Energy Conversion and Management 128 (2016) 160–177

Contents lists available at ScienceDirect

Energy Conversion and Management

journal homepage: www.elsevier.com/locate/enconman

Multi-objective stochastic scheduling optimization model for connecting

a virtual power plant to wind-photovoltaic-electric vehicles considering
uncertainties and demand response
Liwei Ju ⇑, Huanhuan Li, Junwei Zhao, Kangting Chen, Qingkun Tan, Zhongfu Tan
North China Electric Power University, Beijing 102206, China

a r t i c l e i n f o a b s t r a c t

Article history: A stochastic chance-constrained planning method is applied to build a multi-objective optimization
Received 5 May 2016 model for virtual power plant scheduling. Firstly, the implementation cost of demand response is calcu-
Received in revised form 17 September lated using the system income difference. Secondly, a wind power plant, photovoltaic power, an electric
2016
vehicle group and a conventional power plant are aggregated into a virtual power plant. A stochastic
Accepted 23 September 2016
scheduling model is proposed for the virtual power plant, considering uncertainties under three objective
functions. Thirdly, a three-stage hybrid intelligent solution algorithm is proposed, featuring the particle
swarm optimization algorithm, the entropy weight method and the fuzzy satisfaction theory. Finally, the
Keywords:
Virtual power plant
Yunnan distributed power demonstration project in China is utilized for example analysis. Simulation
Demand response results demonstrate that when considering uncertainties, the system will reduce the grid connection
Electric vehicle group of the wind power plant and photovoltaic power to decrease the power shortage punishment cost. The
Multi-objective average reduction of the system power shortage punishment cost and the operation revenue of virtual
Uncertainties power plant are 61.5% and 1.76%, respectively, while the average increase of the system abandoned
energy cost is 40.4%. The output of the virtual power plant exhibits a reverse distribution with the con-
fidence degree of the uncertainty variable. The proposed algorithm rapidly calculates a global optimal set.
The electric vehicle group could provide spinning reserve to ensure stability of the output of the virtual
power plant. Demand response could optimize customers’ power consumption behaviors and improve
the grid connection space of the virtual power plant. The electric vehicle group and demand response
could achieve a linkage optimization effect between the generation side and demand side, achieving opti-
mal system scheduling objectives.
Ó 2016 Elsevier Ltd. All rights reserved.

1. Introduction distributed generators (DGs), energy storage systems and control-

lable loads [4]. In particular, the rapid development of smart grid
Due to traditional energy shortage and environmental pollution technology promotes reasonable resource configuration and pro-
problems, renewable energy power generation has attracted vides solid support for VPP operation.
increased attention. Distributed energy resources (DERs), espe- In recent years, many countries have performed VPP pilot pro-
cially wind power and solar photovoltaic power, are playing jects. In 2007, the University of Kassel formed the largest VPP pilot
increasingly important roles in energy structures [1]. However, project by integrating a wind turbine, a solar energy system, a bio-
limited by the characteristics of small capacity, large quantity gas power plant and a hydro power plant [5]. In 2009, a smart grid
and geographical dispersion, the grid connection cost of DERs is project connecting electric vehicles (EVs) in Denmark took the
relatively high. DER grid connection could also affect the stability uncertainty of large-scale wind power output into consideration
of system operation [2]. Thus, the virtual power plant (VPP) is pro- and introduced VPP technology to manage the EVs’ charging-
posed as a new technology for DERs in the power market [3]. A VPP discharging power [6]. In 2012, the Rhineland Group in Germany
can effectively integrate, aggregate and manage DERs, including began operating the first commercial VPP [7]. In 2008, a distributed
energy station in Guangdong University City, China began opera-
tion to meet the load demand of the University City [8]. In 2011,
⇑ Corresponding author at: North China Electric Power University, Chang Ping the Zhangbei wind-photovoltaic-storage-transmission project
District, Beijing 102206, China. began operation in China, integrating a wind power plant (WPP),
E-mail address: [email protected] (L. Ju).

http://dx.doi.org/10.1016/j.enconman.2016.09.072
0196-8904/Ó 2016 Elsevier Ltd. All rights reserved.
 L. Ju et al. / Energy Conversion and Management 128 (2016) 160–177 161

Nomenclature
L0s load demand before PBDR
Abbreviations DP t electricity price changes after PBDR at time t
VPP virtual power plant d high price compensation rate
DR demand response s discount price rate
PBDR price-based DR u shape factor
IBDR incentive-based DR # scale factor
WPP wind power plant v w;t real-time wind velocity at time t
PV photovoltaic generators gR WPP rated output
CPP conventional power plant v in cut-in wind velocity
EVG plug-in electric vehicle group b6 confidence levels for upstream reserve constraint
GC generating companies b7 confidence levels for downstream reserve constraint
ISO independent system operator r1 reserve coefficient of the load
DERs distributed energy resource r2 reserve coefficient of the WPP
PDF probability density function r3 reserve coefficient of the PV
r4 reserve coefficient of the EV
Set
DLmaxt maximum curtailment load at time t
s, t index for time v out cut-out wind velocity
W index for WPP v rated rated wind velocity
PV index for PV w; f shape parameters of the Beta distribution
g max maximum PV output power
EV index for EV PV

CPP index for CPP lPV average value of the PV output

i, m index for objective functions rPV standard value of the PV output
j index for stochastic constraints f ðg PV;t Þ the PDF of the solar radiation intensity
Ndis
EV;t discharging electric vehicle number at time t
Variable Nchr
EV;t charging electric vehicle number at time t
DLs load demand changes after PBDR at time s ai confidence level of the i-th objective function
Lt load demand after PBDR at time t n stochastic vector
C Pt system power sale income change before and after the x n-dimensional decision vector
PBDR Prfg probability of event occurring in fg
C It system incremental cost of implementing IBDR bj confidence level of constraint g j ðx; nÞ 6 0
DLIt load change amount at time t a1 confidence level of f 1
DLI;u upstream reserve load at time t
qS unit power shortage compensation price
t
a2 confidence level of f 2
DLI;dt downstream reserve load at time t
qW grid connection power prices of the WPP
sI;d
t ; st
I;u
0–1 variable, 1 means reserve load is provided qPV grid connection power prices of the PV
v wind velocity g W actual available outputs of the WPP
g w;t WPP output at time t g PV actual available outputs of PV
g PV;t PV output at time t a3 confidence level of f 3
g EV;t EVG output at time t aCPP ; bCPP ; cCPP cost coefficients of CPP power generation
g dis
EV;t discharging power of the EVG at time t Ncold cold-startup cost of the CPP
CPP
g chr charging power of the EVG at time t
EV;t Nhot hot-startup of the CPP

f minimum value of f i ðx; nÞ
CPP
i
T min
CPP minimum allowed downtime of the CPP
f1 objective function of the minimum system compensa-
tion cost T off
CPP;t continuous downtime of the CPP at time t
DLS;t power shortage load at time t T cold
CPP cold-startup time of the CPP
f2 objective function of the minimum VPP abandoned uW power consumption rate of the WPP
power uPV power consumption rate of the PV
f3 objective function of the maximum VPP operation in- uCPP power consumption rate of the CPP
come g GC;t electricity purchasing from GC at time t
pW;t generation incomes of the WPP at time t b1 confidence level of demand and supply balance con-
pPV;t generation incomes of the PV at time t straint
pEV;t generation incomes of the EV at time t g max
CPP upper limitation of the CPP output
pCPP;t generation incomes of the CPP at time t g min lower limitation of the CPP output
CPP
DCPP;t startup-shutdown cost of the CPP at time t
Dg þCPP upper climbing limitation of the CPP output
g CPP;t power output of the CPP at time t
Dg  lower climbing limitation of the CPP output
GVPP;max
t maximum output power of the VPP CPP
M on shortest startup time of the CPP
RVPP
t installed capacity for reserve demand of the VPP at time CPP

