Liu 2017
Liu 2017
fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2017.2697765, IEEE
Transactions on Power Systems
1
Abstract—This paper studies stability analysis of DC micro- In [12], Nyquist stability criteria is used to determine the
grids with uncertain constant power loads (CPLs). It is well stability of a given equilibrium. In [17], the authors utilize
known that CPLs have negative impedance effects, which may the Jacobian linearization method and analyze the eigenvalues
cause instability in a DC microgrid. Existing works often study
the stability around a given equilibrium based on some nominal of the linearized system. Others approximate the nonlinear
values of CPLs. However, in real applications, the equilibrium DC microgrid system with linear systems [8], [15]. In [8],
of a DC microgrid depends on the loading condition that often the nonlinearity is treated as a bounded noise so that a DC
changes over time. Different from many previous results, this microgrid can be modeled by a linear system with additive
paper develops a framework that can analyze the DC microgrid uncertainty. In [15], fuzzy modeling method is adopted to ap-
stability for a given range of CPLs. The problem is formulated
as a robust stability problem of a polytopic uncertain linear proximate the nonlinear system with a series of linear systems.
system. By exploiting the structure of the problem, we derive With the resulting linear systems, estimations of the region
a set of sufficient conditions that can be efficiently checked of attraction around some known equilibria can be obtained
by solving convex optimization problems to guarantee locally by iteratively exploiting quadratic Lyapunov functions. Still
robust stability. The effectiveness and non-conservativeness of others have devised different forms of Lyapunov-like functions
the proposed framework are demonstrated using simulation
examples. using Brayton-Moser method and Bregman storage functions
to solve problems like local stability analysis and fair power
Index Terms—DC microgrids, uncertain constant power loads, sharing [16], [18].
robust stability, convex optimization problems.
Despite the rich literature on DC microgrids, existing works
have several limitations. First, existing stability analysis meth-
I. I NTRODUCTION ods are based on a fixed equilibrium and require to know its
exact value a priori [8], [12]–[16], [16], [17]. However, in
A DC microgrid is a direct current power network that con-
real applications, the system equilibrium depends on uncertain
sists of locally-controlled sources and loads [1]–[3]. Directly
CPLs and changes over time [19], [20]. Second, existing
connecting DC components allows for a simple integration of
works mostly consider single-bus DC microgrids with parallel
renewable generations and reduces unnecessary power conver-
components [8], [10], [13], [15]. It is unclear how the proposed
sion losses [4]. DC microgrids are finding various applications
methods can be extended to more general topology that often
in more electric aircraft, naval ships, data centers, among
arises in real applications [21]. Third, many methods proposed
others [5]–[7]. With advanced power electronic devices, the
in the literature are ad hoc in nature and are specific to a
transient behavior in the power outputs of many loads can be
particular application scenario [13], [22]–[24]. There lacks a
neglected, and these loads can be modeled as constant power
systematic method to analyze the stability of the closed loop
loads (CPLs) [8]–[10].
system, especially when uncertain CPLs and general topology
Stability analysis problems for DC microgrids with CPLs
are considered.
have been studied in the literature. The CPLs are nonlinear
In this paper, we formulate and study the robust stability
and exhibit a negative impedance V-I characteristic, which
problem of a general DC microgrid with uncertain CPLs. We
may cause instability of a DC microgrid [11]. Researchers
assume that the power of each CPL may take arbitrary values
mainly study the stability of a DC microgrid around a fixed
within a given interval. As a result, the vector of the overall
equilibrium calculated based on some nominal values of
CPL power lies in a polytopic uncertainty set. Different CPL
the CPLs [8], [12]–[16], [16], [17]. One major challenge
vectors in the uncertainty set may lead to different system
lies in the nonlinearity contributed by the CPLs. To tackle
equilibria. We call the system locally robustly stable if all
the nonlinearity, some linearize a DC microgrid at a giv-
the resulting equilibria are locally exponentially stable. We
en equilibrium and analyze the resulting linearized system.
show that checking robust stability of a general DC microgrid
J. Liu is with both the Center for Automotive Research and the Department can be formulated as a robust stability analysis problem of
of Electrical and Computer Engineering, the Ohio State University, Columbus, linear systems with polytopic uncertainties on system matrices.
OH 43210. Email: [email protected]. Existing results on linear uncertain systems [25]–[27] can be
W. Zhang is with the Department of Electrical and Computer Engineering,
the Ohio State University, Columbus, OH 43210. Email: [email protected]. used to derive sufficient locally robust stability conditions;
G. Rizzoni is with both the Center for Automotive Research and the Depart- however, these conditions are computationally challenging
ment of Mechanical and Aerospace Engineering, the Ohio State University, to verify and may be over conservative. In this paper, we
Columbus, OH, 43210. Email: [email protected].
