Downloaded from orbit.dtu.

dk on: ,

Distributed Voltage Regulation of Smart Distribution Networks: Consensus-Based

Information Synchronization and Distributed Model Predictive Control Scheme

Guo, Yifei; Wu, Qiuwei; Gao, Houlei; Shen, Feifan

Published in:
International Journal of Electrical Power & Energy Systems

Link to article, DOI:

10.1016/j.ijepes.2019.03.059

Publication date:
2019

Document Version
Peer reviewed version

Link back to DTU Orbit

Citation (APA):
Guo, Y., Wu, Q., Gao, H., & Shen, F. (2019). Distributed Voltage Regulation of Smart Distribution Networks:
Consensus-Based Information Synchronization and Distributed Model Predictive Control Scheme. International
Journal of Electrical Power & Energy Systems, 111, 58-65. https://doi.org/10.1016/j.ijepes.2019.03.059

General rights
Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright
owners and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights.

 Users may download and print one copy of any publication from the public portal for the purpose of private study or research.
 You may not further distribute the material or use it for any profit-making activity or commercial gain
 You may freely distribute the URL identifying the publication in the public portal

If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately
and investigate your claim.
 Distributed Voltage Regulation of Smart Distribution Networks: Consensus-Based
Information Synchronization and Distributed Model Predictive Control Scheme

Yifei Guoa,b , Qiuwei Wub,c,∗, Houlei Gaoa , Feifan Shenb

a Key Laboratory of Power System Intelligent Dispatch and Control of Ministry of Education, Shandong University, China
b Centre for Electric Power and Energy, Department of Electrical Engineering, Technical University of Denmark, Lyngby, 2800, Denmark
c School of Engineering and Applied Sciences, Harvard University, 29 Oxford Street, Cambridge, MA 02138, USA

Abstract
This paper proposes a distributed voltage control (DVC) scheme for smart distribution networks with high penetration of
inverter-based distributed generators (DGs), aiming to optimally coordinate DG units and on load tap changer (OLTC)
transformer to regulate the voltages within the feasible range. The proposed scheme consists of two important parts:
1) distributed information synchronization (DIS) framework and 2) distributed model predictive control (DMPC)-based
voltage control scheme. The DIS framework is established based on the consensus protocols to synchronize the specific
information about the critical bus voltages and potential OLTC actions. The DMPC-based voltage control scheme is
presented, in which each DG unit only exchanges information with its immediate neighbors and solves the local optimal
control problem. Two control modes are designed to better deal with different operating conditions. In the normal mode,
only the reactive power outputs of DG units are optimized to mitigate the voltage deviations. In the corrective mode, both
of the active and reactive power outputs of DG units are optimally controlled to correct the severe voltage deviations. To
mitigate the mutual interaction between the DGs and OLTC, the potential actions of OLTC are predicted and considered
in the optimization problem of each units. The control performance of the proposed scheme was demonstrated using a
real medium-voltage (MV) distribution network with two feeders under both normal and large-disturbance conditions.
Keywords: consensus protocol, distributed generator (DG), distributed model predictive control (MPC), smart
distribution network, voltage control.

1. Introduction voltage regulators, without considering the flexible and fast

voltage regulation capability of DGs (normally operates with
The integration of distributed generators (DGs) into unity power factor), which cannot well control the voltage
power systems is worldwide encouraged by national policies due to the slow response and discrete actions. Therefore,
to meet the rapidly increasing energy demand and alleviate the voltage regulation of smart distribution networks has a
the environmental problems [1]–[2]. Among the distributed prominent position in the research on smart grids, motivat-
energy resources available today, the inverter-based DGs ing a vast number of research work to tackle this problem [5].
such as solar photovoltaic and wind turbines are the most The existing methods can be grouped into three categories,
common form in medium-voltage (MV) and low-voltage dis- summarized as follows:
tribution networks [3]–[4]. • Local Control. In the local voltage control schemes,
However, the increasing penetration of DG is challeng- the controllers receive the local information from the
ing secure and stable operation of smart distribution net- sensors surrounding them and perform control actions
works. The voltage management problem is one of the on their respective locality based on specific rules
main obstacles against installation of large amounts of DGs. such as droop-based strategies [6]–[8], fuzzy logic-based
The surplus power injections generated by DGs result in methods [9] and sensitivity-based methods [10]. The
reverse power flow and hence voltage rise issue. More- local methods are easy to be implemented and do not
over, the stochastic and intermittent nature of renewable require significant investments of communication in-
resources may cause significant voltage fluctuations. Gen- frastructures compared with the communication-based
erally, the distribution network operators regulate voltage ones. However, such a non-coordinated manner may re-
profile across the network within an appropriate range us- sult in negative interaction/hunting among the voltage
ing the conventional voltage regulation devices including on regulation devices, posing the possibility of undesired
load tap changers (OLTCs), shunt capacitor banks and step islanding [11].
• Centralized Control. Due to the advancement in in-
∗ Corresponding author
formation and communication technologies, the central-
Email address: [email protected],[email protected] ized control schemes are attractive, which utilize the
(Qiuwei Wu) global information of the networks to achieve the coor-

