See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/337360412

Two-Level Distributed Voltage/Var Control of Aggregated PV Inverters in

Distribution Networks

Article in IEEE Transactions on Power Delivery · November 2019

DOI: 10.1109/TPWRD.2019.2955506

CITATIONS READS
55 323

5 authors, including:

Yu Wang Tianyang Zhao

Chongqing University Jinan University (Guangzhou, China)
95 PUBLICATIONS 1,998 CITATIONS 123 PUBLICATIONS 1,792 CITATIONS

SEE PROFILE SEE PROFILE

Chengquan Ju Yan Xu
Nanyang Technological University Nanyang Technological University
10 PUBLICATIONS 390 CITATIONS 433 PUBLICATIONS 14,752 CITATIONS

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

A*STAR Smart Grid Project (SERC Grant No.: 112 120 2022) View project

Electric Vehicle Charging Pricing and Demand Management View project

All content following this page was uploaded by Yu Wang on 20 November 2019.

The user has requested enhancement of the downloaded file.

 1

Two-Level Distributed Volt/Var Control using

Aggregated PV Inverters in Distribution Networks
Yu Wang, Member, IEEE, Tianyang Zhao, Member, IEEE, Chengquan Ju, Member, IEEE, Yan Xu,
Senior Member, IEEE, and Peng Wang, Fellow, IEEE

Abstract— The penetration level of photovoltaic (PV) keeps circumstances, the control systems for electric grids are
increasing in modern distribution networks, which leads to required to be more stable, resilient, flexible and secure [7].
various severe voltage limits violation problems. This paper aims Therefore, how to design an efficient voltage control scheme
to aggregate and utilize the PV inverters for voltage regulation by
with the large-scale user-end PV penetration becomes an
a fully distributed two-level Volt/VAr control (VVC) scheme. In
the lower-level VVC (real-time scale), the rooftop PV inverters emerging and vital question to be investigated.
are aggregated via consensus algorithms and then governed by Given all the voltage regulation resources, the voltage
droop controllers in medium-voltage networks. The droop regulation problem in distribution networks can be typically
controller adjusts the reactive power output of each PV divided into two stages, which aims to control the slow and
aggregator in real-time from its dispatched value depending on fast actuators separately [8], [9]. In the first stage, the voltage
the bus voltage variations. In the upper-level VVC (15-min
regulation is realized by mechanical devices such as on-load
timescale), the reactive power of PV aggregators is dispatched for
power loss minimization, where control signals are set as base tap changers, step voltage regulators, and switched capacitors,
values for PV aggregators. This problem is formulated as second- etc. [9]. In order to reduce the frequency of taps/switches
order cone programming and solved by the alternating direction change and increase their lifetime, these mechanical devices
method of multipliers. The simulation results demonstrate the are suggested to be manipulated in a longer time interval, e.g.,
effectiveness of the proposed method in both short-term and long- hourly or even daily. In the second stage, the power inverters
term scenarios with different system scales.1
have much fast response speed and flexible operation ranges,
Index Terms— Distribution networks, Volt/VAr control, PV
therefore their reactive power is suitable for faster voltage
inverters, consensus algorithms, alternating direction method of regulation. The mechanical devices are to manage the basic
multipliers. network voltage profile based on PV/load predictions. Then
the Volt/VAr control (VVC) of power inverters functions to
I. INTRODUCTION adjust the voltage profiles considering real-time PV/load
fluctuations. This paper focuses on the second-stage problem
A. Background and Motivation
by a novel VVC scheme for aggregated PV inverters.
I N recent years, the penetration level of photovoltaic (PV)
generation in power distribution networks keeps increasing
rapidly [1]. In Australia, 21.6% of all houses now have a PV
B. Related Works
According to the control architectures, VVC can be divided
system installed, where Queensland and South Australia have into three categories: i) decentralized, ii) centralized, and iii)
more than 30% of houses having installed a PV system [2]. In distributed [10]. In decentralized (or called local) control, the
Singapore, the PV penetration level is expected to increase to power inverters can response to local voltage measurement at
25% by the year 2025 [3]. The EcoCampus project at Nanyang the point of common coupling (PCC) rapidly for local VVC.
Technological University, Singapore demonstrates a practical In [11], a PCC voltage droop control based VVC with its
example of the innovative campus with high PV penetration stability issue in distribution networks is presented. In [12], a
[4]. However, high penetration of PV generation will lead to multi-functional local VVC scheme is proposed for voltage
various power quality issues, such as the widely recognized quality issues caused by PV inverters. The local control
voltage rise and voltage fluctuation problems [5], [6]. The schemes have the benefits of fast response speed, while it does
large-scale penetration of small capacity user-end PV systems not require communication infrastructures. However, the
is also challenging the control and operation framework of limitation of decentralized control is the lack of system-wide
distribution networks, where more user-end devices become coordination, which can only provide non-optimal solutions.
controllable via communication systems. In such By contrast, centralized control aims to provide an optimal
solution to VVC. In [9], a multi-time scale coordinated VVC
scheme (hourly and 15 minutes) is proposed for high
Y. Wang is with Rolls-Royce@NTU Corporate Lab, Nanyang renewable-penetrated distribution networks. In [13], based on
Technological University, Singapore, 639798.
T. Zhao and C. Ju are with Energy Research Institute @ Nanyang
robust optimization, a three-stage VVC scheme, which
Technological University, Singapore, 639798. includes mechanical devices scheduling, inverter output
Y. Xu and P. Wang are with School of Electrical and Electronics dispatch, and real-time droop control is proposed. However,
Engineering, Nanyang Technological University, Singapore, 639798.
the central controller suffers extensive computation and
(Corresponding author: Yan Xu, e-mail: [email protected])
communication burdens, especially when dealing with a large
 2

