You are on page 1of 14

This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2019.2917009, IEEE
Transactions on Power Systems
1

A Day-Ahead Optimal Control of PEV Battery


Storage Devices Taking into Account the Voltage
Regulation of the Residential Power Grid
Yulong Huang


Abstract-- This paper first presents an optimization model to PEV Buses set for feeding controllable PEVs.
flexibly control available plug-in electric vehicle (PEV) battery
charging/discharging power based on three-phase power flow  L
PEV / S
PEV Buses and phases union set of unexpected travel
and sensitivity approaches. This model can achieve one of two PEVs and remaining PEVs at time slot t,
goals: (1) minimizing both battery charging/discharging cost and respectively.
a,b,c
extra battery degradation cost due to vehicle-to-grid (V2G)
activities (cost-reduction strategy) or (2) maximizing local peak Phases set of phases a, b, and c.
load shifting and minimizing extra battery degradation cost due kl
to V2G activities (peak-shifting strategy). The first strategy can
Bmn Imaginary components of Ymn between phases k
determine the appropriate charging/discharging rates of an and l.
available PEV battery in order to benefit both the PEV owners kl
Gmn Real components of Ymn between phases k and l.
and the distribution utilities for the day ahead. The second
strategy can reduce the peak loads of the system. Both strategies Ct Electricity price at time slot t.
can improve the power quality at the same time. With the help of C t
Discharging electricity price at time slot t.
d
a sensitivity method, most of the nonlinear constraints are
transformed into linear constraints, and the number of CPEV i nk
v
Battery capacity of the ivnk ’th PEV.
constraints is reduced in the model. An interior point
optimization approach is utilized to solve the optimization model. Costink Battery cost of the ivnk ’th PEV.
v
The optimization model is modified further to address DODitnk
unexpected PEV connections and travel scenarios during v
Depth of discharge (DOD) of the ivnk ’th PEV at
operation. The effectiveness and accuracy of the proposed time slot t.
method are demonstrated and verified on a 75-node test feeder. DODmaxi nk Daily maximum allowable DOD of the ivnk ’th
v

Index Terms--Three-phase power flow, sensitivity, battery PEV.


degradation, plug-in electric vehicles (PEVs), interior point d v ink End time slot of the ivnk ’th PEV.
v
optimization.
t
EPEV i nk v
Energy storage of the ivnk ’th PEV at the end of
NOMENCLATURE time slot t.
t Duration of time slot t.
UB
EPEV i nk v
Energy storage upper band limit of the ivnk ’th P-
 d
PEV ivnk Inverse of the efficiency rate for the battery EV.
LB
discharging of the ivnk ’th plug-in electric vehicle EPEV i nk v
Energy storage lower band limit of the ivnk ’th P-
(PEV). EV.
PEV
c
i nk Charging efficiency rate of the ivnk ’th PEV. FPEV i nk
v
Self-discharge rate of the ivnk ’th PEV.
v

 opt Time slot set of the PEV battery charging/di- ink


v Index of PEV at phase k of bus n.
scharging power optimization period. Im Vector of the phase currents and the neutral cu-
 PEV i nk PEV charging/discharging time slot set of the rrent of bus m.
v

ivnk ’th PEV. I mZIP Injection currents of the constant power, consta-
 valley Time slot set of the valley load period. nt current, and constant-impedance (ZIP) load of
bus m.
I mPEV Injection currents of the PEV charging/dischar-
This work was partially supported by the Fundamental Research Funds for ging load of bus m.
the Central Universities, JNU under Grant 11617359.
Yulong Huang is affiliated with the College of Electrical Engineering and I Rem , I Imm Vectors of the real and imaginary components of
the Institute of Rail Transportation, Jinan University, Zhuhai 519070, China Im.
(e-mail: thuangyulong@jnu.edu.cn).

0885-8950 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2019.2917009, IEEE
Transactions on Power Systems
2

k
I Re k x Vector of the real and imaginary components of
m , I Imm Real and imaginary components of Im at phase k
of bus m. the voltages at phases a, b, c, and the neutral
K Total bus number. node of all buses at all time slots in  opt .
Lti nk Battery cycle life of the ivnk ’th PEV at time slot t. Ymn Admittance matrix from bus m to n.
v

N PEVnk Total number of charging/discharging PEVs at Z LC Charging/discharging cost of all PEVs.


phase k of bus n. ZS Load shape function.
N Charging/discharging time slot number of the Z Cost Extra battery degradation cost due to vehicle-to-
PEV ivnk

ivnk ’th PEV. grid (V2G) activities.

N  opt Time slot number of the PEV battery I. INTRODUCTION


charging/discharging power optimization period.
N
valley 
PEV ivnk
Charging time slot number of the i ’th PEV
nk
v
I
n recent years, the market penetration of electric vehicles
has been growing rapidly around the world. However,
unlike charging at public charging stations, which are
during the valley-load period. deliberately designed for charging plug-in electric vehicles
PmZIP Active injection power corresponding to ZIP lo- (PEVs), the random charging of electric vehicles at home can
ads at phases a, b, and c of bus m. bring several negative consequences to the existing
PnkZIP t Active injection power corresponding to ZIP lo- distribution grid. These negative impacts include an increased
ads at phase k of bus n at time slot t. peak load, the degradation of voltage, the severity of the
PmPEV Vector of total charging power corresponding to power imbalance in the distribution system, and the stress on
the distribution transformers [1]-[4]. These unwanted
all PEV charging/discharging loads at phases a,
consequences have inspired many research studies [5]-[32] on
b, and c of bus m.
how to charge PEVs while creating acceptable influences on
PnkPEV Total charging power at phase k of bus n. the distribution system.
Pcvt i nk The ivnk ’th PEV charging power at time slot t. The optimal charging of PEVs can be broadly classified
v

Pdvt i nk into centralized and distributed charging approaches [5]. The


The iv ’th PEV discharging power at time slot t.
nk
v
centralized PEV charging approach [6]-[17] gives centralized
Ploss Distribution network loss. control and management capability to the aggregator unit
Pnt Total injection power at the three phases of bus n (AU). The main objectives of the major optimization
at time slot t. approaches to the centralized PEV charging method are the
PcvNci nk
enhancement of AU profit, the minimization of costs, and the
Maximum power that can be drawn for charging
v
minimization of the peak load on the distribution system. The
the battery of the ivnk ’th PEV. provision of ancillary services such as voltage control and
PdvNci nk
v
Maximum power that can be released for di- battery state of charge (SOC) is well supported by the
centralized approach through vehicle-to-grid (V2G)
scharging the battery of the ivnk ’th PEV. technology. As described in [18], unidirectional V2G indicates
QmZIP Reactive injection power corresponding to ZIP that PEVs offer ancillary services by controlling their charging
loads at phases a, b, and c of bus m. powers, and bidirectional V2G refers to both charging and
rv ink discharging controls during service provision. In this paper,
Beginning time slot of the ivnk ’th PEV.
v
the term V2G refers to the bidirectional power that flows
SOCinitial ink
v
Initial state of charge (SOC) of the ivnk ’th PEV. between PEVs and grids in order to offer extra services such
SOCdest ink
v
Desired SOC at end of the charging/discharging as voltage control, load flattening, and the minimization of
battery or distribution utility costs.
period of the ivnk ’th PEV. X. Wang et al. [8] established a physical fair-queuing
u Control variable vector. framework for PEV charging in order to solve the under-
Vn Vector of the phase voltages at phases a, b, c, an- voltage and transformer overloading problems. Before each
d the neutral node of bus n. period starts, the central controller relies on the scheduling
VRen , VImn scheme to determine the on or off state for each PEV.
Real and imaginary components of the vector Vn .
However, bidirectional V2G service is not considered in the
VRekt n , VImkt n Real and imaginary components of Vn at phase k control scheme. A probabilistic approach to the optimal
of bus n at time slot t. charging of PEVs in distribution systems was proposed by A.
Vnkt Voltage amplitude of Vn at phase k of bus n at Arias et al. [9] in order to minimize the costs of both energy
time slot t. losses and energy demanded by PEVs and other loads. This
approach does not consider the power imbalance between
Vmax , Vmin Maximum and minimum bus voltage limits.
phases, though. A coordinated charging method was proposed
Wg Weight factor. by K. Clement-Nyns et al. [10] to minimize the power loss and

