IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 34, NO.

3, MAY 2019 1869

Distributed Optimal Tie-Line Power Flow Control

for Multiple Interconnected AC Microgrids
Yun Liu , Member, IEEE, Yuanzheng Li , Member, IEEE, Huanhai Xin , Member, IEEE,
Hoay Beng Gooi , Senior Member, IEEE, and Jianfei Pan , Member, IEEE

Abstract—In a multi-microgrid system (MMG), the microgrids NOMENCLATURE

(MGs) are normally managed by independent operators. Dis-
tributed energy trading/scheduling schemes via interactions of A. Sets and Indices
these MG operators have been extensively investigated. How to SM G /NM G Set/number of the MGs within the MMG.
coordinate these MGs to implement the acquired optimal schedule
in the real time under constant load fluctuation while guaranteeing V/NB Set/number of the DERs within the MMG.
operational stability is seldom reported. Due to the intrinsic advan- Vi /NB i Set/number of the DERs within MG i.
tages of scalability, robustness, and fast response in comparison to E Set of distribution lines within the MMG.
the centralized scheme, a multi-agent based distributed optimal Ei Set of communication links within MG i.
tie-line power flow control strategy is proposed to achieve this ob- Gi Communication network of MG i.
jective, which is facilitated by a regional communication network
overlapping each MG and distributed sensors monitoring the tie- Ni Set of MGs adjacent to MG i.
line power flows. When the MMG is operated in the grid-connected ij Index of DER j in MG i.
mode, the proposed scheme can maintain the scheduled tie-line (ij, ij̃) ∈ E Distribution line within MG i.
power flows among the MGs in the presence of any disturbance (ij, ĩj̃) ∈ E Distribution line connecting MGs i and ĩ.
by adjusting the real-time power outputs of the distributed energy
resources proportionally. When the MMG is islanded, frequency B. Parameters
recovery can be further achieved via a local frequency feedback aG ij , bG ij , cG ij Cost parameters of DER ij.
mechanism. Convergence of the proposed approach is analytically PGmijin /PGmijax Minimum/maximum power of DER ij.
proved. Simulation results in a four-MG system modified from
the IEEE 34-bus test feeder system validate the effectiveness and Ai /wj k i
Adjacency matrix of MG i and its (j, k)th
efficiency of the proposed approach in both grid-connected and entry.
islanded modes. βG ij Participation factor of DER ij.
Index Terms—Consensus, distributed control, frequency regu- Mij Rotor inertia of DER ij.
lation, multi-microgrid system, optimal tie-line power flow. gij, ĩ j̃ Conductance of line (ij, ĩj̃).
bij, ĩ j̃ Susceptance of line (ij, ĩj̃).
α Feedback control gain.
Manuscript received March 19, 2018; revised July 10, 2018 and October 7, kP ij /kI ij Droop/integral gain of DER ij.
2018; accepted November 18, 2018. Date of publication November 26, 2018; C. Control and State Variables
date of current version April 17, 2019. The work of Y. Liu and J. F. Pan was
supported in part by the National Nature Science Foundation of China (NSFC) θij Phase angle of bus ij.
under Grants 51807120 and 51477103, in part by the Guangdong Natural Sci- ωij /ω0 Angular frequency of bus ij and its nominal
ence Foundation under Grants 2015A010106017 and 2016KZDXM007, and in
part by the NSF of SZU Project (2018033). The work of Y. Li was supported in
value.
part by the NSFC under Grant 51707069, and in part by State Key Laboratory PG ij Power output of DER ij.
of Synthetical Automation for Process Industries under Grant PAL-N201806. PL ij Load consumption of bus ij.
The work of H. Xin was supported by the NSFC under Grant U1766206. The
work of J. Ye and H. B. Gooi was supported by the Energy Innovation Program
Peij Net power injection at bus ij.
Office through the National Research Foundation and Singapore Energy Market Pij, ĩ j̃ Power flow from buses ij to ĩj̃.
Authority, under the Project LA/Contract NRF2014EWT-EIRP002-005. Paper Eij Voltage magnitude of bus ij.
no. TPWRS-00393-2018. (Corresponding author: Jianfei Pan.)
Y. Liu and J. F. Pan are with the College of Mechatronics and
λij Incremental cost of DER ij.
Control Engineering, Shenzhen University, Shenzhen 518060, China (e-mail:, ∗i Real-time load forecast error within MG i.
[email protected]; [email protected]). ij Estimation of ∗i at DER ij.
Y. Li is with the School of Automation, Ministry of Education Key Laboratory
of Image Processing and Intelligence Control, Huazhong University of Science
and Technology, Wuhan 430074, China (e-mail:,[email protected]).
H. Xin is with the Department of the Electrical Engineering, Zhejiang Uni-
I. INTRODUCTION
versity, Hangzhou 310027, China (e-mail:,[email protected]). S AN effective way to locally integrate various distributed
H. B. Gooi is with the School of Electrical and Electronic Engineer-
ing, Nanyang Technological University, Singapore 639798 (e-mail:, ehb-
[email protected]).
A energy resources (DERs), microgrids (MGs) are rapidly
expanding recently, e.g., the generation capacity within MGs is
Color versions of one or more of the figures in this paper are available online expected to exceed 1.8 GW by 2018 in the U.S. [1]. However,
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TPWRS.2018.2883407 when a MG, especially one with a high penetration of renewable

0885-8950 © 2018 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.

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY DURGAPUR. Downloaded on June 13,2024 at 14:56:11 UTC from IEEE Xplore. Restrictions apply.
 1870 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 34, NO. 3, MAY 2019

