trev for different CBs
of LV Network Connected DG Units in Fig. 9
A1 A2a A2b B1 B2
In this Section IV.2 the effect of P/f -droop control of tfwd /trev tfwd /trev tfwd tfwd /trev tfwd /trev
LV network connected DG units is presented. In the Fault
future, this P/f -droop control will be required by grid F1 o)
95 X) o)
(173)X) - o)
95 X) o)
(173)X)
o)
codes like [9] from all DG units. In the simulations the F2 95 X) o)
(173)X) - o)
95 X) o)
(173)X)
o)
effect of P/f -droop control of LV network connected DG F4 661 173 X) - o)
209 o)
(570)
o) o)
F5 958 107 - 287 (815)
units during over-frequency situations was simulated F6 926 107 *)
35 *) o)
280 o)
(796)
between 50.3 and 51.5 Hz by linearly increasing the F7 1989 1012 571 o)
1061 o)
(1795)
o)
loading in part of the MV/LV transformers. In the F8 95X) o)
(173)X) - 178 o)
173 X)
o) o)
simulations the initial assumption was that the intentional F9 1356 (2022) - 2211 768
*) X)
time delay of P/f -droop control of LV network , and o) See definitions from Table III
connected DG unit is 0 ms as also proposed in [10].
The reason for this was the fact that in theory with 0 When comparing the protection operation time delays
ms time delay the P/f -droop control might have an effect tfwd and trev between Table III and IV, it can be seen that
on the performance of the proposed protection scheme, they are quite close to each other. Some deviations exist
because the time delays are calculated 30 ms after pick- in faults F6 and F7 (Fig. 9) in trev and tfwdof CBs located
up. Otherwise, the simulated cases were similar with further from the fault (i.e. not closest to the fault).
Section IV.1.More details about some of these new grid
code requirements, like P/f -droop control, can be found IV.3. Case 1 (Radial) – Effect of Reactive Current
from [13]. Injection of MV Network Connected Inverter-
Also in this case with inverter-based DG units the based DG Units
simulations are done with 0 msP/f -droop control
operation time delay as proposed in [10]. However, in In following the effect of voltage support by reactive
[10] it has been stated that “the frequency protection current injection during voltage dips / faults with
shall be insensitive to frequency transients with duration inverter-based MV network connected DG units is
equal or less than 40 ms”. Therefore, from this point of presented. Simultaneously it has been assumed that P/f-
view one could conclude that also minimum reasonable droop control, studied in previous Section IV.2, is used
operation time delay for P/f -droop control could be 40 with 400 ms time delay. In this way it is also easier to see
ms. Because time delays in the proposed protection the only the effect of reactive current injection with
scheme are calculated 30 ms after pick-up, P/f -droop minimum operation time delay more clearly. This
reactive current injection -control is already required by
105
H. Laaksonen
some grid codes in certain countries and in the future it

will be required by grid codes like ENTSO-E RfG [9]
from DG units Type B and larger. In [9] it has been
stated that Type B converter based DG units shall fulfil
the following: a) With regard to reactive power capability
the DSO shall have the right to define the capability of a
DG unit to provide reactive power, b) The DSO/TSO
shall have the right to require fast acting additional
reactive current injection at the connection point in case
of symmetrical (3-phase) faults.
The target value of this additional reactive current
shall be reached with an accuracy of 10% within 60 ms
from the moment the voltage deviation has occurred [9].
In [10] the additional voltage support by reactive
current injection during faults is also required from MV
network connected converter/DFIG based DG units and
the requirement applies to all kinds of faults (1-, 2- and
3-phase). More detail scan be found from [9], [10] and
[13]. In this case the simulations are done with 0 ms
reactive current injection -control time delay, because in
principle it might affect on the performance of the
proposed protection scheme in which the time delays are
calculated 30 ms after pick-up. Otherwise, the simulated
case is similar to Section IV.1.
In Table V, the operation time delays tfwd and trev for Fig. 10. Case 1 with ring operated MV feeders
different CBs are shown during different faults (Table I and fault locations F1–F9 (Table I)
and Fig. 9) from case 1 with inverter-based DG units 1 During ring operation voltage dependent time delay in
and 2 (Fig. 9) when the effect of reactive current trev is equal to tfwd and not dependent from the islanding
injection(Iq_inj) -control of MV network connected DG possibility.
units with 0 ms operation time delay is studied. Possible intended island operation during ring
operation has not been considered as an option and
TABLE V
OPERATION TIME DELAYS TFWDAND TREV FOR DIFFERENT CBS IN CASE 1 therefore the proposed scheme (Fig. 2) applies in the
(IQ_INJ-CONTROL) WITH INVERTER-BASED DG UNITS 1 AND 2 (FIG. 9 AND same manner to reverse direction as in forward direction
TABLE I AND EQ. (1) AND (2)) during ring operation of MV feeders. This also means
Operation time delays tfwd and trev for different CBs in Fig. 9 that in all simulations with ring network topology
A1 A2a A2b B1 B2
tdip_sensitivity_fwd and tdip_sensitivity_rev are 0 ms (Eq. (1) and
tfwd /trev tfwd /trev tfwd tfwd /trev tfwd /trev
Fault (2)).
o)
F1 95 X) o)
(173)X) - o)
95 X) o)
(173)X)
o)
F2 95 X) o)
(173)X) - o)
95 X) o)
(173)X) Case 1 (Ring) – Inverter-based DG Units
o)
F4 678 173 X) - o)
217 o)
(641) In Table VI the operation time delays in forward (tfwd)
o) o)
F5 1000 107 - 288 (833)
F6 950 107 *)
35 *) o)
285 o)
(860) direction for different CBs are presented during different
F7 2003 1024 582 o)
1078 o)
(1815) faults (Table I and Fig. 10) and in Table VII in reverse
o)
F8 95X) o)
(173)X) - 178 o)
173 X) (trev) direction during ring operation of case 1 with
o) o)
F9 1343 (2013) - 2220 771 inverter-based DG units 1 and 2 (Fig. 10). During ring
*) X)
, and o) See definitions from Table III operation over-current pick-up value in both forward and
reverse directions was chosen to be 80 A.
From Table V can be seen, when compared to Table From Tables VI and VII it can be seen, that for
III, that the protection operation time delays tfwd and trev example during F7 fault (Fig. 10) CB A3 operates in
are also in this case quite similar. In general, most of 107ms (Table VII) and initially calculated operation time
deviations exist in faults F4, F5 and F6 (Fig. 9) in trev of delay in F7 fault for CB A2b is 933 ms (Table VI) to
CB B2 located furthest from the fault. However, this is disconnect the faulty line section.
natural because reactive current injection of DG unit 2 However, as stated in Section III.2, when CB with
(Fig. 9) increases the voltage at that point during the shortest operate time opens, time delays of the other
faults F4, F5 and F6 on the parallel feeder. IEDs needs to be re-calculated i.e. operation time delays
tfwdand trev are re-calculated after relatively large change
IV.4. Case 1 (Ring Operated MV Feeders) in magnitude of lowest phase-to-phase voltage
Uph_ph_min_pu(e.g. 0.15 pu) after initial under-voltage based
In Fig. 10 the location of the simulated faults (Table I) pick-up and before the end of initially calculated operate
in case 1 with ring operated MV feeders is presented. times.
106
H. Laaksonen
In this case it would mean that operation time delay of The proposed LVRT curve / network code compatible
A2b would increase (Table III) due to 0.1 pu increase in scheme enables the definition of protection operation
Uph_ph_min_puafter opening of A3 after F7 fault if large time delays during normal, grid connected operation for
enough (e.g. 0.15 pu) limit for change in Uph_ph_min_pu is intelligent electronic devices (IEDs) in MV distribution
not used. network having multiple protection zones without the
need for high-speed communication. The suggested new
TABLE VI
scheme is applicable for MV networks with different
OPERATION TIME DELAYS TFWDFOR DIFFERENT CBS IN RING
OPERATION OF CASE 1 WITH INVERTER-BASED DG UNITS 1 topologies (radial or ring). The proposed scheme takes
AND 2 (FIG. 10 AND TABLE I AND EQ. (1) AND (2)) also into account the possibility of transition to intended
Operation time delays tfwd for different CBs in Fig. 10 island operation. The proposed scheme is generally
A1 A2a A2b A3 B3 applicable with different LVRT curves defined in various
Fault
tfwd tfwd tfwd tfwd tfwd grid codes. Effect of other recent and forthcoming grid
F4 X) 760 - - - - code requirements was also studied in this paper.
F5 X) 1046 178 - - - Reason for this was to confirm that other
F6 X) 1046 178*) 142*) - - requirements, such as converter based DG units’ voltage
F7 o) 1798 1138 933 - - support during voltage dips by reactive current injection
F8 Y) -
- - - -
F9 o) 1798 1138 933 107*) 71*) (Iq_inj), active power/frequency (P/f)- or reactive
*)
During ring operation also blockings needed between A2a and A2b to power/voltage (Q/U) -control of DG units will not affect
maintain selectivity, X) DG unit 1, o) DG unit 2 or Y) DG units 1 and 2 to the proposed protection scheme negatively. However,
(Fig. 10) could get disconnected and transition to island operation could no significant effects were found in the simulations. As
fail due to ENTSO-E NC RfG [9] LVRT requirements one can see from the simulation results during radially
TABLE VII operated MV feeders (Fig. 9) in Tables II–V, the
OPERATION TIME DELAYS TREV FOR DIFFERENT CBS IN RING operation time delays in forward tfwd and reverse trev
OPERATION OF CASE 1 WITH INVERTER-BASED DG UNITS 1 AND 2 directions based on Fig. 3 and 4 may become sufficiently
(FIG. 10 AND TABLE I AND EQ. (1) AND (2)) long. Especially when the faults are at the end of the
Operation time delays trev for different CBs in Fig. 10
radial MV feeders like F7 and F9 faults in Fig. 9.
A2b A3 B1 B2 B3
Fault However, by using stricter tmin (and possibly also tmax)
trev trev trev trev trev
[ms] [ms] [ms] [ms] [ms] operation time delay curves than presented in Figs. 3 and
F4 X) Z)
71*) 197 *) 760 507 260*) Fig. 4 the operation time delays in this kind of faults at
F5 X) 71 231*) 1046 667 305*) the end of certain protection zone could be cleared much
F6 X) - 239*) 1046 675 315*)
F7 o) - 107*) 1798 881 143*)
faster. In Fig. 11 an example about choosing stricter tmin
F8 Y) - - 214 - - curve with inverter-based DG units than in Fig. 3(a)) is
F9 o) - - 1798 881 - presented.
*) X) o)
, , and Y)See definitions from Table VI,
Z)
35 mstrevin A2a*)
On the other hand, in fault F8 CB B1 (Fig. 10)

operates in 214ms (Table VII), but no initial operation
time delay for B2 has not been calculated (Table VI)
because over-current pick-up limit, 80 A in the ring
operated topology, was not exceeded. However, under-
voltage based pick-up limit has been exceeded and after
opening of B1, also over-current pick-up limit will be
exceeded (415 A) and tfwd for B2 (Fig. 10) will be
determined30 ms after this “new” pick-up (119ms) as
Fig. 11. Voltage dependent time delay curve in forward direction for
well as for B3, A3, A2b, A2a and A1 (Table VIII). short-circuit protection when islanding is NOT possible with inverter-
based DG units and when chosen tmin curve is than in Fig. 3 (see also
TABLE VIII Eq. (1))
OPERATION TIME DELAYS TFWDFOR DIFFERENT CBS IN F8 FAULT
AFTER OPENING OF B1 (FIG. 10) In general, it is also worth mentioning that in order to
Operation time delays tfwd for different CBs in Fig. 10
support future active network management schemes with
A1 A2a/A2b A3 B3 B2
tfwd tfwd tfwd tfwd tfwd network topology changes (e.g. from radial to ring), the
Fault proposed scheme requires some central control which
F8o) 2723 2091/1850 768 561 119 could be realized for example at HV/MV substation
o)
See definition from Table VI levelling a similar manner as in [3] where grid
automation controller COM 600 and low-bandwidth
communication issued to change the setting groups of
V. Discussion and Conclusions protection relays based on status of CBs. However, high-
In this paper a future-proof directional short-circuit speed communication for protection purposes with the
protection scheme for Smart Grids has been proposed. proposed scheme would not be needed.
107
H. Laaksonen
References Authors’ information

[1] F. Coffele, M. Dolan, C. Booth, G. Ault, G. Burt, Coordination of ABB Oy, Medium Voltage Products, Vaasa, Finland.
Protection and Active Network Management for Smart
Distribution Networks, CIRED Workshop 2012, May 29-30 2012, H. Laaksonen was born in Vaasa, Finland, on
Lisbon, Portugal. November 22, 1977. He received his MSc
[2] Z. Jiang, J. Fan, J. Mcdonald, Advanced Distribution Automation degree (2004) in Electrical Power Engineering
Management for Active Distribution Systems, CIGRE Session from Tampere University of Technology and
2012, 26-30 August 2012, Paris, France. PhD degree (2011) in Electrical Engineering
[3] H. Laaksonen, D. Ishchenko, A. Oudalov, Adaptive Protection from University of Vaasa. His employment
and Microgrid Control Design for Hailuoto Island, IEEE experience includes working as a research
Transactions on Smart Grid, vol. 5, May 2014, pp. 1486 – 1493. scientist in University of Vaasa, VTT Technical
[4] Laaksonen, H., Kauhaniemi, K., Smart protection concept for LV Research Centre of Finland and Institute of Power Engineering
microgrid, (2010) International Review of Electrical Engineering department at Tampere University of Technology. Currently he works
(IREE), 5 (2), pp. 578-592. with ABB Oy, Medium Voltage Products in Vaasa, Finland. His fields
[5] H. Laaksonen, Advanced Islanding Detection Functionality for of interest are protection of Smart Grids, integration and active
Future Electricity Distribution Networks, IEEE Transactions on management of distributed energy resources in smart distribution
Power Delivery, vol. 28, October 2013, pp. 2056 – 2064. networks and development of new functionalities and algorithms for
[6] Laaksonen, H., Enhanced multi-criteria-based passive islanding future Smart Grid concepts (e.g. microgrids).
detection scheme, (2014) International Review of Electrical
[7] Laaksonen, H., Comparison of possible future passive islanding
detection methods, (2013) International Review of Electrical
[8] A. J.Roscoe, G. M. Burt, C.G. Bright,Avoiding the Non-Detection
Zone of Passive Loss-of-Mains (Islanding) Relays for
Synchronous Generation by using Low Bandwidth Control Loops
and Controlled Reactive Power Mismatches, IEEE Transactions
on Smart Grid, vol. 5, March 2014, 602 – 611.
[9] ENTSO-E, ENTSO-E Network Code for Requirements for Grid
Connection Applicable to all Generators (NC RfG), March 2013.
[Online]. Available: http://networkcodes.entsoe.eu/connection-
codes/
[10] CENELEC, CLC/FprTS 50549-2 Requirements for the
connection of a generating plant to a distribution system - Part 2:
Connection to a MV distribution system, Final Draft Version,
May 2014.
[11] BDEW, Technical Guideline - Generating Plants Connected to the
Medium-Voltage Network, 2008, Germany. [Online]. Available:
http://www.bdew.de/bdew.nsf/id/DE_7B6ERD_NetzCodes_und_
Richtlinien/$file/BDEW_RL_EA-am-MS-Netz_engl.pdf
[12] Technical standard CEI 0-16, Technical conditions for electricity
distribution grid connection with a nominal voltage of greater than
1 kV, December 2012, Italy.
[13] Laaksonen, H., Securing passive islanding detection and enabling
stable islanding with Q/f -droop control of DG unit, (2014)
International Review of Electrical Engineering (IREE), 9 (3), pp.
592-602.
[14] V. Gabrion, F. Colas, L. Karsenti, Risk of Stabilisation of
Islanding Situations by Local Regulations, CIRED 2014
Workshop, 11-12 June, 2014, Rome, Italy.
[15] CIGRE (Joint Working Group B5/C6.26/CIRED), Protection of
Distribution Systems with Distributed Energy Resources, Final
Report, 2014.
108
Voltage Instability Analysis on PV and QV Curves for Radial-Type

and Mesh-Type Electrical Power Networks
Marizan Sulaiman1, Ahmad Fateh Mohamad Nor1, Noor Ropidah Bujal2
Abstract – The analysis of voltage instability in electric power system is very crucial in order to
maintain the equilibrium of the system. This paper presents the analysis of voltage instability of
electric power system by using power-voltage (PV) curve and reactive power-voltage (QV) curve.
This research focuses on the voltage instability analysis using PV and QV curves for radial-type
and mesh-type power networks.
For radial-type power network, the plotting of PV and QV curves will be done by using Graphical
User Interface (GUI) based on MATLAB application software. In the other hand, the power flow
analysis for mesh-type power network will be done by using Power World Simulator and the
plotting of PV and QV curve will be done by using Microsoft Excel. Copyright © 2015 Praise
Keywords: GUI, Power World Simulator, PV and QV Curves, Voltage Instability
C Capacitance Voltage stability can be defined as the ability of a
GUI Graphical user interface power system to keep the voltage at all busses in the
IS Sending-end current system remain steady right after the system is being
IR Receiving-end current subjected to a disturbance. Voltage instability in the other
L Inductance hand is the opposite of voltage stability. Voltage
P Real power instability occurs due to the failure of the power system
PR Receiving-end real power to supply ample power to cover the increased demand of
PRmax Maximum receiving-end real power load. In addition, the characteristics of the load can also
PV Power-voltage contribute to voltage instability [1]-[3].
Q Reactive power Basically, there are three characteristics of loads.
QR Receiving-end reactive power Those characteristics are constant power load, constant
QRmax Maximum receiving-end reactive power current load and constant impedance load [3], [4].
QV Reactive-power voltage The load characteristics are depicted in the form of
R Resistance graph in Fig. 1 [4]. From Figure 1, it can be seen that the
SR Receiving-end apparent power power supplied to constant power load is constant even
SRmax Maximum receiving-end apparent power though the voltage of the load fluctuates. If the voltage
Vgap Voltage gap drops, more current will be drawn in order to maintain
VR Receiving-end voltage the amount of power. Most of industrial electrical motors
Vrated Rated voltage are of constant power load.
VRcri Critical receiving end voltage Fig. 1 also shows that the power consumed by
Vreg Voltage regulation constant current load is proportional to voltage. There are
VRFL Full-load receiving end voltage only a few electrical devices that are of constant current
VRNL No-load receiving end voltage load such as video tape recorders [3]. Lastly, constant
VS Sending end voltage impedance load uses power in proportional to voltage
Y Admittance squared. In other words, electrical devices that are of
Z Series impedance constant impedance load require less power when the
δcr Delta critical voltage is low. Incandescent lamps are one of many
δ'cri Prime critical angle examples of this type of load [3].
δ’ Delta prime Furthermore, Fig. 1 also depicts that the allowable
ϕ Power factor angle system voltage range is within ± 5% of rated voltage, i.e.,
ϕ' Phi Prime 1.0 per unit. This research focuses on constant power
load characteristics since this type of load is favorable to
voltage instability.
109
Marizan Sulaiman, Ahmad Fateh Mohamad Nor, Noor Ropidah Bujal
Fig. 2. Transmission Line Equivalent Circuit

Fig. 1. Load curves for constant power, constant current
and constant impedance characteristics The data for this example are presented as follows
[10]:
Hence, the analysis of voltage instability should be
Z = 24.89 + j112.3 ohm
implemented in order to make sure that the voltage level
Y = 0.00015 + j0.0013 S
at all busses is at stable state. A number of methods to
Type of transmission line = medium
analyze voltage stability have been proposed including
Power factor angle of load = 18.2° (The power factor
the PV and QV curve method [5]-[9]. PV curve and QV
angle is considered positive for lagging and negative for
curve are widely used for this purpose [8].
leading power factor.)
Therefore, the PV and QV curve for both radial-type
Power factor = 0.95
and mesh-type power network will be presented in this
Rated Voltage = 275 kV
paper. The method of plotting PV and QV curve for
radial-type power network will be done by using GUI
based on MATLAB [10]. Meanwhile the method of II.2. PV and QV Algorithm for Radial-Type
plotting PV and QV curve for mesh-type power network Electrical Power System
will be done with the help of Power World Simulator [11]
and Microsoft Excel. According to [10], [15], the values that are used for
plotting the PV and QV curve are transmitted power (P),
receiving-end voltage (V), and reactive power injection
II. Radial-Type Electrical Power System (Q). The algorithm for plotting the PV and QV curve for
radial-type electrical power system has been developed
In electrical power system, a radial-type electrical by [10] and is summarized as follow:
power system can be transmission line or distribution
line [12]. These two types of lines are distinguished by A. Determine the critical point. The critical point is
its length and operating voltages. calculated as;
In Malaysia, the transmission line operates at 132 kV, i. Determine power factor angle:
275 kV and 500 kV; meanwhile, the distribution line
operates at 11 kV, 22kV and 132kV [13]. The length of cos   p  f
transmission is between 80 km to 250 km while the (3)
length of distribution line is less than 80 km.   cos -1  p  f 
II.1. Example of Transmission Line ii. Using the power factor angle, calculate the phi prime,
ϕ' using;
Transmission line has been chosen for this study. The
line parameters that are related to transmission system  '  180         (4)
are Resistance (R), Inductance (L) and Capacitance (C).
Fig. 2 [14] shows the equivalent circuit of the
transmission line. From these R, L and C values, the iii. Determine the maximum receiving-end apparent
value of ABCD parameters can be obtained by using Eq. power, SRmax:
(1) [10], [14]:
Vrated 2
ZY S Rmax  (5)
 ZY   ' 
A  D  1 ; B  Z ; C  Y 1   (1) 4 AB sin 2  
2  4  2
In matrix form, the generalized constant ABCD vi. SRmax is used to determine the maximum receiving-
parameters are: end power, PRmax :
VS   A B  VR 
 I   C D   I R 
(2) PR max  S R max cos  (6)
 S
110
v. SRmax is also used to determine the maximum C. Determine sending-end and receiving-end current,
receiving-end reactive power, QRmax : voltage and power, voltage regulation, voltage gap
and line current.
QR max  S R max sin  (7) i. The first step is to calculate the receiving-end current,
IR :
vi. Then, determine the receiving-end voltage at a critical PR max
point, VRcri : IR  (17)
3 Vrated   cos  
Vrated
VRcri 3 phase  (8) ii. To calculate the sending-end voltage the following
 ' 
2 A sin   calculation has been done:
2
VRcri 3 phase  ZY 
A  D  1  ; B  Z
VRcri 1 phase  (9)  2 
3 (18)
 ZY 
C  1   ; Y  G  jB
vii. The last step is to calculate prime critical power  2 
angle, δ’cri :
' iii. Substitute Eq. (18) into the matrix Eq. (19) to
 'cri  90  (10) determine sending-end voltage (Vs) and sending-end
2
current (Is) for medium-length transmission line.
and the critical power angle, δcri can be obtained by
using:   ZY  
 cri   'cri   (11)  1  2  Z  V 0
Vs       R1 ph
 Is     (19)
  Y  1  ZY   1  ZY   I R    
B. Determine the other points. The other points are     
  4   2 
calculated as;
i. Calculate the delta prime,  '1 and  '2 :
iv. Calculate the sending-end power, Ps:
 '1   ' cr1   ;  '2   ' cr2   (12)
Ps  3VsL-L Is cos  Vs   Is  (20)
ii. Then calculate the receiving-end apparent power, SR1
and SR2: v. Receiving-end power, PR can be obtained by using:
Vrated 2 sin  '   '1  sin  '1 PR  3 VR  I R  cos  0    (21)

S R1  2
AB sin  '
(13) vi. Calculate percentage of voltage regulation at full
2
Vrated sin  '   '2  sin  '2
SR2  load:
AB sin 2  '
VRNL - VRFL
% Vreg   100% (22)
iii. By using the value of SR1 and SR2, determine the VRFL
receiving-end real power, PR1 and PR2 :
vii. Voltage gap between rated voltage and critical point
PR1  S R1 cos  ; PR 2  S R 2 cos  (14) is calculated using:
iv. Calculate the receiving-end reactive power, QR1 and Vrated  Vcri
QR2 : Vgap   100% (23)
Vrated
QR1  S R1 sin  ; QR 2  S R 2 sin  (15)
v. The last step is determining the receiving-end II.3. Graphical User Interface (GUI)
voltage, VR1 and VR2 :
The GUI for analyzing voltage instability for radial-
type electrical power system developed by [10] has been
Vrated sin  '   '1  improved. In the latter version, the users only need to
VR1 
A sin  ' enter the values of R, XL and Y instead of ABCD
(16)
Vrated sin  '   '2  parameters. The GUI based PV and QV calculations are
VR 2  depicted in Fig. 3.
A sin  '
111
TABLE III
TRANSMISSION LINE DATA
From To Limit
R X B B*2
Bus Bus (MVA)
7 8 .0085 .072 .0745 .149 150
9 8 .0119 .1008 .1045 .209 150
7 5 .032 .161 .153 .306 150
9 6 .039 .17 .179 .358 150
5 4 .01 .085 .088 .176 150
6 4 .017 .092 .079 .158 150
TABLE IV
LOAD DATA
Fig. 3. GUI of voltage instability analysis Bus P (MW) Q (MVAR)
for radial-type electrical power system 5 125 50
6 90 30
8 100 35
III. Mesh-Type Electrical Power System
For mesh-type electrical power system, the voltage The methodological steps of plotting the PV and QV
instability analysis process involves a series of power curve for the 9-bus electric power system are listed as
flow solutions. The value of real power (MW) will be follows [3], [15], [18]:
increased constantly for every series of power flow. i. Select a bus. The PV and QV curve will be plotted at
The value of voltages of the bus will change as a this selected bus.
result of the series power flow [15]. Fig. 4 shows the ii. Run the power flow program by using Power World
mesh-type power network that consists of 9-bus power Simulator.
system constructed in Power World Simulator [16], [17]. iii. Increase the value of real power of load (P) by 0.1 per
The data for bus the 9-bus power system as shown in unit in order to plot PV curve. Similarly, increase the
Fig. 4 are presented in Table I until Table IV [16], [17]. value of reactive power of load (Q) by 0.1 per unit in
order to plot QV curve. Then run the power flow
TABLE I program again. A new voltage value at the selected
GENERATOR DATA bus will be obtained. Record the values of P or Q and
Rated Power Rated P Rated Q voltages.
BUS kV
MVA Factor (MW) (MVAR)
1 247.5 16.5 1.00 247.5 0 iv. Repeat step (iii) until the value of desired P or Q
2 192.0 18.0 0.85 163.2 101.1 cannot be delivered by the system.
3 128.0 13.8 0.85 108.8 67.4 v. Repeat step (iii) again for another several times to
obtain the lower point of the PV and QV curve.
TABLE II vi. Finally, use all the recorded value of P or Q obtained
TRANSFORMER DATA
Limit
in step (iii) – (v) and plot it against the voltage of the
From Bus To Bus R X Tap selected bus by using Microsoft Excel.
(MVA)
4 1 0 0.0576 1.00 300 The above steps will be applied on to Bus 5, Bus 6
7 2 0 0.0625 1.00 300 and Bus 8 of the 9-bus power system.
9 3 0 0.0586 1.00 300
Fig. 4. 9-bus electric power system
112
IV. Results and Discussion Fig. 7 shows that the critical voltage points for both
PV and QV curves are 0.7032 per unit and 0.5301 per
IV.1. Results for Radial-Type Electrical Power System unit, respectively. The system will reach critical voltage
Case 1: Results for lagging power factor (18.2°). The point if the power load is increased until 515 MW or if
PV and QV curve is shown in Figure 5. the reactive power load is increased up to 310 MVAr.
Figure 5 shows the PV and QV curve at 0.95 lagging The green lines represent the allowable range of system
power factor which is a good power factor. Real power voltage within ± 5% of rated voltage, that is to say ± 5%
and reactive power are being represented by the black voltage regulation. The system voltage will operates
curve and red curve, respectively. The values of real and within ± 5% range if the value of real power (P) and
reactive power are positive because the loads that operate reactive power (Q) of load do not exceed 305 MW and
at lagging power factor absorb both real and reactive 100 MVAr, respectively.
power. The voltage regulation is 24.3 % and the value of Similarly, Fig. 8 shows critical voltage points for both
voltage gap is 38.5%. The rated voltage (blue line) is PV and QV curves generated at Bus 6 at 0.7394 per unit
operated far from the critical voltage. This is a very good and 0.5259 per unit, respectively. These points can be
situation in order to prevent voltage instability. Voltage reached if the power or reactive power load demand are
instability will occur if the rated voltage operates at or increased up to 470 MW or 299 MVAr, respectively. In
near the critical voltage. order to make sure that the system voltage always
operates within the range of ± 5% of rated voltage, both
Case 2: Results for leading power factor (-18.2°). The of the real power or reactive power of load demand must
PV and QV curve is shown in Figure 6. not exceed 290 MW and 90 MVAr, respectively.
Fig. 6 shows the PV and QV curve at 0.95 leading Finally, Fig. 9 shows voltage instability can occur at
power factor. It can be seen from this curve that the value Bus 8 if the real power or reactive power of load demand
of real power is positive but the value of reactive power rises until 510 MW (corresponding to V = 0.8381 per
is negative. This is due to the loads that operate at unit) and 115 MVAr (corresponding to V = 0.6812 per
leading power factor absorb real power but generate unit), respectively. The system voltage will operate
reactive power. The rated voltage (blue line) also below the ± 5% allowable range if the value of real
operates far from the critical voltage. The voltage power of load demand exceeds 370 MW or the value of
regulation is 13.9 % and the value of voltage gap is reactive power of load demand exceeds 115 MVAr.
20.9%. Lower voltage regulation is due to the voltage It has been shown in Fig. 7, Fig. 8 and Fig. 9 that the
rise at the receiving-end (current injection) and lower increase of real power (P) or reactive power (Q) of load
voltage gap is susceptible to voltage instability compare will force the bus voltages to decrease. Table V shows
to the lagging load conditions. the details of Fig. 7, Fig. 8 and Fig. 9 in terms of the
incremental values of P and Q to reach critical voltage,
ΔP and ΔQ. On the other hand, the incremental values of
IV.2. Results for Mesh-Type Electrical Power System P and Q to reach the minimum limit of allowable system
The generated PV and QV curves for Bus 5, Bus 6 and voltage range (that is 0.95 per unit), ΔP1 and ΔQ1.
Bus 9 are shown in Fig. 7, Fig. 8 and Fig. 9, respectivel.
Fig. 5. PV and QV curve at 0.95 lagging
113
Fig. 6. PV and QV curve at 0.95 lagging
Fig. 9. PV and QV curves at Bus 8