0885-8950 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2019.2917009, IEEE
Transactions on Power Systems
3

to maximize the main grid load factor. Since the exact photovoltaic impacts, such as voltage increases during midday
forecasting of household loads is not possible, stochastic or the support of the evening load peak [29]. However, these
programming is introduced, although the charging cost is not methods still cannot guarantee the voltage limits, especially
considered in this approach. To reduce the negative effects of the control of the neutral potential.
unsteady renewable generation outputs, a stochastic The contributions of this paper include the following. First,
optimization-based dispatch model capable of handling the a flexible day-ahead optimal control (DAOC) model is
uncertain outputs of PEVs and renewable generation was developed based on the three-phase power flow and sensitivity
formulated by G. Wang et al. [11]. However, this model also approaches in order to utilize the available PEV battery
does not consider the power imbalance between phases. F. Wu capacity. This model can achieve one of two beneficial goals:
et al. [12] proposed a centralized electric vehicle charging- (1) minimizing both the battery charging/discharging cost and
control model in order to optimize the PEV-charging loads the extra battery degradation cost due to V2G activities (cost-
after each PEV arrives at the charging station in order to reduction strategy) or (2) maximizing the local peak load
minimize costs. This model captures the use of distributed shifting and minimizing the extra battery degradation cost due
energy resources and uncertainties around PEV arrival times, to V2G (peak-shifting strategy). The cost-reduction strategy
charging demands upon arrival, and other factors, minimizing can determine the appropriate charging/discharging rates of an
the impacts of the resulting load profile on the distribution available PEV battery in order to benefit both the PEV owners
transformer. However, it does not consider the voltage profile and the distribution utilities. The peak-shifting strategy can
or the load imbalance. Based on a stochastic model of a PEV, determine the appropriate charging/discharging rates of an
a two-layer evolution strategy particle swarm optimization available PEV battery in order to provide a maximum level of
algorithm was proposed by J. Tan et al. [13] to deal with the peak-shifting support when the predicted maximum peak load
stochastic nature of the vehicle arrival time, departure time, is unacceptable. Second, both strategies can solve the over-
and daily mileage. W. Tang et al. [14] formulated the charging limit voltage problem, including the alleviation of the neutral-
scheduling problem as a finite-horizon dynamic programming to-ground potential rise. Third, each PEV user can choose to
problem. To avoid the prohibitively high complexity of participate in V2G service with equal charging fairness
solving such a dynamic programming problem, [14] provides according to his or her PEV battery degradation cost and needs,
a model predictive control (MPC)-based algorithm. L. Liu et and each user can further choose to input the arrival time,
al. [15] developed a model minimizing the PEV charge- departure time, travel interval, and desired SOC level. This
discharge fee without considering voltage constraints using a paper explicitly considers PEV owner convenience, which can
dynamic capacity fading model for battery discharge. The be mainly characterized by a desired SOC at the departure
neutral potential rise is not considered in the above control time. Fourth, the proposed scheme only requires one-way
schemes. communication between the utility distribution center and the
T. Shekari et al. [16] developed a generalized energy distributed processing unit. Finally, an optimal control model
management approach for smart microgrids with the modified from the DAOC model is developed to treat
collaboration of PEV charging facilities in both active and unexpected PEV connections or travel scenarios and the actual
reactive power exchanges, which reduces operation cost and SOC deviation from the pre-set value during the execution of
enhances power quality. However, the established model is the DAOC strategy.
tailored only for balanced three-phase networks. C. Sabillon- The DAOC strategy was applied to a 75-node test feeder,
Antunez et al. [17] presented a new approach for solving the and actual PEV data was used. The results indicate that the
electric vehicle charging coordination problem, considering proposed strategy can render the cost of PEV battery charging
Volt-var control, the energy storage device operation, the and the financial losses to the distribution network lower than
on/off control of PEV charging, and the dispatchable those associated with maximum power charging or even
distributed generation available in three-phase unbalanced power charging. It can also improve the quality of the three-
electrical distribution networks. However, this method does phase node voltage, and alleviate the neutral voltage rise.
not take into account the V2G service of PEVs and the neutral Results from the simulation of the peak-shifting strategy show
potential rise. Additionally, the integration of the Volt-var that the proposed strategy can effectively provide peak-
control with a charging control strategy makes the shifting support.
infrastructure investment more expensive and requires fast The rest of the paper is structured as follows. In Section II,
two-way communication between the distribution company a DAOC approach is introduced. Section III discusses the
and the customers. communication infrastructure for the DAOC strategy. The
In the distributed or decentralized PEV charging approach application of the proposed DAOC strategy to a test
[18]-[29] for PEVs, the AU does not make all the decisions distribution network is presented in Section IV. In this section,
regarding the charging or discharging. The main objectives of a comparison with a heuristic charging approach [25] is also
the major optimization approaches to the decentralized PEV presented. Finally, the main conclusions and contributions of
charging method are the avoidance of excessively long waiting the paper are highlighted in Section V.
times at the charging stations [19], the tracking of a desired
total load profile [18], [20]-[24], the minimization of the total II. A DAY-AHEAD OPTIMAL CONTROL APPROACH
cost of charging [21], [25]-[28], and the reduction of the solar
A. Three-phase Power Flow

0885-8950 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2019.2917009, IEEE
Transactions on Power Systems
4

The impact of the increasing integration of PEVs into the B. Objective Functions and Sensitivity Computation
residential distribution grid may include issues such as power In this section, the power loss cost of the distribution
imbalances and voltage rises/drops. A three-phase power flow network, the charging/discharging cost, the extra battery
is adapted here in order to compute the three-phase voltage of degradation cost, and the load shape function are given. The
a residential distribution grid with PEV integration. sensitivities of these variables to the control variable and the
The battery charging time in a day can be divided into 96 sensitivity of the voltage to the control variables are presented.
equal time slots, i.e., the duration of each time slot is 15 These sensitivities transform the nonlinear constraints into
minutes. I m is the vector of the phase currents, as well as the linear constraints in the DAOC.
neutral node current of bus m, as seen in (1), and I m includes According to (7), the sensitivity of x with respect to u,
which is denoted as xu, can be obtained as follows:
I mZIP and I mPEV as given in (2):
xu  -Fx1Fu , (8)
I m  I Rem  jI Imm where Fx and Fu are the sensitivities of equation F with
T T
  I Re
a b c N
  a b c N

m I Rem I Rem I Rem   j  I Imm I Imm I Imm I Imm  , (1)
respect to x and u, respectively.
The sensitivity, Vu , of the voltage amplitude vector
Im   Im  Im .
ZIP PEV
(2)
T
The injection currents of the constant power, constant current, V   Vnat Vnbt Vnct VnNt  , t   opt with respect to the
and constant-impedance (ZIP) loads and the PEV load are
control variable can be obtained using:
respectively computed as follows:
Vu  Vx xu . (9)
 P  jQmZIP  Vm ,
*
I mZIP  m
ZIP
The distribution network loss, Ploss, at any time slot t is
(3)
computed as follows:
 P Vm 
PEV PEV *
I
  V
K K
Ploss   
m m
. (4) l
Gilkj VRek  j
Rei
The vector of the phase voltages, Vn , is defined in (5): i 1 j 1 l  a,b,c,N k  a,b,c,N

Vn =VRen  jVImn VRel i BijlkVImk  j  VImi l BijlkVRek  j  VIml iGijlkVImk  j  . (10)


T T
= V V V V
a
Ren   j V V V V  . (5)
b
Ren
c
Ren
N
Ren
a
Imn
b
Imn
c
Imn
N
Imn
The charging/discharging cost, ZLC, of all PEVs is
calculated using (11), which is the total charging cost minus
Since the battery energy storage system can operate in a
the total discharging income:
constant power charging/discharging mode when the SOC of N PEVnk
the battery is in the normal range [30], a constant power model
is used to model the PEV battery charging/discharging loads
Z LC     
nPEV k  a,b,c
C t P t nk  Cdt P t nk  t .
 cviv dviv 
(11)
ivnk 1 t
PEV ivnk
in this paper. The total charging power, PnkPEV , is obtained
The total power loss cost of the distribution network, ZlossC,
using: within  opt time slots is computed by:

 P 
N PEVnk
PnkPEV  cvivnk
 Pdvink .
v
(6) Z lossC  

t
C P  t  t . t
loss
(12)
ivnk 1 opt

Given I m and the total system admittance matrix, the The sensitivity of the charging/discharging cost (ZLC) with
vector of the system voltages ( Vn ) can be obtained by solving respect to the control variables Pcvt i nk and Pdvt i nk at the
v v

the classical network current equations using the Newton-    PEV i nk ( n PEV ) time slot can be obtained as follows:
v
Raphson algorithm as seen in detail in [31]. It can be seen that
Z LC
the power imbalances between the phase and the neutral  C t , (13)
potential are all considered in this power flow computation. Pcv ink
v
The expansion of the real and imaginary parts of the
Z LC
classical network current equations can be formulated in the  Cd t . (14)
Pdv ink
short term for any time slot t   opt as follows: v

F  x, u   0 ,
The sensitivity of ZLC with respect to the other control
(7)
variables is zero.
T
Battery degradation is one of the major challenges of V2G
where u   Pcvt i Pcvt i Pcvt i Pdvt i Pdvt i Pdvt i  ( ivnk =
 na
v
nb
v
nc
v
na
v
nb
v  nc
v technology [32]. The extra battery degradation cost (ZCost) due
1, …, N PEVnk , k  a,b,c , n PEV , and time slot t ∈  PEV ink ) to V2G activities can be expressed as (15) [15] according to
the relationship (16) between the battery cycle life ( Ltink ) and
v

is the control variable of the DAOC strategy, and v

(n = the depth of discharge (DOD) (17) of the iv ’th PEV at phase


T nk
x   VReatn VRebtn VRectn VReNtn VImatn VImbtn VImctn VImNtn 
k of bus n:
1, .., K, and t ∈  opt ). It should be noted that (7) includes all
 
