You are on page 1of 13

This article has been accepted for publication in IEEE Transactions on Power Systems.

This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2023.3274932

IEEE TRANSACTIONS ON POWER SYSTEMS 1

Distributed Cooperative Control Strategy for Stable


Voltage Restoration and Optimal Power Sharing in
Islanded DC Microgrids
Yuji Zeng, Graduate Student Member, IEEE, Qinjin Zhang, Member, IEEE, Yancheng Liu, Haohao
Guo, Member, IEEE, and Fengkui Zhang

Abstract—Conventionally, voltage regulation and economic In [4], a fully distributed secondary control framework for AC
dispatch in DC microgrid are performed at different time scales, microgrids is proposed, which regulates frequency, voltage, and
which cannot keep the distributed generators (DGs) at the optimal active power by designing three secondary controllers and does
operating point in real time. In this paper, a novel distributed not require additional frequency measurement devices.
cooperative control strategy is proposed to synchronously achieve
stable bus voltage restoration and optimal power sharing in
However, designing multiple PI-based regulation secondary
islanded DC microgrids. The proposed strategy breaks the controllers increases the system order, which tends to lead to
constraints of hierarchical control strategies, and incorporates the system instability. To reduce the communication burden among
objectives of voltage regulation and economic dispatch into a DGs, an event-triggered distributed secondary control is
distributed cooperative control framework to eliminate average proposed in [5] to achieve restoration of frequency and voltage
bus voltage deviation and minimize total power generation costs in in islanded AC microgrids with accurate active power sharing.
real time. The stability of the proposed strategy is verified by However, the control law of local DGs in this method is updated
eigenvalue analysis and steady-state analysis. Furthermore, the frequently at the triggering moments of neighbors, which may
concept of “incremental cost (IC) margin” is introduced in lead to high-frequency oscillations of voltage and current. To
cooperative control considering capacity constraints to avoid the
converter output power violation and performance degradation.
enhance the system operation performance, a finite-time
Finally, the MATLAB/Simulink simulation and StarSim HIL consensus-based secondary frequency control strategy is
experimental results verify the effectiveness and robustness of the proposed in [6], which achieves frequency regulation in finite
proposed strategy in an islanded DC microgrid under various time. However, the bounded settling time of the finite-time
testing scenarios. secondary controller depends on the initial state of the system.
If the initial state information is inaccurate or unavailable, it
Index Terms—DC microgrid, distributed cooperative control, would inevitably lead to controller failure.
economic dispatch, bus voltage restoration, consensus algorithm. For DC microgrids, related research works has mostly
focused on proportional power sharing among DGs and bus
I. INTRODUCTION voltage restoration in DC microgrids [7]-[10]. For example, an

R ECENTLY, the objective of “carbon peaking and carbon


neutrality” has driven the low-carbon transformation of the
energy and power industry. Large-scale deployment of DGs
improved secondary control strategy is proposed in [7] to
achieve load power sharing and average bus voltage restoration
through the coordination between multiple average controllers,
with DC characteristics, such as fuel cells, PV arrays, and but requires monitoring the states of all DGs to calculate the
battery storage units, and aggregation of them with electricity relevant average information. Considering reducing the system
communication burden, a distributed cooperative control
loads to form a microgrid, is considered to be an effective
strategy based on virtual voltage drop consensus is developed
solution for energy conservation and emission reduction [1].
in [8], which achieves voltage restoration and power sharing by
Intuitively, DC microgrid are free of harmonic suppression,
obtaining global consensus of the virtual voltage drop, but
frequency synchronization, and reactive power compensation, requires real-time acquisition of line impedance information.
and has more efficient power and load integration functions, so To reduce the control loop, a distributed cooperative controller
it is widely used in ships, islands, data centers, etc. [2]. with integrated voltage restoration and power sharing is
Categorically, the main control objectives in islanded AC or designed in [9]. On this basis, a distributed cooperative
DC microgrids are voltage/frequency regulation and active controller based on the principle of droop slope regulation and
power dispatch [3]. For AC microgrids, frequency/voltage reference voltage shifting for parallel battery storage units is
recovery and active power sharing are widely studied [4]-[6]. designed in [10] to achieve dynamic State-of-Charge balancing,
This work was supported in part by the National Natural Science Foundation
proportional load power sharing, and average bus voltage
of China under Grant 51979021 and Grant 51709028, in part by the China restoration. However, references [4]–[10] all focus on obtaining
Postdoctoral Science Foundation under Grant 2021M700646, and in part by the proportional power sharing according to the nominal capacity
Dalian High-level Talent Innovation Support Program Project under Grant of DGs. Unfortunately, DGs usually have different generation
2019RQ008. (Corresponding author: Qinjin Zhang.)
The authors are with the College of Marine Engineering, Dalian Maritime
costs, so proportional power sharing based on DGs capacity is
University, Dalian 116026, People's Republic of China (e-mail: not an economical solution.
zengyuji@foxmail.com; zqj20@dlmu.edu.cn; liuyc@dlmu.edu.cn; ghh1984@ Conventionally, hierarchical control is often used to achieve
dlmu.edu.cn; zhangfengkui@dlmu.edu.cn). voltage regulation and economic dispatch in islanded DC

© 2023 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.See https://www.ieee.org/publications/rights/index.html for more information.
Authorized licensed use limited to: Murdoch University. Downloaded on May 14,2023 at 15:52:18 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in IEEE Transactions on Power Systems. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2023.3274932