energy operates in islanded mode, it is susceptible to lose achieved by matching the system dynamics with the primal-dual
stability in the face of severe load fluctuation. Interconnecting gradient algorithm. This approach considers details within each
multiple MGs to form a multi-MG system (MMG) has been control area, but requires distributed communication between
recognized as a solution to enhance stability, reliability, and any pairs of physically connected nodes. Thus it is not robust
supply security under disturbances [2]. to communication interruptions. Moreover, it is undesirable that
A MMG is normally operated in a hierarchical way. Over a the uncontrollable load change should be measured within each
long timescale, the energy scheduling which aims to optimize control area in the real time.
the utility of each MG via energy trading should be achieved In this paper, the real-time power sharing control is to be
based on forecast of renewable energy and load consumption. investigated for the MMG where each MG consists of multi-
Extensive work has been done for energy scheduling of MMGs. ple DERs and certain amount of load. The DERs may include
For instance, the authors of [3] proposed a hierarchical bi-level heterogeneous dispatchable distributed generators and energy
operation scheme for the interactive energy management of the storage systems. With the aid of the up-to-date smart grid tech-
MMG. In [4], a dual-decomposition based energy scheduling al- nologies such as P2P communication and distributed sensors
gorithm was designed. It guarantees fair energy trading among monitoring the tie-line power flows, the MMG is transformed
the MGs while minimizing their aggregate operational cost. into a cyber-physical system (CPS) where each DER is man-
The authors of [5] investigated the optimal operation of MMGs aged by a distributed control agent (CA). The CAs within each
via cooperative energy and reserve scheduling. In [6], a dis- MG form a regional communication network where each CA
tributed energy management and trading scheme of the MMG can exchange information with its neighboring CAs in the real
was investigated where an economic interpretation is provided time. Based on this cyber-physical architecture, a novel dis-
to demonstrate that the trading is fair to each individual MG. tributed optimal tie-line power flow (DOTLPF) control is for-
The above works all aim to minimize the operational cost of the mulated. The control objective is for the MMG to distributively
MMG by optimally scheduling its operational point within each implement the results of robust energy scheduling, e.g., [6], in
energy trading period. How to implement the acquired nominal the real time. The rule is for each MG to assimilate its own
operational point in the real time under constant disturbance load/renewable energy fluctuation by dynamically adjusting the
(e.g., load fluctuation) is not considered. power of DERs in proportion to their participation factors so as
Over a short timescale, power sharing control adjusts the to maintain the scheduled tie-line power flows among the MGs
actual power of the DERs to maintain a supply-demand balance and to prevent disturbance propagation which might lead to a
in the presence of load fluctuation while complying with the system-wide blackout. Compared to the relevant works on dis-
energy schedule. If the MMG operates in islanded mode, a tributed real-time power sharing controls of MGs, the novelties
frequency regulation is further required in the transient after of the DOTLPF control are as follows:
disturbance to maintain system stability. Existing works focus 1) Adaptive mode switch: The DOTLPF scheme can adap-
more on the power sharing control of independent MGs [7]. tively switch between islanded mode and grid-connected
For instance, the authors of [8] proposed a distributed optimal mode depending on whether the MMG is connected to the
resource management scheme for the islanded MG, which can upstream transmission system.
minimize the aggregate generation cost while maintaining the 2) Easy implementation: The DOTLPF scheme only requires
nominal system frequency. On this basis, the authors of [9] a regional communication network for each MG, and the
further took the power flow constraints into consideration. DERs just need to locally measure the system frequency
Recently, the authors of [10] initiated the research on dis- when the MMG is islanded. Thus, this scheme can be
tributed coordination control for MMGs. The control scheme easily implemented with a low cost for control and com-
requires distributed communication among the MG agents. Each munication in practice.
MG agent further coordinates its respective DERs in a central- 3) Robustness and flexibility: Different from the fixed com-
ized manner. However, the control objective of reducing energy munication topology required by the primal-dual based
exchanges among the MGs is not economically efficient for the distributed control, the regional communication network
MMG operation. In a MMG, it is suggested that energy trading of the DOTLPF scheme can be randomly configured,
can further reduce the operational cost of the MGs [4], [6]. As which is plug-and-play for the DERs. Moreover, the topol-
such, the tie-line power flows should be maintained constant ogy can be redundantly designed to enhance the robust-
within each dispatch period to comply with the energy sched- ness to communication interruption.
ule. This indicates that the MMG control objective in islanded 4) Long-timescale optimality: Compared to the economic
mode should be similar to that of automatic generation control dispatch policy for the real-time power sharing control
(AGC) in the transmission network. A few papers have investi- of MGs, the participation factor policy adopted by the
gated the distributed version of AGC [7]. For example, the au- DOTLPF scheme can simultaneously achieve real-time
thors of [11] combined the functions of economic dispatch and control and near optimal operation over a long timescale.
AGC and proposed a completely decentralized solution without Thus, it is applicable to inter-temporal optimal operation
the need for communication. The proposed approach simplifies of the MMG as extensive constraints can be dealt with in
each control area as a single node and requires to measure the the robust energy scheduling process.
derivative of frequency deviations, which can be sensitive to the The rest of the paper is organized as follows: Section II for-
measurement noise. In [12], the similar control objective was mulates the problem of optimal tie-line power flow control. To

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY DURGAPUR. Downloaded on June 13,2024 at 14:56:11 UTC from IEEE Xplore. Restrictions apply.
 LIU et al.: DISTRIBUTED OPTIMAL TIE-LINE POWER FLOW CONTROL FOR MULTIPLE INTERCONNECTED AC MICROGRIDS 1871

each bus is connected with a DER and a local load.1 Any non-
DER bus can be eliminated in the analysis following the Kron
reduction of networks. See [14] for more details. It is assumed
that there are NB i buses within MG i ∈ SM G . The set of buses
is denoted as Vi . Thus, there are altogether NB = N M G
i=1 NB i
DERs in the MMG. Each of these DER buses can be denoted as
ij ∈ Vi (∀j ∈ {1, . . . , NB i }), and its dynamics can be equiva-
lently characterized by the classic swing equations [15]:
Fig. 1. The cyber-physical architecture of the MMG. θ̇ij = ωij − ω0 , (1)
Mij ω̇ij = PG ij − PL ij − Peij . (2)
address the problem, Section III introduces the DOTLPF control Peij can be calculated following the Kirchoff’s Law and AC
approaches in both grid-connected and islanded modes. Simu- power flow equation as
lation results and conclusions are presented in Section IV and  
V, respectively. Peij = Pij,i j̃ + Pij, ĩ j̃ , (3)
With respect to mathematical notations, any variable x with- (ij,i j̃ )∈E (ij, ĩ j̃ )∈E
out superscript denotes the real-time value; x∗ denotes its sched-  
uled (optimal) value; Δx denotes the deviation of x from x∗ ;
Power flows within MG i Power flows among different MGs

0/1 denotes the all-zero/one vector or matrix of proper sizes; Pij, ĩ j̃ = gij, ĩ j̃ Eij Eij − Eĩ j̃ cos (θij − θĩ j̃ )
I denotes the identity matrix of proper sizes; and a saturation
function is defined as − bij, ĩ j̃ Eij Eĩ j̃ sin (θij − θĩ j̃ ). (4)
⎧
⎪
⎨ min if x < min (ij, ĩj̃) ∈ E indicates that there exists a distribution line be-
Sat(x, min, max) = max if x > max . tween buses ij and ĩj̃. i = ĩ means that the distribution line
⎪
⎩x is a tie-line between MGs i and ĩ, otherwise it is an internal
otherwise
line within MG i. The neighboring MG set of MG i is de-
noted by Ni . A MG ĩ ∈ Ni if ∃j, j̃ such that (ij, ĩj̃) ∈ E. MG
II. PROBLEM FORMULATION
0 ∈ Ni if there exists a j such that (0, ij) ∈ E. The bus volt-
In this paper, a CPS refers to a MMG including multiple age magnitude Eij (∀ij) can be restored to the reference value
DERs interconnected by physical distribution networks which via reactive power coordination which has been extensively and
are controlled by computer-aided CAs interconnected by com- independently investigated in the literatures. Interesting read-
munication networks. This architecture, which enjoys the merits ers may refer to [16] for the basic hierarchical reactive power
of fast computing and control, flexibility, scalability, autonomy, control which is also implemented in the case studies due to its
and robustness to single-point-failure/communication interrup- simplicity, and [17], [18] for the advanced distributed reactive
tion [9], [13], is extensively applied for managing complex sys- coordination schemes.
tems. In this section, the physical layer of the MMG is firstly
described. This is then followed by formulation of the main B. Control Objective
control objectives.
Based on the energy scheduling approach of MMGs (e.g., [4],
[19]–[21]), the optimal scheduling results should satisfy
A. Modeling of the MMG
Consider a MMG as an interconnection of NM G MGs, whose 
N Bi 
(PG∗ ij − PL∗ ij ) + ∗
Pij, = 0, ∀i ∈ SM G , (5)
set is denoted as SM G = {1, . . . , NM G }. These MGs can be j =1 ( i j , ĩ j̃ ) ∈E
ĩ j̃

further connected to the main grid through a grid-tied switch, ĩ ∈Ni

as illustrated in Fig. 1. With a slight abuse of notation, the main where PG∗ ij , PL∗ ij and Pij,
∗
are the scheduled values of PG ij ,
grid is indexed as MG 0 and is simplified as a single bus 0. When ĩ j̃
PL ij and Pij, ĩ j̃ .
the switch is closed, the MMG operates in grid-connected mode,
However, because of the time-varying renewable generations
and thus its system frequency is synchronized to the main grid.
and load demand, PL ij may constantly fluctuate around PL∗ ij
When the switch is open, the MMG operates in islanded mode,
within each dispatch time slot. Consequently, the power of DERs
and thus it should achieve frequency regulation by itself. It is
PG ij , ∀ij ∈ Vi should be dynamically controlled to make the
also possible for a certain MG to break away from the MG
DERs within each MG absorb its own load fluctuation following
cluster and operate independently. In this situation, extensive
centralized/distributed control schemes have been proposed to 
N Bi 
N Bi

maintain its operational optimality and stability, e.g., [8], [9] and (PG ij − PG∗ ij ) = (PL ij − PL∗ ij ), (6)
the references therein, which are out of the scope of this work. j =1 j =1

The MMG can be denoted as an undirected graph (V, E)

where V and E ⊆ V × V are the sets of buses and distribution 1 Renewable energy systems, e.g., photovoltaic and wind generators, are
lines, respectively. Without loss of generality, it is assumed that viewed as negative loads without loss of generality.