t T on
CPP;t1 continuous operation time of the CPP at time t1
di information deviation of
f i M off
CPP shortest downtime of the CPP
ki weight coefficient of
f i T off continuous downtime of the CPP at time t1
CPP;t1
uCPP;t status variable of the CPP at time t; 1 means the CPP is
Parameter in operation, and 0 means the CPP is not in operation
est demand-price elasticity Qt storage energy of the EVG at time t
P0t electricity price before PBDR qEV loss rate of charging-discharging power
162 L. Ju et al. / Energy Conversion and Management 128 (2016) 160–177

g
chr
EV;t maximum charging power of the EVG at time t T P;off minimum load downtime time of PBDR
g
dis maximum discharging power of the EVG at time t X I;on
t1 continuous operation time of IBDR
EV;t
b2 confidence levels of the EVG charging power constraint X I;off
t continuous downtime of IBDR
b3 confidence levels of the EVG discharging power con- T I;on minimum load operation time of IBDR
straint T I;off minimum load downtime time of IBDR
DLmax
t maximum curtailment load at time t SI;max upper limitation of the IBDR demand response time
b4 ; b5 confidence levels of the load curtailment constraint DLIt virtual output produced by IBDR at time t
LPD lower climbing limitation of the load produced by PBDR DLPt virtual output produced by PBDR at time t
LPU upper climbing limitation of the load produced by PBDR DLPt maximum output of PBDR at time t
LID lower climbing limitation of the load produced by IBDR DLIt maximum output of IBDR at time t
Q0 storage energy of the EVG at the initial time sPt status variable of PBDR; 1 means PBDR is scheduled, and
X P;on
t1 continuous operation time of PBDR 0 means PBDR is not scheduled
X P;off
t continuous downtime of PBDR sIt status variable of IBDR; 1 means PBDR is scheduled, and
I
LU upper climbing limitation of the load produced by IBDR 0 means PBDR is not scheduled
T P;on minimum load operation time of PBDR

photovoltaic power (PV), an energy storage system and power Yang et al. [25] introduced a robust optimization method to opti-
transmission [9]. In 2014 in Xiaozhongdian, China, a wind- mize the scheduling scheme by adjusting the robust coefficient
photovoltaic-hydro distributed power demonstration project suc- to obtain a scheme set. Braslavsky et al. [26] presented an optimal
cessfully connected the China Southern Power Grid and began scheduling model and construct a simulation method with five
commercial operation [10]. classical scenarios for uncertainty factors. Morteza et al. [27] pre-
Research regarding VPP generally begins from a VPP itself and sented a robust optimization-based decision-making tool for self-
focuses on how to optimize VPP operation based on capacity con- scheduling of a VPP in the uncertain environment of electricity
figuration and operation coordination. Spyros et al. [11] defined markets. The proposed methods and models may be effective for
the concept and control mode of a VPP. Moghaddam et al. [12] overcoming the uncertainty influence, but they can only handle
summarized the essential status quo and development prospect the constraint conditions and do not address objective functions
of domestic and foreign VPPs. Omid et al. [13] described the prin- with stochastic variables. Then, with the rapid development of
ciples of a VPP considering the operational concepts and require- the smart grid, customers can actively respond to system schedul-
ments from participation in active network management. ing. The research regarding how to cooperate with VPP operation
Compared with micro-grids, VPPs particularly emphasize top- by utilizing customers should be further developed.
down control to achieve controllable output through multi- Two-way interactive technology could provide a technological
resource, multi-time and multi-space coordinate scheduling [14]. foundation for demand-side participation in optimal scheduling
Then, mathematical models are constructed for VPP scheduling. of the generation side. Customer guidance measures could opti-
Tascikaraoglu et al. [15] evaluated the virtual power plant strategy, mize customer behavior to respond to system power generation
including a hybrid system consisting of wind, solar, hydrogen and scheduling. Demand response (DR) refers to the market-oriented
thermal power systems. Shayeghi et al. [16] presented the optimal consumption behavior in electricity grids. DR can be classified into
integrated participation model of wind and PV energy, including price-based demand responses (PBDRs) and incentive-based
storage devices and dispatchable distributed generations to demand responses (IBDRs) [20]. Dimopoulos et al. [16] analyzed
increase their revenues. Riveros et al. [17] presented a methodol- the economic benefits of small-scale generation and electric vehi-
ogy to evaluate the optimal bidding strategy of a VPP composed cles under different demand response strategies. Bracco et al.
of a combined heat and power system coupled with district heat- [28] designed a dynamic optimization-based architecture for poly-
ing and renewable-energy sources. The above works from the liter- generation microgrids with tri-generation, storage system demand
ature study the basic concept, operation optimization and response and electrical vehicles. DR could result in load peak-
scheduling model of a VPP but fail to analyze the uncertainty influ- valley shifting and reduce the uncertainty caused by renewable
ence of the WPP, PV and other factors on the system’s stable energy grid connections [29]. In addition, DR plays an increasingly
operation. important role in balancing short-term supply and load demand
Some studies have focused on how to overcome the fluctuation and dealing with fluctuations in renewable energy supplies. How-
problem of VPP output. Perfumo et al. [18] developed a mathemat- ever, the application of the DR model has not been sufficiently dis-
ical model aimed at the design of a model-based feedback control cussed in previous studies of VPP management.
strategy. Zhang et al. [19] studied the optimal scheduling model of Previous studies have already studied the optimization of VPP
a VPP considering smart homes’ energy consumption. Controllable operation and achieved favorable results. However, various insuffi-
units, energy storage systems and controllable loads often provide ciencies still exist. Firstly, some studies took the uncertainty of the
backup service for VPP power generation [20]. Shafie-khah et al. WPP and PV into consideration and analyzed the uncertainty influ-
[21] proposed a market-based VPP model to allow individual DER ence on VPP operation. The robust optimization method is the
units to access the current electricity markets. A VPP can optimally main tool to analyze the uncertainties of the WPP and PV. How-
operate the generation units, including a wind turbine, a solar unit, ever, the method could only handle the constraint conditions with-
a fuel cell and a storage battery [22]. Then, to describe the output out discussing the objective functions with stochastic variables.
uncertainty of the WPP and PV, Peik-Herfeh et al. [23] used a two- The paper uses a stochastic chance-constrained planning method
point assessment method to obtain two values on each side of the to overcome the uncertainty factors, which could handle the
predicted value to present the output fluctuation of the unit. Yang stochastic variables in the constraint conditions and objective
et al. [24] assumed that wind velocity obeys the Weibull distribu- functions. Secondly, DR could optimize customers’ power con-
tion and used the maximum likelihood estimation to obtain the sumption behavior. Customers will increase power consumption
distribution parameters and calculate the wind turbine’s output. in valley periods and decrease power consumption in peak periods.
 L. Ju et al. / Energy Conversion and Management 128 (2016) 160–177 163

