132 IEEE TRANSACTIONS ON SMART GRID, VOL. 9, NO.

1, JANUARY 2018

Stability Analysis and Stabilization Methods of DC

Microgrid With Multiple Parallel-Connected
DC–DC Converters Loaded by CPLs
Mei Su, Zhangjie Liu, Yao Sun, Member, IEEE, Hua Han, and Xiaochao Hou

Abstract—Constant power loads may yield instability due to or nonlinear analysis method because the CPL is a typical
the well-known negative impedance characteristic. This paper nonlinear load.
analyzes the factors that cause instability of a dc microgrid with The first category focuses on stability analysis and stabi-
multiple dc–dc converters. Two stabilization methods are pre-
sented for two operation modes: 1) constant voltage source mode; lization methods for a single converter with CPLs [2]–[9].
and 2) droop mode, and sufficient conditions for the stability It’s found that the ways by increasing damping and reduc-
of the dc microgrid are obtained by identifying the eigenvalues ing negative impedance can mitigate the main bus volt-
of the Jacobian matrix. The key is to transform the eigenvalue age oscillation or collapse [2], [3]. The interconnection and
problem to a quadratic eigenvalue problem. When applying the damping assignment technique has been used for stabil-
methods in practical engineering, the salient feature is that the
stability parameter domains can be estimated by the available ity analysis of a DC microgrid [2]. Reference [4] reviews
constraints, such as the values of capacities, inductances, max- four tools for large signal stability analysis, which includes
imum load power, and distances of the cables. Compared with multi-model approach, Brayton-Moser’s mixed potential [5],
some classical methods, the proposed methods have wider stabil- block diagonalized quadratic Lyapunov function and reverse
ity region. The simulation results based on MATLAB/simulink trajectory tracking. The sliding-mode control [6], nonlinear
platform verify the feasibility of the methods.
feedback [7], phase-plane analysis [8] and Popov criterion [9]
Index Terms—Constant power loads (CPLs), dc microgrid, are also used to solve this problem. To obtain the stability
multi-converter, quadratic eigenvalue problem (QEP), stability. region, a LMI approach is proposed in [10].
The second type considers the stability of two parallel
converters [11]–[17]. Amplitude death solutions are utilized
I. I NTRODUCTION for stabilization of DC microgrid with instantaneous CPLs
ECENTLY, DC microgrid has attracted more and more in [11]. The method based on nonlinear feedback is applied
R attention due to its highlight advantages compared with
AC microgrid: high efficiency, easy control and strong robust-
into two parallel converters in [12]. Synergetic control has
been proposed in [13], which is based on modern mathemat-
ness. As a result, distributed DC generation systems with mul- ics and essential properties of nonlinear dynamic dissipative
tiple DC-DC converters are increasingly used in applications systems. And the special different characters of [12] and [13]
such as aircrafts, spacecrafts and electric vehicles [1]. are compared in [14]. Since increasing damping can enhance
In the DC microgrid, the distributed generations (DGs) are stability, several linear methods based on virtual impedance
usually connected to the DC bus through DC-DC converters are proposed in [15]–[17]. Comparing with [15], the methods
and loaded by various loads. However, it may lead to insta- in [16] and [17] add a voltage recovery term. Comparing with
bility when CPLs are involved. To cope with this issue, many linear methods [15]–[17], nonlinear methods [11]–[13] have
studies have been carried out in terms of modeling, stability a wider stability region. But it need to feedback the size of
analysis, and control strategies [2]–[26]. They can be divided CPLs by communication, which will cause extra cost.
into three categories from the perspective of the number of The third group extends the application and complexity
converters. Moreover, they can also be classified into the linear of the system with n numbers of converters [18]–[26]. The
stability of a reduced-order model of low-voltage DC micro-
Manuscript received August 22, 2015; revised December 8, 2015 grid is analyzed in [18], and the stable ranges of the droop
and January 31, 2016; accepted March 1, 2016. Date of publication coefficients is obtained. Similarly, a reduced-order approx-
March 24, 2016; date of current version December 21, 2017. This work imate model is proposed to predict the system qualitative
was supported in part by the National Natural Science Foundation of
China under Grant 61321003, in part by the National High-Tech Research and behavior in [19], which simplifies the system as a two-
Development Program of China (863 Program) under Grant 2015AA050604, order RLC model to obtain the safe operating regions. But
and in part by the Central South University Innovation-Driven Project. Paper the models in [18] and [19] only work on the droop mode
no. TSG-00985-2015. (Corresponding author: Yao Sun.)
The authors are with the School of Information Science and and ignore the dynamic performances of DC-DC convert-
Engineering, Central South University, Changsha 410083, China (e-mail: ers. In order to predict the dynamic performances of DC-DC
[email protected]). converters, some stability criteria are proposed. The stabil-
Color versions of one or more of the figures in this paper are available
online at http://ieeexplore.ieee.org. ity of the general case of n converters supplied by one
Digital Object Identifier 10.1109/TSG.2016.2546551 source is analyzed in [20]. And the stability criteria are
1949-3053 c 2016 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.
SU et al.: STABILITY ANALYSIS AND STABILIZATION METHODS OF DC MICROGRID 133

developed based on Brayton and Moser’s nonlinear circuit

theory. Similarly, the stability criteria for CPL with multistage
LC filters have been developed based on Brayton and Moser’s
theory [21]. A design method for fault-tolerant stabilizing sys-
tem is proposed in [22]. The salient advantage of the method
is converting the fault-tolerant stabilizing to a constrained
optimization problem which can be solved by LMI. The sta-
bility of multi-converter medium voltage DC (MVDC) power
systems on ships is analyzed in [23]–[26]. In these works,
a linearizing state feedback (LSF) control is proposed and
the stability is analyzed based on order reduction [23], [24].
Reference [25] firstly adapts the backstepping technique to the
MVDC shipboard power systems. The validity of LSF control
in a FGPA hardware is tested in [26].
This paper considers a DC power system which contains n
buck converters. Compared with the shipboard MVDC power Fig. 1. The structure of the DC microgrid.
system in [23] and [24], the distance between converters and
loads is short, so the cable impedance is neglectable. However,
in this paper, the distance is relatively long in microgrid 2) The resistance of the common bus is zero, hence all
because the sources are distributed and far away from the loads can be regarded as one common CPL.
loads. So, the cable impedance can’t be neglected. It leads 3) The pure resistance cable. The impact of cable
that the system in this paper is not able to reduce order. impedance on stability is analyzed in [19]. The results
This paper proposes two linear ways to stabilize the sys- indicate that the cable inductance has no influence on
tem loaded by CPLs, which can operate in two modes, the the system stability and the cable resistance has influ-
constant voltage source mode and droop mode. For the first ence on the system stability. For simplicity, the cable
mode, the classical control method is PI control in [16], impedance is regarded as pure resistance.
however, the stability region is difficult to obtain in a high 4) The load is an ideal CPL. If the response of output reg-
order system. The stability region is obtained in this paper. ulating controllers of POL converters is fast enough,
For the second mode, the proposed control method can not the POL converters attached load can be regarded as
only stabilize the system, but also realize power sharing a CPL [2].
among distributed generators (DGs). The paper has following For the i-th converter, the dynamics of the converter can be
contributions. described by the differential equations as
1) The location of eigenvalues of the load equivalent ⎧
⎪ di
impedance matrix is analyzed. ⎨ Li Li = Vi di − ui
dt (i ∈ {1, 2, · · · , n}) (1)
2) Compared with PI control in [16], the stability region of ⎪
⎩ Ci dui = iL − ii
the proposed constant voltage control method is wider, dt i