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY DURGAPUR. Downloaded on June 13,2024 at 14:56:11 UTC from IEEE Xplore. Restrictions apply.
 1872 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 34, NO. 3, MAY 2019

designed to dynamically and adaptively adjust the actual power

of the DER in proportion to βG ij following
PG ij = PG∗ ij + βG ij ∗i , ∀j ∈ Vi (9)
N B i ∗
 NB i ∗
where j =1 βG ij = 1 and i = j =1 (PL ij − PL ij ) is the
real-time load forecast error within MG i. It should be noted that
control law (9) can be applied to both heterogeneous dispatch-
able distributed generators as well as energy storage devices.
Their corresponding operational constraints do not need to be
considered in the real-time control stage since they have been
included in the robust energy scheduling optimization and will
Fig. 2. Comparison of objectives and timescales of the robust energy schedul- not be violated as long as the actual load fluctuation is within
ing [6] and the proposed DOTLPF control.
the forecast load band.
Remark 1: According to (9), the participation factor policy
so that the power exchanges among the MGs are regulated un- complies with the results of the robust energy scheduling ap-
∗ proach for MMGs, e.g., [6], which aims to achieve the opera-
changed in the real time, i.e., Pij, ĩ j̃ = Pij, ĩ j̃
for all (ij, ĩj̃) ∈ E
tional optimality over a long timescale, e.g., one day. However,
with ĩ ∈ Ni . To satisfy (6), two different power sharing policies
due to the existence of a time-varying load forecast error, the
can be adopted to adjust the power of DERs.
real-time operational point of the MMG always deviates from
Economic dispatch policy: The DERs within each MG satisfy
the optimal energy schedule slightly. In light of this, the real-

N Bi time operational point is termed as near optimality over a long
min Generation Cost = costij (PG ij ), ∀i ∈ SM G (7a) timescale. Contrarily, the economic dispatch policy (7) which
j =1 is adopted extensively in references (e.g., [8], [9], [12]) only
subject to optimizes the DER power outputs based on the current load
level, thus it can only achieve real-time optimality. In the fol-

N Bi  lowing section, it is to be illustrated that the two policies are
∗
(PG ij − PL ij ) + Pij, = 0, ∀i ∈ SM G , (7b)
ĩ j̃ intrinsically the same for control design.
j =1 ( i j , ĩ j̃ ) ∈E
ĩ ∈Ni At this point, the objectives of optimal tie-line power flow
control can be summarized as follows:
PGmijin ≤ PG ij ≤ PGmijax , ∀j ∈ Vi . (7c) 1) Grid-connected mode: In response to any random load
The economic dispatch intends to minimize the total gen- disturbance, the power of DERs can be regulated to a
eration cost where the cost of each DER including both the steady state where either the economic dispatch policy
distributed generator and energy storage system within each or the participation factor policy is satisfied, and the MG
MG can be uniformly modeled as [22], [23] power exchanges are regulated to the scheduled values.
2) Islanded mode: Besides satisfying either of the power
1
aG ij PG2 ij + bG ij PG ij + cG ij .
costij (PG ij ) = (8) sharing policies, the system frequency is further stabilized
2 to the reference value, i.e., limt→+∞ ωij = ω0 , ∀ij.
Constraint (7b) indicates the real-time power balance within
each MG, which can be derived following (5) and (6). Constraint III. THE DOTLPF CONTROL APPROACHES
(7c) is the power limit of each DER.
Participation factor policy [24]: Over a long timescale, the This section first introduces the requirement for the com-
multiple MG operators interact and compete with each other to munication infrastructure design based on which the DOTLPF
optimize the expected operational cost of each MG via energy control approaches in grid-connected and islanded modes are
trading using the robust energy scheduling optimization, e.g., presented, respectively.
[6]. Various operational constraints, e.g., power constraints of
DERs, state of charge constraints of energy storage devices, A. Design of the Cyber Layer
power flow constraints, etc. can be included in this formulation. To facilitate the DOTLPF control, a regional communication
Specifically, the nominal operational point of each DER (PG∗ ij ) network can be set up for each MG i ∈ SM G . Communication
is calculated to serve the forecast load within the specific time between the neighboring MGs is not required, as illustrated in
slot. Meanwhile, a non-negative participation factor βG ij is also Fig. 1. The regional communication network of MG i can be
required to ensure the existence of a feasible solution as long as depicted with an undirected graph Gi = (Vi , Ei , Ai ). Vi which
the load fluctuation is within the forecast uncertainty set. The is the set of buses in the physical layer of MG i also refers to the
function of robust energy scheduling in the overall operational set of MG i’s CAs in the cyber layer. CAs integrate the func-
scheme of a MMG is shown in Fig. 2. Although extensive works tions of real-time control, computation, communication, and
have been done to determine the optimal schedule required by information sensing. Specifically, the power exchange with any
the participation factor policy, how to implement these results neighboring MG j ∈ Ni is required to be measured by at least
effectively and maintain system stability in the real time control one CA within MG i. To alleviate the communication burden,
of a MMG is seldom report. As such, a DOTLPF control will be it is preferred that the CA at the bus which is directly connected
Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY DURGAPUR. Downloaded on June 13,2024 at 14:56:11 UTC from IEEE Xplore. Restrictions apply.
 LIU et al.: DISTRIBUTED OPTIMAL TIE-LINE POWER FLOW CONTROL FOR MULTIPLE INTERCONNECTED AC MICROGRIDS 1873

Fig. 3. Control block diagram of the cyber-physical MMG.

to an adjacent MG should locally measure the corresponding where λ∗i is the optimal IC. It can be observed that (9) and
power flow. For example in Fig. 1, CA 12 can locally measure (12) are equivalent in format by assuming 1/aG ij = βG ij , −
the power exchange between MGs 1 and 2. Moreover, it is also bG ij /aG ij = PG∗ ij , PGmijin = −∞, and PGmijax = +∞. Without
required that each CA should measure the system frequency of loss of generality, the participation factor policy is adopted here-
the local bus when islanded operation is detected. Ei ⊆ Vi × Vi inafter to illustrate the control design and simulation results.
is the set of communication links within MG i. (ij, ij̃) ∈ Ei When the MMG is in grid-connected mode, each DER should
indicates that the CAs at buses ij and ij̃ can exchange infor- operate in grid-feeding mode, i.e., to inject power to the MG fol-
mation using two-way communication technology. Unlike [12], lowing a reference value [27]. Within any of the interconnected
the topology of MG i’s regional communication network is not MG i ∈ NM G , the power of the DERs should be controlled
required to coincide with its physical interconnection topology, following (9) as
which can enhance the flexibility and robustness of the pro-
PG ij = PG∗ ij + βG ij ij , ∀j ∈ Vi . (13)
posed control approach. For instance, based on the cooperative
control theory, a connected communication topology (i.e., there ∗i
Different from (9), (13) replaces by ij which is a local es-
exists at least one path between every pair of the CAs) is a suf- timation of ∗i and is expected to track ∗i in the steady state.
ficient condition for the state of each CA to reach a consensus To achieve this tracking objective, ij can be dynamically ad-
[25]. Therefore, the consensus algorithm can be made robust to justed based on distributed communication and locally measured
communication interruptions by adopting a redundant commu- power flows. For bus ij which is directly connected to a neigh-
nication topology [26]. Ai ∈ RN B i ×N B i is the adjacency matrix boring MG, i.e., ∃(ij, ĩj̃) ∈ E with ĩ ∈ Ni , its ij can be updated
corresponding to graph Gi . Its elements are defined as as
 