In particular, when power supply cannot meet power demand, cus- If the electricity price changes in diverse periods, customers can
tomers can reduce power consumption. The curtailed load can be respond in two ways. In the first way, they can only be on or off in
regarded as virtual power output. The virtual power output can the case that some loads are not able to move from one period to
provide reserve capacity for the WPP and PV, which could ensure another (e.g., illuminating loads). Such loads have sensitivity in a
stable VPP output and optimize VPP scheduling results. However, single period called ‘‘self-elasticity,” which always has a negative
the previous studies rarely discuss the influence of demand value [30]. In the second way, some consumption can be trans-
response on VPPs, especially PBDR. The paper integrates IBDR into ferred from the peak period to off-peak or low periods (e.g., process
VPP components and discusses the influence of PBDR on VPP oper- loads). Such behavior is called multi-period sensitivity and is eval-
ation. Thirdly, because VPP operation has plural requirements, the uated by ‘‘cross-elasticity,” which is always positive [31]. The
VPP scheduling model has multiple objectives. A reasonable math- detailed mathematical description is as follows:
ematical model and solution algorithm are important for optimal
operation of a VPP. The contributions of this study are as follows: (1) If s = t, est is called self-elasticity. Only demand curtailment
can occur in this method. Therefore, est is always negative,
 A multi-objective optimization model for VPP stochastic and the load change is always negative.
scheduling considering uncertainties and DR is proposed. The (2) If s – t, est is called cross-elasticity. Demand shifting can
model considers the uncertain factors of the DR, WPP, PV and occur in this method. est is always positive, and the load
EVG. The minimum system compensation cost, the minimum change is always positive.
abandoned power cost and the maximum VPP operation
income are taken as the optimization objective functions. According to the above analysis, both demand curtailment and
 A stochastic chance-constrained method is introduced to over- demand shift can occur in the PBDR. The load change after the
come the uncertainties of the DR, WPP, PV and EVG. The study PBDR previously calculated in literature [32] does not discuss the
focuses on analyzing the risk level and choosing the optimal detailed calculation method. The load change Lt after PBDR is given
confidence degree of VPP scheduling. Spinning reserve con- by Eq. (2).
straints are satisfied with certain confidence degrees to discuss 8 9
>
> >
>
the risk level of VPP scheduling under different confidence >
> >
>
< ½Pt  P0t  X
24
½Ps  P0s =
degrees and risk coefficients. 0
Lt ¼ Lt  1 þ ett  þ e st  ð2Þ
 The particle swarm optimization algorithm, entropy weight >
> P0t P0s > >
>
> s¼1 >
>
method and fuzzy satisfaction theory are applied to construct : ;
a three-stage hybrid intelligent solution algorithm. Cloud evolu-
s–t
tionary thought and stochastic technology are used to evolve an Combining Eqs. (1) and (2), the system power sale income
excellent particle swarm. The entropy method can solve the sin- change before and after PBDR can be calculated by Eq. (3):
gle objective weight to form a comprehensive optimization
model. The fuzzy satisfaction theory can select the optimal C Pt ¼ P0t L0t  ðP0t þ DPt ÞLt ð3Þ
decision. where in the first term and the second term are the system power
sale income before and after PBDR, respectively.
The remainder of this paper is organized as follows. Section 2
presents the demand response model, including a PBDR model 2.2. Incentive-based DR model
and IBDR model. Section 3 constructs a multi-objective optimiza-
tion model for VPP stochastic scheduling, including model assump- IBDR utilizes incentive policies to encourage customers to par-
tions, uncertainty analyses and model formulation. Section 4 ticipate in market adjustment, which could maintain system sup-
proposes a three-stage hybrid intelligent solution algorithm based ply and demand balance [33]. Customers can participate in
on the particle swarm optimization algorithm, the entropy method system upstream/downstream reserve [32]. The load curtailment
and the fuzzy satisfaction theory. China’s Yunnan distributed and incremental cost of implementing IBDR can be calculated by
energy demonstration project is taken as a case study in Section 5. Eqs. (4) and (5).
Section 6 presents the study’s primary conclusions.
C It ¼ sI;d I;d I;u
t dP t DLt þ st ð1  sÞP t DLt
I;u
ð4Þ
2. Demand response model
DLIt ¼ sI;d I;d I;u
t DLt þ st DLt
I;u
ð5Þ
Demand response mainly includes price-based demand Eq. (4) calculates the system incremental cost of implementing
response (PBDR) and incentive-based demand response (IBDR). IBDR, including upstream reserve cost and downstream reserve
DR can optimize customers’ power consumption behavior to par-
cost. Eq. (5) ensures that sI;d I;u
t and st cannot be 1 simultaneously.
ticipate in the dynamic scheduling of VPP operation.

2.1. Price-based DR model 3. VPP stochastic scheduling optimization model

PBDR can optimize customers’ power consumption behavior This section constructs a stochastic optimization model of VPP
and transfer power consumption period by implementing a time- scheduling. Firstly, the basic structure of a VPP is proposed. Sec-
of-use (TOU) price. The influence of PBDR on customers’ behavior ondly, the uncertainty factors of VPP operation are analyzed, which
can be described by power demand-price elasticity in Eq. (1) [10]. includes VPP uncertainty and DR uncertainty. Thirdly, the dynamic
scheduling model for VPP operation is proposed based on the


DLs =L0s est < 0; if s ¼ t stochastic chance constraint planning theory.
est ¼ ð1Þ
DPt =P0t est P 0; if s – t
3.1. Proposed model
Eq. (1) describes the relationship between load demand and elec-
tricity price. When P 0t , DPt , L0s and est are known, DLs can be In this study, the VPP consists of a WPP, PV, EVG and CPP. A
obtained. day-ahead market structure is used in Fig. 1 [34]. An independent
164 L. Ju et al. / Energy Conversion and Management 128 (2016) 160–177

Incentive-based DR
Price-based DR
(up reserve, down reserve)

change

Reserve
service

signals
signals
Load

Price
Price
Generating companies

Bids Minimization Minimization Maximization VPP

compensation cost abandoned energy cost operation revennue

TRANSCOs
(GC)

ISO
(Day-ahead market)
Schedule

Multi-objective Stochastic Scheduling Optimization

Schedule
Bids
Virtual Power Plant(VPP)
Wind power plant Photovoltaic Conventional gas Electric Vehicle
(WPP) power (PV) turbine (CPP) Group (EVG)

Fig. 1. Market structure with a VPP.

system operator (ISO) receives bids from generating companies 3.2.1. VPP uncertainty
(GC) and the VPP to provide energy. Then, according to customers’ The uncertainty factors of VPP operation are primarily derived
power demand, the ISO sends signals to customers corresponding from the uncertain output of the WPP, PV and EVG. The EVG’s
to the power price adjustment range and reserve price. Customers uncertainty is primarily influenced by customer subjectivity.
change their load demand after receiving price signals. Finally, the
ISO simultaneously schedules energy and reserves in the power (1) WPP output uncertainty
system considering system constraints and different objective
functions. WPP output uncertainty depends on the stochastic characteris-
The WPP and PV are the primary power resources to meet the tic of wind velocity. Though wind velocity appears stochastic in
load demand; the EVG and CPP could provide reserve services for both the short term and long term, large statistical datasets have
power recourses with uncertainty output. The EVG must be regis- shown that wind velocity can be approximately described by the
tered in the VPP central controller before the scheduling period Weibull distribution [32] based on the probability density function
[10]. The VPP central controller can issue orders to the EVG accord- (PDF) in Eq. (6):
ing to the real-time power output state of the WPP and PV by con-
u v u1
f ðv Þ ¼ eðv =#Þ
u
trolling its charging-discharging behavior. Similar to an energy ð6Þ
storage system, the EVG could provide a reserve service via the # #
charging-discharging power. Furthermore, the EVG could also con-
Eq. (6) can be used to gain the average value and variance of the
nect the demand side and generation side to achieve a linked opti-
wind velocity for the corresponding period. After obtaining u and
mization effect. Therefore, this study introduces an EVG instead of
#, the wind velocity can be simulated using MATLAB tools. The rela-
an energy storage system.
tionship between WPP output and real-time wind velocity can be
PBDR and IBDR are both discussed in this study. PBDR optimizes
described by Eq. (7).
customers’ behaviors to optimize VPP operation. IBDR guides cus-
8
tomers to provide reserve services by providing upstream/down- >
> 0; 0 6 v t < v in ; v t > v out
stream spinning reserve price signals. Because VPP scheduling < v 3 v 3
g w;t ðv w;t Þ ¼ t in
v 3rated v 3in gR ; v in 6 v t 6 v rated ð7Þ
depends on perfect communication infrastructure, this study >
>
:
assumes that a smart grid and its communication infrastructure R; v rated 6 v t 6 v out
have already been in operation [35].