Preprint submitted to Journal of LATEX Templates March 3, 2019

 dination of multiple voltage regulation devices. Most of was presented in [38] considering transmission and distribu-
the centralized methods were designed by formulating tion grids as a whole. And then in [39], a distributed voltage
the voltage control problems as the optimal power flow stability assessment method considering DGs is developed
(OPF) models which minimize certain network-wide op- based on the distributed continuation power flow algorithm.
eration objectives including voltage deviations [12], [14], In this paper, a distributed voltage control (DVC) scheme
power losses [13], [14], number of OLTC actions [14], for smart distribution networks with high penetration of
and curtailed energy [13], [17], etc. Moreover, the cen- inverter-based DGs is proposed to better regulate the volt-
tralized model predictive control (MPC)-based voltage age across the network. The proposed scheme includes two
control schemes were proposed in [17, 18, 19], which can important parts: 1) distributed information synchronization
effectively coordinate multiple DGs and the OLTC by (DIS) framework and 2) distributed MPC (DMPC)-based
introducing the prediction mechanism and hence shows voltage control scheme. Compared with the existing works,
good control performance. In [20], a combined central- the main contributions of this paper are three-fold:
ized and local control scheme was proposed where the • A consensus-based DIS framework is proposed in this
centralized controller designed based on the MPC mod- paper to synchronize the information including mea-
ifies the V-Q characteristics of the local control. In [21], sured critical bus voltages and expected OLTC actions.
a two-tier voltage control scheme was proposed for dis- • The concept of DMPC is firstly used in voltage regula-
tribution networks with electric springs (ESs) by com- tion of smart distribution networks with DGs, in which
bining the centralized and distributed control where the each DG unit only exchanges information with imme-
upper-level control is developed based on the MPC to diate neighboring DG units. Based on multi-step opti-
regulate the OLTC and CBs. mization, the proposed controller can smoothen system
• Distributed Control. The distributed voltage control dynamics from the current state to the targeted state.
schemes are widely developed based on the distributed • The coordination between the OLTC with the dis-
control or distributed optimization techniques [5], [22]. tributed voltage control is addressed in this paper. The
In [21], the lower-level control was designed to achieve expected OLTC actions will be considered in the MPC
responsibility sharing of ESs in a distributed manner formulation of each DG to avoid the mutual interaction
using the consensus protocol. In [23], a two-stage dis- with the OLTC.
tributed voltage control scheme was proposed, of which The rest of this paper is organized as follows. Section
the first stage is the local control of each DG based 2 gives a brief overview of the proposed DVC. Section 3
on the sensitivity analysis and the second stage is de- presents the DIS framework. Section 4 presents the dis-
veloped to acquire reactive power support from other tributed MPC formulation of voltage control problem. Sec-
DGs. In [24], a consensus-based cooperative control tion 5 gives the coordination mechanism with the OLTC.
scheme is proposed to regulate the voltages by coor- Simulation results are presented and discussed in Section 6
dinating the plug-in electric vehicles (PEVs) and ac- followed by conclusions.
tive power curtailment of PV arrays. In [25], a dis-
tributed scheme that adjusts the reactive and active
power output of PV inverters was proposed to prevent 2. Overview of the Proposed Distributed Voltage
the over-voltage issues. In [26], a distributed coordi- Control Scheme
nated control of energy storage systems was proposed
for voltage regulation in distribution networks. In [27]– The structure of the proposed DVC scheme is shown in
[28], the distributed voltage control schemes were es- Fig. 1. The scheme is divided into two parts: 1) the DIS
tablished based on the multi-agent techniques with two- framework and 2) the DMPC-based voltage control scheme.
way communications. In [29]–[32], the optimal voltage In the DIS framework, the operating information is synchro-
control problems were formulated as an OPF model and nized (estimated by each DG) using the consensus protocols,
solved in a distributed manner using the alternating di- including the critical bus voltages and accumulated time af-
rection method of multipliers algorithms. In [33], an ter the triggered signal of OLTC. Moreover, the information
optimal voltage control scheme was proposed based on about reactive power outputs of neighboring DG units (in-
the subgradient method. In [34], a consensus-based dis- clude measured and predicted reactive power outputs) can
tributed intelligence algorithm was proposed to achieve be acquired by each DG as well. These necessary informa-
the near-optimal loss minimization performance. The tion will be used in the MPC. Note that, in Fig.1, j ∈ Ni
distributed control is more suitable for future networks denotes there is communication between Unit i and Unit
with a large number of DGs due to its potential advan- j. The DIS framework design is described in Section III
tages in terms of computation burden, communication, in detail. In the second part, the DMPC controller formu-
response speed, plug-and-play capability, etc [5, 35, 36]. lates the voltage control problem as a multi-step constrained
The mutual interaction between distribution and trans- optimization model and then solves it to obtain the power
mission grids for voltage stability problems is also addressed. references of DGs.
A case study to illustrate the interaction between transmis- The voltage sensitivities with respect to power injections
sion and active distribution networks was reported in [37]. A ∂V /∂P , ∂V /∂Q and ∆V /∆Ntap tap changes are related to
master-slave-splitting-based distributed global OPF method the operating point and network parameters. However, the

2
Figure 1: Structure of the proposed distributed voltage control scheme. Figure 2: Distributed information synchronization/estimation frame-
work (x and y denote the information which is required to be synchro-
nized; and x̂(i) and ŷ (i) denote the estimated values at Unit i).

global information could not be available for distributed con-

trollers. According to the numerical analysis, the sensitivi-
3.1.2. Consensus protocols on graphs
ties vary within very limited range during the steady state
operation. Therefore, the voltage sensitivities are calculated Let xi ∈ R denotes the value of node vi (agent
offline based on an analytical sensitivity calculation method i). Let M : RN → R be a function of N variables
[40]. It is expected that the closed-loop nature of MPC will x1 , x2 ..., xN and x1 (0), x2 (0)..., xN (0) denote the initial state
compensate the minor errors of sensitivity coefficients. of the system. The M-consensus problem is to calculate
M (x1 (0), x2 (0)..., xN (0)) in a distributed manner by ap-
Two control modes, namely normal mode and corrective
plying the inputs (state feedback) ui , of which the discrete
mode, are designed in this scheme which are determined by
model of the dynamics of each node vi in the consensus pro-
the voltage conditions. In the normal mode, the DG units
tocols can be described as [45],
operate at the maximum power point tracking (MPPT)
mode to capture more energy. The MPC controller only
xi (k + 1) = f (xi (k), ui (k)) , i∈I (1a)
optimally controls the reactive power outputs of DG invert-


