Distributed Secondary and Optimal Active Power Sharing Control For Islanded Microgrids With Communication Delays
Distributed Secondary and Optimal Active Power Sharing Control For Islanded Microgrids With Communication Delays
fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2017.2785811, IEEE
Transactions on Smart Grid
1
Abstract—As the communication delays in smart grid are con- or distributed way. Motivated by the advantages of distributed
sidered, the novel secondary voltage and frequency restoration, implementation and inspired by the distributed control technol-
the optimal active power sharing, and the accurate reactive power ogy of multi-agent systems (MAS) [11], [12], the distributed
sharing protocols are proposed in this paper. The Lyapunov-
Krasovskii functions are used to analyze the stability of our control approach has attracted a great deal of attention. The
proposed distributed secondary control schemes. Sensitivity anal- tertiary control is usually used for power flow control and
ysis of time delays is also provided. Whether the communication optimal dispatch [9], [10].
delays exist or not, the control protocols always guarantee that This work mainly focuses on the distributed secondary
the output voltage amplitude and frequency of each distributed control layer. The control objectives of secondary control are
generator synchronize to the reference values and maintain the
optimal active power sharing and the accurate reactive power (i) to compensate for the frequency and voltage deviation
sharing properties. Moreover, our proposed control protocols [7], [13], [14], [15], [16], [17], [18], [19], [20], [21], (ii)
are implemented through sparse communication networks and to realize accurate activate and reactive power sharing [7],
thus meet the plug-and-play feature of the future smart grid. [9], [13], [17], [18], [19], [20], [21]. In these works, the
Simulation results are provided to show the effectiveness of the microgrid is considered as a multi-agent system and the DG is
proposed control protocols.
represented as an agent. The active and reactive power sharing
Index Terms—Distributed coordinated control, Multiagent con- problem can be converted to a leaderless consensus problem of
sensus algorithm, Secondary control, Frequency and voltage multi-agent systems and the voltage and frequency restoration
restoration, Time delays, Optimal power sharing.
control problems are transformed to a leader-follower problem
of multi-agent systems [16].
I. I NTRODUCTION In the works [14], [15], [16], the input-output feedback
With the ever growing concerns on energy crisis and envi- linearization approach is used to transform the nonlinear model
ronmental problem, distributed energy resources (DERs) such of microgrid systems to a linear multi-agent system. They
as wind and solar energy have attracted increasing attention. assume only a few DGs can directly obtain the reference
Although the DERs have many advantages and potential values and the frequency and voltages can finally synchronize
benefits, how to make full use of them is a very tricky problem. to the expected values. The work [17] first derives a simplified
One promising solution is through microgrid. Microgrid is a dynamic model of islanded microgrid and then gives a finite-
small power distribution system that can integrate distributed time voltage restoration strategy. A finite time frequency
generator (DG), load, energy storage device and other power regulation strategy with a consensus-based distributed active
electronic equipments [1], [2]. It can operate in islanded or power regulator is presented in [20]. A distributed averaging
grid-connected mode[2], [3]. algorithm is proposed in [13] and [18], which can overcome
For autonomous (islanded) microgrids, the hierarchical con- the frequency and voltage deviations while maintaining the
trol is the basic control and management strategy [4], [5], [6] active power sharing property. In [19], the critical bus voltage
which consists of three levels (i.e., primary control, secondary and frequency can be regulated to the nominal values while
control, and tertiary control). The control objectives of droop- maintaining the equal generation cost incremental value in
based primary control are to guarantee voltage/frequency sta- droop control and the accurate reactive power sharing. The
bility and active/reactive power sharing[5]. However, the droop rapid development of communication technology (such as
control will cause imprecise voltage and frequency regulation GPRS, wireless network, Internet and so on) paves the way for
[6], [7], [8]. One way to deal with these drawbacks is to add the achievement of communication network based secondary
the secondary control [4], [5], [6], [7], [8]. The secondary control [7]- [21]. However, as the communication constraints
control can be implemented in a centralized, decentralized, such as communication delays are considered, the distributed
secondary control problems are not well studied. Only limited
This work was supported by the National Natural Science Foundation of works have discussed the effects of communication delays on
China (61673077, 61273108), the Basic and Advanced Research Project of
Chongqing (CSTC 2016jcyjA0361), and the Fundamental Research Funds for the secondary control [7], [8], [22].
the Central Universities (106112017CDJQJ178827). In this paper, the impacts of communication delays are
G. Chen and Z. Guo are with the Key Laboratory of Dependable Service considered and several control protocols are presented to
Computing in Cyber Physical Society, Ministry of Education, and the College
of Automation, Chongqing University, Chongqing, 400044, China (e-mail: restore the system frequency and terminal voltage to the
[email protected]). nominal values and guarantee the accurate active and reactive
1949-3053 (c) 2017 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/TSG.2017.2785811, IEEE
Transactions on Smart Grid
2
power sharing. As compared with the existing works, the main B. Primary and Secondary Control of Microgrid
contributions of this paper can be summarized as follows: Fig. 1 shows the basic control diagram of primary control.