N PEVnk

    Costink  DODitnk / 2151 , (15)


2.301

the equations at any time slot t ∈  . Z Cost 


PEV ivnk nPEV k  a , b , c ivnk 1 t
v v
PEV ivnk

0885-8950 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2019.2917009, IEEE
Transactions on Power Systems
5

  ZS
2.301 K
Ltink  2151  DODitnk ,DODitnk  0,0.9 , (16)  2  t 
2
 Pi t , (24)
v v v
Pcv ink i 1


 P t nk  d nk  F nk  P t nk  c nk t C nk , 
v

 dviv PEV iv cviv ZS K


 2  t   Pi t .
PEV iv PEV iv PEV iv 2
 (25)
DODink  if Pcvink PEV ink  Pdvink PEV ink  FPEV ink
t t c t d
.(17) 
Pdvink i 1

v v v v v v v

0,otherwise The sensitivity of ZS with respect to the other control variables


 is zero.
It should be mentioned that the charging/discharging cost (ZLC)
and the extra battery degradation cost (ZCost) represent the C. DAOC Model
expense to the PEV owners. The extra battery degradation cost A flexible DAOC model based on three-phase power flow
was seldom considered in previous studies [8]-[10]. is developed as follows:
The sensitivity of the depth of discharge, DODitnk , with min Fun(i+1) (26a)
s.t. Vmin  V    Vu  u  i 1
 ui   Vmax , 
v
i i
respect to the control variable u of the i ’th PEV at phase k nk (26b)
v

0  Pcvt ink 
i 1
of bus n at the    PEV ink time slot can be obtained as follows: t   PEV ink ,
 PcvNcink , (26c)
v v v v

0  Pdvt ink 
i 1
DODink PEV c
t CPEV ink ,DODitnk  0 t   PEV ink ,
 PdvNcink ,
 ivnk (26d)
=
v v v v v v
, (18)
 ,

Pcvink   DODivnk  0
t
Pdtvink 
i 1
v t   valley
 0,  PEV i nk (26e)
v v


DODink t CPEV ink ,DOD  0
d t
Pcvt ink 
i 1
Pdtvink 
i 1
t   PEV ink ,
 0,
 PEV ivnk ivnk (26f)

v v


= . (19) v v v

Pdvink  0, DOD t


 0 i 1
v  iv
nk LB
EPEV i nk
 EPEV
t
i nk
 EPEV
UB
i nk
t   PEV ink ,
, (26g)
v v v v

The sensitivity of the DODitnk with respect to the other control  d i 1
SOCdest ink CPEV ink  EPEV
v iv nk
v
i nk
, (26h)
variables is zero. Therefore, the sensitivity of the extra battery v v v

degradation cost (ZCost) with respect to the control variable u


of the ivnk ’th PEV at phase k of bus n at the    opt time slot t
 DOD t i 
ivnk
 DODitnk u  u
v
i
 i 1
 u 
i
  DOD
maxivnk
,
PEV ivnk

can be found using: (26i)


DODink  PEV i   opt , and (26j)
Z Cost
 
1.301 nk
 Costink  2.301 DODink / 2151 v
, (20) ivnk , n , k
v

Pcv ink v v
Pcv ink
v v
n  1, , K ;k  a,b,c . (26k)
DODink
Z Cost
 
1.301 All the variables with the superscripts (i) and (i+1) in the
 Costink  2.301 DODink / 2151 v
. (21) model represent values obtained at iterations i and i+1,
Pdv ink v v
Pdv ink
v v
respectively. The model is solved iteratively from the first
The total injection power ( Pnt ) at three phases of bus n at iteration where i=0, as explained later in Section II part D. At
time slot t can be computed using: iteration i, the variables with the superscripts (i) are known,
  and the variables with the superscripts (i+1) need to be
 
N PEVnk
Pnt    PnkZIP t   Pcvt ink  Pdvt ink  . (22) computed. This model aims to minimize one of the two Fun
k  a,b,c 
 
v v
ivnk 1 functions, either the cost-reduction objective function or the
The load shape function (ZS) is formulated as follows: peak-shifting objective function. The first objective function,
 K 2
 .
the cost-reduction function, is given by:
ZS  

t
 t  P 
  i 1
i
t (23) Fun=ZLC+ZCost. (27)
opt
The second objective function, the peak-shifting function, is
Given the total amount of energy used in  opt , ZS is minimal expressed as:
when the load profile is as flat as possible [18], [33]. It should Fun=ZS+WgZCost. (28)
The first scheme provides V2G service with minimum
be noted that if t   PEV i nk , the second term inside the
v expense to the PEV owners, i.e., ZLC+ZCost. The second
brackets in (22) is zero. The result of using ZS as an objective scheme also considers the expense to PEV owners, i.e., ZCost.
is that the controller would shift loads towards periods with The charging/discharging cost (ZLC), the extra battery
lower demand. degradation cost (ZCost), and the load shape function (ZS) in the
The sensitivity of the load shape function (ZS) with respect (i+1)’th iteration can be respectively obtained as:
to the control variable u of the ivnk ’th PEV at phase k of bus n Z LCi 1  Z LCi   Z LCui  ui 1  ui  , (29)  
at the    PEV ink time slot can be obtained using (24) and
v
ZCost  i 1
 ZCost  ZCostu i  i 
u  i 1
 ui  , (30)
(25):
ZSi 1  ZSi   ZSui  ui 1  ui  .   (31)

0885-8950 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2019.2917009, IEEE
Transactions on Power Systems
6

r
v i nk D. Implementation
The level of stored energy ( EPEVv ink ) at the end of time slot
v
The developed optimization formula (26) can be easily
rv ink for the (i+1)’th iteration can be calculated as: solved using the interior point optimization method. The step-
v
by-step algorithm for the proposed day-ahead optimal
EPEV ink    Pcvviivnk  PEV v iv   d
r nk i 1 r nk i 1 r nk i 1
v iv c
iv nk t  P
dvivnk  nk t  F
PEV iv nk t
PEV iv charging control model is given as follows:
v v
Step 1) Input data for the next day from the distribution utility,
 SOCinitial ink CPEV ink . (32)
v v such as the forecast loads, the distribution network
Equation (33) ensures that the inter-step energy topology, the distribution network parameters, the
t
conservation ( EPEV ink ) in the (i+1)’th iteration considers drawn controllable PEV charging data, including its capacity,
v
the charging/discharging rated power, and the
or released energy, accounting for the specific efficiency and
historical arrival time, departure time, and travel
self-discharge:
i 1
interval. The charging/discharging power of the
 Pcvt ink  PEV t  Pdvt ink  PEV
i 1 c i 1 d
t
EPEV ivnk ivnk ivnk
t  FPEV ink t uncontrollable PEVs is taken as a constant power
v v v

 t 1 i 1 value and a part of the ZIP loads. The arrival time,
 EPEV ink ,t   PEV ink t  rv ink . (33)
v v v departure time, and travel interval of each PEV and
In the DAOC model (26), constraint (26b) ensures that the initial SOC are either obtained using the statistical
bus voltage is always within the deadband and that the neutral expected value computed from their historical data, or
potential rise is limited, which is not considered in [8]-[11], input by the PEV user in advance. Set iteration i=0.
[17]. Constraints (26c) and (26d) limit the maximum power Step 2) Adjust the settings of all the voltage control devices
that can be drawn or released by the PEV battery. If the PEV according to the traditional method.
user does not participate in V2G service, the discharging upper Step 3) Perform a three-phase power flow. In the first iterati-
limit is set to zero. Constraint (26e) either disables discharging on, initialize the PEV charging power. If the settings
during the valley load period, or alternately deletes the
of all the traditional voltage control devices do not
discharging power variables directly from the control variables
change after the first iteration and before the
during the valley load period. Constraint (26f) guarantees
maximum iteration is reached, or if the maximum
either the charge state or the discharge state at any control
time. Constraint (26g) ensures that the stored energy is always iteration is reached, output the optimal solution and
within the deadband. Constraint (26h) ensures that the SOC is go to Step 6; otherwise, go to Step 4.
not smaller than the target value at the due date. Constraint Step 4) Compute all the sensitivities for each time slot based
(26i) guarantees that the daily DOD is within the allowable on the corresponding distribution network topology,
scope. Constraint (26j) ensures the time slot set for the PEV the predicted load level, and the PEV charging power.
battery’s charging/discharging power optimization is a union Step 5) Solve the optimization model (26) using the interior
of all the PEV charging/discharging time slots. Instead of an point optimization method. Update i=i+1. Go to Step
on/off control [9], the PEV charging/discharging power is 2.
adjusted, which brings more flexible control to the distribution Step 6) When the time slot of the next day is reached, send
system. the corresponding optimal charging/discharging
There are N noneq nonequivalent constraints in the DAOC power value to each available PEV
model, which can be obtained using: charging/discharging device. If the amount of the
N PEV’s stored energy is greater than the energy
N noneq  8KN   6    N   2   N PEVnk . (34)
PEVnk