ers to correct voltage deviations. In the corrective mode, in ui (k) = g xi (k), xj1 (k), xj2 (k), ..., xjni (k) (1b)
addition to reactive power support, necessary active power
curtailment of DG units is also used to correct the severe where the indexes j1 , j2 , ..., jni ∈ Ni ; f (·) and g(·) denote
voltage deviations. The details can be found in Section 4. the rules to solve different consensus problems. The pro-
In this paper, the proposed distributed control scheme is tocols solve the M-consensus problems if and only if there
only investigated from the technical perspective. However, exists a asymptotically stable equilibrium x∗ , which satisfies
the DGs are generally owned by customers instead of the x1 (∞) = x2 (∞) = · · · = xN (∞) = x∗ .
DNOs. Therefore, the economic compensation issues should In this paper, xi represents the voltage and accumulated
be addressed for a practical implementation since additional triggered time of the OLTC, which will be synchronized
reactive power injections can result in power loss inside the using different consensus protocols (average-consensus and
inverters and may accelerate the degradation of the invert- max-consensus, respectively).
ers. Economic incentives via a marketing mechanism should
be clarified to encourage the DG owners to provide voltage 3.1.3. Leaders and followers
support service for distribution networks [41]-[44].
The agents (DG units) are divided into two categories in
the protocols. The DG units having direct access to the ref-
3. Consensus-Based Distributed Information Syn- erence information are defined as leaders, otherwise, defined
chronization as followers. To save the communication investment (nearby
principle), it is worth noting that being leader or follower for
3.1. Preliminaries one unit is not rigid, i.e., one unit may have different roles in
3.1.1. Graph theory different information synchronization problems. The leader
Let G = (V, E, A) be a weighted undirected graph with and follower sets are denoted by L(·) and F (·) , respectively.
the vertex set V = {v1 , v2 , ..., vN }, edge set E ⊆ V × V and The schematic diagram of the distributed information
adjacency matrix A = [aij ]. The vertex indexes belong to a synchronization framework is shown in Fig. 2. In this figure,
finite set I = {1, 2, ..., N }. An edge of G is denoted by eij = Unit 1 and Unit 4 are the leaders in terms of synchronization
(vi , vj ). The adjacency elements associated with the edges of information x and y, respectively.
of the graph are positive, i.e., eij ∈ E ⇔ aij > 0 and assume
aii = 0 . The set of neighbors of node vi is denoted by Ni = 3.2. Synchronization of Monitored Bus Voltages
{j ∈ V|(vi , vj ) ∈ E}. The Laplacian matrix of the graph is
defined as L = D − A where P D = diag(d1 , d2 , · · · , dN ) is the The monitored bus voltages sent to the neighboring units
diagonal matrix with di = j∈Ni aij denoting the degree that are regarded as the leaders. The voltages of monitored
matrix. buses are synchronized by using the average-consensus pro-

3
tocol, of which the discrete model is, 4.1. Modeling

VC , ∀i ∈ LV , 4.1.1. DG
For the inverter-based DG, its control system typically


 N  
(i) (i) P (j) (i)
V̂C (k + 1) = V̂C (k) + µi aij V̂C (k) − V̂C (k) , consists two cascading control loops, i.e., the outer voltage
j=1
loop and inner current loop, resulting in fast and continuous


∀i ∈ F V ,

output power regulation capability. The fast dynamics can
(2) be neglected, i.e.,
(i) (j)
where V̂C and V̂C denote the monitored bus voltage esti- ref
PDG = PDG and QDG = Qref
DG (4)
mated by Unit i and Unit j, respectively; VC is the measured
voltage magnitude; µVi > 0 denote the constant gain. The since the network-level control is generally designed to be
gains aij depend on the information that Unit i can receive. decoupled with the DG’s own control system in terms of
Since the process (2) can converge to VC , each DG can esti- time scale so as to avoid the instability issue. The power
mate the monitored bus voltages which will be used for the outputs of DG units are limited by,
DMPC. 0 ≤ PDG ≤ P DG (5a)
q
3.3. Synchronization of Triggered Information of the OLTC 2 + Q2
PDG DG ≤ SDG (5b)
Similarly, the triggered information of the OLTC is sent
to the neighboring DG units. Here, to realize the distributed ∆QDG ≤ ∆QDG ≤ ∆QDG (5c)
information synchronization based on the consensus proto- where P̄DG is the maximum available power of DG, SDG is
col, the accumulated time interval after the triggered sig- the rated capacity of the inverter, and ∆Q̄DG is its maximum
nal (stops after the tap operation is finished) ∆ttri is syn- reactive power ramping limit.
chronized rather than the triggered time using the max-
consensus protocol. Each unit updates the information ac- 4.1.2. Network Voltage
cording to, In the MPC, the network voltage is predicted based on
∆ttri , ∀i ∈ LOLTC , the first-order approximation model which is obtained by

(i)
∆t̂tri (k + 1) = (j) linearizing the network model around the operating point
max{∆t̂tri (k)}, ∀i ∈ F OLTC , j ∈ Ni
(3) using the sensitivity coefficients, i.e.,