1) The novel control protocols are proposed such that The primary control consists of three control loops, i.e., the
the control objectives are always guaranteed even if power control loop, the voltage control loop and the current
the communication delays are large enough. The re- control loop. The power control loop generates the voltage
sults in [7], [8], [22] are delay-dependent such that references Vod∗ ∗
, Voq according to the set-points provided by
the communication delays should be less than an up- the secondary control. With the voltage references Vod ∗ ∗
, Voq ,
per bound. However, our results are delay-independent, the voltage controller generates the current references i∗Ld, i∗Lq .
which greatly simplifies the implementation of the con- Then, the current control loop is activated to regulate the
trol scheme. outputs of the inverter by sinusoidal pulse width modulation
2) In our work, Lyapunov-Krasovskii based large-signal (SPWM) approach. Droop control is often used in the power
stability analysis approach is presented to analyze the control loop. Droop-based control is an autonomous control,
microgrid system under communication delays. Our which doesn’t need the communication network among DGs.
analysis method is different from the phase-coupled In islanded microgrid, the objective of droop control is to
oscillators approach [18], [21], [25], [29] and the small- maintain the voltage and frequency stability while to keep the
signal stability analysis method for the secondary fre- active and reactive power sharing. The basic primary control
quency and voltage control with communication delays laws for the i-th DG are as follows [5], [6]:
[23], [24], [30]. od
Vi = Vn − mi Qi
3) Our control scheme achieves the optimal active power
Vioq = 0 (1)
sharing control through the secondary control layer
ωi = ωn − ni Pi
rather than the tertiary control layer [9], [10].
4) The proposed secondary control schemes with commu- where Vn , ωn are the rated system frequency and nominal
nication delays are implemented in a fully distributed voltage, respectively; mi and ni are the constant droop co-
way and thus meet the plug-and-play property. efficients; Pi , Qi are the measured active power and reactive
The rest of this paper is organized as follows. In Section power, respectively; Viod , Vioq are the d-axis and q-axis quan-
II, we first give some preliminaries. The distributed secondary tities of voltage and ωi denotes the output q frequency. Since
oq 2 2
control method is given in Section III. Some simulation studies Vi = 0 and the voltage magnitude Vi = Viod + (Vioq ) ,
are performed in Section IV to validate the effectiveness of our we can rewrite the primary droop control law as [16]
proposed scheme. Section V concludes this paper.
Vi = Vn − mi Qi
(2)
ωi = ωn − ni Pi
II. P RELIMINARIES
Conventionally, the secondary control is implemented in a
In this section, we give some useful preliminaries required
centralized way, which is applied to compensate the voltage
in our following analysis.
and frequency deviations caused by the primary control. It has
the PI structure for i-th DG as follows [5], [6], [16]:
A. Graph Theory R
δVi = KP V (Vref − Vi ) + KIV R (Vref − Vi ) dt
Considering an islanded microgrid with multiple DGs. The (3)
δωi = KP ω (ωref − ωi ) + KIω (ωref − ωi ) dt
communication topology among the DGs can be represented as
a graph. The nodes in the graph denote the DGs and the edges where KP V , KIV and KP ω , KIω are the voltage and fre-
represent the communication links among DGs. The graph can quency control parameters, respectively; Vref , ωref are the
be expressed as G = (V, E, A), where V = {1, 2, · · · , N } is a reference values for the voltage and frequency, and δVi , δωi
non-empty finite set; E ⊆ V × V denotes the communication are the deviation offsets that will be sent to the primary control
links; A = (aij )N ×N (i, j ∈ V ) is the coupling adjacency layer. In this paper, we will consider the distributed secondary
matrix whose entry aij means the connection weight. The arc control with a distributed structure as shown in Fig. 2, where
(j, i) ∈ E means that there is an edge form node j to node the DGs are connected through a sparse communication net-
i. The weight aij > 0 if and only if (j, i) ∈ E. Otherwise, work. The distributed secondary controller will be proposed in
aij = 0. The set of neighbors of the i-th node is given by the subsequent section such that the terminal output voltage
Ni = {j ∈ V |(j, i) ∈ E }. Vi and frequency ωi can be restored to their reference values.
Define L = D − A as the graph Laplacian matrix with D = Assumption 1: The graph among DGs is connected and
· · , dN } ∈ RN ×N called as in-degree matrix,
diag {d1 , d2 , ·P balanced. There at least exists a DG that has access to the
where di =
N reference values (given by a virtual leader).
j=1 aij is the weighted in-degree of node i
(that is the i-th row sum of matrix A). The Laplacian matrix
L has all row sums equal to zero, which satisfies L1N = 0. C. Optimal Active Power Sharing
The graph G is called balanced if its Laplacian matrix L meets Assume that the generation cost for the i-th DG is given by
1TN L = 0. A graph is said to have a spanning tree if, there the quadratic function [9], [10]
exists a node (called the root) that has a directed path from
this node to other nodes in the graph [26]. Ci (Pi ) = αi Pi2 + βi Pi + γi (4)
1949-3053 (c) 2017 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/TSG.2017.2785811, IEEE
Transactions on Smart Grid
3
Vdc
2) To achieve the optimal active power sharing, i.e.
i L* q iL Rc Lc
PI Voltage PI Current
controller *
iLd controller Rf Lf
Cf η1 (P1 ) = η2 (P2 ) = · · · = ηN (PN ) , as t → ∞ (9)
VSC Output vb
LC filiter connector 3) i) To restore the output voltage of each DG to the
iLd ˈi Lq
vo io reference value, i.e.,
abc/dq
Vod ˈVoq
abc/dq lim Vi (t) = Vref , i = 1, 2, · · · , N (10)
t→∞
*
Vod *
ˈVoq Pi iod ˈioq ii) or to realize accurate reactive power sharing, i.e.,
Power controller Power
(Droop Control) Calculation
Qi m1 Q1 = m2 Q2 = · · · = mN QN , as t → ∞ (11)
Vi set Ziset
These objectives should be achieved even that the effects of
communication delays are considered.
Fig. 1. The basic control diagram of primary droop control
+ k η̇i (Pi )
Fig. 2. The distributed secondary control architecture
(13)
where uω i is the secondary frequency control input; c 1 is a
where αi , βi , γi are the cost parameters; Pi is the output active positive constant; gi is a pinning gain by which the i-th DG
power for the i-th DG. To minimize the total generation cost is connected to the reference ωref defined by a virtual leader
while maintaining the supply-demand balance, it is required agent; aij is the edge weight between node i and node j; ωi
that the incremental cost of each DG should be equal [9], [19]. denotes the output frequency of the i-th DG; τj denotes the
The incremental cost function is defined as: constant communication delay between node j and node i.