 i ˙ij = wji k (ik − ij ) − α ΔPij, ĩ j̃ , (14a)
wj k > 0 if (ij, ik) ∈ E
[Ai ]j k = [Ai ]k j = . (10) (ij,ik )∈Ei ( i j , ĩ j̃ ) ∈E
0 otherwise ĩ ∈Ni

otherwise,
B. Control Design of the MMG in Grid-Connected Mode 
˙ij = wji k (ik − ij ), (14b)
First of all, the consistency between the economic dispatch (ij,ik )∈Ei
policy and participation factor policy in the control design is to
where α > 0 is a sufficiently small control parameter, and
be validated. Specifically, the incremental cost (IC) of DER ij ∗
ΔPij, ĩ j̃ = Pij, ĩ j̃ − Pij, is the deviation of MG tie-line
in the economic dispatch policy can be defined as ĩ j̃
power
 flow from its scheduled value. Intuitively, the term
dcostij (ij,ik )∈Ei w i
j k ik( −  ij ) in (14) intends to synchronize differ-
λij = = aG ij PG ij + bG ij . (11) ent ij within MG i to a consensus value based on cooperative
dPG ij
control theory [25], while the term ΔPij, ĩ j̃ in (14a) functions as
The optimal solution of (7) follows the well-known equal IC a negative feedback to drive the consensus value to the actual ∗i .
criterion [22] as The block diagram of the cyber-physical MMG under the
  DOTLPF control is shown in Fig. 3. It illustrates the flow of
1 ∗ bG ij m in m ax
PG ij = Sat λ − ,P ,P , (12) information within each MG. Note that when operating in grid-
aG ij i aG ij G ij G ij connected mode, the CAs’ output signals are directly sent to
Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY DURGAPUR. Downloaded on June 13,2024 at 14:56:11 UTC from IEEE Xplore. Restrictions apply.
 1874 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 34, NO. 3, MAY 2019

TABLE II
LOAD DATA OF THE 4-MG SYSTEM

Fig. 4. Single-line diagram of the 4-MG system.

TABLE I
OPERATIONAL PARAMETERS OF THE DERS IN THE 4-MG SYSTEM

the DERs. The convergence property of the proposed control in

grid-connected mode can be summarized as
Theorem 1: Assume that the regional communication net- Fig. 5. Responses of the MMG to load change in grid-connected mode.
work of each MG is connected. The power output of each DER
ij is determined by (13) where the estimated forecast error ij
Otherwise,
is updated following (14) based on distributed communication

and local power flow measurement. By selecting a sufficiently ˙ij = wji k (ik − ij ) − kI ij Δωij , (17b)
small control parameter α, the CPS is locally exponentially sta- (ij,ik )∈Ei
ble, and the system operational point can achieve the control
objective in grid-connected mode in Subsection II-B. where kI ij = βG ij is the integral parameter of DER ij.
The proof is detailed in Appendix A. Since the dynamics of the DOTLPF control (17) is relatively
slow due to the small feedback gains, the transient after dis-
C. Control Design of the MMG in Islanded Mode turbance is mainly governed by the droop control (15). Conse-
quently, the immediate response of the MMG is similar to that
When the MMG is in islanded mode, each DER should op- under the conventional hierarchical control. As time goes on, fre-
erate in grid-supporting mode, i.e., to regulate the system fre- quency deviation is gradually corrected via a negative feedback
quency by controlling the active power delivered to the grid [27]. mechanism to recover the reference frequency ω0 . Specifically,
Specifically, within any of the interconnected MG i ∈ NM G , the when the total reference power of DERs is less than the total
power of the DERs should be controlled following the classic power consumption, the primary frequency control (15) results
droop curve as in Δωij < 0. This can bring up the DER reference power by
PG ij = PGnom
ij − kP ij Δωij . (15) increasing its ij following (17) until the frequency deviation
vanishes. The converse is also true.
PGnom
ij is the nominal power of DER ij determined by The MMG in islanded mode under the DOTLPF control is
∗ shown in Fig. 3 where all the switches should be in islanded
ij = PG ij + βG ij ij , ∀j ∈ Vi ,
PGnom (16)
position. The convergence property of the proposed control in
where ij should be dynamically controlled to track ∗i . For islanded mode can be summarized as
bus ij which is directly connected to a neighboring MG, i.e., Theorem 2: Assume that the regional communication net-
∃(ij, ĩj̃) ∈ E with ĩ ∈ Ni , its ij can be updated as work of each MG is connected. Each of the DERs operates in
  grid-supporting mode (15). The nominal power output of each
˙ij = wji k (ik − ij ) − kI ij Δωij − α ΔPij, ĩ j̃ ,
DER ij is determined by (16) where ij is updated following
(ij,ik )∈Ei ( i j , ĩ j̃ ) ∈E
ĩ ∈Ni (17) based on distributed communication as well as the local
(17a) system frequency and power flow measurement. By selecting

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY DURGAPUR. Downloaded on June 13,2024 at 14:56:11 UTC from IEEE Xplore. Restrictions apply.
 LIU et al.: DISTRIBUTED OPTIMAL TIE-LINE POWER FLOW CONTROL FOR MULTIPLE INTERCONNECTED AC MICROGRIDS 1875

Fig. 6. Responses of the MMG to load change in islanded mode.

Fig. 8. Responses of the MMG to switch on/off of DER 35 in islanded mode.

Fig. 7. Responses of the MMG to switch on/off of DER 35 in grid-connected
mode.