number of control units; besides, it is inherently vulnerable to control framework considering both upper-level optimizations
single-point communication failures; furthermore, the central among aggregators and lower-level cooperative control within
controller needs to gather all the system information, which aggregators. From literature, this kind ‘system of system’
may not be always available in power distribution networks. problem is investigated by some recent research of networked
In recent years, the distributed control has shown its merits microgrids [24], [25]. However, the fully distributed VCC
to overcome those limitations discussed above. The control scheme considering both inter and intra coordination for PV
and optimization objectives can be achieved in a distributed aggregators is of great importance but lack of investigation.
way with peer-to-peer communications, which have good plug- C. Contributions in This Paper
and-play capability and expendability [10]. In [8], a two-stage
This paper proposes a two-level distributed and timescale
distributed control architecture of distributed inverters is
coordinated VVC scheme for PV aggregators, aiming to
proposed for network VVC. The authors in [14] propose a
bridge the gap between the upper-level optimal reactive power
discrete-time consensus control of PV inverters to address
dispatch and lower-level real-time control. Compared to
voltage rise issues in low-voltage networks. In [15], a
existing works in literature, major contributions are as follows:
decentralized and distributed hybrid control scheme for PV
(i) The proposed VVC scheme is fully distributed and
inverters is proposed for both network voltage fluctuation and
includes two timescales coordinated control: i.e. upper-
violation issues. The distributed consensus algorithms have
level and lower-level VVC. The PV uncertainty is handled
also been used for the secondary voltage control of islanded
by two timescale coordination (15 minutes dispatch and
microgrids [16], [17]. It should be noted that the voltage
real-time adjustment), while the control system resilience
control problems in distribution networks and islanded
is enhanced by the fully distributed control structure.
microgrids have some different features: (i) In the hierarchical
(ii) In the lower-level VVC, a distributed control approach is
control framework of microgrids, the controllable DG units are
proposed to group each rooftop PV inverters at low-
operated as voltage sources with both P/f and Q/V droop
voltage (LV) side into aggregators. Then the aggregators
control [18]. However, in distribution networks, the PV units
are governed by the proposed droop controllers in
with maximum power point tracking control are operated as
medium-voltage (MV) networks. Note that most existing
current sources with reactive power related droop control [8],
works only consider the inverters as one single inverter
[14]. (ii) In islanded microgrids, the control objective is to
directly. In addition, the available reactive capacity of
restore all the bus voltages into a nominal value. In distribution
each aggregator is also estimated in a distributed manner.
networks with long feeders and heavy loads, the voltage
(iii) In the upper-level VVC, based on the network conditions
control objectives turn to mitigate the voltage deviations and
and available reactive power capacity, the aggregators are
reduce power loss.
dispatched in a distributed way for network loss
On top of the real-time control, reactive power should be
minimization. This problem is formulated as a SOCP and
optimally dispatched in the longer timescale. This problem can
solved by ADMM. These optimal dispatch signals are
be formulated as the optimal power flow (OPF) problem and
updated every 15 minutes and set as base values for
solved distributedly [19], [22], [23]. Typically, the branch
lower-level droop controls.
flow model (i.e. DistFlow) is used to depict power flow in
This paper is organized as follows. In Section II, the
radial distribution networks [20]. The original OPF based on
architecture of the proposed VVC is introduced. Next, the
DistFlow can be turned into a convexified OPF by relaxing the
lower-level VVC and upper-level VVC are presented in detail
nonlinear power loss term. Then the alternating direction
in Section III and IV, respectively. The case studies with IEEE
method of multipliers (ADMM) can be applied to provide a
test systems are conducted and the simulation results are
distributed and fast-converged solution of the convex
discussed in Section V. Finally, Section VI draws the
optimization problem [21]. Previous research has studied the
conclusion.
convexified OPF problem and its distributed solution, such as
[22], [23]. [19] provides a detailed review of formulation and
II. PROPOSED VVC ARCHITECTURE
solution methods for distributed OPF. In [22], by using the
DistFlow and linearizing the nonlinear terms, a convex A. System Overview
quadratic OPF is formulated to dispatch the reactive power In this paper, a residential distribution network with high
from inverters. In [23], the original OPF is relaxed into PV penetration is considered, as shown in Fig. 1. The rooftop
second-order cone programming (SOCP) problem and PV inverters at the LV side are fed into a point of common
exactness of the solution is discussed. coupling (PCC) and then connected to the MV network
In practice, the capacity of single PV inverters is not through an LV/MV transformer. In urban residual
enough to participate in the reactive power dispatch. Therefore, communities, it is reasonable to assume that the load demands
individual PV inverters should be assembled as aggregators to are evenly distributed in each phase, and the building rooftop
meet upstream network dispatch order. Existing distributed PVs are connected into the grid through three-phase AC/DC
VVC methods either focus on the distributed optimization of inverters [26]. Each group of PV inverters at the downstream
distribution networks [22], [23] or the cooperative control of of an MV/LV transformer is grouped into an aggregator via
PV inverters [14], [15]. This paper aims at a distributed lower-level communication networks. The data exchange of
 3

each bus in MV network is realized by an upper-level dispatched base value. Then, this combined control signal is
communication network. It motivates us to propose a VVC for further shared with each PV inverter within the aggregator in
the above scenario with only peer-to-peer communications. real-time. In addition, the reactive power constraint of each
The state-of-the-art communication infrastructure in residential aggregator is also estimated for the upper-level VVC. The set
distribution networks including both cable communications of time steps is represented as Ts = {0, Δts, 2Δts, …, Ts}.
and wireless technologies such as WiFi and Zigbee can enable
this application [27]. Based on the above electrical and
communication infrastructure, the research objective is to 15-min distributed dispatch among aggregators (Td )
design a two-level fully distributed VVC scheme for

Reactive power
aggregators of user-end PV inverters. td

constraints
td

Dispatch
power
ts ts … ts ts ts … ts
Upper-Level Time
Main Grid Bus-1 Bus-2 Communication Link Bus-N

MV Network
Real-time distributed control within aggregators (Ts )

Fig. 2. Timescale coordination of the proposed two-level VVC.

B = (N, E )

III. LOWER-LEVEL VVC

LV System with
PV Aggregators
PCC PCC PCC

… … In this section, the lower-level VCC is proposed for real-

… time operation and distributed reactive power sharing in each
Lower-Level
Communication
Link G1 = (M1 , D1 ) G 2 = (M 2 , D 2 ) G Nagg = (M Nagg , D Nagg )
PV aggregator. The real-time control takes effect every 0.1s in
this paper, and t T s .
Fig. 1. Overview of the studied system.

