Professional Documents
Culture Documents
Corresponding: bouananeabd@yahoo.fr
ABSTRACT
The unified power flow controller (UPFC) is a FACTS device, which can be used to control the power flow
on a transmission line. This is achieved by regulating the controllable parameters of the system: the line
impedance, the magnitude and phase of the bus voltage. In addition to control of real and reactive power
flow, the UPFC can be employed to enhance power system damping by modulating the converter voltages.
The aim of this article is to design a regulator which enables a power system to track reference signals
precisely and to be robust in the presence of uncertainty of system parameters and disturbances. The
performances of the proposed controllers (NEWELM and NIMC) are based neural adaptive control and
simulated on a two-bus test system and compared with a conventional PI controller with decoupling (PI-D).
The studies are performed based on well-known software package MATLAB/Simulink tool box.
Keyword: FACTS, UPFC, PI-D, NEWELM, NIMC, Neural Adaptive Control, Stability, Robustness.
1
Improving power quality in power grids with a UPFC System for compensation using neural adaptive
controllers (BOUANANE Abdelkrim)
2. MODELING OF UPFC SYSTEM 3.1. PI decoupling control (PI-D):
The equivalent circuit of a UPFC system where the The objective of using the is to provide
series and shunt inverters are represented by voltage independent control of active P and reactive Q
sources vc and vp respectively. The transmission power flow in the system for fixed values of V s and
line is modeled [1], [3], [19] and [28] as a series Vr. This can be achieved by properly controlling the
combination of resistance r and inductance L. The series injected voltage of the UPFC. The voltage,
parameters rp and Lp represent the shunt transformer current and power flow are related through the
resistance and leakage inductance respectively. The following equations:
non-linearity’s caused by the switching of the The principle of this control strategy is to convert
semiconductor devices, transformer saturation and the measured three phase currents and voltages into
controller time delays are neglected in the equivalent d-q values and then to calculate the current
circuit and it is assumed that the transmission system references and measured voltages as follow:
is symmetrical. The simplified circuit of the UPFC = � + � (4)
control and compensation system is shown in Figure
2. = � − � (5)
With
� = � +� and � =� +�
∗� −� ∗ �
�∗ = (6)
∆
∗� −� ∗ �
�∗ = (7)
∆
Figure 2.
Equivalent circuit of UPFC system.
With: ∆= + (8)
The modeling of this circuit is based on hypotheses
simplifying. The dynamic equations of the UPFC are According to the system of equations (1) or (2), one
divided into three systems of equations. By can have the system contains a coupling between the
performing d-q transformation, the current through reactive and active current Id Iq. The interaction
the transmission line can be described by the between current caused by the coupling term (ω)
following equations. Fig. 3.
s
= ωi - i + v -v -v
System
model
L L
{ s
(1) idref
-
+ + id
= -ωi - i + v -v -v
1
PI
+ - + s+r/L
L L
Similarly, the shunt inverter can be described by:
= ωi i + v -v -v
LP L
{ (2) iqref +
PI
+
- 1 iq
= -ωi i + v -v -v - + + s+r/L
LP L
By the use of power balance if one neglects the Figure 3.
losses of the inverter it is possible to express the Control design with PI corrector.
continuous voltage by:
V
= V i + V i -V i -V i (3) To be able to lead to a reliable command of the
CV
The DC-link capacitor C must be selected to be large system, it is indispensable to proceed to a
enough to minimize voltage transients decoupling of the two components. The decoupling
of two loops is obtained by subtracting the term ( )
through a reaction against. It is then conducted to a
3. CONTROLLER DESIGN rule which provides a command with decoupling
The control system of the UPFC consists of the (PI-D) of the currents Id and Iq with a model which
shunt inverter with the control circuit [9], [11] and can be rewritten in the following form:
[20], as well as the series inverter. First, we justify �
= −� � − � + − −
the possibility of separation of the two control { � �
(9)
�
circuits and similarly we are interested in the = −� � − � + − −
� �
adjustment of the inverter for the additional voltage
and more particularly to the setting of the active and The design of the control system must begin with
reactive power transmitted.Then we will develop the the selection of variables to adjust and then that of
different settings considered in this study and we the control variables and their association with
will show the transient behavior of the control variables set. There are various adjustment
circuits using a simulation of the regulators techniques well suited to the PI controller.
considered in the adjustment of the closed loop
UPFC system in order to improve the performances The structure of the PI controller is represented by
in the case of active or reactive power change. the first block diagram of Figure 4.
(change one of the three parameters of the line).