IEEE TRANSACTIONS ON POWER SYSTEMS 2

TABLE I difficult to find the optimal working point. To improve system


COMPARISON OF ECONOMIC DISPATCH STRATEGIES IN MICROGRIDS
robustness, a distributed event-triggered hierarchical control
strategy is proposed in [18] to achieve economic operation of
Number of
Real-time Stability Voltage hybrid AC/DC microgrids, which achieves average bus voltage
Methods communication
optimization analysis restoration
variables restoration and optimal power sharing by designing two
[12]-[15] ✘ ✘ ✘ Many different secondary controllers, but the control structure is more
[16] ✔ ✘ ✘ Two
complex. For DC microgrids, Ref. [19] combined the equal IC
criterion with the subgradient algorithm to develop a distributed
[17] ✘ ✘ ✘ Three
economic dispatch method, which can make full use of
[18] ✔ ✘ ✔ Three renewable energy and reduce the power generation cost of DGs,
[19] ✔ ✘ ✔ One but this method has poor dynamic performance. Considering
[20] ✔ ✘ ✔ Two the improvement of system dynamic response, a hierarchical
cooperative control strategy is proposed in [20] to solve the
[21] ✔ ✘ ✔ Three
problems of voltage restoration and optimal power sharing in
[22] ✔ ✔ ✘ Three real time, but this method relies on complex topology models.
[23] ✔ ✔ ✔ Two On this basis, Ref. [21] extends the research object to DC
Proposed
✔ ✔ ✔ One
microgrid clusters, and realizes voltage regulation and power
method optimization through the coordination between two-layer cyber
networks and multi-level controllers, but a rigorous stability
microgrids. In which, the secondary controller is used for bus proof is not given. Considering the output voltage constraint, an
voltage restoration, whereas the tertiary controller obtains the optimal controller capable of minimizing the generation cost
optimal power sharing, thereby improving the system power and bounded bus voltage regulation is designed in [22].
quality and minimizing the total power generation cost [11]. However, the controller design process is complex and sensitive
Essentially, tertiary controller is a pre-dispatch problem that to communication delays. To reduce the system order, a
requires consideration of DGs capacity constraints and cascaded controller capable of synchronously achieving
allocating the predicted total load demand to DGs based on accurate voltage regulation and optimal power sharing is
generation costs for optimal power sharing. Various designed in [23]. However, this method requires access to
optimization algorithms, such as lambda iteration method [12], common bus voltage information, and the influence of
gradient search method [13], genetic algorithm [14], particle parameter changes on the system stability has not been analyzed.
swarm optimization algorithm [15], have been developed to In summary, related research works require multiple
solve economic dispatch problems in power systems. However, controllers to eliminate the bus voltage deviation and minimize
these optimization methods rely on a central controller to the total power generation cost, respectively. Intuitively, the
handle large amounts of information, which inevitably brings introduction of multiple controllers would inevitably bring
expensive investment costs and single point of failure problems. higher system order, which makes the system stability face
Motivated by the above observations, voltage control and great challenges. Moreover, when the output power of DGs is
power optimization in related research works are usually violated, the above control schemes all constrain its output
performed at different time scales. However, the disconnect power at the power limit, which may deteriorate the converter
between control and optimization makes it difficult to respond performance. Considering the above-mentioned concerns, a
to frequent load power fluctuations in real time, and even small distributed cooperative control strategy is proposed to achieve
disturbances can cause bus voltage and output power to deviate stable average bus voltage restoration and optimal output power
from the optimal operating point. Therefore, economic dispatch sharing in islanded DC microgrids, and avoid DGs power
should also be performed with a real-time control manner. violations. Table I provides a comparison between some
Specifically, output power is optimized by matching ICs economic dispatch strategies and the proposed control strategy
between neighboring DGs, and obeying generation-demand to highlight the advantages of the proposed scheme. The main
power balance and capacity constraints [13]. To solve the real- contributions of this paper are summarized as follows:
time optimization problem of economic dispatch process, many 1) A novel distributed cooperative controller is designed,
scholars have completed various explorations [16]-[23]. In [16], which can unify the objectives of voltage regulation and
an economic dispatch strategy for smart grids based on economic dispatch into a cooperative controller based on
distributed dynamic planning algorithms is proposed, which integral regulation for real-time implementation without
formulates the discrete economic dispatch problem as a designing multiple PI-based secondary controllers, which
conventional knapsack problem solved with distributed greatly simplifies the control structure and improves the system
dynamic planning algorithms to optimally allocate the total stability.
power demand among different DGs. However, this method 2) The proposed strategy is implemented based on a multi-
requires high computational ability of the controller and is agent consensus framework and requires only an average
difficult to be applied in practice. In [17], a two-stage economic virtual state variable to be transmitted between neighboring
dispatch strategy based on leader-follower consensus algorithm DGs, which effectively reduces the system communication
and adaptive weight adjustment technique is proposed, which is burden. In addition, the distributed cooperative controller is
able to achieve incremental cost consensus among DGs in the able to respond quickly to perturbations, thus improving the
presence of unknown communication uncertainty. However, voltage and current regulation performance.
the disconnection between control and optimization makes it 3) The concept of “IC margin” is proposed, which constrains

© 2023 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.See https://www.ieee.org/publications/rights/index.html for more information.
Authorized licensed use limited to: Murdoch University. Downloaded on May 14,2023 at 15:52:18 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in IEEE Transactions on Power Systems. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2023.3274932

IEEE TRANSACTIONS ON POWER SYSTEMS 3

buses, yij=yji=0.
B. Graph Theory and Average Consensus Algorithms
The communication network in Fig. 1 can be represented by
an undirected graph G=(V, E, A), where V={DG1, DG2, …,
DGN} is the node set, E⊆V×V is the edge set [24]. DGi and DGj
are neighbor nodes if and only if (DGi, DGj)∈E, then DGi can
transmit information to DGj. Considering the graph G is
undirected, (DGi, DGj)∈E⇔(DGj, DGi)∈E. The relationship
between nodes and edges in graph G can be described by a
weighted adjacency matrix A=[aij]∈RN×N, where aij is the
communication weight between nodes. If (DGi, DGj)∈E is
Fig. 1. Islanded DC microgrid structure with multiple DGs. satisfied, then aij=1, otherwise aij=0. In addition, the weighted
degree of DGi is di. Therefore, the Laplacian matrix L=D–
the ICs to the limit warning threshold rather than the limit value A=[lij]∈RN×N is:
when the output power of DGs is violated, thereby preventing lii   N aij , i  j
j 1, j i
the converter from working at the power limit for a long time.  (2)
Therefore, the performance of the DGs converter is improved. lij  aij , i j
4) The effectiveness and robustness of the proposed control Without loss of generality, the dynamic average can be
strategy are verified by rigorous stability analysis, steady-state achieved by an average consensus algorithm [25]:
analysis, simulation, and experimental results. xi (t )    aij  xi (t )  x j (t )   ri (t ) (3)
The rest of this paper is organized as follows. Section II j N i
introduces the islanded DC microgrid model, knowledge of where xi(t) and xj(t) are the state variables of DGi and DGj,
graph theory and average consensus algorithms, and the respectively. ri(t) is a local monitoring variable.
economic dispatch problem. In Section III, a novel distributed To obtain global dynamics, rewrite (3) in compact form:
cooperative control strategy is proposed, and rigorous x (t )   Lx (t )  r(t ) (4)
eigenvalue stability analysis and steady-state analysis are T T
where x=[x1, x2, …, xN] and r=[r1, r2, …, rN] are vectors of
completed. Then, simulation tests for six different operating state variables and local monitoring variables, respectively.
conditions is completed in Section IV. Moreover, the StarSim
HIL experimental results are given in Section V. Finally, this C. Economic Dispatch Problem
paper is concluded in Section VI. The economic dispatch problem (EDP) is one of the
important problems in DC microgrids, which is often described
II. PRELIMINARIES AND PROBLEM FORMULATION as an optimization problem [26]. The purpose of solving the
EDP is to find the best combination of output power of DGs and
A. Electrical and Communication Networks to minimize the total generation cost under the premise of
The islanded DC microgrid structure containing multiple obeying generation-demand power balance and capacity
DGs is shown in Fig. 1, which is mainly composed of electrical constraints. In this work, the EDP of dispatchable DGs (e.g.
and communication networks. In the electrical network, each conventional generators, fuel cells, batteries) is considered.
power generation unit and the boost DC-DC converter together Without loss of generality, the power generation cost function
form a DG, and each DG is connected to the local load through of DGs can be approximated by a quadratic function [13]:
a DC bus. Multiple DC bus is connected in series through tie Ci ( Pi )  ai Pi 2  bi Pi  ci (5)
lines to ensure that the power generated by DGs is transmitted where Ci(Pi) is the power generation cost of DGi. Pi is the output
to each load. In the communication network, a neighbor-to- power of DGi. ai, bi, and ci are the coefficients of the cost
neighbor communication topology is selected to share function.
information between neighboring DGs. RLi represents the local Considering the generation-demand power balance and the
load of DGi, rij represents the tie line resistance between DGi maximum/minimum capacity constraints of DGs, the economic
and DGj, the green dashed arrow is the communication link and dispatch problem of DGs can be described as:
min C ( P)  min  i 1 Ci ( Pi )
N
the blue dashed arrow is the energy flow direction between DGs.
To simplify the analysis, the distribution lines are assumed
s.t.  i 1
to be resistive [23]. Let uoi and ioi be the output voltage and  N Pi  PD  PL  PT (6)
current of DGi connected to Busi, respectively. From Fig. 1, the
 Pmin,i  Pi  Pmax,i
relationship between voltage and current is obtained based on
Kirchhoff's Current Law (KCL): where PD and PL are the total load demand power and the total
system line loss, respectively. PT is the total system demand
ioi   y u
jNi
ij oi  uoj   iLi , i  1, 2, , N (1) power. Pmin,i and Pmax,i are the minimum and maximum output
power of DGi, respectively.
where Ni is the neighboring bus of Busi and yij=yji is the If the inequality constraints of (6) are ignored, the augmented
conductance of the distribution line between Busi and Busj, Lagrange function can be used to minimize the total power
yij=1/rij. If there is no physical connection between the two generation cost [13]:

© 2023 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.See https://www.ieee.org/publications/rights/index.html for more information.
Authorized licensed use limited to: Murdoch University. Downloaded on May 14,2023 at 15:52:18 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in IEEE Transactions on Power Systems. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2023.3274932

IEEE TRANSACTIONS ON POWER SYSTEMS 4

N

L( Pi ,  )   i 1 Ci ( Pi )   PT   i 1 Pi
N
 (7)
where λ is the Lagrange multiplier.
From (5), the cost function is strictly convex and
differentiable, and the equality constraints of (6) are convex
functions. Therefore, the conditional extremum problem would
be transformed into an unconstrained optimization problem,
and the solution of the EDP can be obtained by solving (7):
 L
P   i 1  2ai Pi  bi     0
N

i
 (8)
L
  PT  N Pi  0
   i 1
According to the equal IC criterion [13], the optimal IC of
DGi can be obtained from (8) as:
N bi
PT   i 1
2ai
opt,i    (9)
1
 i 1 2ai
N

Furthermore, the optimal output power of DGi is: Fig. 2. Distributed cooperative control structure for DGi.
N bi 1
PT   i 1  bi  i 1
N
voltage calculated locally at the DGi. λi(t) and λj(t) are the ICs
2ai 2ai
Popt,i  (10) of DGi and DGj, respectively.
N 1
ai  i 1 Remark 1: For large-scale DC microgrids, it is difficult to
ai obtain the bus voltage at the common terminal due to factors
However, if the maximum/minimum capacity constraints are such as long transmission distance, high measurement cost, and
considered, the optimal solution can be extended to [27]: low sensor accuracy. Therefore, restoring the average bus
2ai Popt,i  bi  opt,i , for Pmin,i  Popt,i  Pmax,i voltage to the nominal value is a feasible solution [8].
 Furthermore, according to the equal IC criterion [13], the
2ai Popt,i  bi  opt,i , for Popt,i  Pmax,i (11)
optimal output power sharing of DGs can be achieved if and
2ai Popt,i  bi  opt,i , for Popt,i  Pmin,i
 only if ICs converge to the same level.
From (9) and (10), if the total load demand power PT can be Herein, a novel distributed cooperative control strategy is
predicted, the total power generation cost should be easily developed to achieve the two control objectives of (12), where
minimized with a centralized control manner, which is the the control structure of DGi is shown in Fig. 2. Without loss of
conventional practice of the centralized economic dispatch generality, a unidirectional boost DC/DC converter is chosen as
scheme. However, due to the power fluctuation of renewable the connection module between the DGi and the DC bus. In the
energy and loads, the above scheme may often cause DGs to primary control layer, droop control and voltage-current control
deviate from the predicted optimal operating point, so the allow for basic voltage regulation and power distribution.
control effect of the forecast-based economic dispatch scheme Beyond the primary control layer, a distributed cooperative
is poor. Therefore, it is required to develop a distributed control control structure is designed to synchronously realize average
strategy that can realize voltage regulation and economic bus voltage restoration and optimal power sharing. In the cyber
dispatch in real time to smooth frequent power fluctuations and layer, each DG exchanges information with neighboring DGs
minimize the total power generation cost. through a neighbor-to-neighbor communication network. The
following would focus on the design process and
III. DISTRIBUTED COOPERATIVE CONTROL STRATEGY implementation principle of the proposed strategy.
In this section, the control objectives and structure of the B. Proposed Strategy without Capacity Constraints
proposed strategy are first introduced. Then, the design process From Fig. 2, the voltage reference can be determined by:
and realization principle of proposed strategy without capacity uref,i  unom  Rvi ioi  ui (13)
constraint and with capacity constraint are given, respectively.
where Rvi is the virtual resistance of DGi, i.e., droop coefficient.
A. Control Objectives and Structure ioi is the DGi output current. ui is the voltage correction term
In this paper, a novel distributed cooperative control scheme determined by distributed cooperative control.
is developed to synchronously realize the control objectives of Inspired by [9] and [10], the ui can be designed as:
average bus voltage restoration and optimal power sharing:  avg,i 
ui    unom   (14)
lim u nom  u avg,i (t )  0
t   i 
 (12) where κ is the integrator gain. ξavg,i and γi are the average virtual
lim i (t )   j (t )  0
t  state variables and transition factors of DGi, respectively.
where unom is the nominal bus voltage. uavg,i(t) is the average bus Combined with the average consensus algorithm of (3), they

© 2023 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.See https://www.ieee.org/publications/rights/index.html for more information.
Authorized licensed use limited to: Murdoch University. Downloaded on May 14,2023 at 15:52:18 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in IEEE Transactions on Power Systems. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2023.3274932

IEEE TRANSACTIONS ON POWER SYSTEMS 5

can be designed as:


 i t
avg,i  i  N 0  aij avg, j  avg,i  dt
 i jNi
i   i uoi (15)
 i
 i  1   vm Fig. 3. Simplified model for stability analysis.

where ξi is the virtual state variable of DGi. αi is a consensus
gain of DGi. |Ni| is the number of neighbors of DGi. uoi is the where τe is a time constant.
output voltage of DGi. σ is a gain in the range [0, 1] to prevent The output current is connected with a low-pass filter, then:
γi from being zero. λvm is a sufficient large virtual IC, which is c
predetermined before the system runs, which can be expressed Glpf  (20)
c  s
as:
where ωc is a cut-off frequency.
vm  2a vm Pvm  bvm (16)
Furthermore, the output current of DG1 versus the output
where avm and bvm are sufficient large cost function coefficients, voltage can be obtained from Fig. 3:
and Pvm is a sufficient large output power of DGi.
io1   uo1   uo2 (21)
Remark 2: Different from the hierarchical control methods
in [16]-[23], the proposed control strategy only requires a where
distributed cooperative controller to synchronously realize  r2  RL
voltage regulation and economic dispatch, which can perform   r1r2  r1 RL  r2 RL
economic dispatch online and reduce the system order, so as to  (22)
  RL
avoid the “dimension disaster” caused by the parallel DGs.
Therefore, system stability is guaranteed, which would be  r1r2  r1 RL  r2 RL
demonstrated below. where RL is the load resistance.
Remark 3: It is worth pointing out that the proposed According to Fig. 3 and (17)–(22), the output voltage of DG1
distributed cooperative control strategy is also applicable to AC can be deduced:
and AC/DC microgrids, with only some modifications to the   avg,1 
proposed strategy according to the specific control task. In AC uo1  unom  Rv1io1Glpf  Gs  unom  Ge Gd  Gc (23)
microgrids, the use of this distributed cooperative control   1 
strategy can ensure frequency and voltage restoration whereas where Gd is the transfer function of the communication delay,
obtaining active power sharing of the parallel DGs. which can be expressed as:
Theorem 1: For the DC microgrid with multi-parallel DGs, 1  d s
the system stability is guaranteed after applying the proposed Gd  e d s  (24)
distributed cooperative control strategy. 1 d s
Proof: Without loss of generality, taking two parallel DGs as where τd is the communication delay time.
an example, an equivalent model for system stability analysis is Therefore, (23) can be rewritten as:
established, as shown in Fig. 3. Due to the fast switching action   2 
of the converter, the internal controller usually has a low-pass unom 1  Gs   uo2   Rv1Glpf  GeGs Gd  Gc
 2 1 
characteristic with a high bandwidth, which may not affect the uo1   (25)
performance of the droop control loop as long as it is designed  1 
1   Rv1Glpf  Ge GsGd  Gc
properly because their bandwidths are decoupled. As a result,  2 
the dynamic response speed of the internal controller is faster Linearizing the variables of (25), the small-signal model
than the distributed cooperative controller, the transfer function expression can be obtained as:
of the voltage-current controller can be equivalent as [10]:
uˆo1 1  Gs  Gc
1  (26)
Gc  1 (17) uˆnom  1 
1 s uˆo2  0
1    Rv1Glpf  GeGs Gd  Gc
where τ is a time constant. s is a Laplace operator.  2 
According to Fig. 2 and (14), the transfer function of the From (26), the characteristic equation of the closed-loop
integrator can be obtained as: system can be obtained:
 As4  Bs3  Cs2  Ds  E  0 (27)
Gs  (18) where
s
where κ is the integrator gain. A  2 e d
To simplify the analysis, the consensus iteration process for B  2 Rv1 c e d  2 c e d  2  e   d 
ξavg,1 in (15) can be equivalent to the conventional calculation
process of ξavg,1 multiplied by the inertial link, so we have: C  2 Rv1 c  e   d   2 c  e   d   2   d (28)
1 D  2 Rv1 c  2 c     c d
Ge  (19)
1 e s E   c

