You are on page 1of 13

2246 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 28, NO.

5, MAY 2013

Distribution Voltage Control for DC Microgrids


Using Fuzzy Control and Gain-Scheduling Technique
Hiroaki Kakigano, Member, IEEE, Yushi Miura, Member, IEEE, and Toshifumi Ise, Member, IEEE

Abstract—Installation of many distributed generations (DGs) instance, dc power supplies are commonly used in telecommu-
could be detrimental to the power quality of utility grids. Micro- nication buildings and internet data centers where high-quality
grids facilitate effortless installation of DGs in conventional power power is needed [13].
systems. In recent years, dc microgrids have gained popularity be-
cause dc output sources such as photovoltaic systems, fuel cells, and We previously proposed a “low-voltage bipolar-type dc mi-
batteries can be interconnected without ac/dc conversion, which crogrid,” and described the configuration, operation, and con-
contributes to total system efficiency. Moreover, high-quality power trol scheme, which were demonstrated through experimental
can be supplied continuously when voltage sags or blackouts occur results [14], [15]. In the experiment, we used one energy stor-
in utility grids. We had already proposed a “low-voltage bipolar- age unit with a dc/dc converter to sustain dc-bus voltage when
type dc microgrid” and described its configuration, operation, and
control scheme, through experiments. In the experiments, we used the system was in intentional islanding operation. However, it is
one energy storage unit with a dc/dc converter to maintain the desired that the system should have two or more energy storage
dc-bus voltage under intentional islanding operation. However, dc units for redundancy. Several types of droop controls have been
microgrids should have two or more energy storage units for system proposed for the parallel operations among, which some were
redundancy. Therefore, we modified the system by adding another applied for ac/dc microgrids [16]–[21]. Therefore, we added an-
energy storage unit to our experimental system. Several kinds of
droop controls have been proposed for parallel operations, some other energy storage unit in the experimental system and adopted
of which were applied for ac or dc microgrids. If a gain-scheduling gain-scheduling control technique as the droop controls for shar-
control scheme is adopted to share the storage unit outputs, the ing the outputs. In this case, although output power sharing of
storage energy would become unbalanced. This paper therefore two energy storage units was achieved, it was found that the
presents a new voltage control that combines fuzzy control with storage energy control was necessary for the system operation
gain-scheduling techniques to accomplish both power sharing and
energy management. The experimental results show that the dc proposed in [14], [15].
distribution voltages were within 340 V ± 5%, and the ratios of In this paper, we propose a new voltage control method that
the stored energy were approximately equal, which implies that combines fuzzy control with gain-scheduling control technique
dc voltage regulation and stored energy balancing control can be in order to accomplish both power sharing and energy manage-
realized simultaneously. ment simultaneously. Some researches proposed to apply fuzzy
Index Terms—DC power systems, fuzzy control, gain-scheduling controls to microgrid operations. A fuzzy-based controller was
control, microgrids. applied to all DGs in the microgrids, and the system perfor-
mance was analyzed by numerical simulations [22]. A fuzzy
I. INTRODUCTION
control in addition to a PI controller was used to stabilize the
NSTALLATION of distributed generations (DGs) has re-
I markably increased in recent years because of the reduction
in greenhouse gases and depletion of fossil fuels. After the
rate of a diesel generation [23]. A fuzzy control was applied to
maximum power point tracking control of a photovoltaic sys-
tem in an isolated microgrid to improve the response against
2011, Tohoku earthquake and tsunami, the trends have accel- rapidly changing weather conditions [24]. In this paper, we fo-
erated even more in Japan. As it is well known, installation of cus on application of fuzzy control to gain-scheduling control
many DGs could be detrimental to the power quality of util- and demonstrate the validity of the proposed control through
ity grids; therefore, microgrids have been studied as one of the experiments.
candidates to support smooth installation of DGs [1]–[6]. In In our experimental system, electric-double-layer capacitors
particular, dc distribution microgrids have been studied for a (EDLC) were used as the energy storage unit, and each dc/dc
few years [7]–[12]. Apart from a reduction in ac/dc conver- converter of EDLC controlled the dc distribution voltage when
sions losses, dc microgrids can supply continuous high-quality the system was operated under an intentional islanding mode.
power when voltage sags or blackouts occur in utility grids. For The experimental results show that the dc distribution voltages
were within 340 V ± 5%, and the energy ratios of the stor-
age units were approximately equal, implying that dc voltage
Manuscript received April 28, 2012; revised July 4, 2012; accepted August regulation and stored energy balancing control can be realized
27, 2012. Date of current version November 22, 2012. Recommended for pub- simultaneously by applying the proposed control.
lication by Associate Editor L. Chang.
The authors are with the Division of Electrical, Electronic, and In-
formation Engineering, Osaka University, Osaka 565-0871, Japan (e-mail: II. DC MICROGRID FOR A RESIDENTIAL COMPLEX
kakigano@eei.eng.osaka-u.ac.jp; miura@eei.eng.osaka-u.ac.jp; ise@eei.eng.
osaka-u.ac.jp). A. System Configuration
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org. We proposed a dc microgrid for a residential complex, as
Digital Object Identifier 10.1109/TPEL.2012.2217353 shown in Fig. 1 [15]. There are around 50–100 houses in the
0885-8993/$31.00 © 2012 IEEE
KAKIGANO et al.: DISTRIBUTION VOLTAGE CONTROL FOR DC MICROGRIDS USING FUZZY CONTROL AND GAIN-SCHEDULING TECHNIQUE 2247

Fig. 2. Interconnected operation.