2
Improving power quality in power grids with a UPFC System for compensation using neural adaptive
controllers (BOUANANE Abdelkrim)
3.1.2. Simulation results (system UPFC
with PI decoupling control):
Currents Isa,Isb and Isc (A)
2
1.5
0.5
Figure 4. -0.5
Figure 5.
frequency N : Currents waveforms Isa, Isb and Isc (A).
k p 2 N a
(10)
k i N2
There are two well-known empirical approaches
proposed by Ziegler and Tit for determining the
optimal parameters of the PI controller Table 1.
Table 1.
OPTIMAL PARAMETERS OF THE PI CONTROLLER.
Type of Kp Ti Kd
regulator
PI 0.45Kcr 0.83 Pcr 0
3
Improving power quality in power grids with a UPFC System for compensation using neural adaptive
controllers (BOUANANE Abdelkrim)
feedback, is justified by its role, this network can be
interpreted as a state space model nonlinear.
Learning by back propagation algorithm standard is
the law used for identification of the UPFC.
Input layer Hidden layer Output layer
Figure 9.
Structure of the Elman network
Figure 8.
System response in the presence of external disturbance 3.2.1. Neural Adaptive Control by State
Space UPFC System with ERNN
Figure 8 shows when introducing a disturbance of
(NACSSS-ERNN):
duration 25 ms and amplitude 1.5 in the simulation
to further test the stability of our system so we can
The integration of these two approaches (neural
see that the controller (PI-D) rejected the external
adaptive control) in a single hybrid structure, that
disturbance fairly quickly.
each benefit from the other, but to change the
3.2. Neural adaptive control dynamic behavior of the UPFC system was added
against a reaction calculated from the state vector
Currently the adaptive control is of a great (state space) Fig. 7.
importance in the field of control. This command is
dominant in systems that present uncertainties, +
E (t) U (t) UPFC System Y (t)
structural disturbances and changes in the
environment. The main object of the adaptive
control is the synthesis of the act of adaptation, for
K X (t)
the automatic adjustment in real time of the
Figure 10.
regulators of the control loops in order to achieve or
Diagram of a UPFC control by state feedback
maintain certain level of performance when the
parameters of the process to order are difficult to The state feedback control is to consider the
determine or vary with time. The interest of the process model in the form of an equation of state:
adaptive control appears mainly at the level of ∗
= + (11)
disturbance parametric, that is to say are acting on And observation equation:
the characteristics of the process to order, = + (12)
disturbance, act on the variables to regulate or to
order. The interest of the adaptive control appears Where u (t) is the control vector, x (t) the state
mainly at the level of disturbance parametric, that is
vector, and y (t) the output vector of dimension for a
to say are acting on the characteristics of the process
discrete system to the sampling process parameters
to order, disturbance, act on the variables to regulate Te at times of Te sample k are formalized as follows:
or to order. Finally, the combination of the adaptive
control with other types of conventional commands
X + = + (13)
of the automatic has borne fruit and has been the
= + (14)
source of many jobs. The adaptive laws implanted in
The system is of order 1, so it requires a single
the ideal case could lead to instability in the case of
state variable x. this state variable represents the
external disturbances bounded.
output of an integrator as shown in Fig.
In this article we present the Adjustment method
x(t) = y(t) x* (t) = y*(t)
proposed for the UPFC, favoring the classical
approach based on neurons networks. Neural
Network is trained by adaptive learning [3], [22] and
U + y*=x* Y=x
[27], the network ‘learns’ how to do tasks, perform
functions based on the data given for training. The
knowledge learned during training is stored in the
synaptic weights. The standard Neural Network
structures (feed forward and recurrent) are both used
to model the UPFC system. The main task of this
paper is to design a neural network controller which
keeps the UPFC system stabilized. Elman's network Figure 11.
Block diagram of the state representation of the UPFC
[26] said hidden layer network is a recurrent
network, thus better suited for modeling dynamic The transfer function G s = Y s ⁄U s of our
systems. His choice in the neural control by state
4
Improving power quality in power grids with a UPFC System for compensation using neural adaptive
controllers (BOUANANE Abdelkrim)
process UPFC can be written as: Where, Wh ; Wr et Wo are the weight matrices.
G s = (15) Equations are standard descriptions of the state
+ ⁄� space of dynamical systems. The order of the system
depends on the number of states equals the number
We deduce the equations of state representation of of hidden layers. When an input-output data is
the UPFC: presented to the network at iteration k squared error
∗
=− + at the output of the network is defined as:
{ � (16)
= =( − ) (22)
With: U t =e t −kx t For the whole training data u (t), yd (t) de t = 1, 2…
Let: N, the summed squared errors are:
X t+ =[ − ]x t + e t (17) = ∑�= (23)
= (18) The weights are modified at each time step for W 0:
�� �
The dynamics of the process corrected by state =( − )
� �
space is presented based on the characteristic �� �
equation of the matrix [Ad - Bd K], where K is the =( − ). (24)
�
matrix state space-controlled process. Our system is For Wh et Wr,
described in matrix form in the state space:
Where: =− . .