qkagg (t )
Next, notations to describe the studied system are denoted. Time- Time-
Dispatched
The radial MV distribution network is depicted by a tree graph qkagg varying
Base Value
varying
term term
qkd  qkres
B = (N, E ) , which is assumed to be a three-phase balanced
network. N={i |i=1, 2,…, N} represents the set of buses and qkd
E represents the set of distribution lines. The feed bus of the V V V V Vk (t )
distribution network is numbered as 0. It is noted that there is a qkd  qkres

set of Nagg={k |k=1, 2,…, Nagg} ( N agg  N ) PV aggregators qkagg

in the MW network. The upper-level communication network Fig. 3. The proposed droop control curve for PV aggregator.
has the same topology with the MV electrical network.
In aggregator k, the set of PV inverters are denoted as Mk, A. Real-time VVC by Droop Control
which is denoted by Mk = {j |j=1, 2,…,Mk}. The aggregation of The objective of the droop control is to mitigate the
PV inverters is achieved by lower-level communication network voltage violations when the system operation point
networks. The electrical network restraints are simplified as deviates far from the dispatched point. For example, during the
PV inverters are coupled by one PCC. The lower-level period of cloud passing, the PV generation may be
communication network of aggregator k is depicted by an dramatically influenced, then the droop control can function to
undirected graph Gk = (Mk, Dk) with a set of nodes Mk and a set mitigate the voltage deviations. The working principle of
droop control for real-time operation of PV aggregator is
of edges Dk. The nodes correspond to communication agents
shown in Fig. 3. Depending on the bus voltage in the MV
of PV inverters, and edges represent communication links for
network, the reactive power output of the aggregator k is
data exchange.
adjusted around the dispatch base value qkd in each dispatch
B. Architecture of Proposed VVC period. If Vk [V , V ] , the reactive power of the aggregator is
The proposed two-level VVC has two coordinated
kept at the dispatch base value qkd to minimize network loss. If
timescales as shown in Fig.2: 15-mins and real-time. In 15-
mins timescale, the objective is to minimize the network power Vk [V , V ] , a time-varying term has been activated to
loss within the operational voltage limits. The optimized mitigate network voltage deviations. The total reactive power
dispatch signals are set as the base values for PV aggregators, output qkagg of the aggregator k is represented as follows:
which are updated every 15 minutes. The set of time steps is
qkd  Gk (Vk (t )  V ), Vk (t )  V
represented as Td = {0, Δtd, 2Δtd, …, Td}. 
In the real-time scale, the network voltage fluctuations are qkagg (t )   qkd , V  Vk (t )  V (1)
more stochastic and should be mitigated by local droop  d
control. For the droop control of each PV aggregator, a time-  qk  Gk (Vk (t )  V ), Vk (t )  V
varying value based on each bus voltage will be added to the
 4

where Gk is the droop control gain. V and V are the lower and  Mk

upper bound to trigger the time-varying term of droop control.   k , j (t )= qk , j (t ) 



ˆj 1
ak , jjˆ ( k , ˆj (t )   k , j (t ))
Remark 1: The droop control gains are selected as  (5)
Gk   k Lkk , where Lkk = Qk Vk is the sensitivity of bus 
Mk

voltage magnitude to the reactive power injection. To ensure   k , j (t )=qk , j (t )   ak , jjˆ ( k , ˆj (t )   k , j (t ))

 ˆj 1
the stability of the droop-controlled distribution networks, in
general,  k =1 nbus are selected as a reasonable gain value [8]. where qk , j and qk , j are the upper and lower limits of
Besides, the voltage boundary [V , V ] is typically selected as available VAr capacity of inverter j,  k , j and  k , j are the
[0.95, 1.05] according to [6], [9]. estimated average values of upper and lower VAr limits.
It should be also noted that the available capacity for droop Finally, based on (3)-(5), the reactive power output of each
control is the total reactive capacity of the aggregator inverter j is calculated from:
subtracting the dispatched reactive power. Thus, a certain  k , j (t )
amount of capacity can be reserved during the dispatch period.  qk , j (t ), k , j (t )  0
  k , j (t )
By defining a critical voltage range as [V , V ] , the reactive qk , j (t )   (6)
power reserve of an aggregator qkres is calculated from:  k , j (t ) q (t ),  (t )  0
  (t ) k , j k, j
  k, j
qkagg , qkres  qkagg
 where qk , j is the reactive power output of inverter j in

qkres  Gk [max(V  V ,V  V )], qkagg  qkres  qkagg (2)
 aggregator k, which is within its operational limits as
 qkagg , qkres  qkagg qk , j (t )  [ qk , j (t ), qk , j (t )] .
where [V , V ] is selected as [0.94,1.06] in this paper; qkagg and Remark 2: For a set of initial values and connected
communication graph Gk, the steady-state value of qk,j
qkagg are the upper and lower reactive power limits of the converges to
aggregator k.  M k k ,0 (0)
B. Distributed Aggregation of PV Inverters  Mk qk , j (0), k , j (t )  0

The droop controller of an aggregator is associated with a  
qk , j (0)
pinning node (called the leader). The leader updates a  j 1
lim qk , j (t )   (7)
reference state k ,0 for the aggregator k (depicted by graph t 
 M k k ,0 (0) q (0),  (t )  0
Gk) based on the reactive power output from droop control in  Mk k, j k, j

(1) as follows:


q (0)
 j 1 k , j

qkagg (t )
k ,0 (t )= (3) For practical implementation in discrete-time, there exists a
Mk T0=kΔts, for some accuracy level ε1, such that
where k ,0 is the reference state, Mk is the number of PV  k , j (t )  k,0 (t )  1   k , j (t )   k , ˆj (t )  1
inverters in the aggregator k.
This reference state will be broadcasted to each PV inverter   k , j (t )   k , ˆj (t )  1 , t  T0 , j , ˆj  M k , t  T s
in the aggregator via communication graph Gk by the following (8)
leader-follower consensus control law: and the equilibrium in (7) is also achieved. The convergence
Mk analysis is given in Appendix-A.
k , j (t )= a
ˆj 1
k , jjˆ
( k , ˆj (t )  k , j (t ))  g k , j (k ,0 (t )  k , j (t )) (4) C. VAr Capacity of PV Aggregator
The available VAr capacity of each PV inverter j in
where ak , jjˆ represents the weight for information exchanged aggregator k is calculated from [9]:

, j (t )  ( sk , j )  pk , j (t )
qkmax PV 2 2
between inverter j and its adjacent inverter ĵ . ak , jjˆ =1 if (9)

inverters j and ĵ are connected by an edge, otherwise, ak , jjˆ = where skPV

, j and pk , j are the rated capacity and real power

0. g k , j represents the pinning gain, g k , j = 1 if inverter j output of a PV inverter, respectively.