especially when the system order is high.

3) For the droop mode, stabilization and power sharing can where the subscript i represents the i-th converter, Vi and di
be realized simultaneously. What’s more, this paper can are the input voltage and the duty cycle, respectively. The ui ,
obtain a more accurate solution than the corollary of the ii and iLi represent the output voltage , current and inductance
Theorem 5 in [5]. current, respectively.
4) The salient feature of this paper is that the stability Rewriting (1) in the matrix form, the system is given by
⎧
conditions can be estimated by the available constraints. ⎪ di
⎨ L L = Vd − u
The rest is organized as follows. Section II presents the dt (2)
mathematical model of the multi-converter DC microgrid ⎪
⎩C
du
= iL − i
loaded by a CPL. The stability analysis and stabilization dt
methods for two modes are introduced in Section III. The sim-
where u=[u1 u2 . . . un ]T , i=[i1 i2 . . . in ]T , d=[d1 d2 . . . dn ]T ,
ulation results are provided in Section IV. Final conclusions
and V = diag{Vj }, C = diag{Cj }, L = diag{Lj }, j ∈
are drawn in Section V.
{1, 2, . . . , n}.
For a constant power load, the power balance equation is
II. T HE DC M ICROGRID AND BASIC M ODELS as follows.
 n
The DC microgrid with multiple parallel DC-DC convert- u0 i=1 ii = P
ers loaded by CPL is shown in Fig. 1. It mainly consists of (3)
ii = gi (ui − u0 )
three components: sources, loads and cables. And it is modeled
based on the following assumptions: where u0 represents the voltage of the load, P is the power of
1) The sources are buck converters whose inputs can be the CPL and gi represents the conductance of the cable from
regarded as ideal voltage sources. the i-th converter to the load.
134 IEEE TRANSACTIONS ON SMART GRID, VOL. 9, NO. 1, JANUARY 2018

III. S TABILITY A NALYSIS AND S TABILIZED M ETHODS After linearization of equation (2), we have
⎧
Lemma 1 [27]: If λ1 ≤ λ2 ≤ · · · ≤ λn are the eigenvalues ⎪
⎨ L dîL = −û
of a real symmetric matrix A, and β1 ≤ β2 ≤ · · · ≤ βn are the dt (7)
eigenvalues of a real symmetric matrix B, then ⎩ C dû = îL − ∂i û
⎪
λi + β1 ≤ ηi ≤ λi + βn dt ∂u
where “∧” represents the small-signal variation around the
where η1 ≤ η2 ≤ · · · ≤ ηn are the eigenvalues of matrix A+B. operating point of the system. Similarly, with linearizing
(Proof in [27]). equation (3), the following equations are obtained.
Lemma 2 [28]: If A is a real symmetric n-by-n matrix and ⎧  
D is a positive definite real symmetric n-by-n matrix. Then ⎨ n gi ûi − û0 = −  P û0
i=1
u∗0
2
1) AD (or DA) is diagonalizable. (8)
⎩  
2) The eigenvalues of AD (or DA) are real number, and îi = gi ûi − û0
AD (or DA) has the same number of negative (zero, or Then the partial derivative matrix is obtained
positive) eigenvalues with matrix A.
∂i 1
Proof: If D is a positive definite and symmetric matrix, = n
∂u 
gi − P
then there is a nonsingular real symmetric matrix Q to meet i=1
u∗2
0
⎡  
that Q2 =D. Note that M=QADQ−1 =QAQ, so M is a sym- g1
n
gi − P∗2 − g1 −g1 g2 ··· −g1 gn
⎤
⎢ u ⎥
metric matrix and has the same eigenvalues with AD. As M ⎢
⎢
i=1 0
 n

 ⎥
⎥
is real symmetric, M can be written as M=PP−1 , where ⎢
⎢ −g 2 g1 g2 gi − P
∗2
u0
− g2 ··· −g2 gn ⎥
⎥
⎢ i=1 ⎥
=diag{λ(M)} and P is the related orthogonal matrix. Then ⎢
⎢ .
.
.
. .. .
.
⎥
⎥
⎢ . ⎥
it can be inferred that (QP)AD(QP)−1 =, so the conclusion ⎢
⎣
. . 