This work was supported in part by the National Science Foundation under show that for our DC microgrid robust stability problems,
grant ECCS-1309569 and grant CNS-1552838 the parameter uncertainty lies in the diagonal entries of the
0885-8950 (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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2017.2697765, IEEE
Transactions on Power Systems
2
system matrix. We take advantage of this structure and derive A. Dynamic Model for Bus k
a set of sufficient conditions that can be efficiently checked to 1) Voltage Source: We assume that each bus has one con-
guarantee locally robust stability. The conservativeness issue trollable voltage source with standard V-I droop control [28].
is also discussed in the paper, we obtain an approximated Let vk and isk be the terminal voltage and output current of the
quantification of the conservativeness of the proposed work. source on Bus k, respectively. The source connects with the
By virtue of the quantified conservativeness, a quantitative rest of the microgrid through a resistor and an inductor. The
relationship between the CPL power and voltage is derived, current isk is the state variable for the source. The dynamics
which is useful for DC microgrid operations. For example, of isk are given by,
with given CPL power range one can compute a constraint
Lsk didtsk = vk − rsk isk − vbk
for CPL voltage to guarantee stability. To our knowledge, , (1)
the robust stability problem formulated in this paper has not vk = vkref − dk isk
been studied in the DC microgrid literature, and our results where Lsk and rsk are the source inductance and resistance,
provide new insights for stability analysis of DC microgrids vbk is the voltage of the k th DC bus, vkref and dk are the droop
with uncertain CPLs. reference and droop gain, respectively. Here, we have ignored
The rest of the paper is organized as follows, Section II the internal dynamics of the power electronic devices, which
presents a model for a general DC microgrid with CPLs, is a standard assumption for microgrid stability and control
Section III develops the robust stability framework, Section IV problems [5].
uses simulation studies to verify the main results and study the
conservativeness of the framework, and Section V concludes
݅ ݎ
the paper and discusses future research directions.
݅௦ ݅
0885-8950 (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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2017.2697765, IEEE
Transactions on Power Systems
3
where Nk is the set of indices of all the neighboring buses of capacitors are in parallel with the DC bus link capacitors
the k th bus, Cbk and rkj are the DC link capacitance of the and can be directly combined with them. Hence, the resulting
k th bus and the equivalent resistance of the transmission line model is still in the form of (4) with modified Cbk parameters.
between the k th and j th buses, respectively. Note that Nk is For Π-model with equivalent line inductance [29], another
determined by the microgrid network topology. state that represents the line current can be added. It will also
result in a microgrid model similar to (4) with an increased
B. n-bus DC Microgrid state dimension.
We now combine the individual models of the sources, Remark 2. Recently, different control laws have been proposed
CPLs, and the DC link capacitors on all the n buses to obtain to replace the V-I droop control [8], [22]. The modeling
the overall model of the DC microgrid. method presented in this paper and the corresponding stability
Let ix , vx ∈ R2n , p, v, and v ref ∈ Rn , be the vectors of results can be easily extended to DC microgrid with these
all the source and CPL currents, DC bus and CPL voltage, control laws. In Section III-C, we exemplify it by applying the
load power, source voltage, and droop references, respectively. proposed methods to a DC microgrid with distributed control.
They are given as follows, Remark 3. Depending on the control of the power electronic
devices, there are different types of sources [30], [31]. For a
ix=[is1 ,· · ·, isn , il1 ,· · ·, iln ]T,vx=[vB1 , · · · , vBn , vl1 , · · · , vln ]T, current source whose output current is regulated to be linear
p=[p1,· · ·,pn ]T , v=[v1,· · ·,vn ]T, v ref=[v1ref,· · ·,vnref ]T . in the system state, the above described modeling method can
be directly applied. Meanwhile, for a power source [31], it can
In addition, define a vector x ∈ R4n and a nonlinear function
be modeled as a current source whose output current equals
h(·, ·) : Rn × R4n → Rn as follows,
the output power divided by the terminal voltage. Similar to
p1 pn T the CPLs, the model of a power source can be incorporated
x = [iTx , vxT ]T , h(p, x) = [− , · · · , − ] .
vl1 vln in system (4), and the main results of this paper can apply as
The vector x stacks all the current and voltage states of the well.
components, and the nonlinear function h(p, x) contains the
additive nonlinear terms in (2). III. ROBUST S TABILITY F RAMEWORK
Let Dk ∈ R4n×4n , and only its k th diagonal entry is non- A. Robust Stability under Uncertain CPLs
zero and is 1.