For the proposed secondary control input (12) and (13), we
ηi (Pi ) = 2αi Pi + βi (5) have the following result.
Theorem 1. Consider the frequency control system de-
As the optimal active power sharing is considered, the
scribed by (12) with the control input uω i . Under Assumption
following frequency droop control is considered
1, the output frequency of each DG synchronizes to the
ωi = ωn − kηi (Pi ) (6) reference value ωref even though the communication delays
exist.
where k is a positive constant coefficient. By replacing the Proof . Taking the time derivative on both sides of (12)
frequency droop function (2) with (6), we get that yields
ω̇i = −k η̇i (Pi ) + u̇ω
Vi = Vn − mi Qi i (14)
(7)
ωi = ωn − kηi (Pi )
Define the following error function
Remark 1: In the steady-state, the frequency ωi of all DGs
eω
i (t) = ωi (t) − ωref (15)
will be equal and thus ηi (Pi ) are also equal, which meets the
optimal active power sharing [19], [28]. Note that ω̇ref = 0. Combining (13) with (14) and taking the
time derivative of eω
i (t) yield
III. M AIN R ESULTS ėω
i (t)"= ω̇i (t) #
In this paper, the main control objectives of the distributed P
secondary control are as follows: = c1 aij (ωj (t − τj ) − ωi (t)) − gi (ωi (t) − ωref )
1) To restore the frequency of each DG to the reference "j∈Ni #
ω ω ω
P
value, that is = c1 aij ej (t − τj ) − ei (t) − gi ei (t)
j∈Ni
lim ωi (t) = ωref , i = 1, 2, · · · , N (8) (16)
t→∞
1949-3053 (c) 2017 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/TSG.2017.2785811, IEEE
Transactions on Smart Grid
4
Consider the candidate Lyapunov Krasovskii function R(i.e., ωi = ωref ), in light of (12), we have
t P
N N X Z t t0 c 1 [ a ij (ω j (t − τj ) − ωi (t)) − gi (ωi (t) − ωref )]dt =
X 2
X 2 j∈Ni
V (t) = (eω
i (t)) + c a
1 ij eω
j (s) ds ωref −ωn +kηi0 (Pi0 ). That is to say, a bounded compensation
i=1 i=1 j∈Ni t−τj
(17) term uω i is added to restore the systems frequency to the
It is obvious that V (t) ≥ 0, and V (t) = 0 if and only if reference value.
eω ω 2) Sensitivity analysis:
i (t) = 0 and ej (t) = 0, i.e., ωi (t) = ωj (t) = ωref . The
time derivative of this Lyapunov-Krasovskii function along Taking the Laplace transform for (16) yields
(16) is X
seω ω
aij e−τj s eω ω ω
i − ei (0) = c1 j (s) − ei (s) − gi ei (s)
N
j∈Ni
eω (t) ėω
P
V̇ (t) = 2 i i (t) (21)
i=1
N 2 2 Let E(s) = (eω ω T
1 (s), · · · , eN (s)) . Then we further have
eω − eω
P P
+ c1 aij j (t) j (t − τj )
i=1 j∈Ni (sIN + c1 A0 − c1 Aτ (s))E(s) = E(0) (22)
N P
c1 aij eω eω eω
P
=2 i (t) j (t − τj ) − i (t)
i=1 j∈Ni where A0 = diag{di +gi } and the entries of matrix Aτ (s) are
N aij (s) = aij e−τj s . For the simplicity of analysis, we assume
c1 g i e ω ω
P
−2 i (t) ei (t) that τ1 = τ2 = · · · = τN = τ . Then the characteristic matrix
i=1
N 2 2 becomes:
c1 aij eω ω
P P
+ j (t) − e j (t − τj ) ∆(s) = sIN + c1 A0 − c1 e−τ s A (23)
i=1 j∈Ni
(18)
It is obvious that the spectrum of (23) is related to the time
Since the communication graph is balanced, one has
delay. We use the approximation technique in [31] to analyze
N X N X
X 2
X 2 the spectrum of (23), i.e., the eigenvalues of the discretization
c1 aij (eω
i (t)) = c1 aij eω
j (t) (19) matrix F defined as follows
i=1 j∈Ni i=1 j∈Ni " #
Applying (19) yields Ĥ ⊗ IN
F = (24)
N c1 A1 0 · · · 0 −c1 A0
2
c1 gi (eω
P
V̇ (t) = −2 i (t))
i=1 where Ĥ ∈ RM×(M+1) , and Ĥ is composed of the first M
N
+
P P
c1 aij − (eω
2 rows of the matrix H = − 2DτM ∈ R(M+1)×(M+1) . Define
i (t))
i=1 j∈Ni xk = cos( kπ
M ), k = 0, · · · , M . Then the entries of matrix DM
2 are
+2eω
i (t) eω
j (t − τj ) − eω
j (t − τj ) h (−1)i+j
hj (xi −xj ) ,i 6= j
i
N 2
c1 aij eω ω
P P
=− i (t) − ej (t − τj ) − xi 2 ,i = j 6= 0, M
2(1−xi )
i=1 j∈Ni D(i,j) =
N 2M 2 +1
2 ,i = j = 0
c1 gi (eω
P
−2 i (t))
6
i=1 2M 2 +1
(20) − 6 ,i = j = M
From (20), we get that V̇ (t) ≤ 0. The following two cases where h0 = hM = 2 and h1 = h2 = · · · = hM−1 = 1.
are discussed. As we can see from Fig.3, all the eigenvalues locate in the
Case 1: V̇ (t) ≡ 0 left half plane which means that the system is stable although
Since ∃ i such that gi 6= 0, from (20), we get eω i (t) = 0 the time delay is large enough. We also find that the larger the
and eω i (t) = e ω
j (t − τj ). Thus ω i (t) = ω j (t − τj = ωref ,
) time delay is, the more prominent the effect of time delays is.
i.e., the output frequency ωi (t) synchronizes to ωref .