Equivalently, (9) can be written as:
directly exchanges information from the leader, otherwise g k , j
, j (t )  sk , j 1   k , j (t )
qkmax PV
(10)
= 0 [28].
The average values of upper and lower available VAr where k , j (t )  pk , j (t ) skPV
, j is denoted as the PV generation
capacity for each inverter j (  k , j and  k , j ) of can be factor. It is noted that the available VAr capacity is
estimated from:
 5

represented as [ qk , j (t ), qk , j (t )] , which is only a subset of Qi   (Q  x l )  q

iˆ iˆ iˆ
L
i  qid (12c)
iˆC i
[qkmax
, j (t ), qkmax
, j (t )] considering the user’s local utilization.
Pi 2  Qi 2  vi li (12d)
The upper and lower limits of VAr capacity qkagg and qkagg
The reactive power constraint of each PV aggregator is:
in aggregator k are time-varying, which is equal to the sum of
qkagg  qkres  qkd  qkagg  qkres (13)
all PV inverters’ available VAr capacity. However, these
values cannot be directly calculated in a distributed way, as The voltage magnitude should be within the constraints:
each node only knows its neighbors’ information. As vi  vi  vi (14)
mentioned above, the leader can exchange information with The quadratic equality constraint (13d) can be relaxed to a
inverter j if gk,j = 1. Therefore, the upper and lower limits of second-order cone form, and the relaxation is also exact as
VAr capacity of aggregator k can be acquired as follows: discussed by [23]:
 Mk
Pi 2  Qi 2  vi li

(15)
 M k g k , j  k , j (t )
 agg Mk The upper-level VVC is to minimize the network loss
qk (t ) 


qk , j (t ) 
j 1
Mk
, subject to power flow equations and operational constraints.

j 1
gk , j The objective function is defined as:

 Mk
j 1
(11) 
fi ( xi )= ri li
iN

iN
(16)

 Mk
Mk 
g k , j  k , j (t ) Therefore, the network loss minimization problem can be
 agg
 qk (t )  
qk , j (t ) 
j 1
. formulated as:
rl
Mk


 j 1 min i i (17)
gk , j
 iN
 j 1
subject to: (12a)-(12c), (13)-(15)
Thus, the operational limits of an aggregator [qkagg , qkagg ] B. Distributed Solution by ADMM
are obtained from (11). This information is used to update the In this section, the ADMM algorithm is used to solve the
reactive power reserve in (2), and the distributed power relaxed optimization problem (17) in a distributed way [22],
dispatch in the next section. [29], [30]. The method iteratively minimizes the augmented
Lagrangian over three types of variables: i) the primary
IV. UPPER-LEVEL VVC variables; ii) the auxiliary variables to enforce boundary
In the upper-level VVC, the reactive power output from PV conditions among neighboring areas, and iii) the multipliers
aggregators is optimally dispatched in a longer timescale, i.e., for dualizing the relaxed problem. The Lagrangian is designed
15mins, and t  T d . The dispatch results will serve as the base to be separately relative to each type of variable so that we can
value settings in droop control for each aggregator. cyclically minimize each variable while fixing the others. This
A. Power Flow in MV Distributed Networks allows us to solve the problem distibutedly and asymptotically
converge to the same minimum costs obtained with centralized
The feed bus of the MV distribution network is numbered
solvers.
as 1. Each bus i  N \ {1} has a unique ancestor bus Ai, and a For the specific problem in (17), the primary variable of
set of children buses Ci. All the neighbor bus of bus i is node i is defined as xi  {vi , li , si , Si } .
represented by iˆ N . The line from bus Ai to bus i is labeled
i The auxiliary variable yiiˆ denotes variables of node
as line i, and i  E = 1, 2，
..., n . For each bus i N , vi | Vi | 2
iˆ received by node i, which is defined as:
represents the square of bus voltage. si  pi  jqi is the  ( y) ( y) ( y) ( y) ˆ
( Sii , lii , vii , sii ) i  i
complex power injection. For each line i E , the line  ( y) ( y) ˆ
yiiˆ   ( SiC , liC ), i  Ci (18)
impedance is represented by zi  ri  jxi . The complex power 
i i

( y) ˆ
flow from bus Ai to bus i is Si  Pi  jQi . The square of the  (viA ), i  Ai
 i

line current is li | I i | .
2 By substituting the original variables with the auxiliary
variables, the explicit distributed form of (17) can be
For a radial MV distribution network B, the power flow
represented as follows:
equations are represented as follows [22], [23]:
vAi  vi  2(ri Pi  xi Qi )  li (ri2  xi2 ), i  1 (12a)
min rl
iN
i i

Pi   (P  r l ) 
iˆ iˆ iˆ
piL  pid (12b)
over: {xi | i  N },{ yiiˆ | i  N , iˆ  N i }
iˆC i
s.t. vA  vi( y )  2(ri Pi( y )  xi Qi( y ) )  li( y ) (ri2  xi2 ), i  1. (19a)
( y)
i
 6

Pi( y )   (P iˆ
( y)
 riˆliˆ( y ) )  piL( y )  pid ( y ) (19b) variables yiimˆ 1 is calculated from (22). Finally, in S3:  -
iˆCi
update, iimˆ is updated by (23) with only local information.
Qi( y )   (Qiˆ( y )  xiˆliˆ( y ) )  qiL( y )  qid ( y ) (19c)
iˆCi
2 2
Pi( x )  Qi( x )  vi( x) li( x) (19d)

qkagg  qkres  (qkd )( x)  qkagg  qkres (19e)

vi  vi( x )  vi (19f)
xi( x)  yii(ˆy ) (19g) Fig. 4. Data exchange among adjacent nodes in the ADMM.

where the superscripts () ( x ) and () ( y ) on each variable denote Remark 3: The convergence of the upper-level distributed
whether the variable is updated in X-update or Y-update. The optimization depends on the selection of ρ and network scale.
set of constraints in X-update is represented by Si( x ) , and Y- As pointed in [29] and [30], the network diameter has a
stronger impact than the network size on the rate of
update by Si( y ) . convergence. With the low-bandwidth communication
The augmented Lagrangian function of the above problem infrastructures, this algorithm can be converged between two
for each node is represented as: dispatch intervals. Besides, the proposed method is initially

 (  ,x  y
2 designed for balanced MV networks. For the unbalanced MV
L ,i ( xi , y ˆ ,  ˆ )  fi ( xi )  iiˆ i iiˆ
 xi  yiiˆ )
ii ii
2 2 network, the proposed method could be modified based on the
iˆN i
methods in [22], [30]. The droop control and PV aggregation
(20) also need to be performed as per phase voltage conditions.
where iiˆ is the Lagrangian multiplier for (20). …
33-Bus Distribution Network
The optimization problem can be solved by the following 23 24 25
… …
steps based on ADMM:
[S1: X-update] Update primary variables by: 26 27 28 29 30 31 32 33

xim1  arg min( x ) L ,i ( xi , yiimˆ ,  mˆ ) (21)

xi Si ii 1 2 3 4 5 6 7 8 9 10 11 12 13 14

[S2: Y-update] Update auxiliary variables by:

yiimˆ 1  arg min L ,i ( xim 1 , y,  mˆ ) (22) … … …
( y)
ySi ii
19 20 21 22 18 17 16 15
[S3:  -update] Update the multiplier by:
iimˆ 1 =iimˆ + ( xim 1  yiimˆ 1 ) (23) … …