(2) PV output uncertainty

PV output is closely related with solar radiation intensity. The
solar radiation intensity is a stochastic variable that obeys the Beta
3.2. Uncertainty analysis
distribution [32]; therefore, PV output should also obey the Beta
distribution. The PDF is given by Eq. (8).
Multiple uncertainties exist in VPP scheduling models. DR cus-
tomers are quite subjective and change their load demands with  w1  
u g PV;t g PV;t f1
respect to electricity prices. The WPP and PV are uncontrollable f ðg PV;t Þ ¼ 1  max ð8Þ
bðw; fÞ g max
PV g PV
power resources with uncertain power output.
 L. Ju et al. / Energy Conversion and Management 128 (2016) 160–177 165

CðwÞCðfÞ ance to ensure a rational decision. The probability of the decision

bðw; fÞ ¼ ð9Þ
CðwÞ þ CðfÞ meeting the constraints should not be lower than the given confi-
dence level. The multiple-objective chance constraint-planning
wherein b(w, f) is the Beta distribution function. w and f can be cal-
models can be described as:
culated by Eqs. (10) and (11).
8
 



lPV ð1  lPV Þ < min½f 1 ; f 2 ; . . . ; f m 
>
w ¼ lPV 1 ð10Þ s:t: Prðf i ðx; nÞ 6
f i ; i ¼ 1; 2; . . . ; mÞ P ai ð14Þ
r2PV >
:
Prðg j ðx; nÞ 6 0; j ¼ 1; 2; . . . ; mÞ P bj
 
lPV ð1  lPV Þ
f ¼ ð1  lPV ÞlPV 1 ð11Þ
r2PV 3.3.2. VPP stochastic scheduling model
After obtaining the PDF of the local solar radiation intensity DR can smooth the load demand curve and provide more grid
f ðg PV;t Þ, the PV output can be simulated. connection space for the VPP, but DR also influences system
income. The system power sale income difference before and after
(3) EVG output uncertainty PBDR could be regarded as a DR implementation cost. This paper
takes the power shortage cost and DR implementation cost as
The EVG consists of many electric vehicles with different the system compensation cost and takes the minimum system
charging-discharging power behaviors. However, in terms of the compensation cost as the objective function.
overall output, the EVG output should obey a certain PDF. The 8
>
> min
f 1
paper takes the mean output value as the actual output of electric >
>
vehicles. Therefore, the predicted output of the EVG is primarily
< s:t: Prff 1 6
f 1 g P a1
ð15Þ
influenced by an EV’s charging-discharging number. The EVG out- >
> XT
>
> f1 ¼ ½ðC Pt þ C It Þ þ qS DLS;t 
put can be calculated by Eq. (12): :
t¼1

g
EV;t ¼ Ndis dis chr chr

EV;t g EV;t  N EV;t g EV;t ð12Þ wherein the confidence level a1 is used to constrain the probability
of the objective function value being lower than the minimum
wherein g
EV;t is the predicted value of the EVG output. To imitate
value. When a1 equals 1, the system should realize the optimization
the EVG’s charging-discharging behavior, the EVG output can be
scheduling of VPP operation. The decision-makers will hope the
divided into the predicted output and the deviation output:
optimization degree of the objective function is better when a1 is
g EV;t ¼ g
EV;t þ g^EV;t ð13Þ higher. Then, to maximize the energy-saving benefit of VPP opera-
tion, the minimum abandoned energy cost of the VPP should be
According to the EVG’s charging-discharging history data, the
taken as another objective function:
deviation distribution of the EVG output can be obtained. The
8
paper assumes g^EV;t  Nð0; r2EV;t Þ [36]. Then, the EVG output follows >
> min
f 2
>
>

EV;t ; r2EV;t Þ.
g EV;t  Nðg < s:t: Prff 2 6
f 2 g P a2
ð16Þ
>
> XT
>
> f2 ¼ ½qW ðg W;t  g W Þ þ qPV ðg PV;t  g PV Þ
3.2.2. DR uncertainty :
t¼1
Because customers are subjective, they change their load
demands with respect to electricity prices, but the relationship Eq. (16) take the minimum abandoned energy as the objective func-
between the load demand and electricity price is uncertain. Tradi- tion. The abandoned energy mainly includes abandoned wind
tional customers’ psychology model [33] regards curtailed load as a power and abandoned photovoltaic power.
determined value and considers that customers respond to system 8
scheduling according to the expected objective. However, the >
> max
f 3
>
>
>
>
implementation of PBDR depends on customer willingness. There- >
>
> s:t: Prff 3 P
f 3 g P a3
fore, the load change produced by PBDR is influenced by other non- >
>
>
> X
economic factors, which will result in strong uncertainty. IBDR pri- >
>
T
>
> f3 ¼ fpW;t þ pPV;t þ pEV;t þ pCPP;t g
marily constrains customer behavior via a signed contract, which >
>
>
> t¼1
effectively reduces the customer default rate. However, due to >
>
>
> pW;t ¼ qW;t g W;t
>
>
the uncertainty of the external environment, the incomplete infor- >
>
mation, the subjective characteristics of decision-makers and other < pPV;t ¼ qPV;t g PV;t
factors, IBDR also has uncertainty. The uncertainty of PBDR and >
>
>
> pEV;t ¼ Ndis chr chr chr
t qEV;t g EV;t  N t qEV;t g EV;t
dis dis
IBDR can cause power fluctuations, and therefore DLPt and DLIt are >
>
>
> ( )
stochastic variables. >
>
>
>
>
qCPP;t g CPP;t  ½aCPP þ bCPP g CPP;t þ cCPP ðg CPP;t Þ2 
>
>
>
pCPP;t ¼
>
> ½uCPP;t ð1  uCPP;t1 ÞDCPP;t
3.3. Model formulation >
>
>
> 8
>
> < Nhot min off min cold
>
> CPP ; T CPP < T CPP;t 6 T CPP þ T CPP
3.3.1. Stochastic chance constraint planning >
> DCPP;t ¼
>
: : Ncold ; T off
To consider the influence of stochastic variables on system CPP CPP;t > T min cold
CPP þ T CPP
scheduling, this section introduces the stochastic constraint plan-
ð17Þ
ning method [37]. The stochastic chance constraint method is
mainly used to solve decision problems with stochastic variables, Eq. (17) takes the maximum VPP operation income as the objective
wherein the decision must be made before obtaining the real val- function, including the generation incomes of the WPP, PV, EVG and
ues of stochastic variables. Considering the influence of stochastic CPP. The constraints of the proposed model include the demand and
variables, the decisions are allowed to deviate from the constraints supply balance constraint, CPP operation constraint, EVG charging-
to a certain degree to get closer to the actual situation. The confi- discharging constraints, DR operation constraint and system reserve
dence level should be set considering the decision-maker’s toler- constraint.
166 L. Ju et al. / Energy Conversion and Management 128 (2016) 160–177