Case 2: V̇ (t) < 0
In this situation, we get V (t) → 0, i.e., the output frequency B. Distributed Active Power Controller Design
ωi (t) converges to ωref asymptotically. This completes the The previous analysis shows that, to achieve the optimal
proof of Theorem 1. ⋄ active power sharing, one necessary and sufficient condition
Remark 2: At the steady state under the droop-based is to guarantee the incremental costs of DGs are equal, i.e.,
primary control, the incremental cost ηi (Pi ) defined in (5) the control objective can be written as
will be a constant, which can be denoted as ηi0 (Pi0 ). While
the secondary control input uω i is activated, the incremental η1 (P1 ) =η2 (P2 ) = · · · = ηN (PN ) = Ψ (25)
cost ηi (Pi ) will be a time varying variable during the
transient period of frequency regulation. From (13), we where Ψ is a constant value. Inspired by the secondary
can
R t directly calculate the second control input uω i (t) =
frequency controller (12) and (13), the optimal active power
controller is proposed as follows.
P
c
t0 1 [ a ij (ω j (t − τj ) − ω i (t)) − g i (ω i (t) − ω ref )]dt +
j∈Ni
kηi (Pi ) − kηi0 (Pi0 ). At the new steady state ωi = ωn − kηi (Pi ) + uP
i (26)
1949-3053 (c) 2017 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/TSG.2017.2785811, IEEE
Transactions on Smart Grid
5
X Consider the following candidate Lyapunov-Krasovskii func-
u̇P
i =c2
aij (ωj (t − τj ) − ωi (t)) − gi (ωi (t) − ωref ) tion
j∈Ni
N N X Z t
2 2
X X X
+ c2 kaij (ηj (Pj (t − τj )) − ηi (Pi (t))) V (t) = (xi (t)) +c2 aij (xj (s)) ds (31)
j∈Ni i=1 i=1 j∈Ni t−τj
(27)
where uP is the secondary active power control input; c Applying the similar analysis as in the proof of Theorem 1,
i 2 is a
positive gain; aij is the edge weight between node i and node we have
N N
j; τj is the constant communication delay between node j and X 2
X
node i; ηi (Pi ) is the incremental cost function of the i-th DG. V̇ (t) ≤ −2c2 gi (xi (t)) + c2 2gi (xi (t) (ωref + kηi (Pi )))
i=1 i=1
Theorem 2: Consider the system described by (26) with
N N
the control input uP i . Under Assumption 1, when the system
X 2
X 2
≤ −c2 gi (xi (t)) + c2 gi ((ωref + kηi (Pi )))
converges to the steady state, the objectives 1) and 2) can be
i=1 i=1
achieved even though the communication delays exist. (32)
Proof : Taking the time derivative on both sides of the It is seen that V̇ < 0 as xi (t)2 > (ωref + kηi (Pi (t)))2 .
equation (26) yields Thus, the system is stable according to Lyapunov theory.
′
ω̇i = −k η̇i (Pi ) + u̇P (28) Furthermore, at the steady state, i.e. (ωi + kηi (Pi )) = 0,
i
we have ωj (t − τj ) = ωj (t) and ηj (Pj (t − τj )) = ηj (Pj (t))
Substituting (27) into (28) yields because the frequency of all DGs will be equal at steady state.
′ Under Assumption 1 and (29), we have ωi (t) = ωref for
(ωi + kηi (Pi )) gi 6= 0. Therefore, we have ω1 = ω2 = · · · = ωN = ωref ,
P (ωj (t − τj ) + kηj (Pj (t − τj ))) i.e., the objective 1) is achieved.
= c2 aij
j∈Ni − (ωi (t) + kηi (Pi (t))) As ωj (t − τj ) = ωj (t) and ηj (Pj (t − τj )) = ηj (Pj (t)),
−c2 gi (ωi (t) − ωref ) from (29), we have
(29)
Define xi = ωi + kηi (Pi ). Then, we have X
X k η̇i (Pi ) = c2 k aij (ηj (Pj (t)) − ηi (Pi (t))) = 0 (33)
ẋi =c2 aij [xj (t − τj ) − xi (t)] − c2 gi xi j∈Ni
j∈Ni (30)
The compact form of the above equation can be written as
+ c2 gi (ωref + kηi (Pi ))
c2 kLη (P (t)) = 0. Then we have η1 (P1 ) = η2 (P2 ) = · · · =
ηN (PN ) and thus the objective 2) is achieved. Therefore the
proof of Theorem 2 is completed. ⋄
200 τ1 = 0.05s
τ2 = 0.1s
τ3 = 0.5s
C. Distributed Voltage and Reactive Power Controller Design
Imaginary axis
100
τ4 = 1s Due to the impact of line impedance, the voltage-reactive
τ5 = 10s
0 power droop controller is unable to realize the accurate voltage
τ6 = 100s
regulation and the accurate reactive power sharing among DGs
−100 simultaneously. Then we will design the voltage controller and
the reactive power controller separately.