n
. ⎥
⎦
1) is proved. As Q is real symmetric, M is congruent to A. −gn g1 −gn g2 · · · gn gi − P
− gn
u∗2
0
i=1
According to Sylvester’s law of inertia, M has the same index (9)
of inertia with A. The conclusion 2) is proved [27].
Lemma 3: Let Q(λ) = λ2 M+λK+C. Assume that M, K and where g0 = −P(u∗0 )−2 , which is the equivalent conductance
∂i
C are all Hermitian n-by-n matrix, for a quadratic eigenvalue of CPL. Let Y  ∂u , and Y is the admittance matrix of the
problem as network, which is a real symmetric matrix. By linearizing
  equation (3), the follow can be obtained
 2   n
λ M + λK + C = 0 (4)
i=1 îi = − u∗ 2 û0
P
( 0) (10)
If M, K and C are all positive definite, then Re (λ)<0.
ûi = îi r + û0
Proof: Note Q(λ) = λ2 M + λK + C. Let m(x) = xH Mx,
c(x) = xH Cx and k(x) = xH Kx, where x ∈ Cn is nonzero. If x Then the impedance matrix can be obtained
⎧
then xH Q(λ)x = 0, and the roots are given
is an eigenvector, Z=R
by λ = (−c(x) ± c(x)2 − 4m(x)k(x))/2. Because M, K and
⎪
⎪
⎪ ⎡ + R0 ⎤ ⎡ ⎤
⎪
⎪ r1 11 ··· 1
⎨
C are all positive definite, then Re(λ)<0 [28]. ⎢ r2 ⎥ ⎢1 1 ··· 1⎥
⎢ ⎥ ⎢ ⎥ (11)
⎪
⎪ R=⎢ .. ⎥, R0 = g10 ⎢ .. .. .. .. ⎥
⎪
⎪ ⎣ . ⎦ ⎣. . . .⎦
A. Stability Analysis ⎪
⎩
rn 11 ··· 1
The stability of the system with only one converter and
no cable resistance has been analyzed in [1]–[7] and [29]. where ri = g1i is the cable resistance and Z = Y −1 is the
However, for the multi-converter system, the conductance of impedance matrix. Considering that
the cable can’t be neglected. To date, there isn’t a perfect and n
general solution for such a system. The characteristic of equiv- λ1 (R0 ) = , λ2 (R0 ) = · · · = λn (R0 ) = 0 (12)
g0
alent impedance matrix of the network with the CPL has never
According to Lemma 1, the region of the eigenvalues of
been analyzed. These issues will be analyzed in this section.
matrix Z can be obtained
If the duty cycle di∗ is constant, the converters will work in
the mode of constant output voltage. Then, according to the λi (R0 ) + min{ri } ≤ λi (Z) ≤ λi (R0 ) + max{ri } (13)
state equation (2) of the buck converter, the steady operation
point can be obtained by As the load resistance is far larger than the cable’s (the
conductance is just the opposite), then
u∗ = Vd∗ (5) n
λ1 (Z) ≤ + max{ri } < 0 (14)
From the equation (3), the steady-state voltage of the load g0
has two solutions, a high voltage or a low voltage, but the According to equation (12), (13), a corollary below can be
fixed low voltage is unstable [30]. Thus, the voltage of load obtained.
in steady state is described as Corollary 1: Matrix Y has one negative eigenvalue and n-1
n
  positive eigenvalues.
g u∗+ n
gi u∗i − 4P ni=1 gi
i=1 i i i=1 σ1 is the minimal eigenvalue of matrix Y and gm = max{gi }
u∗0 =  (6)
2 ni=1 gi is the maximal conductance. With the Gerschgorin’s disk
SU et al.: STABILITY ANALYSIS AND STABILIZATION METHODS OF DC MICROGRID 135

theorem [28], σ1 satisfies damping to overcome instability and oscillation. However,

gm g0 increasing damping will lead to voltage sags. Therefore, it’s
n ≤ σ1 < 0 (15) necessary to find new method. For this issue, this paper pro-
i=1 gi + g0
poses a linear method in which only current of capacitances
Because g0 is negative and greatly less than the cable’s con- needs to be fed back. Then, the duty cycle is transformed as
ductance, the condition (15) can be approximately rewritten as  
gm g0 d = V −1 L−1 Ca V ∗ − kiC − u + V −1 u (22)
0 > σ1 ≥ n (16)
i=1 gi where iC represents the current through the capacitance, k is
the feedback coefficient, a is a relaxed parameter which can
With the linearized differential equations (7), (8) and (9),
influence the range of k, and V ∗ = v∗ [ 1 1 · · · 1 ]T is the refer-
the Jacobian matrix can be obtained
  ence voltage. With equation (22), the system is reconstructed.
0 −L−1 According to the Kirchhoff’s Current Law, it yields
J= (17)
C−1 −C−1 Y
iC = iL − i (23)
In principle, the stability of the system could be analyzed by
From (2), (22) and (23), the system becomes
inspecting the eigenvalues of the Jacobian matrix J. Usually, ⎧
it is difficult to obtain the analytical expression for the eigen- ⎪ 1 di
⎨ C−1 L = V ∗ − kiC − u
values of matrix J. However, it is possible to assess stability a dt (24)
under some special conditions. If C−1 L−1 = kI, where k is ⎪
⎩C
du
= iL − i
an arbitrary positive parameter, whether J is Hurwitz or not dt
can be judged. As Y is real-symmetric and has one negative Equation (24) has two advantages. Firstly, it’s helpful to
eigenvalue, with Lemma 2, C−1 Y is diagonalizable and has analyze the stability by compensating the asymmetrical com-
the same number of negative eigenvalues with matrix Y. Note ponent of the system parameters, such as, conductors and
that μ1 ≥ μ2 ≥ · · · ≥ μn (μn < 0) are the eigenvalues of capacitors. Secondly, the value of conductors and capacitors
C−1 Y, and  = diag{μi } is the related diagonal matrix. The are usually very small (10−6 ∼10−3 ), the relaxed parameter
eigenvalues of J are obtained as can be used to adjust the stability region and response rate.
  The stabilization mechanism of the system is that iC can influ-
 
|λI − J| = λ2 I + λC−1 Y + kI  = 0 (18) ence the transient process of the system without changing the
steady state.
Equation (18) is a quadratic eigenvalue problem (QEP). In the steady-state, all the state variables will satisfy
Because the quadratic term’s coefficient matrix (I) is non-  ∗
V − kiC − u = 0
singular, the characteristic polynomial of equation (18) is an (25)
iL − i = iC = 0
implicit 2n-order polynomial [28]. Only the 2n-order poly-
nomial can be decoupled into n quadratic equations. As the Solving (25) gives
matrix λ2 I + λC−1 Y + kI is diagonalizable, the analytic solu- u = V∗ (26)
tions can be obtained. Since C−1 Y is diagonalizable, there is
a nonsingular matrix P1 to meet that The stabilization mechanism of the system is that iC can
      influence the transient process of the system without changing
 −1 2   
P1 λ I + λC−1 A + kI P1  = λ2 I + λ + kI  = 0 (19) the steady state (as equation (26)). From (6) and (26), the
steady-state u∗0,A can be calculated
The 2n-order polynomial equation can be decomposed into
   n 
n ∗
n quadratic equations as
∗ v∗ 1 i=1 gi v − 4 i=1 gi P
u0,A = + n (27)
λ2 + λμi + k = 0, i ∈ {1, 2, · · · , n} (20) 2 2 i=1 gi

As μn < 0, J has at least two positive eigenvalues which According to the equation (8) (9) and (24), the correspond-
include a positive real part ing Jacobian matrix is obtained
 

 akC aC − akCY
J1 = − (28)
λ2n−1.2n = −μn ± μ2n − 4k /2 (21) −C−1 C−1 Y
The characteristic polynomial of matrix J1 is given by
Equation (21) indicates that the system may be unstable.    
 