(1) Demand and supply balance constraint  Load climbing constraint

To realize peak-valley shifting, the load change generated by
GVPP
t þ g GC;t ¼ Lt  sI;d I;d
t DLt þ sI;u I;u
t DLt ð18Þ PBDR and IBDR should meet the climbing constraints:

GVPP ¼ g W;t ð1  uW Þ þ g PV;t ð1  uPV Þ þ ðNdis dis sPt LPD 6 DLPt  DLPt1 6 sPt LPU ð32Þ
t EV;t g EV;t

 Nchr
EV;t g EV;t Þ þ g CPP;t ð1  uCPP Þ
chr
ð19Þ sIt LID 6 DLIt  DLIt1 6 sIt LIU ð33Þ
Because the outputs of the WPP, PV and EVG are stochastic vari- sPt sIt
wherein and are the operation states of PBDR and IBDR, respec-
ables, constraint (18) should be rewritten as follows: tively, and are both 0–1 variables. 1 indicates operation, and 0 indi-
PrfGVPP þ g GC;t ¼ Lt  sdt DLdt þ sut DLut g P b1 ð20Þ cates shutdown.
t

 Startup-shutdown time constraint

(2) CPP operation constraint After receiving the price change signal and corresponding
CPP 6 g CPP;t 6 uCPP;t g CPP
uCPP;t g min ð21Þ
max
instructions from the ISO, customers need some time to adjust
their power consumption behaviors. In addition, the load provided
uCPP;t Dg CPP 6 g CPP;t  g CPP;t1 6 uCPP;t Dg þCPP ð22Þ by PBDR and IBDR should also meet the startup-shutdown time
constraints:
ðT on on
CPP;t1  M CPP ÞðuCPP;t1  uCPP;t Þ P 0 ð23Þ P;on
½X t1  T P;on ðsPt1  sPt Þ P 0 ð34Þ

ðT off  M off
CPP ÞðuCPP;t  uCPP;t1 Þ P 0 ð24Þ
CPP;t1 ½X tP;off  T P;off ðsPt  sPt1 Þ P 0 ð35Þ
wherein uCPP;t is a 0–1 variable; 1 indicates that the CPP is sched-
uled, and 0 indicates that the CPP is not scheduled. ½X I;on
t1  T
I;on
ðsIt1  sIt Þ P 0 ð36Þ

(3) EVG charging-discharging constraint ½X I;off

t  T I;off ðsIt  sIt1 Þ P 0 ð37Þ
During the entire scheduling period T, the relationship between
the charging power and discharging power of the EVG should meet
the following constraint:  Operation situation constraint
To increase the convenience for the ISO with regard to making
X
T X
T
the scheduling scheme based on customer demand response, cus-
ðQ 0 þ g chr
EV;t  Q t Þð1  qEV Þ ¼ g dis
EV;t ð25Þ
tomers should choose to participate in PBDR or IBDR. Thus, one
t¼1 t¼1
customer cannot participate in both PBDR and IBDR at the same
Assuming that the EVG cannot charge and discharge at the same time:
time, its charging-discharging behavior should meet the following
constraint. sPt  sIt ¼ 0 ð38Þ
In addition, to ensure stable load demand, customer demand
EV;t  g EV;t ¼ 0
g chr ð26Þ
dis
response times of IBDR should meet:
In addition, the charging-discharging power of the EVG should
X
T
not exceed the corresponding upper limitation because the EVG ðsI;u I;d I;max
t þ st Þ 6 S ð39Þ
output is a stochastic variable that should meet the following t¼1
constraint:

Prf0 6 g chr
chr

EV;t 6 g EV;t g P b2 ð27Þ (5) System reserve constraint
PrfGVPP;max
t  GVPP
t þ DLPt þ DLIt P RVPP
t g P b6 ð40Þ
EV;t 6 g EV;t g P b3
Prf0 6 g dis
dis ð28Þ
PrfGVPP
t  GVPP;min
t P RVPP
t g P b7 ð41Þ

(4) DR operation constraint

RVPP
t ¼ r 1  Lt þ r 2  g W;t þ r3  g PV;t þ r4  g EV;t ð42Þ
DR implementation can change customer behavior and transfer/
curtail part of the load demand. This part of the load demand is Customer upstream spinning reserve indicates customers
regarded as virtual generation units, which have similar con- reducing power consumption to increase power generation. Cus-
straints to those of CPP operation. The primary constraints include tomer downstream spinning reserve indicates customers increas-
the maximum curtailment load constraint, climbing constraint and ing power consumption to reduce power generation.
minimum startup-shutdown time constraint. In the above model, the objective function
f 3 , Eqs. (21)–(24) and
Eqs. (34)–(37) are non-linear objective functions and non-linear
 Load curtailment constraint constraints, respectively. Therefore, to reduce the difficulty of solv-
To ensure the basic load demand, the curtailment load provided ing the model, the solving method of [38] is applied to linearize the
by DR should not exceed the maximum value: objective functions and constraints.

jLt  L0t j 6 DLmax

t ð29Þ 4. Three-stage hybrid intelligent solution algorithm

Prf0 6 DLPt 6 sPt DLPt g P b4 ð30Þ The VPP scheduling model is characterized by multiple con-
straints, multiple variables and non-linearity. Generally, traditional
Prf0 6 DLIt 6 sIt DLIt g P b5 ð31Þ optimization solution methods are slow and have poor
convergence, which make them not able to address large-scale
 L. Ju et al. / Energy Conversion and Management 128 (2016) 160–177 167

calculations in VPP scheduling problems. Swarm intelligent algo- find the optimal solution. Therefore, PSO cannot exceed its limita-
rithms refer to biological group-evolution processes that search tions and is liable to fall into local optima, known as ‘early-
for the optimal solution, which can overcome the insufficiency of maturing.’ During the search progress, the optimal solution and
traditional algorithms. Therefore, this study puts forward a three- other solutions are not completely separated but have a continu-
stage hybrid intelligent algorithm to solve the proposed model. ous relationship, i.e., they obey a stealth law. Cloud evolve thought
In the first stage, an improved particle swarm algorithm with could aid in building the uncertainly model to transfer a qualitative
stochastic imitate technology and cloud evolve thought (SC-PSO) concept into a quantitative value by finding this stealth law [40].
is built to solve the optimization model with a single objective Cloud evolve generates a normally distributed cloud swarm around
function. In the second stage, a decision attribute table of objective the excellent cloud droplets and improves the proportion of excel-
functions is calculated with the results of the single objective mod- lent cloud droplets via iteration and updating. Therefore, cloud
els. Then, the entropy weight method is used to obtain the weight evolve thought can involve early-maturing and identify the global
of each objective. In the third stage, the model decision-making optimal.
method is constructed with the fuzzy satisfaction theory (SFT- The definition of a cloud is as follows: assume that U is a quan-
Decision Making Method) to calculate the satisfactory optimiza- titative discussion domain represented by an accurate numeral. C
tion result. is a qualitative concept on U, and for any element x of the discus-
sion domain, x is a random implementation of the qualitative con-
4.1. First-stage: SC-PSO algorithm cept C. The certainty degree of x is bðxÞ ! ½0; 1, a random variable
with a stable trend.
4.1.1. Particle swarm algorithm
b : U ! ½0; 1 X ! bðxÞ ð43Þ
Particle swarm optimization (PSO) was put forward by Kennedy
and Eberhart in 1995. The PSO algorithm initializes a particle The overall concept feature of the cloud model can be reflected
swarm and adjusts its search direction and position to gain the by digital features of the cloud. The expected value Ex , the entropy
optimal solution by learning from the individual extreme value En and the hyper entropy He can represent a cloud. Using three
and the global value to find the optimal solution. During the search numbers to describe the overall characteristic of the concept, writ-
progress, the particles adjust their search direction and speed and ten as CðEx ; En ; He Þ, the detailed calculation method and cloud
update their position to fine the optimal particle. PSO has four evolve search method is introduced in [41]. This study introduces
characteristics: (1) each particle is initialized with a stochastic cloud evolve thought into particle swarm optimization and
velocity and moves in the entire problem space, (2) each individual chooses excellent particles according to the fitness calculation
has a memory function, (3) individual evolution is primarily real- result. Then, a vector is constructed by those excellent particles,
ized by cooperation and competition between individuals, and and the vectors construct an evolve cloud. Fig. 2 shows the particle
(4) PSO is a simple method and has a deep intelligence background. distribution of the cloud evolve model.
A detailed introduction of PSO is presented in [39].
4.1.3. Stochastic simulation technology
4.1.2. Cloud evolve thought There are two ways to solve the stochastic chance constraint
Particle fitness changes greatly due to the uncertainty output of planning. (1) Transform the chance constraint planning into deter-
the VPP and DR. The majority of the particles are of a low fitness mined planning to solve the problem by determined planning the-
level, which makes it more difficult for the particles to efficiently ory. However, some constraints are difficult to transfer. (2) Address