−200
Case 1 Voltage Regulation
−120 −100 −80 −60 −40 −20 0 Motivated by the frequency controller (12) and (13), the
Real axis
voltage controller is designed as follows:
(a)
Vi = Vn − mi Qi + uVi (34)
15
10 X
u̇Vi =c3 aij (Vj (t − τj ) − Vi (t)) − gi (Vi (t) − Vref )
Imaginary axis
5 j∈Ni
0 + mi Q̇i
(35)
−5
where uVi is the secondary voltage control input; c3 is a
−10 positive constant; gi is a pinning gain by which the i-th DG
−15 is connected to the reference Vref defined by a virtual leader
−15 −10 −5 0 agent; aij is the edge weight between node i and node j; Vi
Real axis
denotes the output voltage of the i-th DG; mi is the voltage-
(b)
reactive power droop coefficient; Qi is the reactive power of
Fig. 3. The impact of time delays on root loci, τ ∈ [0.05, 100]s, M = 5, the i-th DG injected to power network; τj denotes the constant
Fig. 3(b) is the zooming in of Fig. 3(a). communication delay between node j and node i. We have
1949-3053 (c) 2017 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/TSG.2017.2785811, IEEE
Transactions on Smart Grid
6
the following results about the distributed secondary voltage Choose the Lyapunov-Krasovskii function as
controller (34) and (35). N
X 2 XN X Z t 2
Theorem 3: Consider the voltage control system described V (t) = eQ
i (t) + c a
4 ij e Q
j (s) ds
by (34) with the control input uVi . Under Assumption 1, the i=1 i=1 j∈Ni t−τj
voltage amplitude Vi of each DG synchronizes to the reference (40)
Vref even though the communication delays exist. The time derivative of V (t) is
The proof is similar to that of Theorem 1 and is omitted N
eQ Q
P
here. V̇ (t) = 2 i (t) ėi (t)
i=1
Case 2 Reactive Power Sharing N P
P
Q
2
Q
2
The distributed secondary reactive power controller can be +c4 aij ej (t) − ej (t − τj )
i=1 j∈Ni
designed as N P
aij eQ Q Q
P
mi Qi = Vn − Vi + uQ (36) = 2c4 i (t) e j (t − τj ) − e i (t)
i i=1 j∈Ni
N
2
2
P P Q Q
X +c4 aij ej (t) − ej (t − τj )
u̇Q
i = c4 aij (mj Qj (t − τj ) − mi Qi (t)) + V̇i (37) i=1 j∈Ni
j∈Ni (41)
For the balanced topology, one has
where c4 is a constant gain and the other parameters are the
N X 2 N X 2
same with (34). X X
c4 aij eQ
i (t) = c4 aij eQ
j (t) (42)
Theorem 4: Consider the system described by (36) with the
i=1 j∈Ni i=1 j∈Ni
control input uQi . Under Assumption 1, the accurate reactive
power sharing among DGs can be achieved even though the In light of (41) and (42), we have
communication delays exist. N X
X 2
Proof : Taking the time derivative of (36) and applying (37) V̇ (t) = −c4 aij eQ
i (t) − e Q
j (t − τj ) ≤ 0 (43)
yield i=1 j∈Ni
Vdc iL
Vo* * * Rc Lc
PI Voltage iL PI Current VL
controller controller Rf Lf
Cf
VSC Output
Outp vb
LC filiter connec
connector
vo io
wi
Power controller Pi Power
(Droop Control) Calculation
Qi
Primary Control
Switch Unit
U
å a (w ( t - t ) - w ( t ) ) - g (w ( t ) - w )
j ÎN i
ij j j i i i ref c1 c3 å a (V (t - t ) - V (t )) - g (V (t ) - V )
ij j j i i i ref
jÎNi
w V
1 u i u 1
khi (Pi (t)) å i
å mi Qi
s s
å a (w ( t - t ) - w ( t ) ) - g (w ( t ) - w )
ij j j i i i ref Vi
j ÎN i
1 uiP
å c2 uiQ
å a (h ( P (t -t )) -h ( P (t )))
ij j j j i i
s 1
s å c4 å a ( m Q (t - t ) - m Q (t ))
jÎNi
ij j j j i i
jÎNi
Secondary Frequency Control / Active power sharing Secondary Voltage Control / Reactive power sharing
Communication Network
j Î Ni
1949-3053 (c) 2017 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/TSG.2017.2785811, IEEE
Transactions on Smart Grid
7
TABLE I TABLE II
T HE DG S ’ ELECTRICAL PARAMETERS . T HE DG S ’ COST PARAMETERS .
Rline1 Rline 3
DG1 DG4
Remark 3: The control gains c1 , c2 , c3 and c4 are used Lline1 Lline 3
DG2 DG3
to tune the convergence rates of the secondary frequency, Load 2 Load 3
1949-3053 (c) 2017 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/TSG.2017.2785811, IEEE
Transactions on Smart Grid
8
390 390
385 385
380
380
Vo(V)
Vo(V)
360 365
2 4 6 8 10 12 5 10 15 20 25 30 35 40 45 50
Time(sec) Time(sec)
(a) (a)
315 315
314.5 314.5
314 314
ω(rad/s)
ω(rad/s)
313.5 DG1 313.5 DG1
DG2 DG2
313 313 DG3
DG3
DG4
312.5 DG4 312.5
312 312
2 4 6 8 10 12 5 10 15 20 25 30 35 40 45 50
Time(sec) Time(sec)
(b) (b)
4 4
x 10 x 10
8 9
DG1
8 DG2
7
DG3
Active Power(W)
Active Power(W)
7 DG4
6
6
DG1
5
DG2 5
DG3
4 4
DG4
3 3
2 4 6 8 10 12 5 10 15 20 25 30 35 40 45 50
Time(sec) Time(sec)
(c) (c)
20
20
The incremental costs η($/KWh)
η1
The incremental costs η($/KWh)
η
1
η η
2 15 2
15
η3 η
3
η η
4 4
10 10
5 5
0 0
2 4 6 8 10 12 5 10 15 20 25 30 35 40 45 50
Time(sec) Time(sec)
(d) (d)
Fig. 6. The asymptotical tracking performance without time-delay (under Fig. 7. The asymptotical tracking performance with large-delays (under the
the controllers (27) and (35)). (a) the output terminal voltage; (b) the output controllers (27) and (35)). (a) the output terminal voltage; (b) the output
frequency; (c) the output active power; (d) the incremental cost value. frequency; (c) the output active power; (d) the incremental cost value.
respectively. By comparing Figs.7(a)-(c) with Figs. 6(a)-(c), Stage2(3-13s): Secondary control is activated.
it is seen that the corresponding convergence rates become Stage3(13-20s): Constant load Lc = 36×103 +j10×103W
slower, which clearly shows the impacts of time delays on the is added to Load 2.
performances of the secondary control. Stage4(20-30s): Load Lc is removed from Load 2.