The stability analysis of a DC microgrid is crucial since
With the above notation, the overall microgrid dynamics can
excessive CPL might cause instability. In this paper, the
be written in the following compact form:
unknown CPL power p is assumed to be physically bounded
n
X in a polytopic set P defined as follows,
ẋ = Ax + Bv ref + Ch(p, x), A = A0 − B dk Dk , (4)
k=1 P = p : pk ∈ [pk , pk ], k = 1, · · · , n .
where the matrices A, A0 ∈ R4n×4n , and B, C ∈ R4n×n . The We aim at analyzing the stability of system (4) with an
matrix A0 is the system matrix of the open loop DC microgrid arbitrary p ∈ P. The challenge of the analysis lies in that
without the V-I droop control, and it is determined by the the equilibria of the system are difficult to find [32]. With n
system parameters and topology. When the droop control is CPLs, one needs to solve n quadratic equations to obtain the
utilized, the matrix A is the system matrix of the closed loop equilibria of system (4). In general, there are 2n solutions and
system. Other than the parameters and topology, A depends there are no general results to determine which one is the actu-
on the droop gains as well. al equilibrium the system will converge to [33]. Furthermore,
We now briefly discuss how the matrices A, B, and C can the solution of the equations changes with respect to p. With
be constructed from individual bus models (1)-(3) as well as an unknown CPL power profile p, the set of possible equilibria
the system topology. Substituting isk = xk , vbk = x2n+k into cannot be characterized for a general DC microgrid.
(1) yields, Due to the lack of a predetermined equilibrium, we study the
disk 1 ref rsk + dk 1 stability of all the possible equilibria that are feasible for the
= ẋk = v − xk − x2n+k , microgrid. For a vector p ∈ P, let xe ∈ R4n be an equilibrium
dt Lsk k Lsk Lsk
of system (4). At xe , let the steady state voltage of the k th
thus, Akk = −(rskP+dk )/Lsk , Ak(2n+k) = −1/Lsk , Bkk = 1. e
n CPL be vlk , let vle = [vl1
e e T
, · · · , vln ] . Then we can define an
In addition, C = k=1 1/Clk D3n+k from (2). Furthermore, e
operational constraint set Vl in the following form,
from (3), if j ∈ Nk , then A(2n+k)(2n+j) = 1/rkj , and n o
A(2n+k)(2n+k) = −|Nk |/rkj , where | · | gives the cardinality Vle = vle ∈ Rn : vlk e
∈ [vlke e ], k = 1, · · · , n ,
, vlk
of a set. This demonstrates how the matrix A depends on the
system topology. where v elk ≥ vlk
e
≥ 0.
Remark 1. In deriving the microgrid model (4), we have adopt- As p varies in P, xe changes accordingly. Some equilibria
ed a commonly used resistive transmission line model [5]. The might not even be admissible. We thus focus on the following
modeling method can be applied to other transmission line set of equilibria whose corresponding CPL voltage lies in Vle ,
types. For example, regarding transmission line Π-model [29] Xe(P,Vle)= xe∈R4n:Axe+Bv ref+Ch(p, xe ) = 0,p ∈ P,vle ∈ Vle .
that has an equivalent capacitor on each end of the line, the
0885-8950 (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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2017.2697765, IEEE
Transactions on Power Systems
4
Definition 1. System (4) is said to be locally robustly stable if Remark 4. If constraints (7) are all feasible for some P =
any equilibrium in the set Xe (P, Vle ) is locally exponentially P T 0, U (z) = z T P z is a common Lyapunov function for
stable. the vertices in Az . Since Az is a polytopic set, any matrix
Az ∈ Az can be expressed as a convex combination of its
The above defined locally robust stability is not exactly the
vertices [37]. Thus, the function U (z) is a common Lyapunov
same as the definition used in the classical robust control
function for all the matrices in Az as well, and any matrix
literature [34], [35]. The classical robust control typically
Az ∈ Az is Hurwitz stable by Lyapunov method for stability.
studies the stability of one predetermined equilibrium. Note
that such an equilibrium is not available for DC microgrids Checking the condition given in Lemma 1 is a linear matrix
with uncertain CPLs, hence we study all the equilibria lying inequalities (LMIs) feasibility problem, which can be checked
in Xe (P, Vle ). efficiently [38]. There are 2n LMIs involved in the feasibility
problem. Even though LMIs can be checked efficiently, when
n is nontrivial the computation burden could still be huge. This
B. Robust Stability Analysis limits the applicability of Lemma 1 for general DC microgrid
This subsection develops methods to analyze the locally ro- stability analysis.