(i)
∂V
where ∆t̂tri is the accumulated time estimated by Unit i. V = V (0) + ∆u (6)
∂u
(i)
It is worth noting that ∆t̂tri will be reset to “0” after each where u and ∆u denote the control variables including the
control point. power injections of DGs and tap changes and its increment,
Note that, in the DIS framework, the information ex- respectively. This linearized model has been widely used
change among agents is performed all the time with much in the MPC. The MPC problem by formulated as a stan-
faster update rate than the DMPC framework (20 ms vs. dard convex constrained QP problem and can be efficiently
500 ms in this case). This decouples the DIS with the DMPC solved. It is expected that the minor errors of the linearized
in terms of time scale to avoid the instability caused by the model can be compensated by closed-loop nature of MPC.
interaction between the DIS and DMPC. The fast conver-
gence of the DIS can be guaranteed by selecting large gains. 4.2. DMPC Formulation
There is a possibility that the DIS does not perfectly con- The communication-based DMPC is adopted in this pa-
verge when the DMPC is activated. It is expected that the per. At each control step, each unit receives information
closed-loop nature of MPC can compensate the inaccuracy. about the control commands executed by their neighboring
units in the previous time step and other necessary updated
4. Distributed Model Predictive Control Scheme state information. Then, an optimization problem only with
the local cost function is solved where the interaction among
MPC has been successfully used in the voltage control
the neighboring units is considered. If the communication-
which offers several appealing features such as handling mul-
based iterations converge, a Nash equilibrium is achieved
tivariable control problems, ease of tuning and explicit con-
[46]–[47].
sideration of constraints [5]. In the MPC, the control com-
Suppose the prediction and control horizon (steps) are
mands are obtained by solving a discrete-time optimal con-
Hp (Np = Hp /Tc ) and Hc (Nc = Hc /Tc ), respectively. As
trol problem over a given horizon, which is formulated based
known, Np ≥ Nc . From the computational viewpoint, they
on the real-time measurements. An optimal control com-
should be equal unless the controller is required to consider
mand sequence is produced and only the first control in the
changes beyond the control horizon, i.e. Np = Nc [17].
sequence is applied. The MPC can smoothen the system dy-
Firstly, to clearly present the DMPC formulation, the fol-
namics from the current state to the targeted state due to
lowing definitions are provided. Define the measurements of
the multi-step optimization. Besides, the closed-loop nature
active and reactive power of DG units at step k as,
of MPC can effectively account for the model inaccuracy,
T
failure or delays of the control actions [17]. P (k) := [PDG,1 (k), ..., PDG,NDG (k)] ,

4
 T
Q (k) := [QDG,1 (k), ..., QDG,NDG (k)] . constant gain ε is used to adjust the control performance and
guarantee the closed-loop stability. denotes the element-
Define the predicted trajectory vectors of active power, reac- wise multiplication, Ai denotes the ith column of matrix A
tive power and reactive power utilization ratio of DG units and the corresponding term represents the effect of imme-
at step k + l (l = 1, ..., Np ) estimated at step k, as, diate neighbors which can exchange information with DG
T
unit i.
P̂ (k + l|k) := [PDG,1 (k + l|k), ..., PDG,NDG (k + l|k)] ,
T
Q̂ (k + l|k) := [QDG,1 (k + l|k), ..., QDG,NDG (k + l|k)] .
4.2.2. Corrective Mode
Define the vectors of predicted monitored bus voltages and If any monitored bus voltage violates the predefined limit,
voltage sensitivity of monitored bus i ∈ BC where BC de- the MPC controller will switch to the corrective mode. In
notes the set of monitored buses, with respect to active and this mode, both the active and reactive power outputs of DG
reactive power injections by, units are optimized to correct the severe voltage deviations
 T while minimizing the curtailed power. Thus, the control
∂VC,i ∂VC,i
SP
i := , ..., , i ∈ BC problem can be formulated as,
∂PDG,1 ∂PDG,NDG
 T Np
∂VC,i ∂VC,i 2
SQ
X  
:= , ..., , i ∈ BC . (i)
X
i
∂QDG,1 ∂QDG,NDG J2 = W V ∆V̂C,j (k + l|k) + ε V̂C,j − Vref
l=1 j∈BC

4.2.1. Normal Mode Np

X
2
If all monitored bus voltages are within the predefined +WP PDG,i (k + l|k) − P DG,i (9)
limits [Vref − Vth , Vref + Vth ], the controller will be in the l=1
normal mode, where Vth denotes the threshold which is typ-
ically VDB /2 (the predefined deadband of the OLTC, see
Section V) for the MV bus and [0.05, 0.08] p.u. for others. minimize J2 (10a)
Vref denotes the voltage reference which is typically set as PDG,i (k),...,PDG,i (k+Np −1),QDG,i (k),...,QDG,i (k+Np −1)
1.0 p.u.. Based on this design, the OLTC will not participate subject to
in the voltage control in this mode. ∂VC,j
In this mode, the MPC controller aims to regulate the ∆V̂C,j (k + l|k) = (PDG,i (k + l|k) − PDG,i (k))
∂PDG,i
voltage within a predefined range while keeping the fair re-
active power sharing (identical reactive power utilization). ∂VC,j
+ (QDG,i (k + l|k) − QDG,i (k))
Suppose the current time step is k (at the control point), ∂QDG,i

T   
the optimal control problem of DG i can be formulated as, + SP P̂(k + l|k − 1) − P(k) Ai
j
Np 2  T   
+ SQ
X  
(i) Q̂(k + l|k − 1) − Q(k) A
X
J1 = WV ∆V̂C,j (k + l|k) + ε V̂C,j − Vref j i

l=1 j∈BC ∂VC,j

Np + Signtap (k + l) · · ∆Ntap (10b)
X ∂Ntap
+WQ (QDG,i (k + l|k) − QDG,i (k + l − 1|k)) (7)
0 ≤ PDG,i (k + l|k) ≤ P DG,i (10c)
l=1
QDG,i ≤ QDG,i (k + l|k) ≤ QDG,i (10d)

minimize J1 (8a) ∆QDG,i ≤ ∆QDG,i (k + l|k) ≤ ∆QDG,i (10e)