Fig. 7. PV and QV curves at Bus 5
TABLE V
CHANGE OF P AND Q AT LOAD BUSSES
BUS ΔP (MW) ΔQ (MVAR) ΔP1 (MW) ΔQ1 (MVAR)
5 380 220 180 40
6 370 230 210 60
8 410 310 280 80
V. Conclusion
Voltage instability analysis is an important parameter
for monitoring the bus voltage in the electrical power
system. The bus voltage must not operate close to the
critical voltage point at nose point of PV and QV curves.
For lagging loads condition of radial-type power
system, both real and reactive powers of load are
positive. This indicates the load absorbs both P and Q.
On the contrary, in leading load condition, the real power
is positive but the reactive power is negative. This means
that during leading load condition, the load absorbs P but
Fig. 8. PV and QV curves at Bus 6 generates Q.
114
In addition, lagging load conditions have a wider [Online]. Available: http://www.tnb.com.my/business/for-

housing-developers-electrical-contractors/tnb-electricity-
range of voltage gap which represent a much better
system.html. [Accessed: 13-Oct-2014].
voltage operating range before reaching its critical value. [14] H. Saadat, Power System Analysis (McGraw-Hill Inc., 2004).
For the mesh-type electrical power system, lagging [15] P. Kundur, Power System Stability and Control (McGraw-Hill
loads conditions at load busses are considered. The value Inc., 1994).
[16] P. M. Anderson, A. A. Fouad, Power System Control and
of real power P at load busses can be increased much
Stability (John Wiley & Sons, Inc., 2003).
more compared to the value of reactive power Q. [17] W. Ongsakul, D. N. Vo, Artificial Intelligence in Power System
That is to say, the electrical power system can tolerate Optimization (Taylor & Francis Group, LLC, 2013).
an increase in load demand of P much more rather than Q [18] B. Aydin, Voltage security assessment using P-V and Q-V curves,
before reaching both lower limit of allowable operating Master's thesis, Department of Electrical and Electronic Eng.,
Bahcesehir University, Istanbul, Turkey, 2008.
voltage (0.95 per unit) and the critical voltage at nose
point of PV and QV curves. The studies conducted using
Power World Simulator have shown that the system will Authors’ information
experience exceeding of MVA loadings for transmission 1
Faculty of Electrical Engineering, Universiti Teknikal Malaysia,
lines and transformers as the demands of P and Q Melaka.
increase at load busses even before the system voltage
reach its critical value at nose point. 2
Electrical Engineering Department, Merlimau Melaka Polytechnic,
Malaysia.
References Marizan Sulaiman was born in Kuala Besut,

Terengganu, Malaysia on April 16, 1962. He
[1] P. Kundur, J. Paserba, V. Ajjarapu, G. Andersson, A. Bose, C. obtained B.Sc. in Electrical Engineering in
Canizares, N. Hatziargyriou, D. Hill, A. Stankovic, C. Taylor, T. 1984, M.Sc. in Electrical Engineering in 1985
Van Cutsem, V. Vittal, Definition and Classification of Power and Ph.D. in Electrical Engineering in 1989
System Stability IEEE/CIGRE Joint Task Force on Stability from the University of Missouri, USA. He has
Terms and Definitions, IEEE Transactions on Power Systems, held various administrative and academics posts
vol. 19, n. 2, August 2004, pp. 1387 – 1401. including the Deputy Dean of academic and
http://dx.doi.org/10.1109/TPWRS.2004.825981 students’ development at the School of Electrical & Electronic
[2] C. W. Taylor, Power System Voltage Stability (McGraw-Hill, Engineering, Universiti Sains Malaysia (USM), Penang. He is currently
Inc., 1994). the Dean of Faculty of Electrical Engineering, Universiti Teknikal
[3] M. S. Chen, Energy Systems Research Center (University of Malaysia Melaka (UTeM), Malaysia. He has published 8 books,
Texas at Arlington, 1996). authored and co-authored more than 55 journal articles and authored
[4] H. L. Willis, Power Distribution Planning Reference Book and co-authored more than 85 conference papers. His research interests
(Marcel Dekker, Inc., 2004). include power systems, energy efficiency and control & automation.
[5] G.K. Morison, B. Gao, P. Kundur, Voltage Stability Analysis Prof. Marizan is a registered member of Board of Engineers, Malaysia
using Static and Dynamic Approaches, IEEE Transactions on (BEM) and a graduate member of Institutes of Engineers, Malaysia
Power Systems, vol. 8, n. 3, August 1993, pp. 1159 – 1171. (IEM).
http://dx.doi.org/10.1109/59.260881 E-mail: marizan@utem.edu.my
[6] B. Gao, G.K. Morison, P. Kundur, Voltage Stability Evaluation
using Modal Analysis, IEEE Transactions on Power Systems, vol. Ahmad Fateh Mohamad Nor was born in
7, n. 4, Nov 1992, pp. 1529 – 1542. Kuala Lumpur, Malaysia on the 16th of June
http://dx.doi.org/10.1109/59.207377 1988. He received his bachelor degree and
[7] B. Gao, P. Kundur, G. K. Morison, Towards The Development of master’s degree in Electrical Engineering from
A Systematic Approach for Voltage Stability Assessment of Universiti Teknikal Malaysia Melaka (UTeM),
Large-Scale Power Systems, IEEE Transactions on Power Malaysia in 2011 and 2013, respectively.
Systems, vol. 11, n. 3, August 1996, pp. 1314 – 1324. Currently he is a Ph.D. candidate at the Faculty
http://dx.doi.org/10.1109/59.535672 of Electrical Engineering in Universiti Teknikal
[8] A. A. Khan, A Simple Method For Tracing PV Curve Of A Malaysia Melaka (UTeM), Malaysia. His research interests include
Radial Transmission Line, World Academy of Science, electrical power systems and solar electricity. Mr. Fateh is a registered
Engineering and Technology, vol. 2, n. 11, July 2008, pp. 347 – member of the Board of Engineers, Malaysia (BEM).
350. E-mail: ahmadfatehmohamadnor@gmail.com
[9] Chattopadhyay, T.K., Banerjee, S., Chanda, C.K., A simple
approach on voltage stability index of distribution systems for Noor Ropidah Bujal was born in Temerloh,
loads of different types, (2014) International Review of Electrical Pahang, Malaysia on the 1st of October 1980.
Engineering (IREE), 9 (5), pp. 1002-1011. She received her Bachelor of Electrical
http://dx.doi.org/10.15866/iree.v9i5.915 Engineering with Honours from Universiti Tun
[10] Noor Ropidah Bujal, Marizan Sulaiman, Rosli Omar, A graphical Hussein Onn Malaysia (UTHM), Johor,
user interface (GUI) for voltage stability analysis, Conference In Malaysia in 2009 and Master of Electrical
Education - Technical Vocational Education and Training ~Cie- Engineering from Universiti Teknikal Malaysia
TVET 2014~, August 25-26, 2014, Mersing, Johor, Malaysia. Melaka (UTeM), Malaysia in 2014. She is a
[11] PowerWorld Corporation, “PowerWorld Simulator 18 Online lecturer at Electrical Engineering Department, Merlimau Melaka
Help,” 2014. [Online]. Available: Polytechnic, Malaysia. Her main research interest is electrical power
http://www.powerworld.com/WebHelp/. [Accessed: 23-March- systems. Mrs. Noor Ropidah is a registered member of the Board of
2014]. Engineers, Malaysia (BEM).
[12] Millar, R.J., Saarijarvi, E., Lehtonen, M., Hyvarinen, M., E-mail: noorropidah@pmm.edu.my
Niskanen, J., Hamalainen, P., Electricity distribution network
planning algorithm based on efficient initial and radial-to-full
network conversion, (2013) International Review of Electrical
[13] Tenaga Nasional Berhad, “TNB Electricity System,” 2014.
115
P&O and Incremental Conductance MPPT Implementation
Sadik Farhat, Rachid Alaoui, Abdelilah Kahaji, Lahoussine Bouhouch, Ahmed Ihlal
Abstract – The maximum power point (MPP) of a photovoltaic (PV) system varies with
temperature and solar irradiation. The implementation a technical control of the PV power
converter systems to search and make this systems to operate at maximum output power is
required. The realization of the aforementioned converter and digital implementation of this
particular control are presented with experimental results. Hill climbing techniques, which allows
tracking the MPP in real time, is implemented.
This technique involves in the implementation and comparison of the efficiency of both methods:
"Perturb and Observe" (P&O) and the Incremental Conductance (InC). The proposed techniques
are validated using the experimental data of 40Wp PV panel (PVP), controlled by a boost
converter. Following the comparison of both chosen techniques, output power efficiency and
dynamic response are discussed. We also present the advantages and drawbacks of two techniques
by using the performed measurements results. Copyright © 2015 Praise Worthy Prize S.r.l. - All
rights reserved.
Keywords: MPPT, P&O Algorithm, Incremental Conductance, Photovoltaic Power Generation,

Boost Converter, Data Acquisition, MATLAB
Nomenclature The conversion efficiency of solar energy depends on

the intermittent meteorological conditions [1]-[2] such
α Duty cycle temperature, dirt and total or partial shading conditions
A Diode ideality factor [3]-[4], and essentially the solar radiation. It also depends
G Irradiance (W/m2) on the connected load type at the PV panel.
h Radiation proportionality parameter (Am2/W) Because of its nonlinear electrical characteristics,
Io Diode saturation current (A) varying load and meteorological conditions, the MPP is
Id Diode current (A) generating for a single operating point [5]. If this
Iopt MPP output PV current (A) conditions varies, the MPP is changing also and it must
Iph Photocurrent (A) be redefined. The PV systems must be designed to
Ipv PV current (A) operate at their peak output power to extract and
Ish Shunt resistor current (A) maintain this peak, even when climatic conditions are
Kb Boltzmann constant (1.38×10-23 J/°K) unfavorable. The location of MPP is not known. The
ns Cell number in series tracking of the correct MPP is a challenge and becomes
PMPPT Mean value of measured MPP (W) more complicated especially under low irradiation.
q Electron charge (1.6×10-19 C) The PV modules operate with a very low efficiency
Ro Resistor load () [6] when they are directly connected to the load. So, the
Rs Intrinsic resistors () electrical tracking is essential and the MPPT must be
Rsh Equivalent parallel resistor () used into the PV systems to increase their efficiency. This
Tj Junction temperature (°K) is achievable by inserting between the PV and the
Vopt MPP output PV voltage (V) electrical load a power adapter, such as a boost converter.
Vpv Output PV voltage (V) In literature, several methods and MPPT control
Vt Diode thermodynamic potential (V) algorithms are proposed [7]-[8]. This paper presents the
hill climbing MPPT techniques using two methods: P&O
[9] and incremental conductance [10]. Our paper is
I. Introduction organized as follows: After a brief introduction, in
Solar energy is a practically unlimited resource, section 2 we describe the test bench and the measurement
available everywhere. techniques.
The photovoltaic system offers a safe and clean energy In section 3, we expose the PV array modeling.
source, consequently an ecological environment. The PV Section 4, presents the theory of used boost converter.
is one of the renewable energies that attract the major Section 5 exposes the used MPPT algorithms "Perturb &
challenges for scientists and engineers. Observe" and "Incremental conductance".
116
S. Farhat, R. Alaoui, A. Kahaji, L. Bouhouch, A. Ihlal
The experimental tests and the obtained results of PVP Ipv Io

are discussed in section 6. Finally, we conclude our
experimental results. DC/DC
PVP Vpv Boost
Vo Ro
II. Test Bench and Measurements Load

Technique
P&O or InC Drive
The meteorological data such as irradiation G, r
temperature T, are delivered from the data logger CR10X Carrier αref
Campbell, installed at the Higher School of Technology +
in Agadir (Fig. 1). - Comparator
Fig. 2. Implemented system topology
Fig. 1. Meteorological station and data logger CR10X [11]
This station is equipped with various sensors for the

acquisition of global and diffuse radiation, the ambient
temperature, the humidity, speed and wind direction. The
measured values can be presented in various forms
(average, maximum, minimum etc.), depending on the
chosen configuration [11].
The used PVP for validation of our approach and our Fig. 3. Experimental records
algorithms have the specifications summarized in Table I
[11]-[12]. This PVP is fixed on a sloping steel structure. 3
TABLE I Iopt
Ipv (A)
ELECTRICAL CHARACTERISTICS OF PVP [12]

Specifications Values 1
Pnmax: Nominal maximum power 40 W
Voc: Nominal open circuit voltage 21.5 V 0
Isc: Nominal short-circuit current 2.48 A 0 8 Vpv (V) Vopt 20
Vmpp: Nominal maximum output voltage 16.9 V
Impp: Nominal maximum output current 2.34 A 40
MPP
Tn: Nominal temperature 25 °C
Pmax
Gn : Nominal irradiance 1000 W/m2
Ppv (W )
20
We use a test bench that enables us the evaluation of
the efficiency and limitations of two proposed techniques
(P&O and InC). For the validation, we implement and we 0
test their algorithms that appropriate on a PV system 0 8 Vpv (V) Vopt 20
implanted in laboratory.
Both techniques are based on algorithms, which are Fig. 4. Experimentals points for G = 973 W/m2 andT = 24.6°C
implemented under MATLAB environment. Fig. 2
shows the block diagram of system under test. As can be observed there is a single MPP on the curve
Figure 3 presents the realized test bench corresponding to the optimum operating point for each
corresponding to this block diagram of the system. irradiation and temperature value.
In order to obtain the Ipv = f(Vpv) and Ppv = f(Vpv) This MPP corresponds to the maximum power Pmax,
curves, we proceed to a fast variation from 1% to 99% of which is associated to an optimum voltage, Vopt and
boost converter duty ratio (), by fixed step. Typical current, Iopt.
curves thus recorded, are in the form of those shown in To track this point, P&O and InC algorithms are
Fig. 4. implemented in MATLAB environment.
117
III. PV Array Modeling Ipv D
L Io
The literature cites many mathematical models [14]
Vo
representing the highly nonlinear behavior of PVP,
Ce
caused by the photovoltaic effect of semiconductor PN MOSFET Cs Ro
Vpv Pulse
junction [15].
Of this literature review, we retain the complex
physical PV model which is represented by the
equivalent electrical circuit at one diode as shown in Fig.
5. The mathematical expression (1) reflecting the Fig. 6. Boost converter
electrical behavior of this model is given by:
The MPP is reached when Vpv = Vopt and Ipv = Iopt.
I pv  I ph  I d  I sh  Therefore, the optimum duty ratio is given by the Eq.
(5):
  V pv  Rs  I pv   Vopt
 h  G  I o exp  q    1  (1)  opt  1  (5)
  A  ns  K b  T j  
   Ro  I opt
V pv  Rs  I pv

Rsh Given 0 ≤ ≤ 1, the matching condition (6) is realized
when:
Different formula variables (1) are summarized in the Vopt
Ro  (6)
nomenclature. I opt
Ip The elements values of our realized boost converter

v
Id
are given in Table II.
Ish Rs The chopping frequency is 25 kHz.
Iph Vpv TABLE II
Rsh
ELECTRICAL PARAMETERS VALUES OF THE BOOST CONVERTER
Specifications Values
Ce: Input condenser 270 μF
L: Inductor 1 mH
Fig. 5. PV electrical model Cs: Output condenser 220 μF
Ro: Load resistor 30 
IV. Boost Inverter

Schematic of the used boost converter is shown in Fig. V. Hill Climbing Algorithm
6. It is dimensioned to optimum operation of GPV under
test. The MPP of PVP varies with the solar irradiation and
The Load adaptation [16] is based on the automatic cells temperature. So the use of MPPT algorithms is
search the appropriate value of the duty cycle, in order necessary to extract its maximum power.
to maximize the PV output power, because the optimal Among these algorithms, during our investigations,
power transferred to the load cannot be maximized only we opted on the exploitation and the development of two
for this single duty ratio value. below described algorithms.
This boost converter is described by the Eqs. (2) and
(3): V.1. P&O Algorithm
V pv
Vo  (2) A complete review of 19 different MPPT algorithms is
1 provided in [17]. The technique using the P&O algorithm
[18], is the most commonly used because it has the
I o  1    I pv (3) advantage of an easy implementation.
Unfortunately it has also disadvantages [19] that are
From these equations we deduce the PV resistor Rpv listed in section 6. Indeed, this algorithm causes
depending on  and Ro, what gives the relationship (4): disturbances of the duty ratio ensuring control to the
power converter. This disturbances cause an increase or
V pv decrease of the voltage of the DC interface circuit
2
R pv   1    Ro (4) between the PVG and the power converter, as illustrated
I pv in Fig. 7. Fig. 8 shows the flowchart of the first
implemented P&O algorithm.
118
40 V.2. Incremental Conductance

MPP
Our second choice is the incremental conductance
30 algorithm (InC) [20]. This algorithm is based on the
dP/dV<0
instantaneous power derivative calculation to determine
Ppv(W)
20 the MPP [21]. Indeed, observing the Eqs. (7) and (8), the
Increase dP/dV>0
Vpv
MPP is reached when the slope of P-V curve is zero [22]:
10 Decrease
dPpv dI pv
Vpv Ppv  V pv  I pv   I pv  V pv (7)
dV pv dV pv
0
2 4 6 8 10 12 Vopt 1718 20
Vpv(V) dPpv dI pv I pv
0  (8)
dV pv dV pv V pv
Fig. 7. P&OMPPT
Therefore, the used equations analysis above allows

verifying if the PVP is operating at its MPP or far from it
Measure: Vpv(k),Ipv(k) [23].
Practically we notice that the MPP is determined by
comparing instantaneous conductance Ipv/Vpv to the
incremental conductance ΔIpv/ΔVpv.
P(k)=Vpv(k).Ipv(k)
The equations that are involved in this algorithm are
as follows (9):
ΔP(k)=P(k)-P(k-
I pv I pv
 for V pv  Vopt  At MPP
1) V pv V pv
I pv I pv
 for V pv  Vopt  Left MPP (9)
ΔV(k)=Vpv(k)-Vpv(k- V pv V pv
1)
I pv I pv
 for V pv  Vopt  Right MPP
V pv V pv
No Yes
ΔP(k) >
0 The implemented algorithm under MATLAB
environment, calculates dPpv/dVpv. This calculation is
based on the measurement of the PV voltage, and its
ΔV(k) > ΔV(k) >0 incremental output power.
0
If dPpv/dVpv is not close zero, the controller adjusts the
No No
boost converter's duty ratio, step by step until the PVP
Yes Yes
reaches its maximum output power.
α =α - Δα α =α + Δα α =α + Δα α =α - Δα
This implemented InC algorithm, is described by the
above flowchart in Fig. 9.
VI. Experimental Results

k= k + 1 The implemented MPPT performances are studied for
a sunny days with solar irradiance levels ranging from
600 W/m2 to 1000 W/m2. The experimental database was
registered during year 2014. For validation, we measure
Fig. 8. Flowchart of the P&O algorithm the maximum power for each temperature and solar
irradiation values; afterwards we compare it with the
It describes in detail the technique to disturbance of output power extracted by both MPPT techniques; P&O
the PVP operating point. This consists in modifying the and InC.
duty cycle, by a small step-size  and measure the PV As mentioned above, to evaluate the efficiency and
array output power before and after the perturbation. limits of two proposed techniques, we have implemented
If the power increases, the algorithm continues to and tested these two algorithms on PV system located in
perturb the system to the same effect. our laboratory, using a control system whose flowchart is
Else, the system is perturbed to cause opposite effect. shown in Fig. 8 and Fig. 9.
119
integrated to our PV system.

40
Sample Vpv(k),
Ipv(k) a
Power (W)
20
30
ΔV=Vpv(k)-Vpv(k-
1) 28
14 15 16
0
0 5 10 15 20
Vpv (V)
ΔI=Ipv(k)-Ipv(k-1)
40
b
power (W)
31
20 30.5
No Yes 30
ΔV= 0 29.5
16 18 20
0
No Yes 0 2 4 6 8 10 12 14 16 18 20 21.5
Yes No
ΔI/ΔV= -I/V ΔI= 0 Time (s)
Figs. 10. Power (a) and response (b) by P&O algorithm

for G = 973 W/m2 and T = 24.6°C
ΔI/ΔV> - ΔI > 0 40
I/V
30
power (W)
a
31
Yes No No Yes 20
30
α =α - Δα α =α + Δα α =α + Δα α =α - Δα 10
14 14.5 15 15.5
0
0 5 10 15 20
Vpv (V)
40
k=k+1
30
Power(W)
30.8
30.6
20
b 30.4
30.2
10 30
Fig. 9. Flowchart of the Incremental conductance algorithm 29.8
18 19 20 21
0
The used PVP for measurement is exposed to various 0 2 4 6 8 10 12 14 16 18 20 21.5
irradiance levels, and temperature, which vary with the Time (s)
climatic conditions, throughout the experimental
Figs. 11. Power (a) and response (b) by InC algorithm
measurements. The system control algorithm is for G = 973 W/m2 and T = 24.6 °C
implemented under MATLAB environment that
generates code for tracking the MPP in real time. Regarding the dynamic response variation of the
The results of the experimental measurements of the system, Figures 10(b) and 11(b) respectively show a
proposed InC technique will be presented and compared typical response of controllers P&O and InC, for given
with those to traditional P&O method. The experimental values of solar irradiation and temperature. Each of two
tests and measurements are carried by scanning the PVP studied system controllers evaluates the PVP output
electrical characteristic for different voltage levels power, and adjusts the power interface to satisfy the
ranging from 3 V to 19 V. This prevents the PVP to optimal operating conditions.
operate in short-circuit and open circuit. Further, the As shown in Fig. 10(b), in the case of P&O controller,
MPP is located inside this voltage interval. the MPP is achieved at steady state, after a little more
By performing a fast measurement, during than 2 s. While, as illustrated on Fig. 11(b), the MPP is
approximately 20 s, of Ppv = f (Vpv), under irradiance and reached in less than 2 s for the InC controller.
fairly uniform temperature conditions, we obtain the PVP During our investigations, the tests are carried for
powers shown in Figs. 10(a) and 11(a).Under the same solar irradiations varying from 600 W/m2 to 1000 W/m2
experimental conditions, we observe that for both approximately, and temperature values ranging from
proposed methods; P&O and InC, the obtained results 24°C to about 36°C. In these conditions, to evaluate the
converge towards the maximum power point and two studied techniques efficiency; P&O and InC, the
oscillate around this MPP. The PVP powers marked by instantaneous efficiency Effis calculated following the
dotted lines in these figures are obtained by varying duty formula (10). This last is the ratio between PMPPT point
cycle value of signal controlling the boost converter and the MPP.
120
The PMPPT point represents the average value This situation is reversed when the MPPT operates
measured, at steady state, of the extracted power by the with a smaller step. So, a compromise between the
used techniques. The fact of considering this average dynamics and oscillations must be satisfied for the fixed
value allows eliminating the effect of measurement low step MPPT.
noise. For low irradiances levels we observe that the P&O
P efficiency is less than the InC technique. Contrariwise,
E ff  100  MPPT (10) for high irradiances levels, the P&O is more efficient
MPP
compared with the InC technique. We conclude that both
According to Fig. 12 which represents the obtained techniques; P&O and InC, are most efficient methods to
results concerning the efficiency Eff of the two techniques track the MPP. However, we can improve efficiency by
studied; P&O and InC, we estimate that these techniques developing a hybrid algorithm using these two methods
ensure broadly similar controls. and optionally with a variable step.
However, according to this Fig. 12, for low powers, The performances summary of studied MPPTs would
the MPPT of InC type is more efficient compared with serve as a useful guide to choose the appropriate MPPT
P&O. The situation is reversed for the power values method for engineers of photovoltaic industry.
greater than or equal to 28 W. However, as the evaluated
efficiency by (10), is in the range from 97 % to 99 %, we References
can conclude that both methods; P&O and INC, are
highly efficient for MPP tracking. [1] Kavitha, R., Thottungal, R., Cascaded multilevel inverter for
stand alone PV system with maximum power point tracking
technique, (2012) International Review of Electrical Engineering
100 (IREE), 7 (6), pp. 5939-5943.
[2] Khiari, B., Sellami, A., Andoulsi, R., Mami, A., A non linear
Efficiency (%)
98 MPPT control of photovoltaic pumping system based on discrete