∗
= + ℎ ℎ
{ (19)
= + ��
= −( − ). �
. (25)
� ℎ
− − �
�
=− . . �
=[ − ], =[ ], =[ ], =[ ] �
−
− �� � � �
� = −( − ). . (26)
� � � �
=[ ], = [� ] , = [� � ]�
The latter we obtain:
� � � �� � −
3.2.1.1 UPFC SYSTEM IDENTIFICATION � = − + (27)
� � �
USING ERNN (NEWELM): Equation shows that there is a dynamic trace
The identification makes it possible to obtain a of the gradient. This is similar to back propagation
mathematical model that represents as faithfully as through time. Because the general expression for
possible the dynamic behavior of the [5] and [17]. A weight modification in the gradient descent method
��
process identified will then be characterized by the is: � = −�
�
structure of the model, of its order and by the values (28)
of the Settings. It is therefore, a corollary of the The dynamic back propagation algorithm used to
process simulation for which one uses a model and a identify a state space model of the UPFC for
set of coefficients in order to predict the response of NEWELM (Elman network) can be summarized as
the system. The figure shows the network Elman follows:
�
consisting of three layers: a layer of entry, hidden � = −� (29)
layer and a layer of output. The layers of entry and � =� . �
(30)
ℎ
exit interfere with the outside environment, which is
� � � �
not the case for the intermediate layer called hidden � =� . (31)
� �
layer. In this diagram, the entry of the network is the
command U(t) and its output is Y (t). The vector of If the dependence of X (t-1) on Wi is ignored, the
state X(t) from the hidden layer is injected into the above algorithm degretes is the standard back
input layer.
propagation algorithm:
Input layer � � �
Hidden layer Output layer
� = = � − (32)
U(t) �
Wr
W0 y(t) Network with
x(t) ERNN-
Wh
NEWELM
Figure 12.
Structure of the NEWELM
v (t) = id
e (t) = i*sd
UPFC
The state vector X (t) from the hidden layer is With
u (t) = vcd
injected into the input layer. We deduce the SSNN
following equations: K
5
Improving power quality in power grids with a UPFC System for compensation using neural adaptive
controllers (BOUANANE Abdelkrim)
network weights for the controller to start
3.2.1.2. Estimation of the parameters: with.
The errors signal w (t) rebuilt the disturbance D (t) The model M is determined by the Widrow-Hoff
and the error of characterization following the learning law [25]. The training of the network
command u (t): consists in modifying, with each step, the weights
= �� . + (35) and bias in order to minimize the quadratic errors at
output by using the law of Windrow-Hoff. With
So, we can say that the stability of the whole when each step of training, the error at output is calculated
̂ are stable
R(s) and � as the difference between the required target t and
the output y of the network. The quantity to be
For an ideal model ̂
� =� (36) minimized, with each step of training k, is the
variance of the error at the output of the network.
and
� � � �
̂
=� −
a perfect pursuit of trajectory � = � � = � � + � � − � � (38)
= , ∀ d (t) Te: The sampling time is equal to 1ms. This time will
An online estimate of unmeasured disturbances be maintained throughout this study.
̃ = (37) The estimation of output and the error are calculated
by:
3.2.2.2. Neural Inverse Model Control
� = [− �− � �− ][ ]+� ;
Design:
+ −
= − − (40)
+
Where:
̂ = + − − − (41)
with =[ ]
and
�=[ − − ] (42)
(c)
Time (s)
Evolution of the Weights NACIMS
Figure 19.
Powers (P and Q) of the control system with
8
Improving power quality in power grids with a UPFC System for compensation using neural adaptive
controllers (BOUANANE Abdelkrim)
4. COMPARATIVE SIMULATION STUDIES [2] B.A. Renz, A. Keri, A.S. Mehraben, C. Schauder, E.
BETWEEN NACSS-NEWELM AND Stacey, I. Kovalsky, L. Gyugyi, and A. Edris, 1999,
NACIMS: “AEP Unified Power Flow Controller Performance”,
IEEE Trans. on Power Delivery, 14(4), pp. 1374-
1381.
Adaptive Neural Feedback Control (ANFC) [3] Bouanane A., Chaker A. and Amara M. “adaptive
is a hybrid control that has allowed them to control control with SSNN of UPFC system for
any variation in tracking, regulation or stability. The compensation of active and reactive power”. Res J.
results of the simulation showed the strength of our Appl. Sci. Eng. Technol. Vol 6(4):739-747,2013.
neural adaptive controller (NAC). [4] Nicolas, Boi K. Nguyen “Impact of FACTS devices
on voltage and transient stability of a power system
We can say that the decoupled PI regulator would including long transmission lines’’, IEEE
be ideal for the UPFC system control if the ± 30% Transactions on Power Engineering Society. Vol. 3,
variation of the reactance did not degrade its pp. 1906 - 1911,2000.
dynamic performance, as pointed out in this article. [5] Delgado,A. and al (1995) “dynamic recurrent neural
networks for system identification and control” IEE
The process model is never perfect.