QDG,i (k),...,QDG,i (k+Np −1)

subject to where the first term in the cost function is used to penalize
∂VC,j the voltage deviations and the second term is to minimize
∆V̂C,j (k + l|k) = (QDG,i (k + l|k) − QDG,i (k)) the curtailed energy of DGs where WP denotes the weighting
∂QDG,i
 T    factor for active power curtailment. Signtap denotes if there
+ SQ j Q̂(k + l|k − 1) − Q(k) A i (8b) is a potential tap change within the step k +l. When solving
the problem, the reactive power limit is firstly considered as
QDG,i ≤ QDG,i (k + l|k) ≤ QDG,i (8c) the rated capacity of the inverter. If the solution exceed
∆QDG,i ≤ ∆QDG,i (k + l|k) ≤ ∆QDG,i (8d) the rated capacity limit, the closest point within the limit
is selected as the solution. The potential OLTC action is
where the first term in the cost function J1 is used to mit- considered in this mode which is predicted based on the
igate the voltage deviations, the second term is used to method presented in Section V.
achieve fair reactive power sharing among DGs and the third The presented MPC problem can be transformed into a
term is to smoothen the reactive power variations. WV , Wα standard quadratic programming problem and can be effi-
and WQ denote the corresponding weighting factors. The ciently solved in milliseconds by the commercial solvers.

5
 Signtap (k) can be obtained by,

1, if t0 < t̂act ≤ t0 + k × Tc
Signtap (k) = (11)
0, otherwise.

6. Case Study
In this section, the control performance of the proposed
DVC is demonstrated by using a real Finnish distribution
network consisting of two 20 kV feeders [13]. The network
topology as well as the communication network topology is
Figure 3: The principle of OLTC operation. presented in Fig. 4 and the corresponding adjacent matrix
is,
 
0 1 0 0 1 0 0 0 0 0
5. Coordination With OLTC  1 0 1 0 0 0 0 0 0 0 
 
 0 1 0 1 0 0 0 0 0 0 
 
 0 0 1 0 0 0 0 0 0 1 
 
The OLTC is an efficient voltage control device which can  1 0 0 0 0 1 0 0 0 0 
directly change the voltage level of the whole distribution A =  0 0 0 0 1 0 1 0 0 0 .
  (12)
 
network. The OLTC is controlled in a local manner instead  0 0 0 0 0 1 0 1 0 0 
 
of being optimized together with DGs. The reasons are as  0 0 0 0 0 0 1 0 1 0 
 
follows:  0 0 0 0 0 0 0 1 0 1 
• The DMPC scheme is designed in a much faster time 0 0 0 1 0 0 0 0 1 0
scale than the mechanical time delay of OLTC actions,
This means if aij = 1, then Unit i and Unit j can exchange
which is generally in several seconds. The OLTC ac-
information using the two-way communication. As shown in
tions cannot be finished during one control period.
Fig. 4, 10×1 MW inverter-based DG units (DG 01∼DG 10)
• In this scheme, it just needs time information sent from
are placed at Bus 05, 07, 15, 20, 24, 31, 36, 39, 41 and 45,
the OLTC controller rather than changing the existing
respectively. Each DG is equipped with a smart agent which
control structure of OLTC, implying less extra invest-
can send/receive information and solve the control problems.
ment.
In order to guarantee the robustness of the DIS, the accu-
• It could avoid more computation complexity due to the
mulated time of the OLTC is sent to DG 01 and DG 05,
introduction of discrete control variables, which can re-
respectively and the measurements of monitored bus volt-
sult in a mix-integer nonlinear programming problem.
ages are sent to the closest two DG units using the one-way
The principle of OLTC operation is illustrated in Fig. communications. According to [48], the minimum informa-
3. The OLTC will perform a tap change if the controlled tion update interval in IEEE 802.11 (Wi-Fi) is of the order
bus voltage violates the predefined deadband VDB for longer of 10 ms, which is adequate for the DIS system. Therefore,
than a predefine time delay Tdelay . VDB and Tdelay are intro- the information update rate of the DIS is set as 20 ms. The
duced in order to avoid frequent and unnecessary switching details of the leaders in the DIS framework for each targeted
ref
around the reference voltage VMV , which may result in the information is listed in Table. 1. The deadband V
DB and
reduction of OLTC lifetime. VDB is often designed symmet- predefined time delay T are set as 4% and 3 s. The tap
delay
rical around the reference. The mechanical time delay Tmech , changing range of the OLTC is ±9 × 1.67% and the mechan-
typically in 3 ∼ 10 s, is required for the OLTC to move the ical delay is 5 s. The threshold for other buses V is set
th
taps by one position. The non-sequential mode is adopted as 0.05 p.u.. Based on the radial topology of the network,
for the OLTC, in which the OLTC makes no distinction be- Buses 07, 15, 20, 24, 31, 36, 45 and the MV side bus of the
tween the first and subsequent tap changes. Thus, suppose main transformer are selected as the monitored buses.
the estimated accumulated time is ∆t̂tri , the time of the tap In order to have stable operation, the network control
change can be estimated as t̂act = t0 − ∆t̂tri + Tmech . has to be slowed down to have at least 5–10 times lower
In this paper, the voltage of the MV side bus of the bandwidth than the DG units. Thus, the control period of
main transformer VMV is controlled by the OLTC. At each the DMPC controllers is designed as 0.5 s. The prediction
control point, if the tap action has been triggered and and control steps are designed as Np = Nc = 5.
the current tap position is not at the minimum N tap (for
VMV > Vref + VDB /2) or maximum position N tap (for 6.1. Normal Operation
VMV < Vref − VDB /2), ∆ttri will be sent to the closest agents In this subsection, the control performance of the pro-
(if not, ∆ttri = 0 will be sent). Then, each MPC controller posed DVC (DMPC) under normal operation is presented
will detect if there is a potential tap change within the pre- and compared with the conventional local constant power
diction horizon. Suppose the current time is t0 , the indica- factor control (PFC) and centralized MPC (CMPC). Con-
tion of the potential tap change for the kth prediction step sidering the negligible communication delay and solution

6
 (a)

(b)

Figure 4: Network topology of the test system.