So it is necessary to take measures to enhance the stability. |λI − J1 | = λ2 I + λ akC + C−1 Y + aI  = 0 (29)
Moreover, the DC microgrid needs more than one operating Equation (29) is also a QEP, which has high similarity with
mode to adapt to the varying environments. To this end, two equation (18). Thus, a similar algorithm can be used to solve it.
stabilization methods have been proposed in the following Note that M = akC+C−1 Y, N = akI+C−1 YC−1 , then M=NC.
sections. As N is a real symmetric matrix and C is a positive diagonal
matrix, according to Lemma 2, M is diagonalizable, whose
B. Stabilization Methods of the Basic Model eigenvalues are all real. Equation (29) can be diagonalized by
 
1) Mode A (Constant Voltage Source): Based on the above  2 
λ I + λ1 + I  = 0 (30)
analyses, the converter may be unstable without any measures.
Usually, the effective and convenient method is increasing where 1 = diag{λ(M)}.
136 IEEE TRANSACTIONS ON SMART GRID, VOL. 9, NO. 1, JANUARY 2018

If diagonal matrix 1 is positive definite, the system is sta- has the same function with that in (22). From (2) and (32),
ble. In other words, only if matrix M is Hurwitz, the system is the state equations operating in droop mode is obtained
stable. Considering that M = NC, C is real symmetric positive ⎧
⎪ 1 di
⎨ L L = V ∗ − KiL − u
definite and N is real symmetric, so when N is positive definite,
a dt (37)
M is Hurwitz. Then a sufficient condition which maintains the ⎪
⎩C
du
= iL − i
system stable is obtained dt
  
ak > − min λ C−1 YC−1 (31) When the system is in steady state, it yields iL = i.
So, in steady state, the relation between output voltage and
Note σ2 =min{λ(C−1 YC−1 )}. With Sylvester’s law of current is
inertia [27], C−1 YC−1 has the same index of inertia with u = V ∗ − Ki (38)
matrix Y, so σ2 < 0. According to the Rayleigh’s mini-max
principle [27], σ2 can be written as Equation (38) indicates that ki would act as an output
resistance in steady state, which will share the load power pro-
xT Yx portionally. According to (3) and (34), the equilibrium (u∗0,B )
σ2 = min (32)
xT x =0 xT C2 x can be obtained

Consider that n   n   2 n  
 gi
v∗ + gi
v∗ − 4P gi
min xT Yx = σ1 xT x 1+gi ki 1+gi ki 1+gi ki
(33) i=1 i=1 i=1
min xT C2 x = min Ci2 xT x u∗0,B = n  
 gi
and σ1 < 0 (in equation (16)), it can be inferred that 2 1+gi ki
i=1
σ1 xT x σ1 (39)
σ2 ≥  2 =   (34)
min Ci x x T min Ci2 Usually, the droop coefficient ki is more than the line
Finally, the stability condition (31) can be rewritten as resistance (g−1
i ), then the result of equation (35) can be
σ1 approximately treated by
k>−   (35)  " n $−1
a min Ci2 ∗ #
∗ v 1 ! ∗2
From above analysis, there exists a value of k to guarantee u0,B ≈ + v − 4P ki (40)
2 2
the system stability in the constant voltage source mode. In i=1

general, the constant voltage source mode can be applied to If the equilibrium state exits, it will meet that
the following cases: " n $−1
# P
1) The high-quality voltage demand of loads. For example, ki ≤ ∗2 (41)
the constant voltage charging of non-inverter loads. 4v
i=1
2) The small power rating of loads. The small power
According to (8) (9) and (37), the Jacobian matrix can be
demand doesn’t need power sharing among the sources.
obtained
3) The concentrated distribution of all sources. The cable  −1 
resistances are almost equal, which the power sharing is aL K aL−1
J2 = − (42)
satisfied automatically. −C−1 C−1 Y
4) The small change of the CPL. The voltage of the According to Theorem 5 in [5], if the equilibrium state
common bus is almost constant the change of the CPL. exits and
However, the constant voltage source mode isn’t always    %
1 1 1 ki
−2 −2
applicable in the practical scenarios of the microgrid, where λmin C YC + min >0 (43)
a Li
the micro-sources are distributed, and the power rating of the
loads are big and changed frequently. So, it’s necessary to for all iL and uC , then all the solutions of (37) approach the sta-
take measures to improve power sharing accuracy to avoid ble equilibrium state as t → ∞. In this paper, a more accurate
overloading of a converter. stability conditions are obtained as
   %
2) Mode B (Droop Control): In microgrid, power sharing 1 1 1 ki
is often taken into account. In mode A, all output voltages are λmin C− 2 YC− 2 + max >0 (44)
a Li
constant in the steady state, so it can’t realize power sharing
Proof: Note that
and may lead to ineffective source consumption. Droop con- ⎡ ⎤
& ' & ' 1
trol is an effective optimal control strategy in the condition of I I aL−1 K aL−1 C− 2 ⎦
avoiding communications. If the droop control is incorporated, J3 = 1 J2 1 = −⎣
C− 2
1 1 1
the duty cycles are C2 −C− 2 C− 2 YC− 2
  (45)
d = V −1 aV ∗ − aKiL − au + V −1 u (36)
where J3 and J2 is isospectral [27]. The sufficient and nec-
where K = diag{ki } and ki is the droop coefficient of i-th
  the system stable is that J3 is
essary condition of keeping
converter. The ki can be regarded as a virtual resistance in
series with filter conductor. The relaxed parameter a in (36) Hurwitz. Note q = max Lkii . Assume that q lies in m-th row
SU et al.: STABILITY ANALYSIS AND STABILIZATION METHODS OF DC MICROGRID 137

and m-th column in matrix L−1 K. As matrix C−1/2 YC−1/2 is According to the Rayleigh’s mini-max principle [28], μn
real symmetric, there is an orthogonal matrix P2 to make that can be written as
     