A. Response of the MMG to Load Variations

a sufficiently small control parameters α, the MMG is locally In the scenario of grid-connected mode, L2 in MG 3 in-
Lyapunov stable, and the system operational point can achieve creases by 0.2 MW at t = 10 s, and L5 in MG 1 increases by
the control objective in islanded mode in Subsection II-B. 0.405 MW at t = 100 s. The responses of the MMG are de-
The proof is detailed in Appendix B. picted in Fig. 5 which indicates that a new steady state can be
reached after approximately 300 seconds. Figs. 5(a)–5(d) show
the variations of the local estimation of load forecast errors in
IV. SIMULATION RESULTS
MGs 1-4, respectively. It can be observed from Figs. 5(a)–5(d)
In this section, the DOTLPF control is validated in a 4-MG that the responses of load forecast errors estimated at different
system modified from the IEEE standard 34-bus test feeder with DERs within a same MG are almost identical, which indicates
DIgSILENT PowerFactory. To simulate the scenario of real im- that the power adjustments of the DERs are in proportion to
plementation, the DOTLPF control law is discretized using the their participation factors within the same MG following (13).
trapezoid rule method [28] with a step size of 0.05 second. In the figures hereinafter, only the forecast error of one DER
The single-line diagram is shown in Fig. 4 where MGs 1 and within each MG will be plotted to represent the consensus load
2 have 3 DERs each, MG 3 has 5 DERs, while MG 4 has forecast errors of MGs 1 ∼ 4 (denoted as 1 ∼ 4 ) unless oth-
4 DERs. The power capacity of each DER is uniformly set erwise stated. Moreover, consistent to the control objective, the
to 0.5 MW. Initially, DER 35 is disconnected from MG 3 to DER power outputs within MG 3 and MG 1 increase together
test the plug-and-play capability in Section IV-B. These DERs with the consensus load forecast errors 3 and 1 respectively at
are coordinated based on the participation factor policy. The t = 10 s and t = 100 s to counteract their internal load increases,
nominal power output (PG∗ ij ) and participation factor (βG ij ) while those of MGs 2 and 4 remain unchanged after a transient
of each DER is optimized using the distributed robust energy caused by its adjacent MGs as no internal load change happens.
management scheme proposed in [6]. The operational parame- Fig. 5(e) illustrates the responses of DER power outputs within
ters of the DERs in the dispatch time slot considered are tab- MG 1. It indicates that the power differences before and after
ulated in Table I. The control parameter α is tuned as 0.05. the disturbance for DERs 11–13 are 0.067 MW, 0.133 MW, and
Loads L1-L9 are connected to the the 4-MG system from dif- 0.203 MW, respectively, which exactly covers the 0.405 MW
ferent locations as in Fig. 4. The initial load data is tabulated in load increase within MG 1. Due to the tie-line power flow con-
Table II. In the following, the proposed control will be validated trol, the power exchanges among the MGs can be recovered to
in grid-connected mode and islanded mode, respectively. The their respective initial levels, as shown in Fig. 5(f).
response of the MMG to mode switch between the two will be In the scenario of islanded mode, L1 in MG 1 decreases by
investigated as well. 0.4 MW at t = 20 s, and L2 in MG 3 decreases by 0.2 MW at
Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY DURGAPUR. Downloaded on June 13,2024 at 14:56:11 UTC from IEEE Xplore. Restrictions apply.
 1876 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 34, NO. 3, MAY 2019

Fig. 9. Islanding and reconnection of the MMG to the main grid.

t = 100 s. The responses of the MMG are depicted in Fig. 6 When the MMG is in islanded mode, a simulation is per-
which indicates that a new steady state can be reached after formed in the same scenario, and the results are shown in Fig. 8.
approximately 200 seconds. Fig. 6(a) plots the consensus load The analysis is similar to that of grid-connected mode except
forecast errors 1 ∼ 4 . Similar to grid-connected mode, the de- that a frequency fluctuation can be observed before it recovers
creases of 1 and 3 lead to decreases of the corresponding DER to 60 Hz. Fig. 8(c) further indicates that when a communication
power outputs, which counteract the load decreases within MG delay of 0.4 second is considered, the frequency fluctuation is
1 and MG 3. On the other hand, 2 and 4 remain unchanged only slightly severer.
as no internal load change takes place. The function of tie-line
power flow recovery can also be achieved in islanded mode, as C. Islanding and Reconnection of the MMG to the Main Grid
indicated by Fig. 6(b). Fig. 6(c) shows that the system frequency
Finally, the effect of islanding and reconnection to the main
can be recovered to ω0 = 60 Hz after the transient caused by
grid is tested. At t = 50 s, the MMG is disconnected from the
the load variations. It also investigates the influence of commu-
main grid due to a certain main grid fault. After the main grid
nication delay on the control performance. Specifically, when a
fault is cleared, the MMG is resynchronized and reconnected to
communication delay of 0.4 second which is sufficiently longer
the main grid at t = 450 s. The simulation results are depicted
than that adopted in [8] is considered, a more severe frequency
in Fig. 9. Fig. 9(a) plots the consensus load forecast errors of the
fluctuation can be observed, but the frequency recovery can still
4 MGs. It can be observed that before islanding of the MMG,
be achieved.
MG 1 exports 0.4 MW power to the main grid. After islanding,
It is also worth mentioning that measurement errors are in-
a frequency fluctuation can be observed in Fig. 9(c) before re-
evitable in practical implementation of the DOTLPF control
covering to 60 Hz after about 300 seconds. After this transient,
[29], which may cause the operational states to fluctuate slightly
the consensus load forecast error 1 decreases to some extent as
around the steady states. Nevertheless, this fluctuation is nor-
no power is further exported to the main grid, while the consen-
mally small and will only cause a trivial influence to the opera-
sus load forecast errors 2 ∼ 4 are approximately recovered to
tion of the MMG.
their initial levels to maintain the MG power exchanges. After
the MMG is reconnected to the main grid, it takes approximately
400 seconds for 1 ∼ 4 and the power exchanges among the
B. Test of the Plug-and-Play Capability MGs/the main grid to be recovered to their initial levels at the
steady state.
Next, the plug-and-play capability of the proposed control
approach is to be validated. In grid-connected mode, a backup
V. CONCLUSION
DER 35 is connected to MG 3 at t = 50 s and disconnected
again at t = 350 s. The responses of the MMG are depicted in In this paper, a distributed optimal tie-line power flow con-
Fig. 7. Specifically, Fig. 7(a) shows the consensus load forecast trol approach was proposed for the multiple-microgrid system.
errors of MGs 1, 2, and 4, whose steady-state values remain The proposed approach was implemented in the framework of
unchanged as no internal disturbances occur. The estimated load a multi-agent cyber-physical system based on the regional com-
forecast errors and power outputs of the DERs within MG 3 are munication within each microgrid and distributed sensors moni-
depicted in Fig. 7(b) and (c). It can be observed that when toring the tie-line power flows. The objective is to implement the
DER 35 is plugged in, it gradually picks up load and shares the optimal energy schedule of each time slot in the real time while
power together with DERs 31–34. Consequently, 31 ∼ 34 will maintaining system stability. This approach is applicable to the
slightly decrease in about 150 seconds. Since DER 31 and DER multiple-microgrid system operated in both grid-connected and
32 have lower participation factors than DER 33 and DER 34, islanded modes. In grid-connected mode, the proposed control
their power outputs decrease less in this process. When DER can dynamically regulate the power exchanges to the sched-
35 is switched off, it takes approximately 150 seconds again for uled values among the MGs, while distributed energy resources
31 ∼ 34 to recover to the initial level. Fig. 7(d) shows that the within each microgrid can share its internal load fluctuation
power exchanges among the MGs stay unchanged in the steady based on the participation factor policy or economic dispatch
state. policy. In islanded mode, the system frequency can be further

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY DURGAPUR. Downloaded on June 13,2024 at 14:56:11 UTC from IEEE Xplore. Restrictions apply.
 LIU et al.: DISTRIBUTED OPTIMAL TIE-LINE POWER FLOW CONTROL FOR MULTIPLE INTERCONNECTED AC MICROGRIDS 1877

regulated to the nominal value. Simulation results based on a By selecting the bus at MG 0 as the reference bus satisfying
4-microgrid system modified from the IEEE 34-bus test feeder Δθ0 = 0, (20) can be reduced to
demonstrated the effectiveness of the proposed approach in both
ΔP e = BΔθ, (22)
operational modes.
One of the future works is to explore how to systematically where B ∈ RN B ×N B is equivalent to B excluding the first row
integrate this real-time control scheme and the robust energy and column.
scheduling scheme over a long timescale by efficiently and re- Define ΔF in = [ΔF in1 , . . . , ΔF inN 
M G ] with ΔF ini =

liably sharing the same communication infrastructure so that a [ΔFini1 , . . . , ΔFiniN B i ] . ΔFinij = (ij,i j̃ )∈E ΔPij,i j̃ indi-
fully distributed and robust operation architecture can be con- cates power flowing from bus ij to other buses within MG
structed for the multi-microgrid system. i. Following from (18), ΔF in can be calculated as
ΔF in = Bin Δθ, (23)
APPENDIX A
PROOF OF THEOREM 1 where
⎧
The proof of Theorem 1 can be proceeded in two steps: ⎪
⎨ −β if (ij, ij̃) ∈ E
 ij,i j̃
Step 1. Linearization: Linearizing (4) around the optimal op- [Bin ]ij,i j̃ = (ij,il)∈E βij,il if ij = ĩj̃ .
⎪
⎩0
eration point leads to otherwise
ΔPij, ĩ j̃ = βij, ĩ j̃ (Δθij − Δθĩ j̃ ) + bij, ĩ j̃ ΔEij + eij, ĩ j̃ ΔEĩ j̃ Similarly, define ΔF out = [ΔF out1 , . . . , ΔF outN M
G
] with