Excellent particle

Ordinary particles

Progeny particles

Fig. 2. Particle distribution of the cloud evolutionary model.

168 L. Ju et al. / Energy Conversion and Management 128 (2016) 160–177

Table 1
The decision attribute table of objective functions.

Annotation: ⁄denotes setting the objective as the optimization objective and solving the model.

8
the chance constraints with stochastic imitate technology and then >
>
> max ~f
>
>
Prf~f P
f g P minfai g
solve the problem with an intelligent algorithm [41]. This study >
>
< s:t:
chooses the second method to address the stochastic constraints.
Xk ð50Þ
Prðg j ðx; nÞ 6 0; j ¼ 1; 2; . . . ; mÞ P bj ð44Þ >
> ~f ¼ ki
f 0i
>
>
>
>
>
:
i¼1
For an unconditional variable x, set N1 = 0 and then generate N i ¼ 1; 2; 3
random variables according to the PDF. If the random variable n
and the decision variable x meet the inequality wherein
f 0i is the objective function that is obtained from Eqs. (45)
g j ðx; nÞ 6 0; j ¼ 1; 2; . . . ; m, then increment N1 by 1. When N is large and (46).
enough, Eq. (44) comes into play according to the big-number law; Then, Eq. (50) is set as the optimization objective function with
thus, N 1 =N P bj . Conversely, if N 1 =N < bj , the passed particle the constraints from Eqs. (19)–(42). The SC-PSO algorithm is used
would not meet the confidence level of Eq. (44), indicating that to solve the model and obtain the global optimal solution set.
the particle should be abandoned and replaced by a new particle
until all particles meet the constraint of N 1 =N P bj .
4.3. Third-stage: SFT-decision making method
4.2. Second-stage: function weight model
The method in Sections 4.1 and 4.2 could obtain a global opti-
Based on the SC-PSO algorithm described in Section 4.1, this mal solution set considering all objectives, namely the Pareto opti-
mal frontier. Then, the decision-maker needs to choose the best
study sets
f i ði ¼ 1; 2; . . . ; kÞ as the optimization objectives for
compromise solution according to the specific preferences for dif-
obtaining the solution set. The decision attribute table of the objec-
ferent applications. The fuzzy satisfaction method is introduced to
tive functions is listed in Table 1.
build the decision-making method [42]. The membership function
According to the decision attribute table, the decision matrix of
of objective function i in Pareto optimization scheme r is defined as
the objectives can be obtained as ½r ij kk ði; j ¼ 1; 2; . . . ; kÞ. Both cost-
uri . The drop-gradient membership function is introduced to
objective and benefit-objective exist in the objective set. Different
address the cost-objective, and the ascending half-line member-
objectives have different orders of magnitude, which require a pre-
ship function is introduced to address the benefit-objective, as
handle before weighting the objectives. The detailed handle
follows:
method is as follows.
8
f r 6 minð
f n Þ
>
> 1
 For cost-objective, they should be handled by Eq. (45): < n
maxð
f n Þ
f rn
urn ¼ minð
f n Þ 6
f rn 6 maxð
f n Þ ; n ¼ 1; 2; . . . ; k
> maxð
f n Þminð
f n Þ
>
maxi r ij  r ij :
f r P maxð
f n Þ
r 0ij ¼ ; j ¼ 1; 2; . . . ; k ð45Þ 0; n
maxi r ij  mini r ij ð51Þ
 For benefit-objective, they should be handled by Eq. (46):
8
r ij  mini r ij > 1;
f r P maxð
f m Þ
r 0ij ¼ ; j ¼ 1; 2; . . . ; k ð46Þ >
>
<
m
maxi r ij  mini r ij
f r f min
urm ¼ i 3
; minð
f m Þ 6
f rm 6 maxð
f m Þ ; m – n; m ¼ 1; 2; . . . ; k
>
> f max f min
After Eqs. (45) and (46), the decision matrix after the pre-handle >
:
3

0;
f r 6 minð
f m Þ
½r 0ij kk can be obtained. Then, the entropy weight method is intro- m

ð52Þ
duced to solve the weights of the objectives [10]:
 Calculate the entropy Ei of objective
f i :
X
n X
k
Ei ¼ w rij lnðr 0ij Þ; i ¼ 1; 2; . . . ; k ð47Þ ki  uri
i¼1
j¼1
ur ¼ ð53Þ
X
R X
k
wherein w ¼ 1= lnðnÞ, a constant related to the sample number, to ki  uri
Pn 0
make Ei 2 ½0; 1, r 0ij satisfies 0 < r 0ij < 1 and j¼1 r ij ¼ 1, and when
r¼1 i¼1

r ij ¼ 0, r ij lnðrij Þ ¼ 0.
where in m and n are the indexes of the objective function, i = n + m,
 Calculate the weights of the objective functions:
r is the index for the Pareto solution set, R is the Pareto solution set,
di ¼ 1  Ei ; i ¼ 1; 2; . . . ; k ð48Þ
and
f ri and uri are the objective values
f ri and its membership degree
di value Pareto optimization scheme r, respectively. The solution with
ki ¼ Pi ; i ¼ 1; 2; . . . ; k ð49Þ the maximum membership function is the most preferred compro-
i¼1 di
mise solution based on the adopted weight factors, which is also
Then, the weights of the single objectives with their weight selected as the best Pareto-optimal solution or the final solution
coefficients are used to synthesize the objective function as of the optimization problem. Fig. 3 shows the three-stage hybrid
follows: intelligent solution algorithm flow chart.
 L. Ju et al. / Energy Conversion and Management 128 (2016) 160–177 169

Start

1.Time division and tariffs

2.Parameters of IBDR and PBDR
3.Parameters of Conventional gas turbine Input parameter
4.Predicted values of WP, PV, load
5. Parameters of plug-in electric vehicle

1.Parameter of random chance constrained The minimum cost of The minimum cost of VPP The maximum revenue of
programming system implementing DR abandoning Energy VPP generation power
2.Parameter of the objective function
confidence level Objective functions
3.Parameter of constraints confidence level