I I aL−1 K aL−1 C−1/2 P2 xT Yx σ1
J4 = J3 = μn = min ≥ (54)
P2 PT2 −P2 C−1/2 2 x x =0
T x T Cx min{Ci}
(46) To sum up, the stability region of the system under mode B
where 2 is a diagonal matrix, and the minimal element lies is shown in (41) and (53).
in m-th row and m-th column. Obviously, J4 is also isospectral
with J2 and J3 . The characteristic polynomial of matrix J4 can C. Design of the Coefficients and Stabilization Measures
be obtained by According to the stability conditions (27), (31) and (44), it
  is found that the critical point of the coefficients depend on
 
|λI − J4 | = λ2 I + λQ1 + Q2  = 0 (47) matrix L, C and A. Clearly, L and C are easy to be obtained
for all DGs. However, the conductance of lines and the power
where Q1 and Q2 are both real symmetric, and satisfies
 of CPL are difficult to know. So it’s necessary to estimate the
Q1 = aL −1
 −1K + 2  (48)
feasible region by using the information which is easy to be
Q2 = a L K2 + P2 L−1 C−1 PT2 obtained.
For mode A, the stability conditions is written as (35),
Equation (40) is also a real and symmetric QEP. Note that together with (16), the stability condition (35) can be given
Q(λ) = λ2 I + λQ1 + Q2 . According to Lemma 3, if Q1 , Q2
are both positive definite, then Re(λ) < 0. The system will be gm g0
ak > −    n (55)
stable, and the stability conditions can be written as min Ci2 i=1 gi
 −1
aL K + 2 > 0 In conditions (55), only the value of Ci is the local
(49)
L−1 K2 + P2 L−1 C−1 PT2 > 0 information. It is difficult to obtain the information of the
cable’s conductance, load’s voltage and power without com-
If all the parameters are available, then inequalities (42) munications. For the lines with the same material and cross
can be solved by Linear Matrix Inequality (LMI). However, sectional area, the resistances of them are proportional to the
because not all parameters are available, LMI will fail in this length. Neglecting other tiny factors, it can be approximately
situation. Then, only the conservative sufficient solutions of treated as
equation (44) are given by " n $−1
⎧ gm #  
⎨ aL−1 K + 2 > 0 n ≈ li−1 max li−1 (56)
1 (50) i=1 gi i=1
⎩ L−1 K2 + I>0
max{Li Ci } where li represents the length of lines of the i-th DG. The
1 1 instantaneous power of load is difficult, but the maximal power
Since C−1 Y and C− 2 YC− 2 are isospectral, matrix C−1 Y of load (Pm ) can be estimated by the statistical data. According
has one negative and n-1 positive eigenvalues, and to (27), the load’s voltage should satisfy that
 

1 1   
μn = λmin C− 2 YC− 2 (51) v ∗ 1
n
i=1 gi v∗ − 4 ni=1 gi P v∗
u∗0,A = + n ≥ (57)
2 2 i=1 gi 2
Where μn is the minimal eigenvalue of matrix C−1 Y.
Then conditions (50) can be expressed as Then, it can be inferred that
⎧  % P 4Pm
⎪
⎪ ki g0 = − ≥ − ∗2 (58)
⎨ a max L + μn > 0
⎪ u∗2
0,A
v
i
 % (52)
⎪
⎪ ki 1 According to (47), (50), (56), and (58), the stable region can
⎪
⎩ max μn + >0
Li max{Li Ci } be obtained by using the limited available information. Since
that
Simplifying (47), it yields  
 % gm g0 max li−1 −g0
μn ki 1 −  2  n ≈ n −1 ·  2
− < max <− (53) max Ci
a Li max{Li Ci }μn i=1 gi i=1 li  min Ci

−1
The proof is accomplished. max li 4Pm
Remarks: By Brayton and Moser Criterion, the stability ≤ n −1   (59)
∗2
v min Ci2
region is presented in (43), where minimal value of all Lkii must i=1 li

be bigger than the value − μan . However, this paper obtains Equation (55) can be rewritten as
a wider stability region that the maximal value of all Lkii must  
max li−1
be bigger than the value − μan . If the droop coefficient is k > n −1
4Pm
  (60)
smaller, the maximal of the DG’s output power is higher. ∗2
a i=1 li v min Ci2
138 IEEE TRANSACTIONS ON SMART GRID, VOL. 9, NO. 1, JANUARY 2018

Remark: According to (60), the stability region can be

adjusted by choosing a proper relaxed parameter a.
For mode B, similarly, only Ci , Li can be easily obtained.
According to conditions (16), (53) (54) and (57), it can be
obtained as follow
μn −σ1 −g0 gm
− ≤ ≤ 
a a min{Ci } a min{Ci } ni=1 gi
 
4Pm gm 4Pm max li−1
≤  ≈  (61)
min{Ci }av∗2 ni=1 gi min{Ci }av∗2 ni=1 l−1
n i
1 1 min{Ci } 1 min{Ci } i=1 gi 1
− ≥ ≥
max{Li Ci } μn max{Li Ci } −σ1 max{Li Ci } gm −g0
n n −1
min{Ci } v∗2 i=1 gi ≈ min{Ci } v
∗2
i=1 l Fig. 2. The simulink structure of the DC microgrid.
≥  i 
max{Li Ci } 4Pm gm max{Li Ci } 4Pm max l−1 TABLE I
i
T HE S IMULINK PARAMETERS OF THE DC M ICROGRID
(62)
Then, the stability conditions (53) can be written as
⎧  
⎪
⎪  % max li−1
⎪
⎪ k 4P
⎪ i m
⎨ max Li > av∗2 min{Ci } n l−1
⎪
i=1 i
 % n −1 (63)
⎪
⎪ ∗2 min{C }
⎪
⎪
ki v i i=1 li
⎪ max <  
⎪
⎩ Li 4Pm max{Li Ci } max l−1
i