bust stability defined in Definition 1. We derive sufficient con- This limitation can be eliminated by making use of the
ditions to guarantee the locally robust stability of system (4). special structure of the matrix Az . One can observe that each
Computationally efficient convex optimization problems are δk lies on one diagonal entry of Az . We take advantage of this
formulated to facilitate the analysis. structure and transform the condition from checking 2n LMIs
Let xe be an arbitrary equilibrium in Xe (P, Vle ). Linearizing to solving a convex optimization problem whose complexity
system (4) around xe yields, only increases polynomially with n.
n
X Let Az ∈ Az be the element-wise maximum Pn of all the
ż = Az z, Az = A + δk D3n+k , (5) matrices in Az , i.e. it is given by Az = A + k=1 δ k D3n+k .
k=1 Let the largest value of all δ k , k = 1, · · · , n, be δ max . It is
worth pointing out that Az and δ max are deterministic and can
where z ∈ R4n is the state variable of the linearized sys-
be obtained easily. In addition, let ∆δk =Pδ k − δk and let Qj
tem, Az ∈ R4n×4n is the linearized system matrix, and n
δk = pk /(Clk (vlk e 2
) ), which is obtained by linearizing the k th Pan matrix definedTas follows, Qj = P k=1 ∆δk D3n+k +
be
e ( k=1 ∆δk D3n+k ) P.
entry of Ch(p, x) at xe . Since pk and vlk vary in [pk , pk ] and
e e
[vlk , v lk ], respectively, δk takes value in an interval, [δk , δ k ], Lemma 2. System (4) is locally robustly stable if ∃P = P T
where δk = pk /(Clk (vlk e )2 ) and δ = p /(C (v e )2 ). Note
k k lk lk 0, γ > 0, and t > 0 such that the following optimization
e problem is feasible,
that the overall uncertainty of pk and vlk is captured by δk .
The system matrix Az is an uncertain matrix, and it belongs
min t − γ (8)
to a set Az defined as follows, P,t,γ
( n
) T
4n×4n
X subj. to: P Az + Az P −γI4n ,
Az= Az∈R : Az=A+ δk D3n+k , δk ∈ [δk , δ k ] . (6)
P tI4n , 2t · δ max < γ,
k=1
T
P = P , P 0, γ > 0, t > 0.
The set Az is a polytopic set, and it has 2n vertices. Let the
matrix Avj be the j th vertex of the set Az . It can be written in Remark 5. Instead of solving the feasibility problem given
Pn
the form: Avj = A+ k=1 δk D3n+k with δk = δ k or δk = δk . in Lemma 1 that involves 2n LMIs, the condition derived
Recall that a matrix is Hurwitz stable if all of its eigenvalues in Lemma 2 allows us to verify the locally robust stability
have negative real parts. Given a CPL uncertainty set P and a by solving a convex optimization problem with only one set
voltage constraint set Vle , all the possible equilibria lie inside of LMIs. One explanation is that for Lemma 1, we need to
Xe (P, Vle ) and all the possible linearized system matrices lie check the stability of 2n DC microgrids with different CPL
inside Az . Therefore, system (4) is locally robustly stable if power and steady state voltage. In Lemma 2, we only check the
all the matrices in Az are Hurwitz stable. Notice that there are stability of one DC microgrid in an extreme case, represented
infinitely many elements in the set Az . As a standard result in by Az . In the meantime, we find that by lower bounding the
linear uncertain systems, to ensure stability of all the matrices smallest eigenvalue of every Qj , the Hurwitz stability of Az
in Az , it suffices to have a common Lyapunov function for all ensures the stability of all the matrices in Az . By exploring
the vertices of Az [36]. into the structure of the system matrix, this lower bound
For a matrix M , let M 0 and M ≺ 0 represent that holds when upper bounding the largest possible parameter
M is positive definite and negative definite, respectively, let uncertainty δ max . The detailed proof of Lemma 2 is reported
M 0 and M 0 represent that M is positive semidefinite in [39].
and negative semidefinite, respectively.
The feasibility of problem (8) guarantees the locally robust
Lemma 1. System (4) is locally robustly stable if ∃P = P T stability of system (4). The cost function of the problem is
0 satisfying motivated by the structure of the constraints. When there exists
multiple solutions, cost function (8) selects the “optimal”
P Avj + (Avj )T P ≺ 0, j = 1, · · · , 2n . (7) solution that minimizes the function t − γ. The cost function,
0885-8950 (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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2017.2697765, IEEE
Transactions on Power Systems
5
however, does not affects the feasibility of the problem, and computational efficiency and less conservativeness, it has
any other convex cost functions can be utilized. better applicability for practical DC microgrid applications.