PSO algorithm Equality constraints:(19)、(20)、(25)、(26)、(39)、(40)

1. Initialize population size Inequality constraints:(21)、(22)-(24)、(27)、(28)、(30)-(38)、(40)-(43)
2. Initialize particle iterations number
3. Initialize other parameters

VPP scheduling model calculations under single objective

function, gain particle fitness value and current individual Decision property sheet and decision matrix of
optimal solution set objective functions base on the results of single
objective optimization model

Cloud evolutionary search PSO algorithm

1.construct excellent particle swarm by 1. Construct ordinary particle group
fitness value with m outstanding particle without m outstanding particle
Pretreatment for efficiency objective function
2. Generating n cloud droplets around 2.Check whether the maximum and vost objective function
Excellent particle around by using cloud number of iterations, yes, stop
conditions generator iteration; no, go to step 3
3.CHange particle flying weight, iterative 3.Using an ordinary PSO update
update particle by take cloud strategy iteration particle Apply entropy method to calculate the weight
coefficients of objective function

Updated particle swarm after cloud

evolutionary search and PSO search Weight each target function with all the
objective function weights, construct objective
function optimization model

Whether particles
meet the confidence level N
Requirement?
Obtain model optimization results , the solution
could reconcile the various optimization
Y objectives, namely Pareto frontier

N Whether reach the

maximum iteration? Apply satisfaction fuzzy theory to calculate the
membership function of objective function i (i =
Y 1,2, ..., k) in pareto optimization Scheme r

VPP scheduling model calculations, gain particle fitness value and

current individual optimal solution set Weight Objective function membership function
In different Pareto optimization r, forming a
comprehensive membership function

Whether solve the

weighted hybrid model
N

Y Result

Fig. 3. Three-stage hybrid intelligent solution algorithm flow chart.

170 L. Ju et al. / Energy Conversion and Management 128 (2016) 160–177

5. Case study

To verify the validity and applicability of the proposed model,

this section constructs the simulation scenario set. Then, the Yun-
nan distributed power demonstration engineering project in China
is taken as the simulation object for example analysis.

5.1. Scenario set

Case 1: The basic scenario, VPP scheduling without the EVG and
DR. This scenario did not consider the EVG and DR but analyzed the
applicability of the proposed stochastic scheduling model and cor-
responding algorithm in overcoming the uncertainty factors of VPP
operation.
Case 2: EVG scenario, VPP scheduling with only the EVG. This Fig. 4. Predicted values (24 h) for load, WPP and PV.

scenario only considers VPP scheduling with the EVG, assuming

with a similar installed capacity. The CPP contains one TAURUS60,
all EVs are plug-in electric vehicles. The charging-discharging
one CENTAUR50 and one CENTAUR40 gas turbine [44] and divides
power and the total charging capacity of EVs are 1.8 kW and
the cost curve into two segments to linearize it. The detailed
10.8 kW h. The energy loss rate of the charging-discharging power
parameters are listed in Table 3.
is 5.6%. The number of EVs is 1000 [43]. Assume that the charging-
This engineering project was built to meet the demand load in
discharging power deviation of the EVG obeys the normal distribu-
Diqing, Yunnan province. Its typical demand load curve is described
tion N(0, 0.122) [36]. The charging-discharging power price of the
in [43]. The minimum values of the daily minimum demand load
EVG follows the TOU price [8].
rate in Diqing are 57% in March. The prices of the WPP and PV
Case 3: DR scenario, VPP scheduling with only DR. This scenario
are set by the government as 610 Yuan/MW h and 1000 Yuan/
considers the VPP scheduling scheme with PBDR and IBDR. Assume
MW h, respectively [43]. According to [45], the price of the CPP is
that the electric price before PBDR is 590 Yuan/MW h. Referring to
set as 560 Yuan/MW h. In addition, historical data of wind velocity
[32], this study sets the electric price at a load peak, float and valley
and illumination intensity in Diqing were selected for the regres-
period and demand-price elasticity. Table 2 contains the time-of-
sion fitting of the probability density functions [43]. The parame-
use period and price parameters. To avoid peak-valley upside-
ters of the wind turbine are v in ¼ 3 m=s, v rated ¼ 14 m=s and
down, this case limits the fluctuation load of PBDR to not more
than 15% of the original load, namely 3 MW.
v out ¼ 25 m=s, and pthe shape parameter and scale parameter are
ffiffiffiffi
For IBDR, the upstream/downstream spinning reserve price is u ¼ 2 and # ¼ 2v
= p [10], respectively. The values of a and b
800 Yuan/MW h and 160 Yuan/MW h [10]. The maximum load are 0.32 and 8.14, respectively. The method proposed by [46] is
and load fluctuation of IBDR participating in spinning reserve is used to simulate the generation scenarios of the WPP and PV. 10
not more than 3.5 MW and 1.5 MW. typical scenarios are obtained, with the mean value of all scenarios
Case 4: Comprehensive scenario, VPP scheduling with the EVG taken as the input value. The details are shown in Fig. 4.
and DR. This scenario considers scheduling schemes with both The initial population size is set to 2000, the learning factor
the EVG and DR. The detailed parameters are the same as in the c1 = c2 = 0.2, the maximum inertia weight wmax = 0.8, the minimum
other cases. inertia weight wmin = 0.8, the maximum flight speed Vmax = 10, and
the maximum iteration number is 2000 [38]. During the cloud
evolve search process, five excellent particles are chosen according
5.2. Basic data
to their fitness values. In addition, five particle swarms are gener-
ated around those five particles by using the cloud evolve search
China’s Yunnan distributed power demonstration engineering
method. The optimal confidence is when the error of the two con-
project is taken as the simulation object [43]. The project includes
tinuous generations is within the allowed range [42]. MATLAB
a 4.5 MW PV, a 6.5 MW WPP and 18 MW hydro power units in
2011a was used as the solving software on an Ideapad-Y450 com-
Xiaozhongdian, Shuoduogang river. Similar to hydro units, the CPP
puter with a 3.4-GHz CPU and 4 GB RAM.
has rapid startup-shutdown speeds and low pollutant emission
levels. Therefore, this study selects the CPP to replace hydro units
5.3. Example results
Table 2
Time-of-use period and price parameters. 5.3.1. Case 1: VPP scheduling without EVG and DR
The March load with the maximum load rate is selected as the
Parameter Price with PBDR/(Yuan/MW h)
basic data to study the application of stochastic constraint plan-
Period Valley period Normal period Peak period
ning in overcoming the inner uncertainty of the WPP and PV.
0:00–5:00,21:00– 5:00–8:00,14:00– 8:00–14:00,19:00–
24:00 19:00 21:00
Assuming that the prediction error of VPP power generation is
Price 260 590 920 10%, the electricity-cut compensation cost is 1.2 Yuan/kW h. The
confidence parameters of a and b are 0.9 and 0.92, respectively.

Table 3
CPP operation parameters.

Unit type g min

CPP /MW
g max
CPP /MW Dg CPP /MW DCPP;t /Yuan on=off
M CPP /h Slope of segment 1 /(Yuan/MW) Slope of segment 2/(Yuan/MW)

TAURUS60 2.5 5.67 3 204.8 2 239 273.2

CENTAUR50 2 4.6 2.5 136.3 1.5 150.25 307.3
CENTAUR40 1 3.515 1.8 122.9 1 136.6 341.5
 L. Ju et al. / Energy Conversion and Management 128 (2016) 160–177 171

Table 4
The decision attribute table of the objective functions (unit: Yuan).