It should be noted that the stability condition shown in (63)

is the subset of (53) after the conservative estimates in mode B.
Similarly, the stability condition shown in (60) is the subset
of the equation (31) in mode A. Hence, if (41) and (63) holds,
the system will be stable.
From the stability conditions (60) and (63), there are several
measures which can enhance the stability.
1) Increasing the value of filter inductance and capacity.
2) Improving the voltage of common bus. When it doesn’t take any stabilization measure, the system
3) Increasing the number of distributed generators. based on the basic model is tested. According to the Section II,
4) Decreasing the power of CPL. the state equations of this system are given by
⎧
⎪ diL
⎪
⎪ Li 1 = v∗ − ui
D. Comparison Analysis ⎪
⎪ dt
⎪
⎪
1) Compared With Czieskis Work in [10]: The control law ⎨ du1
Ci = iLi − ii , i = 1, 2, 3 (64)
of the buck converter in [10] contains the non-local infor- ⎪ dt
⎪
⎪
mation. DG1 needs the voltage and current of DG2. So, it ⎪
⎪ ii = gi (ui − u0 )
⎪
⎪
will cause an extra cost with communication. However, the ⎩ 
proposed control doesn’t need. u0 ii = P
2) Compared With Magnes Work in [12]: If the informa- In the equation (53), the nonlinear ingredient in constraint
tion of cable’s resistance is available, the method in [12] is equations makes the differential-algebraic equations difficult
feasible. However, in practice it is difficult to obtain this to solve. So it is necessary to simulate by Matlab/Simulink.
parameter accurately. In the proposed method, the stabil- The results are given in Fig. 3.
ity region can be obtained as equation (63) without the The simulation results from Fig. 3 show that the voltages
information of the cable’s resistance. are oscillating, which confirm the instability of CPL. These
3) Compared With Brayton Moser Criterion in [5]: oscillations are caused by energy imbalances that occurred
The stability region of this study is wider than that during the transient period when LC input and output powers
obtained by Brayton Moser Criterion (detail in above, equa- are not equal in steady state. So it is necessary to take measures
tion (43) and (53)). to enhance the stability. The performance of the system in two
modes will be tested in the following.
IV. S IMULATION
To verify the aforementioned analyses, the DC microgrid A. Mode A: Constant Voltage Source
with three DGs and one CPL are carried out in Fig. 2, and the In mode A, the control method needs to feedback the cur-
parameters of the system are given in Table I. rent information of the capacitances. With the equation (18),
SU et al.: STABILITY ANALYSIS AND STABILIZATION METHODS OF DC MICROGRID 139

Fig. 3. Transient response of voltages without stabilization measure. Fig. 4. Transient response of voltages in mode A.

the duty cycle of the i-th converter should be written as

follows
Li Ci  ∗    ui
di = a v − k iLi − ii − ui + (65)
Vi Vi
Then the state equation can be written as (20). According
to the stability conditions (46) and (49), the feedback gain k
should satisfy
gm 4Pm
k>   2  n ∗2
(66)
a min Ci i=1 gi v
From Table I, the parameters can be obtained as
  # Fig. 5. Transient response of output power in mode A.
min Ci2 = 1.6 × 10−7 , gm = 2, gi = 3.5,
Pm = 3000, v∗ = 500 B. Mode B: Droop Control
Substitute them into (55), it can be written as In this mode, the control method should get feedback on
the current of inductances. With the equation (32), the duty
ak > 1.7 × 105 (67) cycle of the DC-DC converters are given by
Then, take a = 105 , then the stability region can be  
di = Vi−1 av∗ − aKiL − au + Vi−1 u, i = 1, 2, 3 (69)
obtained as
Then the state equation can be written as (33). According
k > 1.7 (68)
to the parameters in Table I, max{Ci } = 4 × 10−4 , Pm =3000,
Take k=5, and the system will be stable. The simulate results max{Li Ci } = 2×10−7 , v∗ = 500, min{Li } = 2×10−4 stability
are followed in Fig. 4 and Fig. 5. From 0s∼0.5s, the system conditions (61) and (62) can be calculated as
doesn’t take any stabilization measures, which indicate that ⎧  %
⎪
⎪ ki 30
the output voltage and power are divergently oscillating. But, ⎨ a max >
once the stabilization of mode A is adopted from 0.5s, the  % Li 7 (70)
⎪
⎪ ki
system is stable. The power of load is changed from 2kW to ⎩ max < 6.67 × 105
Li
3kW at t=2s and restored at t=4s. The response curves indicate
that the system is stable, and it can keep stable as long as that Take a = 0.01, and require that all the DGs share the same
load power is less than the upper bound. Moreover, the system power, take k1 = k2 = k3 . Then the stability region can be
has a well dynamic traceability. expressed as
These curves verify that when the system operates in con- 0.085 < k1 < 133.4 (71)
stant voltage source mode, the performance of output voltage
is great, but the contribution of each DG depends on the line Take k1 = k2 = k3 = 10, the system will be stable
impedance and may be inefficient. So, the constant voltage and meet power sharing. The simulate results are followed in
source mode can be applied to the scenes, such as, the high- Fig. 6 and Fig. 7. From 0∼0.5s, the system doesn’t take any
quality voltage demand of loads and the concentrated sources stabilization measures, which indicates that the output voltage
spread. However, it can’t realize power sharing and may lead and power are divergently oscillating. But, once the stabiliza-
to ineffective source consumption. To realize energy opti- tion of mode B is adopted from 0.5s, the system is stable.
mization, power sharing should be taken into account. Droop The power of load is changed from 2 kW to 3 kW at t=2s and
control is a very effective optimal control strategy without restored at t=4s. The response curves indicate that the system
communications. So, it’s adopted in mode B. is stable, and it can keep stable as long as that load power is
140 IEEE TRANSACTIONS ON SMART GRID, VOL. 9, NO. 1, JANUARY 2018

Fig. 8. Transient response of output voltage of DG1 in mode B.

Fig. 6. Transient response of voltages in mode B.

Fig. 9. Transient response of output voltages with normal switching

frequency in mode A.

Fig. 7. Transient response of output power in mode B.

less than the upper bound. Moreover, the system has a well
dynamic traceability.
These curves verify that when the system operates in droop
control mode, the range of output voltage change is suit-
able, and the accuracy of power sharing is greatly enhanced
compared with the Fig. 5.
Fig. 10. Transient response of output voltages with low switching frequency
It should be noted that the above simulations are based on in mode A.
the average modeling inside the Matlab/Simulink to verify the
effectiveness of two stabilization methods in mode A and B.
Therefore, the switching converters are missing. In order to test
the proposed control solution with a low switching frequency,
following simulations are carried out in consideration of the
switching converters. The simulink structure and parameters
of the DC microgrid are also seem to the above simulink in
Fig. 2 and Table I, respectively.

C. The Wider Stability Regions Than That in

Brayton and Moser Criterion
In order to verify the wider stability regions in this paper Fig. 11. Transient response of output voltages with normal switching
frequency in mode B.
than that in Brayton and Moser Criterion, the case that
k1 = 5, k2 = k3 = 0 is carried out based on the equation (53)
under the condition of normal switching frequency ( fnor =10k).
If the system is still stable in this condition, equation (53) can transient state; 2) in steady state, the output voltage maintains
be verified. Fig. 8 shows the output voltage of DG1 to present around 500V; 3) the higher switching frequency, the more ideal
clearly the variations in mode B. the results are.