Two metrics are defined as follows to check the

convergence: Fig. 5. Diagram of the IEEE 33-bus radial distribution network.
1/2
 
iN

r m1   ( xim1  yim1 )2 

(24a) 0 (75)
1
(50)
2
(80)
3
(35)
4
(60)
5
(a) Graph of aggregator
1/2 with 10 PVs
 

10 9 8 7 6
s m1    ( yim1  yim )2  (24b) (80) (25) (30) (45) (65)
iN 
(32) (16) (70) (15) (30)
where r and s are primal and dual residuals. The iteration of 1 2 3 4 5
[S1]-[S3] continues until r and s converge to the satisfied
accuracy level ε2. (40) 10 9 8 7 6 (20) (b) Graph of
Fig. 4 illustrates the information exchange between adjacent 0 (15) (10) (14) aggregator with 20
(34) (10) (20) PVs
buses during this process. In S1: X-update, yiimˆ and iimˆ of the (43) 11 12 13 14 15 (30)

previous step from both children buses and ancestor buses are
20 19 18 17 16
received by bus i. The argument of the minimum regarding (40) (10) (20) (31) (45)
primary variables is calculated from (21). In S2: Y-update, the
Fig. 6. The communication graphs of the PV aggregators: (a) aggregator with
updated primary variable xim 1 is exchanged to its adjacent 10 PVs, (b) aggregator with 20 PVs.
buses. The argument of the minimum regarding auxiliary
 7

V. SIMULATION RESULTS algorithms is highly related to the weights and topology of

communication networks as well as the location of the leader.
A. Test System and Parameter Settings
The convergence of distributed aggregation control for PV
In this paper, two IEEE test distribution networks i.e. 33- aggregators with the communication graph in Fig. 6 (a) is
bus system and 69-bus system are used to test the proposed demonstrated in Fig. 8. The PV generation factor is considered
VVC method [31]. The single line diagram of the IEEE 33-bus as 0.8, therefore, the reactive capacity of each aggregator is
system is shown in Fig. 5. There are 8 PV aggregators located
at different buses. It should be noted that the placement and 550  1  0.82  330 kVA according to (10). The total
size of PV units and the inverters have an impact on results, reactive power increases from 0 to 165 kVA at the start of the
which is determined at the planning stage [32]. In this paper, simulation. The results demonstrate that the distributed control
two types of PV aggregators with the capacity of 550kVA are of PV aggregators can reach the steady-state within 100
considered: (a) aggregator with 10 PVs and (b) aggregator iterations, and the power-sharing is proportional to their
with 20 PVs. The communication graphs and the capacity of capacity given in Fig. 6 (a). As the lower-level VVC is a
each PV inverter (in kW) are shown in Fig. 6. In this test, the dynamic algorithm, the reference signals from droop control
substation voltage is 1.0 p.u., the operational voltage range can be continuously updated without waiting for convergence.
is [V , V ]  [0.95, 1.05] and the critical voltage range is In the upper-level dispatch, the convergence of ADMM is
related to the network scale and parameter of ρ. The primal
[V , V ]  [0.94, 1.06] . Besides, the single line diagram of the and dual residuals (r and s) of the upper-level optimization for
IEEE 69-bus system is shown in Fig. 7, where there are 15 PV the 33-bus system are shown in Fig. 9. The results show that
aggregators with the capacity of 550kVA spread in the system. with a properly chosen of ρ, the algorithm can reach a high
The simulation is conducted in Matlab with Gurobi solver. accuracy in hundreds of iterations. For the same signal
In the following subsections, the convergence for each exchange rate of 0.1s as above, the algorithm can surely find
control level of the proposed VVC is discussed first. Then the optimal solution within the 15-min dispatch interval of
both long timescale (24-hour) and short timescale (one-hour) upper-level control. In Table I, the iteration number respect to
simulations are carried out to validate the proposed VVC. Two different parameters ρ and two networks for primal and dual
case studies are conducted with different test systems (i) Case residuals to reach the accuracy of ε2<2e-3 are given.
A: IEEE 33-bus system. (ii) Case B: IEEE 69-bus system. Generally, a smaller ρ will lead to fast convergence, but the
69-Bus Distribution Network solution accuracy will be influenced if ρ is too small.

0.3   100   500   750   1000   1500

Primal Residual

… …
…
…
0.2

0.1
… … … … …
…
0
0 500 1000 1500 2000
Iteratives
… … … 0.3   100   500   750   1000   1500
Dual Residual

… …
0.2
Fig. 7. Diagram of the IEEE 69-bus radial distribution network.
30 0.1
Reactive Power (kVA)

PV No.

20 0 0
500 1000 1500 2000
Iteratives
10 Fig. 9. Primal and dual residuals of distributed optimization based on
ADMM.

0
Table I Iteration Number for Convergence of ADMM
0 50 100 150 200 Value of ρ 200 500 750 1000 1500
Iterations
Fig. 8. Convergence of distributed aggregation control for the aggregator
Iterations for 691 705 890 1238 >2000
with 10 PVs. ε2<2e-3 (33-
bus network)
B. Convergence Iterations for 786 795 962 1350 >2000
As an important concern of distributed algorithms, the ε2<2e-3 (69-
algorithm convergence is investigated in this section. bus network)
Generally, the convergence of the distributed consensus
 8

100 C. Case A: 33-bus Networks

Timescale of The 24-hour PV and load data with 15-min sampling rate
80 PV
PV/Load (%)

one-hour
simulation
are shown in Fig. 10 [33], [34]. The high-resolution PV and
60 load data from 10:00 to 11:00 with 1-second sampling rate is
40 Load shown in Fig. 11. These two timescales of PV and load data
are used for short-term and long-term simulations.
20 1) Case A-1: Long-Term Simulation
0 The results of 24-hour simulation in Case A-1 are shown in
0 4 8 12 16 20 24
Time (Hour) Figs. 12-14. The voltage profiles of the 33-bus distribution
Fig. 10. 24-hour PV and load profiles with 15-minute sampling rate. network with and without the proposed control scheme are
100
compared in Fig. 12. All the bus voltages are regulated within
[0.95, 1.05] by the upper-level dispatch. Correspondingly,
80
PV/Load (%)

each aggregator will provide the dispatched reactive power

PV
60 output, as shown in Fig. 13. During high PV generation
40 periods, a portion of aggregators in the distribution network
needs to absorb some reactive power for voltage rise
20 regulation. The network power loss, as the optimization
Load
0 objective of the upper-level, is minimized during the operation
10:00 10:15 10:30 10:45 11:00
Time (H:M) period. Fig. 14 compares the network power loss (i) with the
Fig. 11. One-hour PV and load profiles with 1-second sampling rate. proposed VVC (ii) without the proposed VVC and (iii) with
1.1 1.1 (b) only Q-V droop control. The 24-hour total network energy loss
(a)
(time integration of power loss) is 1.68 MWh, which is far less
1.05 1.05 than that without the proposed VVC of 3.36MWh. The total
Voltage (p.u.)
Voltage (p.u.)

network energy loss with only droop control is 2.65MWh.