CASE C. To implement our proposed control protocols with
small-delays The frequency response is shown in Fig.9(b). During t =
The performances of the active power controller (27), the 3 − 13s, the frequency is restored to the reference value under
voltage controller (35), and the reactive power controller (37) the action of secondary control. The steady state frequency of
are test in this case, respectively. The comparatively small the microgrid remains at 314.16rad/s no matter new constant
delay values are chosen as τ1 = τ2 = τ3 = τ4 = 0.05s. load Lc is connected to or disconnected from DG1, even
The communication weight is set as aij = 1 except for though there are transient deviations as shown in Fig. 9(b).
a34 = a43 = 2. The other simulation parameters are the same During t = 13−20s, the increased output active power is equal
with CASE A. The simulation is divided into 4 stages: to the added load and the demand and supply is balanced.
Stage1(0-3s): Only the primary control is applied. During the stages 2- 4, Fig. 9(d) shows that the incremental
1949-3053 (c) 2017 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/TSG.2017.2785811, IEEE
Transactions on Smart Grid
9
Vo(V)
voltage regulation under the voltage controller (35). The DG1
375
reactive power sharing shown in Fig.9(e) is poor because the DG2
DG3
370
precise voltage regulation will lead to the errors in reactive DG4
9(a) and Fig. 9(e) respectively, it indicates that there exists 314
ω(rad/s)
a conflict between the accurate voltage regulation and the 313.5 DG1
DG2
accurate reactive power sharing. 313 DG3
DG4
312.5
312
390 5 10 15 20 25 30
Time(sec)
385
(b)
380 4
x 10
Vo(V)
10
375 DG1
9 DG2
370 DG3
Active Power(W)
DG1 8
DG2 DG4
365 DG3 7
DG4 6
360
5 10 15 20 25 30
Time(sec) 5
4
(a) 3
5 10 15 20 25 30
4 Time(sec)
x 10
1.8
DG1
DG2
(c)
1.6
Reactive Power(VAr)
DG3
20
η1
The incremental costs η($/KWh)
1.4 DG4
η2
1.2 15
η3
1 η4
10
0.8
0.6 5
5 10 15 20 25 30
Time(sec)
0
5 10 15 20 25 30
(b) Time(sec)
Fig. 8. The asymptotical tracking performance with small-delays (under the (d)
controllers (27) and (37)). (a) the output terminal voltage; (b) the output 4
x 10
2.5
reactive power. DG1
DG2
Reactive Power(VAr)
2
DG3
DG4
Since the dynamics of the voltage controller, the current 1.5
controller and the LC filter are much faster than the dynamics 1
of the power controller, we have neglected the fast dynamics of 0.5
the voltage and current controllers and the LC filter during our
0
modeling process. However, during our simulation process, we 5 10 15
Time(sec)
20 25 30
0
neglect the dynamics of voltage and current controller and LC
filter. Compared with Fig.9, it shows our proposed modeling −200
approach is acceptable and effective.
−400
As discussed in Remark 3, the controller gains c2 , c3 , c4 2.9 2.95 3
Time(sec)
3.05 3.1
1949-3053 (c) 2017 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/TSG.2017.2785811, IEEE
Transactions on Smart Grid
10
Reference Reference
power, incremental cost, and output reactive power curves have value value
390
385
380
DG2 DG3 DG2 DG3
Vo(V)
375
DG1 (a) (b)
370 DG2
DG3
365 DG4
Fig. 11. The communication topology with communication link failure
360
5 10 15 20 25 30 between DG1 and DG4
Time(sec)
(a)
tion topology is changed from Fig. 11(a) to Fig. 11(b) at t =
315
10s and then recovered at t=15s. That is, the communication
314.5 link between DG4 and DG1 is disconnected at =10s and then
reconnected at t = 15s. The communication weight among DGs
ω(rad/s)
314 DG1
DG2 is set as aij = 1. The delay values, controllers and simulation
DG3
313.5 DG4 parameters are the same as that of Case C. The simulation
results in Fig. 14 show that our proposed distributed secondary
313
5 10 15
Time(sec)
20 25 30 control protocols enjoy good robustness performances against
(b)
communication link failures.
4
x 10
The plug-and-play functionality of the microgrid system
8
DG1 was tested by disconnecting DG2 at t = 10s and reconnecting
DG2
7
DG3 it at t = 15s. As DG2 is disconnected, the communication
Active Power(W)
DG4
6 topology is changed from Fig.13(a) to Fig.13(b). The simula-
5 tion parameters are the same as that of Case C. Fig. 15 shows
4
the simulation results. It is shown that the DGs adaptively
adjust the frequency, voltage, and active power outputs with
3
5 10 15
Time(sec)
20 25 30 the plug-and-play of DGs.
CASE E. Comparison study
(c)
In this case, we make a comparison with the distributed
20
secondary control algorithm provided in [16]. Under the same
The incremental costs η($/KWh)
η1
η2
15
η3
control parameters chosen in Case C, we run the algorithm
η4 proposed in [16] and get the simulation results as shown in Fig.
10
16. Fig. 16(b) shows that the method in [16] realizes the active
5 power sharing among DGs according to droop coefficients.