The price we pay for such a complexity reduction is the
increased conservativeness coming from the constraint trans-
C. Discussions of the Proposed Framework
formation. In other words, there can be locally robustly stable
microgrids that satisfy the conditions in Lemma 1, but fail to The robust stability framework provides a set of sufficient
satisfy the condition in Lemma 2. The conservativeness can conditions to determine the locally robust stability of sys-
be reduced when we directly focus on the smallest eigenvalue tem (4). In this subsection, we show two insights into the
of each matrix Qj in the program by introducing constraints results.
such as −γI4n ≺ Qj , j = 1, · · · , 2n . Nevertheless, this again 1) Critical Case Interpretation: To solve problem (8) or
increases the computational complexity by adding another 2n problem (9), we only need to know the matrix Az . By
constraints. We next show that this can be accomplished by definition, the matrix Az can be recognized as the linearized
adding only a polynomial number of constraints. system matrix of the DC microgrid in a case where each CPL’s
To simplify notation, we define a matrix Gk as follows, power reaches the upper bound and each CPL’s steady state
Gk = P D3n+k + D3n+k P. voltage reaches the lower bound. Given operational constraint
sets P and Vle , Az is simple to obtain. We call this case the
Proposition 1. System (4) is locally robustly stable if ∃P = critical case of the DC microgrid with constraint sets P and
P T 0, γ0 , · · · , γn > 0, η1 , · · · , ηn > 0, such that the Vle . It is convenient to interpret the robust stability framework
following optimization problem is feasible, as checking the critical case to guarantee the stability of the
n
X DC microgrid in all possible scenarios.
min − γk (9) Due to its dependence on the operational constraint sets, the
P,γ0 ,··· ,γn ,η1 ,··· ,ηn
k=0 critical case alters when the constraint sets are changed, and
n
T X the stability analysis made with the previous critical case may
subj. to: P Az + Az P −γ0 I4n , η k = γ0 ,
no longer hold. We next derive a condition to guarantee the
k=1
locally robust stability of the post-change system by simply
δ k · Gk (γk − ηk )I4n , k = 1, · · · , n,
comparing the critical cases.
γk > 0, ηk > 0, k = 1, · · · , n, Suppose the constraint sets are changed into P̂ and V̂le ,
P = P T , P 0, γ0 > 0. respectively, we study the equilibria of system (4) that lie
inside an operationally feasible set X̂e (P̂, V̂le ). Let the lin-
Proof. We assume that problem (9) is feasible. Let P = P T earized system
0, γ0 , · · · , γn > 0, and η1 , · · · , ηn > 0 be the solution of Pn matrix of the post-change critical case be
Âz = A + k=1 δ̂k D3n+k , where δ̂k is the new upper bound
the problem. From Lemma 2, to prove that system (4) is of each δk .
locally robustly stable, we only need to show that the smallest
To simplify discussion we use P1 (Az ) and P2 (Az ) to
eigenvalue of Qj is greater than γ0 .
represent problem (8) and problem (9), respectively, when Az
The matrix Qj can be expressed in terms of Gk and ∆δk
is utilized in the problems.
as follows,
n
X n
X n
X Corollary 1. System (4) is locally robustly stable with
Qj =P ∆δk D3n+k+( ∆δk D3n+k )TP= ∆δk Gk . (10) X̂e (P̂, V̂le ) if the problem Pl (Az ) is feasible, l = 1 or 2, and
k=1 k=1 k=1 δ k ≥ δ̂k , ∀k = 1, · · · , n.
Since Qj is symmetric, its smallest eigenvalue is given by Remark 6. Corollary 1 presents a condition that reduces the
minyT y=1 y T Qj y. Substituting (10) into minyT y=1 y T Qj y repetitive efforts of checking the robust stability framework.
yields, If the pre-change system is proved to be locally robustly
stable using the proposed framework, the post-change system
n
!