D. Mode A With the Normal and Low Switching Frequency E. Mode B With the Normal and Low Switching Frequency
Three points can be summarized in figures 9 and 10 as Compared with the Mode A in Fig. 9 and Fig. 10, the
follows: 1) during 0 to 0.001s, the system starts operation on voltages of mode B are slightly sag in Fig. 11 and Fig. 12.
SU et al.: STABILITY ANALYSIS AND STABILIZATION METHODS OF DC MICROGRID 141

Fig. 12. Transient response of output voltages with low switching frequency Fig. 15. Transient response of output currents with non-ideal CPL in mode B.
in mode B.
an effective linear method without the voltage sags is presented
for the 2n-order system. For the second mode, this paper
proposes modified droop method based on a more accurate 2n-
order model and the dynamic properties of DC-DC converters,
which can identify system stability and enhance power sharing
accuracy among converters. Compared with Brayton Moser
Criterion, this paper updates the infimum of the stability region
by using QEP theory, so this paper has wider stability region.
Mostly, the salient feature of two methods is that the stability
conditions can be estimated by the available limited informa-
tion. The views of this paper are verified through simulations
Fig. 13. Transient response of output voltages with non-ideal CPL in mode A. under different switching frequencies and non-ideal CPL.

R EFERENCES
[1] H. Kakigano, Y. Miura, and T. Ise, “Low-voltage bipolar-type DC micro-
grid for super high quality distribution,” IEEE Trans. Power Electron.,
vol. 25, no. 12, pp. 3066–3075, Dec. 2010.
[2] A. Kwasinski and C. N. Onwuchekwa, “Dynamic behavior and stabiliza-
tion of DC microgrids with instantaneous constant-power loads,” IEEE
Trans. Power Electron., vol. 26, no. 3, pp. 822–834, Mar. 2011.
[3] S. Sanchez and M. Molinas, “Assessment of a stability analysis tool for
constant power loads in DC-grids,” in Proc. 15th Int. Power Electron.
Motion Control Conf. (EPE/PEMC), Novi Sad, Serbia, Sep. 2012,
pp. DS3b.2-1–DS3b.2-5.
[4] D. Marx, P. Magne, B. Nahid-Mobarakeh, S. Pierfederici, and B. Davat,
Fig. 14. Transient response of output voltages with non-ideal CPL in mode B. “Large signal stability analysis tools in DC power systems with constant
power loads and variable power loads—A review,” IEEE Trans. Power
Electron., vol. 27, no. 4, pp. 1773–1787, Apr. 2012.
[5] R. K. Brayton and J. K. Moser, “A theory of nonlinear networks—I,”
Fig. 12 also presents the effectiveness in low switching Quart. Appl. Math., vol. 22, no. 1, pp. 1–33, Apr. 1964.
frequency of the converter. [6] Y. Zhao, W. Qiao, and D. Ha, “A sliding-mode duty-ratio controller for
DC/DC buck converters with constant power loads,” IEEE Trans. Ind.
Appl., vol. 50, no. 2, pp. 1448–1458, Mar./Apr. 2014.
F. Mode A and Mode B With Non-Ideal CPL [7] A. M. Rahimi, G. A. Williamson, and A. Emadi, “Loop-cancellation
The CPL in the simulations above is ideal. In reality, the technique: A novel nonlinear feedback to overcome the destabilizing
effect of constant-power loads,” IEEE Trans. Veh. Technol., vol. 59,
ideal CPL is not available, especially when the switching fre- no. 2, pp. 650–661, Feb. 2010.
quency of the CPL is not high enough. The recent study shows [8] C. Rivetta, A. Emadi, G. A. Williamson, R. Jayabalan, and B. Fahimi,
that ideal CPL is not the worst case [31]. Therefore, the pro- “Analysis and control of a buck DC-DC converter operating with con-
stant power load in sea and undersea vehicles,” IEEE Trans. Ind. Appl.,
posed method under non-ideal CPL [31] will be tested in this vol. 42, no. 2, pp. 559–572, Mar./Apr. 2006.
section. [9] D. Karimipour and F. R. Salmasi, “Stability analysis of AC microgrids
Fig. 13, Fig. 14 and Fig. 15 shows that the proposed with constant power loads based on Popov’s absolute stability criterion,”
IEEE Trans. Circuits Syst. II, Exp. Briefs, vol. 62, no. 7, pp. 696–700,
method is also valid. The voltage stability and power shar- Jul. 2015.
ing are guaranteed. In the transient response, the overshoots [10] P. Magne, B. Nahid-Mobarakeh, and S. Pierfederici, “DC-link voltage
in Fig. 13 and Fig. 14 are a little larger than those in large signal stabilization and transient control using a virtual capacitor,”
in Proc. IEEE Ind. Appl. Soc. Annu. Meeting (IAS), Houston, TX, USA,
Fig. 10 and Fig. 11, respectively. Oct. 2010, pp. 1–8.
[11] S. R. Huddy and J. D. Skufca, “Amplitude death solutions for stabiliza-
tion of DC microgrids with instantaneous constant-power loads,” IEEE
V. C ONCLUSION Trans. Power Electron., vol. 28, no. 1, pp. 247–253, Jan. 2013.
This paper proposes two stabilization methods for two oper- [12] J. G. Ciezki and R. W. Ashton, “The design of stabilizing controls
for shipboard DC-to-DC buck choppers using feedback linearization
ation modes: the constant voltage source mode and droop techniques,” in Proc. IEEE 29th Annu. Power Electron. Spec. Conf.,
mode in DC microgrid. For the first mode, this paper presents Fukuoka, Japan, May 1998, pp. 335–341.
142 IEEE TRANSACTIONS ON SMART GRID, VOL. 9, NO. 1, JANUARY 2018