sliding mode, (2012) International Review of Electrical
96
InC MPPT Engineering (IREE), 7 (6), pp. 6129-6136.
P&O MPPT [3] K.Ishaque, Z.Salam, A review of maximum power point tracking
techniques of PV system for uniform insolation and partial
94 shading condition, Renewable and Sustainable Energy Reviews,
Vol. 19,2013, pp. 475-488.
92 [4] K.H.Hussein, I.Muta, T.Hoshino, M.Osakada, Maximum
20 22 24 26 28 30 photovoltaic power tracking: an algorithm for rapidly changing
MPP (W) atmospheric conditions, Generation, Transmission and
Distribution, IEE Proceedings, Vol. 142, Issue 1, 1995, pp. 59-64.
Fig. 12. MPPT efficiency [5] M. Berrera, A. Dolara, R. Faranda, S. Leva, Experimental test
ofseven widely-adopted MPPT algorithm, PowerTech
Conference, IEEE Bucharest, June 28, July 2-2009, pp. 1-8.
VII. Conclusion [6] R.R. King, A. Boca, W. Hong, X.-Q. Liu, D. Bhusari, D.
Larrabee, K.M.Edmondson, D.C. Law, C.M. Fetzer, S.
The PV system is studied under the same profiles of Mesropian, N.H. Karam, Band Gap-Engineered Architectures for
high-Efficiency Multi-junction Concentrator Solar Cells, 24th
irradiation and temperature. During our work, we note European Photovoltaic Solar Energy Conference, Hamburg,
that, on the one hand, the two methods; P&O and InC, Germany, 21-25 September 2009, pp. 55-61.
converge to the MPP, throughout the experiment. On the [7] P.Midya, P.T.Krein, R.J.Turnbull, R.Reppa, J.Kimball, Dynamic
other hand, the proposed methods scan all the existing Maximum Power Point Tracker for Photovoltaic Applications,
Power Electronics Specialists Conference, PESC'96 Record,27th
MPPs. Note that the important aspect of the implemented Annual IEEE, Vol. 2, Baveno23-27 Jun 1996, pp. 1710-1716.
algorithms principle is their independence of PVP [8] V. Salas, E. Olías, A.Barrado, A.Lázaro, Review of the maximum
parameters. The efficiency of both adopted techniques is power point tracking algorithms for stand-alone photovoltaic
within the range from 97 % to 99 %. According the systems, Solar Energy Materials and Solar Cells, Vol. 90, Issue
11, 2006, pp. 1555-1578.
experimental results, the P&O and InC algorithms have [9] Tsai-Fu Wu, Chh-Lung Shen, Chien-Hsuan Chang, Jeiyang Chiu,
very similar output power performance, with a slight 1φ3W grid-connection PV power inverter with partial active
advantage in terms of dynamic response time for the InC power filter, Aerospace and Electronic Systems, IEEE
algorithm (about 1.5 s). Transactions on, Vol. 39, Issue 2, 2003, pp. 635-646.
[10] O.Wasynezuk, Dynamic Behavior of a Class of Photovoltaic
Both implemented algorithms perform a continuous Power Systems, Power Apparatus and Systems, IEEE
scan of the P-V characteristic at a value close to the MPP. Transactions on, Vol.PAS-102,Issue 9, 1983, pp. 3031-3037.
However, we observe that the measured power point [11] Farhat, S., Alaoui, R., Kahaji, A., Bouhouch, L., MPPT efficiency
oscillate around the MPP due to varying the duty cycle test by neural networks and P&O algorithm, (2013) International
Review of Electrical Engineering (IREE), 8 (5), pp. 1548-1555.
by a fixed step. [12] KYOCERA Corporation, KC40, High Efficiency Multicrystal
The step size significantly affects the fast tracked Photovoltaic Module,
MPP. Indeed, a greater rapidity can be obtained with a http://www.kyocerasolar.com/assets/001/5158.pdf
bigger step. But, in this case the operating point will [13] National Instruments Corporation©, USB Data Acquisition,
http://sine.ni.com/np/app/main/p/bot/no/ap/daq/lang/en/pg/2/sn/n1
oscillate broadly around the MPP. Consequently, the 7:daq,n24:USB/
peak-to-peak size of this oscillations increases, which [14] HairulNissahZainudin, SaadMekhilef, Comparison Study of
leads to low efficiency and instability of the system. Maximum Power Point Tracker Techniques for PV Systems,
121
Farhat, R. Alaoui, A. Kahaji, L. Bouhouch,
S. Farhat, Bouhouch, A. Ihlal
Proceedings of the 14th International Middle East Power Systems Abdelilah Kahaji Teacher of Electrical
Conference, MEPCON'10,
MEPCON'10, Cairo University, Egypt Egypt, December Engineering, with a Bachelor (1993) in
19-21,
19 21, 2010, pp. 750-755.
750 755. electronics of Higher Normal School of
[15] Weidong Xiao, W.G. Dunford, P.R. Palmer, A. Capel, Technology (ENSET), Mohammadia, Morocco.
Application of Centered Differentiation and Steepest Descent to Master of Business Administration (MBA)
Maximum Power Point TrackingTracking,, Industrial Electronics, IEEE (2002) at the University of Muncton Canada
Transactions on on, Vol. 54,Issue
54,Issue 5, 2007, pp. 2539
2539-2549.
2549. and a post graduate degree in (2005), of
[16] Henry Shu Shu-Hung
Hung Chung, K.K. Tse, S.Y. Ron Hui, Industrial Engineering in the National School of
C.M.Mok,M.T.Ho,A
C. M.Mok,M.T.Ho,A Novel Maximum Power Point Tracking Applied Sciences (ENSA) of Agadir, Moro
Morocco.
cco. His research, in the
Technique for Solar Panels Using a SEPIC or Cuk Converter,
Technique Converte context of national doctoral thesis, focuses on the thematic of
Power Electronics,
Electronics IEEE Transactions on, Vol.18,
Vol. Issue 3, 2003, Renewable Energies. The doctoral investigations took place in the
717 724.
pp. 717-724. Research Team in Advanced Technologies and Engineering of
[17] T. Esram, P.L. Chapman, Comparison of Photovoltaic Array Renewable Energies (ERTAIER) Agadir, Moroc
Morocco.
co.
Maximum Power Point Tracking Techniques, Energy Conversion,
IEEE Transac
Transactions
tions on,
on, Vol. 22, Issue 2, 2007, pp. 439-449.
439 449. Lahoussine Bouhouch Professorof
Professorof higher
[18] N. Femia, G. Petrone, G. Spagnuolo, M. Vitelli, Optimization of education at the ESTA (High School of
Perturb
erturb and O Observe
bserve maximum power point tracking method, Technologies of Agadir), IbnZohr University,
Power Electronics, IEEE Transactions on, on Vol. 20, Iss
Issue
ue 4, 2005, Agadir, Morocco. PhD Electrical Engineering at
963 973.
pp. 963-973. the Nancy I University, France in 1988 and state
[19] M.A.Eltawil, Z Z.Zhao,MPPT
Zhao,MPPT techniques for photovoltaic doctorate in Electrical Engineering in 2007.
applications
applications, Renewable and Sustainable Energy Reviews, Reviews, Vo ol. Responsible of the res
research
earch team ERTAIER
25, 2013
2013,, pp. 793
793-813
813. (Research Team inn Advanced Technologies and
[20] Chihming, Comparative study of peak
Hua Chihchiang, Shen Chihming, Engineering of Renewable Energies). His research focuses on topics
powertracking techniques for solar storage systems, Applied related to renewable energy, instrumentation and electromagnetic
Power Electr
Electronics
onics Conference
Conference and Expo Exposition,
sition, APEC
APEC'98.
'98. compatibility (EMC).
Conference Proceedings IEEE IEEE,, Thirteenth Annual
Annual, Vol.
Vol
2,Anaheim
Anaheim
Anaheim, CA,CA 15-19
15 19 Feb 1998, 697 685.
1998 pp. 697-685 Ahmed Ihlal, was born and brought up iinn
[21] D.P.Hohm
D.P. Hohm, M.E.Rop
Hohm, M.E.Ropp, p, Comparative Study of Maximum Morocco. He studied Physics and Chemistry
PowerPoint Tracking Algorithms Using an Experimental, and holds, in 1984, his BSc degree (Licence
(LicenceEs--
Programmable, Maximum Power Point Tracking Test Bed Bed, Sciences
ciences Physique)
Physique) in Solid
olid State
tate Physics
Physics from
Photovoltaic Specialists Conference, Conference Record of the the University Mohamed V, Rabat - Morocco
Morocco..
Twenty Eighth IEEE,
Twenty-Eighth IEEE, Anchorage, AK, 15--22 22 Sept. 20002000, He then joined Paris VII University–
University– France,
France,
1699 1702.
pp. 1699-1702. where he got, in 1985, a MSc. ddegreeegree (DEA:
[22] Issam Houssamo, Fabrice Locment, Manuela Sechilariu, Sechilariu Diplome des Etudes Approfondies) in Solar
Maximum power tracking for photovoltaic power system: Energy. He pursued his research on the studies and got, in 1988, a PhD
Development
evelopment and experimental comparison of two algorithms, degree from the University of Caen BasseNormandie - France. Dr. A.
Renewable Energy,
Energy, Vol.
Vol 35, Issue 10, 2010, pp. 2381-2387.
2381 2387. Ihlal started his teaching career on I988 as Assistant Professor in the
[23] Li Jiyong, Wang Honghua
Honghua, A n novel
ovel stand-alone
stand alone PV generation faculty of Science at University IbnZohr. Then he holds a
system based on variablestep size INC MPPT and SVPWM "Doctoratd'Etat" thesis in 1995. He is currently Full Professor in
control Powe
control, Powerr Electronics and M otion Control Conference,
Motion Faculty of Sciences, University IbnZohr, Agadir - Morocco. He is head
IPEMC'09, IEEE 6th International
IPEMC'09, International,, Wuhan, 17 17--20
20 May of the group working on ddeveloping
eveloping cost effective processes for the
2009, pp. 2155
2009,pp. 2155--2160
2160. fabrication of CIGS and CZTS absorber layers, buffer layers and TCOs.
He is working on PV and CSP systems as well. He has published 60
scientific papers, and acted as a referee for numerous international
Authors' information
Authors' journals. He has contributed to ththee organization of numerous national
and international conferences and was a member of scientific
Sadik Farhat Associate Professor ((2001) 2001) committees for several international conferences. He is supervising
inElectrical
inElectrical Engineering with a Bachelor (1996) PhD, MSc as well as BSc students in the field of PV and CSP. He is an
in electronics in Higher Normal School of expert of the CNRST iinn the field of renewable energies.
Technology (ENSET), Rabat, Morocco and a
post graduate degree in (2007), of Energy and
Environment in the National School of Applied
Sciences (ENSA) of Agadir, Morocco. His
research, in the context of national doctoral
thesis, focuses on the thematic of Renewable Energies. The doctoral
investigations took place in the Research Team in Advanced
Technologies and Engineering of Renewable Energies (ERTAIER)
Agadir, Morocco
Agadir, Morocco.
My Rachid El Moutawakil Alaoui Professor

at the University IbnZohrAgadir,Morocco, with
a Bachelor in Electronics ENSET, Rabat and
Masters in Industrial Engineering at the ENSA
Agadir. His doctoral thesis focused on the
Defensive Software processors. Permanent
member of the Research T Team
eam of Advanced
Technologies and Engineering Renewa Renewable
ble
Energy ERTAIER. Responsible for electromagnetic compatibility
(CEM) research focus.
122
Optimization FOU of Interval Type-2 Fuzzy Inference System

Using Big Bang – Big Crunch Algorithm for Short Term
Load Forecasting on National Holiday Case Study:
South and Central Kalimantan-Indonesia
Akhmad Ramadhani1, Agus Dharma2, Imam Robandi1
Abstract – The application interval type-2 fuzzy inference systems (IT2FIS) has been attention
for a short-term load forecasting problems solution. This paper present optimization membership
function of antecedent (X,Y) and consequent (Z) interval type-2 Fuzzy Logic System using Big
Bang – Big Crunch Algorithm for application short term load forecasting on national holiday.
This method has implemented on the historical peak load data during 14 national holiday case
study in South and Central Kalimantan – Indonesia electrical power system on year 2008. The Big
Bang–Big Crunch (BB–BC) algorithm will applied to optimization interval footprint of uncertainty
(FOU) membership functions of interval type-2 fuzzy logic system. BB-BC Algorithm is a global
optimization algorithm and has a low computational cost and a high convergence speed, so very
efficient in this research. The test result showed main absolute percentage error (MAPE) less than
interval type-2 fuzzy inference system (FIS) and type-1 fuzzy inference system (FIS). Finally, this
paper defined main absolute percentage error (MAPE) 0.58045% for type-1 FIS, 0.53906% for
interval type-2 FIS, and 0.52421% for optimization interval type-2 FIS-Big Bang Big Crunch
Algorithm. Copyright © 2015 Praise Worthy Prize S.r.l. - All rights reserved.
Keywords: Interval Type-2 Fuzzy Inference System, Big Bang - Big Crunch Algorithm, STLF,
MAPE
BB-BC Big Bang – Big Crunch Load forecasting is very important in power system,
GA Genetic Algorithm because it could be estimate the power consumption at
IT2FIS Interval Type-2 Fuzzy Inference any given time. Accuracy in forecasting power loads can
System increase the security of power systems especially in the
IT2FLS Interval Type-2 Fuzzy Logic System operation of power systems such as power delivery (load
MAPE Main Absolut Percentage Error flow), maintenance unit and unit commitment.
MF Membership Function Therefore, the load forecasting can help in the
UMF Upper Membership Function reduction of operating costs and reliable case of electric
LMF Lower Membership Function power systems. Load characteristics of each area were a
STLF Short Term Load Forecasting different depending on the season and behavior of
Ã Type-2 Fuzzy Set society itself. Peak load on a special day such as a
Jx Primary membership of primary national holiday also has different characteristics from a
variable, x normal day for the particular holiday is always changing
FOU (Ã) Footprint of Uncertainty of IT2FS Ã either the day, date, month and season each year.
 A  x  ,  A  x  Lower MF, Upper MF of IT2FS Ã Generally, peak load on national holidays tends to
decrease compared to normal days throughout the year.
yl Switch point left In this study, there are 14 national holidays were been
yr Switch point right
c used in short-term load forecasting on the electrical
xi Center of Mass power system of South and Central Kalimantan-
f Fitness function Indonesia.
l Upper and lower limit parameter Professor Zadeh from California University first
r Normal random introduced fuzzy logic in 1965.
k Step iteration Fuzzy logic describing mathematical calculations
xnew Upper and lower bounded based on set theory to describe the uncertainty in the
N Population Size form of linguistic variables, in another sense the theory
123
Akhmad Ramadhani, Agus Dharma, Imam Robandi
of fuzzy logic to develop set theory Boolean (0 and 1) be II.1. Interval Type-2 Fuzzy Set
the set of which has a membership value which is
An interval type-2 fuzzy sets (IT2FS) Ã symbolized
continuous and ambiguity (between 0 and 1) [1].
by membership function μÃ where x∈X and u∈Jx⊆ [0,
Professor Zadeh first introduced the concept of
1] characteristics could be recognize in the following
uncertainty of the type-2 fuzzy sets in 1975 as a
equation [4]:
development of the concept of ordinary fuzzy sets of
"fuzzy-fuzzy" or fuzzy-fuzzy type 1 are at right again [1].
Due to a system of type-1 fuzzy logic knowledge base   A  x,u 
A xX xJ  x,u  Jx   0.1 (1)
is often used to build rules in a fuzzy logic system (FLS) x
was erratic. In this study developed a load forecasting

technique that has been studied previously using interval x is the primary variable has domain X; u∈U, secondary
type-2 fuzzy logic system (IT2FLS) for a national
variable, has domain Jx for every x∈X; Jx is called the
holiday [2], [4].
primary membership of x. The uncertainty of the Ã
However, in this study the membership function (MF)
expressed by the union of all primary membership (Jx)
which is the footprint of uncertainty will be optimized
was been called the footprint of uncertainty (FOU) of Ã
using big bang-big crunch algorithm [8]. This method
(see Fig. 1), i.e.
will tested using the data of peak load 14 a national
The equation shown as follows:
holiday last 4 years on the electrical power system of
South and Central Kalimantan-Indonesia.
FOU   
A  Jx   x,u  ;u  Jx   0,1 (2)
x X
II. Interval Type-2 Fuzzy Logic System
where Jx is an interval set, i.e.:
The concept of uncertainty of the type-2 fuzzy sets
was first introduced by Zadeh in 1975 as a development
of the concept of ordinary fuzzy sets of "fuzzy-fuzzy" or
the fuzzy type 1 in fuzzy again. Due to a system of type-
Jx   x,u  ;u   
A
 x  ,  A  x  (3)
1 fuzzy logic knowledge base is often used to build rules

in a fuzzy logic system (FLS) was erratic. So from Eq (3) FOU (Ã) can also expresses as:
There are three reasons why the uncertainty of the
rules can occur [1], [2]: FOU 
A         x  ,  A  x  
 A  (4)
1. The word used as the antecedents and consequents of x X
rules can have different meanings to different people.
2. Consequents obtained from polling a group of experts Jx = Primary membership of x
will often differ on the same rule because the experts  A = Lower Membership Function (LMF) from Ã;
do not necessarily all agree on the rules.
3. The training data that contain a lot of noise  A = Upper Membership Function (LMF) from Ã.
Uncertainty in the antecedent or consequent
uncertainty translated in the antecedent or consequent 1
membership functions. Fuzzy logic systems are type-1
fuzzy set membership function with the form of type-1,
cannot directly handle the uncertainties rules type-2
fuzzy logic system, while the antecedent or consequent UMF(Ã) UMF(Ã)
membership functions of type-2 fuzzy set, capable of
uncertainty rules.
Type-2 fuzzy set has a membership level that they FOU(Ã) FOU(Ã)
themselves are fuzzy.
A level in type-2 fuzzy set is a subset of secondary
membership. Same with the Type-1 Fuzzy Logic System, LMF(Ã)
Type-2 Fuzzy Logic System also include FIS
membership functions and defuzzification. The 0 x
difference is that before the defuzzification process there
Fig. 1. FOU (grey), LMF (line dotted), UMF (line solid)
are some the reduction process types (type reduction)
which have some methods like Kernik Mendel
Encryption algorithm (KMA), Mendel Kernik Enhance II.2. Structure Interval Type-2 Fuzzy Logic System
Encryption (EKMA), Enhance Kernik Mendel Algorithm
with Initialization (EKMANI), Iterative Algorithm with a Structure of Interval Type-2 Fuzzy Logic System
Stop Condition (IASC), and Enhanced Iterative (IT2FLS) in Fig. 2 below shows the mapping of
Algorithm with a Stop Condition (EIASC) [6]. processes IT2FIS from input value crisp x into output
that is expressed in equation Y=f(x).
124
k n N
Fuzzifier Defuzzifier
 n 1
f y n   n  k 1 f n y n
Output yl  mink1,N 1 n
Input k N
crisp x T1FS Y=f(x)  n1 f   nk 1 f n (7)
L n N
IT2 FS Inference Type-  n 1
f y n   n  L 1 f n y n
Engine Reducer  L n N
 n1 f   n L 1 f n
k n N n n
Rule Base  n 1
f n y   n  k 1 f y
yr  maxk1,N 1 k N n
 n1 f n   nk 1 f (8)
Fig. 2. Structure of Interval Type-2 Fuzzy Logic System (IT2FLS) R n N n n

 n1 f n y   n R 1 f y
R N n
II.3. Membership Function and Fuzzy Rules  n1 f n   n R 1 f
Fuzzy rules IF-THEN used in this method for
forecasting maximum load. where switch point of L and R:
In this paper input membership function (antecedent)
that is X ,Y and Output membership function y L  yl  y L 1
(consequent) is Z for short term load forecasting by (9)
R R 1
following by Eq. (5) as follows in: y  yr  y
IF X is Ai AND Y is Bi THEN Z is Ci (5)

The centroid obtained from this equation:
Fuzzy set Ai, Bi, and Ci takes eleven term set:
Negative Very Big (UNVB and LNVB)  yl  yr 
Centroid  (10)
Negative Big (UNB and LNB) 2
Negative Medium (UNM and LNM)
Negative Small (UNS and LNS)
II.5. Overview of Big Bang-Big Crunch Algorithm
Negative Very Small (UNVS and LNVS)
Zero (UZE and LZE) In this study, interval of Footprint of Uncertainty
Positive Very Small (UPVS and LPVS) (FOU) from Interval Type-2 Fuzzy Logic System in
Positive Small (UPS and LPS) shown Fig. 2 has been optimized using Big Bang-Big
Positive Medium (UPM and LPM) Crunch Algorithm. This optimization method has a low
Positive Big (UPB and LPB) computer cost and high convergence speed [8].
Positive Very Big (UPVB and LPVB) The Big Bang–Big Crunch (BB–BC) optimization
method developed by Erol and Eksin [8] consist of two
phases: a Big Bang Phase and a Big Crunch Phase.
II.4. Type Reducer with Kernik Mendel Algorithm The first step is big bang phase where candidate
Before the defuzzification process are the reduction solutions were random distributed over the search space.
process types (type reduction) which has some methods Randomness could see as equivalent to the energy
like Kernik Mendel Encryption algorithm (KMA), dissipation in nature while convergence to a local or
Mendel Kernik Enhance Encryption (EKMA), Enhance global optimum point viewed as gravitational attraction.
Kernik Mendel Algorithm with Initialization (EKMANI), Since energy dissipation creates disorder from ordered
Iterative Algorithm with a Stop Condition (IASC), particles, we will use randomness as a transformation
Enhanced Iterative Algorithm with a Stop Condition from a converged solution (order) to the birth of new
(EIASC) [6]. solution candidates (disorder or chaos). The next step is
In this paper, defuzzification process using the big crunch phase where a contraction procedure
centroid method in the Interval Type-2 FLS had calculates a center of mass for the population.
proposed by Kernik and Mendel [6]. All subsequent big bang phases were randomly
Procedures of proposed is: distribute about the center of mass or the best fit
individual in a similar fashion. After the Big Bang phase,
N a contraction procedure applied such as the Big Crunch
 f n yn phase to form a center or a representative point for
YCos  x'    n
n 1
N
  yl, yr  (6) further Big Bang operations. In this phase, the
contraction operator takes the current positions of each
n
n
f F
y n Y n
x 
'
f candidate solution in the population and its associated
n 1
cost function value and computes the center of mass.
125
The best fit individual, as stated in the center of mass, START

could be compute as:
 Calculate Peak load data from Day-4 up to MaxSD (MaxWD)
N 1  Calculate Load Differences (LDmax)
 i 1 f i
i
x
 Calculate Typical Load Differences (TLDmax)
xc  (11)  Calculate Variable Load Differences (VLDmax)
N 1
 i 1 f
i
 Define membership function Antecedent (X,Y) and Consequent
(Z) interval type-2 fuzzy logic system (IT2FLS)
where xc is the position of the center of mass, xi is the
position of the candidate within a p-dimensional search
space, fi is the fitness function value of the ith candidate,
p is the number of parameters to be optimized, N is the
population size in Big Bang phase.
The new candidates around the center of mass were NO Optimization of membership function Antecedent (X,Y)
been calculated by adding or subtracting a normal and Consequent (Z) interval type-2 fuzzy logic system
(IT2FLS) with Big Bang- Big Crunch Algorithm for
random number whose value decreases as the iterations define value of footprint of uncertainty (FOU)
elapse. It could be formalize as:
x new  xc  lr / k (12)
YES
where xc stands for center of mass, l is the upper limit of
 Define fuzzy rules interval type-2 fuzzy inference system
the parameter, r is a normal random number and k is the
iteration step. Then new point xnew was upper and lower  Apply Operation of AND into interval type-2 fuzzy inference system
 Apply Implication method MIN into FIS
bounded. The BB–BC optimization is approach takes the
 Apply aggregation method MAX into FIS
following steps [8]:
 Calculate defuzzification via type reducer using Kernik Mendel
 Step 1 Form an initial generation of N candidates in a
Algorithm for define value of Forecast VLDmax
random manner. Respect the limits of the search
space.
 Step 2 Calculate the fitness function values of all the  Calculate Forecast Load Differences
candidate solutions.  Calculate Load Forecast (Pmax forecast)
 Step 3 Find the center of mass according to Eq. (11).  Calculate Error Forecast and Main Absolute Percent Error
Best fitness individual could be chose as the center of

mass. AND
 Step 4 Calculate new candidates around the center of
mass by adding or subtracting a normal random Fig. 3. Flow chart implementation IT2FIS-BBBC Algorithm
number whose value decreases as the iterations elapse for STLF on national holiday
of using eq. (12).
 Step 5 Return to Step 2 until stopping criteria has Single line diagram of South and Central Kalimantan
been meeting. electrical power system obtained from PT PLN (Persero)
Region South and Central Kalimantan is like the Fig. 4.
Preprocessing peak load from the Day-4 to MaxSD on
III. Implementation of IT2FIS-BBBC the day of holidays based on the date used for forecasting
Algorithm for Short Term Load the peak load of each holiday [1], [3]. Peak load data
Forecasting on National Holiday from Day-4 up to MaxSD on national holiday based on
the date of the year 2005-2008 had used to forecast in the
Flow chart implementation of optimization IT2FIS- same holiday in 2008.
BBBC Algorithm into Short Time Load Forecasting on Firstly preprocessing data by computing MaxWD (i) is
national holiday presented on Fig. 3. the average maximum load of previous four weekdays
which is given on Eq. (13):
III.1. Preprocessing Peak Load South and Central
Kalimantan Power System Electricity WD i  h  4  WD i  h 3  
 
The data used in this research is quantitative data daily  WD i  h  2  WD i  h 1 
MaxWD i   (13)
electricity load in units of Kilowatts on the electrical 4
system of South and Central Kalimantan- Indonesia for a
national holiday and four days before the holiday in a Load Differences (LDs) for maximum loads on
span of four years each hour for 24 hours. holydays are define as the difference between the holiday
(MaxSD) and MaxWD:
126
Fig. 4. Single Line Diagram South and Central Kalimantan Electrical Power System
MaxSD  i   MaxWD  i  VLDMAX(i) for the same holidays on previous forecasted

LDMAX  i    100 (14) years, Y is VLDMAX(i) the previous holiday next to
MaxWD  i 
forecasted year and Z is forecast VLDMAX(i).
Optimization of FOU the candidate 44% until 54 %
The Typical Load Differences (TLDs) calculated by uncertainty by giving ± 22% until 27 % of their upper
averaging LDs of the same special day in the same day- and lower fuzzy type-1 membership function. As result
type collected from the historical load data. The TLDs of the BB–BC optimization search, the parameters of the
employed as base values for forecasting the maximum interval type-2 fuzzy sets in the fuzzy rule base are
loads for each a special day type. obtained with a cost function value of 50 % FOU can be
The Variation of Load Differences (VLDs) defined as seen on Fig. 6 and Fig. 7.
the amount of difference between the load behavior of Next, Ai, Bi and Ci were group into 22 regions upper
the special day and the typical behaviors of the same and lower of triangle membership function as mention on
special days with same day type. Then the values of Sub-section 2.3, and then some fuzzy rules bases made.
VLDmax were been calculated as shown in this equation: In 2008 there were 14 fuzzy rules show in Table II, one
example of choosing fuzzy set using max rule by taking
VLDMAX  i   LDMAX  i   TLDMAX  i  (15) the biggest value of corresponding membership degree
(µ) of variable input (X,Y) and output (Z) in Isra Mi’raj
Peak load data had used to calculate the VLDmax show in Table III. After all some basic requirement for
respectively and calculation the results from 2005 until variable input and output is determined then it was been
2008 was been presented on Table I. implemented on IT2FLS toolbox used in Matlab
software environment by typing it2fuzzy in command
line. This crisp value obtained finally called Forecast
III.2. Processing Optimization Interval Type-2 FIS VLDMAX (i).
Using Big Bang-Big Crunch Algorithm
Operation of Interval Type-2 FLS fuzzy set is III.3. Post Processing Optimization IT2FIS -
identical with an operation on Type-1 fuzzy set, but Big Bang-Big Crunch Algorithm
Interval Type-2 FLS fuzzy have FOU, UMF and LMF to
produce firing strength. Design of fuzzification of input After processing of interval type-2 FIS Forecast Load
for X and Y using IT2MF Editor, there were 2 trapezoidal Difference has calculated by Eq. (16) i.e.:
and 9 triangular membership functions within the range
of -6 to 6 for input processes, while all 11 triangular Forecast LDMAX  i  
membership functions is used for output Z. All input (16)
 ForecastVLDMAX  i   TLDMAX  i 
values of variables X, Y and output values of variable Z
by definition were changed into its values, where X is
127
Fig. 5. Comparing error for Application STLF with previous method
Peak Load Forecasted in year ith (MW) has calculated Indonesia electrical power system in year 2008 using
by Eq. (17) i.e.: optimization interval type-2 Fuzzy Inference System –
Big Bang Big Crunch algorithm is found on MAPE were
P'MAX  MaxWD  i   relative smaller than interval type-2 FIS, and type-1 FIS
(17) method for load forecasting.
 ForecastLDMAX ( i )  MaxWD( i ) This method forecasting using optimization interval

100 type-2 Fuzzy Inference System – Big Bang Big Crunch
algorithm have MAPE is less than 1.5%, This result was
The percentage of error is the difference between relatively small compare to standard MAPE permitted.
values with the calculation method that has real value.
The percentage of error had calculated by Eq. (18) i.e.:
P'MAX  i   MaxSD  i 
Error%   100 (18)
MaxSD  i 
IV. Result and Discussion

The calculation results of forecasting error IT2FLS-
BBBC Algorithm using data from various types of load
conditions in the holidays where this result is just a case
of forecasting in year 2008 show in Table IV.
The result application STLF on national holiday study Fig. 6. Membership function input variables X, Y
case South and Central Kalimantan-Indonesia electrical with optimization BBBC Algorithm
power system in year 2008 using optimization interval
type-2 fuzzy inference system-Big Bang–Big Crunch
Algorithm has been main absolute percentage error
(MAPE) less than interval type-2 fuzzy inference system
(FIS) and type-1 fuzzy inference system (FIS) method.
Main absolute percentage error MAPE in year 2008
using type-1 FIS is 0.58045%, interval type-2 FIS is
0.53906%, and optimization interval type-2 FIS-Big
Bang-Big Crunch Algorithm is 0.52421%. Comparing
error forecasting with previous method show in Fig. 5.
V. Conclusion
Fig. 7. Membership function output variables Z
Application sort term load forecasting on national with optimization BBBC Algorithm
holiday case study in South and Central Kalimantan-
128
TABLE I
CALCULATION VLDMAX FROM YEAR 2005 UNTIL 2008
Name Holiday Year in 2005 Year in 2006 Year in 2007 Year in 2008
Type of Holiday
MaxWD LDMAX MaxWD LDMAX MaxWD LDMAX TLDMAX VLDMAX MaxWD LDMAX TLDMAX VLDMAX
1.New Year 211 -0.947 221.355 -6.5483 213.663 3.4716 -3.7481 7.21968 241.275 -0.52844 -1.3415 0.81308
Public Holiday
2.Independence Day 217.28 -5.2098 229.8475 -0.3078 242.838 1.2817 -2.7588 4.04056 236.5875 1.01971 -1.412 2.43169
1.Idul Adha 213.0225 -13.962 224.375 -11.8529 211.10 10.4453 -12.9075 23.3528 219.725 -3.2199 -5.1233 1.9033
Islamic Holiday
2.Islamic New Year 212.62 -5.0606 229.5975 -5.4998 222.77 -1.0504 -5.2803 4.2298 245.85 0.46776 -3.8703 4.33807
3.Maulid 224.945 -2.0382 221.3575 0.1231 242.793 -3.12303 -0.9576 -2.1655 249.1 -1.3649 -1.6794 0.31449
4.Isra Mi'raj 227.565 2.2345 235.51 -1.6602 243.563 1.98614 0.28715 1.69899 229.6625 4.23992 0.85348 3.38644
5.Idul Fitri I 208.515 -3.5129 229.145 -2.0751 234.675 -3.5262 -2.79402 -0.7321 235.625 -6.77984 -3.0381 -3.74177
6.Idul Fitri II 206.2725 -0.4084 226.4025 -3.4419 231.788 -1.8713 -1.9252 0.0538 231.5125 -5.815 -1.9072 -3.9078
1.Good Friday 216.475 1.7669 224.0625 0.41841 236.35 -1.9589 1.09268 -3.0516 247.25 1.19312 0.07547 1.11766
Christian Holiday
2.Ascention 224.3075 -5.0054 223.09 3.9222 241.475 0.01035 -0.5416 0.55196 241.9125 -4.8416 -0.3576 -4.484
3.Crismas 195.9425 4.2780 214.8825 3.14009 243.713 -1.9131 3.70906 -5.6222 227.75 5.59824 1.835 3.76324
1.Silent Day 223.1175 -1.7065 223.1175 -1.7065 231.068 -5.07535 -1.7065 -3.3689 218.4625 1.7566 -2.82945 4.58605
Hindus /Budhis
2. Imlek 216.5612 -8.3585 228.73 2.0417 185.398 5.0176 -3.1584 8.17599 240.7125 3.98297 -0.43306 4.41603
/Tionghoa
3.Waisak 222.5 1.9326 235.385 -1.5315 237.713 3.48635 0.20053 3.28583 242.5375 -2.03576 1.2958 -3.3316
TABLE II
FUZZY RULES BASE FOR LOAD FORECASTING IN YEAR 2008
Antecedent Consequent
Number of Rules
X Y Z
1 PVB PS PVS
2 PB PVS PS
3 PVB NB PS
4 PB PS PB
5 NS PB ZE
6 PS ZE PM
7 NVS PM NB
8 ZE NB NB
9 NM NB PVS
10 PVS PVS NB
11 NVB NB PB
12 NM PB PVB
13 PVB PVB PB
14 PM PB NM
TABLE III
FUZZY RULES BASE FOR LOAD FORECASTING IN YEAR 2008
Holiday Forecast Membership function degree of Fuzzy Set (μ) Rules
Type of Holiday Variabel VLDMAX
in year 2008 NVB NB NM NS NVS ZE PVS PS PM PB PVB (MAX)
Input X 1.69899 0 0 0 0 0 0 0.3010 0.6989 0 0 0 PS
Islamic
Isra Mi’raj Input Y 0.31448 0 0 0 0 0 0.6855 0.3144 0 0 0 0 ZE
Holiday
Output Z 3.38643 0 0 0 0 0 0 0 0 0.6135 0.3864 0 PM
TABLE IV
THE RESULT APPLICATION STLF USING IT2FIS-BBBC ALGORITHM WITH PREVIOUS METHODE IN YEAR 2008
Interval Type-2 FLS-BBBC
Type-1 FLS Interval Type-2 FLS
Name of holiday in Algorithm
Type of holiday
year 2008 P’MAX Actual Error P’MAX Actual Error P’MAX Actual Error
(MW) (MW) (%) (MW) (MW) (%) (MW) (MW) (%)
1.New Year 240.457 240 0.19042 240.4560 240.00 0.19002 240.447 240 0.1864
Public Holiday
2.Independence Day 237.974 239 0.42912 238.1220 239.00 0.36735 238.090 239 0.38051
1.Idul Adha 212.858 212.65 0.08193 212.8597 212.65 0.09865 212.8586 212.65 0.09803
Islamic Holiday
2.Islamic New Year 243.851 247 1.27481 244.4134 247.00 1.04717 244.2581 247 1.05424
3.Maulid 248.076 245.70 0.96742 247.9633 245.70 0.92118 248.1045 245.70 0.91246
4.Isra Mi'raj 232.696 239.40 2.80031 233.2612 239.40 2.56422 233.7479 239.40 2.34654
5.Idul Fitri I 219.038 219.65 0.27851 219.0401 219.65 0.27765 219.4659 219.65 0.27786
6.Idul Fitri II 217.833 218.05 0.09948 217.8344 218.05 0.09884 217.8393 218.05 0.09895
1.Good Friday 249.913 250.20 0.11439 249.9128 250.20 0.11479 249.908 250.20 0.11647
Christian Holiday
2.Ascention 231.364 230.20 0.50579 231.3650 230.20 0.50611 231.365 230.20 0.50642
3.Crismas 241.043 240.50 0.22601 241.0426 240.50 0.22563 241.043 240.50 0.22601
1.Silent Day 223.2004 222.30 0.40504 223.1840 222.30 0.39767 223.184 222.30 0.39786
Hindus /Buddhist
2. Imlek 249.302 250.30 0.39846 249.3012 250.30 0.39904 249.302 250.30 0.39885
/Chines
3. Waisak 238.404 237.60 0.33846 238.4041 237.60 0.33846 238.404 237.60 0.33846
MAPE 0.58045 0.53906 0.52421
129
Acknowledgements Authors’ information