X
T T
min y Qj y = min y ∆δk Gk y. remains locally robustly stable when the upper bound of
y T y=1 y T y=1
k=1 each parameter uncertainty does not increase. Recall that
e 2
By Weyl’s inequality, δ k = pk /(Clk · (vlk ) ). One implication from the corollary
n n is to increase the load capacitance, Clk . This provides one
X ∆δk X
explanation for the observations that larger capacitor increases
min y T Qj y > − ηk ≥ − ηk = −γ0 ,
y T y=1
k=1
δk k=1 the stability region of DC microgrid [8]. Another method to
keep the locally robust stability is to modify the operational
thus, the smallest eigenvalue of Qj is greater than −γ0 , and
bounds of the CPLs. When the upper bound of CPL power
this completes the proof.
is smaller or the lower bound of CPL voltage is regulated to
Proposition 1 directly uses the smallest eigenvalues of the be higher, the system remains locally robustly stable. It is in
matrices, which can potentially reduce the conservativeness. line with the common intuition that a smaller CPL always
In addition, the number of constraints in convex problem (9) makes it easier for the system to remain stable, and a higher
grows polynomially with respect to the number of buses, n. level of voltage in the DC microgrid is helpful [33]. Notice
This makes the condition numerically more tractable than that the condition given in Corollary 1 does not impose any
methods like Lemma 1. Due to its advantages in terms of additional constraint on the lower bound of δk . Recall that
0885-8950 (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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2017.2697765, IEEE
Transactions on Power Systems
6
δk = pk /(Clk (v elk )2 ), it implies that the values of pk and parameters of the k th bus are shown in TABLE I. Suppose that
v elk have less influences on the locally robust stability of the each load power is uncertain and may take values in the range
DC microgrid. The proof of Corollary 1 is similar to that of [5kW, 20kW]. Let pnk be the nominal power of the k th CPL,
Proposition 1 and is thus omitted here. and it is given by pnk = 15kW. In addition, the operationally
e
2) Application to Distributed Control: Recently, some dis- feasible range of each steady state CPL voltage vlk is set
tributed controllers have been proposed to replace the V-I as [360V, 440V]. The simulation studies are conducted on a
droop controllers for DC microgrids [22], [23]. The modeling laptop with Intel Core i7. The model of the DC microgrid is
method presented in this paper and the corresponding stability built and simulated by MATLAB/Simulink, and the convex
results can be easily extended to DC microgrids with these problems given in the proposed robust stability framework are
distributed controllers. In the following we use an example to solved by MATLAB convex optimization toolbox CVX [41],
demonstrate this. [42]. Notice that any other simulation and optimization tools
Similar to that of [22], we consider a distributed control law can be used for the case studies.
for voltage regulation. We adopt a communication network
L G
like [40], which enables each source to communicate with the L G L G
distributed control law, v becomes a control state, and every Fig. 3. The nine-bus DC microgrid
equilibrium of v makes sure that the weighted sum of the
tracking errors to be zero. Seeing that the goal of voltage We study the stability of the DC microgrid’s equilibri-
regulation is accomplished when the equilibrium is stable, the a that lie in an operationally feasible set as discussed in
locally robust stability analysis is of significance to study. Section III-A. Existing works only consider the stability of
The main results of the paper can be applied for the analysis. one given equilibrium which requires the full knowledge of
The state variable of the closed loop system is augmented the CPL power profiles [8], [12], [13], [15]. Furthermore,
by the control state v. Let ζ = [v T , xT ]T be the augmented their results cannot guarantee the stability of all the possible
state variable. The dynamics of v are linear in ζ, and the equilibria in the feasible set. We demonstrate this by using
dynamics of x has a linear term in ζ and a nonlinear term simulation results. With nine CPLs in the example microgrid
concerning the CPLs as well. Thus, the dynamics of ζ can be there are 512 equilibria. Sorting out the one equilibrium that
characterized by a system similar to system (4). The problem the DC microgrid operates at is in general difficult [33].
can also be formulated as a stability analysis problem of a Numerical methods [8], [13] finds that the system operates at a
polytopic uncertain linear system with uncertainties lying on stable equilibrium with nominal power 15kW. However, when
the diagonal entries of the system matrix. We can apply the the CPL power increase from 5kW to 20kW, the simulation
proposed robust stability framework for this problem. results, which are shown in Fig. 4, indicate that the microgrid
becomes unstable when the load power increases to around
IV. C ASE S TUDY 16kW.
In this section, we demonstrate the effectiveness of the ro-
bust stability framework proposed in Section III and study the TABLE I
S PECIFICATIONS OF THE NINE - BUS DC MICROGRID
conservativeness of the proposed robust analysis framework.