[13] I. Kondratiev, E. Santi, R. Dougal, and G. Veselov, “Synergetic control Mei Su was born in Hunan, China, in 1967. She
for DC-DC buck converters with constant power load,” in Proc. IEEE received the B.S., M.S., and Ph.D. degrees from
35th Annu. Power Electron. Spec. Conf., Aachen, Germany, Jun. 2004, the School of Information Science and Engineering,
pp. 3758–3764. Central South University, Changsha, China, in 1989,
[14] M. Cupelli, M. Moghimi, A. Riccobono, and A. Monti, “A compari- 1992, and 2005, respectively.
son between synergetic control and feedback linearization for stabilizing Since 2006, she has been a Professor with the
MVDC microgrids with constant power load,” in Proc. IEEE PES School of Information Science and Engineering,
Innov. Smart Grid Technol. Europe (ISGT), Istanbul, Turkey, Oct. 2014, Central South University. Her research interests
pp. 1–6. include matrix converter, adjustable speed drives,
[15] X. Lu et al., “Stability enhancement based on virtual impedance for DC and wind energy conversion system.
microgrids with constant power loads,” IEEE Trans. Smart Grid, vol. 6,
no. 6, pp. 2770–2783, Nov. 2015.
[16] M. Srinivasan and A. Kwasinski, “Decentralized control of a vehicular
microgrid with constant power loads” in Proc. IEEE Int. Elect. Veh.
Conf. (IEVC), Florence, Italy, Dec. 2014, pp. 1–8.
Zhangjie Liu received the B.S. degree in detection
[17] A. Kwasinski and P. T. Krein, “Stabilization of constant power loads
guidance and control techniques from Central South
in DC-DC converters using passivity-based control,” in Proc. 29th Int.
University, Changsha, China, in 2013, where he is
Telecommun. Energy Conf. (INTELEC), Rome, Italy, Sep./Oct. 2007,
currently pursuing the Ph.D. degree in control engi-
pp. 867–874.
neering.
[18] S. Anand and B. G. Fernandes, “Reduced-order model and stability anal-
His research interests include renewable energy
ysis of low-voltage DC microgrid,” IEEE Trans. Ind. Electron., vol. 60,
systems, distributed generation, and dc micro-grid.
no. 11, pp. 5040–5049, Nov. 2013.
[19] A. P. N. Tahim, D. J. Pagano, E. Lenz, and V. Stramosk, “Modeling
and stability analysis of islanded DC microgrids under droop con-
trol,” IEEE Trans. Power Electron., vol. 30, no. 8, pp. 4597–4607,
Aug. 2015.
[20] M. Belkhayat, R. Cooley, and A. Witulski, “Large signal stability cri-
teria for distributed systems with constant power loads,” in Proc. IEEE
Annu. 26th Power Electron. Spec. Conf. (PESC), vol. 2. Atlanta, GA, Yao Sun (M’13) was born in Hunan, China, in 1981.
USA, Jun. 1995, pp. 1333–1338. He received the B.S., M.S., and Ph.D. degrees from
[21] X. Liu, Y. Zhou, W. Zhang, and S. Ma, “Stability criteria for constant the School of Information Science and Engineering,
power loads with multistage LC filters,” IEEE Trans. Veh. Technol., Central South University, Changsha, China, in 2004,
vol. 60, no. 5, pp. 2042–2049, Jun. 2011. 2007, and 2010, respectively. He is currently an
[22] P. Magne, B. Nahid-Mobarakeh, and S. Pierfederici, “Dynamic consider- Associate Professor with the School of Information
ation of DC microgrids with constant power loads and active damping Science and Engineering, Central South University.
system—A design method for fault-tolerant stabilizing system,” IEEE His research interests include matrix converter,
J. Emerg. Sel. Topics Power Electron., vol. 2, no. 3, pp. 562–570, micro-grid, and wind energy conversion system.
Sep. 2014.
[23] G. Sulligoi, D. Bosich, L. Zhu, M. Cupelli, and A. Monti, “Linearizing
control of shipboard multi-machine MVDC power systems feeding con-
stant power loads,” in Proc. Energy Convers. Congr. Expo. (ECCE),
Raleigh, NC, USA, 2012, pp. 691–697.
[24] G. Sulligoi et al., “Multiconverter medium voltage DC power systems Hua Han was born in Hunan, China, in 1970. She
on ships: Constant-power loads instability solution using linearization received the M.S. and Ph.D. degrees from the School
via state feedback control,” IEEE Trans. Smart Grid, vol. 5, no. 5, of Information Science and Engineering, Central
pp. 2543–2552, Sep. 2014. South University, Changsha, China, in 1998 and
[25] M. Cupelli, M. Mirz, and A. Monti, “Application of backstepping 2008, respectively. She was a Visiting Scholar at the
to MVDC ship power systems with constant power loads,” in Proc. University of Central Florida, Orlando, FL, USA,
IEEE Elect. Syst. Aircraft Railway Ship Propul. Road Vehicles (ESARS), from 2011 to 2012. She is currently an Associate
Aachen, Germany, 2015, pp. 1–6. Professor with the School of Information Science
[26] M. Cupelli, M. de Paz Carro, and A. Monti, “Hardware in the loop and Engineering, Central South University.
implementation of linearizing state feedback on MVDC ship systems Her research interests include microgrid, renew-
and the significance of longitudinal parameters,” in Proc. IEEE Elect. able energy power generation system, and power
Syst. Aircraft Railway Ship Propul. Road Vehicles (ESARS), Aachen, electronic equipment.
Germany, 2015, pp. 1–6.
[27] C. D. Meyer, Matrix Analysis and Applied Linear Algebra. Philadelphia,
PA, USA: Soc. Ind. Appl. Math., 2000.
[28] F. Tisseur and K. Meerbergen, “The quadratic eigenvalue problem,”
Soc. Ind. Appl. Math. Rev., vol. 43, no. 2, pp. 235–286, 2001.
[29] A. Khaligh, “Realization of parasitics in stability of DC–DC converters Xiaochao Hou received the B.S. degree in automa-
loaded by constant power loads in advanced multiconverter automotive tion from Central South University, Changsha,
systems,” IEEE Trans. Ind. Electron., vol. 55, no. 6, pp. 2295–2305, China, in 2014, where he is currently pursuing the
Jun. 2008. M.S. degree in electrical engineering.
[30] J. W. Simpson-Porco, F. Dörfler, and F. Bullo, “Voltage stabilization His research interests include renewable energy
in microgrids via quadratic droop control,” in Proc. IEEE 52nd Annu. systems, distributed generation, and micro-grid.
Conf. Decis. Control, Florence, Italy, Dec. 2013, pp. 7582–7589.
[31] M. Cupelli, L. Zhu, and A. Monti, “Why ideal constant power loads are
not the worst case condition from a control standpoint,” IEEE Trans.
Smart Grid, vol. 6, no. 6, pp. 2596–2606, Nov. 2015.