© 2023 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.See https://www.ieee.org/publications/rights/index.html for more information.
Authorized licensed use limited to: Murdoch University. Downloaded on May 14,2023 at 15:52:18 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in IEEE Transactions on Power Systems. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2023.3274932

IEEE TRANSACTIONS ON POWER SYSTEMS 6

i i
ξavg  ξ  Aξavg  Dξavg
Ni Ni
(30)
i 
 ξ   D  A ξavg  ξ  i Lξavg
Ni Ni
where ξ=[ξ1, ξ2, …, ξN]T and ξavg=[ξavg,1, ξavg,2, …, ξavg,N]T are
vectors of virtual state variables and average virtual state
variables, respectively.
Performing Laplace transform on (30):
i
sΞavg  Ξavg (0)  sΞ  Ξ (0)  LΞavg (31)
Ni
where Ξavg and Ξ are the Laplace transforms of ξavg and ξ,
respectively.
By (28), Ξ(0)=Ξavg(0), so (31) can be simplified as:
Fig. 4. Characteristic root loci with parameters changing. (a) Rv1 increases 1
from 0.01 to 10. (b) κ increases from 1 to 50. (c) τe increases from 10 to  i 
Ξavg  s  sI N  L  Ξ  H obs Ξ (32)
1000ms. (d) τd increases from 10 to 500ms.
 Ni 
Without loss of generality, setting RL=80Ω, r1=1Ω, r2=0.5Ω, where IN and Hobs are the N-order identity matrix and the
ωc=50rad/s, the closed-loop pole root loci as Rv1, κ, τe and τd observer transfer function matrix, respectively.
varies can be plotted separately, as shown in Fig. 4. From Figs. Finally, applying the final value theorem to (32) [28]:
1
4(a)(c), the dominant poles ρ1 and ρ2 lies in the left half-plane  i 
of the imaginary axis and away from it when the relevant lim sξavg (t )  lim sΞavg  lim s  sI N  L   lim sΞ
t  s 0 s 0  Ni  s 0 (33)
parameters are increased, which shows that the system has a 
satisfactory stability margin. On the other hand, the conjugate  QN Ξ  Ξ 1N
ss ss
poles ρ3 and ρ4 vary less and neither of them crosses the ss
imaginary axis to the right half-plane, so the system stability is where Ξ is the steady state vector of the virtual state variable.
almost unaffected. From Fig. 4(b), the dominant poles ρ1 and ρ2 1N is a vector whose elements are all one. QN is an average
gradually move away from the imaginary axis as κ increases, matrix whose elements are all 1/N, and:
1 N
which indicate that the system enters the steady state more Ξ ss   i 1 iss (34)
quickly. However, the dominant poles ρ3 and ρ4 gradually move N
away from the real axis, which may lead to voltage and current From (33), the average virtual state variables of DGs reach a
oscillations. To make matters worse, when κ>23, ρ3 and ρ4 global consensus in steady state. This completes the proof.
crosses the imaginary axis to the right half-plane, which would Remark 4: The convergence rate of the observer depends on
lead to system instability. Therefore, the value of κ should not the consensus gain αi. Too small αi would seriously slow down
exceed 23 and requires the selection of a suitable value to the convergence speed of the observer, and too large αi may
ensure a compromise between the dynamic and steady state bring about the convergence oscillation problem. Therefore, the
performance of the system. From Fig. 4(d), the dominant poles choice of αi should consider the trade-off between steady-state
ρ1 and ρ2 gradually approach the imaginary axis as τd increases, and dynamic performance.
which would reduce the system stability margin. Unfortunately, Theorem 3: Through the regulation of (14), the control
when τd>260ms, the conjugate poles ρ3 and ρ4 crosses the objectives of average bus voltage restoration and optimal power
imaginary axis to the right half-plane, which lead to system sharing can be achieved synchronously in steady state.
instability. Thus, the proposed strategy has a communication Proof: Transfer (13) from the time domain to the frequency
delay margin of 260ms, which satisfies the requirements of domain:
engineering applications. This completes the proof. Uref  Unom  U  Rv Io (35)
Theorem 2: Applying the average virtual state variable
where Uref is the Laplace transform of uref=[uref,1, uref,2, …,
observer in (15) allows for the global consensus of the average
uref,N]T. Unom=(unom/s)1N is the Laplace transform of the nominal
virtual state variable in the steady state.
voltage vector. U is the Laplace transform of u=[u1, u2, …, uN]T.
Proof: From Theorem 1, the system stability is guaranteed,
Io is the Laplace transform of io=[io1, io2, …, ioN]T. Rv=diag{Rvi}
so the proposed distributed cooperative control strategy can be
is the droop coefficient matrix.
analyzed for steady state in the following. Differentiate the first
Combining (14) and (32), U can be expressed as:
term of (15):
i  Ξavg   H obs Ξ 
U  G  U nom    G  U nom 
avg,i  i 
(36)
Ni
 
jNi
avg, j  avg,i   Η   Η 

(29) where Gκ=diag{Gκ} is the transfer function matrix of the


i   integral regulator, Gκ=κ/s. H is the Laplace transform of γ=[γ1,
 i    avg, j  diavg,i  γ2, …, γN]T.
Ni  jNi  Furthermore, transfer the second term of (15) to the
Rewrite (29) in compact form: frequency domain:

© 2023 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.See https://www.ieee.org/publications/rights/index.html for more information.
Authorized licensed use limited to: Murdoch University. Downloaded on May 14,2023 at 15:52:18 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in IEEE Transactions on Power Systems. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2023.3274932

IEEE TRANSACTIONS ON POWER SYSTEMS 7

Ξ  HUo (37) 1  2   N (53)