1
The authors wish a highly gratitude to Indonesia Department of Electrical Engineering, Sepuluh Nopember Institute of
Technology ITS Sukolilo, Surabaya.
Electrical Company PT.PLN (Persero), Area Pengatur
Postal code: 60111.
dan Pengaturan Beban (AP2B) Region South and Central
Kalimantan-Indonesia. We also sincerely thank to Prof. 2
Department of Electrical Engineering, Udayana University, Bali.
Oscar Castillo from Tijuana Institute of Technology, E-mails: dhaniel.electricianrsudulin@gmail.com
ag2bali@gmail.com
Mexico, who permitting us to use his Interval Type-2
Fuzzy Toolbox for used on Matlab, to support our Akhmad Ramadhani was born in Banjarmasin,
research. In addition, thanks to prof. Osman K Erol from Indonesia June 1st, 1984. He received the
Istanbul Technical University for guidance about Big Informatic Technology undergraduate degree
Bang-Big Crunch Algorithm. from STIMIK, Banjarbaru, Indonesia. He is
electrician and programmer on Hospital
Government at Province South Kalimantan,
Indonesia. He research interest is Artificial
References Intelligence, Renewable Energy, Intelligent
Power System Control and Management of
[1] J. M. Mendel and R. I. John, “Type-2 fuzzy sets made simple,” Energy. Since 2013, He is as student at electrical engineering magister
IEEE Trans. on Fuzzy Systems, vol. 10, no. 2, pp. 117–127, 2002. program in Sepuluh Nopember Institute of Technology.
[2] A.Dharma, I.Robandi, M.H.Purnomo,”Application of Short Term
Load Forecasting On Special Days Using Interval Type-2 Fuzzy Agus Dharma received Doctor in Electrical
Inference System; Study Case in Bali Indonesia”, Journal of Engineering from Sepuluh Nopember Institute
Theorical And Applied Information and Technology, Vol.49 no.2, of Technology in 2013. He is lecturer in
March, 2013 Electrical Engineering Department University of
[3] Soedibyo, Stephani, R., Aprilely, A.F., Ratih, M.S., Primaditya, Udayana, Bali, Indonesia. His research about
S., Suyanto, Power optimization for adaptive wind turbine: Case Compress Air System Optimization (CASO),
study on islanded and grid connected, (2014) International Management and Energy Conversion, Electrical
Review of Electrical Engineering (IREE), 9 (4), pp. 835-843. Load Forecasting solution using Soft Computing
[4] K.-Bin Song, Y.-Sik Baek, D. Hun Hong, and G. Jang, “ Sort - and a member of IEEE and Power System
Term Load Forecasting for the holidays using fuzzy liniear Operation and Control Laboratory Sepuluh Nopember Institute of
regression method.” IEEE Trans.Power Syst, Vol.20, no.1,pp.96- Technology.
101, Feb.2005
[5] K.Kim, H.Youn, and Y-C.Kang ” Short-term load forecasting for Imam Robandi received Doctor in Design and
special days in anomalous load condition using neural networks Information Engineering, Tottori Univ., Japan,
and fuzzy inference method”, IEEE Trans.Power System, vol.15, March 2002. Since 2008, he has been Professor
no.2, pp. 559-565, May, 2000 at Electrical Engineering Sepuluh Nopember
[6] D. Wu and M. Nie, “Comparison and practical implementation of Institute of Technology. His present reseach
type-reduction algorithms for type-2 fuzzy sets and systems,” in about Dynamic stability problem of large scale
Proc. IEEE Int’l Conf. on Fuzzy Systems, Taipei, Taiwan, June electrical power system with Intelligent Control
2011. Strategies, Optimization Control, Adaptive
[7] Engin Yesil, “Interval Type-2 Fuzzy PID Load Frequency Control and Intelligent Control Technologies
Controller Using Big Bang-Big Crunch Optimization”, Applied (Fuzzy Logic Control, Genetic Algorithm, Artificial Immune System,
Soft Computing, Elsevier, pp.100-112, 15-Feb.2014 Ant Colony, Honey Bee Colony, Birds Swarm Optimization, Bacteria
[8] Osman K.Erol, Ibrahim Eksin, ”A New Optimization Method : Algorithm, And Shark Schooling). He has Award for Encourment of
Big Bang – Big Crunch”, Advance in Engineering Software, Research and Development from The Institute of Electrical Engineer of
Elsevier, pp.106-111, Feb.2006 Japan (IEEJ) in year 2000.
[9] Imam Robandi. “Desain Sistem Tenaga Modern” Andi Publisher ,
Yogyakarta, June 2006
[10] J.R.Castro and O.Castilo, “Interval Type-2 Fuzzy Logic For
Intelligent Control Application”, Fuzzy Information Processing
Sociaty,2007. NAFIPS’07. Annual Meeting of the North
American, 2007, pp.592-297
[11] The Tijuana Institute of Technology and Baja California
Autonomous University, Interval Type-2 Fuzzy Logic Toolbox for
Use with MATLAB user guide Version 1.0.
[12] Imam Robandi, “Modern Power System Control” Andi Publisher,
Yogyakarta, December 2008.
130
Photovoltaic System Equipped with a Reliable

and Efficient Regulator/MPPT and Energy Flow Controller
K. Hirech, M. Melhaoui, E. Baghaz, R. Malek, K. Kassmi
Abstract – The work presented in this paper concerns the conception, implementation and
testing of a photovoltaic system equipped with a new concept of controller 'regulator/ MPPT'. The
role of the controller is, firstly, to exploit the maximum power by Hill climbing improved algorithm
that considers optimal electrical characteristics of PV panels (variation of the optimal voltage of
the panels during operation around the maximum power point) regardless of the system
perturbation such as sudden variation of illumination. Secondly, to manage the energy between
panels and batteries and also, between panels and grid, in order to control the charge / discharge
process with taking into consideration the different situations that can occur in an installation
(charge cycle, discharge cycle, charge/discharge cycle). Thirdly, to inject the surplus of energy
provided by panels into the grid when the batteries are fully charged and even when they are in
charge phases (second and third phases of charge while regulation is enabled; a part of the power
delivered by the panels is not absorbed by the batteries).
And also, its role is to estimate the state of charge with precision by the combination of two
methods: the OCV method and the coulomb counting method. In addition, the system ensures all of
the following functions: display of parameters on LCD, acquisition and presentation of results on
computer, supervision and so on.
The results obtained during days of experimentation, under weather disturbances (variation of the
irradiance and temperature) show the robustness of the PV system, the good control of batteries
charge/discharge process under the maximum of energy provided by the PV panels and the
protection against overcharging and deep discharging. The surplus of energy provided by panels
and unabsorbed by the batteries is transferred into the grid. The estimation of the state of charge
SoC is evaluated from the measured parameters in real time; it shows an improvement of around
5% compared to the conventional technique. It is conceived so that it can be adapted to any
battery system of type lead. Copyright © 2015 Praise Worthy Prize S.r.l. - All rights reserved.
Keywords: Photovoltaic Systems, Battery, Regulator, MPPT, Boost/Buck Converter,

Charge/Discharge, SoC
I. Introduction A good SoC estimation offers many advantages such

as: following the charge / discharge process, life of
Nowadays, there are many works about photovoltaic batteries and state of health [4]-[5]. It can also be used
systems (PV) that operate the energy in order to inject it for the modelling, management and reliability of the
into the grid "Grid- Connected Systems" [1] or in the batteries use [4]. The operation of solar controllers
stand-alone installations "Stand- Alone Systems" [2]. available on the market are principally based on the
Concerning the stand-alone PV applications, the major regulation from the threshold voltages and currents.
problem is the storage of the energy supplied by the PV These parameters are set by manufacturers that make
panels in solar batteries that suffer by several types of it impossible to optimize the installation according to
degradation due to their malfunctioning, bad regulation user needs or characteristics of batteries [6]-[7].
of charge/discharge, impact of the temperature, necessity Thus for controlling deep is charge they are based on
of management and ageing [3]. voltage values of the batteries, which are not hollowing
Therefore, optimizing battery life requires control of for a good optimization of operation. These voltage
its functioning by regulators, taking into consideration values do not reflect the true capacity of the battery
the different situations that the installations may undergo, during operation, because it depends, all the time, on the
good detection of thresholds in order to better control the value of the discharge current, [6]-[7].
charge/discharge process, storage of optimum energy Conventional methods of the SoC estimation have
supplied by the panels and the determination and precise limitations that reduce and prevent their uses as an ideal
knowledge of the state of charge (SoC) which reflects the and effective solution to both, protection of batteries and
quantity of electricity stored in batteries [4]. optimization of their use [8]-[9].
131
In general, they are not precise and have their climbing algorithm, taking into consideration the optimal
limitations [8]-[10]. electrical characteristics (voltage, current) of cells and
These methods include: PV panels, this algorithm presents are markable rapidity
 The specific gravity method [10] and the method at the level of convergence towards the MPP point
based on the measurement of the open circuit voltage compared to the classical algorithm [15]-[16].
(OCV) [8]. Both methods are not practical for an The determination of the SoC is achieved by the
online estimation since measurements are made after combination of two methods to correct estimation errors
a relatively long period of stabilization and and to increase precision, taking into consideration the
relaxation. faradic efficiency of charging/discharging [8]. The
 Coulomb counting method based on the integral over initialization of SoC is fixed by the method of open
time of the current entering (charging) or leaving circuit voltage (OCV) [8] when the batteries are at rest: a
(discharging) the cells [9]. The precision of this first recalibration of the state of charge is applied.
method is closely related to the techniques of The quantity of charge absorbed or delivered by the
measurement and integration of the current, which is batteries is calculated by the Coulomb counting method
variable and depends on weather conditions [8]. in real time [9] and a second recalibration of the state of
 The method of measuring the internal impedance of charge is applied when the batteries reaches full charge
the battery is not reliable since the impedance is conditions. The proposed method is characterized by its
temperature, acid density and ageing dependent, simplicity, it is easy to implement and does not require
which increases the complexity of using it [11]. the use of additional hardware. So, the SoC is estimated
Photovoltaic systems connected to the grid [12] have with precision in real time. The proposed power system
for the main role the injection of energy supplied to the is designed in order to manage the energy flow and
grid. Among their structures, one that integrates elements optimize the system functioning between panels and
of storage (batteries) [12]-[13] either to satisfy needs of battery as well as between panels and grid. This power
local energy demand [14] or to facilitate coupling block is controlled by PWM signals generated by a
(presence of intermediate) and to resolve problems of microcontroller. The various measurements and
adaptation between production and demand [12]-[14]. parameters will be displayed on an LCD and transferred
The main challenges for these structures: through a serial connection to a computer [6].
 Protection of batteries used in the process charge /
discharge
 The stability and energy storage when the batteries II. Structure and Functioning
are fully charged of the System
 The operating performance of the energy supplied
II.1. Synoptic Diagram
during the charging phases (part of delivered energy
is not absorbed during the phases of battery charging. The synoptic diagram of the PV system designed and
This paper proposes and studies a structure of a tested in this work (Fig. 1) consists of various blocks and
photovoltaic system equipped with a digital MPPT functional subsystems:
regulator which guarantees the storage of the energy  Two PV panels mounted in parallel [6].
produced by the PV panels in lead acid batteries, under  Two lead acid solar batteries in series.
optimal conditions, with a good precision of the state of  DC or AC load through a DC/AC converter (lamp,
charge estimation. This structure maximizes the power pump, ...)
transferred to the converter output, increases the rapidity  A DC/DC Boost (or Buck) converter to adapt the PV
of charging and improves efficiencies compared to some panels to the batteries (functioning around the
commercial regulators. A new concept is proposed to maximum power point) [17].
control and manage the energy supplied by panels in the  The management and supervision system (MSS),
charging/discharging process by a regulation system. which is based on the use of a microcontroller to
This latter is based firstly on the hash of current perform all the tasks of: optimization of the system
supplied by panels via a PWM signal controlling the functioning by the MPPT control, regulation of
switch Buck or Boost DC/DC converter to maximize the charge/discharge process, control of the energy
power to the output by the execution of an algorithm that transfer between various blocs, acquisition and
performs the function MPPT. And secondly, it is based display of different quantities (voltages, currents,
on the hash of current supplied to the output of the powers, SoC, performance, illumination, temperature)
converter to charge the batteries with a second PWM on the screen of a PC or LCD and communication to
signal controlling the charge circuit. A part of energy the computer by a serial link [18].
produced and unabsorbed by the batteries during Block of energy flow controller, formed by three
regulation (batteries charge) is injected into the grid with power circuits (circuits 1 to 3) presented in Fig. 1, that
the minimum of energy loss by using a power circuit control the charge/discharge process of the batteries and
controlled via a third PWM signal. When the batteries are inject the surplus of energy into the grid with the
charged, all of energy produced is injected into the grid. minimum of energy loss either during the charging
The MPPT control is ensured by an improved Hill phases, or when the batteries are fully charged.
132
II.2. The MPPT Control

To ensure the MPPT control, the MSS system
generates a PWM signal with variable duty cycle α to
converge instantly the operating point of the PV panels
around the MPP, taking into consideration the
performance of the PV panels and the duty cycle,
regardless of weather or load variations [18]-[19].
The principle of the MPPT control (Fig. 2), is to
increase or decrease the duty cycle α according to how
the PV system evolves. The algorithm used during this
work is based on the calculation of the power, which its
derivative is going to allow us to determine the evolution
of the system. When the value of dP/dt is positive; the
Fig. 1. Synoptic diagram of the PV system designed
system thus evolves towards the maximal point of power
and when dP/dt is negative; the PV system loses the
power while the system did not still find the MPP, or it
oscillates around it. According to the study made on the
photovoltaic systems, the optimal voltage of PV panels
varies between two values Vmin and Vmax, regarding
the intensity of the illumination and temperature. Thus,
the algorithm forces the search for the MPP using the
voltage range [Vmin, Vmax]:
 If Vmin<V<Vmax: (V is the voltage of PV panels)
the system oscillates all around or close to the MPP.
In this case, a temporization is fixed during every
change of the evolution of the PV system. When
TMP=1 (temporization or delay elapsed): the Vstate
signal is inverted (the evolution state includes two
states: 1 or 0). This signal allows us at the end of the
algorithm to increment the duty cycle if Vstate=1, or
decrease the duty cycle if Vstate=0. The timer will be
reset in this case using TMP variable. Otherwise, no
operation is performed.
 If (V<Vmin): the PV voltage derivative is calculated
Fig. 2. The MPPT algorithm used in this work [19]
to determine effectively the evolution of the PV
system (The PV system evolves toward or away from
the MPP). When ∆V<0 and TMP = 1; the Vstate In our case, the charge process is done by using the
signal is inverted and the timer will be reset using technical characteristics of the batteries shown in Table I,
TMP variable. Otherwise, no operation is performed. following the phases mentioned below:
 If (V>Vmax): the PV voltage derivative is calculated  Boost phase: The batteries are charged by the totality
to determine effectively the evolution of the PV of energy supplied by the PV panels until VBAT
system (The PV system evolves toward or away from (batteries voltage) reaches the regulation voltage VR=
the MPP). When ∆V>0 and TMP = 1; the Vstate 28.4V and no energy is transferred into the grid.
signal is inverted and the timer will be reset using  Absorption phase: The charge continues to fight
TMP variable. Otherwise, no operation is performed. against the phenomenon of overcharging. The
algorithm sets VBAT to 28.4V and the charging current
At the end of the algorithm and using the Vstate
signal, the duty cycle is increased if (Vstate equals to1 IBAT decreases gradually until IOCT = 0.8A (the
“High state”) or decreased if (Vstate equals to 0 “Low batteries do not absorb all of the energy supplied and
state”) with the fixed step determined by the algorithm. that absorbed decreases gradually). At this point, the
batteries are fully charged, the SoC is over than 95 %
and the energy unabsorbed by the batteries is
II.3. The Energy Transfer Controller transferred into the grid.
 Floating phase: To compensate the self-discharge, the
The method proposed in this work for the control of
batteries absorb a minimum of energy only to fix the
energy provided by the PV system is carried out by
voltage VBAT around VFLT = 27.2V. This phase is
hashing current supplied to the output of the converter by
required to manage energy consumption and protect
PWM signals.
the batteries against the causes of premature ageing.
This energy is used to charge the batteries following
Almost the totality of the energy is transferred into
three phases described by the algorithm of Fig. 3 and the
the grid.
surplus is injected into the grid.
133
Fig. 4. Control algorithm of the batteries discharge
The initial state of charge So Cii sestimated from the

open circuit voltage (OCV) of batteries. This voltage
Fig. 3. Control algorithm of the batteries charge
ischaracterized by itsstabilitywhilethe batteries are in
long rest (more than 24 hours) that reflects an image of
The discharging control of batteries is based on the the actual batteries state ofcharge.
situation of the installation using the algorithm of Fig. 4. The quantity of charge absorbed or delivered is
This process is activated/deactivated on the basis of measured by an algorithm that integrates the current
batteries state of charge SoC estimation: during the charge (IBAT) and discharge (Idech) as a
 If the SOC reaches the limit value SoCmin.D, then, function of time and taking into consideration the
the process of discharge will stop,
faradaic efficiency (). Arecalibration of thestate of
 If the SOC is higher than SoCmin.R, then, the
charge SOCis activated whenthe batteries reachthe
discharge process is activated,
conditions of thefinal chargethat theyare fully chargedto
 If SoCmin.D<SoC< SoCmin.R, then, the situation of
minimizeerrorsof integration and theeffects of
the installation determines the charge / discharge
temperature. During the charge/discharge, the expression
process. If the discharge takes place following the
of the SoC is [21], [22]:
normal cycle (dSoC/dt≤0) then the batteries discharge
until SoC=SoCmin.D to protect against deep SoC  t   SoCi  100% / Qo 
discharging (Table I). If a new charge cycle takes (1)
place (dSoC/dt>0), then the discharge is blocked until     Ibat  t   Idech  t   dt 

the SoC reaches SoCmin.R in order to ensure a  
minimum amount of charging.
where:
TABLE I Qo The nominal capacity of the battery
ELECTRICAL CHARACTERISTICS AND VALUES CHARGE/DISCHARGE  Faradaic efficiency
PARAMETERS OF BATTERIES AT 25 °C Ibat Charge current
Characteristics and conditions Batteries of Two batteries on Idech Discharge current
of charge discharge 6 cells series (2*6 cells)
Nominal voltage 12V 24V SoCi The initial state of charge
Nominal capacity 110Ah 110Ah
Maximum voltage 14.8V 29.6V The initial state of charge thatvaries very little with
Maximum current 22A 22A temperature T (Eq. (2)) [23] , is determined from the
Voltage of regulation VR 14.2V 28.4V
Voltage of floating VFLT 13.6V 27.2V open circuit voltage (OCV) following the empirical
Current of over charge terminate 0.8A 0.8A equation (Eq. (3)) [16]:
IOCT
Limit Voltage of discharge LVD 11.4V 22.8V SoCi  T  =SoC  25C   1  0.003  T  25  (2)
State of charge. minimum to 40% 40%
disconnect SoCmin.D
State of charge. minimum to 45% 45% SoCi  25°C  =50  OCV  1170 (3)
reconnect SoCmin.R
On Fig. 6 the appearance of the initial state of charge

II.4. State of Charge Estimation (SoCi) variation is presented. This variation depends on
the voltage OCV and temperature (0 ° C, 25 ° C and 50 °
In order to get a high precision, the state of charge
C). It appears:
SoC, which the prediction model is presented in Fig. 5,
 When the temperature is T = 25 ° C and the voltage
was estimated by combining both methods: Coulomb
OCV ranges from 24 V to 25 V (4% variation) then
counting [20] and the open circuit voltage (OCV) [4],
SoCi varies from 30% to 80% (variation by a factor
[5], [20].
of 2.6).
134
 When the voltage OCV is 25 V and the temperature controlling the DC/DC converter. The second one
increases (decreases) 15 ° C around T = 25 ° C (40% (PWM1) controls the switch 1 (circuit 1) in order to
variation) then SoCi increases (decreases) to 83.6% control and manage the batteries charge. And the third
(76.4%) (a variation of 3.6%). one (PWM2) controls switch 2 (circuit 2) to supervise
and manage the transfer of energy surplus produced into
the grid. The discharge of the batteries is provided by the
circuit 3controlled by a square signal (S.C.D). The
overalls results show:
 During the functioning of the complete PV system,
The MSS system generates the PWM signal (Fig. 8)
of a duty cycle, for this sample, coincides with a
value of 0.7. This allows adapting the PV panels to
the load (Batteries and grid) and extracting the
maximum of power from PV panels by following the
point MPP.
Fig. 5. The proposed method to estimate the state of charge  During the charging cycle:
SOC during the charge/ discharge of the batteries
 During the Boost phase (Fig. 7(a)), the batteries
120 are charging with IBAT current. However the VBAT
voltage and the state of charge SoC increase
100
linearly with time. During this phase, the
80 management system MSS generates the PWM1
SoCi (%)
60
signal (Fig. 9) of a duty cycle of 0.98, to activate
correctly the power circuit 1 (circuit 1 in Fig. 1).
40 This latter controls the batteries charge and the
SoCi (0°C)
20 SoCi (25°C)
signal PWM2 (Fig. 9) of a 0.02 duty cycle
SoCi (50°C) (inverse of PWM1) to deactivate correctly the
0
power circuit 2 (circuit 2 in Fig. 1) which controls
23,0 23,5 24,0 24,5 25,0 25,5 the transfer of power to the grid. During this
OCV (V) phase, the power supplied by the PV panels is
Fig. 6. The initial state of charge SoCi based on the open circuit voltage
OCV for a 24V battery at temperatures of 0°C, 25°C and 50°C
completely absorbed by the batteries, until a
voltage VR that is limited to 28.4V in order to
protect batteries against overcharging. At this
III. Simulation Results voltage, a capacity of 85 to 90% is recovered and
the batteries are relatively charged that allows the
To simulate the various blocks of the system, the PV system to generates a S.C.D signal in state 1 (state
system of Fig. 1 is implemented as blocks in ON) to control circuit 3 in order to discharge the
ISISPROTEUS [24], and then simulated the system batteries if a load is connected.
operation taking into consideration the algorithms of Fig.  During the absorption phase (Fig. 7(a)), the
2 to Fig. 5. In the corresponding program where all voltage VBAT is limited around VR=28.4V. The
instructions, equations defined above and functions that SoC continues to increase gradually depending on
as sure the various tasks are programmed and injected the power absorbed that decreases, due to the
into a micro controller which plays the role of a system decrease of the charging current, to achieve full
processor. The parameters related tour system: panel charge (VBAT = 28.4V and SoC> 95%). In this
type, batteries (Table I), acquisition,... are set to optimize case, the management system MSS generates two
the operation, and then extracted the graphics that PWM signals. The first signal (PWM1) (Fig. 10),
illustrate the different operating modes: functioning of of a variable duty cycle (0.02 to 0.98) is to adjust
the MPPT control, control of energy flow, and estimating the VBAT voltage around VR, by using pulses of
the state of charge. The results of all possible modes of current generated by the hashing of the current
the batteries operation, when the energy is transferred IBAT through the power circuit1. And the second
into the batteries (charging cycle, discharging cycle and signal (PWM2) (Fig. 10), of a variable duty cycle
charging/discharging cycle), are presented in Fig. 7(a) (0.02 to 0.98) which is practically the opposite of
and Fig. 7(b). They show the typical evolution of the PWM1,is to activate or deactivate correctly the
voltage VBAT, the state of charge SoC, the charge current power circuit 2 to inject the surplus of energy into
IBAT and the discharge current Idech during different the grid.
cycles. During this process, the management and During this phase, the charging of batteries
supervision system controls the different power circuits continues through the current pulses IBAT
(Fig. 1) by PWM signalsof10kHzfrequencyandvariable generated by the PWM1 until a SoC which
duty cycle to ensure the good functioning of the system. exceeds 95% and a current IBAT less than IOCT. The
The Fig. 8 to Fig. 12 extract three PWM signals. A first surplus (energy supplied) is transferred into the
one (PWM) ensures the MPPT function of the system by
135
grid via the circuit of power 2 controlled by the conditions, a power of 65W at a voltage of 14.2V and
PWM2 signal. At this stage, the batteries are fully a current of 4.4 A. [6]-[25].
charged and the system continues to generate the  Two lead acid solar batteries (flat plat) mounted in
S.C.D signal in state 1 (state ON) to control series. Each battery has a nominal voltage and
circuit 3 in order to discharge the batteries if a capacity of 12V and 110Ah.
load is connected.  A DC / DC boost converter dimensioned to operate at
 During the floating phase (Fig. 7(a)), the charging a frequency of 10 kHz, a power of about 200 W and a
current decreases above IOCT and the SoC is over current ranging between 1A and 10A [6].
95 %. The batteries keep on charging through  A management and supervision system (MSS) (Figs.
current pulses to stabilize the voltage VBAT around 14), which provides communication between the
VFLT=27.3V. During this phase, the management different blocks of the PV system. It processes the
system MSS generates the PWM1 signal (Fig. data, acquires and displays all the electrical
11), of a variable duty cycle around 0.02 to parameters and controls the switches of the electronic
activate correctly power circuit 1, which controls circuits. This ensures the good functioning of the
the charge. A low energy supplied by the PV MPPT control, the regulation and control the energy
panels is transferred to the batteries in order to flow, the estimation of the state of charge and the
compensate the self-discharge, and almost all of management and drives a LCD display. A serial link
the energy is injected into the grid. Since the is used to transfer, present and storage data into a
batteries are fully charged, the discharge may be computer. The application developed in this work
performed if a load is connected via power circuit applies to all stand-alone PV systems equipped by
3, which is controlled by the S.C.D signal lead acid batteries. The algorithms 2 to 4 may be
generated by the MSS system, to the highest state adapted for the system (choice of components, panels
(state ON). and batteries’ characteristics, conditions
 During the discharge cycle (Fig. 7(a)), the square charge/discharge.) so that the whole installation is
S.C.D signal controls power circuit 3 to supply the reliable and optimized.
load from the batteries (Fig. 12). The voltage VBAT
and the SoC decrease gradually to the threshold
SoCmin.D set by the algorithm of discharge (Fig. 4).
In addition, the S.C.D signal changes to 0 (state OFF)
because the SoC reached SoCmin.D in order to stop
the discharge. To initiate battery charging, the system
generates a PWM1 signal with a duty cycle of the
order of 0.98 that controls power circuit 1 in order to
transfer all the energy into the batteries.
 During the tests of the batteries operating modes:
charge cycle, discharge cycle and charge/discharge
cycle, when the charge is carried by IBAT = 2.3A and
the discharge by Idech = 5A, the results are shown in (a) the charging phases and simple discharge
Fig. 7(b):
 During the charge cycle, VBAT and SoC increase
progressively until SoC = 70%.
 During the discharge cycle, VBAT decreases and so
does the SoC until reaching the minimum state of
charge (SoCmin.D); then the discharge is
disabled.
 During charging/discharging cycle, the MSS
system activates the discharge when a 5% of the
capacity is recovered (SoC = SoCmin.R = 45%).
IV. Experimental Results (b) charge, discharge and charge/discharge cycles