ΔF outi = [ΔFouti1 , . . . , ΔFoutiN B i ] . ΔFoutij = ( i j , ĩ j̃ ) ∈E
= βij, ĩ j̃ (Δθij − Δθĩ j̃ ). (18) ĩ ∈Ni
ΔPij, ĩ j̃ indicates power flowing from bus ij to other MGs.
For each of the buses within the MMG, distributed voltage ΔF out can be calculated as
restoration control approach can be applied to recover its volt-
ΔF out = Bout Δθ, (24)
age to the reference value, i.e., ΔEij ≈ 0 for all ij, which
leads to the second equality of (18). Parameter βij, ĩ j̃ can be where ΔF out = ΔP e − ΔF in and Bout = B − Bin .
calculated as Step 2. Stability of the Closed-Loop Error System: Define
Δ = [Δ1 , . . . , ΔN M G ] with Δi = [Δi1 , . . . , ΔiN B i ] .
β−
i j , ĩ j̃
 The updating of forecast error (14) linearized around the optimal
∗
βij, ĩ j̃ = Eij Eĩ∗j̃ gij, ĩ j̃ ∗
sin (θij − θĩ∗j̃ ) operation point can be written in the matrix form as
∗
−Eij Eĩ∗j̃ bij, ĩ j̃ cos (θij
∗
− θĩ∗j̃ ) . (19) Δ˙ = −LΔ − αΔF out = −LΔ − αBout Δθ, (25)
 where L = diag(L1 , . . . , LN M G ) with Li being the Laplacian
β+
i j , ĩ j̃ matrix of the communication topology corresponding to MG
i. The Laplacian matrix is defined as Li = Di − Ai with the
In an MV or LV distribution network, the line impedance is
degree matrix Di = diag(Ai 1).
not purely inductive, but has a non-negligible resistance part.
− With Δωij = 0 across the MMG in grid-connected mode, it
Nevertheless, βij, ĩ j̃
which accounts for line loss can gener-
follows from (2) and (22) that
ally be ignored in the analysis for the following reasons: i)
the output impedance of a inverter-based DER is typically in- ΔP G − ΔP L = ΔP e = BΔθ, (26)
ductive, which makes the inductive part dominate the resis-
where ΔP G and ΔP L are similarly defined as ΔP e . In the
tive part, i.e., |bij, ĩ j̃ | > |gij, ĩ j̃ | [14]; ii) the voltage angle differ-
following analysis, it is assumed that ΔP L = 0 (ΔP L = 0 is
ences of the distribution lines are normally within 3◦ [30], thus
∗ equivalent to the shift of ΔP G in the steady state). It can be de-
cos(θij − θĩ∗j̃ )  sin(θij ∗
− θĩ∗j̃ ). Therefore, the MMG can be
rived from (13) that ΔP G = ΛΔ, where Λ is a diagonal matrix
+
treated as a lossless system by assuming βij, ĩ j̃ ≈ βij, ĩ j̃
. Based with the entry corresponding to bus ij being βG ij . Therefore,
on (18), (3) can be written in a matrix form as considering (25) and (26), the dynamics of the closed-loop error
    system can be described as
ΔPe0 Δθ0
=B , (20) Δ˙ = −(L + αBout B −1 Λ)Δ. (27)
ΔP e Δθ
The stability of (27) can be characterized by its system ma-
where ΔP e = [ΔP e1 , . . . , ΔP eN M G ] and Δθ = [Δθ 1 , . . . , trix. Define L̃(α) = L + αBout B −1 Λ which can be viewed as
Δθ N M G ] ; P ei and θ i can be further expanded as P ei = L perturbed by αBout B −1 Λ. In the sequel, it is to be illustrated
[Pei1 , . . . , PeiN B i ] and θ i = [θi1 , . . . , θiN B i ] ; and matrix B ∈ that all the eigenvalues of the system matrix have negative real
R(N B +1)×(N B +1) is defined as parts using matrix perturbation theory [31, Lemma 7].
⎧
⎪ Due to the connectivity of MG ith’s regional communi-
⎨ −β  ij, ĩ j̃
if (ij, ĩj̃) ∈ E
cation network, its corresponding Laplacian matrix Li has a
[B]ij, ĩ j̃ = (ij,k l)∈E βij,k l if ij = ĩj̃ . (21)
⎪
⎩0 simple eigenvalue 0 and NB i − 1 positive eigenvalues [25,
otherwise Theorem 4.31]. It is straightforward that the eigenvalues of L

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY DURGAPUR. Downloaded on June 13,2024 at 14:56:11 UTC from IEEE Xplore. Restrictions apply.
 1878 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 34, NO. 3, MAY 2019

are the union of the eigenvalues of Li (∀i ∈ NM G ). Therefore, corresponding to the ith eigenvalue of B, i.e., λi (B). Specif-
L has NM G eigenvalues 0 and the other eigenvalues are posi- ically, B has a simple eigenvalue λ1 (B) = 0 with its corre-
tive. Construct matrices X = [e1 , . . . , eN M G ] ∈ RN B ×N M G and sponding left eigenvector ν 1 = 1q since it is a symmetrical
Y = ΓX  , where Γ ∈ RN M G ×N M G is a diagonal matrix with and irreducible matrix [25]. Consider (29) and apply the state
the ith entry being 1/NB i , and ei is a vector with its entries transformation to Δθ G as TB Δθ = [0, δθ  ] ⇒ Δθ = TB δθ,
corresponding to the buses in MG i being 1 while other entries where δθ ∈ RN B −1 is the transformed state vector. Thus, (32)
being 0. It can be readily verified that Y X = I. The columns can be rewritten as
(rows) of X (Y ) are the right (left) eigenvectors of eigenvalues 1   1 1
0 of L. Based on the matrix perturbation theory, the positive V = δθ B δθ + 2 Δω  M Δω + 2 Δ2
α2 α α
eigenvalues of L remains positive for L̃(α) when α is suffi-
2
ciently small, and the perturbation on the zeros eigenvalues can + δθ  TB M Δω − Δ M Δω, (33)
be characterized by the eigenvalues of α