opt
PEV n  nk
v
PEV ivnk
k  a , b , c i 1 PEV n k  a ,b,c
storage lower band limit, discharge the PEV
The first term in (34) is the number of linear-inequality according to the DAOC optimization value. If the
constraints (26b), the second term is the number of linear- amount of the PEV’s stored energy is less than the
inequality constraints (26c), (26d), and (26g), and the third energy storage upper band limit, charge the PEV
term is the number of linear-inequality constraints (26h) and according to the DAOC optimization value.
(26i). Moreover, when the next day begins, go to Step 1 and
There are N eq equivalent constraints in the DAOC model, perform the next round of computation with the new
input data for the following day.
which can be obtained using: Traditional voltage control in the distribution system is
N N
N eq     N      N
PEVnk PEVnk
. (35) usually achieved by incorporating an on-load tap changer
nk valley nk 
n k  a , b , c i 1
nk PEV iv
n nk k  a , b , c i 1
PEV iv
PEV v PEV v (OLTC), switched capacitors, and a step voltage regulator
The first term in (35) is the number of nonlinear equality (SVR). The traditional control system of a tap changer-based
constraints (26f), and the second term is the number of linear voltage regulator measures the voltage and the load current,
equality constraints (26e). estimates the voltage at the remote point, and triggers the tap
Without the introduction of the sensitivity method, all the change when the estimated voltage is out of boundary, usually
constraints in model (26) are nonlinear with respect to the employing the traditional line drop compensation method [34].
control variables. Furthermore, the three-phase power flow However, the traditional control system may fail under a high
equations must be added to the constraints of the model (26), penetration of PEVs since it is difficult to predict the load
which greatly increases the computational burden of the model. current of the intermittent charging/discharging power without

0885-8950 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2019.2917009, IEEE
Transactions on Power Systems
7

any charging/discharging control. In Step 2 of the proposed where variables with the superscript 0 represent values
DAOC approach, all the voltage control devices act according obtained by the DAOC model (26). Those charging powers
to traditional method, and then the computation of the model are only controlled if Pcvt ink  0 ; other charging powers remain
v
(26) is performed based on the obtained tap positions of the
OLTC and the step-voltage regulator, as well as the switch equal to zero. This means that the obtained discharging power
position of the capacitors. Since all the traditional voltage in model (26) is unchanged and the constraints of the maximal
control devices act according to a controlled daily DOD are not broken. The cost reduction objective
charging/discharging power and a predicted load level, the function is taken as ZLC, which only computes the value for
voltage control effect will be highly improved. Moreover, one time slot in (11). Constraint (36d) guarantees that the
unlike the integration of the Volt-var control with the charging adjusted charging power is not less than the charging power
control strategy in [17], the proposed DAOC strategy needs obtained in model (26). All constraints include the constraints
less infrastructure investment. corresponding to only one time slot. Subsequently, the number
of constraints is reduced to N noneq  8K  4  N PEVnk , and the
E. Treatment of Unexpected Scenarios  n , k SPEV
However, in reality, there are three kinds of unexpected computation burden of the revised optimal control model is
scenarios that need to be addressed in Step 6 of the DAOC greatly reduced. The computation of model (36) is performed
algorithm. Firstly, if a certain number of PEVs unexpectedly based on the actions of the voltage control devices according
plug in to the distribution system, the DAOC obtained above to the traditional method, too. If the objective function value
provides no optimal control signals for these extra plugged-in change is less than the threshold, or the maximum iteration is
PEVs. Secondly, if a certain number of PEVs are reached, the optimal solution is output. Then the
unexpectedly used for travel, i.e., they do not travel according corresponding optimal charging/discharging power value is
to the input data at Step 2 of the DAOC implementation immediately sent to each available PEV charging/discharging
procedure, the voltage limitation cannot be guaranteed any device. After the PEVs travel unexpectedly or the actual SOC
longer with the execution of the optimal control obtained initial level deviates from the pre-set value, the revised
above. Lastly, when the actual SOC initial level is smaller optimal control model (36) is used to optimize the next time
than the pre-set value, the PEV may not be able to be slot, recurrently.
discharged by following the obtained DAOC strategy due to
the energy storage lower band limitation. If the actual SOC III. COMMUNICATION INFRASTRUCTURE FOR THE DAOC
initial level is greater than the pre-set value, the PEV can be MODEL
charged to its desired level earlier. These PEVs cannot The communication infrastructure for the proposed DAOC
charge/discharge according to the DAOC optimization value. model between the distribution utility and the PEV is as
Although most of these kinds of events occur with a small shown in Fig. 1. The quarterly forecasted loads and the
probability, the usage of the proposed DAOC model was distribution network topology are available from the database
further developed in order to deal with these cases. maintained by the distribution utility. The distribution utility
If a certain number of PEVs unexpectedly plug in, the sends this data to the corresponding distributed processing
obtained charging/discharging power of the original PEVs units that are used for the DAOC cyclic computation by means
remain unchanged, and the newly plugged-in PEVs are only of RS-485 telecommunication technology, in which only one-
optimized with the DAOC model. After waiting for the period way communication is required between the utility distribution
of the DAOC optimization computation, the extra plugged-in center and the distributed processing unit. Similar to the
PEVs begin to charge/discharge. Furthermore, if the PEVs do communication method used in [8], the DAOC controller
not have time to wait, a simple control can be made to average collects status information from the PEV charging/discharging
out the total charging power among their charging periods, and devices and dispatches the optimization results to the PEV
their charging periods are better chosen in the valley load charging/discharging devices using ZigBee technology. Inbuilt
period. actuators of the PEV charging/discharging devices
If the PEVs unexpectedly travel or the actual SOC deviates automatically adjust the charging/discharging power of the
from the day-ahead pre-set value, all bus voltages are checked PEV after receiving the control signals in real time. It should
for the next time slot recurrently. If the voltage limitation is be noted that since the proposed DAOC strategy is optimized
violated at time slot t, the DAOC model is revised as follows while considering the implementation of the traditional
in order to optimize the remaining control variable using the voltage control, the conventional voltage control and the
cost-reduction strategy: proposed DAOC strategy can work simultaneously, and the
min Fun(i+1) (36a) conventional voltage control system does not need to be

s.t. Vmin  V i   Vui  ui 1  ui   Vmax , (36b)  updated.
0 P t
cvivnk
i 1
P Nc
cv ivnk
t   PEV ink ,  n, k  SPEV ,
, (36c) IV. APPLICATION EXAMPLE OF THE PROPOSED DAOC
v

i 1 MODEL
E t
PEV ivnk
0
E t
PEV ivnk
E UB
PEV ivnk
t   PEV ink ,

v
A. Residential Distribution System Under Test
 n, k  SPEV , (36d) The residential distribution grid studied here is based on the

0885-8950 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2019.2917009, IEEE
Transactions on Power Systems
8

75-node feeder topology extended from an IEEE 34-node test yuan/kWh, while during the peak load period at 7:00–22:45,
feeder [35], as shown in Fig. 2. The non-PEV load profile is s- the electricity price is 3.11 yuan/kWh. The discharging
TABLE I
Utility Distribution Center
RATED CHARGING POWER AND CAPACITY OF EACH PEV
Available Database
Forecasted Load
Distribution Network Topology PcvNcina CPEV ina PcvNcinb CPEV inb PcvNcinc CPEV inc
PEV Charging Data Node v v v v v v

(kW) (kWh) (kW) (kWh) (kW) (kWh)

RS-485 37,39,42,4
4,48,52,55 8 47.5 7.5 41.4 8 47.5
Distributed Processing
Unit and DAOC Controller ,59,67,75

ZigBee
electricity price is 2.5 yuan/kWh.  valley is the time slot set

PEV PEV PEV PEV


during 22:45–7:00. Assuming SOCdest ink = 0.85,
v
Battery Battery Battery Battery
UB LB
Fig. 1. Communication infrastructure between the distribution utility and the EPEV i nk
CPEV ink = 0.9, EPEV i nk
CPEV ink = 0.2, and DODmaxink
v v v v v
PEV.
= 0.2, the battery cost is 3000 yuan/kWh, the upper and lower
49 50 51 52 bus voltage limits are 1.1 p.u. and 0.9 p.u., respectively, and
24
53 54 55 the upper and lower neutral voltages are 0.0025 p.u. and zero
23

22
45 46 47 48 p.u., respectively. The benchmark power is 333.3 kVA. The
11
39
21
56 57 58 59
maximum iteration number of the DAOC is 10. The maximum
38 32
Zone
10
9 29
17 30 18 20 68 69 70 71 72 73 74 75 iteration number of the interior point optimization is 9. The
Substation