Fig.16(c) shows the method in [16] cannot achieve the equal
0
5 10 15
Time(sec)
20 25 30 incremental cost. Our results shown in Fig. 6(d) and Fig. 9(d)
achieve the consensus incremental cost, i.e., the optimal active
(d)
4
power sharing objective (9) is achieved.
x 10
2.5
DG1
To further test the time delay impact on the performance of
DG2 active power sharing, the active power controller proposed by
Reactive Power(VAr)
2 DG3
DG4 [16] is revised as folllows
1.5 X
uPi =c
P
aij (nj Pj (t − τj ) − ni Pi (t − τi )) (44)
1 j∈Ni
0.5
5 10 15 20 25 30
The delay values are chosen as τ1 = τ2 = τ3 = τ4 = τ5 =
Time(sec)
0.095s, which is smaller than the delay value 1s considered
(e) in Fig. 7. The simulation result is shown in Fig.16(d). It
Fig. 10. The asymptotical tracking performance with small-delays under is seen that the control systems become unstable while the
our proposed model (under the controllers (27) and (35)). (a) the output communication delays are considered.
terminal voltage; (b) the output frequency; (c) the output active power; (d)
the incremental cost value; (e) the output reactive power.
V. C ONCLUSION
CASE D. To test the robustness of our proposed control In this paper, we consider the effects of time delays and
protocols against communication link failures and DGs’ plug- provide the secondary control schemes to i) restore the fre-
and-play quency of each DG to the reference value; ii) achieve the
To test the robustness of our proposed protocols against optimal active power sharing; iii) restore the output voltage
communication link failures, we assume that the communica- of each DG to the reference value; iv) achieve the accurate
1949-3053 (c) 2017 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/TSG.2017.2785811, IEEE
Transactions on Smart Grid
11
Reference Reference
value value
385
380
Vo(V)
360 Fig. 13. The communication topology with the disconnected DG2
2 4 6 8 10 12 14 16 18 20 22
Time(sec)
(a)
315
314.5
314
ω(rad/s)
390
313.5 DG1
DG2 385
313 DG3
DG4
312.5 380
Vo(V)
312 DG1
2 4 6 8 10 12 14 16 18 20 22 375
DG2
Time(sec) DG3
370 DG4
(b)
365
x 10
4 2 4 6 8 10 12 14 16 18 20
10 Time(sec)
DG1
9 DG2
DG3
(a)
Active Power(W)
8
DG4
315
7
6 314.5
5 314
ω(rad/s)
4 313.5 DG1
3 DG2
2 4 6 8 10 12 14 16 18 20 22 313 DG3
Time(sec)
DG4
312.5
(c)
312
2 4 6 8 10 12 14 16 18 20
20 Time(sec)
η1
The incremental costs η($/KWh)
η2
15 (b)
η3
4
η4 8
x 10
10 DG1
DG2
7 DG3
Active Power(W)
5 DG4
6
0 5
2 4 6 8 10 12 14 16 18 20 22
Time(sec)
4
(d)
4
3
x 10 2 4 6 8 10 12 14 16 18 20
3 Time(sec)
DG1
2.5 DG2
(c)
Reactive Power(VAr)
DG3
2 DG4
20
The incremental costs η($/KWh)
η
1
1.5
η2
15
1 η3
0.5 η4
10
0
2 4 6 8 10 12 14 16 18 20 22
Time(sec) 5
(e)
0
2 4 6 8 10 12 14 16 18 20
Time(sec)
Fig. 12. The asymptotical tracking performance with small-delays and control
gains c2 = c3 = c4 = 50 (under the controllers (27) and (35)). (a) the output (d)
terminal voltage; (b) the output frequency; (c) the output active power; (d)
the incremental cost value; (e) the output reactive power. Fig. 14. The robustness of our proposed method under the communication
failure between DG1 and DG4 (under the controllers (27) and (35)). (a) the
output terminal voltage; (b) the output frequency; (c) the output active power;
(d) the incremental cost value.
1949-3053 (c) 2017 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/TSG.2017.2785811, IEEE
Transactions on Smart Grid
12
4
390 x 10
10
DG1
385 DG2
8 DG3
Active Power(W)
380 DG4
Vo(V)
375 6
DG1
370 DG2
DG3
4
365 DG4
360 2
2 4 6 8 10 12 14 16 18 20 22 2 4 6 8 10 12 14 16 18 20 22
Time(sec) Time(sec)
(a) (a)
316
m1*P1
315.5 DG1 1.6
DG2 m2*P2
315 DG3 1.4 m3*P3
DG4
ω(rad/s)
314.5 m4*P4
mi*Pi
1.2
314
1
313.5
0.8
313
0.6
312.5
2 4 6 8 10 12 14 16 18 20 22 2 4 6 8 10 12 14 16 18 20 22
Time(sec) Time(sec)
(b) (b)
4
x 10 20
10
8 η2
15
η3
Active Power(W)
6 η
4
10
4
DG1
DG2 5
2 DG3
DG4
0 0
2 4 6 8 10 12 14 16 18 20 22 5 10 15 20
Time(sec) Time(sec)
(c) (c)
5
20 x 10
η1
The incremental costs η($/KWh)
3
η2
15 2
η3
Active Power(W)
η4 1
10
0
DG1
5 DG2
−1
DG3
DG4
0 −2
2 4 6 8 10 12 14 16 18 20 22 2 4 6 8 10 12 14 16 18 20 22
Time(sec) Time(sec)
(d) (d)
Fig. 15. The DGs’ plug-and-play property with disconnected and reconnected Fig. 16. Comparison between the proposed method and the approach in
DG2 (under the controllers (27) and (35)). (a) the output terminal voltage; [16].(a) the output active power without communication delays; (b) the active
(b) the output frequency; (c) the output active power; (d) the incremental cost power sharing; (c) the incremental cost value; (d) the output active power
value. with communication delays.