1 1
2) Case A-2: Short-Term Simulation
0.95 0.95 The results of one-hour simulation in Case A-2 are
demonstrated in Figs. 15-18. As shown in Fig. 15, the
0.9 0.9 proposed VVC can effectively control the network voltages,
0 6 12 18 24 0 6 12 18 24 even though there are fluctuations between two dispatch points.
Time (Hour) Time (Hour)
Fig. 12. Voltage profiles of the 33-bus distribution network in Case A-1. (a) The droop control defined in (1) will take effect when the bus
with the proposed VVC; (b) without the proposed method. voltage is out of the normal operational range
800 Agg [V , V ]  [0.95, 1.05] . For each dispatch time instant (every 15
Reactive Power (kVar)

Bus No.
minutes), the optimal reactive power output of each aggregator
400 will be dispatched by the upper control level. If the voltage
cannot be maintained within [0.95, 1.05] , the droop control is
0
functioned to mitigate the PV fluctuations during one dispatch
Reactive power period. It can be observed that in Fig. 16 the base value of
absorption for voltage
rise regulation each aggregator is changed every 15 minutes. The reactive
-400
0 4 8 12 16 20 24 power is further shared by each PV inverter within one
Time (Hour) aggregator. As an example, the reactive power-sharing by each
Fig. 13. Reactive power dispatched to each aggregator in Case A-1. PV inverter of the PV aggregator at bus 17 (Aggregator-17) is
250
shown in Fig. 17. Each inverter in Aggregator-17 shares the
w/o proposed method
reactive power requirements from the proposed VVC
Network Loss (kW)

with proposed method

200 only droop method according to their capacity rating given in Fig. 6 (a).
150 In addition, to illustrate that the proposed control structure
is resilient with plug-and-play of PVs caused by either manual
100
operations or communication failures, a comparison study has
50 been made and shown as Fig. 18. In this condition, PV-10 is
0 plugged out from 10:13 to 10:19, while PV-7 is plugged out
0 4 8 12 16 20 24 from 10:32 to 10:38. It can be found that during the periods of
Time (Hour)
one PV plugged out, the rest of the PVs will increase their
Fig. 14. Network power loss (i) with the proposed VVC (ii) without the
proposed VVC and (iii) with only droop control in Case A-1. outputs to meet the total reactive requirement. It indicates the
outside characteristics of the PV aggregators can remain the
same even though there are PV plugged in and out operations.
 9

1.08 compared in Fig. 19. All the bus voltages are regulated within
[0.95, 1.05] by the upper-level dispatch. Each aggregator can
provide the dispatched reactive power output, as shown in Fig.
Voltage (p.u.)

1.04
20. Fig. 21 compares the network power loss (i) with the
proposed VVC (ii) without the proposed VVC and (iii) with
1 only Q-V droop control. The 24-hour total network energy loss
(areas of power loss) is 2.16 MWh, which is far less than that
0.96 without the proposed VVC of 3.1MWh. The total network
10:00 10:15 10:30 10:45 11:00
Time (H:M) energy loss of with droop control is 2.8MWh.
Fig. 15. Voltage profiles of the 33-bus distribution network in Case A-2. 1.1 1.1 (b)
(a)
400 Agg
Bus No.
Reactive Power (kVar)

1.05 1.05

Voltage (p.u.)
Voltage (p.u.)
200
1 1
0
0.95 0.95
-200
0.9 0.9
-400 0 6 12 18 24 0 6 12 18 24
10:00 10:15 10:30 10:45 11:00 Time (Hour) Time (Hour)
Time (H:M) Fig. 19. Voltage profiles of the 69-bus distribution network in Case B-1. (a)
Fig. 16. Reactive power injection of each aggregator in MV distribution with the proposed VVC; (b) without the proposed method.
networks in Case A-2. Agg
600 Bus No.

60 Reactive Power (kVar)

PV No. 400
Reactive Power (kVar)

40 200
0
20 -200
-400
0
-600
0 4 8 12 16 20 24
-20 Time (Hour)
10:00 10:15 10:30 10:45 11:00 Fig. 20. Reactive power dispatched to each aggregator in Case B-1.
Time (H:M) 300 w/o proposed method
Fig. 17. PV reactive power absorption in Aggregator-17 with 10 PVs in Case
A-2. 250 with proposed method
Network Loss (kW)

only droop method

60 Other PVs output 200

increase PV No.
Reactive Power (kVar)

150
40
100
50
20
0 0
4 8 12 16 20 24
0 Time (Hour)
Fig. 21. Network power loss (i) with the proposed VVC (ii) without the
PV-10 plugged out PV-7 plugged out
-20 proposed VVC and (iii) with only droop control in Case B-1.
10:00 10:15 10:30 10:45 11:00
Time (H:M) 2) Case B-2: Short-Term Simulation
Fig. 18. Plug-and-play operation of PVs in Case A-2. The results of one-hour simulation in Case B-2 are shown
D. Case B: 69-bus Networks in Figs. 22-24. Similar to Case A, as shown in Fig. 22, the
proposed VVC can effectively control the network voltage
In Case B, the proposed method is validated in a larger
variations. In Fig. 23, the base value of each aggregator is
network IEEE 69-bus system with 15 PV aggregators, as
changed for every 15 minutes, and the droop control takes
shown in Fig. 7. The 24-hour and one hour PV and load data
effect for the aggregators when there are bus voltage violations.
in Case A given by Figs. 9 and 10 are still utilized in this test
In addition, the larger PV aggregator with 20 PVs is further
case.
considered in this case. The capacity and communication
1) Case B-1: Long-Term Simulation graph of the studied PV aggregator are shown in Fig. 6 (b). As
The results of 24-hour simulation in Case B-1 are shown in shown in Fig. 24, each PV can provide reactive power
Figs. 19-21. The voltage profiles of the 69-bus distribution proportional to their capacity to meet the total reactive power
network with and without the proposed control scheme are requirement of Aggregator-27 in Fig. 23.
 10

In summary, the proposed VVC can handle both short-term shared by each PV inverter according to their capacity.
and long-term voltage regulation in a fully distributed manner. Besides, the convergence, flexibility and scalability issues are
All the reactive power reserves from PV inverters are also discussed. The proposed method provides a feasible
effectively utilized to handle the voltage problems bring along solution for fully distributed control and management of PV
with high PV penetration. The total network loss as an inverters in power distribution networks.
objective of the proposed VCC is minimized. The proposed
APPENDIX
method demonstrates a reliable and efficient way to manage
and control of PV inverters in future high PV penetrated A. Converge Analysis
distribution networks.
The distributed control law in (4) is based on leader-
1.08 follower consensus algorithms [28]. If the communication
1.06 graph Gk is connected, (4) can converge to a unique solution:
lim k , j (t )=k ,0 (0)
Voltage (p.u.)