16
31
19 43 44
convergence errors for the control variable and the constraints
1 2 3 4 6 7 25 8 12 13 33 34
are 10-10 and 10-12, respectively.
5 26
35
60
61
During the initial condition, no PEV participates in the
14 15 27 28
36 62 V2G service, and all PEVs begin to charge to their maximum
37 63
40 64 charging power until they are fully charged when people
41
42
65
return home at 19:30, which is a PEV owner’s desired battery-
66
67 charging mode. In Part E of this section, the PEV charging
Fig. 2. Topology of the studied residential distribution grid. power initialization is changed in order to show the
convergence effect of the DAOC algorithm.
caled from [36]. An idealistic value of the neutral to ground
resistance is determined as 10-4 ohms for a solidly grounded B. Cost-Reduction Strategy
neutral of a four-wire bus [31], and this value is chosen in the In this strategy, the objective function is taken as (27). The
test. non-PEV load is also increased by 16% to show the control
It was assumed that 30 houses with 30 PEVs were in the effects, and some of the optimization results are listed in Table
network (Table I). Two types of PEVs were considered: the II. The load levels before and after the increase are named load
‘Build Your Dreams (BYD) model EV300’ with a 47.5 kWh level 1 and load level 2, respectively. Before the optimal
battery capacity [37] and the ‘Beijing Automotive Industry control in load level 1, the phase ‘b’ voltage of node 7 is the
Corporation (BAIC) model EU260’ with a 41.4 kWh battery highest. After optimization, the highest voltage decreases by
capacity [38]. The charging and discharging efficiency of each 0.26% from 1.1023 p.u. to 1.0994 p.u., and the lowest phase ‘b’
PEV battery was considered to be 88%. The self-discharge voltage of node 7 increases 0.57% from 1.0401 p.u. to 1.0460
rate of each PEV battery was 0.1% capacity each day [39]. p.u. The voltage comparison is shown in Fig. 3. Before the
According to the 2009 National Household Travel Survey optimal control in load level 2, the phase ‘a’ voltage of node
[40], some curve fitting results for the trip attributes show that 34 is the smallest. After optimization, the smallest voltage
both the arrival and departure times of the vehicles follow a increases 2.30% from 0.8789 p.u. to 0.8991 p.u. The voltage
normal distribution [13]. According to an investigation of the comparison is shown in Fig. 4. After the optimization of the
resident driving habits in our community in China, the arrival two load levels, almost all of the bus voltages are limited to
and departure times of the vehicles follow normal distributions the range of [0.9, 1.1] p.u.
too, with mean values of approximately 19:30 and 7:30 and Before the optimal control at load level 1, the neutral
standard deviations of 0.51 and 0.48 hours, respectively. voltage of node 46 is the highest, and it reaches 2.5161×10-3
Therefore, the charging start time and the end time are set as p.u. After optimization, the voltage decreases by 0.61% and
19:30 and 7:30, respectively, in the test case, and the the neutral voltage is almost controlled within 2.5×10-3 p.u..
temporary departure time is neglected. It is assumed that all The neutral voltage of node 46 decreases by 2.08% at load
PEVs can participate in V2G service and only one vehicle per level 2. Fig. 5 shows the comparison of the neutral voltage of
household can be charged, i.e., ivnk = 1 in the test case. node 46 before and after optimization at load level 2.
Two electricity price levels are adopted. During the valley The comparison of the load variation before and after
load period at 22:45–7:00, the electricity price is 0.95 optimization is shown in Fig. 6 at load level 1. After

0885-8950 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2019.2917009, IEEE
Transactions on Power Systems
9

optimization, the load variation range decreases, the minimal Instead of centralized charging almost in the peak load period,
load increases from 1.42766 p.u. to 1.48466 p.u., and the max- PEV battery charging occurs most often in the valley load
period, which helps to achieve peak shaving and valley filling.
The SOC variation curve of all PEVs after optimization at
load level 1 is shown in Fig. 8. After optimization for the two
load levels, all SOC values are limited within the range of [0.2,
0.9], and the SOC of almost all PEVs for the last time slot is
Fig. 3. Comparison of the phase ‘b’ voltage of node 7 before and after not less than 0.85. All PEVs can be recharged to the desired
optimization at load level 1.
SOC level when they depart. Moreover, most of the discharge
service is provided from 20:45 to 22:30.

Fig. 4. Comparison of the phase ‘a’ voltage of node 34 before and after
optimization at load level 2.

Fig. 6. Comparison of the load variation before and after optimization using
the objective function (27) at load level 1.

Fig. 5. Comparison of the neutral voltage of node 46 before and after


optimization at load level 2.

TABLE II
COMPARISON OF OPTIMAL CONTROL RESULTS USING THE TWO
STRATEGIES WITH MAXIMUM CHARGING POWER INITIALIZATION
Fig. 7. Comparison of the total charging power of the PEVs before and after
Cost-reduction optimization using the objective function (27) at load level 1.
Strategy Peak-shifting strategy
strategy
Objective function ZLC+ZCost ZS+WgZCost
Load level Level 1 Level 2 Level 1 Level 2
Initial ZCost (yuan) 0
ZCost after
0.04 0.04 0.05 0.01
optimization (yuan)
Initial charging cost Fig. 8. SOC variation of all PEVs after optimization.
2692.33
(yuan)
Charging cost after
1325.93 1302.82 1331.54 1299.67
optimization (yuan) The sensitivity Vu of the neutral voltage of Bus 46 with
Initial discharging
benefits (yuan)
0 respect to the control variables at Bus 37 at time 20:45, is
Discharging benefits listed in Table III at load level 1, from which it can be seen
after optimization 22.46 24.41 23.06 20.30 that increasing the charging power at phase ‘c’ and decreasing
(yuan) the charging power at phases ‘a’ and ‘b’ will decrease the
Network loss cost
before optimization 2419.28 3092.47 2419.28 3092.47 neutral voltage of Bus 46. After one optimization iteration of
(yuan) the DAOC model, the charging power at phases a, b, and c
Network loss cost decreases from 8 kW to 0.81 kW, from 7.5 kW to 0.81 kW,
after optimization 2090.74 2732.66 2090.87 2733.16
(yuan)
and from 8 kW to 0.76 kW, respectively, and the neutral
ZLC+ZCost (yuan) 1303.51 1278.45 1308.53 1279.38 voltage of Bus 46 decreases 5.29% from 2.5161×10-3 p.u. to
Peak-valley load 2.3830 × 10-3 p.u. when the settings of all the traditional
difference before 71.38 70.41 71.38 70.41 voltage control devices do not change with the charging power
optimization (%) variation.
Peak-valley load
difference after 66.72 66.87 66.52 66.76
optimization (%) TABLE III
PART VOLTAGE SENSITIVITY OF BUS 46 AT TIME 20:45.
Total iterations 3 2 3 2
u of Node
Pcv6ina Pcv6inb Pcv6inc Pdv6 ina Pdv6 inb Pdv6 inc
37 v v v v v v

imal load decreases from 4.98843 p.u. to 4.46170 p.u. A Vu (10-4


1.19 0.22 -1.44 -1.19 -0.22 1.44
comparison of the total charging power is depicted in Fig. 7. p.u./p.u.)

0885-8950 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2019.2917009, IEEE
Transactions on Power Systems
10