IV.1. Experimental Procedure Figs. 7. The evolution of VBAT, IBAT, SoC and Idech
The Figs. 13 show both: the fully automated bench of
electrical measurement and the complete PV system
designed and tested in this work. The whole equipment
consists of:
 Two monocrystalline photovoltaic panels mounted in
parallel. Each one provides, under optimum
Fig. 8. A sample of The PWM signal that controls the boost converter
136
 A weather station equipped with a pyranometer and

temperature sensor (thermistor) to know the intensity
of illumination and temperature while the PV systems
are functioning.
 The measuring instruments composed of ammeters,
voltmeters, oscilloscopes and Keithley millimetre
connected to a computer to determine the different
electrical parameters of the PV system such as
current, voltage, power...in real-time.
Fig. 9. Samples of the: S.C.D, PWM1and PWM2 signals IV.2. Results and Discussion
during the Boost charging phase
To validate all the simulation results, the PV system
designed and carried out in this work (Fig. 13) is tested
in the normal functioning of installation (production of
energy, batteries charge, injection into the grid) during 8
days (D1 to D8) successive and globally illuminated in
order to fully charge the batteries.
Fig. 10. Samples of the: S.C.D, PWM1 and PWM2 signals during the
Absorption charging phase
(a)
(b)
Fig. 11. Samples of the: S.C.D, PWM1and PWM2 signals

during the Floating charging phase
(c)
Fig. 12. Samples of the: S.C.D, PWM1and PWM2 signals

during the discharging phase
 The power circuits (circuits 1 to 3) have a basic

structure based on power switches controlled by (d) (e)
PWM signals whose form ensures the good
functioning of energy transfer (charge/discharge of Figs. 13. The photovoltaic system designed and digital bench of
batteries, injection of surplus energy provided to the measurement installed at the laboratory: (a): photovoltaic panels; (b):
pyranometer; (c): designed system (batteries, DC /DC converter,
grid through a DC/AC converter, supplying a DC or measurement instrument, LCD…); (d): keithleymultimeter;
AC load through a DC/AC converter). (e): digital oscilloscope
137
When the system starts functioning at first first,, the

acquisition interface performs measurements of batteries
voltage under the conditions of open circuit ((OCV OCV
=24.28V
=24.28V), ), and then calculates the SoC by the OCV
method
method, using Eq. (2) and Eq. (3). The interface displays
an initial state of charge SoCi =42 =42% %.. The management
and supervision System (MSS) generates PWM signals
of a 10kHz frequency and 5V amplitude to control the
different power switches designed
designed. Thishis is not possible at
(a) this amplitude,
amplitude, as these signals aren’t adapted to the
system. Then
system. hen the appropriate circuits are introduced and
used to amplify,
amplify, filter and adapt these signals to the
requirements of the switches to operate the system in the
normal state
state. The
The different results obtained are presented
in Fig. 16 and Fig. 17,
17 and they show the following:
following:
 During Boost phase (0h to 38h) (Fig. (Fig. 16):
16): the
batteries are charged with a current IBAT that reaches
2 A towards middaymidday.. The voltage VBAT and SoC
(b) increase until the threshold limit VBAT = VR . The SoC
shows that more than 90% of the battery capacity is
recovered during this phase ((Fig. Fig. 17).
17). In the same
figure of the SoC, the values obtained by the two
methods: that proposed in this work (Eq.(1)) Eq.(1)) and
OCV method [8] [8] are traced. This latter (the (the OCV
method
method)) Requires a very long period of rest and even
with 14 hours of rest (the (the time between the last
measures in the end of a day when the illumination is
very low and the morning of the next day)the
(c) batteries voltage is not fully stable and it is slightly
higher than that real.
real. This reflected the estimation
estimation
Fig
Figs.. 14. Management and supervision system: (A) global
interface.,, (B) system parameters.
interface. parameters.,, (C) electrical
electrical quantities
errors observed. However, the proposed method does
visualization not require a rest period except in the calculation of
the initial state SoCi (calculated in a first test from the
The discharge tests were performed with a resistive Vbat voltage that must be completely stable). So it
load. To present the different results, we plotted curves appears a difference of the order of 5% between the
of data corresponding to tests in which the system is two methods because our method takes into
turned on for 7 to 8 hours a day (morning to evening). consideration the nature of the batteries, temperature
The nights are not reflected
reflected in the presentation of and the charge / discharge current, then it can
results since the illumination is practically zero. The estimate the improvement of obtaining SoC by our
temperature, illumination, different electrical quantities method, compared to the conventional m method
ethod of the
(voltage, current, power) of PV panels and batteries, order of 5%. The SoC values estimated by the
PWM signals (MPPT control, power circuit 1 to 3) and proposed method coincide perfectly with those
efficiencies (PV panels, DC / DC converter and
efficiencies concluded from the batteries when these last are
Complete PV system) are inferred from measuring entering in a long period of rest.
instruments such as multimeters, Keithley, pyranometer During this phase, the management and supervision
and meteorological station. system MSS permipermitted
tted the storage in batteries of the
They are compared to those presented by the entire energy supplied by the PV panels. To do this,
acquisition interface in order to confirm the congruence the MSS system optimizes the functioning of the
between values concluded. Concerning the functioning of system by generation: the PWM signal controlling the
the PV panels and DC/DC converter, simulations [26] Boost converter ((Fig.
Fig. 18
18(aa))) of a 0.7 duty cycle and
have been carried out to ensure the optimal operation oof the PWM1 signal (Fig. (Fig. 18
18(a)))) of a 0.98 duty cycle
the PV system. and 36V amplitude in order to activate the power
circuit 1 designed in this work. Practically no energy
is injected into the grid. The MSS system generates
IV.3. Energy Flow Management in the PV System also the PWM2 signal ((Fig. Fig. 1818((c))) of a 0.02 duty
a. Batteries charging control cycle and 18V amplitude , to deactivate the power
The batteries charging system conceived is analyzed circuit 2.
(Fig
Figs.. 13
13)) during the days where the illumination reaches  During the absorption phase (38h (38h-50h)
50h) (Fig.
(Fig. 16
16):
): .
800 W/m2 (Fig.Fig. 15
15).
). The batteries are charged by current pulses IBAT in
Copyright © 201
2015 Praise Worthy Prize S.r.l. - All rights reserved International Review of Electrical Engineering, Vol. 10,, N. 1
138
order to fix VBAT around 28.4V to complete the

charging of batteries up to SoC = 95% (without
reaching overcharge). At this phase (Fig. 17), the SoC
increases slowly due to the progressive decrease of
the charge current absorbed by the batteries. The end
of this phase is determined by a SoC of over 95% and
a current above IOCT = 0.8A. During this phase, while
the charging of batteries is being regulated, the
surplus of energy is injected into the grid. This is
achieved by generating the PWM1 signal (Fig. 19(a))
with a variable duty cycle (0.02 to 0.98) and 36V
amplitude to control the power circuit 1 in order to
adjust VBAT around VR. And a PWM2 (Fig. 19(b))
signal almost inversely to PWM1 and 18V amplitude Fig. 17. The variation of the state of charge SoC as a function of time
to control the power circuit 2.
 During the floating phase (50h to 55h) (Fig. 16): the
batteries are fully charged (Fig. 17). To compensate
self-discharging, the system provides the charge
through current pulses of low value (0.3A) around the
float voltage VFLT=27.3V (away from the gassing
range). So, Almost the totality of the energy is
injected into the grid (In this phase the batteries
absorb a low quantity of energy).This is achieved by
generating the PWM1 signal and 36V amplitude (Fig.
20(a)) to control the power circuit 1 in order to adjust
VBAT around VFLT. And the PWM2 signal (Fig. 20(b))
with a duty cycle nearly the opposite of PWM1 and in Figs. 18. Samples of the signals: (a): PWM of the duty cycle of 0.7
the order of 0.98 and 18V amplitude to activate the controlling the Boost converter, (b): PWM1 of the duty cycle of 0.98
controlling the batteries charge, (c): PWM2 of the duty cycle of 0.02
power circuit 2 which controls the injection of surplus controlling the injection of energy into the grid
energy into the grid.
Fig. 15. The typical illumination variation during the days Figs. 19. Samples of the signals: (a): PWM1 of the duty cycle of 0.68
the experimentation took place controlling the batteries charge, (b): PWM2 of the duty cycle of 0.32
controlling the injection of energy into the grid
Figs. 20. Samples of the signals:

Fig. 16. Variation of the charging voltage and current (a): PWM1 of the duty cycle of 0.05 controlling the batteries charge,
of the batteries (VBAT, IBAT) as a function of time (b): PWM2 of the duty cycle of 0.95 controlling the injection of energy
into the grid
139
b. Operation modes of the batteries

To take into consideration the different situations that
can occur in PV installations, the overall PV system is
experienced in different mode of control to manage the
transfer of energy between panels and batteries and also
between batteries and Loads: charging Cycle,
discharging cycle and charging/discharging cycle.
The Figs. 21 show the typical results. In this case, the
charging current IBAT is approximately 4.5A and the
initial battery voltage VBAT=12.27V. While this
installation was tested, the management and supervision (a)
system (MSS) performs the following operations:
 The initial state of charge estimation SoCi= 57%.
 During the charge cycle (between 0 and 3 h), when
the charge current is IBAT=4.5A and discharge current
Idech=0A (Fig. 21(b)), an increase of VBAT and SoCis
observed (Fig. 21(a)). While the state of charge
reaches SoC= 67%.
 During the discharge cycle (between 3 and 8h), when
the current Idech=6.5A and IBAT=0A (Fig. 21(b)) a
decrease of VBAT and SoC is observed (Fig. 21(a))
and when SoCmin.D reached SoC = 40%, the system
stops the discharge to protect the batteries (Idech =
0A). (b)
 During the charge/discharge cycle (between 8 and Figs. 21. The evolution of electrical characteristics of the batteries
11h), as long as the SoC is lower than a threshold of during a charge cycle.
reconnection SoCmin.R = 45% (Fig. 21(a)), the (a): Variation of VBAT and SoC.
discharge is blocked. Above this threshold and (b): Variation of IBAT and Idech.
depending on the demand, a discharge can take place.
When the discharge current is low (Idech=1A) All the results obtained (Fig. 22 to Fig. 28) show that:
between 9.5h and 10h (Fig. 21(b)), VBAT and SoC  There is a very good agreement between the optimal
increase slowly since some of the power injected into simulation results and the experimental ones.
the batteries is consumed by the load (Fig. 21(a)).  The duty cycle of the PWM signal generated via the
When the discharge current is high (Idech=6A) MSS system by using HILL CLIMBING improved
between 10h and 11h, VBAT and SoC decrease, since algorithm, to ensure the MPPT control (Fig. 22) is
the power injected in the batteries does not practically the optimum one. The rapidity of the
compensate the energy consumed by the load. system convergence to the MPP is tested by
All the results obtained in this section show the good experimentation and comparison of this algorithm to
control of energy transferred during charge cycle, the classical one presented in references [16]-[17].
discharge cycle and charge/discharge cycle of batteries The results (Fig. 23) showed that the search time ratio
depending on the use. This demonstrates the validation of of the classical (Tcla) and improved algorithms
the simulation results and therefore of the proposed (Timp) clearly depends on the irradiance and the
charge/discharge technique of the PV system conceived position point of the functioning PV panel. For strong
(Figs. 13). irradiances (940 W/m2), the search times are virtually
identical when the initial conditions are close to the
MPP, that the initial voltage of the PV Panel is in the
IV.4. Optimal Functioning of the PV System optimal range of panels (12 V-16 V). Outside this
Concerning the optimal functioning of the PV system, range, the improved algorithm is much faster and the
where HILL CLIMBING improved algorithm is adopted search time can reach 50% compared to the search
(Fig. 4), the PWM signal that ensures the MPPT control, time using the classical algorithm. For low irradiance,
the search time that converge to the MPP and the the improved algorithm is much faster compared to
electrical quantities (voltage, current and power) of PV the classic alone, even if the initial conditions are in
panels and batteries during their charging/discharging the optimum voltage range. The search time ratio can
process are extracted. From these results, the efficiency achieve a factor of 2 (Tcla/Timp).
 The electrical quantities (voltages, currents and
of the panels (panel), DC/DC converter (cnv) and
powers) of the PV panels (Figs. 24) are practically the
global PV system (sysPV) are inferred. Taking into
optimum ones, throughout all the batteries charging
consideration the same figure, the results of optimal
process. This shows that the MPPT controller has
simulation obtained in Pspice under the same conditions
performed its role. It optimizes the functioning of the
are reported [26].
panels without divergence.
140
 The efficiency of the PV panels is around 9.5 % (Fig.

25). This shows, firstly, a good agreement between
the type of PV panels used and their ageing [17] and,
secondly, low losses of the power supplied by the PV
panels while the installation is working.
 During the first phase (Boost) of charging (0- 38h),
the DC/DC converter efficiency (Fig. 26) is the
efficiency of the battery charging since all the power
supplied by the batteries is absorbed. This efficiency,
which is about 85%, is very satisfactory and close to
optimum.
 During the second and third phase of charging (38h - Fig. 22. The evolution of the duty cycle as a function of time
55h), the efficiency while injecting into the grid 3,0
increases and that of the batteries charge decreases
2,5 2
(Fig. 26).This is so due to the decrease (increase) of 400 W/m
Tcla / Timp
2,0
the power absorbed (injected) by the batteries (grid). 2
700 W/m
 The global efficiency of the PV system during the 1,5
2
940 W/m
charging and the injection into the grid (Fig. 27) is 1,0
around 8.5%. It is very satisfactory and considered as 0,5

the best percentage according tithe literature [27]- 0,0
[28]. 0 2 4 6 8 10 12 14 16 18 20
Vpv (V)
Relatively to the feasibility and the power/energy gain
of the battery charging, we have tested again, for 4 days Fig. 23. Relationship between the search times of the MPP in the
(D1toD4), the system designed face to a commercial presence of the classic and improved algorithms in function of the PV
regulator to compare their performances during the panel voltage and the illumination
charging of a battery of 12Vand110Ah, two panels are
connected in series with every system. The Fig. 28 shows
the results of the power variation supplied by the two
systems at the same battery. During the first day of
charging (D1), the designed system provided a power
(Pcnv) 4 times greater than the commercial regulator
(Preg). The power Pcnv reached maximum valuesin the
middle of the day when the irradiance is maximal. The
commercial regulator stops supplying power to the
battery when the voltage reaches13.7Vtostart the 2nd (a)
charging phase(regulation around 13.7V), while the
system designed continuous to operate under optimum
conditionsuntilthe4thday (D4) where the battery voltage
reaches VR =14.2V and the 2nd phase begins (absorption
phase). During this operation, the system conceived has
transferred 1,45 kWh of energy to the battery, however,
that transferred by the commercial regulator is only 0.13
kWh. Consequently the energy transferred by the system
designed is 11 times greater compared to that of
commerce.
This allowed us to optimize the functioning and (b)
accelerate the charging. All these results show that the
batteries are charged according to the three phases fixed
by the algorithm of Fig. 4, taking into consideration the
parameters set (thresholds charges, MPPT control, power
circuits…). The efficiencies presented in Fig. 25 to Fig.
27, demonstrate the good functioning and performance of
the various PV system blocks: DC / DC converters,
power circuit and so on.
(c)
V. Conclusion
Figs. 24. The evolution of: (a): experimental and simulated (Optimum)
In this paper, the design, implementation and results of Vpv voltage; (b): experimental and simulated (Optimum)
results of Ipv current; (c): experimental and simulated (Optimum)
functioning of a photovoltaic system are studied. results of Ppv power
141
This system is equipped with a management system

and digital controller, which take into consideration a
specific installation (panels, batteries, geographical
location and temperature). The designed system ensures
the pursuit of the maximum power point of the PV panels
by a HILL CLIMBING improved algorithm, which
requires knowing the variation range of the optimal
voltage of panels during their normal operation.
The regulation of the charge/discharge process is
provided in three phases, the estimation of the state of
charge of the solar lead acid batteries type strongly
Fig. 25. Efficiency of the PV Panels depends on the initial state of charge precision. The
control of the injection of surplus energy produced into
the grid manage the demand of electricity by AC and DC
loads. The typical results show:
 That the PV system designed works under optimal
conditions while the installation is working by
following the maximum power point(storage phases,
injection into the grid and managing the consumption
by the AC and DC loads).
 The search time of the MPP is significantly improved
by 50% compared to the search time by use the
classical algorithm. The PV system becomes fast and
accurate.
 The batteries are fully charged in three phases, with
Fig. 26. Efficiency of DC/DC converter (experimental and simulation maximum power which allowed to optimize and
(optimum) results) during the charging of batteries and the injection
into the grid accelerate the charge.
 A good estimate of the state of charge (around 5% of
improvement compared to the traditional method).
 A good performance of the DC/DC converter and the
global PV system (85% and 8.5%).
 That the control circuit of energetic flow allows to
efficiently manage the energy supplied by the PV
panels and to inject the surplus into the grid when the
batteries are charged and even during the charging
phases while the regulation is in order to minimize
losses and for better use of energy.
The global results obtained in this work demonstrate
the reliability and optimization of the PV system.
Firstly, the digital card of the management and
supervision system is configured according to the
technical characteristics of this equipment (PV panels,
Fig. 27. The global efficiency of PV system (experimental and
simulation (optimum) results) during the charging of batteries and the
batteries ...) and, secondly, he uses algorithms which are
injection into the grid adaptable to all situations such as changes in
illumination, loads and temperature.
Acknowledgements
This work is supported by:
 Belgian Development Agency CTB (Project
MIP/012/010), Morocco.
 United Nations Development Programme UNDP Art
Gold Morocco, ENV 2008 2 OO.
 Cooperation Moroccan-Belgian "Institutional
University Commission", IUC, Oujda, 2008-2012
(Water and Environment Activity / Sub-Activity
Renewable Energy).
 Moroccan-Tunisian Cooperation, SCIENTIFIC
Fig. 28. The evolution of the powers provided by the system conceived RESEARCH AND TECHNOLOGY Project
(Pcnv) and a commercial regulator (Preg) during the battery charging
142
(11/MT/38). [17] M. Melhaoui, E. Baghaz, K. Hirech, K. Kassmi, Protection of the

photovoltaic (PV) systems by a dysfunction detection circuit of
the PV system (DDCS), International Journal of Emerging
Technologies in Computational and Applied Sciences (IJETCAS)
References JETCAS 13- 509, 2013, IJETCAS.
[1] S. K. Khadem, M. Basu and M. F. Conlon, Power Quality in Grid [18] M. Melhaoui, K. Hirech, R. Gaamouche, M.F.Yaden, E. Baghaz,
Connected Renewable Energy Systems: Role of Custom Power K. Kassmi, Conception et réalisation d’un système photovoltaïque
Devices, International Conference on Renewable Energy and menu des commandes de contrôle et d’acquisition numériques,
Power Quality (ICREPQ’10), Granada, Spain, 23rd to 25th March International Renewable and SustainableEnergyConference,
2010. Congress Palace, Ouarzazate, Morocco, March 7-9, 2013
[2] V. Salas E. Olías, A. Barrado, A. Lázaro, Review of the [19] Melhaoui, M., Baghaz, E., Hirech, K., Yaden, F., Kassmi, K.,
maximum power point tracking algorithms for stand-alone Contribution to the improvement of the MPPT control functioning
photovoltaic systems, Solar Energy Materials and Solar Cells of photovoltaic systems, (2014) International Review of Electrical
Volume 90, Issue 11, 6 July 2006, Pages 1555–1578. Engineering (IREE), 9 (2), pp. 393-400.
[3] Kais Brik, Faouzi ben Ammar, Causal tree analysis of depth [20] Kong-Soon Ng, Yao Feng Huang, Chin-Sien Moo, and Yao-
degradation of the lead acid battery. Journal of Power Sources, Ching Hsieh, An Enhanced Coulomb Counting Method for
Volume 228, 15 April 2013, Pages 39-46. Estimating State-of-Charge and State-ofHealth of Lead-Acid
[4] Xiaopeng Chen, Weixiang Shen, Zhenwei Cao, Ajay Kapoor, Batteries, IEEE transaction. Published in Telecommunications
Adaptive gain sliding mode observer for state of charge Energy Conference, 2009. INTELEC 2009. 31st International.
estimation based on combined battery equivalent circuit model, [21] Hongxing Yang , Zhaohong Fang, Wei Zhou, Battery behavior
Original Research Article Computers & Chemical Engineering, prediction and battery working states analysis of a hybrid solar–
Volume 64, 7 May 2014, Pages 114-123. wind power generation system, Renewable Energy 33, 2008,
[5] W. Waag, D.U. Sauer, Secondary Batteries – Lead– Acid Systems 1413–1423.
| State-of-Charge/Health, Reference Module in Chemistry, [22] Martin Coleman, William Gerard Hurley, and Chin Kwan Lee,
Molecular Sciences and Chemical Engineering, from An Improved Battery Characterization Method Using a Two-
Encyclopaedia of Electrochemical Power Sources, 2009, Pages Pulse Load Test, 708 IEEE Transactions on Energy Conversion,
793-804 Vol. 23, No. 2, June 2008
[6] K. Hirech, M. Melhaoui, F. Yaden, E. Baghaz, K. Kassmi, Design [23] N. Achaiboua, M. Haddadi, A.Malek, Lead acid batteries
and realization of an autonomous system equipped with a simulation including experimental validation, Journal of Power
regulator of charge/discharge and digital MPPT command, The Sources 185 (2008) 1484–1491.
Mediterranean Green Energy Forum 2013, MGEF-13 [24] Web-page: http://www.labcenter.com/index.cfm, 2014
[7] Mrs Jaya et al. PIC Based Solar Charging Controller For Battery. [25] Web-page:
International Journal of Engineering Science and Technology www.solarcellsales.com/techinfo/docs/Shell_SP75.pdf, 2014
(IJEST).Vol. 4 No.02 February 2012. [26] Web-page : http://www.cadence.com, 2014
[8] K. W. E. Cheng, B. P. Diwakar, Hongjie Wu, Kai Ding, and Ho [27] Jung Hun So, Young SeokJung, Gwon Jong Yu, JuYeop Choi, Jae
Fai Ho, Battery-Management System (BMS) and SOC Ho Choi, Performance results and analysis of 3 kW grid-
Development for Electrical Vehicles, IEEE Transactions on connected PV systems, Renewable Energy 32, 2007, 1858–1872.
Vehicular Technology, Vol. 60, No. 1, January 2011. [28] Danny H.W. Li, K.L. Cheung, Tony N.T. Lam, Wilco W.H.
[9] Martin Coleman, Chi Kwan Lee, Chunbo Zhu, and William Chan, A study of grid-connected photovoltaic (PV) system in
Gerard Hurley, Fellow, IEEE, State-of-Charge Determination Hong Kong, Applied Energy 90, 2012, 122–127. Contents lists
From EMF Voltage Estimation: Using Impedance, Terminal available at Science Direct Applied Energy.
Voltage, and Current for Lead-Acid and Lithium-Ion Batteries,
IEEE Transactions on Industrial Electronics, Vol. 54, No. 5,
October 2007. Authors’ information
[10] KorayKutluay, YigitÇadirci, Yakup S. Özkazanç, Member, IEEE,
and IsikÇadirci, A New Online State-of-Charge Estimation and Kamal Hirech graduated with Research
Monitoring System for Sealed Lead Acid Batteries in Master’s degree in electronic and
Telecommunication Power Supplies, IEEE Transactions on Communication System from Mohammed 1st
Industrial Electronics, Vol. 52, No. 5, October 2005. University, Oujda, Morocco in 2010. Currently,
[11] NovieAyub, Windarkoand, Jaeho Choi, SOC Estimation Based on he is pursuing his Ph.D. degree in Physics and
OCV for NiMH Batteries Using an Improved Takacs Model JPE Engineering (Electronic and photovoltaic), in
10-2-11, Journal of Power Electronics, Vol. 10, No. 2, March Electronic components and Photovoltaic
2010 Energy’s Team (CEEP) in LETAS Laboratory
[12] Yann Riffonneau, SeddikBacha, Member, IEEE, Franck Barruel, at Mohammed 1st University, Oujda, Morocco. His areas of research
and Stephane Ploix, Optimal Power Flow Management for Grid include the analysis is of the functioning and improving the
Connected PV Systems With Batteries, IEEE Transactions on performance of a solar regulator including the MPPT control in stand-
Sustainable Energy, Vol. 2, No. 3, July 2011 309 alone systems.
[13] Sang young Park1, Yanzhi Wang2, Younghyun Kim1, Naehyuck
Chang1 and Massoud, Battery Management for Grid-Connected Mustapha Melhaoui received his B.Eng degree
PV Systems with a Battery, Pedram2ISLPED’12, July 30–August in Computer Science and Electronics from
1, 2012, Redondo Beach, CA, USA. Copyright 2012 ACM 978-1- Engineer school “ESIREM”, Dijon –France, in
4503-1249-3/12/07 2009. He also obtained his Research Master’s
[14] Mehdi Dali a, Jamel Belhadj, Xavier Roboam, Hybrid solarewind degree in Computer Science from Bourgogne
system with battery storage operating in grid–connected and University, France, in 2009. Currently, he is
standalone mode: Control and energy management e pursuing his Ph.D. in Physics and Engineering
Experimental investigation, Energy 35 (2010) 2587e2595 journal (Electronic and photovoltaic), in Electronic
homepage: www.elsevier.com/locate/energy Components and Photovoltaic Energy’s team (CEEP) in LETAS
[15] Yaden M.F; Melhaoui M; Gaamouche R, Hirech K; Baghaz EL.H Laboratory at Mohammed 1st University, Oujda – Morocco. His
and Kassmi K. Photovoltaic System Equipped with Digital research interest include the optimization, analysis and control of
Command Control and Acquisition. Electronics 2013, 2, 192-211. photovoltaic system, and the visualization, processing and the control
[16] E. Baghaz, M. Melhaoui, F. Yaden, K. Hirech, K. Kassmi.Design, of system behavior.
realization and optimization of the photovoltaic systems equipped
with analog and digital MPPT commands .Energy Procedia 42
(2013) 270 – 279.
143
Elhadi Baghaz graduated with Research

Master’s degree in electronic and
Communication System from Mohammed 1st
University, Oujda, Morocco in 2010. Currently,
he is pursuing his Ph.D. degree in Physics and
Engineering (Electronic and photovoltaic), in
Electronic components and Photovoltaic
Energy’s Team (CEEP) in LETAS Laboratory
at Mohammed 1st University, Oujda, Morocco. His areas of research
are in the functioning Study of the photovoltaic station realized in
Douar Zragta of the rural Oujda Angad Prefecture.
Rachid Malek is an Associate Professor at

Mohamed Premier University, National School
of Applied Sciences – Oujda, Morocco. He got
his Ph.D. in 1997 (Paul Sabatier University,
Toulouse – France) in the field of
Nanotechnologies Simulation and Modeling. He
was a post doc at Ohio University (USA) for 02
years and at Technical University of Munich
(Germany) for 01 year. He is a member of Electronic Components &
Photovoltaic Energies of LETAS Lab, Faculty of Sciences – Oujda
(Morocco). He is participating actively in supervising Ph.D. students in
the research team in the field of Materials Science and Electronic
Components, Cells and Photovoltaic Systems.
Khalil Kassmi, born in Casablanca (Morocco)

in 1963, made all the higher education in the
University Paul Sabatier (UPS) of Toulouse
(France) from 1983 until 1991. In 1991, He
obtained his PhD degree in Electronics from
UPS in France. His different research were
realized in the Laboratory of Automatic and
Analysis of the Systems LAAS / CNRST
(Toulouse) in MOS technology team. In 1992, he was recruited at SGS
Thomson-Casablanca (Engineer), where he was responsible of the
assembly chain of the powers electronic components (TO220,…).
Then, in 1993, he integrated the Mohammed 1st University in Oujda,
Morocco, as Professor-researcher, and he obtained his PhD degree in
Electronics in 1996. Since 2000, he is Professor at the Mohammed 1st
University of Oujda. He is responsible for different training:
electronics, power electronics and renewable energies 'Photovoltaic
Systems ' in Professional License and MASTERS. He is responsible of
the research team ‘Electronic components and Photovoltaic Energy
CEEP’ of LETAS laboratory. He has been in charge of the direction of
the LETAS Laboratory during the period of 2009-2011. In the frame of
the CEEP team, He makes researches and manages PhD student thesis
on materials and electronic components, cells and photovoltaic systems.
He is an author of a chapter in the book ' The renewable energies in
Morocco, The debate was launched in 2007 with more than 100
publications and international communications in the field of the
photovoltaic renewable energies (materials, cells and photovoltaic
systems). Since 2002, he has been responsible for cooperation’s
national and international projects in the field of the Photovoltaic
renewable energies (CNRST (PROTARS III D43/06 D43/06), Franco-
Moroccan (Integrated Actions: MA/06/78, MA/09/205), Morocco-
Tunisian (11/MT/38), Morocco-Belgian (Interuniversity Cooperation
for the Developments CUD)). In the framework of the opening of the
University to itssocioeconomic environment, since 2009, he became
responsible of the project: United Nations Development Program
UNDP art Gold Morocco and the Agency of Belgian Development in
Morocco CTB (MIP/01 /010).
144
Modeling and Control of a Hybrid Microgrid by Multi-Agent System
Lahoussine Elmahni, Lahoussine Bouhouch, Rachid Alaoui, Ali Moudden
Abstract – For economic and ecological considerations and following reasons for the depletion
of certain oil resources, the coupling of renewable energy sources with the production of hydrogen
by electrolysis of water contributes to the development of intelligent networks called "Smart Grid"
(SG). In this context, this paper aims to develop an algorithm for the management and control of a
hybrid microgrid by multi-agent system (MAS). This hybrid Microgrid is composed of a
photovoltaic source (PVECS), a fuel cell (PEMFC), an electrolyzer (ELZ), and a hydrogen storage
system (H2TANK). Oversight of this microgrid is under the control of MAS and ensures optimum
performance, taking into account constraints such as intermittent renewable energy, supply /
demand balance and the price per kWh. Copyright © 2015 Praise Worthy Prize S.r.l. - All rights
reserved.
Keywords: Electrolyzer, Fuel Cell, Multi-Agent, MACSimJX, Microgrid, Photovoltaic
Nomenclature NH2 Hydrogen moles per sin storage tank (kmol/s)