Y Bout B −1 ΛX = Y (B − Bin )B −1 ΛX where B  = TB Br TB = diag{[λ2 (Br ), . . . , λN B (Br )]} is posi-

tive definite. When α → 0+ , the positive definite terms of the
= Y ΛX − Y Bin B −1 ΛX = Y ΛX. (28) state vectors (in the first line of (33)) have higher magnitudes,
The third equality holds considering Y Bin = 0. It can be verified which can bound the indefinite terms with lower magnitudes (in
ΛX ∈ RN M G ×N M G is a diagonal matrix with its ith entry
that Y  the second line of (33)). Therefore, the Lyapunov function V is
being N Bi −1 positive definite.
j =1 aG ij /NB i . Therefore, the zero eigenvalues of L̃(0)
Define another state transformation matrix TL = [TL 1 , TL 2 ] ,
become positive for L̃(α) for a sufficiently small α. Since all where TL 1 = [e1 , . . . , eN M G ] and TL 2 ∈ RN B ×(N B −N M G ) are
the eigenvalues of −L̃(α) are negative, the error system (27) is the sets of left eigenvectors corresponding to the 0 eigenvalues
stable. This concludes the proof of Theorem 1. and positive eigenvalues of L, respectively. The state trans-
formation leads to TL Δ = [δ1 , δ2 ] ⇒ Δ = TL 1 δ1 +
APPENDIX B
TL 2 δ2 , where δ1 ∈ RN M G and δ2 ∈ RN B −N M G are the
PROOF OF THEOREM 2
transformed states.
In the optimal steady state, the synchronized motion of Δθ is The time derivative of the Lyapunov function (32) along the
not unique, i.e., the optimal phase angle vector can be θ ∗ + 1c trajectory of (31) can be calculated as
for any c ∈ R. Rather than picking a reference bus as in the  
proof of Theorem 1, θ ∗ is selected to satisfy 2 2
V̇ = − Δθ  BΔθ − Δω  RΔω − Δ L + Λ Δ
α2 α
1 Δθ = 1 θ − 1 θ ∗ ≡ 0. (29)  
2
∗
Hence, 1 θ̇ = 1 θ̇ = 1 Δω. To guarantee that P ∗G ≡ Bθ ∗ , + Δθ  (2Bout M − KP )Δω + Δθ  Bin + Λ Δ
α
each entry of θ ∗ should have the same changing rate, i.e.,  
  2 2
∗ 1 1 + Δω  KP + M L Δ
θ̇ = 1Δω ⇒ Δθ̇ = I − 1 Δω. (30) α α
NB NB
2
In islanded mode, the dynamics of the closed-loop error sys- = − δθ  B  δθ − Δω  RΔω − δ1 TLT1 ΛTL 1 δ1
α
tem can be characterized by  
⎡ ⎤ ⎡ ⎤⎡ ⎤ 2 2
1 − δ2 TLT2 L + Λ TL 2 δ2
Δθ̇ I− 1 Δθ α2 α
⎢ ⎥ ⎢ ⎥
⎣ Δω̇ ⎦ = ⎢
NB ⎥⎢ ⎥
⎣ −M −1 B −M −1 KP M −1 Λ ⎦ ⎣ Δω ⎦ . + δθ  TB (2Bout M − KP ) Δω + δθ  TB ΛTL 1 δ1
Δ˙ −αBout −Λ −L Δ  
  2 2
(31) + δθ TB Bin + Λ TL 2 δ2 + Δω  KP TL 1 δ1
α α
where Δω = [Δω 1 , . . . , Δω N M G ] with Δω i = [Δωi1 , . . . , 2
ΔωiN B i ] ; M and KP are diagonal matrices with the entry + Δω  (KP + M L)TL 2 δ2 (34)
α
corresponding to DER ij being Mij and kP ij , respectively. The
dynamics of Δθ follow from (30); the dynamics of Δω follow where R = α22 KP − α2 M Λ − M (I − N1B 1). The second
from (2), (22), (15) and (16); and the dynamics of Δ follow equality follows from the state transformations by considering
from (17). that Bin TL 1 = 0 and LTL 1 = 0. When α → 0+ , the negative
Choose a Lyapunov candidate as definite terms of the state vectors (in the first two lines of the
second equality of (34)) have higher magnitudes, which can
1 1
V = 2
Δθ  BΔθ + 2 Δω  M Δω bound the indefinite terms with lower magnitudes (in the last
α α
three lines of the second equality of (34)). Therefore, the Lya-
1 2
+ 2 Δ2 − Δ M Δω + Δθ  M Δω. (32) punov function V̇ is negative definite, and thus the error system
α α (31) is locally Lyapunov stable. Considering condition (29),
Define a state transformation matrix TB = [ν 1 , TB ] , [Δθ  , Δω  , Δ ] → 0 in the steady state. This concludes the
where TB = [ν 2 , . . . , ν N B ] and ν i is the left eigenvector proof of Theorem 2.
Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY DURGAPUR. Downloaded on June 13,2024 at 14:56:11 UTC from IEEE Xplore. Restrictions apply.
 LIU et al.: DISTRIBUTED OPTIMAL TIE-LINE POWER FLOW CONTROL FOR MULTIPLE INTERCONNECTED AC MICROGRIDS 1879