C. Peak-Shifting Strategy are 4.06% and 4.03%, respectively, for load level 1. The
In this strategy, the objective function is taken as (28). The difference between them is small.
weight factor Wg is taken as 10. The optimization results for The weight Wg in the peak-shifting function is chosen by
the peak-shifting strategy considering load levels 1 and 2 are experience in order to consider the battery degradation cost of
listed in Table II. Here, the peak-valley load difference  is a PEV’s V2G service. Table IV lists the results for load level 1
under two other conditions: Wg = 103 and Wg = 105. ZCost
used to measure the load variation degree, and it is defined in
decreases with the increase of Wg. As the weight factor Wg
(37) as the proportion of the difference between the peak load
decreases from 103 to 10, the peak-valley load difference after
Ppeak and the valley load Pvalley to the peak load:
optimization decreases a little, from 66.53% to 66.52%. Also,
   Ppeak  Pvalley  Ppeak . (37) as the weight factor Wg increases from 103 to 105, the peak-
After optimization using the objective function (28) at load valley load difference after optimization decreases a little,
level 1, the highest voltage of the phase ‘b’ voltage of node 7 from 66.53% to 66.30%. This is a result of the fact that the
decreases 0.26%. Almost all of the bus voltages are limited DAOC model is nonlinear and nonconvex. However, the
within [0.9, 1.1] p.u. After optimization, the highest neutral weight value produces only a small impact on the DAOC
voltage of node 46 decreases 1.38% and the neutral voltages optimization results since the battery degradation cost ZCost is
much smaller than ZLC.
of all the low voltage buses are controlled within 2.5×10-3 p.u.
After optimization, the load variation scope becomes narrow, TABLE IV
the minimal load increases from 1.42766 p.u. to 1.48875 p.u., COMPARISON OF OPTIMAL CONTROL RESULTS USING DIFFERENT
and the maximal load decreases from 4.98843 p.u. to 4.44736 PARAMETER SETTINGS AT LOAD LEVEL 1
p.u. The load variation comparison is shown in Fig. 9. The Strategy Cost-reduction strategy Peak-shifting strategy
total charging and discharging power of each time slot is
Objective function ZLC+ZCost ZS+WgZCost
depicted in Fig. 10. PEV battery charging is also largely
Computational Heuristic
shifted to the load valley period. Even power Wg=103 Wg=105
condition power
Initial ZCost (yuan) 0
ZCost after optimization
0.00 0.19 0.05 0.04
(yuan)
Initial charging cost
1763.13 1042.91 2692.33
(yuan)
Charging cost after
1585.76 1265.11 1331.46 1340.98
optimization (yuan)
Initial discharging
0
benefits (yuan)
Discharging benefits
after optimization 0.05 81.12 23.00 21.75
(yuan)
Fig. 9. Comparison of the load variation before and after optimization using Network loss cost
the objective function (28) at load level 1. before optimization 2170.84 2039.47 2419.28
(yuan)
Network loss cost after
2158.28 2140.35 2159.42 2161.42
optimization (yuan)
ZLC+ZCost (yuan) 1585.71 1184.18 1308.51 1319.27
Peak-valley load
difference before 62.86 66.24 71.38
optimization (%)
Fig. 10. Comparison of the total charging power of the PEVs before and after Peak-valley load
optimization using the objective function (28) at load level 1. difference after 63.34 66.86 66.53 66.30
optimization (%)
After optimization at load level 2, almost all of the bus Total iterations 4 2 3 3
voltages are limited within [0.9, 1.1] p.u., and the load
variation scope becomes narrow. The smallest voltage of the D. Comparison of the Two Optimization Cases
phase ‘a’ voltage of node 34 increases 2.31% from 0.8789 p.u. As shown in Table II, the cost reduction is lower when the
to 0.8992 p.u. The highest neutral voltage of node 46 objective function is taken as (27), and μ is lower when the
decreases 2.11%. The minimal load increases from 1.65402 objective function is taken as (28). From the two objective
p.u. to 1.67466 p.u., and the maximal load decreases from functions, it seems that the first strategy benefits PEV owners,
5.58910 p.u. to 5.03906 p.u. while the second strategy benefits the Power Supply Company.
After optimization for the two load levels, all SOC values However, it can also be seen from Table II that the two
are limited within the range of [0.2, 0.9], and the SOC of all objective functions are not contradictory. The second strategy
the PEVs at the last time slot is not less than 0.85. All the can also benefit PEV users. In the second strategy, the cost
PEVs can be recharged to the desired SOC level when they reduction function is also computed and it reaches 1308.53
depart, as well. The maximum total DODs of all time slots yuan and 1279.38 yuan for load levels 1 and 2, respectively.
using the peak-shifting strategy and the cost-reduction strategy The difference between the cost reduction function values for

0885-8950 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2019.2917009, IEEE
Transactions on Power Systems
11

the two strategies is not large. The first strategy can also shown in (34) and (35). Two other distribution grids were used
benefit the Power Supply Company. The daily saved network here to check the computation time variation: the revised IEEE
loss costs are 328.54 yuan and 359.81 yuan for load levels 1 123 Node Test Feeder [35] and a 237 Node Feeder adapted
and 2, respectively, and the peak-valley load difference from the IEEE 123 Node Test Feeder. It was assumed that the
decreases to from 71.38% to 66.72% and from 70.41% to 30 PEVs listed in Table I were in each of the two distribution
66.87% after optimization for load levels 1 and 2, respectively. grids. The objective function is taken as (27). It takes up to
The difference between the peak-valley load differences after 43.5 minutes and 195.1 minutes each iteration to solve the
optimization for the two strategies is not large either. Both problem for the IEEE 123 Node Test Feeder and the 237 Node
strategies bring considerable economic benefits to both PEV Feeder, respectively. Both iterative algorithms converge in
users and the Power Supply Company. In addition, the extra two iterations. The computation time will become longer than
battery degradation cost is negligible compared to the other 24 hours when the network size is larger than 237 nodes and
costs. However, when the discharging electricity price PEV number and charging time slot number are even greater,
increases from 2.5 to 4.0 yuan/kWh, ZCost after optimization of and then the computation time could not meet the requirement
the cost-reduction strategy increases from 1.32 yuan to 1.41 of control: the optimization results should be obtained before
yuan, and the charging/discharging cost after optimization the next day. This problem could be hopefully solved by three
decreases from 289.05 yuan to -66.72 yuan with initial SOC methods including the sensitivity of voltage bands [42],
levels of all PEVs are set at 0.5 and DODmaxink = 0.5. When parallel computation [43]-[44], and sparse matrix [45]
v
techniques in future research work. First, the sensitivity of
the discharging electricity price increases further to 4.5
voltage upper and lower bounds may be useful to reduce the
yuan/kWh, ZCost decreases to 1.39 yuan, and the
computation of sensitivity Vu and the number of voltage
charging/discharging cost decreases to -122.88 yuan. If ZCost is
removed from the cost-reduction function, ZCost after constraints (26b) in the DAOC model. Second, the
optimization of the cost-reduction strategy increases from 1.32 computational tasks of the three-phase power flow and all the
yuan to 1.67 yuan when the discharging electricity price sensitivities for all time slots during optimization period can
increases from 2.5 to 4.5 yuan/kWh. The results mean that the be performed concurrently and thus multithreading techniques
benefit of including the battery degradation cost in the cost- [43] and graphics processing unit (GPU) acceleration
reduction function lies in constraining battery degradation cost approach [44] could be used to implement these distributed
and at the same time reducing charging/discharging cost. computing tasks. Furthermore since the proposed DAOC is
The combined support available from a number of PEVs in solved by the interior point method, it may be accelerated by
the network can also provide significant network support by using GPU as coprocessor similar to the GPU acceleration
reducing the SVR tap operation. The minimum and maximum approach used in [44] too. Finally, the Newton modified-
tap positions of the SVR between bus 16 and bus 26 based on augmented-nodal-analysis formulation method [45] based on
the second PEV’s charging strategy at load level 1 when Wg is sparse matrix solver could be utilized for the computation of
taken as 105 are −9 and −2, respectively. Conversely, the three-phase power flow.
minimum and maximum SVR tap positions based on the E. Comparison of Different Initializations and Comparison of
maximum charging power are −10 and −1, respectively. Using a Heuristic Charging Method
the second PEV’s charging strategy, the SVR needs an As noted in [46], noticeably different solutions could be
average of 25 tap changing operations. However, using the obtained when the optimal power flow was simply initialized
maximum PEV’s charging strategy, the SVR needs an average from different starting points. However, any differences
of 32 tap changing operations. should be within the tolerances associated with the
The computer is configured with an Intel Xeon 12-core 3.4- convergence criteria. In this section, the objective function is
GHz CPU and 16 GB of RAM. All codes are implemented in still taken as (27) with two different initialization points in
a MATLAB environment. The computation time of each order to show the differences in the DAOC convergence time
iteration of the proposed DAOC model is mainly composed of and the optimal solution. In the second initialization, the total
three parts: the three-phase power flow computation, the charging power is averaged out among the charging periods.
sensitivity computation, and the interior point optimization In the third initialization, the PEV charging power is
computation time. The MATLAB function ‘fmincon’ is used determined by the heuristic charging approach proposed in
to perform the interior point optimization [41]. There are [25], which minimizes the charging cost. Table IV lists the
N PEVnk
2   
nPEV k  a , b , c ivnk 1
N
PEV ivnk
= 2880 control variables, 34428
optimal control results using the second and third
initializations.
linear nonequivalent constraints, 990 linear equivalent From Table II and Table IV, it can be seen that the optimal
constraints, and 1440 nonlinear equivalent constraints in the objective function (27) after DAOC regulation with different
test case. As shown in Table II and Table IV, this iterative initializations is different, and the biggest relative difference is
algorithm converges quickly. The computation time of each 33.9%. The objective function after DAOC regulation with
iteration is approximately 14 min. However, the computation heuristic power initialization is the smallest. This difference
time increases with the PEV number, their results from the fact that the interior point optimization
charging/discharging time slot number, and network size as method can only find the minima near the initialization point.