Table 1: Leaders in the DIS framework

Information Leader
OLTC Info.(∆ttri ) DG01, DG05
Bus voltage (VMV ) DG01, DG05
Bus voltage (VBus07 ) DG01, DG02 (c)
Bus voltage (VBus15 ) DG02, DG03
Bus voltage (VBus20 ) DG04, DG10 Figure 5: Voltage profile of Feeder I with the different control schemes
(Different colors represent different bus voltages). (a) PFC, (b) CMPC,
Bus voltage (VBus24 ) DG05, DG06
and (c) DMPC.
Bus voltage (VBus31 ) DG05, DG06
Bus voltage (VBus36 ) DG07, DG09
Bus voltage (VBus45 ) DG09, DG10 namely a sudden step increase of slack bus voltage (at the
HV side of the main transformer).

time of the centralized control methods, the control period 6.2.1. Information Synchronization Performance
of the CMPC is designed as 2 s. Dynamic voltage pro- The synchronization procedures of the voltage of Bus 20
files of Feeder I are shown in the Fig. 5. As can be seen (monitored bus) VBus20 and the accumulated time ∆ttri are
from Fig. 5(a), the local PFC fails to regulate the voltages illustrated in Figs. 6 and 7, respectively. As can be seen,
within the predefined range of [0.95, 1.05] p.u.. However, the the DIS system can fast track the synchronized information
CMPC and DMPC can both effectively regulate the voltages of monitored bus voltages and OLTC actions, providing ac-
within the range of [0.98, 1.02] p.u., implying the CMPC and curate data feedback for the MPC controller. This validates
DMPC have the similar control performances under normal the effectiveness of the propose DIS framework. In Fig. 7,
operation. Moreover, the results can verify that the network the solid lines represent the accumulated time estimated by
voltage profile can be well regulated only based on several DG01-DG08, respectively. The dash line represents its ac-
monitored voltage bus instead of all bus voltage measure- tual value. And to be noticed, it is reset to “0” after each
ments and feedback in the network. This would be helpful control point of the DMPC controller (such as t = 53.5 s
to reduce the communication and computation burdens of and t = 54 s). This effectively validates the max-consensus
the system. protocol designed in (3).

6.2. Large-Disturbance Operation 6.2.2. Voltage Regulation Performance

In this subsection, the control performance under large The voltage performances under the emergency operation
disturbances in the external grid is examined and com- are illustrated in Figs. 8–9. Firstly, as can be seen from Fig.
pared with the PFC, CMPC and the conventional one-step 8, after the disturbance at t = 50 s, the network voltages
optimization-based distributed optimal control (DOPC) violate the predefined range and go beyond 1.1 p.u.. Then,
without prediction mechanism, of which the control period the OLTC and DVC can cooperatively correct the severe
is designed as 0.5 s. At t = 50 s, the distribution network voltage deviations within the feasible range [0.98, 1.02] p.u.
is affected by a significant disturbance in the external grid, until t = 90 s.

7
 Figure 8: Voltage profile across the network with the DMPC.
Figure 6: Synchronization of VBus20 (average-consensus).

Figure 7: Synchronization of ∆ttri (max-consensus).

As shown in Fig. 9, the optimal control methods, i.e., the

Figure 9: Comparison of voltages of Bus 20 under emergency operation
CMPC, DOPC and DMPC, can effectively accelerate the with the PFC, CMPC, DOPC and DMPC.
voltage recovery. Comparably, the DMPC can help voltages
recover much faster.
Acknowledgment

This work is supported in part by the ForskEL program

7. Conclusion through the ‘IDE4L-DK Top-Up’ project, in part by the
National Key Research and Development Program of China
(2016YFB0900603), in part by National Natural Science
This paper proposes a DVC scheme for smart distribution Foundation of China under grant 51877127, and in part by
networks with inverter-based DGs to regulate the voltage the China Scholarship Council (CSC).
across the network within the feasible range. The proposed
scheme includes two parts: the DIS framework and DMPC-
based voltage control scheme. The former is developed us- References
ing the consensus protocol to synchronize the monitored bus
voltages and triggered time information of the OLTC. The [1] J. Tan and L. Wang, “A game-theoretic framework
synchronization performance has been proven to be effective for vehicle-to-grid frequency regulation considering smart
and fast enough. Simulation results validate the effective- charging mechanism,”, IEEE Trans. Smart Grid, vol. 8, no.
ness of the proposed DVC. Under normal operation, it can 5, pp. 2358-2369, Sep. 2017.
effectively regulate the voltages within the feasible range. [2] Yang H, Li S, Li Q, et al. Hierarchical distributed control for
decentralized battery energy storage system based on con-
Under large-disturbance conditions, it is also able to cor-
sensus algorithm with pinning node. Protection and Control
rect the voltage deviations and avoid the hunting between
of Modern Power Systems, vol. 3, no. 1, pp. 1-9, 2018.
the DGs and OLTC. Compared with the PFC, CMPC and [3] M. J. E. Alam, K. M. Muttaqi, and D. Sutanto, “Mitigation
DOPC, it can better accelerate and smoothen the voltage of rooftop solar PV impacts and evening peak support by
recovery. The distributed control is suitable due to its good managing available capacity of distributed energy storage
scalability and flexibility for the future smart distribution systems,” IEEE Trans. Power Syst., vol. 28, no. 4, pp. 3874-
networks with a large number of DGs. 3884, Nov. 2013.