Substituting (36) and (37) into (35) to obtain: From (53), according to the equal IC criterion [13], the
U ref  U nom  G U nom  H obsU o   Rv I o (38) proposed strategy can achieve optimal power sharing of DGs.
Therefore, through the regulation of (14), the average bus
From Fig. 2, the global dynamics of the output voltage can
voltage restoration and optimal power sharing can be realized
be expressed as:
synchronously in steady state. This completes the proof.
Uo  GcUref (39) Remark 5: Intuitively, although the proposed strategy
where Gc=diag{Gci} is a transfer function matrix. requires only one controller to achieve the objectives of (12),
From (1), the output voltage is related to the output current the average bus voltage restoration and optimal power sharing
through the system admittance matrix: exhibit different time constants. Therefore, the dynamic
Io  YUo  IL (40) decoupling of the voltage regulation and the economic dispatch
Combined with (38)-(40), the global system dynamic model can be obtained through the bandwidth difference design.
can be obtained: C. Proposed Strategy with Capacity Constraints
Uo   Gc1  G H obs  RvY    I N  G  Unom  Rv I L  (41)
1
Considering the capacity constraints of DGs, the transition
Then, the steady-state output voltage vector can be derived factor γi of (15) is redesigned in this subsection. Specifically,
through applying the final value theorem to (41): the concept of “IC margin” is proposed to constrain ICs to the
uoss  lim suo (t )  lim sU o (s) limit warning threshold when the output power of DGs is
t  s 0 violated, thereby indirectly preventing the converter from
 lim  sG  sG H obs  sRvY 
1
1
c (42) working at the power generation limit for a long time. Therefore,
s 0
γi is redesigned as:
  s 2  I N  G  U nom  sRv I L   i
Since the voltage and current control of the main control 1      i  i   H,i  , for i   H,i
 vm
layer can respond quickly, its closed-loop transfer function can  
be approximated as [28]:  i  1   i , for  L,i  i   H,i (54)
  vm
Gc (0)  I N  Gc1 (0) (43)
 i
Moreover, the following constraints can be obtained from the 1     i  i   L,i  , for  i   L,i
definition of the observer transfer function:   vm
lim H obs  QN (44) where λH,i and λL,i are the upper and lower warning thresholds
s0
of IC, respectively, and εi is a limiting gain. λH,i and λL,i can be
Substituting (43) and (44) into (42) to get: expressed as:
uoss  QN1unom 1N (45)
H,i  max,i  marg,i
or, equivalently:  (55)
QN uoss  unom 1N (46) L,i   marg,i   min,i
where λmax,i=2aiPmax,i+bi and λmin,i=2aiPmin,i+bi are the
According to the definition of QN [28]:
maximum and minimum IC of DGi. λmarg,i=2aiPmarg,i+bi is the IC
QN x  x 1N (47) margin of DGi, and Pmarg,i is the power margin of DGi.
Combining (46) and (47) leads to: From (54), equation (15) is still valid if and only if the ICs
1 does not trigger the limit warning threshold. Once the ICs
uoss   i 1 uossi  unom
N
(48) triggers the upper or lower warning threshold, the ICs would be
N
constrained to the warning threshold instead of reaching the
From (48), the proposed strategy is able to restore the average maximum value. Therefore, the proposed strategy can satisfy
bus voltage to the nominal value in steady state. the capacity constraints while avoiding a series of adverse
On this basis, apply the final value theorem to (36): consequences caused by power violations of DGs.
uss  lim su(t )  lim sU ( s) The selection of the limiting gain εi becomes critical when
t  s 0
(49) the ICs trigger the limit warning threshold. The following
 lim  sG U nom  sG H obsUo 
s 0 would take the ICs trigger upper warning threshold of DGs as
or, equivalently: an example to deduce the selection range of εi. Since ICs cannot
uss  unom 1N  QN uoss   unom  uoss  1N  0 (50) reach its maximum value, it yields that:
1   H,i   i
Therefore, the following relationship holds in steady state by i    max,i (56)
combining (13) and (38): 
i
avg,1 avg,2 avg, N  vm
unom    (51) From (15), γ among DGs can quickly reach an average
1 2 N
consensus. Therefore, the selection range of εi can be deduced:
Substituting (15) into (51), we can infer that:
1  max,i
1 2 N 0  i   (57)
  (52)  marg,i  vm  m arg,i
 vm  vm  vm
Remark 6: Compared with existing economic dispatch
or, equivalently:
strategies considering capacity constraints, the proposed

© 2023 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.See https://www.ieee.org/publications/rights/index.html for more information.
Authorized licensed use limited to: Murdoch University. Downloaded on May 14,2023 at 15:52:18 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in IEEE Transactions on Power Systems. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2023.3274932

IEEE TRANSACTIONS ON POWER SYSTEMS 8

TABLE II
SYSTEM PARAMETERS
Islanded DC microgrid
Parameters DG1 DG2 DG3 DG4
uin,i(V) 450 450 450 450
unom(V) 700
rij(Ω) 0.19 0.25 0.28 0.23
ai(¢/W2) 0.178 0.132 0.196 0.145
bi(¢/W) 4.9 6.5 5.3 3.4
ci(¢) 110 120 85 95
amax(¢/W2) 0.196
bmax(¢/W) 6.5
cmax(¢) 120
Pmax,i(W) 4500 4500 4500 4500
Pmin,i(W) 400 400 400 400
Primary control
Parameters DG1 DG2 DG3 DG4
Rvi 1 1 1 1
KUP, KUI 1.5, 20
KIP, KII 0.05, 1
fk 20kHz
Distributed cooperative control
Parameters DG1 DG2 DG3 DG4 Fig. 5. Simulation results of Case I. (a) Output powers. (b) Incremental costs.
κ 10 (c) Total cost. (d) Output voltages.
αi 200 200 200 200
σ 0.5
εi 0.01 0.01 0.01 0.01
Pvm(W) 10000
Pmarg,i(W) 100 100 100 100

strategy can limit the IC of DGs violating capacity constraints


to the limit warning threshold rather than the limit value, which
can effectively improve the performance of DGs converter. To
be mentioned, the DGs no longer participate in the consensus
protocol of (15) when ICs trigger the limit warning threshold,
but constrain their ICs to the limit warning threshold to avoid
power violation. Moreover, the remaining DGs would continue
to complete the tasks of voltage regulation and economic
dispatch based on the consensus protocol. In this case, the
output power of each DG is no longer optimal, but there is a
slight deviation from the optimal operating point.

IV. SIMULATION RESULTS


In this section, an islanded DC microgrid consisting of four Fig. 6. Simulation results of Case II. (a) Output powers. (b) Incremental costs.
DGs and multiple loads is developed in the MATLAB/Simulink (c) Total cost. (d) Output voltages.
environment, as shown in Fig. 1. Six different cases are
considered, such as normal operation, capacity constraints, deviation. In addition, when the loads power changes, the bus
plug-and-play, communication delays, methods comparison, voltage fluctuation amplitude is only 1.3V and 1.9V, and it
time-varying loads simulation etc., to verify the effectiveness recovers smoothly in a short time. Therefore, the proposed
and superiority of the proposed strategy. Moreover, the strategy can achieve average bus voltage restoration and
simulation parameters are given in Table II, and the simulation optimal output power sharing, which keep the DGs at the
step is 1 microsecond. optimal operating point in real time.
A. Simulation Case I: Normal System Operation B. Simulation Case II: Capacity Constraints
In this case, the proposed strategy is tested under normal This case mainly explores the performance of the proposed
system operation. The initial total load power is 5kW, which strategy when a DG reaches its own capacity constraint. The
rises to 7.5kW at t=5s and drops to 4kW at t=10s. initial total load demand power is 10kW, which rises to 14kW
From Figs. 5(a)(b), the proposed strategy can promote the at t=5s and drops to 8kW at t=10s.
convergence of ICs of DGs and ensure that their output power From Figs. 6(a)(b)(c), the proposed strategy achieves the
track to the optimal value. When loads demand power changes optimal output power sharing before t=5s. At t=5s, since the IC
suddenly, the output power of DGs re-tracks to the optimal of DG2 triggers the upper warning threshold λH,2=1168.1¢/W, it
value, and their ICs maintain a consistent state. From Figs. no longer participates in the consensus protocol in (15). In other
5(c)(d), the proposed strategy guarantees to minimize the total words, DG2 becomes an intermediary for information exchange
power generation cost and eliminate the average bus voltage between other DGs. Therefore, the IC and output power of DG2