0885-8950 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2019.2917009, IEEE
Transactions on Power Systems
12

Since the heuristic method is trying to achieve the minimum (36) is used to optimize the charging power of the remaining
charging cost without considering the voltage constraints, it PEVs for the time slots between 6:00 and 7:30. The SOC
can achieve the smallest charging cost, but it cannot guarantee curves of the remaining PEVs are shown in Fig. 13, and it can
the voltage level. The voltages of some buses are over the be seen that the remaining PEVs reach a fully charged state
limit using the heuristic charging power after considering the earlier. In Figs. 12 and 13, the lines marked with circles are
traditional voltage control. Using the heuristic charging at load the SOC curves after the DAOC regulation from model (26),
level 1, the phase ‘b’ voltage of node 7 is still the highest, and without any change to the arrival or departure times, and the
the highest voltage reaches 1.1022 p.u., but using the heuristic solid lines without any marks are actual SOC curves that take
charging at load level 2, the phase ‘a’ voltage of node 34 is into account the changes to the arrival and departure times.
still the smallest and the smallest voltage reaches 0.8796 p.u. The maximum iteration number of the revised optimal
After DAOC regulation, the highest voltage of the phase ‘b’ control model (36) is 5. There are 596 linear nonequivalent
voltage at node 7 at load level 1 decreases to 1.0990 p.u., the constraints in the model (36), and the optimization
smallest voltage of the phase ‘a’ voltage at node 34 increases computation time for one time slot is between 24 and 87
to 0.8917 p.u., and the degrees of all bus voltages over the seconds, which is much less than one time slot duration.
limits are relieved. Fig. 11 shows the comparison of the phase
‘a’ voltage of node 34 before and after optimization for load TABLE V
SCENARIOS OF THE ARRIVAL AND DEPARTURE TIMES
level 2. It should also be noted that the network loss cost after
Scenario Arrival time Departure time
optimization with the maximal charging power initialization is
A 18:00 9:00
the smallest, and the peak-valley load difference after B 18:00 6:00
optimization with an even power initialization is smallest. C 21:00 9:00
D 21:00 6:00

Fig. 11. Comparison of phase ‘a’ voltage at node 34 before and after
optimization at load level 2.
Fig. 12. SOC variation of the late arriving PEVs in Scenario C.
F. Discussion of Unexpected Plug-Ins or Travel
Since the arrival time and departure time of some PEVs are
obtained using the statistical expected value, the impact of the
variance of arrival and departure times on the DAOC
regulation effect is presented in the following paragraphs.
The arrival and departure times of 15 PEVs at Buses 52, 55, Fig. 13. SOC variation of the remaining PEVs in Scenario D.
59, 67, and 75 are chosen as random variables and changed by
three standard deviations from their means as shown in Table Since the arrival and departure time variation are non-
V. In model (26), the arrival and departure times are set negligible, the negative effects analyzed above would
beforehand, using the cost-reduction DAOC strategy for load deteriorate the DAOC regulation effects. Therefore, it would
level 1. When executing the obtained DAOC strategy on the be better for the PEV user to guarantee that his or her PEV is
next day, four extreme scenarios are considered in Table V. In available between the time intervals, as they are pre-set, and
Scenario A, all PEVs are available between 19:30 and 7:30, so the actual SOC initial level is not lower than the pre-set day-
they are charged/discharged according to the obtained DAOC ahead value.
strategy from model (26). In Scenario B, the minimum SOC In addition, if three PEVs plug in at Bus 37 unexpectedly,
value of the 15 early departing PEVs is 0.77 when they depart the extra optimization computation using the peak-shifting
at 6:00; the remaining PEVs reach a fully charged state earlier strategy only takes 2998.55 seconds. Therefore, these three
using the DAOC strategy obtained from model (26) between PEVs need to wait at least 2998.55 seconds, not counting
19:30 to 6:00 and the strategy obtained from model (36) communication time delay, in order to provide V2G service.
between 6:00 and 7:30. In Scenarios C and D, due to the
energy storage lower band limitation, some PEVs cannot V. CONCLUSIONS
discharge from 21:00 to 22:30 according to the DAOC An optimization model designed to flexibly control the
discharging strategy obtained from model (26). At 7:30, none available PEV battery charging/discharging power, taking into
of the late arriving PEVs reach the desired SOC level in account the voltage regulation of the residential power grid,
Scenarios C and D, and the minimal SOC level at 6:00 is 0.70. was presented in order to achieve one of two goals: (1)
Therefore, in Scenario C these late arriving PEVs are minimizing both the battery charging/discharging cost and the
optimized using model (36) for the time slots between 7:30 to extra battery degradation cost due to V2G activities (cost-
9:00. Then all these PEVs reach the desired SOC level at 8:00, reduction strategy) or (2) maximizing the local peak load
with the SOC curves depicted in Fig. 12. In Scenario D, model shifting and minimizing the extra battery degradation cost due

0885-8950 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2019.2917009, IEEE
Transactions on Power Systems
13