In addition, we discuss the importance of system parameters to
robust stability and the relationship of CPL power with voltage rsk 0.05 Ω rlk 0.05 Ω Lsk 0.9mH
Llk 0.9mH Cbk 0.75mF Clk 0.7mF
levels. [pk , pk ] [5kW, 20kW] rjk 1Ω pn
k 15kW
e , ve ]
[vlk [360V, 440V] vkref 400V dk 0.06
lk
A. Simulation Validation
The case study is based on a nine-bus DC microgrid shown In fact, the DC microgrid may operate at every equilibrium
in Fig. 3. Each source is under the V-I droop control. The in the feasible set. There needs a numerically efficient method
0885-8950 (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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2017.2697765, IEEE
Transactions on Power Systems
7
L G L
G G L
Bus 7 Bus 8
L L
0885-8950 (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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2017.2697765, IEEE
Transactions on Power Systems
8
0885-8950 (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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2017.2697765, IEEE
Transactions on Power Systems
9
[15] D. Marx et al., “Large signal stability analysis tools in DC power [40] J. W. Simpson-Porco, F. Dörfler, and F. Bullo, “Synchronization and
systems with constant power loads and variable power loads ;a review,” power sharing for droop-controlled inverters in islanded microgrids,”
IEEE Trans. Power Electron., vol. 27, no. 4, pp. 1773–1787, April 2012. Automatica, vol. 49, no. 9, pp. 2603 – 2611, 2013.
[16] J. A. Belk et al., “Stability and control of ad hoc dc microgrids,” in [41] M. Grant and S. Boyd, “CVX: Matlab software for disciplined convex
proc. IEEE 55th Annu. CDC, Dec 2016, pp. 3271–3278. programming, version 2.1,” http://cvxr.com/cvx, Mar. 2014.
[17] D. Zonetti, R. Ortega, and J. Schiffer, “A tool for stability and [42] ——, “Graph implementations for nonsmooth convex programs,”
power sharing analysis of a generalized class of droop controllers in Recent Advances in Learning and Control, ser. Lecture Notes
for high-voltage direct-current transmission systems,” arXiv preprint in Control and Information Sciences, V. Blondel, S. Boyd, and
arXiv:1609.03149, 2016. H. Kimura, Eds. Springer-Verlag Limited, 2008, pp. 95–110,
[18] C. D. Persis et al., “A lyapunov approach to control of microgrids with http://stanford.edu/∼boyd/graph dcp.html.
a network-preserved differential-algebraic model,” in proc. IEEE 55th
Annu. CDC, Dec 2016, pp. 2595–2600.
[19] J. Liu et al., “Planning and control of electric vehicles using dynamic
energy capacity models,” in proc. IEEE 53rd Annu. CDC, 2013, pp.
379–384.
Jianzhe Liu (S’13) received the B.E. degree in elec-
[20] D. T. Nguyen and L. B. Le, “Risk-constrained profit maximization for trical engineering from the Huazhong University of
microgrid aggregators with demand response,” IEEE Trans. Smart Grid, Science and Technology, Wuhan, China, in 2012. He
vol. 6, no. 1, pp. 135–146, 2015. is currently pursuing doctorate degree in electrical
[21] D. Boroyevich et al., “Intergrid: A future electronic energy network?” engineering from the Ohio State University, Colum-
IEEE J. Emerg. Sel. Topics Power Electron.,, vol. 1, no. 3, pp. 127–138, bus, OH. His research interests include modeling,
2013. energy management, and control of energy systems.
[22] V. Nasirian et al., “Distributed cooperative control of DC microgrids,”
IEEE Trans. Power Electron., vol. 30, no. 4, pp. 2288–2303, April 2015.
[23] S. Anand, B. G. Fernandes, and M. Guerrero, “Distributed control to
ensure proportional load sharing and improve voltage regulation in low-
voltage DC microgrids,” IEEE Trans. Power Electron., vol. 28, no. 4,
pp. 1900–1913, April 2013.
[24] Y. Gu, X. Xiang, W. Li, and X. He, “Mode-adaptive decentralized control
for renewable dc microgrid with enhanced reliability and flexibility,”
IEEE Trans. Power Electron., vol. 29, no. 9, pp. 5072–5080, Sept 2014. Wei Zhang (S’07-M’10) received the B.E. degree
[25] C. Scherer, “LPV control and full block multipliers,” Automatica, in automatic control from the University of Science
vol. 37, no. 3, pp. 361 – 375, 2001. [Online]. Available: and Technology of China, Hefei, China, in 2003, and
http://www.sciencedirect.com/science/article/pii/S000510980000176X the M.S. degree in statistics and the Ph.D. degree in
[26] P. Gahinet, P. Apkarian, and M. Chilali, “Affine parameter-dependent electrical engineering from Purdue University, West
lyapunov functions and real parametric uncertainty,” IEEE Trans. Autom. Lafayette, IN, USA, in 2009. Between 2010 and
Control, vol. 41, no. 3, pp. 436–442, Mar 1996. 2011, he was a Postdoctoral Researcher with the
[27] Y. He et al., “Parameter-dependent lyapunov functional for stability Department of Electrical Engineering and Computer
of time-delay systems with polytopic-type uncertainties,” IEEE Trans. Sciences, University of California, Berkeley, CA,
Autom. Control, vol. 49, no. 5, pp. 828–832, May 2004. USA. He is currently an Assistant Professor with the
Department of Electrical and Computer Engineering,
[28] J. M. Guerrero et al., “Hierarchical control of droop-controlled AC and
Ohio State University, Columbus, OH, USA. His research focuses on control
DC microgrids ;a general approach toward standardization,” IEEE Trans.
and game theory with applications in power systems, robotics, and intelligent
Ind. Electron., vol. 58, no. 1, pp. 158–172, Jan 2011.
transportations.