1949-3053 (c) 2017 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/TSG.2017.2785811, IEEE
Transactions on Smart Grid
13
Control, vol. 49, no. 9, pp. 1520-1533, Sep. 2004. Gang Chen(M’16) received the Ph.D. degree
[12] W. Wang and J. J. E. Slotine, “Contraction analysis of time-delayed in control engineering from Zhejiang University,
communications and group cooperation,” IEEE Trans. Autom. Control, Hangzhou, China, in 2006. Since 2006, he has been
vol.51, no. 4, pp. 712-717, Apr. 2006. with the College of Automation, Chongqing Univer-
[13] J. W. Simpson-Porco, Q. Shafiee, F. Dörfler, et al., “Secondary frequency sity, Chongqing. From 2009 to 2010, he was a visit-
and voltage control of islanded microgrids via distributed averaging,” ing scholar at the Automation and Robotics Research
IEEE Trans. Ind. Electron., vol. 62, no. 11, pp. 7025-7038, Nov. 2015. Institute, the University of Texas at Arlington. He is
[14] A. Bidram, A. Davoudi, F. L. Lewis, et al., “Distributed cooperative currently a professor at the College of Automation,
secondary control of microgrids using feedback linearization,” IEEE Chongqing University. His research interests include
Trans. Power Syst., vol. 28, no. 3, pp. 3462-3470, Aug. 2013. distributed control, cooperative control, intelligent
[15] A. Bidram, A. Davoudi, and F. L. Lewis, “A multiobjective distributed control, cyber-physical system, smart grid, nonlinear
control framework for islanded ac microgrids,” IEEE Trans. Ind. Infor- control and control applications.
mat., vol. 10, no. 3, pp. 1785-1798, Aug. 2014.
[16] A. Bidram, A. Davoudi, F. L. Lewis, et al., “Secondary control of
microgrids based on distributed cooperative control of multi-agent
systems,” IET Gener. Transm. Distrib., vol. 7, no. 8, pp. 822-831, Aug.
2013.
[17] F. Guo, C. Wen, J. Mao, et al., “Distributed secondary voltage and
frequency restoration control of droop-controlled inverter-based micro-
grids,” IEEE Trans. Ind. Electron., vol. 62, no. 7, pp. 4355-4364, Jul.
2015.
[18] J. W. Simpson-Porco, F. Dörfler, and F. Bullo, “Synchronization and
power sharing for droop-controlled inverters in islanded microgrids,”
Automatica, vol. 49, no. 9, pp. 2603-2611, Sep. 2013.
[19] X. Wu, C. Shen, and R. Iravani, “A distributed, cooperative frequency
and voltage control for microgrids,“ IEEE Trans. Smart Grid, DOI
10.1109/TSG.2016.2619486, 2016.
[20] X. Lu, X. Yu, J. Lai, et al., “A novel distributed secondary coordination
control approach for islanded microgrids,” IEEE Trans. Smart Grid, DOI
10.1109/TSG.2016.2618120, 2016.
[21] J. Schiffer, T. Seel, J. Raisch, et al., “Voltage stability and reactive power
sharing in inverter-based microgrids with consensus-based distributed
voltage control,” IEEE Trans. Control Syst. Technol., vol. 24, no. 1, pp.
96-109, Jan. 2016.
[22] J. Lai, H. Zhou, X. Lu, et al., “Distributed power control for DERs
based on networked multiagent systems with communication delays,”
Neurocomputing, vol. 179, pp. 135-143, 2016.
[23] E. A. A. Coelho, D. Wu, J. M. Guerrero, et al., “Small-signal analysis
of the microgrid secondary control considering a communication time
delay,” IEEE Trans. Ind. Electron., vol. 63, no. 10, pp. 6257-6269, Oct.
2016.
[24] S. Liu, X. Wang, and P. X. Liu, “Impact of communication delays on Zhijun Guo was born in Jiangxi, China, in 1991.
secondary frequency control in an islanded microgrid,” IEEE Trans. Ind. He received the M.S. from Chongqing University,
Electron., vol. 62, no. 4, pp. 2021-2031, Apr. 2015. Chongqing, China, in 2017. He is currently pursu-
[25] J. Schiffer, R. Ortega, A. Astolfi, et al., “Conditions for stability of ing the Ph.D degree in control theory and control
droop-controlled inverter-based microgrids,” Automatica, vol. 50, no. 10, engineering from the same university. His current re-
pp. 2457-2469, Oct. 2014. search interests include distributed cooperative con-
[26] Z. Qu, J. Wang, and R. A. Hull, “Cooperative control of dynamical trol, time-delay system, finite-time control, event-
systems with applications to autonomous vehicles,” IEEE Trans. Autom. triggered control and their applications in micro-
Control, vol. 53, no. 4, pp. 894-911, May. 2008. grids.
[27] N. Chopra and M. W. Spong, “Passivity-based control of multi-agent
systems,“ In Advances in Robot Control, Springer Berlin Heidelberg,
pp. 107-134, 2006.
[28] H. Xin, L. Zhang, Z. Wang, et al., “Control of island ac microgrids
using a fully distributed approach,” IEEE Trans. Smart Grid, vol. 6, no.
2, pp. 943-945, Mar. 2015.
[29] J. Schiffer, E. Fridman, R. Ortega, et al, “Stability of a class of delayed
port-Hamiltonian systems with application to microgrids with distributed
rotational and electronic generation,” Automatica, vol. 74, pp. 71-79,
Dec. 2016.
[30] C. Ahumada, R. Cárdenas, D. Sáez, et al., “Secondary control strategies
for frequency restoration in islanded microgrids with consideration of
communication delays,” IEEE Trans. Smart Grid, vol. 7, no. 3, pp. 1430-
1441, May. 2016.
[31] F. Milano, ”Small-Signal Stability Analysis of Large Power Systems
With Inclusion of Multiple Delays,” IEEE Trans. Power Systems, vol.
31, no. 4, pp. 3257-3266, Jul. 2016.
1949-3053 (c) 2017 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.