to V2G activities (peak-shifting strategy). The proposed [7] M.H.K. Tushar, C. Assi, M. Maier, and M.F. Uddin, “Smart microgrids:
optimal joint scheduling for electric vehicles and home appliances,” IEEE
formulation was tested on an electrical distribution network Trans. Smart Grid, vol. 5, no. 1, pp. 239–250, Jan. 2014.
with 75 nodes. The results showed that the optimized [8] X. Wang, Y. Pi, and A. Tang, “Scheduling of electric vehicle charging via
operation of the PEV charge or discharge produced an overall multi-server fair queueing,” IEEE Trans. Parallel Distrib. Syst., vol. 28,
no. 11, pp. 3298-3312. Nov. 2017
battery charging/discharging cost reduction with the use of the
[9] A. Arias, M. Granada, and C. A. Castro, “Optimal probabilistic charging
cost-reduction strategy when compared with maximum power of electric vehicles in distribution systems,” IET Electr. Syst. Transp., vol.
charging or even power charging. The optimized operation 7, no. 3, pp. 246-251, May 2017.
also produced a peak load reduction with the use of the peak- [10] K. Clement-Nyns, E. Haesen, and J. Driesen, “The impact of charging
Plug-In hybrid electric vehicles on a residential distribution grid,” IEEE
shifting strategy, and the degrees of all bus voltages over the Trans. Power Syst., vol. 25, no. 1, pp. 371-380, Feb. 2010.
limits were relieved by using both strategies. The results also [11] G. Wang, J. Zhao, F. Wen, Y. Xue, and G. Ledwich, “Dispatch strategy
showed that the neutral-to-ground potential rise of the of PHEVs to mitigate selected patterns of seasonally varying outputs from
renewable generation,” IEEE Trans. Smart Grid, vol. 6, no. 2, pp. 627–
distribution network was alleviated after the optimal control of 639, Mar. 2015.
the PEV charge. In fact, it depends on the non-PEV load level, [12] F. Wu and R. Sioshansi, “A two-stage stochastic optimization model for
the controllable PEVs’ charging/discharging power range, and scheduling electric vehicle charging loads to relieve distribution-system
constraints,” Transp. Res. Pt. B-Methodol., Vol 102, pp 55-82, Aug. 2017.
reactive power reserve whether the proposed DAOC can keep [13] J. Tan and L. Wang, “Integration of plug-in hybrid electric vehicles into
all bus voltages within acceptable limits. If an error in the residential distribution grid based on two-layer intelligent optimization,”
forecasting of the daily load profiles is considered, the control IEEE Trans. Smart Grid, vol. 5, no. 4, pp. 1774-1784, Jul. 2014.
[14] W. Tang and Y. J. Zhang, “A model predictive control approach for low-
effect of the proposed DAOC may degrade. This may be complexity electric vehicle charging scheduling: Optimality and
addressed by stochastic programming in the future work. scalability.” IEEE Trans. Power Syst., vol. 32, no. 2, pp. 1050–1063, Mar.
Moreover, the results demonstrated that according to the 2017.
[15] L. Liu, T. Liu, T. Zhang, and J. Liu, “Orderly charging and discharging
current battery cost and electricity price in China, the extra
strategy optimization for electric vehicles considering dynamic battery-
battery degradation cost due to V2G was negligible compared wear model,” Automation Electr. Power Syst., vol. 40, no. 5, pp. 83-90,
to the saved cost of all PEVs and the saved network loss cost Mar. 2016.
after DAOC charging instead of maximum power charging or [16] T. Shekari, S. Golshannavaz, and F. Aminifar, “Techno-economic
collaboration of PEV fleets in energy management of microgrids,” IEEE
even power charging, which would motivate PEV users to Trans. Power Syst., vol 32, no 5, pp. 3833-3841, Sep. 2017.
participate in V2G service. Since the conventional voltage [17] C. Sabillon-Antunez, O. D. Melgar-Dominguez, J. F. Franco, M.
control and the proposed DAOC strategy can work Lavorato, and M. J. Rider, “Volt-var control and energy storage device
operation to improve the electric vehicle charging coordination in
simultaneously, the proposed DAOC strategy needs less unbalanced distribution networks,” IEEE Trans. Sustain. Energy, vol 8,
infrastructure investment. no. 4, pp. 1560-1570, Oct. 2017.
The DAOC model was developed further in order to deal [18] X. Wang and Q. Liang, “Energy management strategy for plug-in hybrid
electric vehicles via bidirectional vehicle-to-grid,” IEEE Syst. J., vol. 11,
with PEVs being unexpectedly plugged in or travelling and to no. 3, pp. 1789–1798, Sep. 2017.
address any SOC deviation from the day-ahead pre-set value, [19] A. Gusrialdi, Z. Qu, and M. A. Simaan, “Distributed scheduling and
and the computation time on the test electrical distribution cooperative control for charging of electric vehicles at highway service
stations,” IEEE Trans. Intell. Transp. Syst., vol. 18, no.10, pp. 2713-2727,
network indicated its feasibility. Oct. 2017.
ACKNOWLEDGEMENT [20] M. J. E. Alam, K. M. Muttaqi, and D. Sutanto. “A controllable local peak-
The author would like to express special thanks to the shaving strategy for effective utilization of PEV battery capacity for
distribution network support,” IEEE Trans. Ind. Appl., vol. 51, no. 3, pp.
editor and reviewers whose comments led to valuable
2030-2037, May 2015.
improvements of the manuscript. [21] W.J. Ma, V. Gupta, and U. Topcu, “Distributed charging control of
electric vehicles using online learning,” IEEE Trans. Autom. Control, vol.
VI. REFERENCES 62, pp. 10, pp. 5289-5295, Oct. 2017.
[22] A. Malhotra, G. Binetti, A. Davoudi, and I. D.Schizas, “Distributed
[1] S. Shafiee, M. Fotuhi-Firuzabad, and M. Rastegar, “Investigating the power profile tracking for heterogeneous charging of electric vehicles,”
impacts of plug-in hybrid electric vehicles on power distribution systems,” IEEE Trans. Smart Grid, vol. 8, no. 5, pp. 2090-2099, Sep. 2017.
IEEE Trans. Smart Grid, vol. 4, no. 3, pp. 1351–1360, Sep. 2013. [23] L. Gan, U. Topcu, and S. Low, “Optimal decentralized protocol for
[2] L. P. Fernández, T. G. S. Román, R. Cossent, C. M. Domingo, and P. electric vehicle charging,” IEEE Trans. Power Syst., vol. 28, no. 2, pp.
Frías, “Assessment of the impact of plug-in electric vehicles on 940–951, May 2013.
distribution networks,” IEEE Trans. Power Syst., vol. 26, no. 1, pp. 206– [24] G. Binetti, A. Davoudi, D. Naso, B. Turchiano, and F. L. Lewis,
213, Feb. 2011. “Scalable real-time electric vehicles charging with discrete charging rates,”
[3] Q. Gong, S. Midlam-Mohler, V. Marano, and G. Rizzoni, “Study of PEV IEEE Trans. Smart Grid, vol. 6, no. 5, pp. 2211–2220, Sep. 2015.
charging on residential distribution transformer life,” IEEE Trans. Smart [25] Y. Cao, S. Tang, C. Li, , P. Zhang, Y. Tan, Z. Zhang, and J. Li, “An
Grid, vol. 3, no. 1, pp. 404–412, Mar. 2012. optimized EV charging model considering TOU price and SOC curve,”
[4] E. Veldman and R. A. Verzijlbergh, “Distribution grid impacts of smart IEEE Trans. Smart Grid, vol. 3, no. 1, pp. 388–393, Feb. 2012.
electric vehicle charging from different perspectives,” IEEE Trans. Power [26] Z. Ma, S. Zou, L. Ran, X. Shi, I. A. Hiskens, “Efficient decentralized
Syst., vol. 6, no. 1, pp. 333–342, Jan. 2015. coordination of large-scale plug-in electric vehicle charging,” Automatica,
[5] M. Amjad, A. Ahmad, M. H. Rehmani, and T. Umer, “A review of EVs vol 69, pp. 35-37, Jul. 2016.
charging: From the perspective of energy optimization, optimization [27] Y. Xu, “Optimal distributed charging rate control of plug-in electric
approaches, and charging techniques,” Transport. Res. Part D, vol 62, no. vehicles for demand management,” IEEE Trans. Power Syst., vol. 30, no.
1, pp. 386–417, Jul. 2018. 3, pp. 1536-1545, May 2015.
[6] D.T. Nguyen and L.B. Le, “Joint optimization of electric vehicle and [28] A. Cortés and S. Martínez, “A hierarchical algorithm for optimal plug-in
home energy scheduling considering user comfort preference,” IEEE electric vehicle charging with usage constraints,” Automatica, vol 68, pp.
Trans. Smart Grid, vol. 5, no. 1, pp. 188–199, Jan. 2014. 119-131, Jun. 2016.
[29] M. J. E. Alam, K. M. Muttaqi, and D. Sutanto, “Effective utilization of
available PEV battery capacity for mitigation of solar PV impact and grid

0885-8950 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2019.2917009, IEEE
Transactions on Power Systems
14

support with integrated V2G functionality,” IEEE Trans. Smart Grid, vol.
7, no. 3, pp. 1562-1571, May 2016.
[30] B. Feng, H. Lin, and X. Wang, “Modulation and control of AC/DC matrix
converter for battery energy storage application,” IET Power Electron.,
vol. 8, no. 9, pp. 1583-1594, Mar. 2015.
[31] M. J. E. Alam, K. M. Muttaqi, and D. Sutanto, “A three-phase power flow
approach for integrated 3-wire MV and 4-wire multigrounded LV
networks with rooftop solar PV,” IEEE Trans. Power Syst., vol. 28, no. 2,
pp. 1728-1737, May 2013.
[32] W. Su, H. Eichi, W. Zeng, and M. Chow, “A survey on the electrification
of transportation in a smart grid environment,” IEEE Trans. Ind. Inform.,
vol. 8, no. 1, pp. 1–10, Feb. 2012.
[33] A. Giusti, M. Salani, G. A. D. Caro, A. E. Rizzoli, and L. M.
Gambardella, “Restricted neighborhood communication improves
decentralized demand-side load management. ,” IEEE Trans. Smart Grid,
vol. 5, no. 1, pp. 92–101, Jan. 2014.
[34] W. H. Kersting, “Distribution system modeling and analysis,” 1st ed.
New York: CRC press, 2002, cha.6-8, pp. 125-245.
[35] W. H. Kersting, “Radial distribution test feeders,” in Proc. IEEE Power
Eng. Soc. Winter Meeting, Jan. 2001, vol. 2, pp. 908-912.
[36] H. Hino, H. Shen, N. Murata, S. Wakao, and Y. Hayashi, “A versatile
clustering method for electricity consumption pattern analysis in
households,” IEEE Trans. Smart Grid, vol. 4, no. 2, pp. 1048-1057, Jun.
2013.
[37] BYD model specifications. Jan. 2017. [Online]. Available:
http://car.diandong.com/duibi/chexi/208/?niankuan=2016
[38] BAIC model specifications. Nov. 2017. [Online]. Available:
http://car.diandong.com/duibi/chexi/352/?niankuan=2017
[39] Y. Ghiassi-Farrokhfal, S. Keshav and C. Rosenberg. “Toward a realistic
performance analysis of storage systems in smart grids.” IEEE Trans.
Smart Grid, vol. 6, no. 1, pp. 402–410, Jan. 2015.
[40] National Household Travel Survey 2009. The Federal Highway
Administration, USA, [Online]. Available: http://nhts.ornl.gov/
[41] Mathworks, “Optimization toolbox user’s guide,” The Mathworks Inc.,
2016.
[42] K. Jhala, B. Natarajan, and A. Pahwa, “Probabilistic voltage sensitivity
analysis (PVSA)—A novel approach to quantify impact of active
consumers,” IEEE Trans. Power Syst., vol. 33, no. 3, pp. 2518-2527, May
2018.
[43] G. Geng and Q. Jiang, “A two-level parallel decomposition approach for
transient stability constrained optimal power flow,” IEEE Trans. Power
Syst., vol. 27, no. 4, pp. 2063-2073, Nov. 2012.
[44] G. Geng, Q. Jiang, and Y. Sun. “Parallel transient stability-constrained
optimal power flow using GPU as coprocessor,” IEEE Trans. Smart Grid,
vol. 8, no. 3, pp. 1436-1445, May 2017.
[45] I. Kocar, J. Mahseredjian, U. Karaagac, G. Soykan, and O. Saad.
“Multiphase load-flow solution for large-scale distribution systems using
MANA,” IEEE Trans. Power Del., vol. 29, no. 2, pp. 908-915, Apr. 2014.
[46] A. D. Papalexopoulos, C. F. Imparato, and F. F. Wu, “Large-scale
optimal power flow: Effects of initialization, decoupling and
discretization,” IEEE Trans. Power Syst., vol. 4, pp. 748–759, May 1989.

Yulong Huang received the Ph.D. degree in


electrical engineering from the South China
University of Technology, Guangzhou, China, in
2012. In the 2010–2011 academic year, he was a
visiting scholar at the University of Waterloo. Since
September 2012, he has been with the College of
Electrical Engineering at Jinan University, Zhuhai,
China. His research interests mainly include power
system optimization, operation, and control.

0885-8950 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

You might also like