© 2023 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.See https://www.ieee.org/publications/rights/index.html for more information.
Authorized licensed use limited to: Murdoch University. Downloaded on May 14,2023 at 15:52:18 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in IEEE Transactions on Power Systems. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2023.3274932

IEEE TRANSACTIONS ON POWER SYSTEMS 9

Fig. 8. Simulation results of Case IV. (a) Output powers of method in Ref.
[21]. (b) Output voltages of method in Ref. [21]. (c) Output powers of
proposed method. (d) Output voltages of proposed method.

can simultaneously achieve average bus voltage restoration and


Fig. 7. Simulation results of Case III. (a) Output powers. (b) Incremental costs. total power generation cost minimization.
(c) Total cost. (d) Output voltages. (e) Total cost.
D. Simulation Case IV: Comparison with state-of-the-art
methods
are constrained to be around λ2=1167¢/W and P2=4400W,
respectively, and the output power that DG2 cannot afford is In this case, the proposed strategy is compared with the state-
shared by other DGs. To be mentioned, the output power of of-the-art method in [21]. It is assumed that the local load power
DG2 is no longer optimal, but the ICs of the remaining DGs of each DG in the initial stage is PL1=1kW, PL2=1kW, PL3=1kW,
continue to maintain a matching state, so the total power and PL4=2kW. In addition, PL1 decreases from 1kW to 0 at t=5s
generation cost is still close to the optimal value. At t=10s, the and PL3 increases from 1kW to 2kW at t=10 s.
capacity constraint state of DG2 is lifted, so each DG re-tracks From Figs. 8(a)(b), the method of [21] is able to achieve
to the best working point. In addition, the average bus voltage optimal output power sharing and DC bus voltage restoration.
fluctuation amplitude is only 2.7V and 3.7V when the loads However, the output power ripple of DGs is large and there is
changes, which is far below the limit (i.e. ±35V), and the an overshoot of 22.5V and a drop of 13.8V in the bus voltage
average bus voltage can quickly recover to the nominal value. during startup process, which can easily lead to system collapse
due to relay protection malfunction. In addition, the local load
C. Simulation Case III: Comparison of different types of connected to DG1 exits leading to high frequency oscillation of
methods the DC bus voltage, which is not conducive to stable system
In this case, the proposed strategy is compared with droop operation. From Figs. 8(c)(d), the application of the proposed
control and the proportional power sharing method in [9]. The strategy is also able to achieve optimal output power sharing
total load demand power is 10kW. Droop control, proportional and DC bus voltage restoration. More importantly, the output
power sharing method, and the proposed strategy are applied at power ripple of DGs is small, the bus voltage quickly recovers
0s, 5s and 10s, respectively. smoothly after a 2.1V overshoot during startup process, and it
From Figs. 7(a)(b)(c), droop control cannot achieve optimal shows better robustness in all other cases.
output power sharing and proportional power sharing. Although From the above observations, the introduction of multiple PI-
the method in [9] can realize output power sharing of DGs, the based secondary controllers in the method of [21] tends to lead
total power generation cost is higher, which is not an to the deterioration of the system dynamic and steady-state
economical solution. On the contrary, the proposed strategy performance, and the coupling effect between PI controllers
enables DGs to track the optimal operating point and minimize leads to a more difficult selection of PI parameters, which may
the total power generation cost. Furthermore, the droop control cause system instability. Fortunately, the proposed strategy is
causes the average bus voltage to drop by about 3.6V, while the able to combine the economic dispatch and voltage regulation
proportional power sharing method and the proposed strategy functions in a distributed cooperative controller based on
are able to stable the average bus voltage at the nominal value. integral regulation, which can effectively reduce the system
For the sake of intuition, Fig. 7(e) presents the histogram of the order and thus improve the controller dynamic response speed
total power generation cost of above three methods, it can be and ensure excellent steady-state performance.
observed that the total power generation cost of the proposed
strategy is significantly smaller than the other two methods, and
is equal to the optimal value. Therefore, the proposed strategy

© 2023 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.See https://www.ieee.org/publications/rights/index.html for more information.
Authorized licensed use limited to: Murdoch University. Downloaded on May 14,2023 at 15:52:18 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in IEEE Transactions on Power Systems. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2023.3274932

IEEE TRANSACTIONS ON POWER SYSTEMS 10

Fig. 9. Simulation results of Case V. (a) Output powers. (b) Incremental costs.
(c) Total cost. (d) Output voltages.

Fig. 11. Simulation results of Case VII. (a) Total load power. (b) Output
powers. (c) Incremental costs. (d) Total cost. (e) Output voltages.

at t=2s and restored at t=4s. DG4 is disconnected from the


system at t=5s, and reconnected at t=10s.
From Fig. 9, since the communication network still has a
minimum directed spanning tree when the communication link
between DG1 and DG2 is disconnected, the control effect is not
affected. Similarly, communication restoration has little effect
on the system performance. At t=5s, DG4 is out of operation due
to failure, and it no longer participates in the consensus protocol
in (15), which is equivalent to DG4 being removed from the
communication network. In this case, the remaining DGs are
required to bear the total load demand power, so they cannot
continue to operate at the optimal working point. However, the
ICs of the remaining DGs can still be matched in real time,
which keeps the total power generation cost as small as possible.
At t=10s, DG4 is re-inserted, which causes the average bus
voltage to fluctuate with an amplitude of about 7.6V.
Fortunately, the optimal output power sharing and average bus
voltage restoration can be re-implemented in a short time,
which validates the excellent plug-and-play performance of the
proposed strategy.
F. Simulation Case VI: Communication Delays
In this case, the proposed strategy is tested with different
Fig. 10. Simulation results of Case VI. (a1) Output powers with τd=5ms. (b1)
Incremental costs with τd=5ms. (a2) Output powers with τd=10ms. (b2) communication delays. The droop control and the proposed
Incremental costs with τd=10ms. (a3) Output powers with τd=15ms. (b3) strategy are activated at 0s and 5s, respectively. The initial total
Incremental costs with τd=15ms. load demand power is 5kW, which rises to 7.5kW at t=10s.
From Figs. 10(a1)(b1), the 5ms communication delay has no
significant effect on the control performance of the proposed
E. Simulation Case V: Communication Abnormal and Plug strategy. If the communication delay increases to 10ms, the
and Play output power and ICs of DGs oscillate during the control
This case discusses the control performance of the proposed method and the loads change process, but the steady state can
strategy with communication interruption, communication still be restored quickly, as shown in Figs. 10(a2)(b2). However,
restoration, and plug-and-play. The total load demand power is when the communication delay is increased to 15ms, the output
5kW. The communication link between DG1 and DG2 is broken power and ICs of DGs oscillate violently, and it is difficult to

© 2023 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.See https://www.ieee.org/publications/rights/index.html for more information.
Authorized licensed use limited to: Murdoch University. Downloaded on May 14,2023 at 15:52:18 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in IEEE Transactions on Power Systems. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2023.3274932

IEEE TRANSACTIONS ON POWER SYSTEMS 11

Fig. 12. Experimental platform.