Fig. 1. System configuration of the dc microgrid for a residential complex.

system, each having a micro combined heat and power unit


(micro-CHP unit) such as a gas engine or a fuel cell. The micro-
CHP units are connected to a dc distribution line (three wire,
±170 V), and the electric power is shared among the houses.
Cogenerated hot water is either used by individual house or
shared between adjacent houses. The utility grid is connected
to the system by a rectifier. At the load side, various forms of
electric power (such as ac 100 V and dc 48 V) can be obtained
from the converters. EDLCs are used as the main energy storage
unit because of their advantages such as fast response, safety
(especially compared with Li-ion batteries), easy measurement
of the stored energy, and no toxicity of the constituent materials.
In a verification test of energy storage system using an EDLC
unit, the voltage, and maximum energy were 500 V and 5 MJ,
respectively [25]. From this, EDLC is considered viable as an
Fig. 3. Intentional islanding operation.
energy storage system in a small grid. The capital cost per
kilowatthour (kWh) for EDLC is 300–2000 dollars, while that
for lead-acid and Li-ion batteries is 200–400 and 600–2500
micro-CHP units such that the power from the utility grid is
dollars, respectively [26]. In addition, EDLCs can handle many
within the contract demand. In addition, the supervisor computer
charge–discharge cycles, and therefore has a low-cost per cycle
decides the order of the operation of the micro-CHP units so as
considering its long life. The capital cost per kWh-per cycle
to meet the heat commitment.
for EDLC is 2–4 cents, while that for lead-acid and Li-ion
When the system is disconnected from the utility grid, the sur-
batteries is 20–100 and 15–100 cents, respectively [26]. The
plus, or deficient power is compensated by the EDLC as shown
disadvantage of EDLC is its low-energy density. If a large energy
in Fig. 3. In intentional islanding operation, the EDLC converter
capacity is needed for a microgrid, a relatively large EDLC
controls the dc distribution voltage, and the number of the op-
is required. However, a large-energy capacity is not necessary
erating micro-CHP units is determined by not only the load
for the proposed dc microgrid because the micro-CHP units
consumption, but also the stored energy of the EDLCs. When
are operated to prevent over charge/discharge of the EDLCs as
the stored energy exceeds a maximum limit, the system stops
described in the following section.
one of the operating micro-CHP units. Then, the total output
of the micro-CHP units becomes less than the load consump-
B. System Operation
tion, and the EDLC discharges until the stored energy becomes
The total output power of micro-CHP units can be controlled less than the minimum limit. On the contrary, when the stored
by changing the number of running micro-CHP units. When energy falls under the minimum limit, the system starts a micro-
the system is connected to the utility grid, any deficiency in the CHP unit. Then, the total output of micro-CHP units becomes
power supplied by the microgrid is compensated by the power more than the load consumption, and the EDLC charges until
from the utility grid as shown in Fig. 2. In the interconnected the stored energy exceeds the maximum limit. These two modes
operation, the rectifier controls the dc distribution voltage, and are repeated alternately in intentional islanding operation. The
the supervisor computer changes the number of the running heat commitment is not fundamentally considered in intentional
2248 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 28, NO. 5, MAY 2013

Fig. 4. Configuration of the experimental system in the laboratory.

islanding operation because continuous electricity supply is a


priority during the operation. Fig. 5. Control for dc/dc converter for energy storage.
The energy storage unit does not need a large capacity, if this
system chooses the suitable micro-CHP units that can start up
in a few minutes. For instance, one of the commercial micro
gas engine cogeneration (GEC) systems can start up in about
5 min. Most of the heat in the starting-up period is utilized in
heating the catalyst for the exhaust gas. Therefore, it is feasible
that the GEC system will be able to start in a few minutes every
time, after improving the exhauster for quick start up. In this
case, we assumed the energy of the storage system to be less
than 10 kWh and the charging/discharging time to be less than
1/2 h, although the parameters depend on the specification of
each system in practice. Therefore, EDLC can be a candidate of Fig. 6. Droop control feature.
the main energy storage unit in this system [27].
Fig. 4 shows a configuration of the experimental system used
in our laboratory. It is assumed that there are three households,
and each house has a dc power source such as a fuel cell. The more energy storage units and those converters can be operated
feature of the dc system is to adopt a three-wire dc distribution, in parallel, it contributes to the voltage regulation and system
which consists of a +170-V line, a neutral line, and a −170-V redundancy. Droop control is a well-known method for voltage
line. In a three-wire composition, the voltage relative to ground control when two or more converters are used. In general, the
becomes low compared with that in two-wire system, where droop controller detects the output power or current as a feed-
one of the wires is grounded, and when half bridge inverters are back parameter, and the deviation of dc voltage is controlled
adopted for it, one of the single-phase 100-V output lines can in proportion to the output power. However, if the converters
also become a grounded neutral line as long as abiding to the connect to energy storage units, the controller should consider
Japanese standard [15]. In addition, load side dc/dc converters not only the output power balance, but also the stored energy.
can choose the source voltage from +340, +170, or −170 V. A In particular, stored energy balance is important to carry out
rectifier connected to the utility grid is controlled to maintain the the operation described in the previous section because micro-
dc voltage constant (340 V) in the interconnected operation. To CHP units have to start or stop frequently under an unbalanced
balance the positive voltage (+170 V) with a negative voltage condition of the stored energy. Therefore, we propose a novel
(−170 V), a voltage balancer is connected at the dc side of control method combining gain scheduling and fuzzy control,
the rectifier. Two EDLC banks are connected through a dc/dc which accomplishes good voltage regulation, load sharing, and
converter as energy storage. energy balance simultaneously.

C. DC Voltage Control
As mentioned in the previous section, the dc distribution volt- III. CONTROL STRATEGY OF CONVERTER FOR ENERGY
STORAGE UNIT
age is normally controlled by a grid connected rectifier in the
interconnected operation. The dc/dc converters of the storage Fig. 5 shows the circuit and proposed control diagram of
systems are controlled to maintain the dc distribution voltage a dc/dc converter for EDLC banks. We designed the circuit
within a specified range [15]. In intentional islanding operation, to be symmetric with respect to the neutral line, because the
the dc/dc converters of the storage systems need to maintain the converter is supposed to function as a voltage balancer under
dc distribution voltage. Therefore, if the dc microgrid has two or the appropriate control.
KAKIGANO et al.: DISTRIBUTION VOLTAGE CONTROL FOR DC MICROGRIDS USING FUZZY CONTROL AND GAIN-SCHEDULING TECHNIQUE 2249

Fig. 7. Circuit and control diagrams to obtain the relation between the steady-
state error and the gain K c .

TABLE I
PARAMETERS

A. Gain-Scheduling Control for Output Power Sharing


Fig. 8. Gain-dc/dc converter output power characteristics. (Voltage variation
When dc voltage is controlled by several converters, it is dif- 2%).
ficult to achieve good voltage regulation and sharing at the same
time. As illustrated in Fig. 6, better voltage regulation requires
a higher dc gain, but it can decrease load sharing. Therefore,
we adopt a gain-scheduling control that changes the gain Kc
according to the output power in order to obtain better voltage
regulation and load sharing simultaneously [16].
We examined the steady-state dc voltage error when the gain
Kc or the output power was changed by numerical simulations.
Fig. 7 and Table I show the circuit, control diagram, and param-
eters for the simulation. The rated output voltage and capacity
were 340 V and 3 kW, respectively. Fig. 8 shows the results,
which indicated that the steady-state error of the dc voltage be-
comes larger when the load is heavier or the gain Kc is smaller.
When the variation in the dc voltage is permitted within 2%, we
can obtain the relation between the gain Kc and output power
(p.u.) as shown in Fig. 8. The gain Kc can be expressed by a
linear function of the output power. Similarly, we can obtain
the relation between the gain and input power when the voltage
variation is permitted within 2%. Fig. 9 shows the results, ex-
pressing it with a linear function. From the results, the equation
between the output power (p.u.) Pout and the gain Kc can be
obtained as follows:

Kc = max(1.3 |Pout | , 0.1) (1) Fig. 9. Gain-dc/dc converter input power characteristics. (Voltage variation
2%).

where the input power is expressed by a negative number, and the


B. Proposed Control for Stored Energy Balancing
minimum value of Kc is determined as 0.1. This gain-scheduling
control allows better load sharing and voltage regulation simul- We propose a dc-voltage control that incorporates a fuzzy
taneously. However, a stored energy control is also needed for control that changes the dc voltage reference to balance the
the converter energy storage unit device to prevent a surplus or stored energy. Fig. 10 shows the membership functions of fuzzy
deficient of stored energy. control. The input Xk is the ratio of the stored energy and the
2250 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 28, NO. 5, MAY 2013

Fig. 11. Simulation circuit.


Fig. 10. Control rule of fuzzy control.
TABLE II
MAIN PARAMETERS
average of all stored energies as shown
Wk
Xk =  n  (2)
1
n −1 ×( i=1 Wi − W k )

Wk is the charged energy of the kth EDLC bank, which is


described as follows:

2 Ck vk − 2 Ck Vm in
1 2 1 2
Wk =
2 Ck Vm ax − 2 Ck Vm in
1 2 1 2

vk2 − Vm2 in
= (3)
Vm2 ax − Vm2 in
where the value n refers to the total number of EDLC banks. Ck
and vk are the capacitance and voltage of the kth EDLC bank,
respectively. Vm ax and Vm in are the maximum and minimum
operation voltage of EDLC banks, respectively.

To obtain the dc voltage reference Vdc , Xk is initially calcu-
lated from (2) and (3). Then, a membership value is calculated
from the membership function of PB, PS, Z, NS, and NB. Fi-
Fig. 12. Events of the simulation (initial condition W 2 /W 1 ≈ 2).
nally, the dc voltage reference is obtained from the resulting
membership function. For example, if X1 is 1.7, the values of
PS and PB in the antecedent membership function are 0.4 and

0.6, respectively. Then, Vdc of EDLC1 is given from the conse-
quent membership function as follows:

Vdc = 345 × 0.4 + 350 × 0.6 = 348 [V]. (4)
We assumed that the dc-distribution voltage has a tolerance
of ±5% with reference to ac systems. In this case, the upper Fig. 13. Droop control to obtain the voltage reference.
and lower reference should be within 3% because the gain-
scheduling control includes a tolerance of ±2%. Therefore, the
demonstrated through simulations and experiments described in
upper and lower limit was selected 350 and 330 V in the mem-
the following sections.
bership function of consequent, respectively. An isosceles tri-
angle is usually used as a membership function. However, to
IV. SIMULATION
realize a good voltage regulation, the peaks of PS and NS in
the antecedent are shifted to 1.1 and 0.9, respectively. We first To confirm the performance of the proposed method, simula-
obtained this membership from computer numerical simulation tions were conducted using PSCAD/EMTDC. The circuit and
and then applied it to experiments. Although this fuzzy rule the main parameters are shown in Fig. 11 and Table II, espec-
was determined on the basis of experience, its effectiveness was tively. The configuration and the parameters are based on the
KAKIGANO et al.: DISTRIBUTION VOLTAGE CONTROL FOR DC MICROGRIDS USING FUZZY CONTROL AND GAIN-SCHEDULING TECHNIQUE 2251

Fig. 14. Simulation results (gain-scheduling control only) (Initial condition Fig. 15. Simulation results (gain-scheduling control and fuzzy control) (Initial
W 2 /W 1 ≈ 2). condition W 2 /W 1 ≈ 2).

Although the fuzzy control was chosen for the stored energy
experimental system shown in Fig. 4. Resistances and current
management in this paper, the other control could be a candidate
sources were used as the loads and the DGs, respectively. The ∗
to obtain the voltage reference Vdc . For example, a simple droop
resistances were set to 115.6 Ω, which consumed 1 kW at 340 V.
control can be illustrated in Fig. 13. The control diagram is
The output power of the current source was changed gradually
simpler than the proposed fuzzy control, and the only gain Kv
from 0 to 1 kW or from 1 to 0 kW for 1 s.
is needed as the control parameter. In other words, it has only
Two EDLCs were connected through dc/dc converters:
one degree of freedom to adjust the control performance.
EDLC1 (rated voltage 160 V, and rated capacitance 18 F) and
EDLC2 (rated voltage 216 V, and rated capacitance 18.75 F).
The minimum voltage of both EDLCs was set to be 100 V. A. Simulation Results of Proposed Method
Therefore, EDLC1 and EDLC2 can store about 140 kJ and Fig. 14 shows the simulation results (the dc distribution volt-
344 kJ, respectively. age, the stored energy ratio, the current from EDLC, and the
Fig. 12 shows the ON/OFF events of the loads and DGs in terminal voltage of EDLC) when only the gain-scheduling con-
the simulation. All loads and DGs were turned ON in the first trol technique was used. The maximum voltage of Vdc (EDLC1)
half, and the loads and DGs were turned OFF in the last half. and Vdc (EDLC2) was 346.7 V which was about 340 V + 2%,
The initial voltages of EDLC1 and EDLC2 were 125 and 200 V, and the minimum voltage of them was 332.9 V which was about
respectively. Therefore, the initial stored energy ratio (W2 /W1 ) 340 V – 2%. Therefore, the distribution voltages at the output
was about 2. converter of the EDLCs were within 340 V ± 2%. However, the
2252 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 28, NO. 5, MAY 2013

Fig. 16. Simulation results (gain-scheduling control and droop control, Fig. 17. Simulation results (gain-scheduling control and droop control,
K v = 10) (Initial condition W 2 /W 1 ≈ 2). K v = 50) (Initial condition W 2 /W 1 ≈ 2).

B. Simulation Results of Droop Control


stored energy of EDLCs was not balanced, and the maximum
of the ratio (W2 /W1 ) reached 4.2. To compare the proposed fuzzy control described in the previ-
Fig. 15 shows the simulation results when the proposed con- ous section and the droop control shown in Fig. 13, simulations
trol was used. The stored energy ratio tended to be 1. The fea- of the droop control were conducted in the same condition.
tures of EDLC currents were different from those in Fig. 14 be- Fig. 16 shows the simulation results when the droop control
cause EDLC2 had 1.3 times energy at the initial state. EDLC2 was used, and Kv was set to 10. Although the stored energy
initially discharged its power, and then the energies of both ratio tended to be about 1, it took about 160 s, which was twice
EDLCs balanced at around 80 s. The reason for the difference as the result of the fuzzy control. Fig. 17 shows the simulation
in the EDLC terminal voltages, VEDLC1 and VEDLC2 , despite an results when Kv was set to 50. The stored energy ratio also
almost unity ratio of EDLC charge (W2 /W1 ), is the difference tended to be 1, and it took about 80 s, which was almost the
in the maximum voltages of EDLC1 and EDLC2. Regarding the same as the result of the fuzzy control shown in Fig. 15. In this
distribution voltages at the output converters of the EDLCs, the case, the maximum and minimum voltage of the output convert-
maximum and the minimum were 349.9 and 331.7 V, respec- ers of the EDLCs was 349.9 and 331.6 V, respectively. From the
tively. Although the range of the voltage was a little wider than result, it seems that the proposed fuzzy control can be replaced
the previous results, the range was within 340 V ± 3%, which by the simple droop control. To confirm the advantage of the
was satisfied with the assumed specification (340 V ± 5%) proposed fuzzy control, another simulation was conducted in
described in Section III-B. the following section.
KAKIGANO et al.: DISTRIBUTION VOLTAGE CONTROL FOR DC MICROGRIDS USING FUZZY CONTROL AND GAIN-SCHEDULING TECHNIQUE 2253

Fig. 18. Simulation results (gain-scheduling control and droop control, Fig. 19. Simulation results (gain-scheduling control and fuzzy control) (Initial
K v = 50) (Initial condition W 2 /W 1 ≈ 0.5). condition W 2 /W 1 ≈ 0.5).

droop control, the results after 120 s were almost the same as
C. Comparison Between Fuzzy Control and Droop Control
the previous droop control.
In this simulation, the configuration, parameters and events Fig. 20 shows that the relation between input Xk and the

were the same as the previous simulation. The only difference output Vdc in each control. The fuzzy control changes the slope
was that the all parameters of EDLC1 and EDLC2 were ex- in the range of Xk between 0.5 and 2.0. On the contrary, the
changed: EDLC1 (rated voltage, 216 V and rated capacitance droop control has one slope in the shorter range of Xk between

18.75 F) and EDLC2 (rated voltage 160 V and rated capaci- 0.8 and 1.2, and the Vdc was fixed on the minimum (330 V) or
tance, 18 F). The minimum voltage of both EDLCs was also set the maximum (350 V) in the other area.
to be 100 V, and the initial voltages of EDLC1 and EDLC2 were The characteristics of the two controls can be seen from the
200 and 125 V, respectively. Therefore, the initial stored energy results in Figs. 18 and 19. Because the initial value of Xk was
ratio (W2 /W1 ) was about 0.5. set to 0.5 in each case, the power from EDLC1 was supplied
Fig. 18 shows the simulation results when the droop control to EDLC2 in order to balance the stored energy. In Fig. 18,
(Kv = 50) was used. Although the stored energy ratio tended the current from the converter of EDLC1 and EDLC2 (IEDLC1
to be 1, it exceeded up to 1.05 around 75 s. After the stored and IEDLC2 ) were constant from 0 to 20 s because Xk was
energy ratio returned to 1, it dropped to 0.95 around 180 s. lower than 0.8, and the voltage references Vdc (EDLC1)∗ of
Fig. 19 shows the simulation results when the fuzzy control was and Vdc (EDLC2)∗ were fixed to the upper and lower limit, re-
used. Although there was no overshoot like the results of the spectively. On the other hand, the absolute value of IEDLC1
2254 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 28, NO. 5, MAY 2013

Fig. 22. Circuit of experimental system.

of the droop control, while the integral numbers of the square


of IEDLC2 at 250 s was almost the same. It indicates that the
loss in case of the fuzzy control is lower than that of the droop
control. Therefore, this paper proposes the fuzzy control as the
Fig. 20. Relations between input X k and voltage reference V d∗c . energy balance control.

V. EXPERIMENT
Two experiments were conducted to confirm the performance
of the proposed method practically. The circuit and the main
parameters are shown in Fig. 22 and Table III, respectively. The
system configuration and parameters are almost the same as
the simulation circuit. In house3, we used a commercial GEC,
and the rated capacity was 1 kW. The generator outputs ac
340 V, 307.5 Hz, and the ac power was converted into a dc 390–
400 V by a thyristor-diode rectifier. In general, the dc power
is converted to a single-phase ac 200 V and is supplied to the
utility grid, but we modified it to accept the dc power directly. In
the other two houses, we substituted the dc power supply for the
real GEC, which outputs 400-V dc constantly. Buck choppers
were connected between GECs and the dc distribution lines
because the dc voltage of the gas engine was different from the
distribution voltage. The output power was changed gradually
from 0 to 1 kW or from 1 to 0 kW for 1 s. The ratings of the two
EDLCs were the same as the simulation: EDLC1 (rated voltage
160 V, and rated capacitance 18 F) and EDLC2 (rated voltage
216 V and rated capacitance 18.75 F). The minimum voltage of
both EDLCs was also set to be 100 V. For the dc loads, we used
three electronic loads under constant resistance mode, each with
Fig. 21. Integral of the square of the current (IE D L C 1 and IE D L C 2 ). a rated capacity of 1 kW.

and IEDLC2 in Fig. 19 were decreasing during the same period, A. Case (a)
which contributed to reduce the losses due to the line resistances The ON/OFF events of the loads and DGs in case (a) were
(Rline1 and Rline2 ) and the inner resistances of both EDLCs. the same as the previous simulation shown in Fig. 12. The
Fig. 21 shows the integral of the square of IEDLC1 and IEDLC2 initial voltages of EDLC1 and EDLC2 were 125 and 200 V,
in each case. In case of the fuzzy control, the integral numbers of respectively. Then, the initial stored energy ratio (W2 /W1 ) was
the square of IEDLC1 at 250 s was 20.1% lower than the results about 2. Fig. 23 shows the experimental results when only the
KAKIGANO et al.: DISTRIBUTION VOLTAGE CONTROL FOR DC MICROGRIDS USING FUZZY CONTROL AND GAIN-SCHEDULING TECHNIQUE 2255

TABLE III
MAIN EXPERIMENT PARAMETERS

Fig. 24. Experimental results of case (a) (gain-scheduling control and fuzzy
control).

same as the simulation results. However, the distribution volt-


age Vdc (EDLC2) and the current IEDLC2 in Fig. 23 were higher
than the simulation results in Fig. 14 during the first 120 s. The
reason is conjectured that the feedback value of Vdc (EDLC2)
were a little lower than the real values in the experiment, which
Fig. 23. Experimental results of case (a) (gain-scheduling control only). phenomenon was confirmed by numerical simulations. Apart
from it, the experimental results demonstrated that the validity
of the proposed methods. Regarding the performance during the
gain-scheduling control technique was used. The dc distribu- change of loads/sources, there were no disturbances or instabil-
tion voltage was controlled within 340 V ± 5%, but the stored ities due to the gain-scheduling control or the fuzzy control. In
energy of EDLCs was not balanced and the peak ratio was case of the fuzzy control, the sign of IEDLC1 and IEDLC2 were
around 3.5. Fig. 24 shows the experimental results when the pro- opposite at first to balance their energy. Then, the envelopes of
posed control was used. The energies of both EDLCs balanced IEDLC1 and IEDLC2 were gradually close each other during the
at around 100 s. The both experimental results were almost the change of loads/sources.
2256 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 28, NO. 5, MAY 2013

Fig. 25. Experimental results of case (b) (gain-scheduling control only).


Fig. 26. Experimental results of case (b) (gain-scheduling control and fuzzy
control).
B. Case (b)
In this experiment, Load1 and the output of DG3 were turned
ON from 0 to 1 kW at 5 s. Other loads and DGs were turned distribution voltage, EDLC1 supplied Load1 with electricity,
OFF through the experiment. The initial voltage of EDLC1 and and EDLC2 was charged with power from DG3. Therefore, the
EDLC2 were 130 and 180 V, respectively. Therefore, the initial stored energy ratio was increasing.
stored energy ratio (W2 /W1 ) was about 1.3. Fig. 26 shows the experimental results when the proposed
Fig. 25 shows the experimental result when only the gain- control was used. The stored energy ratio (W2 /W1 ) approaches
scheduling control technique was used. If there had been no line 1, and the energy ratios of each EDLC became equal. The volt-
resistances between house1 and house3, the distribution voltage ages of EDLC1 and EDLC2 decreased slightly because of the
Vdc (house1) would have been the same as Vdc (EDLC2) after presence of switching losses of the converters and the line resis-
the event. In this case, the output power of DG3 would have tance. From this result, we confirmed that the stored energy ratio
been supplied to the Load1 directly, and the both EDLCs would converged by the proposed control. The response took around
have not outputted their power ideally. However, there were line 150 s in this case, which is satisfactory for the dc microgrids as-
resistances R2 (1 Ω × 2) between house1 and house3 in the sumed in this paper. If there is a case where the faster response
experimental system, and the distribution voltage Vdc (house1) is needed, the member ship function of antecedent has to be
and Vdc (EDLC1) became lower than Vdc (EDLC2). To keep the shrunk horizontally centering on X = 1.
KAKIGANO et al.: DISTRIBUTION VOLTAGE CONTROL FOR DC MICROGRIDS USING FUZZY CONTROL AND GAIN-SCHEDULING TECHNIQUE 2257

VI. CONCLUSION [12] A. A. A. Radwan and Y. A.-R. I. Mohamed, “Linear active stabilization
of converter-dominated DC microgrids,” IEEE Trans. Smart Grid, vol. 3,
This paper presented a new dc distribution voltage control for no. 1, pp. 203–216, Mar. 2012.
dc/dc converters with an energy storage unit. The proposed con- [13] M. Noritake, K. Hirose, M. Yamasaki, T. Oosawa, and H. Mikami, “Eval-
uation results of power supply to ICT equipment using HVDC distribution
trol combines a gain-scheduling technique with fuzzy control. system,” in Proc. Int. Telecom. Energy Conf., Jun. 2010, pp. 1–8.
The experimental results show that the dc distribution voltage [14] H. Kakigano, Y. Miura, T. Ise, T. Momose, and H. Hayakawa, “Funda-
was within 340 V ± 5%, and the ratios of the storage units were mental characteristics of DC microgrid for residential houses with cogen-
eration system in each house,” in Proc. IEEE Power Energy Soc. General
approximately equal. This indicates that dc voltage regulation Meeting, Jul. 2008, Paper 08GM0500, pp. 1–8.
and stored energy balancing control are realized simultaneously. [15] H. Kakigano, Y. Miura, and T. Ise, “Low-voltage bipolar-type DC micro-
The main advantage of the proposed control is presented for grid for super high quality distribution,” IEEE Trans. Power Electron.,
vol. 25, no. 12, pp. 3066–3075, Dec. 2010.
the cases where the model is unknown or is mathematically [16] Z. Ye, D. Boroyevich, K. Xing, and F. C. Lee, “Design of parallel sources
complex. Therefore, the proposed control is relatively easy to in DC distributed power systems by using gain-scheduling technique,” in
introduce to real-life projects compared with the modern control Proc. IEEE Power Electron. Spec. Conf., Aug. 1999, vol. 1, pp. 161–165.
[17] W. Tang and R. H. Lasseter, “An LVDC industrial power distribution
theories that utilize the time-domain state space representation. system without central control unit,” in Proc. IEEE Power Electron. Spec.
However, for this purpose, trial and error methods might be Conf., 2000, vol. 2, pp. 979–984.
adopted to adjust the membership functions in practice, which [18] Y. Ito, Y. Zhongqing, and H. Akagi, “DC microgrid based distribution
power generation system,” in Proc. IEEE Int. Power Electron. Motion
is a time consuming process. Our future study will attempt to Control Conf., Aug. 2004, vol. 3, pp. 1740–1745.
establish the design procedure of the membership functions. [19] Y. A. I. Mohamed and E. F. El-Saadany, “Adaptive decentralized droop
In actual application, if a dc microgrid is extended and another controller to preserve power sharing stability of paralleled inverters in
distributed generation microgrids,” IEEE Trans. Power Electron., vol. 23,
energy storage system is included, it would require a communi- no. 6, pp. 2806–2816, Nov. 2008.
cation line to obtain state information and detect faults. In this [20] A. M. Roslan, K. H. Ahmed, S. J. Finney, and B. W. Williams, “Im-
case, equation (2) must be updated such that the new input Xk proved instantaneous average current-sharing control scheme for parallel-
connected inverter considering line impedance impact in microgrid net-
can be calculated. Although the membership functions appear works,” IEEE Trans. Power Electron, vol. 26, no. 3, pp. 702–716, Mar.
to be the same as shown in Fig. 10 from the simulation analysis, 2011.
we will address this in another study after the accomplishment [21] T. M. Haileselassie and K. Uhlen, “Impact of DC line voltage drops on
power flow of MTDC using droop control,” IEEE Trans. Power Systems,
of design procedure for the membership functions. vol. 27, no. 1, pp. 1441–1449, Aug. 2012.
[22] C. N. Papadimitriou and N. A. Vovos, “A fuzzy control scheme for in-
tegration of DGs into a microgrid,” in Proc. IEEE Mediterranean Elec-
trotechnical Conf., Apr. 2010, pp. 872–877.
REFERENCES [23] A. Elmitwally and M. Rashed, “Flexible operation strategy for an isolated
PV-diesel microgrid without energy storage unit,” IEEE Trans. Energy
[1] N. Hatziargyriou, J. Driesen, F. Katiraei, B. Kroposki, R. Lasseter, T. Ise, Convers., vol. 26, no. 1, pp. 235–244, Mar. 2011.
S. Morozumi, S. Papathanassiou, R. Iravani, A. Dimeas, C. Marnay, H. [24] B. N. Alajmi, K. H. Ahmed, S. J. Finney, and B. W. Williams, “Fuzzy-
Asano, S. Papathanassiou, G. Strbac, and G. Venkataramanan, “Micro- logic-control approach of a modified Hill-Climbing method for maximum
grids,” IEEE Power Energy Mag., vol. 6, no. 3, pp. 26–94, 2008. power point in microgrid standalone photovoltaic system,” IEEE Trans.
[2] Lawrence Berkeley National Laboratory. Microgrid Symposium. (Sep. Power Electron., vol. 26, no. 4, pp. 1022–1030, Apr. 2011.
2012). [Online]. Available: http://der.lbl.gov/microgrid-symposiums [25] T. Konishi, S. Hase, and Y. Nakamichi, “Verification tests of energy storage
[3] Q. Zhang, R. Callanan, M. K. Das, S. Ryu, A. K. Agarwal, and system for DC electrified railways using EDLC,” Quart. Rep. Railway
J. W. Palmour, “SiC power devices for microgrids,” IEEE Trans. Power Tech. Res. Inst., vol. 48, no. 2, pp. 71–77, 2007.
Electron, vol. 25, no. 12, pp. 2889–2896, Dec. 2010. [26] H. Chen, T. N. Cong, W. Yang, C. Tan, Y. Li, and Y. Ding, “Progress in
[4] T. Hornik and Q. Zhong, “A current-control strategy for voltage-source electrical energy storage system: A critical review,” Progress Naural Sci.,
inverters in microgrids based on H∞ and repetitive control,” IEEE Trans. vol. 19, no. 3, pp. 291–312, 2009.
Power Electron, vol. 26, no. 3, pp. 943–952, Mar. 2011. [27] S. M. Muyeen, R. Takahashi, T. Murata, and J. Tamura, “Integration of
[5] P. Tenti, H. K. M. Paredes, and P. Mattavelli, “Conservative power theory, an energy capacitor system with a variable-speed wind generator,” IEEE
a framework to approach control and accountability issues in smart mi- Trans. Energy Convers., vol. 24, no. 3, pp. 740–749, Sep. 2009.
crogrids,” IEEE Trans. Power Electron., vol. 26, no. 3, pp. 664–673, Mar.
2011.
[6] J. Rocabert, G. M. S. Azevedo, A. Luna, J. M. Guerrero, J. I. Candela, and
P. Rodriguez, “Intelligent connection agent for three-phase grid-connected
microgrids,” IEEE Trans. Power Electron., vol. 26, no. 10, pp. 2993–3005,
Oct. 2011.
[7] J. M. Guerrero, J. C. Vasquez, J. Matas, L. G. Vicuna, and M. Castilla, “Hi-
erarchical control of droop-controlled AC and DC microgrids—A general
approach toward standardization,” IEEE Trans. Ind. Electron., vol. 58,
no. 1, pp. 158–172, Jan. 2011.
[8] A. Kwansinski and C. N. Onwuchekwa, “Dynamic behavior and stabiliza- Hiroaki Kakigano (M’06) was born in 1976. He
tion of DC microgrids with instantaneous constant-power loads,” IEEE received the B.S. and M.S. degrees in nuclear engi-
Trans. Power Electron., vol. 26, no. 3, pp. 822–834, Mar. 2011. neering from Nagoya University, Nagoya, Japan, in
[9] A. Kwansinski, “Quantitative evaluation of DC microgrids availability: 1999 and 2001, respectively. He received the Ph.D.
Effects of system architecture and converter topology design choices,” degree in electrical engineering from Osaka Univer-
IEEE Trans. Power Electron., vol. 26, no. 3, pp. 835–851, Mar. 2011. sity, Osaka, Japan, in 2009.
[10] R. S. Balog and P. T. Krein, “Busselection in multibus DC microgrids,” After he worked at an Electric Company, he joined
IEEE Trans. Power Electron., vol. 26, no. 3, pp. 860–867, Mar. 2011. Osaka University in 2004 for a doctorate course,
[11] T. Wu, K. Sun, C. Kuo, and C. Chang, “Predictive current controlled 5-kW where he is currently an Assistant Professor. From
single-phase bidirectional inverter with wide inductance variation for DC- August 2011 to March 2012, he was a Visiting Profes-
microgrid applications,” IEEE Trans. Power Electron., vol. 25, no. 12, sor with KU Leuven, Leuven, Belgium. His research
pp. 3076–3084, Dec. 2010. interests include power electronics, microgrids, and dc distribution systems.
2258 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 28, NO. 5, MAY 2013

Yushi Miura (M’06) received the Doctorate de- Toshifumi Ise (M’87) was born in 1957. He received
gree in electrical and electronic engineering from the the Bachelor, Master, and Doctor of Engineering de-
Tokyo Institute of Technology, Tokyo, Japan, in 1995. grees in electrical engineering from Osaka Univer-
From 1995 to 2004, he was at Japan Atomic sity, Osaka, Japan, in 1980, 1982, and 1986, respec-
Energy Research Institute as a Researcher where tively.
he developed power supplies and superconducting He joined , Osaka University in 1990, where he
coils for nuclear fusion reactors. Since 2004, he has is currently a Professor in the Division of Electrical,
been an Associate Professor in the Division of Elec- Electronic, and Information Engineering, Faculty of
trical, Electronic, and Information Engineering at Engineering. From 1986 to 1990, he was at the Nara
Osaka University, Osaka, Japan. His research inter- National College of Technology, Nara, Japan. His re-
ests include applications of power electronics and search interests include the areas of power electronics
superconducting technology. Currently he is interested in control of distributed and applied superconductivity including superconducting magnetic energy stor-
generations and energy storages in the power systems. ages and new distribution systems.
Dr. Ise is a Member of the Institute of Electrical Engineers of Japan and the
Japan Society for Power Electronics.

You might also like