8
 [4] Zhang D, Li J, and Hui D. Coordinated control for voltage trol of distribution networks hosting dispersed generation,”
regulation of distribution network voltage regulation by dis- in Proc. International Conference and Exhibition on Elec-
tributed energy storage systems. Protection and Control of tricity Distribution, Stockholm, Sweden, 2013.
Modern Power Systems, vol. 3, no. 1, pp. 1-8, 2018. [20] H. S. Bidgoli, and T. Van Cutsem, “Combined local and
[5] K. E. Antoniadou-Plytaria, I. N. Kouveliotis-Lysikatos, P. centralized voltage control in active distribution networks,”
S. Georgilakis, and N. D. Hatziargyriou, “Distributed and IEEE Trans. Power Syst., vol. 33, no. 2, pp. 1374-1384, Mar.
decentralized voltage control of smart distribution networks: 2018.
Models, methods, and future research,” IEEE Trans. Smart [21] Y. Zheng, D. J. Hill, K. Meng and S. Y. Hui,“Critical bus
Grid, vol. 8, no. 6, pp. 2999-3008, Nov. 2017. voltage support in distribution systems with electric springs
[6] G. Mokhtari, A. Ghosh, G. Nourbakhsh, and G. Ledwich, and responsibility sharing”, IEEE Trans. Power Syst., vol.
“Smart robust resources control in LV network to deal with 32, no. 5, pp. 3584-3593, Sep. 2017.
voltage rise issue,” IEEE Trans. Sustain. Energy, vol. 4, no. [22] M. Yazdanian and A. Mehrizi-Sani, “Distributed control
4, pp. 1043-1050, Oct. 2013. techniques in microgrids,” IEEE Trans. Smart Grid, vol.
[7] E. Demirok, P. González, K. H. B. Frederiksen, D. Sera, P. 5, no. 6, pp. 2901-2909, Nov. 2014.
Rodriguez, and R. Teodorescu, “Local reactive power con- [23] B. A. Robbins, C. N. Hadjicostis, and A. D. Domı́nguez-
trol methods for overvoltage prevention of distributed solar Garcı́a, “A two-stage distributed architecture for voltage
inverters in low-voltage grids,” IEEE J. Photovoltaics, vol. control in power distribution systems,” IEEE Trans. Power
1, pp. 174-182, 2011. Syst., vol. 28, no. 2, pp. 1470-1482, May 2012.
[8] K. Turitsyn, P. Šulc, S. Backhaus, and M. Chertkov, “Op- [24] M. Zeraati, M. E. Golshan, and J. M. Guerrero, “A
tions for control of reactive power by distributed photo- consensus-based cooperative control of PEV battery and PV
voltaic generators,” Proc. IEEE, vol. 99, no. 6, pp. 1063- active power curtailment for voltage regulation in distribu-
1073, Jun. 2011. tion networks,” IEEE Trans. Smart Grid, in press.
[9] V. Calderaro, G. Conio, V. Galdi, and A. Piccolo, “Reactive [25] F. Olivierx, P. Aristidou, D. Ernst, and T. Van Cutsem,
power control for improving voltage profiles: A comparison “Active management of low-voltage networks for mitigating
between two decentralized approaches,” Electr. Power Syst. overvoltages due to photovoltaic units,” IEEE Trans. Smart
Res., vol. 83, no. 1, pp. 247-254, 2012. Grid, vol. 7, no. 2, pp. 926-936, Mar. 2016.
[10] V. Calderaro, G. Conio, V. Galdi, G. Massa, and A. Pic- [26] Y. Wang, K. T. Tan, X. Y. Peng, and P. L. So, “Coor-
colo, “Optimal decentralized voltage control for distribution dinated control of distributed energy-storage systems for
systems with inverter-based distributed generators,” IEEE voltage regulation in distribution networks,” IEEE Trans.
Trans. Power Syst., vol. 29, no. 1, pp. 230-241, Jan. 2014 Power Del., vol. 31, no. 3, pp. 1132-1141, Jun. 2016.
[11] N. Mahmud and A. Zahedi, “Review of control strategies for [27] H. E. Farag, E. F. El-Saadany, and R. Seethapathy, “A
voltage regulation of the smart distribution network with two ways communication-based distributed control for volt-
high penetration of renewable distributed generation,” Re- age regulation in smart distribution feeders,” IEEE Trans.
new. Sustain. Energy Rev., vol. pp. 582-595, 2016. Smart Grid, vol. 3, no. 1, pp. 271-281, Mar. 2012.
[12] L. Cheng, Y. Chang, and R. Huang, “Mitigating voltage [28] H. E. Farag and E. F. El-Saadany, “A novel cooperative
problem in distribution system with distributed solar gener- protocol for distributed voltage control in active distribu-
ation using electric vehicles,” IEEE Trans. Sustainable En- tion systems,” IEEE Trans. Power Syst., vol. 28, no. 2, pp.
ergy, vol. 6, no. 4, pp. 1475-1484, Oct. 2015. 1645?1656, May 2013.
[13] A. Kulmala, S. Repo, and P. Jarventausta, “Coordinated [29] B. A. Robbins and A. D. Domı́nguez-Garcı́a, “Optimal re-
voltage control in distribution networks including several active power dispatch for voltage regulation in unbalanced
distributed energy resources,” IEEE Trans. Smart Grid, vol. distribution systems,” IEEE Trans. Power Syst., vol. 31, no.
5, no. 4, pp. 2010-2020, Jul. 2014. 4, pp. 2903-2913, Jul. 2016.
[14] P. Li, H. Ji, C. Wang J. Zhao, G. Song, F. Ding, and J. [30] B. A. Robbins, H. Zhu, and A. D. Domı́nguez-Garcı́a, “Op-
Wu, “A coordinated control method of voltage and reactive timal tap setting of voltage regulation transformers in un-
power for active distribution networks based on soft open balanced distribution systems,” IEEE Trans. Power Syst.,
point,” IEEE Trans. Sustain. Energy, vol. 8, no. 4, pp. 1430- vol. 31, no. 1, pp. 256-267, Jan. 2016.
1442, Oct. 2017. [31] P. Šulc, S. Backhaus, and M. Chertkov, “Optimal dis-
[15] Y. P. Agalgaonkar, B. C. Pal, and R. A. Jabr, “Distribu- tributed control of reactive power via the alternating direc-
tion voltage control considering the impact of PV generation tion method of multipliers,” IEEE Trans. Energy Convers.,
on tap changers and autonomous regulators,” IEEE Trans. vol. 29, no. 4, pp. 968-977, Dec. 2014.
Power Syst., vol. 29, no. 1, pp. 182-192, Jan. 2014. [32] H. J. Liu, W. Shi, and H. Zhu,“Distributed voltage con-
[16] S. Weckx, C. Gonzalez, and J. Driesen, “Combined central trol in distribution networks: online and robust implemen-
and local active and reactive power control of PV inverters,” tations”, IEEE Trans. Smart Grid, in press.
IEEE Trans. Sustain. Energy, vol. 5, no. 3, pp. 776-784, Jul. [33] A. Maknouninejad and Q. Zhihua, “Realizing unified mi-
2014. crogrid voltage profile and loss minimization: A coopera-
[17] G. Valverde and T. Van Cutsem, “Model predictive control tive distributed optimization and control approach,” IEEE
of voltages in active distribution networks,” IEEE Trans. Trans. Smart Grid, vol. 5, no. 4, pp. 1621-1630, Apr. 2014.
Smart Grid, vol. 4, no. 4, pp. 2152-2161, Dec. 2013. [34] K. Utkarsh, A. Trivedi, D. Srinivasan, and T. Reindl,
[18] G. Valverde and T. Van Cutsem. “Control of dispersed “A consensus-based distributed computational intelligence
generation to regulate distribution and support transmis- technique for real-time optimal control in smart distribution
sion voltages,” in Proc. IEEE PowerTech, Grenoble, France, grids,” IEEE Trans. Emerg. Topics Comput. Intell., vol. 1,
2013. no. 1, pp. 51-60, Feb. 2017.
[19] T. Van Cutsem and G. Valverde. “Coordinated voltage con- [35] Yazdanian M, Mehrizi-Sani A, “Distributed control tech-