Fig. 14. Experimental results of Case II. (a) Incremental costs. (b) Output
currents. (d) Output voltages.

V. EXPERIMENTAL RESULTS
In this section, an islanded DC microgrid experimental model
based on StarSim HIL is built, as shown in Fig. 12. Herein, the
islanded DC microgrid consisting of four DGs and several loads
is simulated by NI-PXIe-FPGA-7846R. Besides, the proposed
distributed cooperative strategy is implemented by DSP
controllers of TMS320F28379D with a sampling period of 100
microseconds. Moreover, the experimental system parameters
are given in Table II.
Fig. 13. Experimental results of Case I. (a) Incremental costs. (b) Output
currents. (d) Output voltages. A. Experimental Case I: Capacity Constraints
The experimental results with the DG2 capacity constraint are
fully restore to steady state, but the established control given in Fig. 13. The initial total load demand power is 10kW,
objectives can still be achieved, as shown in Figs. 10(a3)(b3). which rises to 15kW at t=3s and drops to 10kW at t=7s.
Although the proposed strategy is proved to be stable in the From Figs. 13(a)(b)(c), the proposed strategy is able to
range of 10-260ms communication delay in Theorem 1, the realize stable bus voltage restoration and optimal output power
constraints of voltage fluctuation range, current ripple, dynamic sharing until t=3s. When the load demand power suddenly rises,
response time and other technical indicators cannot be ignored, the IC and output current of DG2 are limited to λ2=1163¢/W and
so safety and reliability should be considered on the basis of io2=6.25A due to trigger upper warning threshold, whereas the
system stability. On the other hand, stability analysis using remaining DGs still complete the voltage regulation and
simplified models has certain limitations, and the actual economic dispatch tasks based on the consensus protocol.
controller may have unknown parameters, unmodeled When the load demand power suddenly drops, the capacity
dynamics, external disturbances and other uncertainties, which constraint state is lifted, so each DG re-tracks to the optimal
can lead to deviations between theoretical analysis and operating point. The DC bus voltage produces a drift during the
simulation results. Therefore, only 15ms or less communication sudden change of load power, but can recover to the nominal
delay can be tolerated in practical applications. value quickly. The above experimental results are similar to the
results of simulation Case II, which verifies the effectiveness of
G. Simulation Case VII: Simulate Time-Varying Loads
the proposed strategy with a certain DG capacity constraint.
In this case, the effectiveness of the proposed strategy under
time-varying load conditions is tested. Due to the limitation of B. Experimental Case II: Methods Comparison
computer data storage, the total load demand power data within Fig. 14 presents the experimental results under the
24h in an islanded DC microgrid is proportionally reduced to comparison between different methods. The total load demand
24s, as shown in Fig. 11(a). power is 10kW. The droop control, proportional power sharing
From Figs. 11(b)(c)(d), the output power and total power method, and the proposed strategy are applied at 0s, 3s and 7s,
generation cost of DGs can still dynamically track to the respectively.
optimal value under time-varying load conditions. Moreover, From Figs. 14(a)(b)(c), both droop control and proportional
the proposed strategy can still keep the average bus voltage power sharing methods cannot achieve bus voltage restoration
around 700V, as shown in Fig. 11(e). Therefore, the proposed and optimal power sharing at the same time, whereas the
strategy can realize average bus voltage restoration and optimal proposed strategy can solve this problem well. Therefore, DGs
output power sharing under time-varying load conditions, and are able to track the optimal operating point in real time to
exhibits strong robustness and practicability. eliminate average bus voltage deviation and minimize total
power generation cost, which is similar to the results of

© 2023 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.See https://www.ieee.org/publications/rights/index.html for more information.
Authorized licensed use limited to: Murdoch University. Downloaded on May 14,2023 at 15:52:18 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in IEEE Transactions on Power Systems. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2023.3274932