NS Cells connected in series
Abbreviations p Pressure (Pa)
DER Distributed Energy Resources P Power (W)
ELZ Electrolyzer Pb Pressure of tank (Pa)
H2TANK Hydrogen Tank Pbi Initial pressure of the storage tank (Pa)
HEM Home Energy Management PEMFC Power of the fuel cell (W)
JADE Java Agent DEvelopment Framework PELZ Power of electrolyzer(W)
MAS Multi-Agent System PG Grid power (W)
MG Micro grid PL Load power (W)
MGCC The Micro grid Central Coordinator PLC,PLNC Power of critical and non-critical loads (W)
PEMFC Proton Exchange Membrane Fuel Cell PV Photovoltaic power (W)
PVECS Photovoltaic Energy Conversion System q Electron charge: 1.602176565×10-19(C)
SG Smart Grid r Ohmic resistance parameter (m2)
Rg Gas constant R: 8.3144621 (JK−1mol−1)
Symbols Rohm0 Ohmic resistance constant ()
A Electrode surface area (m2) Rs,Rsh Photovoltaic serial and shunt resistors ()
a, b Constants terms (V/°K)
S Over voltage coefficients on electrodes (V)
E Cell reversible potential (V)
t Over voltage coefficients on electrodes (m2/A)
F Faradays constant: 96485.3399(C mole−1)
T Temperature (°K)
FELZ Produced hydrogen moles per second (mol/s)
V, Vd Voltage (V)
FPEMFC Consumed hydrogen moles per second (mol/s)
Vb Volume of the tank (m3)
Fgaz Gas molar flow rate (mol/s)
VELZ The electrolyze voltage (V)
I Current (A)
VPEMFC Voltage of the fuel cell (V)
Icc Photo generated current (A)
Vrev Reversible cell voltage (V)
Ico Current consumed by PEMFC (A)
Vt Diode thermodynamic potential(V)
Id Diode current (A)
z z = 2 for H2,z = 4 for O2
Ilimit Limitation current (A)
zc Compressibility factor as a function of
Ip Current produced by ELZ (A)
pressure
Is Reverse saturation current (A)
α Diode ideality factor
Ish Shunt resistor current(A)
ηF Faraday efficiency (%)
K Boltzman constant: 1.3806503×10-23 (J/°K)
KRI Empirical constant (/A)
KRT Empirical constant (/°K) I. Introduction
MH2 Molar mass of hydrogen (kg/kmol)
nc Cells in series per stack Internationally, governments and multinational
companies are investing huge funds in the development
145
of energy production from renewable sources and The contribution of our work is to model and develop
particularly those based on hydrogen, which will become an approach to the management of a microgrid using
in the near future an essential solution to energy and MAS.
environmental problems. In addition, the massive Our paper is organized as follows: after the
proliferation of these distributed resources and the arrival nomenclature and the introduction, section 3 outlines the
of electric vehicle require innovative and intelligent context and work tools. Section 4 describes each physical
solutions for the control and management of the and software component of the microgrid. In section 5
electricity grid. This intelligent control of power systems the focus will be on the control and management of the
is called "Smart Grid" (SG) [1]. network by MAS. In section 6, we present the results of
The balance between production and consumption is simulations, as well as the interest and the reliability of
provided by SG implementing parallel systems of the chosen configuration. Finally a conclusion and a list
communication and integrating distributed information of references will complete our paper.
networks.
Recently, control and management of MGs by MAS
[2] have indeed attracted the attention of many II. Context and Work Tools
researchers, given the perfect concordance of their II.1. Microgrid Subtest
characteristics with those of MGs. Indeed:
- The distributed nature gives the MAS system the During our investigations, we are interested in
ability to adapt well to SG. studying the microgrid (MG) ensuring the supply of a
- The autonomy of the MAS: MAS must take proactive village of 1300 people located 50 km north of Agadir.
steps in seeking to meet their needs based on their For this village, the daily consumption is estimated at
roles and their limits. For example in the case of a 330 kWh. An extract from the daily consumption is
charge that seeks to be powered via a source, MAS presented in Fig. 1; where you notice a peak
are trying to minimize costs while maintaining consumption of 24 kWh goshawks of 20 h [17].
operational constraints. Meteorological data concerning this village are extracted
- The opening of the system: MAS easily adapt to from PVGIS (Photovoltaic Geographical Information
changes in the network, even in the event of a fault, System).
an addition or removal of elements. This procedure is The model that we have developed in this work is
ensured even if these operations were not included in designed to optimize production and consumption using
the system design. the microgrid configuration shown in Fig. 2.
The application of multi-agent systems is also LOAD PROFILE
developed for solving optimization problems [3], the 25
restoration of the initial state of MG [4], the integration
of electric vehicles [5] and cyber security in smart Grid
Power (Kw)
20
[6].
According to some authors of scientific literature,
several areas implement management based on the MAS. 15
For example, Dimeas [7] introduced a general
framework for control of microgrid based on technology
of MAS. Pipattanasomporn [8] has proposed a 10
0 5 10 15 20
framework for the response to the request that can be Tim e (h)
implemented within a network of an intelligent House.
Logentthiran [9] introduced the operation in real time of Fig. 1. Typical load profile
a microgrid using MAS. Nagata [10] introduced a
method to keep the system voltage in the optimal range GRID
in order to avoid voltage instability phenomena using AC
multiple agents. Wang [11] proposed models of fuel
cells, wind turbines and photovoltaic-based systems.
Muhammad Ali [12] studied architecture which INVERTER
LOAD AC/DC
upgrades the existing system by monitoring all control 330 KWh/d
signals through which they pass between production DC
units, servers, connected loads and the protective devices
by using the TCP/IP. FUEL CELL
On modeling and simulation of decentralized 25 KW
renewable energy systems, several studies have been

presented in scientific and technical literature [13], [14], PHOTOVOLTAIC H2 TANK ELECTROLYZER
[15] and [16]. These studies have sufficiently detailed 125 KW 20 KG 80 KW
such systems models: PVECS, PEM, ELZ and
H2TANK. Fig. 2. Configuration of the microgrid (physical layer)
146
This configuration consists of photovoltaic panels

=  
(PVECS) with an output of 125 kW, of a fuel cell
(PEMFC) of power 25 kW, an alkaline electrolyzer
(ELZ) with a power of 80 kW and storage of hydrogen
(H2TANK) originally half full and with a maximum III.1.2. Model of the Fuel Cell and the Electrolyzer
capacity of 20 kg. The equations describing the workings of a fuel cell
are identical to those governing the electrolyzer. In this
II.2. Investigative Tools work we will look at the equations:
a. The flow of hydrogen and oxygen consumed by a fuel
Aware of our investigations, we use two tools that are cell or those produced by an electrolyzer. These flows
widely used in models with attractive to MAS: are directly proportional to current delivered. Indeed
a. The platform JADE multi-agent which comprises a the molar flow is given by Eqs. (6) and (7) [22]:
set of tools necessary for the construction and
commissioning of agents within a specific 1
environment [18]. = (6)
b. Matlab / Simulink used to simulate the functioning
and control of a physical system by using
mathematical equations [19]. = (7)
These tools are used to develop decentralized and
autonomous systems that collaborate with another tool b. The voltages of the PEMFC and that of ELZ.
called MACSimJX developed in Java by Charles R. Particularly, the relationship between the voltage and
Robinson, [20]. the current delivered by PEMFC or consumed by the
The overlay provides the interface, between Simulink ELZ, are given by (8) and (9) [22]:
and JADE as shown in Fig. 3.
= − + +
SIMULINK JADE (8)
− + − + (1 − )
MACSimJX MACSimJX
Client Server
= + + +1 (9)
Fig. 3. Interfacing provided by MACSimJX III.1.3. Hydrogen Storage System Model (H2TANK)
One of the hydrogen storage techniques is physical
III. Microgrid Configuration hydrogen storage which is to store hydrogen in the form
of compressed gas or liquid.
III.1. Description of the Physical System The difference between the amount of hydrogen
The study articles of scientific and technical literature, produced by the ELZ and consumed by the PEMFC is
mentioned in the introduction, for modeling and sent to the storage tank.
simulation of decentralized renewable energy systems, In this study, the model of hydrogen storage is based
led us in the remainder of this document, to be limited to on Eq. (10) [23] which directly calculates the tank
simplified models to accelerate code execution time, pressure using the ratio between the flow of hydrogen
under the Matlab environment. The models we have and the volume of the tank:
chosen will be described below.
− = (10)
III.1.1. Model of PVECS
Fig. 4 summarizes the electric model chosen for the
photovoltaic cell whose Eqs. (1)-(5) characterized the
behavior [21]:
= + + (1)
= [ − 1] (2)
= + (3)
+
= − [ − 1] − (4) Fig. 4. Model of the photovoltaic cell
147
III.1.4. Model of DC-DC Converters Indeed, we assign to each component of the MG an

agent who takes his name as shown in Figure 7. Then we
One of the converters that we incorporate in to our
assign behavior, requirements and tasks for each of these
models is elevator type or parallel (Boost).
agents.
This converter, which is shown schematically in Fig.
MGCC_Agent: Agent that coordinates the operation of
5, is controlled by a PI corrector using the MPPT
agents to accomplish the overall task of the system.
controller P & O (Observe and disturbed) that is widely
GRID_Agent: Agent that informs other agents selling
treated in literature, for example in [24]. This converter is
prices and purchasing power, and the PG power
used to level the voltage of the fuel cell to the DC bus.
exchanged with the MG. PG is considered positive if the
network of the power absorbed and negativein the
opposite case.
PVECS_Agent andPEMFC_Agent:Agents that
exchange information about the unit (PVECS or
PEMFC) with other agents for example: source PVECS
or PEMFC type, identification number, power activated,
fuel, price with which the user agrees to sell or buy
energy, etc.
ELZ_Agent: Agent that monitors the amount of
hydrogen produced while informing the other agents.
H2TANK_Agent: Agent that ensures the monitoring of
Fig. 5. Boost DC-DC converter the State of the volume of the tank of hydrogen. To
simplify the study and the programming code, the
The second type of converters implemented in our influence of the temperature and the pressure of the gas
modeling is step-down type or series (Buck). The latter, are not taken into account in this work. We will also
as Fig. 6 shows, has the role to ensure the adaptation assume that the hydrogen container is dimensioned to
levels of voltage-current of the electrolyzer and the DC ensure the maximum volume produced by hydrogen
bus [25]. storage.
This converter uses a corrector of the classic PI. LOAD_Agent: Agent that monitor and controls
expenses, according to the preferences of the user. It
connects or disconnects the loads depending on whether
they are critical or non-critical and real-time by energy
price.
III.2.2. Overview of the Developed Application

As shown in Fig. 8, a graphical user interface was
developed with the aim of carrying out different tasks,
Fig. 6. Buck DC - DC converter namely; the execution of the circuit in Simulink, the
launch of the agents under JADE and the visualization of
simulation results.
III.2. Management System As shown in Fig. 9, the agents in action are visible on
The control and management of our MG under study the graphic interface of JADE.
are provided by a set of software entities called agents. Following the launch of the agents under JADE Figure
For JADE tool; those agents work together in partnership 10 highlights a few messages exchanged between the
for the execution of their duties, in order to achieve the different agents that come into action.
realization of the overall objective of the system; witch
constitutes an MAS. ELZ_AGENT
Each of these agents must be able to handle multiple
tasks simultaneously in response to various external GRID_AGENT H2TANK_AGENT
events. To make this management more effective, each
JADE agent is composed of a single thread and every MGCC_AGENT
behavior, that the latter (thread) is composed of, is an
object of type Behavior [18].
PVECS_AGENT PEMFC_AGENT
III.2.1. Description of Agents Associated

with MG Under Study LOAD_AGENT
The implementation of an MAS at the JADE platform

is first to set up our agents MG. Fig. 7. Software configuration of the MAS associated with MG studied
148
Indeed, these rated powers PL, PLC, PLNC, PG, PV, PELZ
and PEMFC are respectively the power load, critical load,
non-critical loads, network (GRID) of photovoltaic
(PVECS ),the electrolyzer and the PEMFC. These
powers are linked via relationship (11) and (12):
+ + = + (11)
The loads to be supplied are classified into two types;

those that are critical: PLC and those that are non-critical
power PLNC, such as:
= + (12)
The flowchart in Fig. 12 describes the control

Fig. 8. Graphical user interface for application
algorithm that we have developed for the overall
operation and management of MG under test.
The block load management contained in our
organizational chart, allows good management and
ensures a balance between supply and demand, by
disconnecting or shifting the operation of non-critical
loads.
The flowchart under study seeks to fully exploit the
PVECS; Indeed, when PV > PL, solar panels provide
power to the loads completely; while the PEMFC is
switched off and the excess power (PG < 0) can be
injected to the grid.
In case PLC < PV < PL, it feeds only critical loads,
while putting off the PEMFC and ELZ. However, the
appeal of the grid or the PEMFC is done only when the
Fig. 9. Graphical user interface for JADE condition PV < PLC is satisfied in this case: PEMFC > 0 and
PG ≥ 0.
V. Simulation Results
The MAS system developed with JADE at base of the
flowchart in Fig. 12 and the physical system under
SIMULINK (Fig. 11), has enabled us to develop a
control strategy using collaborative agents for the overall
task of the system.
All statements submitted subsequently relate to July
and November 2012.
TABLE I
BLOCKS MODELING MG UNDER STUDY
Fig. 10. Graphical user interface of the agent sniffer jade Bloc Model
1 Meteorological data
2 Photovoltaic panels
III.3. Global Model 3 Power inverter
4 Filter
The global model that we developed under Simulink is 5 Step-up transformer 240 V/25 kV
represented on Fig. 11. It is composed of a set of blocks 6 Grid
that Table I summarizes. 7 Loads
8 Step-down converter or series (Buck)
9 Electrolyzer
10 Hydrogen tank
IV. Management Algorithm Microgrid 11 Truck or parallel converter (Boost)
12 PEMFC
Managing the flow of energy in the SG, according to a 13 Interfaces measurement
control algorithm, provides smart features; eg load 14 S-FunctionMACSimJX
management and operation of island mode (autonomous). 15 Breakers
It is therefore interesting to bring equation, and ignoring 16 Visualization Tools
losses, power flows involved in the SG.
149
Fig. 11. SIMULINK diagram
Start
Read powers
Pv, PLC, PLNC, PEMFC, PELZ, PG
YES NO
PV ≥ PL
NO YES
PL>PV ≥ PLC
100 % working PV
PEMFC= 0 ;PG< 0 PEMFC =0 ;PELZ = 0 ; PG= 0
PV<PLC
YES Load Management
NO
PEMFC > 0 ;PG≥ 0
Load Management
Fig. 12. Management flow chart of microgrid
Before presenting the simulation results of the MG July 1st particularly.

under test, we set, depending on weather data Figure 13,  As shown in Figure 17, the electrolyzer operates at its
the typical daily horizontal radiation chosen for the maximum power during the month of July;
months of July and November. It is clear, in this curve Particularly between 10:00 and 16:00 of the 1st of
that for the month of July, the radiation and the duration July.
of the insulation are more important than those of the Figure 18 shows the simulation results for the absolute
month of November. value of the excess power of the MG during the first five
Figs. 14, 15, 16 and 17 illustrate the results of days of July and November.
simulation on daily developments produced by the This excess power will be injected to the public
PVECS and the PEMFC powers, as well as those network. Fig. 19 shows a zoom on the 1st day of the
consumed by the ELZ and the loads. months July and November.
These simulation results show the following facts: According to these curves, we note that during the
 All the results that summarized in Figs. 14 to 17, months of, July injected average power is more important
shows that the average photovoltaic power produced than injected during the month of November, which
in July is obviously most higher than that November. generally is the case since the month of July is generally
 According to Figures 16 and 17, the fuel cell works more informed and hot by report in November.
only when the PVECS is unable to provide the energy Regarding the simulated evolution of the amount of
required by users or between 19 h and 06 h for the 1st hydrogen stored in the tank during the first 5 days of July
of November and between 19:00 and 07:00 for the and November, it is shown in Fig. 20.
150
140
H o u r l y G l o b a l S o l a r R a d i a ti o n ( k w /m 2 )
1.4
JULY NOVEMBER 120
1.2 PVEC S
PEMFC
100
1 EL Z
Po w e r (KW)
80 L OAD
0.8
60
0.6
40
0.4 20
0.2 0
0 4 8 12 16 20 23
0 Ti m e (h )
0 4 8 12 16 20 23
Ti m e (h ) Fig. 17. Evolution of daily power for the July 1st
14
Fig. 13. Global horizontal radiation daily in July and November
November
12 july
PVECS PEMFC ELZ LOAD
120 10
Power (KW)
100 8
80 6
Po w e r (kw )
60 4
2
40
0
20 0 20 40 60 80 100
Ti m e (h )
0
0 23 47 71 95 119 Fig. 18. Daily power injected to the grid in July and November
Ti m e (h )
12
Fig. 14. Evolution daily power in November
Novem ber
P o w e r s o l d to g r i d ( K W )
PVECS PEMFC ELZ LOAD 10

july
140
8
120
6
100
P o w e r (K W )
80 4
60 2
40
0
0 5 10 15 20
20
Tim e (h)
0
0 23 47 71 95 119 Fig. 19. Daily power injected to the grid the first day
Tim e (h)
of November and July
Fig. 15. Evolution daily power in July November July

H y d r o g e n T a n k s to r a g e l e v e l ( k g )
20
120
100 PVECS
15
PEMFC
80 ELZ
P o w e r (kw )
LOAD 10
60
40
5
20
0
0 0 20 40 60 80 100
0 4 8 12 16 20 23
Tim e (h) Tim e (h)
Fig. 16. Evolution of daily power for the November 1st Fig. 20. Quantity of hydrogen in the tank in Novemberand July
151
Fig. 21 matches the 1st day of the months July and System for Demand Side Management in Smart Grid, Power
Electronics and Drive Systems (PEDS), IEEE2011 Ninth
November. This curve shows that the hydrogen is stored
International Conference on, Singapore 5-8, Dec. 2011, pp. 424-
when there is an excess of solar energy, that is to say, 429.
between 7 AM and 5 PM. [4] T. Logenthiran, D. Srinivasan, A.M. Khambadkone, H.N. Aung,
This hydrogen is re-used in the opposite case, that is to Scalable Multi-Agent System (MAS) for Operation of a
Microgrid in Islanded Mode, Power Electronics, Drives and
say, between 5 PM and 7 AM. It is also important to note Energy Systems (PEDES) & 2010 Power India, IEEE2010 Joint
that the hydrogen tank is never empty, which can meet International Conference on, New Delhi 20-23, Dec. 2010, pp. 1-
the demand of consumers. 6.
[5] Shao Shengnan, M. Pipattanasomporn, S. Rahman, Grid
200 20 Integration of Electric Vehicles and Demand Response With
July Hydrogen Tank level Customer Choice, Smart Grid, IEEE Transactions on, Vol. 3,
Issue 1, 2012, pp. 543-550.
Novemeber Hydrogen Tank level
H yd ro g e n Ta n k L e ve l (kg )
15 [6] W. Y. Wang, Z. Lu, Cyber security in the Smart Grid: Survey and
July PV Power challenges, Computer Networks (Amsterdam, Netherlands :
PV Po w e r(kw )
1999), Vol. 57, Issue 5, 2013, pp. 1344-1371.

November PV Power [7] A.L. Dimeas, N.D. Hatziargyriou, Multi-agent Reinforcement
100 10
Learning for Microgrids, Power and Energy Society General
Meeting, IEEE2010, Minneapolis, MN 25-29 July, 2010, pp. 1-8.
[8] M. Pipattanasomporn, M. Kuzlu, S. Rahman, An Algorithm for
5 Intelligent Home Energy Management and Demand Response
Analysis, Smart Grid, IEEE Transactions on, Vol. 3, Issue 4,
2012, pp. 2166-2173.
0 0 [9] T. Logenthiran, D. Srinivasan, A.M. Khambadkone, HtayNwe
0 5 10 15 20 25
Aung, Multiagent System for Real-Time Operation of a Microgrid
Ti m e (h )
in Real-Time Digital Simulator,Smart Grid, IEEE Transactions
on, Vol. 3, Issue 2, 2012, pp. 925-933.
Fig. 21. Quantity of hydrogen in the tank with PV power [10] T. Nagata, M. Ishida, An Autonomous Agent Approach to Power
for the 1st day of November and July System Voltage Control, Environment and Electrical Engineering
(EEEIC), IEEE2011 10th International Conference on, Rome 8-11
May, 2011, pp. 1-4.
VI. Conclusion [11] Caisheng Wang, Modeling and control of hybrid
wind/phovoltaic/fuel cell distributed generation systems,Ph.D.
This paper focuses on the monitoring and modeling of MontanaState University, Bozeman-Montana, College of
decentralized renewable energy sources. So, a part of this Engineering, 2006.
article is devoted to the modeling of different elements of [12] M. Ali, M. Zakariya, M. Asif, A. Ullah, TCP/IP Based Intelligent
Load Management System in Micro-Grids Network Using
the Microgrid. The rest of the paper outlines the MATLAB/Simulink, Energy and Power Engineering (EPE),
development of the control strategy, focusing on the Vol. 4, No.4, 2012, pp. 283-289.
behavior of agents located at the JADE platform and the [13] Nowdeh, S.A., Tousi, B., Zoraghchian, A.A., Hajibeigy, M.,
collaboration between these agents to accomplish the Optimal PV and FC application in a hybrid power system with the
aim of selling electrical energy to distribution network, (2011)
overall task of the system. International Review on Modelling and Simulations (IREMOS), 4
Following our investigations, we have been able to (5), pp. 2392-2401.
demonstrate the ability of multi-agent systems for [14] Sebt-Ahmadi, S.M., Ebrahimi, S., Pirnazari, A., Ranjbar, A.-M.,
managing a microgrid of solar panels, a fuel cell, an An optimal hybrid power generation scheme in a power system -
A practical experience, (2013) International Review of Electrical
electrolyzer and a hydrogen storage element. We can Engineering (IREE), 8 (5), pp. 1566-1577.
confirm that the MAS systems are able to provide energy [15] M. Nazari, M. Abedi, G. B. Gharehpetian, H. Toodeji,
supply for the village under test while minimizing the Photovoltaic Array, Fuel Cell and Electrolyzer Connection to Grid
costs involved in controlling the balance between by Direct Non-Linear Controlled H-Bridge Multilevel Inverter,
(2010) International Review of Automatic Control (IREACO), 3
demand and electricity production, and the effective (6), pp. 633-640.
exploitation techniques HEM aimed at improving [16] Mellit, A., A soft-computing technique for modeling and
comfort, safety and mastery of energy consumption. simulation of stand-alone photovoltaic power supply system,
MAS systems can also be programmed to have the ability (2007) International Review of Electrical Engineering (IREE), 2
(4), pp. 480-488.
to isolate the microgrid in case of occurrence of a failure. [17] FaziaBaghdadi, Modélisation et simulation des performances
This creates an effective system and self-healing. d'une installation hybride de conversion d’énergies
renouvelables,Ph.D., Université Mouloud Mammeri de Tizi-
Ouzou, 2011.
References [18] Fabio Luigi Bellifemine, Giovanni Caire, Dominic Greenwood,
Developing Multi-Agent Systems with JADE, Vol. 7, John Wiley
[1] Amirat, Y., Benbouzid, M., Wang, T., Turri, S., An ensemble & Sons, 2007,p. 300.
empirical mode decomposition approach for voltage sag detection [19] SimPowerSystems™, User's GuideVol. R2013a, Hydro-Québec
in a smart grid context, (2013) International Review of Electrical and the Math Works, Inc., 2013.
Engineering (IREE), 8 (5), pp. 1503-1508. [20] C.R. Robinson, P. Mendham, T. Clarke, MACSimJX: A Tool for
[2] M. Merdan, A. Prostejovsky, I. Hegny, W. Lepuschitz, F. Andrén, Enabling Agent Modelling with Simulink Using JADE,Journal of
T. Strasser, Power Distribution Control using Multi-Agent physical agents, Vol. 4, n. 3, September 2010, pp. 1-7.
Systems, Recent Advances in Robotics and Automation, Studies in [21] Farhat, S., Alaoui, R., Kahaji, A., Bouhouch, L., MPPT efficiency
Computational Intelligence, Vol. 480, Springer-Verlag, Berlin test by neural networks and P&O algorithm, (2013) International
Heidelberg, 2013, pp. 323-333. Review of Electrical Engineering (IREE), 8 (5), pp. 1548-1555.
[3] T. Logenthiran, D. Srinivasan, Tan Zong Shun, Multi-Agent [22] W. Gao, V. Zheglov, G. Wang, S.M. Mahajan, PV - Wind - Fuel
152
Cell - Electrolyzer Micro-Grid Modeling and Control in Real My Rachid El Moutawakil Alaoui Professor at
Time Digital Simulator, Clean Electrical Power, IEEE2009 the University Ibn Zohr Agadir,Morocco, with a
International Conference on, Capri 9-11 June, 2009, pp. 29-34. Bachelor in Electronics ENSET, Rabat and
[23] M. Uzunoglu, O.C. Onar, M.S. Alam, Modeling, Control and Masters in Industrial Engineering at the ENSA
Simulation of a PV/FC/UC based Hybrid Power Generation Agadir. His doctoral thesis focused on the
System for Stand-Alone Applications, Renewable Energy, Vol. Defensive Software processors. Permanent
34, Issue 3, 2009, pp. 509-520. member of the Research Team of Advanced
[24] Ciresan, A., Draghici, D., Gurbina, M., Lascu, D., A new boost Technologies and Engineering Renewable
DC-DC converter exhibiting low stresses and high efficiency, Energy ERTAIER. Responsible for electromagnetic compatibility
(2013) International Review of Electrical Engineering (IREE), 8 (CEM) research focus.
(6), pp. 1694-1700.
[25] Ajaamoum, M., Kourchi, M., Bouachrine, B., Ihlal, A., Pr. Ali Moudden was born in 1958 in Morocco.
Bouhouch, L., Photovoltaic panel emulators, design and He obtained (1987) the "Doctorat d'état" in
implementation using rapid prototyping technique, (2014) physics at the University of Caen (France). He
International Review of Electrical Engineering (IREE), 9 (5), pp. founded with other colleagues in 1991, the
1012-1020. laboratory of instrumentation and measurements
at the Faculty of Sciences of Agadir (Morocco),
that became in 2005 the laboratory of metrology
Authors’ information and information processing. Laboratory
activities are focused mainly on the development of physical techniques
Lahoussine Elmahni Was born in Agadir, for materials evaluation and processes monitoring. He was director of
Morocco in 1968; Professor (1993) in electrical higher school of technology of Agadir during the period 2005-2011.
engineering with bachelor(1989) in electronics
in higher normal school of technology
(ENSET), Rabat and a post graduate degree in
(2008), of energy and environment in the
National School of Applied Sciences (ENSA) of
Agadir, His research is focused on Smart Grid,
electric vehicles, demand response, energy efficiency, renewable
energy integration, energy storage and distributed resources.
Lahoussine Bouhouch Professor of higher

education at the ESTA (High School of
Technologies of Agadir), Ibn Zohr University,
Agadir, Morocco. PhD Electrical Engineering at
the Nancy I University, France in 1988 and state
doctorate in Electrical Engineering in 2007.
Responsible of the research team ERTAIER
(Research Team in Advanced Technologies and
Engineering of Renewable Energies). His research focuses on topics
related to renewable energy, instrumentation and electromagnetic
compatibility (EMC).
153
Load Frequency Control of Multi Area SSSC and CES Based System
Under Deregulation Using Particle Swarm Optimization
P. Marimuthu, B. Basavaraja, Subhransu Sekhar Dash
Abstract – This paper deals with the application of various evolutionary optimization techniques
for designing the gain of integral controller in a multi area hydrothermal system under
deregulation.
The various techniques which have been used are Simulated Annealing (SA), Genetic Algorithm
(GA) and Particle Swarm Optimization (PSO). A two area hydrothermal system in the presence of
Static Synchronous Series Compensator (SSSC) and Capacitive Energy Storage (CES) under
deregulation is considered to exemplify the optimum parameter search. The integral of square of
error (performance) index is considered in the search of optimal Automatic Generation Control
(AGC) parameters. The results presented in this paper demonstrate the superior working of PSO
over others in the tuning of AGC. Copyright © 2015 Praise Worthy Prize S.r.l. - All rights
reserved.
Keywords: Simulated Annealing, Genetic Algorithm, Particle Swarm Optimization, Static