(25)
1.04 t 

1.02
The distributed control law in (5) is based on dynamic
average consensus algorithms [35]. If the communication
1 graph Gk is connected, (5) can converge to the unique
0.98 solutions:
10:00 10:15 10:30 10:45 11:00 Mk

q
Time (H:M) 1
Fig. 22. Voltage profiles of the 69-bus distribution network in Case B-2. lim k , j (t )= k , j (0) (26)
t  Mk
Agg j 1
300 Bus No.
Mk
Reactive Power (kVar)

q
1
200 lim k , j (t )= k , j (0) (27)
t  Mk
100 j 1

0 According to the proposed control in (6), if k , j (t )  0 , it

-100 can be derived that
-200
-300
10:00
lim k , j (t ) k ,0 (0)
10:15 10:30 10:45 11:00 t 
lim qk , j (t )  qk , j (0)  qk , j (0) (28)
Time (H:M) t  lim  k , j (t ) Mk

q
Fig. 23. Reactive power injection of each aggregator in MV distribution t 
1
networks in Case B-2. k , j (0)
Mk j 1
60
Similarly, if k , j (t )  0 , it can be derived that
PV No.
Reactive Power (kVar)

40 lim k , j (t ) k ,0 (0)
t 
lim qk , j (t )  qk , j (0)  qk , j (0) (29)
t  lim  k , j (t ) Mk

q
20 t 
1
k , j (0)
Mk j 1
0
Therefore, combining (28) and (29), we can obtain the
-20 steady-state equilibrium in (7).
10:00 10:15 10:30 10:45 11:00
Time (H:M)
Fig. 24. PV reactive power absorption in Aggregator-27 with 20 PVs in Case REFERENCES
B-2. [1] Renewables 2018 Golbal Status Report. [Online]. Available:
http://www.ren21.net/status-of-renewables/global-status-report/.
VI. CONCLUSION [2] Solar Trends Report for Solar Citizens. [Online]. Available:
http://apvi.org.au/wp-content/uploads/2018/12/Solar-Trends-Report-
In this paper, a two-level distributed VVC scheme for high for-Solar-Citizens-FINAL_11-12-18_2_logos.pdf.
PV penetrated distribution networks is proposed. Compared [3] Eco-Business website. [Online]. https://www.eco-
with centralized approaches, the proposed VVC enhances the business.com/news/singapore-could-be-25-solar-powered-by-2025/.
[4] Ecocampus project. [Online]. Available: http://ecocampus.ntu.edu.sg/
system resilience as its distributed architecture. Various case Pages/index.aspx.
studies have been conducted to validate the effectiveness of [5] A. Samadi, et al. “Coordinated active power-dependent voltage
the proposed control scheme. The results demonstrate that the regulation in distribution grids with PV systems,” IEEE Transactions
on Power Delivery, vol. 29, no .3, pp. 1454-1464, Jan. 2014.
network voltage regulation is achieved in two timescales by [6] Y. Wang, et al. “Coordinated control of distributed energy-storage
the proposed two-level VVC. The network loss is reduced with systems for voltage regulation in distribution networks,” IEEE
the upper-level dispatch of aggregators within the voltage Transactions on Power Delivery, vol. 31, no .3, pp. 1132-1141, Jul.
2015.
constraints, while the PV fluctuations between the dispatch [7] N. Mahmud, and A. Zahedi, “Review of control strategies for voltage
intervals are handled by the droop control in the lower-level in regulation of the smart distribution network with high penetration of
real-time. The reactive power of one aggregator is further
 11