REFERENCES [24] R. A. Jabr, “Adjustable robust OPF with renewable energy sources,” IEEE
Trans. Power Syst., vol. 28, no. 4, pp. 4742–4751, Nov. 2013.
[1] M. Munsell, “US microgrid capacity will exceed 1.8 GW by 2018,” [25] Z. Qu, Cooperative Control of Dynamical Systems, 1st ed. London, U.K.:
[Online]. Available: http://www.greentechmedia.com/articles/read/US- Springer-Verlag, 2009.
Microgrid-Capacity-Will-Exceed-1.8-GW-by-2018 [26] H. Xin, M. Zhang, J. Seuss, Z. Wang, and D. Gan, “A real-time power
[2] L. Meng et al., “Review on control of dc microgrids and multiple microgrid allocation algorithm and its communication optimization for geographi-
clusters,” IEEE Trans. Emerg. Sel. Topics Power Electron., vol. 5, no. 3, cally dispersed energy storage systems,” IEEE Trans. Power Syst., vol. 28,
pp. 928–948, Sep. 2017. no. 4, pp. 4732–4741, Aug. 2013.
[3] M. Marzband, N. Parhizi, M. Savaghebi, and J. M. Guerrero, “Distributed [27] J. Rocabert, A. Luna, F. Blaabjerg, and P. Rodriguez, “Control of power
smart decision-making for a multi-microgrid system based on a hierarchi- converters in AC microgrids,” IEEE Trans. Power Electron., vol. 27,
cal interactive architecture,” IEEE Trans. Energy Convers., vol. 31, no. 2, no. 11, pp. 4734–4749, Nov. 2012.
pp. 637–648, Jun. 2016. [28] A. C. Luo, Discretization and Implicit Mapping Dynamics. New York,
[4] D. Gregoratti and J. Matamoros, “Distributed energy trading: the multiple- NY, USA: Springer, 2015.
microgrid case,” IEEE Trans. Ind. Electron., vol. 62, no. 4, pp. 2551–2559, [29] P. Ferraro, E. Crisostomi, R. Shorten, and F. Milano, “Stochastic frequency
Apr. 2015. control of grid-connected microgrids,” IEEE Trans. Power Syst., vol. 33,
[5] Y. Li et al., “Optimal operation of multi-microgrids via cooperative energy no. 5, pp. 5704–5713, Sep. 2018.
and reserve scheduling,” IEEE Trans. Ind. Inform., vol. 14, no. 8, pp. 3459– [30] K. Purchala, L. Meeus, D. Van Dommelen, and R. Belmans, “Usefulness
3468, Aug. 2018. of DC power flow for active power flow analysis,” in Proc. Power Eng.
[6] Y. Liu et al., “Distributed robust energy management of a multi-microgrid Soc. Gen. Meeting, Jun. 12–15, 2005, pp. 454–459.
system in the real-time energy market,” IEEE Trans. Sustain. Energy, [31] K. Cai and H. Ishii, “Average consensus on general strongly connected
vol. 10, no. 1, pp. 396–406, Jan. 2019. digraphs,” Automatica, vol. 48, no. 11, pp. 2750–2761, Nov. 2012.
[7] D. Molzahn et al., “A survey of distributed optimization and control
algorithms for electric power systems,” IEEE Trans. Smart Grid, vol. 8,
no. 6, pp. 2941–2962, Nov. 2017.
[8] Z. Wang, W. Wu, and B. Zhang, “A fully distributed power dispatch
method for fast frequency recovery and minimal generation cost in au-
tonomous microgrids,” IEEE Trans. Smart Grid, vol. 7, no. 1, pp. 19–31,
Jan. 2016. Yun Liu (S’14–M’16) received the B.Eng. (with
[9] Y. Liu, Z. Qu, H. Xin, and D. Gan, “Distributed real-time optimal power first class honors) and Ph.D. degrees from the Col-
flow control in smart grid,” IEEE Trans. Power Syst., vol. 32, no. 5, lege of Electrical Engineering, Zhejiang University,
pp. 3403–3414, Sep. 2017. Hangzhou, China, in 2011 and 2016, respectively.
[10] W. Liu, W. Gu, J. Wang, W. Yu, and X. Xi, “Game theoretic non- From August 2016 to September 2017, he was with
cooperative distributed coordination control for multi-microgrids,” IEEE the Energy Research Institute, Nanyang Technolog-
Trans. Smart Grid, vol. 9, no. 6, pp. 6986–6997, Nov. 2018. ical University, Singapore, as a research fellow. He
[11] Z. Wang, F. Liu, S. H. Low, C. Zhao, and S. Mei, “Distributed frequency is currently an Assistant Professor with the College
control with operational constraints, part I: Per-node power balance,” of Mechatronics and Control Engineering, Shenzhen
IEEE Trans. Smart Grid, vol. 10, no. 1, pp. 40–52, Jan. 2019. University, Shenzhen, China. His research interests
[12] N. Li, C. Zhao, and L. Chen, “Connecting automatic generation control include power system stability analysis, photovoltaic
and economic dispatch from an optimization view,” IEEE Trans. Control generation, microgrid, distributed control/optimization, and cyber-physical
Netw. Syst., vol. 3, no. 3, pp. 254–264, Sep. 2016. security.
[13] M. Wooldridge, An Introduction to Multiagent Systems, 2nd ed. West
Sussex, U.K.: Wiley, 2009.
[14] J. Schiffer, R. Ortega, A. Astolfi, J. Raisch, and T. Sezi, “Conditions
for stability of droop-controlled inverter-based microgrids,” Automatica,
vol. 50, no. 10, pp. 2457–2469, Oct. 2014.
[15] S. D’Arco and J. A. Suul, “Equivalence of virtual synchronous machines
and frequency-droops for converter-based microgrids,” IEEE Trans. Smart
Grid, vol. 5, no. 1, pp. 394–395, Jan. 2014. Yuanzheng Li (M’18) received the M.S. degree from
[16] A. M. Bouzid, J. M. Guerrero, A. Cheriti, M. Bouhamida, P. Sicard, and M. the Huazhong University of Science and Technology,
Benghanem, “A survey on control of electric power distributed generation Wuhan, China, in 2011, and the Ph.D. degree in elec-
systems for microgrid applications,” Renewable Sustain. Energy Rev., trical engineering from the South China University
vol. 44, pp. 751–766, 2015. of Technology, Guangzhou, China, in 2015. His cur-
[17] A. Maknouninejad and Z. Qu, “Realizing unified microgrid voltage profile rent research interests include optimal power system
and loss minimization: A cooperative distributed optimization and control dispatch and decision making energy management
approach,” IEEE Trans. Smart Grid, vol. 5, no. 4, pp. 1621–1630, Jul. system of microgrids, energy internet, etc.
2014.
[18] S. Bolognani, R. Carli, G. Cavraro, and S. Zampieri, “Distributed reactive
power feedback control for voltage regulation and loss minimization,”
IEEE Trans. Autom. Control, vol. 60, no. 4, pp. 966–981, Oct. 2015.
[19] J. Lee, J. Guo, J. K. Choi, and M. Zukerman, “Distributed energy trading in
microgrids: A game-theoretic model and its equilibrium analysis,” IEEE
Trans. Ind. Electron., vol. 62, no. 6, pp. 3524–3533, Jun. 2015.
[20] S. Park, J. Lee, S. Bae, G. Hwang, and J. K. Choi, “Contribution-based
energy-trading mechanism in microgrids for future smart grid: A game
theoretic approach,” IEEE Trans. Ind. Electron., vol. 63, no. 7, pp. 4255– Huanhai Xin (M’14) was born in Jiangxi, China, in
4265, Jul. 2016. 1981. He received the Ph.D. degree from Zhejiang
[21] H. Wang, T. Huang, X. Liao, H. Abu-Rub, and G. Chen, “Reinforcement University, Hangzhou, China, in 2007. He is cur-
learning in energy trading game among smart microgrids,” IEEE Trans. rently a Professor with the Department of Electrical
Ind. Electron., vol. 63, no. 8, pp. 5109–5119, Aug. 2016. Engineering, Zhejiang University. He was a Postdoc-
[22] Y. Xu and Z. Li, “Distributed optimal resource management based on the toral Researcher with the Electrical Engineering and
consensus algorithm in a microgrid,” IEEE Trans. Ind. Electron., vol. 62, Computer Science Department, University of Central
no. 4, pp. 2584–2592, Apr. 2015. Florida, Orlando, FL, USA, from June 2009 to July
[23] I. U. Nutkani, P. C. Loh, P. Wang, and F. Blaabjerg, “Linear decentralized 2010. His research interests include power system
power sharing schemes for economic operation of AC microgrids,” IEEE stability analysis and renewable energy integration.
Trans. Ind. Electron., vol. 63, no. 1, pp. 225–234, Jan. 2016.

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY DURGAPUR. Downloaded on June 13,2024 at 14:56:11 UTC from IEEE Xplore. Restrictions apply.
 1880 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 34, NO. 3, MAY 2019

Hoay Beng Gooi (SM’95) received the B.S. degree Jianfei Pan (M’15) received the Ph.D. degree from
in electrical engineering from the National Taiwan Hong Kong Polytechnic University, Hong Kong,
University, Taipei City, Taiwan, in 1978, the M.S. de- in 2006. He is currently with the College of
gree in electrical engineering from the University of Mechatronics and Control Engineering, Shenzhen
New Brunswick, Fredericton, NB Canada, in 1980, University, Shenzhen, China. His main research in-
and the Ph.D. degree in electrical engineering from terests include design and analysis of motors and gen-
The Ohio State University, Columbus, OH, USA, in erators, distributed control, and energy management.
1983. From 1983 to 1985, he was an Assistant Profes-
sor with Lafayette College, Easton, PA, USA. From
1985 to 1991, he was a Senior Engineer with Empros
(now Siemens), Minneapolis, MN, USA, where he
was responsible for the design and testing coordination of domestic and inter-
national energy management system projects. In 1991, he joined the School
of Electrical and Electronic Engineering, Nanyang Technological University,
Singapore, as a Senior Lecturer, where he has been an Associate Professor since
1999. He was the Deputy Head of Power Engineering Division during 2008–
2014. He has been an Editor for the IEEE TRANSACTIONS ON POWER SYSTEMS
since 2016. His current research interests include microgrid energy management
systems, electricity markets, spinning reserve, energy efficiency, and renewable
energy sources.

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY DURGAPUR. Downloaded on June 13,2024 at 14:56:11 UTC from IEEE Xplore. Restrictions apply.