Synchronous Series Compensator, Capacitive Energy Storage, Deregulation
Nomenclature The various entities in such a type of system are

generating companies (GENCOs), transmission
α, β Weighing factors companies (TRANSCOs) and distribution companies
∆f1 Change in frequency in area 1 (DISCOs).
∆f2 Change in frequency in area 2 The fundamental role of LFC in this type of power
J Performance index system structure is to enable exchange of power and also
Vm Voltage magnitude of the bus m to provide better conditions for trading of power. The
Vn Voltage magnitude of the bus n concept of control of generation in power systems under
Xs Transformer leakage reactance open market scenario has been presented in [3].
Smn Power flow from the bus m to n The implementation of independent system operator
θc Phase angle of the current (ISO) to balance the economics along with efficient
K Boltzmann constant working of the system also has been presented in [4], [5].
apf Area participation factor A detailed review of how the concept of LFC under
open market scenario can be simulated and assessed has
been presented in [6], [7]. On the other hand, various
I. Introduction
FACTS devices have been widely employed for the
The process of maintaining several sets of balances control of power systems which provide better stability
form the process of successful operation of power [8]-[11].
system. The two balances which are predominant factors Static Synchronous Series Compensator (SSSC) and
to keep frequency constant are load-generation Capacitive Energy Storage (CES) are seen as effective
scheduling and actual tie line flows [1], [2]. devices to bring about better dynamic response of the
A good operation of system is identified by its system. To compare the improvement of dynamic
constant frequency of operation and quality of supplied performance of the system by optimizing the gain of
power to its consumers. integral controller in a SSSC and CES based
This is generally carried out by varying the generation hydrothermal system under deregulation, various
as per the load demand. Generally if frequency is low, evolutionary techniques like SA, GA, and PSO have
then the amount of generation is increased and if the been employed.
frequency is high, the generation is decreased. The Simulation results depict that the PSO technique
electric industry at current stage is undergoing change ([16]-[20]) gives good results in terms of Peak time,
from a form of vertically integrated utility system which settling time and overshoot over the other techniques and
provides power at regulated rates to a system that also the performance index value is also less as compared
constitutes companies competing with each other to sell to the other methods.
power at lower rates.
154
II. System Investigation Nominal parameters of the system are given in the
Appendix.
The AGC system investigated is composed of an The performance index (PI) namely, integral of square
interconnection of two areas. Area 1 comprises of a of error (ISE) given by:
reheat system and Area 2 comprises of hydro system.
The detailed transfer function models of speed t
   f1 
2
governors and turbines are discussed and developed in J    f 22  Ptie2 12 (1)
the IEEE committee report on Dynamic models for 0
Steam and Hydro turbines in Power systems [12].
The detailed small perturbation transfer function block is considered in this work to compare the performance of
diagram model of two area hydrothermal system under various optimization methods.
open market scenario is shown in Fig. 1.
Cpf11 Cpf12
+ +
Cpf21 + Cpf22
p.u load of + p.u load of
Disco 1 Disco 2
Cpf31 Cpf32
+ +
Cpf41 + + Cpf42
B 1 1
R1 R2 PD1(s )
Thermal area
PCES
ACE
+ - + -
1 1 1  sK r Tr + - K p1
Ki/s a
p - 1  s Tg 1  s Tt 1  s Tr 1  s T p1
+ +
+
+
- +
Demand of
1 1 1  sK r Tr discos in
a area 1 to
p - 1  s Tg 1  s Tt 1  s Tr Gencos in
+ area 2
+ +
+ 2  T12
+ s
- PSSSC -
+ + Demand of
-1 discos in
- 1 1  sTR 1 - sTw
a area 2 to
p 1  s T1 1  s T2 1  0.5s Tw Gencos in
-1 area 1
-
+ + + +
- 1 1  sTR 1 - sTw KP2
Ki/s a
1  s T1 1  s T2 1  0.5s Tw + - 1  s TP 2
p
+ -
-
ACE + Hydro area
PCES
1 1 PD 2 (s )
B R3 R4
Cpf13 Cpf14
+ +
Cpf23 + Cpf24
p.u load of + p.u load of
Disco 3 Disco 4
Cpf33 Cpf34
+ +
Cpf43 + Cpf44
Fig. 1. Two Area SSSC and CES based Hydrothermal System under Deregulation
155
III. Design of SSSC Vm  Vs  Vn Vm  Vn   Vs 

I     I o  I (4)
Fig. 2 shows the schematic diagram of SSSC located in jX T  jX T   jX T 
series with the tie line between the interconnected areas.
Generally the equivalent circuit is represented by a series The term I is an additional current term due to
connected voltage source V s along with a transformer SSSC voltage Vs . The power flow from bus m to bus n
leakage reactance X s as shown in Fig. 3. The main can be written as Smn  Vm I   Smno  S mn which
controllable parameter is V s , which in fact represents the implies:
magnitude of injected voltage. The phasor diagram of
system is shown in Fig. 4 by taking into account the Pmn  jQmn   Pmno  Pmn   j  Qmno  Qmn  (5)
operating conditions of SSSC. Based on the above Fig. 4,
when Vs  0 , the current I o of the system can be written
where Pmno and Qmno are the real and reactive power
as:
flows respectively when Vs  0 . The change in real
Vm  Vn power flow caused by SSSC voltage is given by:
Io  (2)
jX T
VmVs
Pmn  sin  m    (6)
XT
where X T  X L  X S . The phase angle of the current
can be expressed as:
when Vs lags the current by 90 , Pmn can be written
 V cos  n  Vm cos  m  as:
 c  tan 1  n  (3)
 Vm sin  m  Vn sin  n  VmVs
Pmn  cos  m   c  (7)
XT
But Eq. (2) can be expressed in a generalized form as:
Fig. 2. Schematic of SSSC applied to two - area interconnected system
Fig. 3. Equivalent circuit of SSSC
Fig. 4. Phasor diagram at (a) Vs=0; (b) Vs lagging I by 90°
156
Fig. 5. Structure of SSSC
Fig. 6. Block diagram of CES applied to LFC
From Eq. (7) the term cos  m   c  can be written as: Pmn  Ptie  PSSSC
which implies:
Vn
cos  m   c   cos  n   c  (8)
Vm  VmVn 
 X sin  mn  
  
T
Referring to Fig. 4 it can be written as: PSSSC (15)
Vs 
 
yw  V  V  2V V cos 
2 2 
cos  n   c   (9)  m n m n mn 
xy
Based on Eq. (15) it can be observed that the power
and it can be seen as: output of SSSC can be controlled by varying the SSSC
voltage Vs and thus the frequency and tie line
yw  Vm sin  mn (10) deviations can be controlled. Fig. 5 shows the structure
of SSSC to be incorporated along the hydrothermal
also: system. It can be seen from Fig. 5 that the structure of
xy  Vm2  Vn2  2VmVn cos  mn SSSC consists of gain block K SSSC , time constant TSSSC
(11)
and two stage phase compensation blocks having time
and  mn   m   n
constants T1 , T2 , T3 , T4 respectively.
Using these relationships Eq. (11) can be modified as:
VmVn Vs IV. Design of CES

Pmn  sin  mn  (12)
XT Vm2  Vn2  2VmVn cos  mn The frequency deviation or Area control error signal
is the control signal for the CES unit [9]. Fig. 6 shows the
From Eq. (12) it can be written as Pmn  Pmno  Pmn block diagram representation of CES unit employed in
the system. The structure of CES consists of gain K CES ,
which implies:
time constant TCES and two stage phase compensation
V V blocks having time constants T1 , T2 , T3 and T4 .
Pmn  m n sin  mn 
XT
V V Vs  (13)
  m n sin  mn   V. Evolutionary Techniques Employed
 XT 2 2
Vm  Vn  2VmVn cos  mn 
  Various evolutionary techniques employed in the
work have been listed below:
Linearizing Eq. (13) about an operating point it can be
written as: a) Simulated Annealing [13]:
The SA procedure simulates the process of slow
VmVn cooling of molten metal to achieve the minimum
Pmn  cos  m   n   m   n   function value in a minimization problem. The cooling
XT
phenomenon is simulated by controlling a temperature
V V Vs  (14) with the concept of Boltzmann probability distribution.
  m n sin  mn   According to the Boltzmann probability distribution, a
 XT 2 2
Vm  Vn  2VmVn cos  mn 
  system in thermal equilibrium at a temperature T has its
157
energy distributed probabilistically according to Eq. (16): that are highly adapted to their environments. GA is an
algorithm that operates on a similar principle.
 E 
P  E   exp   (16) Algorithm:
 kT 
 Initialization of binary chromosome strings of
population, each consisting of the gains.
where k is the Boltzmann constant. Therefore by
 Decoding of strings and calculation of the fitness
controlling the temperature T and assuming that the
function.
search process follows the Boltzmann probability
 Selection of strings into the mating pool based on
distribution, the convergence of the algorithm can be
Roulette wheel selection criteria
controlled.
According to Metropolis et al. (1953) if at any instant  Performing the Crossover and mutation to generate
off springs.
the current point is xt and the function value at that point
 Genetic cycle updating and stopping criteria.
is E  t   f  xt  , then the probability of the next point  The end of last iteration, the values corresponding to
being at xt 1 depends on the difference in the function the string having minimum fitness value is identified.
values at these two points i.e. E  E  t  1  E  t  and
c) Particle Swarm Optimization [15]:
is calculated using the Boltzmann probability distribution Particle swarm optimization (PSO) is a computation
given by Eq. (17): technique developed by Eberhart and Kennedy based on
the analogy of swarm of birds and school of fish. PSO
  E   mimics the behaviour of individuals in a swarm to
P  E  t  1   min 1,exp   (17) maximize the survival of the species.
  kT  
It is similar to other evolutionary computation
techniques in conducting searching for optima using an
If E  0 , this probability is one and the point xt 1 initial population of individuals ([16]-[20]).
is accepted. But if E  0 it implies that xt 1 is worse The individuals of this initial population are then
than that of xt . In this case according to Metropolis updated according to some kind of process such that they
are moved to a better solution area. In PSO system, each
algorithm, there is some finite probability of selecting the
individual adjusts its flying in a multi-dimensional search
point xt 1 even though it is worse than xt .
space according to its own flying experience and its
This probability depends on relative magnitude of E companions flying experience. Each individual is
and T . referred to as “particle” which represents a candidate
solution to the problem. Each particle is treated as a point
Algorithm:
in a D-dimensional space. The i th particle is represented
 SA is a point by point method. The algorithm begins
with an initial point and a high temperature T . as X i   xi1 ,xi 2 ,xi 3 ,.....xiD  .
 A second point is created at random in the vicinity of The best previous position (giving the best fitness
the initial point and the difference in the function value) of any particle is recorded and represented as
values f at these two points is calculated. Here in Pi   Pi1 ,Pi 2 ,Pi 3 ...PiD  . The index of the best particle
this work the difference in the function values is among all the particles in the population is represented
taken as f instead of E so as to correlate with by the symbol ' g' . The rate of position change (velocity)
performance index. for particle ‘ i ’ is represented as Vi  Vi1 ,Vi 2 ,Vi 3 ....ViD  .
 If the second point has a smaller function value, the The particles are manipulated according to Eq. (18)
point is accepted otherwise the point is accepted with and (19):
a probability of (16). In order to do this, another point
is created at random in the neighborhood of the
current point and the Metropolis algorithm is used to 
ViD  ViD  c1r1  PiD  X iD   c2 r2 PgD  X iD  (18)
accept or reject the point.
 In order to simulate the thermal equilibrium at every X iD  X iD  ViD (19)
temperature, a number of points ' n' is tested at a
particular temperature before reducing the where c1 and c2 are positive constants and r1 and r2 are
temperature. The algorithm is terminated when
sufficiently small temperature and small enough uniformly distributed random numbers in the range (0,
change in function values are obtained. 1).
b) Genetic Algorithm [14]: Algorithm:

In nature, a combination of natural selection and  In a typical run of the PSO, an initial population of
procreation permits the development of living species random solutions is generated. Each particle keeps
track of its coordinates in hyperspace which are
158
associated with the fittest solution it has achieved so of performance index obtained during various
far. optimization techniques.
 The value of the objective function Pi is also tracked. It can be observed from Table I and Table II that the
The global version of the PSO keeps track of the performance of PSO is better than that of other
overall best value and its best location obtained so far optimization techniques in terms of Peak time, settling
by any particle in the population, which is called Pg . time and overshoot and also the performance index of the
system is very less as compared to the other optimization
The PSO at each step changes the velocity and techniques.
position of each particle towards its Pi and Pg . Fig. 7 shows the comparison of frequency deviation of
 The particle swarm optimization then updates the both areas and tie line power error deviation with respect
particle coordinates based on (18) and (19) to all techniques for normal case. Fig. 8 shows the
 These above two steps are repeated from population various generations of Gencos in thermal area during
to population until a stopping criterion terminates the normal case.
search producing the optimum gains. Fig. 9 shows the generations of Gencos in hydro area
during normal case.
VI. Results and Discussions

A multi area SSSC and CES based hydrothermal
system under open market scenario has been considered
for the study. It is to be noted that each Genco
participates in LFC as per the following participation
factors: apf1 = 0.5, apf2 = 0.25, apf3 = 0.25, apf4 = 0.5,
apf5 = 0.25, apf6 = 0.25. A step load disturbance of 0.4%
is considered in either of the areas.
A special case of contract violation has also been
considered in which an additional load of 0.3% is
considered in both the areas after the time span of 35 and
70s. It is to be noted that this additional load is taken up
by the Gencos which lie in the area in which contract
violation has occurred.
The Disco Participation Matrix Considered in this
work is given by:
Table I shows the peak time, overshoot and settling

time for the frequency deviations obtained during various
values of gain of integral controller obtained through Fig. 7. Frequency and tie line power error deviations
various optimization methods. Table II shows the value during normal case
TABLE I
COMPARISON OF PERFORMANCE OF OPTIMIZATION TECHNIQUES
Optimization Peak time Settling
S.No Integral Gain values Area Considered Overshoot (Hz)
Technique (s) Time (s)
Simulated ki1  1.35 Thermal area 2.73 0.00294557 7.04
1
Annealing ki 2  0.267 Hydro area 0.88 0.00590737 3.525
Genetic ki1  0.2307 Thermal area 2.18 0.00322513 4.34

2
Algorithm ki 2  0.7918 Hydro area 0.9 0.0060065 2.915
Particle Swarm ki1  0.3578 Thermal area 2.155 0.00314339 4.12

3
Optimization ki 2  0.9335 Hydro area 0.895 0.00598582 2.76
159
Fig. 10. Frequency and tie line power error deviations during
Fig. 8. Generation of Gencos of Thermal area during normal case
contract violation case
Fig. 9. Generation of Gencos of Hydro area Fig. 11. Generation of Gencos of Thermal area
during normal case during Contract violation case
160
Fig. 10 depicts the comparison of frequency deviation by Particle Swarm Optimization brings about better
of both areas and tie line power error deviation with response in terms of overshoot and settling time as
respect to all techniques for contract violation case. shown in the simulation.
Fig. 11 shows the various generations of Gencos in
thermal area during contract violation case. Fig. 12
shows the generations of Gencos in hydro area during
contract violation case. Fig. 13 and 14 depict the
performance index of the system for various techniques
during normal case and contract violation case.
TABLE II
COMPARISON OF PERFORMANCE INDEX VALUES FOR NORMAL CASE
S. No. Optimization Technique Performance Index
5
1 Simulated Annealing 3.241  10
Fig. 13. Comparison of Performance index values
5
2 Genetic Algorithm 2.15  10 during Normal case
Particle Swarm 5
3 1.91  10
Optimization
Fig. 14. Comparison of Performance index values

during Contract Violation
Appendix
(a) System data
T p1 , T p 2 =20s; K p1 , K p 2 =120Hz/p.u. Mw;
Pr1 , Pr 2 =1200Mw; Tt =0.3s; T g =0.08s, Tw =1s;
Tr  5s , T1  41.6s , T2 =0.513s; R1 , R2 = 2.4Hz/pu
Mw; T12 = 0.0866s; B1 , B2  0.4249p.u Mw/Hz;
(b) CES data:

T1=0.279s; T2=0.026s; T3=0.411s; T4=0.1s;
KCES=0.3; TCES=0.0352s
(c) SSSC data:

Fig. 12. Generation of Gencos of Hydro area T1=0.188s; T2=0.039s; T3=0.542s; T4=0.14s;
during Contract violation case
KSSSC=0.292; TSSSC=0.03s
VII. Conclusion References

Various Evolutionary Optimization techniques have [1] C. Concordia and L.K.Kirchmayer, “Tie-Line Power and
been successfully applied to tune the gain of integral Frequency Control of Electric Power System - Part II”, AIEE
controller. A two- area hydrothermal system under Transaction, vol. 73, Part- 111-A, pp. 133-146, April 1954.
deregulated scenario in the presence of SSSC and CES [2] M.L.Kothari, B.L.Kaul and J.Nanda, “Automatic Generation
Control of Hydro-Thermal system”, journal of Institute of
has been used to demonstrate the method. The Engineers(India), vo1.61, pt EL2, pp85-91, Oct 1980.
performance index namely ISE has been found in the [3] Robert P. Schulte, “Automatic generation Control modification
application of different optimization techniques. for present demands on interconnected power systems”, IEEE
Out of all the methods proposed Particle Swarm Trans. On Power Systems, August 1996,pp. 1286- 1291.
[4] Jayant Kumar, Kah-Koeng and Gerald Sheble, “AGC simulator
Optimization shows better results in terms of minimum for price based operation Part1”, IEEE Transactions on Power
performance index and also the integral controller tuned Systems, vol.12,no.2, May 1997,pp. 527-532.
161
[5] Jayant Kumar, Kah-Hoeng and Gerald Sheble, “AGC simulator Authors’ information
for price based operation part- 2”, IEEE Transactions on Power
Systems, Vol.12, no. 2, May1997, pp 533-538. P. Marimuthu was born in 1977. He received
[6] Bjorn H.Bakken and OvesGrande, “Automatic generation control his B.E. in Electrical and Electronics
in a deregulated environment”, IEEE Transactions on Power Engineering from Madras University, India in
Systems, vol.13, no.4, Nov1998,pp. 1401-1406. 1998, his M.E. in Power Systems from Anna
[7] V. Donde, M. A. Pai and I. A. Hiskens, “Simulation of bilateral University, Chennai, India in 2005. He is
contracts in an AGC system after restructuring”, Power systems currently working towards his Ph.D. in the field
engineering research center, Oct 2000. of Improvement of Dynamic performance of
[8] Rao, C. Srinivasa; Nagaraju, S. Siva; Raju, P. Sangameswara, A Multi Area Hydro Thermal System at Jawaharlal
modified genetic approach to hydrothermal system with thyristor Nehru Technological University (JNTU), Hyderabad, India. He is a
controlled phase shifter under open market system, (2007) member of the faculty in the Department of Electrical and Electronics
International Review of Electrical Engineering (IREE), 2 (4), pp. Engineering, Annapoorana Engineering College, Salem, India. His
507-514. research interest includes dynamic response of hydro thermal system,
[9] Marimuthu, P., Govindaraju, C., Load frequency control of optimal location and Power flow control with FACTS devices.
hydrothermal system under open market considering capacitive
energy storage, (2012) International Review on Modelling and Dr. B. Basavaraja was born in 1970. He
Simulations (IREMOS), 5 (5), pp. 2307-2313. obtained his B.Tech (EEE) degree from
[10] P.Marimuthu, S.S.Dash, B.Basavaraja “Performance Analysis of Gulbarga University, India, M.Tech from
CES and CES-SSSC in Load Frequency Control of Multi Area Karantaka University, India and PhD from NIT
System under Open Market Scenario” International Journal of Warangal, India. He is presently working as a
Distributed Energy Resources and Smart Grids, vol. 10 Number 1 Professor & Chairman (HOD), EEE, University
(2014) Page No. 1-13. of BDT Engineering College, Davanagere,
[11] Praghnesh Bhatt, S. P. Ghoshal, Ranjit Roy, “Optimized (Visvesvaraya Technological University
Automatic Generation Control by SSSC and TCPS in Belgaum) Davanagere, Karnataka, India. He is Executive Committee
Coordination with SMES for Two-Area Hydro-hydro Power Member of ISTE and Senior IEEE member. His area of interest
System”, Proceedings of the 2009 International Conference on includes power electronics and drives, High voltage Engineering and
Advances in Computing, Control, and Telecommunication EMTP applications.
Technologies, 2009, pp 474-480.
[12] “Dynamic models for steam and hydro turbines in power system Dr. Subhransu Sekhar Dash received the M.E
studies” IEEE committee report, IEEE Trans, PAS-92, 1973, pp degree in Electrical Engineering from UCE
1904-1915. Burla, Orissa, India and Ph.D degree in
[13] Kalyanmoy Deb, “Optimization for engineering design”, Prentice Electrical Engineering from Anna University in
Hall of India, 2000. 1996 and 2006 respectively. He is presently
[14] H. Shayeghi, H. A. Shayanfar, A. Jalili and M. Sivandian, “A working as Professor and Head of the
genetic algorithm based AGC of a restructured power system”, department of Electrical Engineering in SRM
International conference on Artificial Intelligence, Jun 2006, pp University, Chennai, India. His area of interest
237-242. includes Power Quality, Converters, Multilevel Inverters, Power
[15] L. Youssef Abdel-Magid and M. A. Abido, “AGC tuning of System Operation & Control, Stability and Intelligent controlling
interconnected reheat thermal systems with particle swarm Techniques.
optimization”, Proc of IEEE Conf. Electronics, Circuits and
Systems, 2003, pp 376-379.
[16] Shankar, T., Shanmugavel, S., Karthikeyan, A., Hybrid approach
for energy optimization in wireless sensor networks using PSO,
(2013) International Journal on Communications Antenna and
Propagation (IRECAP), 3 (4), pp. 221-226.
[17] Barzegari, A., Barforoushi, T., Asgharpour, H., Lesan, S., A new
algorithm based on particle swarm optimization for solving power
economic dispatch considering valve-point effects and emission
constraints, (2011) International Review on Modelling and
Simulations (IREMOS), 4 (3), pp. 1303-1311.
[18] Sakthivel, S., Mary, D., Voltage stability limit improvement by
static VAR compensator (SVC) under line outage contingencies
through particle swarm optimization algorithm, (2011)
International Review on Modelling and Simulations (IREMOS), 4
(2), pp. 766-771.
[19] Paramasivam, B., Chidambaram, I.A., Design of a load-frequency
controller using craziness based PSO for an interconnected power
system with SSSC and RFB, (2012) International Review of
Automatic Control (IREACO), 5 (2), pp. 102-112.
[20] Ibrahim, H.E.A., Elnady, M.A., A comparative study of PID,
fuzzy, fuzzy-PID, PSO-PID, PSO-fuzzy, and PSO-fuzzy-PID
controllers for speed control of DC motor drive, (2013)
International Review of Automatic Control (IREACO), 6 (4), pp.
393-403.
162
Power Line Inspections and Dedicated Flight Envelopes
Sandra C. R. Antunes, K. Bousson, João G. Mota
Abstract – Verification of electrical lines and their infrastructure is an essential activity to

ensure the undisturbed operation of the electrical grid and supply energy to final consumers.
Performing the survey work in safety is the primary objective of the inspection teams and with this
doctoral program to create tools that help in this task. For this purpose, the development of
versatile monitoring systems to inspect high voltage overhead lines represents an invaluable
element for the maintenance of these, for corrective actions and especially preventive. Copyright
© 2015 Praise Worthy Prize S.r.l. - All rights reserved.
Keywords: Autonomous, Flight Envelopes, Overhead Power Lines, Robotic, Supervisory Control
Nomenclature The pioneering work consisted in using manned

helicopters with automated video system for tracking the
AGL Altitude above ground level distribution network poles of the power line that
DHL Distance of helicopter to the line deserved more careful inspection [8]. That experimental
DLG Distance of the Line to the ground and laboratorial work concluded on the viability of using
DLO Distance of the line to the obstacle automated aerial vehicles for power line inspection, but
GPS Global Position System exhibited limitations due to environmental conditions as
IAS Instant Air Speed well as vibrations and translational motion of the vehicle.
INS Inertial Navigation System These limitations were exploited to improve
LiDAR Light Detection and Ranging distribution pole tracking later on based on better corner
MIMO Multiple Inputs, Multiple Outputs detection and matching [3]-[4], [9].
PLMI Power Line Measurement inspection Unmanned aerial vehicles (UAVs) enable better
UAV Unmanned Aerial Vehicle accuracy with less expensive operations for overhead line
VTG Velocity relative to the ground inspection compared with manned vehicles as described
above [21]-[23]. Besides, they may deal better with
keeping the safety distance from the towers and
I. Introduction
conductors. Meanwhile, Jones and colleagues [10]
Overhead lines inspection focuses on detecting and described the requirements of a UAV for overhead line
assessing defects on overhead lines, which are structures inspection and outlined some of the difficulties
made of conductors supported on towers [1]-[4]; the unit associated with that concept, and suggested the use of
of one tower with connecting conductors (in one small, electrically-driven rotorcrafts which draw their
direction) is defined as a span and repeats across large power from the phase conductors themselves. In current
territories. Evaluating distance of the vegetation to the approach for UAV-based overhead line inspection, the
power lines for outage prevention and bushfire mitigation vehicle is equipped with gyro-stabilized cameras,
and safety purposes [5]-[7] is a key feature of inspection. navigation and position regulation, a computer for image
Common defects on conductors and towers are and other sensor data processing and a communication
mechanical damage and corrosion of steel reinforced link. The main research objectives are to design a system
aluminum strands and damaged insulators. for position and attitude control of the vehicle, to design
Such defects can be visually detected (by human a guidance system for flying over the obstacles and re-
vision) or using video or photo cameras, ultra-violet or attaching to the power line, as well as detecting power
infrared cameras, depending on the resulting physical line defects.
phenomena. Internal corrosion cannot be visually From the standpoint of position and attitude control,
detected; furthermore, it is not even detectable by infra- different estimation and control methods are presented in
red methods in some cases until the conductor is near to [7]-[9], [11] based on real-time imaging of the
failure [1], which may be too late for the most cost conductors and towers. Also, an algorithm for power line
effective action to be taken. Recent approaches enhance tracking is also proposed in [9]. Most of the methods
or complement human vision with computer vision used in the aforementioned work for position estimation
methods. Rotorcrafts, mainly helicopters, have been the and control resort to Kalman filtering [20] without
preferred systems for airborne inspection of overhead appropriate modeling of the stochastic characteristics of
lines. the system dynamics and the measurement noise since
163
S. Antunes, K. Bousson, J. Mota
the system model error and measurement noise area and the surrounding environment characteristics.
covariance matrices are just set subjectively. For that The proposed obstacle avoidance methods for
reason, accurate attitude estimation based on model overhead line inspection are far from being optimal since
uncertainty approach has been proposed by Bousson and they do not deal with minimizing the distance between
colleagues [12] for flight control of unmanned aerial the flight tracks during the avoidance process to the
vehicles. The Kalman filtering method is explained planned waypoint trajectory that was initially scheduled.
hereafter. Let a continuous-time system be described by Recently advances in optimal collision avoidance
the following state-space equations: methods and optimal 4D trajectory generation conducted
by Bousson and colleagues [16]-[18] offer interesting
 x  Ax  Bu  w  t  insight that could improve the collision avoidance task
 (1) for airborne inspection of power lines.
 y  Cx  v  t  The critical problems in airborne overhead line
inspection are related not only to acquiring accurate
where x is the state vector, u the control vector, and y video imaging for efficient defect detection on overhead
the output vector. Vectors w  t  and v  t  are lines and vegetation assessment but also, and mainly, to
flight safety.
respectively the system model uncertainty dynamics and Indeed, due to the low flight altitude and speed of the
the observation noise. Let Q  E wwT   and R  vvT   aircraft that are required for precise operations, most of
the flight will be held outside regular safety
be defined as the covariance matrices of w  t  and v  t  , configurations for aircraft operations on minimum
respectively. Then, assuming the system to be distance to ground and obstacles, altitude above ground
controllable and observable, the Kalman filter model is to instant air speed ratio, time for decision on
given as follows: contingencies, especially when hovering to assess the
The state estimate model: line conditions or when tracking the lines, span after
span. Current research activities about airborne
ˆx  Axˆ  Bu  L  y  Cxˆ  ; x  t0   ˆx0 (2) inspection of power lines have not dealt yet with these
serious flight safety problems. Safe fight operations are
where gain matrix L is defined as: achievable by supervising the aircraft maneuvers and
monitoring them within the flight envelope during the
inspection.
L  PC T R 1 (3)
Another aspect that has not got sufficient attention in
airborne power line inspection is real-time telemetry, that
where P is the solution of the following Riccati matrix
is, sending inspection data as well as flight characteristics
differential equation:
(state, parameters, navigation trajectory, ...) to a distant
station to help improve monitoring of the inspection
P  AP  PAT  Q  PC T R 1CP (4) process while the aircraft is still in the air, thus allowing
the inspection crew on the ground to interact with the
Similarly, in the case of discrete-time system given vehicle (and airborne crew, if present) when more visual
by: information of the inspected item needs to be performed.
 xk 1  uk  uk  wk Besides, corrosion detection shall be based on mixed
 (5) approach built on pattern recognition methods and
 yk  Cxk  vk physical principles [19]. The proposed project will tackle
these issues by providing a better understanding of safe
with the estimative: supervisory control and monitoring strategies for
unmanned aerial vehicle operations with respect to
ˆxk 1  ˆxk  u k  Lk  yk  Cxˆ k  ; x  t0   ˆx0 (6)
stringent flight modes within tight flight operational
envelopes during power line inspection.
The gain was given by:
1 II. Overhead Line Inspections

Lk  Pk C T R  CPk C T  (7)
The constant evolution of technology, and need for the
with Riccati solution as follow: operator to perform missions safely, suggests the use of
autonomous platforms replacing humans in some of their
1 works in order to increase safety of involved platforms,