9
 niques in microgrids,” IEEE Trans. Smart Grid, vol. 5, no.
6, pp. 2901-2909, 2014.
[36] Mahmud N, and Zahedi A, “Review of control strategies for
voltage regulation of the smart distribution network with
high penetration of renewable distributed generation,” Re-
newable and Sustainable Energy Reviews, vol. 64, pp. 582-
595, 2016.
[37] P. Aristidou, G. Valverde, and T. V. Cutsem, “Contribution
of distribution network control to voltage stability: A case
study,” IEEE Trans. Smart Grid, vol. 8, no. 1, Jan. 2017.
[38] H. Sun, Q. Guo, B. Zhang, Y. Guo, Z. Li, and J. Wang,
“Master-slave-splitting based distributed global power flow
method for integrated transmission and distribution analy-
sis,” IEEE Trans. Smart Grid, vol. 5, no. 3, May 2015.
[39] Z. Li, Q. Guo, H. Sun, J. Wang, Y. Xu, and M. Fan, “A
distributed transmission-distribution-coupled static voltage
stability assessment method,” IEEE Trans. Power Syst., in
press.
[40] Y. Guo, H. Gao, Q. Wu, H. Zhao, J. Østergaard, and S.
Mohammad,“Enhanced voltage control of VSC-HVDC con-
nected offshore wind farms based on model predictive con-
trol,” IEEE Trans. Sustain. Energy, vol. 9. no. 1, pp. 474-
487, Jan. 2018.
[41] M. Behrangrad, “A review of demand side management
business models in the electricity market,” Renew. Sustain.
Energy Rev., vol. 47, pp.70-83, 2015.
[42] H. Sugihara, K. Yokoyama, O. Saeki, K. Tsuji, and T. Fu-
naki, “Economic and efficient voltage management using
customer-owned energy storage systems in a distribution
network with high penetration of photovoltaic systems,”
IEEE Trans. Power Syst., vol. 28, no. 1, pp. 102-111, Feb.
2013.
[43] M. Pasetti, S. Rinaldi, D. Manerba, “A virtual power plant
architecture for the demand-side management of Smart Pro-
sumers,” Appl. Sci. vol. 8, no. 3, pp. 1-20, 2018.
[44] Setlhaolo D, Xia X, “Combined residential demand side
management strategies with coordination and economic
analysis,” Int. J. Electr. Power Energy Syst., vol. 79, pp.
150-160, 2016.
[45] R. Olfati-Saber, and R. M. Murray, “Consensus problems
in networks of agents with switching topology and time-
delays,” IEEE Trans. Autom. Control, vol. 49, no. 9, pp.
1520-1533, Sep. 2004.
[46] E. Camponogara, D. Jia, B. Krogh, and S. Talukdar,
“Distributed model predictive control,” IEEE Control Syst.
Mag., vol. 22, no. 1, pp. 44-52, Feb. 2002.
[47] A. Venkat, I. Hiskens, J. Rawlings, and S. Wright, “Dis-
tributed MPC strategies with application to power system
automatic generation control,” IEEE Trans. Control Syst.
Technol., vol. 16, no. 6, pp. 1192-1206, Nov. 2008.
[48] Q. Shafiee, C. Stefanovic, T. Dragicevic, P. Popovski, J. C.
Vasquez, and J. M. Guerrero, “Robust networked control
scheme for distributed sec- ondary control of islandedmi-
crogrids,” IEEE Trans. Ind. Electron., vol. 61, no. 10, pp.
5363-5374, Oct. 2014.

10