[29] P. Kundur, N. J. Balu, and M. G. Lauby, Power system stability and
control. McGraw-hill New York, 1994, vol. 7.
[30] M. P. Kazmierkowski and L. Malesani, “Current control techniques for
three-phase voltage-source PWM converters: a survey,” IEEE Trans. Ind.
Electron., vol. 45, no. 5, pp. 691–703, Oct 1998.
[31] E. Koutroulis, K. Kalaitzakis, and N. C. Voulgaris, “Development of Giorgio Rizzoni (S’83-M’85-SM’02-F’04) received
a microcontroller-based, photovoltaic maximum power point tracking the B.S., M.S., and Ph.D. degrees from the Univer-
control system,” IEEE Trans. Power Electron., vol. 16, no. 1, pp. 46– sity of Michigan, Ann Arbor, MI, USA, in 1980,
54, Jan 2001. 1982, and 1986, respectively, all in electronics and
[32] G. S. Gajani, A. Brambilla, and A. Premoli, “Numerical determination communication engineering.
of possible multiple DC solutions of nonlinear circuits,” IEEE Trans. He is the Chair of Ford Motor Company in
Circuits and Syst. I, Reg. Papers, vol. 55, no. 4, pp. 1074–1083, May ElectroMechanical Systems. He is a Professor of
2008. Mechanical and Aerospace Engineering and Elec-
[33] J. W. Simpson-Porco, F. Dörfler, and F. Bullo, “On resistive networks of trical and Computer Engineering with the Ohio
constant-power devices,” IEEE Trans. Circuits and Syst. II, Exp. Briefs, State University (OSU), Columbus, OH, USA. Since
vol. 62, no. 8, pp. 811–815, Aug 2015. 1999, he has been the Director of the Center for
[34] G. Leitmann, “Guaranteed asymptotic stability for some linear systems Automotive Research (CAR), Ohio State University, an interdisciplinary
with bounded uncertainties,” Journal of Dynamic Systems, Measurement, university research center in the OSU College of Engineering. He has
and Control, vol. 101, no. 3, pp. 212–216, 1979. contributed to the development of graduate curricula in these areas, and
has served as the Director of three U.S. Department of Energy Graduate
[35] P. P. Khargonekar, I. R. Petersen, and K. Zhou, “Robust stabilization
Automotive Technology Education Centers of Excellence: Hybrid Drivetrains
of uncertain linear systems: quadratic stabilizability and H-∞ control
and Control Systems from 1998 to 2004, Advanced Propulsion Systems from
theory,” IEEE Trans. Autom. Control, vol. 35, no. 3, pp. 356–361, Mar
2005 to 2011, and Energy Efficient Vehicles for Sustainable Mobility from
1990.
2011 to 2016. Since January 2011, he has been the OSU Site Director for the
[36] J. Elizondo et al., “Robust small signal stability for microgrids under
U.S. Department of Energy China-USA Clean Energy Research Center-Clean
uncertainty,” in proc. IEEE 6th PEDG, June 2015, pp. 1–8.
Vehicles. His research interests include modeling, control and diagnosis of
[37] M. Mansour, “Sufficient conditions for the asymptotic stability of advanced vehicles, energy efficiency, alternative fuels, the interaction between
interval matrices,” International Journal of Control, vol. 47, no. 6, pp. vehicles and the electric power grid, vehicle safety and intelligence, and policy
1973–1974, 1988. and economic analysis of alternative fuels and vehicle fuel economy.
[38] S. Boyd and L. Vandenberghe, Convex Optimization. Cambridge Prof. Rizzoni was elected a Fellow of IEEE in 2004 and a Fellow of
University Press, 2004. SAE in 2005. He was the recipient of the 1991 National Science Foundation
[39] J. Liu, W. Zhang, and G. Rizzoni, “Robust stability analysis of dc mi- Presidential Young Investigator Award, and of several other technical and
crogrids with constant power loads,” arXiv preprint arXiv:1702.07580, teaching awards
2017.
0885-8950 (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.