Proc.Control Theory Applications, Vol
The results of this analysis of the two controls by 142,n°4,pp.307-314.
neural networks at ± 30% of XL fig.19 are [6] Gyungui. L, FIEE. (2008) ‘Unified Power Flow
summarized in the following points: Control concept for Flexible AC Transmission
System’ IEE proceedings- C, Vol.139, No.4
[7] H. Johal and D. Divan, “Design considerations for
series –connected distributed FACTS converters”,
All control strategies indicate that the IEEE Transactions on industry applications, Vol.43,
proposed regulators have better dynamic PP: 1609-1618, Nov/Dec. 2007.
performance and are much more robust [8] Henriques, J. and A. Dourado, 1999. A hybrid
than the traditional PI controller. They neural- decoupling pole placement controller and its
seem to be very high-performance application. Proceeding of 5th European Control
dynamic regulators. Conference (ECC99). Karlsruhe, Germany.
[9] Hideaki Fujita, Yasuhiro Watanabe, Hirofumi
Thanks to the generalization capacity of
Akagi, “Control and Analysis of a Unified Power
network neurons, the use of a neural Flow Controller”, IEEE TRANSACTIONS ON
identification regulator allows an POWER ELECTRONICS, VOL. 14, NO. 6
improvement in the dynamic November 1999, pp.1021-1027.
performance of the regulator [10] J. Bian, D.G. Ramey, R.J. Nelson, and A. Edris,
approached. It has even been 1997, “A Study of Equipment Sizes and Constraints
demonstrated by simulation that a neural for a Unified Power Flow Controller”, IEEE Trans.
identification regulator can solve the on Power Delivery, 12(3), pp. 1385-1391.
problem of the incapacity of parametric [11] Kannan. S, Shesha Jayaram and M.M.A.Salama.
(2007) ‘Real and Reactive Power Coordination for a
variations regulator PI of the line.
Unified Power Flow Controller’ IEEE Transactions
on Power Systems, 2007, vol.19. No.3, pp. 1454 –
5. CONCLUSION 1461.
[12] K.K. Sen, 1998, “SSSC – Static Synchronous Series
In this article, our UPFC system based on Compensator: Theory, Modeling and Application”,
three robust control methods has been proposed [(PI- IEEE Trans. on Power Delivery, 13(1), pp. 241-246.
D), (NEWELM and NIMC) based neural adaptive [13] L. Gyugyi, C.D. Schauder, S.I. Williams, T.R.
control]. These control strategies introduce enough Reitman, D.R. Torgerson, and A. Edris, 1995, “The
flexibility to set the desired level of stability and Unified Power Flow Controller: A new approach to
power transmission control”, IEEE Trans. on Power
performance. Practical constraints were considered
Delivery, 10(2), pp. 1085-1097.
by introducing appropriate uncertainties. The [14] L. Gyugyi, 1994, “Dynamic Compensation of AC
methods above have been applied to a typical test of Transmission Line by Solid State Synchronous
a single-phase power system bridge. Simulation Voltage Sources,” IEEE Transactions on Power
results showed that designed regulators were able to Delivery, 9(22), pp. 904-911.
ensure a robust stability and performance a wide [15] M. I. Marei, "A unified control strategy based on
range of uncertainty of parameters of transmission. phase angle estimation for matrix converter interface
But line orders two hybrid (NACSSS-ERNN and system," IEEE Systems Journal, vol. 6, no. 2, pp.
NACIMS) in case of modification of the parameters 278-286, 2012.
[16] Narain, G.H. and G. Laszlo, 1999. Understanding
of the system and an excellent ability to improve the
FACTS: Concepts and Technology of Flexible AC
stability of the system under small disturbances. Trnasmission Systems. Wiley-IEEE Press
Marketing, pp: 297-352; 407-424.
[17] Narendra K.,Parthasarathy K.(1990) “identification
REFERENCES and control of dynamical systems using neural
networks” ,IEEE Trans. Neural Networks ,Vol 1
,pp4-27 .
[18] Nguyen, D. and B. Widrow, 1990. Neural networks
[1] Alf Isaksson” Model based control design Supplied for self-learning control systems. IEEE Control Syst.
as supplement to course book inAutomatic Control Mag., 10(3): 18-23.
Basic” course (Reglerteknik AK) September, 1999.