Pk 1  Q  Pk T  Pk C T R  CPk C T  CPk T (8) protect human lives, extend time for carrying out
missions and acquire data, and reduce the risk factors
Obstacle detection and avoidance methods based on such as human fatigue when exposed to danger work.
computer vision are dealt with in [13]-[15] with some Albatroz Engineering designed a system to inspect
relative success, depending on the illumination of the overhead lines using different aerial and ground vehicles.
164
In the airborne specialized work context, it has been maneuvering,

developing innovative tools for monitoring critical Medium voltage lines are the most difficult to inspect
variables associated with flight risks in the vicinity of as the electrical grid unfolds like a tree: each line has its
overhead lines and reports them, in real time, to the roots at a substation grows along a common trunk and
inspection teams inside the helicopter. The concern with spans into branches, which also divide into more
the safe operation of manned aircraft was a cornerstone branches.
specification to Albatroz Engineering. The research and Such inspections require constant tight maneuvering
development culminated in the system called PLMI since spans are shorter than 200m and aircraft fly very
(Power Line Maintenance Inspection), in which flight close to the ground as the height of most towers is below
characteristics are modified according to safety needs, 20m, and sometimes lower than 10m.
inspection functions and are monitored in real time. The main use for distance measurement sensors is the
The overhead line inspection systems consists of a set clearance inspection that establishes the distance to the
of sensors that, together, allows for the precise location obstacle classified into one of four categories: trees,
of points of interest detected and assign them to a degree buildings, roads and ground. Fig. 1 shows the live
of severity that takes the network operator to take the interface inside a helicopter with a large window for
decision to perform maintenance actions or wait for objects detected with LiDAR and classified in real time;
another inspection to intervene conductors or towers. At there is GPS localization on the bottom left and
the base of this equipment there are: synchronous video on the top right insert window and
- Video Camera, with visual field of view of dials for mission control. Depending on the conductors
conductors and surroundings; section and distance to the line, sensors may fail to detect
- Laser LiDAR (Laser Imaging Detection and Ranging) conductors. The modeling of the surrounding of lines
allows the measurement of distances to object in front depends also on the height above ground and the velocity
of a transmitter-receiver pair; of the aircraft as it influences the spatial sampling
- GPS/INS to locate the aircraft (coupled with LiDAR, frequency. The three critical variables used to measure
it locates all objects around in world coordinates). performance are: distance to the ground, distance to the
line, and velocity to ground (left to right in bottom of
TABLE I
Fig. 1). These coincide with the three variables needed to
CRITICAL FLIGHT CHARACTERISTICS FOR DIFFERENT TYPES
OF INSPECT define safe flight envelopes. Fig. 1 shows the first
Inspection Type FLIGHT CHARACTERISTICS interface drawn in 2006 where the three variables take a
Visual Inspection Hover / Low altitude prominent role. Interviews were conducted with pilots
Thermographic Inspection Hover / Low altitude and inspectors to create heuristics to coloring indicative
Inspection Corona Hover / Low altitude / Good visibility
bars of ground clearance and line clearance, where red
Detailed Inspection and Low altitude / Low velocity
Registration means shorter distances (riskier) and green means larger
Inspection Measuring Low altitude / Low velocity distances (safety). It was found that velocity is a
distances characteristic that pilots and inspectors would like to
Environmental Inspection Low altitude / Low velocity hold constant at each instant. Reference values depend on
the line geometry; therefore it was decided not to color
Inspection systems that include distance measurement
the dial. During post-flight debriefing, the three variables
are relevant to this work because they provide the
are combined with others to analyze performance and
sensory data necessary for monitoring and control
safety and learn lessons to improve future inspections.
variables within flight envelops. Other sensors are
Several methods from literature develop control of the
omitted, as they are not involved in flight safety.
platforms, but, given the present scenario, the
The main variable to define inspection characteristics
applications of some methods may provide more benefit
is the line voltage. In the text, medium voltage is defined
than others. The project is intended to provide solutions
form 10kV to 50kV, high voltage from 50kV to 100kV
that will be integrated into future PLMI systems,
and very high-voltage above that value. The voltage of
currently in use in manned helicopters. In the doctoral
an overhead line depends on the role it plays in the
program is planned to extend the monitoring and control
electrical grid and grid topology and line geometry
of the same variables to unmanned aircraft, solution that
change accordingly. Higher voltages are associated with
with must more emphasis in security sector.
higher energy transits, implying higher currents and
conductor cables with larger sections. Very high voltage
lines are used in point-to-point connections with typical III. Autonomous Flight Tools
spans (distance between towers) of 500m or longer.
Distances from the top of the towers to the ground are To know the performance characteristics of aircraft,
40m or above, although conductors could be much closer and make detailed modeling of the standard maneuvers,
to ground. It is common to have straight line section taking into account the technical and practical knowledge
longer than 5km. High voltage lines have shorter spans, of overhead line inspections developed with manned
typically 200m to 300m, and the top of towers are 15m to aircrafts, is an advantage when to drawing flight
35m above ground. Branches occur rarely, straight line maneuvers and different monitoring missions for
sections are shorter and angles at towers require more different types of aircraft.
165
The autonomy of these helicopters daily work round

the 4 to 5 hours per pilot (by regulation), reaching an
effective average of 540km to 100km in medium voltage
lines, 100km to 200km per day for high voltage and
200km to 350km per day in very high voltage lines,
depending on environmental factors and the inspections
specifications. In many situations these helicopters are
used to perform visual inspections or visual and
thermography close-up to the towers in order to check
the details and structural integrity. In some cases, the use
of UAVs is likely to provide a better insight into the state
of the equipment, allowing an inspection of safety and
bring benefits to the cost of some types of inspection.
Fig. 1. Albatroz engineering aerial HMI Difficult access locations such as density lines too near to
residential areas, valleys with very steep slopes may be
The choice of flight deck requires knowledge about instances where the inspection with unmanned vehicles
the type of missions and the intended results. While brings wider benefits.
maximizing safety. Flight produces require: But there are other limitations that hinder the use of
- Low speed – To achieve a good spatial sampling, the unmanned aircraft in the daily inspections to overhead
aircraft speed range is in the order of 20-90km/h lines, like a structural standpoint aircraft, range, laws and
(5.5m/s to 25m/s); permits to fly within European airspace. A large group of
- Low Altitude - Data collection requires closeness to operators has been working on the technology of
the target, and that is between 15m and 120m above autonomous systems to overcome these barriers to create
ground level; a market for these and other systems that are already
- Conditioned by the weather – Usually aircraft fly known to offer advantages in replacing manned aircraft
along the lines and pilot must ensure a safety distance in some operations.
(monitored by LiDAR in real time) is kept so they
should maintain the aircraft windward, to prevent the
wind to "push" it against the line; visual constraints IV. Dedicated Flight Envelope
combined with wind directions determine the way the For dedicated flight envelope is introduced some
line is inspected. Rain or fog may require to modify magnitudes crucial to the development of the work.
or suspend inspections when visibility is reduced. These are shown in a graphical way in Fig. 2. The
Wind and rain contribute to cool metallic parts, speed over ground (VTG) is used in place of Instant Air
affecting the results of thermography. Condensation Speed [IAS], because the estimates provided by Pitot
of water in drops or droplets is detected by the tubes 30kt below (about 56km /h or 15.4 m /s) have low
LiDAR beams and is considered as an obstacle, and reliability.
neighboring objects can obscure the real line; To draw flight envelope dedicated to overhead line
affecting measurement inspections; inspections the following data are needed:
- Under Human Fatigue - Any type of inspection - Vehicle speed in flight to estimate support (lift) that
mission leads to an natural tiredness of the airborne can be approximated by VTG;
teams limitations of hours per day for this type of - Distance from the helicopter to the inspection target
work per pilot exist and depends on the aeronautical (DHL) and nearby obstacles it divided by VTG
authorities of each country; health and safety estimates the time for reaction and maneuver;
constraints usually extend also to inspectors inside - Distance from the helicopter to the ground (AGL) and
the aircraft. VTG veiled an estimate of the feasibility of
For missions requiring a low altitude and low speed autorotation;
the best platforms that can be used are rotorcrafts: - Diagram flight envelope 'altitude vs. speed' provided
helicopters or multi-rotors, because they allow to hover by manufacturer with flight safeguard curves given
around the point of interest subject to inspection and be by characteristics of the aircraft and other variables of
stationary, if required. For simpler inspections where data state.
is acquired while flying along the lines, it may be According to the inspection methodology used in
economically more advantageous to use simpler manned vehicles, several schemes are designed with the
platforms with more range, such as fixed wing aircraft. stored information on flight: Vehicle trajectories;
Each year, thousands of miles of overhead lines are anomalies detected around the lines; minimum distances
inspected. Currently, these missions rely on expensive of obstacles to the line; altitudes of the vehicle, line and
aerial platforms as manned helicopters with hazard ground, or distances from vehicle to ground. With this
factors as explained, and yet they still prove to be one of data we can determine the cross-line spacing between
the best investments regarding the ability to perform points in the same scan; and the instantaneous vehicle
workload per unit time. speed to estimate the longitudinal spacing between scans.
166
Fig. 4. Instantaneous vehicle speed during overhead

manned mission inspection
Fig. 2. Safety distances in inspection missions
The green zone ("SAFE") shows the optimal values of
In Fig. 3 is represented the different distances stored the variables; however, due to constraints of inspection it
during the flight. In aerial vehicles, the pilot use the is not always possible to operate only within this zone.
common sense and experience to make a safe flight, but, To use unmanned aerial inspection of power lines,
in this inspection case, there is no system to indicate which aim is to define specific flight procedures based on
what is or not to exceed the safety limits taking into the same variables.
account the flight envelope of the aircraft, for that As there is not our objective work in a designing of an
specific situation. This work aims to provide tools to aircraft, will just defined a payload within structural
monitor and optimize the safety of flight. limits, dynamic, and configurations and introduce them
in a existing aircraft. So, do not needs to be revised
specifications, including diagrams Altitude vs. Speed, we
want to use them not to redefine them.
Known thresholds safety variables, we can set up
qualitatively and quantitatively the distances of approach
to targets to perform inspections, maintaining a certain
speed or even performing a maneuver 'hover' (hover over
the support) allowing the pilot on the ground performing
evasive maneuvers.
Fig. 3. Altitude vs. Distance during overhead manned mission

inspection
Following the sequence of graphs below, is

represented the instantaneous speed during the flight
inspection mission, for a medium voltage line. We can
see that during the mission were achieved speeds around
25 to 60 knots (Fig. 4).
In the case of a rotary-wing aircraft, the diagram
makers provide is typically on the altitude as a function
of speed is shown in Fig. 5.
As illustrated, the red area ("AVOID") represents the
area where the pilot should avoid flying, and it is called
the 'dead man's curve'. The yellow zone ("CAUTION") is
a low risk zone in which the duration of flight should be
as short as possible. Fig. 5. Flight envelope for a sample manned rotorcraft
167
In the case of the intended application with the use of Enter Exit
unmanned aerial vehicles, the aim is to make flight
inspection respecting the safety limits taking into account Inspection System
flight procedures specific.
Although the payload applied to aircraft provided it is
within the maximum load will not be necessary to check
the structural limits of the aircraft, it should be noted is
not a goal designing an aircraft, but to verify Control
qualitatively and quantitatively how we can approach the
target to effect inspecting, maintaining a certain speed or
even effecting a ploy to 'hover' without exceeding the
limits and safety margins allowing the pilot to make Surveillance
flight maneuver not to hit the object analysis.
V. Flight Envelopes and Supervision

Supervisory Control
Identified the problem and its variables, is allowed to
draw the appropriate control solution. Fig. 6. Supervision control working scheme
The robust control allows us to have the ability to
control our system when it is subject to constant external Considering the cascade represented in the scheme are
disturbances. The interest to being able to control the following the procedures:
system after this disruption experience is the system can - Inspection system: is composed of multiple
get back to its initial state, quickly, if possible. If our instrumentation equipment whose data are processed
aerial platform is a rotary wing system itself is subject to and translated to magnitudes symbolic high level and
various perturbations means when operated on physical sent to a control module of the system;
media as well as noise from the mechanical operation. In - Control: Receive real-time magnitudes calculated by
Kalman filter method, for example, is possible to control the inspection system. The received signals are
some of this disturbances, because Kalman is processed sequentially, and can cause some changes
an algorithm that uses a series of measurements observed in the configuration of the inspection system and
over time, containing noise, and other inaccuracies, and transmit them to the next level of supervision;
produces estimates of unknown variables that tend to be - Surveillance: The symbolic quantities are stored,
more precise than those based on a single measurement processed and then sent the information from the
alone. Using Kalman filter method it is possible to processes and control devices with the aim of
preview the dynamic attitude of aircraft and control them responding to the requirements of the human
in order to avoid that get out of safe flight envelope. supervisor. Thus, this level manipulates databases,
A Supervised Control is a method used to monitor the uses graphical routines and offers communication
system permanently and constant. protocols.
With some tests with the aid of an unmanned aircraft - Supervision: Supervision manipulates symbolic
equipped with some sensors will be possible with the information programmed form declarative, and aim to
data collected, draw a preliminary flight envelope with help the operator intelligently. This serves as between
minimal safety distances must meet the helicopter during the user interface and control systems. It assesses the
the inspection missions. quality of situation and proposes actions to correct
The initial implementation of this mathematical model any anomalies mission.
will consider only two degrees of freedom, being thus Before the test technician will be required to draw a
possible to determine the safe operating area between qualitative model that allow us to identify a flight
DHG (distance from the helicopter to the ground) and envelope of securities considered state variables and
DHL (distance from the helicopter to the line or control in a general case, i.e., considering the use of
infrastructure). rotary wing air assets. Thus, as expected result, it is
The preparation and characterization of maneuvering intended to obtain a quantitative model with real data,
flight envelope is based on a method of supervised taking into account a previously qualitative model trace.
control. This is a set of tasks, including monitoring, At the moment we have notion of environment work
surveillance and supervision, combined with each other over the power lines inspections and problems associated
that the operator should consider when collecting and at this missions, can be an advantage to use other control
processing data. That is, considering a MIMO (multiple methods associated to the supervision method such as
inputs, multiple outputs) with several input variables and Kalman filters and/or a robust control method and take it
output, the user will process the input data in the possible to protect the flight envelopes for the external
inspection system, considered the interaction of control noises or other external disturbances that may affect the
parameters, monitoring and supervision, see Fig. 6. mission.
168
VI. Flight Monitoring Expected Results

Given the described scenario, it is possible to trace a
preliminary scheme with the expected flight envelops for
helicopters. In the following diagram in Fig. 5 are
outlined curves with higher and lower risk degrees.
As mentioned above, the curves which depict the
flight envelops of an aircraft depends not only on the
performance of the aircraft – There are a number of
variables which confine the missions to a certain
boundary conditions that are part of this design of our
Fig. 8. Flight envelope dedicated to overhead inspection
envelopes.
Generally, the manufacturer provides a final diagram On right diagram, translates the relationship between
for a specific configuration and conditions of flight, but velocity and the distance of the aircraft to conductor. For
to obtain the final diagram had to consider factors such very short distances there is a high risk, that grows with
as the aircraft dynamic, which limits to flight with much speed and pilot have a little time to maneuver
maximum and minimum altitude and speed; weather that safely the helicopter. However, when the distance
with some disturbances of winds, pressure, humidity, increases the risk decreases faster than the distance,
temperatures or others that allow us to decide if it is because the flight is parallel to the support and is
possible to fly or not; the kinematics of mission; proportional to velocity and avoidance maneuvers tend to
dedicated operation missions; human factors, in this case move away from the plane of the target.
if it is impossible for pilots to comply the mission, with The same problem is considered with limitations
some requirements of flight, may be is impractical imposed by sensors: if you fly too far, it is not possible to
determined aircraft for certain services or flight. make inspection in good condition. The speed increase
To perform the final diagram is an offshoot of factors will lead to a set of three thresholds to the right of the
and variables that makes us possible to get a general graph that have also been omitted for simplicity (next be
diagram of the limits to obey during the flight. included).
The diagram in Fig. 7 reflects the relationship between
altitude and velocity, in manned helicopters is one
specification of the aircraft manufacturer. The scales of VII. Conclusion
all diagrams are merely qualitative.
Several conditions can influence the performance and
the limits of flight of an aircraft. This study aims to join
the boundaries of flight envelope of the aircraft to the
limits and performance of systems used in overhead lines
to make these safer missions.
As mentioned, during missions, there are no warning
systems to indicate to the pilot and occupants the
boundaries and risks conditions of flight, for the aircraft,
when they trying to approach an ideal condition
suggested by inspection sensors, in this case for better
collection of data, only the common sense and
experience of the pilot is that it can trace a flight done
safely. Seeing in the average instantaneous velocity
performed during a normal mission in a manned flight
chart, as seen in Fig. 4, the flight was carried out at 60
knots, on average. Ideally the same flight should be done
in the order at 40 knots, but it is necessary to take into
Fig. 7. Flight envelope dedicated to overhead inspection account the aircraft altitude is and whether it is able to
make an autorotation if necessary. The minimum
On the left diagram of Fig. 8 it is reflected the distances from the helicopter to the obstacle depends on
relationship between the distance from the aircraft to the the detection capability of the LiDAR system.
ground and the aircraft to the target. Naturally, how As higher is the voltage of the lines farther them the
much higher it is the flight, the risk is reduced; and the aircraft can fly, that’s because the diameter of the lines is
further away we fly to the obstacle the risk is also lower. larger and easier to detect. To get to the plan outlined is
The same occurs with the verse of the problem, how necessary to characterize the inherent risk maneuvers in
too far from the target, when DHL increases, the inspections, and thus create methodologies that culminate
performance decreases because is not possible to observe in one way, giving the means we use to monitor, oversee
the obstacle with the set of sensors used to perform and supervise the monitoring of flights over electric
inspection mission. lines. As a final goal, we intend to elaborate envelopes of
169
flight in order to develop inspection missions with [Online]. Available: http://dx.doi.org/10.1109/CACSD-CCA-

ISIC.2006.4776719
reliable, robust, and always safely inspection teams and
[10] D. Jones, “Power line inspection – a UAV concept”, in The IEE
infrastructures that integrate aerial electrical lines. Forum on Autonomous Systems 2005 (Ref. No. 2005/11271), pp.
In general the limits of a helicopter flight envelopes 8-18, Nov. 2005.
are difficult to trace, is difficult to be detected by the [11] D. Golightly, D. Jones, “Visual control of an unmanned aerial
vehicle for power line inspection”, in Proceedings of 12th
operators, not only by the dynamic component as well
International Conference of Advanced Robotics (ICAR ’05), pp.
physical aspects relating to the integrity of the aircraft, 288-296, Jul. 2005. [Online]. Available:
and data in our case the sensors. I.e., when designing any http://dx.doi.org/10.1109/ICAR.2005.1507426
component to integrate the helicopter there are a number [12] K. Bousson, Nuno F. L. Maia, António S. Lebres, Model
of rules and requirements to take into account. Uncertainty Approach to Aircraft Angular Rate Estimation using
Kalman-Bucy Filtering, (2011) International Review of
In the next step, given the data and knowledge Aerospace Engineering (IREASE), 4 (6), pp. 307-314.
acquired, intends to proceed the development of [13] C. Liu, L. Wang, C. Liu, "Mission planning of the flying robot for
algorithms that allow drawing the envelope of flight power line inspection", Progress in Natural Science, vol. 19 issue
control in view of the supervised flight data collected 10, pp.1357-1363, Oct. 2009. [Online]. Available:
http://dx.doi.org/10.1016/j.pnsc.2009.02.006
during missions. [14] L.Wang, L. Fang, "Obstacle-navigation control of power
transmission lines inspection robot". Proceedings of the IEEE
International Conference on Robotics and Biomimetics (ROBIO
Acknowledgements 2007), pp. 706-711, Dec. 2007. [Online]. Available:
http://dx.doi.org/10.1109/ROBIO.2007.4522249
I would like to thank the mentors of Albatroz [15] M. Williams, D. I. Jones, G. K. Earp, "Obstacle avoidance during
Engineering and University of Beira Interior for the aerial inspection of power lines", EMERALD Aircraft
Engineering and Aerospace Technology, vol. 73 issue 5, pp. 472-
paper reviews; they provide valuable comments and 479, Jan. 2001. [Online]. Available:
suggestions to improve the quality of the paper. http://dx.doi.org/10.1108/00022660110403023
[16] K. Bousson, “Model Predictive Control Approach to Global Air
Collision Avoidance”, Aircraft Engineering and Aerospace
References Technology vol. 80 issue 6, pp. 605-612, Oct. 2008. [Online].
Available: http://dx.doi.org/10.1108/00022660810911545
[1] P. H. Schwabe, D. Pike, “The measurement of corrosion in [17] K. Bousson, P. Machado, “4D Flight Trajectory Optimization
overhead power lines”, Anti-Corrosion Methods and Materials, Based on Pseudospectral Methods”, Proceedings of The World
vol. 35 issue 7, pp. 4-5, Dec. 1988. [Online]. Available: Academy of Science, Engineering and Technology, vol.45, pp.
http://dx.doi.org/10.1108/eb020672 482-488, Sep. 2010.
[2] B. Avidar "Electronic airborne inspection method for overhead [18] K. Bousson, P. Machado, “4D Trajectory Generation and Control
transmission power-lines". Proceedings from ESMO-93 in 6th for Waypoint-Based Aerial Navigation”, WSEAS Transactions on
International Conference of Transmission and Distribution Systems and Control, vol. 8, issue 3, pp. 105-119, Jul. 2013.
Construction and Live Line Maintenance, pp. 89-93, Sep. 1993. [19] H. Silva, T. G. Amaral, O. P. Dias “Automatic Optical Inspection
[Online]. Available: for Detecting Defective Solders on Printed Circuit Boards”, The
http://dx.doi.org/10.1109/TDCLLM.1993.316255 36th Annual Conference of the IEEE Industrial Electronics
[3] D. I. Jones, G. K. Earp, "Camera sightline pointing requirements Society (IECON’10), Phoenix, USA, pp. 1081-1085, Nov. 2010.
for aerial inspection of overhead power lines" ELSEVIER [Online]. Available:
Electric Power Systems Research, vol. 57 issue 2, pp. 73-82, Mar. http://dx.doi.org/10.1109/IECON.2010.5675520
2001. [Online]. Available: http://dx.doi.org/10.1016/S0378- [20] P. Bauer, G. Ritzinger, A. Soumelidis, J. Bokor, “LQ Servo
7796(01)00100-6 Control design with Kalman filter for a Quadrotor UAV”,
[4] I. Golightly, D. Jones, “Corner detection and matching for visual Transportation Engineering – Periodica Polytchenica, vol.36 issue
tracking during power line inspection”, Image Vision Computing, 1-2, pp. 9-14, Dec. 2009. [Online]. Available:
vol. 21 issue 9, pp. 827-840, Sep. 2003. [Online]. Available: http://dx.doi.org/10.3311/pp.tr.2008-1-2.02
http://dx.doi.org/10.1016/S0262-8856(03)00097-0 [21] Gumerova, M., Ismagilov, F., Khairullin, I., Vavilov, V.,
[5] C. Sun, R. Jones, H. Talbot, X. Wu “Measuring the distance of Electrodynamic brakes for unmanned aerial vehicles, (2014)
vegetation from power lines using stereo vision”, ISPRS Journal International Review of Aerospace Engineering, 7 (6), pp. 202-
of Photogrammetry and Remote Sensing, vol. 60 issue 4, pp. 269- 206.
283, Jun. 2006. [Online]. Available: [22] Deif, T.N., Kassem, A.H., El Baioumi, G.M., Modeling and
http://dx.doi.org/10.1016/j.isprsjprs.2006.03.004 attitude stabilization of indoor quad rotor, (2014) International
[6] L. Zhengrong, H. Ross, W. Rodney, “A Biologically Inspired Review of Aerospace Engineering, 7 (2), pp. 55-60.
Object Spectral-Texture Descriptor and Its Application to [23] Deif, T.N., Kassem, A.H., El Baioumi, G.M., Modeling,
Vegetation Classification in Power-Line Corridors”, IEEE robustness, and attitude stabilization of indoor quad rotor using
Geoscience and Remote Sensing Letters, vol. 8 issue 4, pp. 631- fuzzy logic control, (2014) International Review of Aerospace
635, Jul. 2011. [Online]. Available: Engineering, 7 (6), pp. 197-201.
http://dx.doi.org/10.1109/LGRS.2010.2098391
[7] L. Zhengrong, T. S. Bruggemann, J. J. Ford, L. Mejias, Y. Liu,
“Toward automated power line corridor monitoring using
advanced aircraft control and multisource feature fusion”, Journal
of Field Robotics vol. 29 issue 1, pp. 4-24, Dec. 2011. [Online]. Sandra C. R. Antunes holds a MEng degree in
Available: http://dx.doi.org/10.1002/rob.20424 aeronautical engineering in the Department of
[8] C. C. Whitworth, A. W. G. Duller, D. I. Jones, G. K. Earp, Aerospace Sciences at the University of Beira
“Aerial Video Inspection of Overhead Power Lines” IEEE Power Interior, Covilhã, Portugal, in July 2009. Her
Engineering Journal, vol.15 issue 1, pp. 25-32, Feb. 2001. current research interests include optimal robust
[Online]. Available: http://dx.doi.org/10.1049/pe:20010103 control with applications to flight dynamics,
[9] D. Jones, I. Golightly, J. Roberts, K. Usher, "Modeling and avionics, nonlinear vibrations and chaotic
Control of a Robotic Power Line Inspection Vehicle", IEEE systems. She is a PhD student and researcher in
International Conference on control Applications, Oct. 2006. the Aeronautics and Astronautics Research Center (AeroG) belonging
170
at LAETA group in the University of Beira Interior, Covilhã, Portugal,

from 2008 at the current. Sandra C. R. Antunes works with structural
analysis for airborne equipments, flight kinematics observation and
control, applied to manned helicopters as well as UAV. She have
European Aviation Safety Agency (EASA) qualification as Design
Assurance System Auditor in Albatroz Engenharia, Lisbon, Portugal,
from 2010 at the current. Sandra C. R. Antunes is a member of
Business and Aviation Technical Commission (RTCA).
K. Bousson received an MEng degree in

aeronautical engineering from the Ecole
Nationale de l’Aviation Civile (ENAC) in 1988,
an MSc degree in computer science (with
emphasis on artificial intelligence) from Paul
Sabatier University in 1989, and a PhD degree
in control & computer engineering from the
Institut National des Sciences Appliquées
(INSA) in 1993, all in Toulouse, France. He was a researcher at the
LAAS Laboratory of the French National Council for Scientific
Research (CNRS) in Toulouse, from 1993 to 1995, and has been a
professor in the Department of Aerospace Sciences at the University of
Beira Interior, Covilhã, Portugal, since 1995. His current research
activities include trajectory optimization, guidance & control of
aerospace vehicles, chaotic and uncertain systems, and machine
learning. Dr. K. Bousson is a Member of the Portuguese Association
for Automatic Control (APCA) and a Senior Member of the American
Institute of Aeronautics and Astronautics (AIAA).
João G. Mota is head of research and general

manager of Albatroz Engenharia. He holds a
MEng in Electrical Engineering at the Technical
University of Lisbon (IST), Lisbon, Portugal, in
November 2001. He has a background on R&D
in mobile robotics and 3D reconstruction with
LiDAR. He worked at the R&D unit of Energias
de Portugal (EDP), the major electrical operator
in Portugal, where he realised the need for innovation on power line
maintenance inspection.
171
1827-6679(201501/02)10:1;1-K

Load Frequency Control of Multi Area

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Load Frequency Control of Multi Area

Uploaded by

Copyright:

Available Formats

International Review of

Optimum Design of a Surface Mounted Fractional Slots Permanent Magnet Motor 28

Phase Drift Phenomenon in Dual Active Bridge Converter –

Vesa Väisänen, Jani Hiltunen, Raimo Juntunen

Copyright © 2015 Praise Worthy Prize S.r.l. - All rights reserved

low passive component count, soft-switching capabilities Vin2 Vout   

Vin = 100 V, Vout = 250 V

Vin = 130 V, Vout = 250 V

Vin = 130 V, Vout = 210 V

At time instant t1 transistor PA+ is turned off. The

III.1. Charge Based Compensation

Q pri VDS  Qsec VDS 

The phase drift can be compensated by adding drift to

III.2. Controller Based Compensation

Input power [W]

needs to be known. This paper presents two phase drift http://dx.doi.org/10.1109/TPEL.2013.2271511.

Raimo Juntunen received the M.Sc. degree in

Replacing Copper with New Carbon Nanomaterials

J. Pyrhönen, J. Montonen, P. Lindh, J. J. Vauterin, Marcin J. Otto

I. Introduction permanent magnet materials which provide low-loss

Copyright © 2015 Praise Worthy Prize S.r.l. - All rights reserved

V. The Design and Construction of a

V.1. The Test Machine

V.2. No Load and Load Simulations 5

A finite element analysis (FEA) was performed with 4

Induced voltage [V]

harmonic spectrum. The peak value of the induced -2

Induced voltage [V]

0 50 100 150 200 250 300 350 2

Fig. 6. Phase voltage under generating to 30 W resistive load

0 The total loss of the CNT-yarn machine under rated

Rated output power, W 30

Fig. 7. Phase current under generating to 30 W resistive load

V.3. Machine Design Summary and

The results indicate that the Teijin Aramid-

Voltage [V], Current [A]

0 0.5 1 1.5 2 2.5 3 3.5 4

Fig. 13. Voltage of the motor at the speed 15000 min-1

resistors the machine being at room temperature.

V.4.3. No-load Measurement as Motor -1

Fig. 14. Current of a motor phase at the speed 15000 min-1

[7] G. -A. Capolino, A. Cavagnino, “New Trends in Electrical Authors’ information

Computation of Electromagnetic Losses in Double-Rotor Vernier

Huijuan Liu1, Yue Hao1, Shuangxia Niu2, Weinong Fu2

Nomenclature Kadd Coefficient of additional loss

Copyright © 2015 Praise Worthy Prize S.r.l. - All rights reserved

III. Constructions and Operation

TABLE I The computed core loss curves versus time at different

140 Figs. 4. Core-loss distribution at full-load operation

V. Conclusion May 2005, pp. 10E311(1-3).

Huijuan Liu received Ph. D. in Electrical

Weinong Fu received the Ph.D. in electrical

Optimum Design of a Surface Mounted

A. Mansouri, H. Msaddek, H. Trabelsi

Keywords: Fractional Slots SMPM, Analytical Modeling, Optimization, Efficiency, Optimal

Nomenclature kc Carter factor

Copyright © 2015 Praise Worthy Prize S.r.l. - All rights reserved

I. Introduction In the presented work, the genetic algorithm is used to

Inner stator slot width, bss1:

Outer stator slot width, bss2:

Stator slot pitch, s:

Ratio of the slot opening over the slot width, kopen:

hry in Eq. (8) defines the rotor yoke height.

s Acu is the copper area: