2890 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 31, NO.

4, APRIL 2016

A Virtual Impedance Optimization Method for

Reactive Power Sharing in Networked Microgrid
Yixin Zhu, Student Member, IEEE, Fang Zhuo, Member, IEEE, Feng Wang, Member, IEEE,
Baoquan Liu, Student Member, IEEE, Ruifeng Gou, and Yangjie Zhao

Abstract—Unlike the typical microgrid with a common ac bus, mode or grid-connected mode. Microgrid can not only solve the
networked microgrid always suffers more serious reactive power problem of large scale DG coupling but also play full advantages
sharing issues due to its complex inner structure. In such case, of DG units, improve power supply reliability and bring more
the system reactive power sharing error cannot be easily evalu-
ated and eliminated. So, this paper proposes a wireless control benefits to its users. However, microgrid also faces some chal-
strategy that employs optimized virtual impedance controllers and lenging problems, such as control stability and power sharing
local load measurements for the reactive power sharing in net- issues.
worked microgrid. First, from the modeling of microgrid network, In an islanded microgrid, loads must be properly shared by
an estimation method for network reactive power sharing error multiple DG units. Conventionally, the frequency and voltage
is derived. Through the estimation method-based network fea-
ture analyses, corresponding design for virtual impedance con- magnitude droop control method is adopted, which mimics the
troller is presented. Then, by introducing genetic algorithm, virtual behavior of synchronous machines in power systems [4]–[9].
impedance controller parameters of each distributed generation The droop control technique provides a decentralized control
unit are optimized, which aims to minimize the microgrid global capability, which makes the microgrid operation independent
reactive power sharing error. The parameter optimization process of communication links. This wireless control manner enables
is performed offline in microgrid configuration stage. By using
these optimized virtual impedance controllers, the reactive power the “plug and play” interfacing of DG units and also enhances
sharing performance of a networked microgird can be greatly im- the reliability of system. Although the frequency droop control
proved. Finally, the accuracy of the estimation method is validated always achieves accurate real power sharing, the voltage droop
by MATLAB simulation results, and the feasibility of the proposed control typically results in poor reactive power sharing due to
virtual impedance optimization method is verified through real the mismatch in feeder impedances, and also to the different
power experiments.
offsets of local loads. Moreover, when the microgrid structure
Index Terms—Estimation method, genetic algorithm (GA), net- is networked, the reactive power sharing issue will be further
worked microgrid, reactive power sharing, virtual impedance. aggravated due to the irregular distribution of DG units.
I. INTRODUCTION To improve the reactive power sharing performance of droop
control, many solutions have been developed [10]–[18]. In [10],
ITH the increased concerns on environment and clean
W energy, more and more renewable energy sources such
as photovoltaic cells, wind turbines, and microturbines are in-
reactive power sharing errors are reduced through the injection
of a small ac voltage signal in the system. However, this ap-
proach may reduce the quality of DG output voltages and line
tegrated into the power grid in the form of distributed genera- currents. As an important tool for droop control method, the
tion (DG) units. These DG units are normally interfaced to the well-known virtual impedance method can enhance the system
grid through power electronic converters. To effectively man- stability and reactive power sharing accuracy at the same time.
age these DG units, the microgrid concept is proposed, which In [11] and [12], predominant virtual inductors are placed at
is considered to pave the way to the future smart grid [1]–[3]. DG unit outputs, which mainly aimed to prevent the power con-
Microgrid usually connects to the power grid through the point trol instability. In [13]–[16], a comprehensive treatment about
of common coupling (PCC); it can operate in either islanded virtual impedance concept is presented. The focus has been on
the mismatched output impedances of closed-loop controlled
Manuscript received December 6, 2014; revised May 4, 2015 and Febru- DG units, while the mismatch in feeder impedances has not
ary 21, 2015; accepted June 12, 2015. Date of publication June 26, 2015; been considered. Thus, reactive power sharing issues cannot
date of current version November 30, 2015. This work was supported by Na-
tional High-tech Research and Development Program (863 program) under be solved completely. In [17], reactive power sharing errors
Grant 2015AA050606 and Natural Science Foundation of China under Grant are reduced through the modification of droop control slopes.
51177130. Recommended for publication by Associate Editor J. M. Guerrero. With the preset virtual inductor and estimated feeder impedance
Y. Zhu, F. Zhuo, F. Wang, and B. Liu are with the State Key Laboratory
of Electrical Insulation and Power Equipment, School of Electrical Engineer- value, the slope of Q–V droop control is modified to compensate
ing Xi’an Jiaotong University, Xi’an 710049, China (e-mail: zhuyixin1987@ the mismatch in feeder voltage drops. In [18], communication
163.com; [email protected]; [email protected]; comeliu is introduced to facilitate the estimation of feeder impedance
[email protected]).
R. Gou and Y. Zhao are with the Xian High Voltage Apparatus Research value, making the system reactive power sharing more accurate.
Institute, XiDian Group, Xi’an 710049, China (e-mail: [email protected]; Aforementioned reactive power sharing strategies have consid-
[email protected]). ered the impacts of DG output impedances, feeders, and local
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org. loads. However, in a networked (multi-bus) microgrid, besides
Digital Object Identifier 10.1109/TPEL.2015.2450360 aforementioned mismatched factors, the networked structure is

0885-8993 © 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications standards/publications/rights/index.html for more information.
ZHU et al.: VIRTUAL IMPEDANCE OPTIMIZATION METHOD FOR REACTIVE POWER SHARING IN NETWORKED MICROGRID 2891

also an important factor that affects the system reactive power

sharing. Without a complete consideration of these mismatched
factors, previous methods cannot work properly as desired in
networked microgrids.
To solve the reactive power sharing issue in networked mi-
crogrids, communication-based solutions become shortcuts. In
[19] and [20], intercommunication among DG units is utilized
to improve the reactive power sharing accuracy. However, this
manner is subject to the DG install locations. Synchronizing
signals are utilized in [21]–[23] to trigger an extra regulation for
reactive power sharing, but the control performance is easy to
be influenced if load changes during the regulation period. With
the communication between microgrid central controller and
DG units, reactive power sharing methods based on set-point
regulation, also known as secondary control [24]–[28], can be
well applied in networked microgrids. With a similar commu-
nication mechanism, virtual impedances can also be adjusted to Fig. 1. Illustration of a networked microgrid.
realize reactive power sharing [29]–[31]. However, it seems that
the communication-based virtual impedance methods have no
significant benefits compared with set-point regulation methods. microgrid can also operate in either islanded mode or grid-
Although communication can help improving the reactive connected mode. In grid-connected mode, as the voltage of
power sharing accuracy, it also introduces some new issues, microgrid is supported by the main grid, the power sharing issue
such as cost increase and expansion capability decrease. In can be easily solved by adopting power tracking techniques.
this paper, a wireless reactive power sharing control strategy However, in islanded mode, the power sharing relies on the
that employs optimized virtual impedance controllers and lo- cooperation among multiple DG units.
cal load measurements is proposed for networked microgrids. Normally, for islanded operation, DG units can employ the
Based on the microgrid network modeling, the cause of reac- conventional real power–frequency droop control and reactive
tive power sharing issue is revealed. Meanwhile, an estimation power–voltage magnitude droop control as
method for network reactive power sharing error is proposed.
ωi∗ = ω0 − DP i Pi (1)
Through the estimation method-based network feature analy-
ses, virtual impedance controller is designed accordingly; then Ei∗ = E0 − DQ i Qi (2)
detailed controller parameters are optimized by using genetic
algorithm (GA), which is to introduced to minimize the system where ω0 and E0 are the initial values of DG angular frequency
global reactive power sharing error. The optimization process is and voltage magnitude, Pi and Qi are the measured real and
performed offline in microgrid configuration stage. With the op- reactive powers after the first-order low-pass filtering, and DP i
timized virtual impedance controllers, DG units can well share and DQ i are the real and reactive power droop slopes, respec-
the loads in microgrid. Throughout the whole strategy, commu- tively. With the derived angular frequency and voltage magni-
nication is not used; instead, local load measurement is added tude from (1) and (2), the instantaneous voltage reference can
for the virtual impedance control, which has no effect on the be obtained accordingly.
wireless manner of droop control. Finally, simulation and ex- Normally, larger capacity DG units are set with relatively
perimental results are provided to verify the proposals. smaller droop slopes. So that when all the DG units operate
under the same frequency and voltage magnitude, larger ca-
pacity DG units could output more real and reactive powers
II. OPERATION AND MODELING OF NETWORKED MICROGRID according to (1) and (2). In practical application, due to the
A. Operation of Networked Microgrid consistent frequency among DG units, the P–ω droop control
always achieves accurate real power sharing. However, as the
With the fast development of microgrid, more “plug and play”
voltage magnitudes of DG units can hardly be unified due to
DG units are applied into the microgrid. This process leads
the mismatch in network, the Q–V droop control always suffers
to some changes in the configuration of traditional single-bus
reactive power sharing issues.
microgrid and forms the networked microgrid as illustrated in
Fig. 1. In such microgrid, DG units can be connected to any
B. Modeling of Microgrid Network
node of the network. Some loads connected directly to DG
units are called local loads, and the others are public loads. In a single-bus microgrid, all the DG units are connected to
Considering that the focus of this paper is the fundamental the common ac bus through their respective DG feeders. By
real and reactive power control, the amount of nonlinear loads comparing the impedance values of these feeders, the reactive
is limited in the microgrid. Like the traditional microgrid, by power sharing performance of system can be basically obtained.
controlling the static transfer switch at the PCC, a networked However, for a networked microgrid, it is not straightforward
2892 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 31, NO. 4, APRIL 2016

nodes j to k)
     
Vn D j Vn D k Vn D i
− = Af i (4)
Vn Q j Vn Q k Vn Q i
where
 
Rf i −ωcom Lf i
Af i = .
ωcom Lf i Rf i
In (4), Vn D j and Vn Q j are the DQ axis voltages of node j;
Vn D k and Vn Q k are the DQ axis voltages of node k; If D i and
If Q i are the DQ axis currents of feeder i; Rf i and Lf i are the
resistance and reactance of feeder i. It is to be noted that line i
can be either a DG feeder or a network feeder, and node j can
be either a DVS node or a network node. By combining all the
feeder voltages together, the whole feeder model can be derived
as
Fig. 2. Sketch of the networked microgrid model.
BF Vn D Q = AF If D Q (5)
where
⎡ ⎤
to evaluate the system reactive power sharing performance due Af 1 0 ... 0
⎢ ⎥
to its complex structure. To address this problem, more detailed ⎢ 0 Af 2 ... 0 ⎥
⎢ ⎥
power sharing analyses are needed. Fig. 2 depicts a conceptual Af = ⎢ ⎥
⎢ ... ... ... ... ⎥
networked microgrid model, and the dashed part indicates the ⎣ ⎦
research object in this paper: equivalent network, containing 0 0 ... Af (n +s)
physical network and DG feeders. For the sake of analysis, only
T
linear load condition is considered, and all the DG units are set Vn D Q = Vn D 1 Vn Q 1 ... Vn D (m +s) Vn Q (m +s)
to the same power rating. T
As can be seen from Fig. 2, the whole microgrid system is If D Q = If D 1 If Q 1 ... If D (n +s) If Q (n +s) .
composed of a physical network and several DG units. Each In (5), the mapping matrix BF is size of 2(n+s) × 2(n+s),
DG unit can be equivalent to a droop controlled voltage source which reflects the connecting relationship between nodes and
(DVS) in series with a DG feeder. The DG feeder can be either feeders. For example, if line i connects between nodes j and k,
a coupling inductor or a virtual feeder if virtual impedance the element BF (i,j) and BF (i,k) will be +1 or –1 (depending on
method is applied. By separating DVS units from the system, the current direction), and all the other elements in that row will
the equivalent network of microgrid is obtained. The detailed be 0. In the equivalent network, some nodes are DVS nodes,
network model will be explained in the following part of this and the voltages of these nodes are treated as the inputs of this
section. It is worth noting that in the modeling process, the network model. By extracting them from vector Vn D Q , equation
three-phase voltages and currents are all represented as vectors (5) can be revised as
in d–q rotating reference frame. In this paper, the reference
frame of DVS 1 is regarded as the common reference frame, BF 1 VdvsD Q + BF 2 Vpn D Q = AF If D Q (6)
whose rotating frequency is ωcom , while the system variables where
in other individual reference frames are all translated to this T
common reference frame using the following equations: EdvsD Q = EdvsD 1 EdvsQ 1 ... EdvsD s EdvsQ s
     Vpn D Q = Vpn D 1 Vpn Q 1 ... Vpn D m Epn Q m
T
.
FD cos δi − sin δi Fdi
= (3)
FQ sin δi cos δi Fq i In (6), EdvsD i and EdvsQ i are the DQ axis voltages of DVS
i; the mapping matrixes BF 1 and BF 2 are separated from BF ,
where FD and FQ are the DQ axis components of variable F and they are size of 2(n+s) × 2s and 2(n+s) × 2m, respectively.
in the common reference frame; Fdi and Fq i are the dq axis 2) Load Model: The voltage of the RL load connected to
components of variable F in the reference frame of DVS i; and network node j can be expressed as
δi is the angle difference between the two reference frames.    
Vpn D j IlD j
Back to the microgrid model shown in Fig. 2, the equivalent = Alj (7)
network totally contains s DG feeders, n network feeders, s DVS Vpn Q j IlQ j
nodes and m network nodes, where s is the number of DG units. where
The detailed modeling process is shown as follows:  
Rlj −ωcom Llj
1) Feeder Model: The voltage of feeder line i between nodes Alj = .
j and k can be expressed as (assuming that currents flow from ωcom Llj Rlj
ZHU et al.: VIRTUAL IMPEDANCE OPTIMIZATION METHOD FOR REACTIVE POWER SHARING IN NETWORKED MICROGRID 2893

In (7), IlD j and IlQ j are the DQ axis currents of load j; In (12), YNET is presented as a partition matrix. Its diagonal
Rlj and Llj are the resistance and reactance values of load j, submatrix Yli is the input admittance of DVS i, which reflects
respectively. If there is no load connected to node j, a sufficiently the network port characteristics. Its off-diagonal submatrix Yij
large ground resistance can be introduced to mimic the no load and Yj i reflect the interconnection between DVS i and j, and
characteristic. Hence, the whole model of m loads can be given they are always equal. Thus, the network matrix is a partition
by symmetric matrix. By observing the diagonal elements in YNET ,
the mismatch in network can be basically obtained. When the
Vpn D Q = AL IlD Q (8) input admittances of DVS units are equal, the corresponding
where network will have an ideal reactive power sharing performance.
⎡ ⎤ However, this is not a necessary condition for accurate reactive
Al1 0 ... 0 power sharing, as will be discussed in the following.
⎢ ⎥
⎢ 0 Al2 ... 0 ⎥
⎢ ⎥
AL = ⎢ ⎥ C. Estimation of Reactive Power Sharing Errors
⎢... ··· ... ... ⎥
⎣ ⎦
As mentioned before, DG units are all set to the same power
0 0 ... Alm rating for a convenient comparison. That means each DG unit
IlD Q = IlD 1 IlQ 1 ... IlD m IlQ m
T
. should output the same real and reactive powers in microgrid.
According to (11) and (12), the DVS DQ axis currents can be
3) Complete Model of Equivalent Network: According to the calculated as
network connection, the relationship between feeder currents
and load currents can be expressed as IdvsD = GNET EdvsD − BNET EdvsQ (13)
IdvsQ = BNET EdvsD + GNET EdvsQ (14)
IlD Q = AF 2L If D Q . (9)
where
In (9), matrix AF 2L is size of 2m × 2n maps the connection
between loads and feeders. For example, when lines i and j EdvsD = [EdvsD 1 EdvsD 2 ... EdvsD s ]T
are connected to the network node k, the element AF 2L (k,i)
and AF 2L (k,j) will be +1 or –1 (depending on whether the EdvsQ = [EdvsQ 1 EdvsQ 2 ... EdvsQ s ]T
line current is entering or leaving the node), and all the other IdvsD = [IdvsD 1 IdvsD 2 ... IdvsD s ]T
elements in that row will be 0. By transforming (6), (8), and (9),
the complete equivalent network model can be derived as IdvsQ = [IdvsQ 1 IdvsQ 2 ... IdvsQ s ]T .
If D Q = ANET EdvsD Q (10) In (13) and (14), IdvsD i and IdvsQ i are the DQ axis currents
of DVS i; EdvsD i and EdvsQ i are the DQ axis voltages of DVS
where i; GNET and BNET are the system conductance and susceptance
ANET = (AF − BF 2 AL AF 2L )−1 BF 1 . matrixes obtained from matrix YNET . According to the common
reference frame setting, the Q axis voltage of DVS 1 should be
By extracting DVS currents from vector If D Q , following 0. Then, equation (13) can be rewritten as
equation can be derived:  
IdvsD = GNET EdvsD − BNET EdvsQ (15)
IdvsD Q = CNET If D Q = YNET EdvsD Q (11)

where matrix BNET is size of s×(s–1), which is obtained from

where BNET by eliminating the first column; Similarly, vector EdvsQ
is obtained from EdvsQ by eliminating the first element EdvsQ 1 .
YNET = CNET ANET .
As the real power sharing of the system using the droop control
In (11), CNET is of size 2s × 2n, which is used to pick method is always accurate, the elements in IdvsD should be
out DVS currents from feeder currents. YNET is the admittance equal. Thus, equation (13) can be transformed as
matrix of equivalent network, and its detailed form is shown  
Tr e IdvsD = Tr e GNET EdvsD − Tr e BNET EdvsQ =0 (16)
next
⎡ ⎤ where
Y11 Y12 . . . Y1s
⎢ ⎥ ⎡ ⎤
⎢ Y21 Y22 . . . Y2s ⎥ 1 −1 0 ... 0
⎢ ⎥ ⎢ ⎥
YNET = ⎢ ⎥ (12) ⎢ 1 −1 ⎥
⎢... ... ... ... ⎥ ⎢ 0 ... 0 ⎥
⎣ ⎦ Tr e =⎢ ⎥ .
⎢ ... ... ... ... ... ⎥
Ys1 Ys2 . . . Yss ⎣ ⎦
where 1 0 0 ... −1 s×s
 
Glj −Blj Then, the relationship between EdvsD and EdvsQ is derived
Ylj = .
Blj Glj EdvsQ = Ted 2eq EdvsD (17)
2894 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 31, NO. 4, APRIL 2016

where
 −1
Ted 2eq = Tin (Tr e BNET ) Tr e GNET
and
⎡ ⎤
0 0 ... 0
⎢ ⎥
⎢ 1 0 ... 0 ⎥
⎢ ⎥
⎢ ⎥
Tr e =⎢
⎢ 0 1 ... 0 ⎥
⎥ . Fig. 3. Structure of the target networked system in this paper.
⎢ ⎥
⎢... ... ... ...⎥
⎣ ⎦ TABLE I
0 0 ... 0 s×(s−1) RELATED PARAMETERS OF THE TARGET MICROGRID SYSTEM

In (17), Ted2eq is a relation matrix determined by the equiva- Parameters Value

lent network. It reflects the relationship between DVS D and Q
Feeder impedance/km R f : 1 Ω, X f : 0 Ω
axis voltages. By substituting EdvsQ in (13) using (17), follow- Length of each feeder f1: 100 m, f2: 400 m, f3: 200 m, f4: 100 m
ing equation can be obtained: Power range of each load P: 0.9–4.5 kW, Q: 0.9–4.5 kVar
Rated power of each load P: 2.7 kW, Q: 2.7 kVar
IdvsQ = MNET EdvsD (18) Coupling inductor of each DG X c : 0.3 Ω
Initial virtual impedance R v : 0 Ω, X v : 0.2 Ω
where Nominal voltage of microgrid E l∗−l ( r m s ) : 190 V

MNET = BNET + GNET Ted 2eq .

In (18), MNET is the estimation matrix of network, which
the corresponding network can be called automatic power shar-
is also determined by the equivalent network. Considering that
ing network (APSN), which can realize accurate power sharing
in steady state, all the DVS D axis voltages are approximately
by only adopting the conventional droop control method. From
equal to the nominal voltage of microgrid E∗ , the Q axis current
the deduce process, it can be seen that Qer n etj % is determined
of each DVS can be estimated as
⎡ ⎤ ⎡ ∗⎤ by the equivalent network. When the physical network is def-
IdvsQ 1 E inite, by regulating virtual impedances properly, the network
⎢ ⎥ ⎢ ∗⎥
⎢ IdvsQ 2 ⎥ ⎢E ⎥ reactive power sharing performance can be improved.
⎢ ⎥ ⎢ ⎥
⎢ ⎥ ≈ MNET ⎢ ⎥ ≈ CAPSN E ∗ (19)
⎢ ... ⎥ ⎢...⎥
⎣ ⎦ ⎣ ⎦ III. VIRTUAL IMPEDANCE CONTROLLER DESIGN BASED ON
IdvsQ s E ∗ NETWORK FEATURE ANALYSIS

where A. Target System

To analyze the relationship between network parameters and
CAPSN = [M1 M2 ... Ms ]T .
reactive power sharing error concretely, a networked microgrid
In (19), Mi is the sum of the elements in row i of MNET , is introduced as the target system in this paper. Its structure is
the vector CAPSN composed of Mi is defined as the estimation illustrated in Fig. 3, with related parameters listed in Table I.
array in this paper, which can reflect the output reactive power of The prototype of the target system is a low-voltage microgrid
each DVS. To compare the reactive powers more conveniently, (three-phase three-wire ac system), which mainly serves the
the estimation array CAPSN can be modified as follows: lighting system in an office park. In low-voltage microgrids, as
sCAPSN distribution feeders are mainly resistive, the real and reactive
∗
CAPSN = s = [M1∗ M1∗ ... M1∗ ]T . (20) power coupling issue may occur, leading to the poor transient
i=1 Mi (or even poor stability) of the system. However, when coupling
Then, the reactive power sharing error of a single DG unit inductor is placed at the DG unit output, this phenomenon can be
can be estimated as follows: prevented. To further enhance the stability of system, the virtual
impedance method is also used. In this paper, the potential of
Qer dg i % = (M1∗ − 1) × 100%. (21)
virtual impedance method will be further dig out to solve the
Thus, the total reactive power sharing error of a network can reactive power sharing issue.
also be calculated accordingly

s B. Network Feature Analyses
Qer n etj % = |Qer dg i %|. (22) As previously stated, the reactive power sharing issue is as-
i=1
sociated with network factors, containing loads, feeders, and
In the equation, Qer n etj % reflects the reactive power sharing structure. For an established microgrid, the latter two factors are
capability of a certain network, and it is estimated from the net- definite, and only the load factor is active. So, the relationship
work matrix YNET . The estimation accuracy will be validated by between loads and system reactive power sharing performance
MATLAB simulation results. If Qer n etj % is relatively small, is discussed here.
ZHU et al.: VIRTUAL IMPEDANCE OPTIMIZATION METHOD FOR REACTIVE POWER SHARING IN NETWORKED MICROGRID 2895

Fig. 4. Qe r dg 1 % variation with different load changes.

Fig. 5. Comparison between k q x i and k q r i : E d v sd i = 159 V, V p n d i =
157 V, V p n q i = 3 V.
TABLE II
DG REACTIVE POWER SHARING ERROR CHANGE RATE TO THE DIFFERENT
LOAD CHANGE
For a single DG unit, its real and reactive powers can be
Load 1 Load 2 Load 3 Load 4 Load 5 calculated as follows:
Edvsdi Rf i − Vpn di Rf i + Vpn q i Xf i
ke p 1 i 0.62 –0.47 –0.20 0.34 –0.07 Pi = Edvsdi (23)
ke q 1 i –0.11 –0.06 –0.07 –0.09 –0.08 Rf2 i + Xf2 i

Edvsdi Xf i − Vpn di Xf i + Vpn q i Rf i

Qi = Edvsdi (24)
Fig. 4 shows the reactive power sharing error of DG unit 1 Rf2 i + Xf2 i
under different load combinations (termed microgrid operation where Vpn di and Vpn q i are the dq axis voltages of DG feeder
points in this paper). In the figure, Surface S1 depicts the value terminal in the reference frame of DVS i; Edvsdi is the d axis
variation of Qer dg 1 % according to the power change of load 1; voltage of DVS i. As the DG feeder is composed of coupling
meanwhile, the other loads keep their rated powers. inductor and virtual feeder, its impedance value is adjustable.
It can be observed that the load change impact on reactive From (23) and (24), the partial derivatives of Qdvsi with respect
power sharing error of a DG unit is nearly linear. And the change to Rf i and Xf i can be obtained as follows:
rates of Qer dg 1 % resulted from different load changes are com-
pared in Table II. The slope kep1 i represents the change rate of (Rf2 i − Xf2 i )(Edvsdi − Vpn di ) + 2Rf i Xf i Vpn q i
kq xi = Edvsdi
Qer dg 1 % according to the real power change of load i. During Rf2 i + Xf2 i
this stage, the load i reactive power and the other load powers (25)
keep their rated values. In a similar way, slope keq 1 i is obtained
accordingly.
(Rf2 i − Xf2 i )Vpn q i + 2Rf i Xf i (Edvsdi − Vpn di )
From the change rates listed in Table II, it can be concluded kq r i = Edvsdi
that for the given low-voltage system, reactive power sharing Rf2 i + Xf2 i
error of a DG unit is mainly related to the load real power change, (26)
while the load reactive power change has a much smaller impact. where kq xi and kq r i are partial derivatives according to reac-
Meanwhile, the real power change of local load has the most tive power, which can reflect the power regulation capability of
impact. With a similar analysis way, it also can be proved that virtual reactance and resistance. For example, kq xi reflects the
in a microgrid with mainly inductive feeders, the DG reactive reactive power change with the virtual reactance variation. If
power sharing error is mainly related to the load reactive power kq xi value is positive, that means the reactive power output of a
change, especially the reactive power change of local load. DG unit could be raised by increasing virtual reactance value.
The comparison between kq xi and kq r i under different virtual
impedance values is shown in Fig. 5. In most cases, to maintain
C. Control Variable Selection a favorable stability for the system, Xf i is designed larger than
Normally, virtual impedance contains two components: vir- Rf i .
tual resistance and virtual reactance. Both of them have the As shown in the figure, the virtual resistance value Rv i varies
ability to regulate DG reactive power; however, their regulation between 0.02 and 0.2 Ω, while the virtual reactance value Xv i
effects are different. For better power control, it is necessary to varies between 0.12 and 0.32 Ω. This range selection ensures
investigate their respective power regulation effects. a high X/R ratio for DG feeder. From the comparison, it can
2896 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 31, NO. 4, APRIL 2016

Fig. 7. Local control scheme based on the proposed method.

Fig. 6. Position change of S1 with different virtual resistance values.

be seen that kq r i value is negative definite, while kq xi value in

some areas is positive. Based on this, two main conclusions can
be obtained.
1) The reactive power regulation effects of virtual resistance
and reactance are mismatched. So, it is not proper to reg-
ulate them equally.
2) The regulation effect of virtual reactance may be reversed
due to the value change of virtual impedance. Thus, it Fig. 8. Introduction of the method process.
may be counterproductive in practice if virtual reactance
is selected to regulate reactive power.
So, in the proposed strategy, virtual reactance is fixed, while local load measurement is not needed. In fact, virtual impedance
virtual resistance is adjustable for reactive power regulation. By controller needs to be designed flexibly according to the actual
increasing (or decreasing) the virtual resistance value, the DG network features. The control scheme of a DG unit with the
reactive power can be lowered (or raised). designed virtual resistance controller is depicted in Fig. 7.
In the control scheme, local load measurement is added to
D. Virtual Impedance Controller Design facilitate the tuning virtual resistance. With the measured local
load real power Plo cal , the feeder transmitted power Pfeeder can
By tuning the virtual resistance in DG unit 1, the position of be obtained at the same time. The virtual voltage is calculated
Surface S1 depicted in Fig. 4 will be changed accordingly as as follows:
shown in Fig. 6.     
It can be seen that for a certain operation point, if virtual Udv i Rv i −Xv i Iodi
= (28)
resistance value is set proper, DG reactive power sharing error Uq v i Xv i Rv i Ioq i
can be eliminated to zero. And even a small variation in virtual
resistance value can lead to an obvious regulation effect. Thus, where Iodi and Ioq i are the dq axis output currents of DG unit.
the variation range of virtual resistance need not be set too wide. In steady state, each load power is constant, and the proposed
The figure also shows that the relationship between proper Rv control will work just as the conventional method with fixed
value and load power is approximate liner. Combining with the virtual impedance. When load changes in microgrid, virtual
conclusion in network feature analyses, the virtual resistance resistance will be tuned accordingly to adapt the new microgrid
control and power information can be associated as operation point. The value selection of each decision factor (fai ,
fbi , and fci ) is related to the network configurations, and will
Rv i = fai Plo cali + fbi Pfeederi + fci + fdi Qlo cali (27) be discussed in the following.
where Plo cali and Qlo cali are the measured real and reactive
powers of DG local load; Pfeederi is the real power transmitted IV. PARAMETER OPTIMIZATION USING GA
through DG feeder; fai , fbi , fci , and fdi are decision factors, To present the proposed method more clearly, the method
which are the variables to be optimized. Considering that the and steps are briefly illustrated in Fig. 8. As the figure shows,
guiding significance of load reactive power is relatively small for the proposed method contains two stages, and the main work is
the target system, decision factor fdi could be omitted. If local done in the first stage, containing virtual impedance controller
load is not connected, fai should also be omitted and even the design and parameter optimization. Step 2 has been discussed
ZHU et al.: VIRTUAL IMPEDANCE OPTIMIZATION METHOD FOR REACTIVE POWER SHARING IN NETWORKED MICROGRID 2897

previously. In the designed controller, virtual reactance is fixed,

while virtual resistance is adjustable. In this section, controller
parameters (decision factors) in each DG unit will be optimized
to improve the system reactive power sharing performance.

A. Objective Function of the Optimization

As DG units locate in different positions of the microgrid,
their respective network input characteristics are distinct. The
parameter optimization of each virtual impedance controller
should be treated differentially. But, before that, the related
optimization problem should be designed. In this paper, the
global reactive power sharing error of microgrid is set as the
optimization objective, which is defined as follows:


np
Qer global % = Qer n etj % (29)
j =1

where np is the number of the preset operation points. As can

be seen from (29), Qer global % provides an overview of the
reactive power sharing situation in microgrid, and its value, the
smaller the batter. Fig. 9. Function FQ (x): calculation of Q e r g lo b a l %.

At the beginning, each load is set with three real power opera-
tion points (1.5, 3 and 4.5 kW), and two reactive power operation
points (2 and 4 kVar). So, each load has six operation points, and
the system has total 6m (m is the load number) operation points
accordingly. This number is too large for Qer global % calcula-
tion, and in practice, the number may be even larger. However,
some operation points are not so valuable for the optimization.
To make the optimization more effective, the focus should be
on those operation points associated with local load change,
while the other load changes can be neglected properly. Finally,
there are five operation points for each local load and two for
each public load. Thus, total 500 operation points are selected
as the preset operation points for Qer global % calculation. The
calculation function FQ (x) based on the global reactive power
sharing error is established in MATLAB as illustrated in Fig. 9.
The output of function FQ (x) is the optimization objective
Qer global %, and the decision factors in each virtual resistance Fig. 10. Illustration of GA in MATLAB.
controller, which need to be optimized, are set as the inputs. In
a microgrid, different operation points correspond to different
networks. As the figure shows, by accumulating reactive power in this paper can be solved. Related configuration parameters
sharing errors of different networks, the global reactive power used in GA toolbox are listed in Table III.
sharing error value is obtained. When optimization algorithm Some other parameters keep their default values in MATLAB.
is utilized, the best values for these decision factors, which can The virtual impedance controller optimizations for simulation
minimize the output value Qer global %, will be obtained. and experiment systems are separate, and their optimization re-
sults are listed in Tables IV and VIII, respectively. With the opti-
B. Optimization Algorithm and Configuration mized virtual impedance controllers, the reactive power sharing
performance of the target system can be improved, and this will
To deal with such a nonlinear optimization problem, GA
be verified by simulation and experimental results.
is a good choice [32], [33]. GA is a search technique used
in computer, which mimics Darwinian natural selection. The
V. SIMULATION VERIFICATION
operation process of GA in MATLAB is described in Fig. 10.
When the network structure and feeder impedance values are To verify the effectiveness of the proposed control strategy,
known, corresponding FQ (x) can be established. By calling the as well as the reactive power sharing evaluation method, the
optimization toolbox in MATLAB, choosing GA as the solver microgrid system depicted in Fig. 3 is established in MATLAB
and FQ (x) as the objective function, the optimization problem simulation. The DG unit configurations and initial microgrid
2898 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 31, NO. 4, APRIL 2016

TABLE III
CORRESPONDING SETTING IN GA OPTIMIZATION

Key parameter setting in MATLAB GA tool

Population size 20 Scaling Rank

Elite count 2 Selection Stochastic uniform
Crossover fraction 0.8 Crossover Constraint dependent
Migration fraction 0.2 Mutation Scattered
Constraint condition (simulation system E∗ l −l = 190 V)
Inequality constraints 4500f a + f c ≤ 0.2Ω; 4500f b + f c ≥ −0.1Ω
Boundary conditions −3e−5 ࣘ fa , fb ࣘ 4.5e−5 ; −0.05 ࣘ fc ࣘ 0.1
Constraint condition (experimental system E∗ l −l = 95 V)
Inequality constraints 1100f a + f c ≤ 0.2Ω; 1100f b + f c ≥ −0.1Ω
Boundary conditions −2e −5 ≤ f a , f b ≤ 2e −5 ; −0.1 ≤ f c ≤ 0.2

TABLE IV
DG UNIT AND MICROGRID PARAMETERS USED IN MATLAB SIMULATION

DG unit parameters

LCL filter L f : 1.5 mH; C f : 60 μF; L o : 0.95 mH

Maximum power P m a x : 12 kW; Q m a x : 12 kVar
Drop slopes D P : 6e−3 rad/W; D Q : 2.5e−4 V/Var
Initial virtual impedance R v : 0.02 Ω; X v : 0.3 Ω Fig. 11. Simulated power sharing performance of the conventional fixed vir-
Rated voltage (rms) E l∗−l ( r m s ) : 190 V tual impedance method. (a) Real power. (b) Reactive power.
Initial microgrid operation point
Load 1 P: 3.8 kW; Q: 4.5 kVar; Zl : 4+4.7j Ω
Load 2 P: 3.0 kW; Q: 1.4 kVar; Zl : 10+4.7j Ω
Load 3 P: 3.6 kW; Q: 2.0 kVar; Zl : 8+4.7j Ω
Load 4 P: 3.7 kW; Q: 2.9 kVar; Zl : 6+4.7j Ω
Load 5 P: 3.7 kW; Q: 2.9 kVar; Zl : 6+4.7j Ω
Optimized virtual impedance controller
DG unit 1 R v : 2.77e −5 P l o c a l + 2.32e −5 P f e e d e r + 0.0752; X v : 0.3Ω
DG unit 2 R v : 4.43e −5 P l o c a l + 5.18e −6 P f e e d e r + 0.0005; X v : 0.3Ω
DG unit 3 R v : 3.48e −5 P l o c a l + 8.36e −7 P f e e d e r − 0.0283; X v : 0.3Ω

operation point are listed in Table IV. With the same power
rating, the three DG units shall share the load equally. Fig. 12. Simulated currents of DG units with the conventional method under
the initial operation point.

A. Performance of the Conventional Method

The performances of the system using the conventional droop power sharing performance, it can be seen that the network
control method are illustrated in Fig. 11. It is noted that fixed mismatch has not been compensated completely, leaving the
virtual impedances are added, or else, the reactive power sharing reactive power sharing issue unsolved.
performance would be even worse.
At t = 1.6 s, load 5 is disconnected from the network, and
the simulation results are consistent with the previous analysis. B. Performance of the Proposed Method
Fig. 11(a) shows that the real power sharing is always accurate. When DG units enable the optimized virtual impedance con-
However, due to the mismatch in network, there are significant trollers, the power sharing performance of the system under
errors in reactive power sharing as shown in Fig. 11(b). The different operation points is shown in Fig. 14. Compared with
voltage magnitudes of those inner DVS units can be reflected the performances shown in Fig. 11, its improvement in reactive
according to the Q–V droop control relationship. power sharing is significant.
Fig. 12 shows the associated DG currents under the initial The associated DG line currents are presented in Fig. 15. It
operation point. With the conventional method, the magnitude shows that these currents are almost equal after the optimized
and phase of DG currents are not the same. The output voltage virtual resistances being added, and the system current sharing
magnitudes of the three DG units are also obtained. performance is improved.
Fig. 13 shows the changes of DG voltage magnitudes dur- The voltage magnitudes of DG units are presented in Fig. 16.
ing this process. When reactive load decreases in the network, As the figure shows, there is a little voltage drop at each DG
DG output voltages are raised. From the presented reactive unit output due to the introduced virtual resistances. However,
ZHU et al.: VIRTUAL IMPEDANCE OPTIMIZATION METHOD FOR REACTIVE POWER SHARING IN NETWORKED MICROGRID 2899

Fig. 13. Simulated DG voltage magnitudes with the conventional method. Fig. 15. Simulated currents of DG units with the optimized virtual impedance
method under the initial operation point.

Fig. 16. Simulated DG voltage magnitudes with the proposed method.

Fig. 14. Simulated power sharing performance of the proposed control

method. (a) Real power. (b) Reactive power.

by raising the initial set-point of Q–V droop control, this issue

can be solved, and more important is that the system reactive Fig. 17. Virtual resistance values under different operation points.
power sharing performance is well improved. The associated
DG virtual impedance values are illustrated in Fig. 17.
With the conventional method, DG unit 2 always outputs less TABLE V
SYSTEM MATRIXES OF DIFFERENT EQUIVALENT NETWORK IN SIMULATION
reactive power due to its higher feeder impedance and also to
the smaller local load. However, with the proposed method, DG
GN E T BN E T
unit 2 will be given a smaller virtual resistance compared with
⎡ ⎤ ⎡ ⎤
the other two DG units, making its reactive power raised accord- 3.17 −0.99 −1.97 0.18 0.01 0.02
Physical network ⎣ −0.99 209 −0.99 ⎦ ⎣ 0.01 0.04 0.01 ⎦
ingly. Due to the adaptively tuned virtual impedances, reactive without virtual −1.97 −0.99 3.11 0.02 0.01 0.07
power sharing errors are limited in a small range. These virtual impedances ⎡ ⎤ ⎡ ⎤
impedance values are recorded and will be used in the verifi- 0.54 −0.26 −0.16 1.13 −0.41 −0.56
Network with fixed ⎣ −0.26 0.67 −0.25 ⎦ ⎣ −0.41 0.92 −0.42 ⎦
cation of the proposed reactive power sharing error estimation virtual impedances −0.16 −0.25 0.57 0.56 −0.42 1.11
method. ⎡ ⎤ ⎡ ⎤
0.69 −0.28 −0.27 0.84 −0.31 −0.39
Network with ⎣ −0.28 0.70 −0.27 ⎦ ⎣ −0.31 0.78 −0.36 ⎦
C. Verification of the Estimation Method optimized virtual −0.27 −0.27 0.69 −0.39 −0.36 0.88
impedances
The system matrixes of the initial network with different
virtual impedances are shown in Table V.
2900 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 31, NO. 4, APRIL 2016

TABLE VI
COMPARISON BETWEEN ESTIMATION AND SIMULATION RESULTS

Network with fixed Network with optimized

virtual impedances virtual impedance

Estimation C A∗ P S N [1.18 0.79 1.03] [1.03 0.95 1.01]

results
Qe r dgi % [18.3 − 20.8 2.5] [3.4 − 4.8 1.4]

Simulation Qi [5.37 3.63 4.65] [4.45 4.26 4.43]

results
Q i /Q a v e [1.18 0.80 1.022] [1.02 0.97 1.01]
Fig. 18. Q e r n e t j % comparison between the conventional method and pro-
Qe r dgi % [18.0 − 20.2 2.2] [1.6 − 2.7 1.1] posed method under different operation points.

TABLE VII
FEEDER IMPEDANCE CHANGES IN “VARIANT” MICROGRID SYSTEMS
From the change in GNET and BNET , it can be seen that,
at the beginning, the input admittance of each DG unit is large
Feeder 1 Feeder 2 Feeder 3 Feeder 4
and significantly different. And these mismatched character-
istics lead to the poor reactive power sharing performance of MG 0 0.1 Ω 0.4 Ω 0.2 Ω 0.1 Ω
MG 1 +15% +15% +15% +15%
system. When fixed virtual impedance method is applied, DVS
MG 2 −15% −15% −15% −15%
input admittances are decreased, and the match degree of net- MG 3 +10% −15% −10% +15%
work is improved relatively. But these fixed virtual impedances
cannot compensate the mismatch in network specifically. On
the other hand, the input admittance magnitude of each DVS
unit can be calculated from GNET and BNET . It will be found
that the DVS unit with larger input admittance value always
outputs more reactive power, and this phenomenon obeys the
laws of physics. With the system matrixes obtained from the
equivalent network, reactive power sharing can be analyzed
qualitatively. And with the further deduced evaluation array,
the reactive power sharing can even be estimated quantitatively.
The reactive power sharing error estimation results derived from
network matrixes are listed in Table VI.
In the table, Qav e is the averaged value of DG unit reactive Fig. 19.
power outputs, and Qi /Qav e is calculated for a better com- Performances of the optimized virtual impedance method while applied in mis-
∗
parison with CAPSN . As the aforementioned table shows, the matched systems.
proposed estimation method can well reflect the reactive power
sharing state of a microgrid network, and its accuracy is favor- is limited fewer than 20%, when optimized virtual impedance
able. Compared with simulation results, the estimation results controllers are utilized.
can be obtained much faster. As the estimation method has a
good accuracy, it is feasible to use the estimation results in the
E. Performance in Mismatched Systems
calculation of Qer global %, which serves the parameter opti-
mization of virtual impedance controller. To test the fault-tolerant capability of the proposed method,
the optimized virtual impedance controllers designed for the
original target system will be applied to several “variant” mi-
D. Performance Comparison With Different Operation Points crogrid systems, which are created from the original system by
To further compare the reactive power sharing performance changing feeder impedance values. The changed feeder param-
between the conventional method and proposed method, the eters of each “variant” system are listed in Table VII.
target microgrid will operate under more different operation In the table, MG 0 represents the original microgrid system,
points, and the corresponding comparison results are shown in and MG 1–3 represent the “variant” microgrid systems. In MG
Fig. 18. 1 and 2, feeder impedance values are enlarged and reduced
In the comparison, according to the load range, 30 opera- uniformly. While in MG 3, feeder impedance values are changed
tion points are randomly generated. As Fig. 18 shows, the net- irregularly. The performances of the original virtual impedance
work reactive power sharing error Qer n etj % is always high with controllers in different “variant” systems are shown in Fig. 19.
fixed virtual impedances, while the system using optimized vir- As can be seen from the figure, these virtual impedance
tual impedances has a much better performance. It also can be controllers, which are designed for the original target system,
seen that the microgrid network reactive power sharing error are also suitable for “variant” systems. The maximal network
ZHU et al.: VIRTUAL IMPEDANCE OPTIMIZATION METHOD FOR REACTIVE POWER SHARING IN NETWORKED MICROGRID 2901

Fig. 22. Experimental apparatus.

Fig. 20. DG currents of each phase under the proposed method in unbalanced TABLE VIII
load condition. DG UNIT AND MICROGRID PARAMETERS USED IN EXPERIMENT

DG unit parameters

LCL filter L f : 0.72 mH; Cf : 60 μF; L c : 0.95 mH

Maximum power P m a x : 3 kW; Q m a x : 3 kVar
Drop slopes DP : 1e−3 rad/W; DQ : 8e−4 V/Var
Initial virtual impedance R v : 0.02 Ω; X v : 0.5 Ω
Rated voltage (rms) E l∗−l ( r m s ) : 95 V
Optimized virtual impedance controller
DG unit 1 R v : 1.60e −4 P l o c a l + 1.10e −4 P f e e d e r + 0.02; X v : 0.5 Ω
DG unit 2 R v : 1.75e −4 P l o c a l − 1.74e −5 P f e e d e r − 0.08; X v : 0.5 Ω

and the presented performance of harmonic current sharing is

Fig. 21. DG currents with the proposed method under nonlinear load condi-
tion. acceptable.
Similar to the situation of reactive power sharing issue, mis-
matched feeder impedances in a microgrid also affect the im-
balance power and harmonic power sharing performance [34].
reactive power sharing error Qer n etj % is lower than 25%, which However, as the proposed virtual impedance method is estab-
is acceptable compared with the performance of conventional lished based on the optimization of network features, it also
methods. This result well proves that the proposed method has has some capability to solve the imbalance power and harmonic
some fault-tolerant capability to the line impedance errors. power sharing issues.

F. Special Load Condition VI. EXPERIMENTAL VERIFICATION

1) Unbalanced Load Condition: To test the unbalanced load Experiments are performed to validate the effectiveness of
sharing capability of the proposed method, loads in microgrid the proposed reactive power sharing method. A test microgrid
will be changed to create unbalanced load condition. To realize system is established in laboratory with two identical DG units
this, loads in phase-B are reduced by half. The related current- (DG units 1 and 2) as shown in Fig. 22.
sharing performance is shown in Fig. 20. The network structure and parameters can refer to Fig. 3.
It can be seen that due to the load change in phase-B, the Compared with the simulated microgrid, DG unit 3 is removed
three phase currents are distinct, and the currents in phase-B is from the system, and the theoretical parameters of the loads
smaller than that of the other two phases. However, the current are the same as that in simulation. The two DG units used in
in each phase is still well shared by the three DG units. the experiment are both controlled by DSP TMSF2812 chips;
2) Nonlinear Load Condition: To observe the harmonic load each DG unit is composed of a programmable power supplies
sharing performance of the proposed method, nonlinear loads (5 kVA) and a three-phase inverter. Detailed parameters of the
are added to the microgrid. Considering the voltage distortion DG main circuit and controller are listed in Table VIII.
issue in microgrid, the proportion of nonlinear power is limited
in 40%. In the simulation, loads 4 and 5 are replaced by two
A. Experimental Network Analysis
three-phase diode rectifiers, respectively. The phase-A currents
of the three DG units are compared in Fig. 21. It can be seen The theoretical system matrixes of the experimental network
that there are no significant differences among the DG currents, with different virtual impedances are shown in Table IX.
2902 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 31, NO. 4, APRIL 2016

TABLE IX
SYSTEM MATRIXES OF DIFFERENT EQUIVALENT NETWORK IN EXPERIMENTS

GN E T BN E T
   
1.93 −1.61 0.23 0.04
Physical network without virtual impedances
−1.61 1.77 0.04 0.06
   
0.28 −0.12 0.66 −0.45
Network with fixed virtual impedances
−0.12 0.32 −0.45 0.60
   
0.34 −0.17 0.57 −0.40
Network with optimized virtual impedances
−0.17 0.35 −0.04 0.58

Fig. 24. Experimental DG voltage and current waveforms with the conven-
tional control method. (a) Original figure. (b) Enlarge figure.

Fig. 23. Power sharing performance with the conventional control method in
experiment. (Before the load change, P1 , P2 are each 1800 W, and Q1 , Q2 are
2000 and 1300 Var, respectively; then, P1 , P2 drop to 1500 W, and Q1 , Q2 drop
to 1500 and 1000 Var, respectively.)

As the table shows, the system admittances are reduced when

virtual impedances added; meanwhile, the system match degree
is increased relatively, which can mitigate the reactive power
sharing issue to some extent. But the residual network mis-
match can still cause obvious reactive power sharing errors.
When optimized virtual impedance controllers are enabled, Fig. 25. Power sharing performance with the proposed control method in
the improved system matrixes can be calculated accordingly experiment. (Before the load change, P1 , P2 are each 1800 W, and Q1 , Q2 are
1800 and 1500 Var, respectively; then, P1 , P2 drop to 1500 W, and Q1 , Q2 drop
(with Rv 1 = 0.25 W and Rv 2 = 0.01 W ). From the results, it to 1300 and 1200 Var, respectively.)
can be seen that the DG input characteristics become closer,
which benefits the reactive power sharing performance of the
system.
C. Performance of the Proposed Method
When the optimized virtual impedance method is applied,
B. Performance of the Conventional Method reactive power sharing errors are reduced as Fig. 25 shows.
The performance of the system using the conventional droop Consistent with the network feature analysis, the optimized
control method is illustrated in Fig. 23. The load action in ex- virtual impedances can further improve the reactive power shar-
periment is the same as that in simulation. As can be seen from ing performance compared with fixed virtual impedances. Be-
the figure, due to the P–ω droop control, the real power sharing fore the adjustable virtual resistance introduced, DG unit 1 out-
is always accurate. However, due to the mismatch in network, puts more reactive power due to its larger local load, and also to
reactive power sharing errors are obvious. its lower feeder impedance. According to the optimized virtual
The DG line current and voltage waveforms corresponding impedance controller listed in Table IX, a larger virtual resis-
to Fig. 23 are obtained in Fig. 24(a), and the zoom-in current tance will be given to DG unit 1, which lowers the output voltage
waveforms are shown in Fig. 24(b). and reactive power relatively.
It can be seen that, due to the poor reactive power sharing The DG line current and voltage waveforms corresponding
performance, the corresponding current sharing of system is not to Fig. 25 are obtained in Fig. 26(a), and the zoom-in current
satisfactory. waveforms are shown in Fig. 26(b). It can be seen that the DG
ZHU et al.: VIRTUAL IMPEDANCE OPTIMIZATION METHOD FOR REACTIVE POWER SHARING IN NETWORKED MICROGRID 2903

droop-controlled parallel three-phase voltage source inverters,” IEEE

Trans. Ind. Electron., vol. 60, no. 4, pp. 1271–1280, Apr. 2013.
[5] Q. C. Zhong, “Robust droop controller for accurate proportional load
sharing among inverters operated in parallel,” IEEE Trans. Ind. Electron.,
vol. 60, no. 4, pp. 1254–1262, Apr. 2013.
[6] C. N. Rowe, T. J. Summers, R. E. Betz, D. J. Cornforth, and T. G. Moore,
“Arctan power-frequency droop for improved microgrid stability,” IEEE
Trans. Power Electron., vol. 28, no. 8, pp. 3747–3759, Aug. 2013.
[7] M. Savaghebi, A. Jalilian, J. C. Vasquez, and J. M. Guerrero, “Autonomous
voltage unbalance compensation in an islanded droop controlled mi-
crogrid,” IEEE Trans. Power Electron., vol. 60, no. 4, pp. 1390–1402,
Apr. 2013.
[8] M. A. Abusara, J. M. Guerrero, and S. M. Sharkh, “Line-interactive UPS
for microgrids,” IEEE Trans. Ind. Electron., vol. 61, no. 3, pp. 1292–1300,
Mar. 2014.
[9] U. K. Madawala, M. Neath, and D. J. Thrimawithana, “A power–frequency
controller for bidirectional inductive power transfer systems,” IEEE Trans.
Ind. Electron., vol. 60, no. 1, pp. 310–317, Jan. 2013.
[10] Y. Li and Y. W. Li, “Power management of inverter interfaced autonomous
microgrid based on virtual frequency-voltage frame,” IEEE Trans. Smart
Grid., vol. 2, no. 1, pp. 30–40, Mar. 2011.
[11] J. M. Guerrero, L. G. Vicuna, J. Matas, M. Castilla, and J. Miret, “Output
impedance design of parallel-connected UPS inverters with wireless load
sharing control,” IEEE Trans. Ind. Electron., vol. 52, no. 4, pp. 1126–1135,
Aug. 2005.
[12] J. M. Guerrero, L. G. Vicuna, J. Matas, M. Castilla, and J. Miret, “A
Fig. 26. Experimental DG current waveforms with the proposed control wireless controller to enhance dynamic performance of parallel inverters
method. (a) Original figure. (b) Enlarge figure. in distributed generation systems,” IEEE Trans. Power Electron., vol. 19,
no. 4, pp. 1205–1213, Sep. 2004.
[13] J. Kim, J. M. Guerrero, P. Rodriguez, R. Teodorescu, and K. Nam, “Mode
adaptive droop control with virtual output impedances for an inverter-
currents are almost identical after the application of optimized based flexible AC microgrid,” IEEE Trans. Power Electron., vol. 26,
no. 3, pp. 689–701, Mar. 2011.
virtual impedance controllers. [14] W. Yao, M. Chen, J. Matas, J. M. Guerrero, and Z.-M. Qian, “Design and
analysis of the droop control method for parallel inverters considering the
impact of the complex impedance on the power sharing,” IEEE Trans. Ind.
VII. CONCLUSION Electron., vol. 58, no. 2, pp. 576–588, Feb. 2011.
[15] J. M. Guerrero, J. Matas, L. G. De Vicuna, M. Castilla, and J. Miret,
A reactive power sharing strategy that employs optimized “Decentralized control for parallel operation of distributed generation
virtual impedance controllers and local load measurements is inverters using resistive impedance,” IEEE Trans. Power Electron.,
proposed and validated in this paper, as well as an estimation vol. 54, no. 2, pp. 994–1004, Apr. 2007.
[16] J. M. Guerrero, L. G. De Vicuna, J. Matas, M. Castilla, and J. Miret,
method for microgrid reactive power sharing. Based on the esti- “Output impedance design of parallel-connected UPS inverters with wire-
mation method, the relationship between reactive power sharing less load-sharing control,” IEEE Trans. Power Electron., vol. 52, no. 4,
and microgrid network is discussed, which is utilized to design pp. 1126–1135, Aug. 2005.
[17] Y. W. Li and C. N. Kao, “An accurate power control strategy for power
virtual impedance controller. With the help of GA, the decision electronics-interfaced distributed generation units operation in a low volt-
factors in each virtual impedance controller are optimized to age multibus microgrid,” IEEE Trans. Power Electron., vol. 24, no. 12,
minimize the global reactive power sharing error. By adopting pp. 2977–2988, Dec. 2009.
[18] J. He, Y. W. Li, J. M. Guerrero, F. Blaabjerg, and J. C. Vasquez, “An island-
the optimized virtual impedance controllers, DG units can well ing microgrid power sharing approach using enhanced virtual impedance
share the loads in networked microgrid. Through simulation re- control scheme,” IEEE Trans. Power Electron., vol. 28, no. 11, pp. 5272–
sults, it is also proved that the proposed method can share the 5282, Nov. 2013.
[19] X. Yu, A. M. Khambadkone, H. Wang, and S. T. S. Terence, “Control of
unbalanced power and harmonic power to some extent. In ad- parallel-connected power converters for low-voltage microgrid—Part I: A
dition, communication has not been used, making the proposed hybrid control architecture,” IEEE Trans. Power Electron., vol. 25, no. 12,
method more convenient to be applied; local load measurement pp. 2962–2970, Dec. 2010.
[20] A. M. Roslan, K. H. Ahmed, S. J. Finney, and B. W. Williams, “Im-
can be removed if there is no local load. proved instantaneous average current-sharing control scheme for parallel-
connected inverter considering line impedance impact in microgrid net-
REFERENCES works,” IEEE Trans. Power Electron., vol. 26, no. 3, pp. 702–716, Mar.
2011.
[1] M. A. Zamani, T. Sidhu, and A. Yazdani, “Investigations into the control [21] J. W. He and Y. W. Li, “An enhanced microgrid load demand sharing
and protection of an existing distribution network to operate as a microgrid: strategy,” IEEE Trans. Power Electron., vol. 27, no. 9, pp. 3984–3995,
A case study,” IEEE Trans. Ind. Electron., vol. 61, no. 4, pp. 1904–1915, Sep. 2012.
Apr. 2014. [22] H. Han, Y. Liu, Y. Sun, M. Su, and J. M. Guerrero, “An improved droop
[2] J. W. He, Y. W. Li, and F. Blaabjerg, “Flexible microgrid power control strategy for reactive power sharing in islanded microgrid,” IEEE
quality enhancement using adaptive hybrid voltage and current con- Trans. Power Electron., vol. 30, no. 6, pp. 3133–3141, Jun. 2015.
troller,” IEEE Trans. Ind. Electron., vol. 61, no. 6, pp. 2784–2794, [23] Y. X. Zhu, F. Zhuo, B. Q. Liu, and H. Yi, “An enhanced load power
Jun. 2014. sharing strategy for low-voltage microgrids based on inverse-droop control
[3] M. Sechilariu, B. C. Wang, and F. Locment, “Building integrated photo- method,” in Proc. Int. Power Electron. Conf. ECCE Asia, Hiroshima,
voltaic system with energy storage and smart grid communication,” IEEE Japan, 2014, pp. 3546–3552.
Trans. Ind. Electron., vol. 60, no. 4, pp. 1607–1618, Apr. 2013. [24] Q. Shafiee, J. M. Guerrero, and J. C. Vasquez, “Distributed secondary
[4] J. C. Vasquez, J. M. Guerrero, M. Savaghebi, J. Eloy-Garcia, and R. control for islanded microgrids—A novel approach,” IEEE Trans. Power
Teodorescu, “Modeling, analysis, and design of stationary reference frame Electron., vol. 29, no. 2, pp. 1018–1031, Feb. 2014.
2904 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 31, NO. 4, APRIL 2016