IEEE TRANSACTIONS ON POWER SYSTEMS 12

simulation Case III. [12] C. E. Lin and G. L. Viviani, “Hierarchical economic dispatch for
piecewise quadratic cost functions,” IEEE Trans. Power App. Syst., vol.
PAS-103, pp. 1170–1175, Jun. 1984.
VI. CONCLUSION [13] A. J. Wood and B. F. Wollenberg, Power Generation, Operation, and
Control, 2nd ed. New York: Knovel, 1996.
A novel distributed cooperative control strategy is developed
[14] P. Chen, and H. Chang, “Large-scale economic dispatch by genetic
for stable bus voltage restoration and optimal power sharing in algorithm,” IEEE Trans. Power Syst., vol. 10, no. 4, pp. 1919–1926, Nov.
islanded DC microgrids. Different from the centralized control 1995.
or hierarchical control, the proposed strategy can unify the [15] P. Jong-Bae, J. Yun-Won, S. Joong-Rin, and K. Y. Lee, “An improved
objectives of voltage regulation and economic dispatch in particle swarm optimization for nonconvex economic dispatch problems,”
IEEE Trans. Power Syst., vol. 25, no. 1, pp. 156–166, Feb. 2010.
single controller with a fully distributed manner, thereby [16] Y. Xu, W. Zhang, and W. Liu, “Distributed dynamic programming-based
eliminating the average bus voltage deviation and minimizing approach for economic dispatch in smart grids,” IEEE Trans Ind.
the total power generation cost in real time. Furthermore, the Informat., vol. 11, no. 1, pp. 166–175, Feb. 2015.
concept of “IC margin” is introduced to constrain the ICs to the [17] G. Wen, X. Yu, Z. Liu, and W. Yu, “Adaptive consensus-based robust
strategy for economic dispatch of smart grids subject to communication
limit warning threshold when the output power of DGs is uncertainties,” IEEE Trans Ind. Informat., vol. 14, no. 6, pp. 2484–2496,
violated, thus avoiding the deterioration of the converter Jun. 2018.
performance. Finally, the simulation and experimental results [18] Z. Li, Z. Cheng, J. Si, and S. Li, “Distributed event-triggered hierarchical
control to improve economic operation of hybrid AC/DC microgrids,”
verify the effectiveness of the proposed strategy in an islanded
IEEE Trans. Power Syst., vol. 37, no. 5, pp. 3653–3668, Sep. 2022.
DC microgrid under various testing scenarios. [19] Z. Wang, W. Wu, and B. Zhang, “A distributed control method with
It is worth pointing out that communication delays have minimum generation cost for DC microgrids,” IEEE Trans. Energy
become an important factor restricting the development of Convers., vol. 31, no. 4, pp. 1462–1470, Dec. 2016.
distributed control strategies, and strengthening the influence of [20] J. Hu, J. Duan, H. Ma, and M. Y. Chow, “Distributed adaptive droop
control for optimal power dispatch in DC microgrid,” IEEE Trans. Ind
time-varying communication delays on the proposed strategy Electron., vol. 65, no. 1, pp. 778–789, Jan. 2018.
would be the focus of future research work. [21] M. Zaery, P. Wang, W. Wang, and, D. Xu, “Distributed global
economical load sharing for a cluster of DC microgrids,” IEEE Trans.
Power Syst., vol. 35, no. 5, pp. 3410–3420, Sep. 2020.
REFERENCES [22] J. Peng, B. Fan, and W. Liu, “Voltage-based distributed optimal control
[1] J. M. Guerrero, J. C. Vasquez, J. Matas, L. G. Vicuña, and M. Castilla, for generation cost minimization and bounded bus voltage regulation in
“Hierarchical control of droop-controlled AC and DC microgrids—a DC microgrids,” IEEE Trans. Smart Grid., vol. 12, no. 1, pp. 106–116,
general approach toward standardization,” IEEE Trans. Ind Electron., Jan. 2021.
vol. 58, no. 1, pp. 158–172, Jan. 2011. [23] Y. Dou, M. Chi, Z. W. Liu, G. Wen, and Q. Sun, “Distributed secondary
[2] T. Dragičevič, X. Lu, J. C. Vasquez, and J. M. Guerrero, “DC control for voltage regulation and optimal power sharing in DC
microgrids—part I: a review of control strategies and stabilization microgrids,” IEEE Trans. Control Syst. Technol., vol. 30, no. 6, pp.
techniques,” IEEE Trans. Power Electron., vol. 31, no. 7, pp. 4876–4891, 2561–2572, Nov. 2022.
Jul. 2016. [24] Y. Zeng, Q. Zhang, Y. Liu, X. Zhuang, X. Lv, and H. Wang, “An
[3] X. Lu, J. M. Guerrero, K. Sun, and J. C. Vasquez, “An improved droop improved distributed secondary control strategy for battery storage
control method for DC microgrids based on low bandwidth system in DC shipboard microgrid,” IEEE Trans. Ind Appl., vol. 58, no.
communication with DC bus voltage restoration and enhanced current 3, pp. 4062–4075, May. 2022.
sharing accuracy,” IEEE Trans. Power Electron., vol. 29, no. 4, pp. [25] R. Olfati-Saber, and R. M. Murray, “Consensus problems in networks of
1800–1812, Apr. 2014. agents with switching topology and time-delays,” IEEE Trans. Autom.
[4] Q. Shafiee, V. Nasirian, J. C. Vasquez, J. M. Guerrero and A. Davoudi, Control, vol. 49, no. 9, pp. 1520–1533, Sep. 2004.
“A multi-functional fully distributed control framework for AC [26] Z. Fan, B. Fan, and W. Liu, “Distributed control of DC microgrids for
microgrids,” IEEE Trans. Smart Grid., vol. 9, no. 4, pp. 3247–3258, Jul. optimal coordination of conventional and renewable generators,” IEEE
2018. Trans. Smart Grid., vol. 12, no. 6, pp. 4607–4615, Nov. 2021.
[5] M. Chen, X. Xiao, and J. M. Guerrero, “Secondary restoration control of [27] S. Yang, S. Tan, and J. X. Xu, “Consensus based approach for economic
islanded microgrids with a decentralized event-triggered strategy,” IEEE dispatch problem in a smart grid,” IEEE Trans. Power Syst., vol. 28, no.
Trans Ind. Informat., vol. 14, no. 9, pp. 3870–3880, Sep. 2018. 4, pp. 4416–4426, Nov. 2013.
[6] S. Deng, L. Chen, X. Lu, T. Zheng, and S. Mei, “Distributed finite-time [28] V. Nasirian, S. Moayedi, A. Davoudi, and F. L. Lewis, “Distributed
secondary frequency control of islanded microgrids with enhanced cooperative control of DC microgrids,” IEEE Trans. Power Electron.,
operational flexibility,” IEEE Trans. Energy Convers., vol. 36, no. 3, pp. vol. 30, no. 4, pp. 2288–2303, Apr. 2015.
1733–1742, Sep. 2021.
[7] P. Wang, X. Lu, X. Yang, W. Wang and, D. Xu, “An improved
distributed secondary control method for DC microgrids with enhanced
dynamic current sharing performance,” IEEE Trans. Power Electron.,
vol. 31, no. 9, pp. 6658–6673, Sep. 2016.
[8] L. Xing, Y. Mishra, F. Guo, P. Lin, Y. Yang, G. Ledwich, and Y. Tian,
“Distributed secondary control for current sharing and voltage restoration
in DC microgrid,” IEEE Trans. Smart Grid., vol. 11, no. 3, pp. 2487–
2497, May. 2020.
[9] W. W. A. G. Silva, T. R. Oliveira, and P. F. Donoso-Garcia, “An
improved voltage-shifting strategy to attain concomitant accurate power
sharing and voltage restoration in droop-controlled DC microgrids,”
IEEE Trans. Power Electron., vol. 36, no. 2, pp. 2396–2406, Feb. 2021.
[10] Q. Zhang, Y. Zeng, Y. Liu, X. Zhuang, H. Zhang, W. Hu and H. Guo,
“An improved distributed cooperative control strategy for multiple
energy storages parallel in islanded DC microgrid,” IEEE J. Sel. Topics
Power Electron, vol. 10, no. 1, pp. 455–468, Feb. 2022.
[11] A. A. Hamad, M. A. Azzouz, and E. F. El-Saadany, “Multiagent
supervisory control for power management in DC microgrids,” IEEE
Trans. Smart Grid., vol. 7, no. 2, pp. 1057–1068, Mar. 2016.

© 2023 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.See https://www.ieee.org/publications/rights/index.html for more information.
Authorized licensed use limited to: Murdoch University. Downloaded on May 14,2023 at 15:52:18 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in IEEE Transactions on Power Systems. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2023.3274932

IEEE TRANSACTIONS ON POWER SYSTEMS 13

Yuji Zeng (Graduate Student Member, IEEE) was born


in Guangdong, China, in 1996. He received the B.S.
degree in marine engineering from Guangdong Ocean
University, Zhanjiang, China, in 2018.
He is currently working toward the Ph.D. degree in
marine engineering from Dalian Maritime University,
Dalian, China. His current research interests include dc
microgrids, energy storage system, fuel cell system,
energy management system, and shipboard hybrid
power system.

Qinjin Zhang (Member, IEEE) was born in Jiangsu,


China, in 1986. He received the B.S. degree in
electrical engineering and automation and the M.S.
and Ph.D. degrees in marine engineering from Dalian
Maritime University, Dalian, China, in 2009, 2011,
and 2015, respectively.
Since 2020, he has been an Associate Professor with
the Marine Engineering College, Dalian Maritime
University. His research interests include distributed
energy generation technology, power electronics
converters, microgrids, and ship dc integrated power
system.

Yancheng Liu was born in Liaoning, China. He


received the B.S. and M.S. degrees in electrical
engineering from Harbin Industrial University, Harbin,
China, in 1985 and 1988, respectively, and the Ph.D.
degree in marine engineering from Dalian Maritime
University, Dalian, China, in 2002. He is currently a
Professor with Marine Engineering College, Dalian
Maritime University. His research interests include
A.C. motor control, power electronic converters, ship
electrical propulsion technology, renewable energy
systems, and microgrids.

Haohao Guo (Member, IEEE) was born in Jiangxi,


China, in 1984. He received the M.S. and B.S. degrees
from Marine Engineering College, Dalian Maritime
University, Dalian, China, in 2007 and 2012,
respectively. He is currently a Lecturer with Marine
Engineering College, Dalian Maritime University.
Since 2019, he has been a Visiting Scholar with
Aalborg University, Aalborg, Denmark. His research
interests include ship electrical propulsion technology
and PMSM control.

Fengkui Zhang was born in Hebei Province, China,


in 1990. He received the B.Eng. and M.Eng. degrees
in marine engineering from Dalian Maritime
University, Dalian, China, in 2013 and 2016,
respectively He is currently an Experimental Teacher
with Dalian Maritime University. His research
focuses on control of marine permanent magnet
propulsion motor.

© 2023 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.See https://www.ieee.org/publications/rights/index.html for more information.
Authorized licensed use limited to: Murdoch University. Downloaded on May 14,2023 at 15:52:18 UTC from IEEE Xplore. Restrictions apply.

You might also like