renewable distributed generation,” Renewable Sustainable Energy [28] H. Zhang, F. L. Lewis, and Z. Qu, “Lyapunov, adaptive, and optimal
Reviews, vol. 64, pp: 582-595, Oct. 2016. design techniques for cooperative systems on directed communication
[8] B. A. Robbins, et al. “A two-stage distributed architecture for voltage graphs,” IEEE Transactions on Industrial Electronics, vol. 59, no. 7,
control in power distribution systems,” IEEE Transactions on Power pp. 3026–3041, 2012.
Systems, vol. 28, no. 2, pp.1470-1482, May 2013. [29] Q. Peng and S. H. Low, “Distributed optimal power flow algorithm
[9] Y. Xu, Z. Y. Dong, R. Zhang, and D. J. Hill, “Multi-timescale for radial networks, I: Balanced single phase case,” IEEE
coordinated voltage/var control of high renewable-penetrated Transactions on Smart Grid, vol. 9, no. 1, pp.111-121, Apr. 2016.
distribution networks,” IEEE Transactions on Power Systems, vol. [30] Q. Peng and S. H. Low, “Distributed algorithm for optimal power
32, no. 6, pp. 4398-4408, Feb. 2017. flow on an unbalanced radial network,” 54th IEEE Conference on
[10] K. E. Antoniadou-Plytaria, et al. “Distributed and decentralized Decision and Control (CDC), 2015.
voltage control of smart distribution networks: models, methods, and [31] M. E. Baran and F. F. Wu, “Network reconfiguration in distribution
future research,” IEEE Transactions on Smart Grid, vol. 8, no .6, pp. systems for loss reduction and load balancing,” IEEE Transactions on
2999-3008, Nov. 2017. Power Delivery, vol.4, no. 2, pp. 1401–1407, Apr. 1989.
[11] P. Jahangiri and D. C. Aliprantis, “Distributed volt/var control by PV [32] Y. Zhang, Y. Xu, H.M. Yang, and Z.Y. Dong, “Voltage Regulation-
inverters,” IEEE Transactions on Power Systems, vol. 28, no. 3, pp. Oriented Co-Planning of Distributed Generators and Battery Storage
3429-3439, Aug. 2013. in Distribution Networks,” Int. J. Electrical Power and Energy
[12] M. J. E. Alam, K. M. Muttaqi, and D. Sutanto, “A multi-mode control Systems, vol. 105, pp. 79-88, Feb. 2019.
strategy for VAr support by solar PV inverters in distribution [33] 1-second resolution data, EPRI. [Online]. Available:
networks,” IEEE Transactions on Power Systems, vol. 30, no. 3, pp. http://dpv.epri.com/measurement_data.html.
1316-1326, May 2015. [34] J. A. Jardini, et al. “Daily load profiles for residential, commercial and
[13] C. Zhang, Y. Xu, Z. Dong, and J. Ravishankar “Three-Stage Robust industrial low voltage consumers,” IEEE Transactions on Power
Inverter-Based Voltage/Var Control for Distribution Networks with Delivery, vol. 15, no .1, pp. 375-380, 2000.
High-Level PV,” IEEE Transactions on Smart Grid, vol. 10, no. 1, [35] R. Olfati-Saber, J. A. Fax, and R. M. Murray, “Consensus and
pp. 782-793, Sep. 2017. Cooperation in Networked Multi-Agent Systems,” Proceedings of the
[14] G. Mokhtari, et al. “Smart robust resources control in LV network to IEEE, vol. 95, no. 1, pp. 215-233, Mar. 2007.
deal with voltage rise issue,” IEEE Transactions on Sustainable
Energy, vol. 4, no .4, pp. 1043-1050, Oct. 2013.
[15] Y. Wang, et al. “Decentralized-distributed hybrid voltage regulation Yu Wang (S’12-M’17) received the B.Eng.
of power distribution networks based on power inverters,” IET degree from Wuhan University, Wuhan China in
Generation, Transmission and Distribution, vol. 13, no. 3, pp. 444- 2011, and the M.Sc. and Ph.D. degree from
451, Nov. 2018. Nanyang Technological University, Singapore in
[16] A. Bidram, A. Davoudi, F. L. Lewis, and J. M. Guerrero, “Distributed 2012 and 2017, respectively. He is currently a
cooperative secondary control of MGs using feedback linearization,” research fellow in Nanyang Technological
IEEE Transactions on Power Systems, vol. 28, no. 3, pp. 3462-3470, University, Singapore. His research interests
Aug. 2013. include distributed control and optimization,
[17] F. Guo, et al. “Distributed secondary voltage and frequency energy storage systems, microgrids and smart
restoration control of droop-controlled inverter-based MGs,” IEEE grids.
Transactions on Industrial Electronics, vol. 62, no .7, pp. 4355-4364,
Jul. 2015.
[18] Y. Wang, et. al. “Cyber-Physical Design and Implementation of
Distributed Event-Triggered Secondary Control in Islanded Tianyang Zhao (S’14–M’18) received the
Microgrids,” IEEE Transactions on Industry Applications, vol. 55, B.Eng., M.Eng. and Ph.D. degrees in electrical
no. 6, pp. 5631-5642, Nov.-Dec. 2019. engineering from North China Electric Power
[19] D. K. Molzahn, et al. “A survey of distributed optimization and University, Beijing China in 2011, 2013 and
control algorithms for electric power systems,” IEEE Transactions on 2017, respectively. Currently, he is a Postdoctoral
Smart Grid, vol. 8, no .6, pp. 2941-2962, Nov. 2017. Research Fellow with Energy Research Institute,
[20] M. Baran and F. Wu, “Optimal capacitor placement on radial Nanyang Technological University, Singapore.
distribution systems,” IEEE Transactions on Power Delivery, vol. 4, His research interests include power system
no. 1, pp. 725–734, Jan. 1989. operation optimization, game theory and
[21] S. Boyd, N. Parikh, E. Chu, B. Peleato, and J.Eckstein, “Distributed reliability.
optimization and statistical learning via the alternating direction
method of multipliers,” Foundations and Trends® in Machine
learning, vol. 3, no. 1, pp. 1–122, 2011.
[22] B. A. Robbins, and A. D. Domínguez-García, “Optimal reactive Chengquan Ju received the B.Eng. degree in
power dispatch for voltage regulation in unbalanced distribution electrical engineering from Wuhan University,
systems,” IEEE Transactions on Power Systems, vol. 31, no .4, pp. Wuhan, China, in 2012 and the M.Sc degree in
2903-2913, Jul. 2016. power engineering from Nanyang Technological
[23] M. Farivar and S. H. Low, “Branch flow model: Relaxations and University (NTU), Singapore, in 2013. He is
convexification—Part I,” IEEE Transactions on Power Systems, vol. currently an optimization engineer in Envision
28, no. 3, pp. 2554–2564, Aug. 2013. Digital. His research interests include energy
[24] W. Zhang, and Y. Xu, “Distributed Optimal Control for Multiple management, optimal power flow, energy storage
Microgrids in a Distribution Network,” IEEE Transactions on Smart planning, hybrid energy system and hierarchical
Grid, in press, 2018. and distributed optimization.
[25] Y. Liu, et al. “Distributed Optimal Tie-Line Power Flow Control for
Multiple Interconnected AC Microgrids,” IEEE Transactions on
Power Systems, in press, 2018.
[26] Solar inverter solutions for building applications, ABB. [Online].
Available:https://library.e.abb.com/public/26cbca93f85d4ffc9652cc4c
d93c83c3/ABB_BCB00139%20Rev.C_Brochure_Solar%20Inverter
%20solutions%20for%20building%20applications_EN_2019.pdf.
[27] V. C. Gungor, et al. “Smart grid technologies: Communication
technologies and standards,” IEEE Transactions on Industrial
Informatics, vol. 7, no .4, pp. 529-539, Nov. 2011.
 12

Yan Xu (S’10-M’13) received the B.E. and

M.E degrees from South China University of
Technology, Guangzhou, China in 2008 and
2011, respectively, and the Ph.D. degree from
The University of Newcastle, Australia, in
2013. He is now the Nanyang Assistant
Professor at School of Electrical and
Electronic Engineering, Nanyang
Technological University (NTU), and a Cluster
Director at Energy Research Institute @ NTU
(ERI@N), Singapore. Previously, he held The University of Sydney
Postdoctoral Fellowship in Australia. His research interests include power
system stability and control, microgrid, and data-analytics for smart grid
applications. Dr Xu is an Editor for IEEE TRANSACTIONS ON SMART GRID,
CSEE Journal of Power and Energy Systems, and an Associate Editor for IET
Generation, Transmission & Distribution.

Peng Wang (M'00-SM'11-F'17) received his

B.Sc. degree from Xi'an Jiaotong University,
China, in 1978, the M. Sc. degree from Taiyuan
University of Technology, China, in 1987, and
the M. Sc. and Ph.D. degrees from the
University of Saskatchewan, Canada, in 1995
and 1998 respectively. Currently, he is a
professor of Nanyang Technological University,
Singapore.

View publication stats