[25] Y. X. Zhu, F. Zhuo, and H. T. Shi, “Accurate power sharing strategy for Feng Wang (S’08–M’13) received the B.S., M.S.,
complex microgrid based on droop control method,” in Proc. IEEE Conf. and Ph.D. degrees in electrical engineering from
ECCE Asia, Downunder, 2013, pp. 344–350. Xi’an Jiaotong University, Xi’an, China, in 2005,
[26] T. L. Vandoorn, J. D. M. De Kooning, B. Meersman, and L. Vandevelde, 2009 and 2013, respectively.
“Communication-based secondary control in microgrids with voltage- Between 2010 and 2012, he was an exchang-
based droop control,” in Proc. IEEE Transmiss. Distrib. Conf. Expo., ing Ph.D. student in the Center for Power Elec-
2012, pp. 1–6. tronics Systems at Virginia Polytechnic Institute and
[27] J. M. Guerrero, J. C. Vasquez, J. Matas, L. G. De Vicuna, and M. State University (Virginia Tech), Blacksburg, USA.
Castilla, “Hierarchical control of droop-controlled ac and dc microgrids— In November 2013, he joined Xi’an Jiaotong Univer-
A general approach toward standardization,” IEEE Trans. Ind. Electron., sity as a Postdoctoral Fellow, where he is currently
vol. 58, no. 1, pp. 158–172, Jan. 2011. with the State Key Laboratory of Electrical Insulation
[28] M. Savaghebi, A. Jalilian, J. C. Vasquez, and J. M. Guerrero, “Secondary and Power Equipment, School of Electrical Engineering. His research interests
control scheme for voltage unbalanced compensation in an islanded droop include dc/dc conversion, digital control of switched converters, especially in
controlled microgrid,” IEEE Trans. Smart Grid, vol. 3, no. 2, pp. 797–807, renewable energy generation fields.
Jun. 2012.
[29] H. Mahmood, D. Michaelson, and J. Jiang, “Accurate reactive power
sharing in an islanded microgrid using adaptive virtual impedances,” IEEE
Trans. Power Electron., vol. 30, no. 3, pp. 1605–1617, Mar. 2015.
[30] Y. Zhang and H. Ma, “Analysis of networked control schemes and data-
processing method for parallel inverters,” IEEE Trans. Ind. Electron.,
vol. 61, no. 4, pp. 1834–1844, Apr. 2014.
[31] Y. Zhang and H. Ma, “Theoretical and experimental investigation of net-
worked control for parallel operation of inverters,” IEEE Trans. Ind. Elec- Baoquan Liu (S’13) received the B.S. degree in elec-
tron., vol. 59, no. 4, pp. 1961–1970, Apr. 2012. trical engineering from the Harbin Institute of Tech-
[32] M. B. Shadmand and R. S. Balog, “Multi-objective optimization and nology, Harbin, China, in 2009, and the M.S. de-
design of photovoltaic-wind hybrid system for community smart dc gree from Xi’an Jiaotong University, Xi’an, China, in
microgrid,” IEEE Trans. Smart Grid, vol. 5, no. 5, pp. 2635–2643, 2012, where he is currently working toward the Ph.D.
Sep. 2014. degree in the State Key Laboratory of Electrical In-
[33] S. Jazebi, M. M. Haji, and R. A. Naghizadeh, “Distribution network re- sulation and Power Equipment, School of Electrical
configuration in the presence of harmonic loads: optimization techniques Engineering.
and analysis,” IEEE Trans. Smart Grid, vol. 5, no. 4, pp. 1929–1937, His research interests include system operation
Apr. 2012. and energy management of microgrids, especially hy-
[34] J. W. He, Y. W. Li, and F. Blaabjerg, “An enhanced islanding microgrid brid energy storage systems in microgrids.
reactive power, imbalance power, and harmonic power sharing scheme,”
IEEE Trans. Power Electron., vol. 30, no. 6, pp. 3389–3401, Jun. 2015.

Yixin Zhu (S’11) received the B.S. and M.S. degrees

in electrical engineering from Xi’an Jiaotong Univer-
sity, Xi’an, China, in 2009 and 2011, respectively, he
is currently working toward the Ph.D. degree under
the direction of Prof. Fang Zhuo. Ruifeng Gou received the B.S. degree in electrical
He is currently with the State Key Laboratory of engineering from Xi’an Jiaotong University, Xi’an,
Electrical Insulation and Power Equipment, School China, in 1982. Now he is with the Xi’an High Volt-
of Electrical Engineering, Xi..an Jiaotong Univer- age Appliance Research Institute, Xi’an, China, as a
sity. His research interests include the design, control, Senior Engineer.
and application of the high power active power filter, His research interests include system operation
the photovoltaic grid-connected inverter, and also the and power management of microgrids, especially the
modeling, analysis, and power management of the microgrid. microgrid mode switch techniques.

Fang Zhuo (M’00) was born in Shanghai, China, in

May 1962. He received the B.S. degree in automatic
control and the M.S. and Ph.D. degrees in automation
and electrical engineering all from Xi’an Jiaotong
University, Xi’an, China, in 1984, 1989, and 2001,
respectively.
In 2004, he was a Visiting Scholar with Nanyang
Yangjie Zhao received the B.S. and M.S. degrees in
Technological University. He was an Associate Pro-
electrical engineering from Xi’an Jiaotong Univer-
fessor with Xi’an Jiaotong University in 1996, and
sity, Xi’an, China, in 2006 and 2009, respectively.
a Full Professor in power electronics and drives in He is currently with the Xi’an High Voltage Ap-
2004. Then, he was a Supervisor of Ph.D. student.
pliance Research Institute, Xi’an. His research inter-
He is also the Associate Dean of the Faculty of Industry Automation. His re-
ests include system operation and energy control of
search interests include power electronics, power quality, active power filter,
microgrids, especially the microgrid connection tech-
reactive power compensation, inverters for distributed power generation, etc.
niques.
Dr. Zhuo is a member of China Electro Technical Society, Automation So-
ciety and Power Supply Society. Also, he is the Power Quality Professional
Chairman of Power Supply Society in China.