16 reduce the power output of the WPP and PV. Based on the decision
(a)
14 attribute table (Table 3), to analyze the VPP output structure with
and without uncertainty, this study used the entropy weight
12
method (Eqs. (44)–(53)) to weight multiple objective functions
10 and make the optimal decision. Then, the VPP operation scheduling
8 is obtained as shown in Fig. 5 (Schedule 1).
6
If considering uncertainty, the power output of the WPP and PV
would decrease, whereas the CPP output would increase. The aban-
4
Power output/MW

doned power of the WPP and PV increased from 9.90 MW h to

2 15.82 MW h and 3.61 MW h to 6.15 MW h, respectively. The grid
0 connection power of the CPP increased from 310.61 MW h to
319.55 MW h. To reduce the power shortage cost due to the VPP
16
(b) uncertainty, the system would reduce the output of the WPP and
14 PV and increase the CPP output to meet the load demand.
12 Finally, set a ¼ b to analyze the influence of the confidence level
10 on system scheduling. The objective values and VPP grid connec-
8
tion power under different confidence levels are obtained by the
sensitivity analysis. The detailed results are listed in Table 5.
6
As the confidence degree increases, the decision-maker’s will-
4 ingness to bear the power shortage cost will decrease. The VPP grid
2 connection power decreased from 464.21 MW h to 452.08 MW h.
0 The abandoned energy cost and system operation cost increase,
1 3 5 7 9 11 13 15 17 19 21 23 but the VPP scheduling income decreases. That is, the stochastic
Time/h change constraint planning method can effectively control the
uncertainty of the VPP output. The system power shortage costs
Fig. 5. The scheduling result comparison of the VPP in Case 1 with (a) and without and VPP operation risk will decrease, which could provide a pre-
uncertainty (b). vention and control tool for the decision-maker.

Using the proposed algorithm to solve the models with and with-
out uncertainty, the time to solve the model is 70 s and 78 s, which 5.3.2. Case 2: VPP scheduling with only EVG
are less than the solution times in the basic PSO algorithm (95 s This section discusses the EVG’s optimization effect on VPP
and 102 s) and the non-dominated sorting genetic-II (NSGA-II) scheduling. The parameter values are similar to those in Case 1.
algorithm (85 s and 97 s) [47]. The computational time for the Then, the optimization result shows that the system operation
other three cases are 85 s, 88 s and 95 s. Table 4 is the decision cost, abandoned energy cost and VPP operation income are
attribute table of the objective functions. 6599.91 Yuan, 12222.41 Yuan and 184563.1 Yuan, respectively.
Compared with the objective function value without uncer- According to the simulation results, the EVG can only alleviate
tainty, the system operation cost is reduced from 183.95 million the influence of uncertainty on system power generation. How-
to 71.33 million Yuan when the uncertainty is considered. How- ever, compared with Case 1, the system operation cost and aban-
ever, the system abandoned energy cost would increase, which doned energy cost are markedly reduced, whereas the VPP
indicates that the grid connection power and operation income operation income is increased. The output of the WPP and PV
of the VPP are reduced. To reduce the electricity shortage cost increased by 2.97 MW h and 0.94 MW h, respectively. Fig. 6 is
due to the VPP output uncertainty, system scheduling would the scheduling results of VPP in Case 2.

Table 5
System scheduling results under different confidence levels.

ða; bÞ Objective value/Yuan VPP grid connection/MW h

f
f
f
1 2 3

0.8 7548.24 15028.18 188245.39 464.21

0.84 7391.56 15335.26 186012.48 461.32
0.88 7273.45 15548.27 183585.19 458.57
0.92 7132.84 15806.03 182081.81 455.29
0.96 6937.95 16028.45 179587.60 452.08
172 L. Ju et al. / Energy Conversion and Management 128 (2016) 160–177

operation income increases. The EVG could substitute the CPP to

provide reserve service. Therefore, the CPP output decreased from
319.55 MW h to 316.28 MW h. In particular, during the period
from 1:00 to 8:00, the CPP output decreased by 45.27 MW h. The
EVG charging-discharging distribution with 1000 EVs (benchmark
value) is shown in Fig. 7 (Schedule 2).
According to Fig. 7, the EVG primarily charges power during the
valley periods and discharges power during the peak periods. Some
EVs charge power while others discharge power during flat peri-
ods. This shows that the EV charging-discharging time distribution
is opposite the demand load distribution. Furthermore, sensitivity
analysis was performed to analyze the influence of the EVG grid
connection number on the load demand and VPP grid connection
power. The sensitivity analysis results is shown in Fig. 8 (Schedule
3).
Fig. 6. The scheduling results of the VPP in Case 2. According to Fig. 8, with an increase in the number of EVG grid
connections, the load demand curve becomes smoother. When the
number reaches 2000, the peak load and valley load are 25.31 MW
1.2
and 16.45 MW, respectively. The peak-valley ratio is 1.59. Com-
Charge power Discharge power
pared with the original results, the peak load decreased by
1
560 kW, the valley load increased by 1.01 MW, and the peak-
valley ratio decreased by 0.16. Regarding the GC output power,
EV number/pu

0.8
the system generally reduces the power purchasing from the GC,
0.6
which indicates that the system power shortage is alleviated.
Therefore, the VPP could make use of the EVG to meet the load
0.4 demand with EV charging power during the valley load periods
and discharging power during the peak load periods.
0.2

5.3.3. Case 3: VPP scheduling with only DR

0
1 3 5 7 9 11 13 15 17 19 21 23 This section analyzes the influence of PBDR and IBDR on VPP
generation scheduling. The parameter values are the same as in
Time/h
Case 1. The simulation results show that the system operation cost,
Fig. 7. Electric vehicle charging-discharging distribution. abandoned energy cost and VPP operation income are
7836.48 Yuan, 7415.78 Yuan and 186368.45 Yuan, respectively.
DR could help smooth the load demand curve, improve the VPP
grid connection, decrease the abandoned energy cost and increase
28
the VPP operation income. However, DR also expanded the uncer-
26 tainty influences of the WPP and PV on system power generation.
Compared with Case 1, the system operation cost increases by
Load demand/MW·h

24
703.56 Yuan. The load demand curve before and after PBDR and
22 IBDR is shown in Fig. 9.
According to Fig. 9, both PBDR and IBDR show good peak-valley
20
shifting effects. Compared with the original load demand curve, the
18 peak-valley ratio of the load demand curves after PBDR and IBDR
16
are 1.57 and 1.55, respectively. IBDR has a more obvious peak-
valley shifting effect on the load demand curve. For PBDR, cus-
14 tomer responses lag behind the TOU price. Therefore, some load
12 will also shift and increase during the float load period. For IBDR,
the load demand is almost constant during the float period.
GC power output/MW

10

0
1 3 5 7 9 11 13 15 17 19 21 23
Time/h

Fig. 8. Sensitivity analysis results for the electric vehicle number in Case 3.

In Case 2, the VPP consists of a WPP, PV, CPP and EVG. The EVG
and CPP both provide reserve services for the WPP and PV. Com-
pared with Case 1, the system operation cost decreases and VPP Fig. 9. Load demand curve before and after PBDR and IBDR.
 L. Ju et al. / Energy Conversion and Management 128 (2016) 160–177 173

Therefore, the load demand curve after IBDR is smoother than that maximizes the VPP grid connection power. The load demand distri-
after PBDR. When both PBDR and IBDR are introduced, the load bution for the four cases is shown in Fig. 11 (Schedule 5).
demand curve is smoothed to the maximum extent. The peak load According to Fig. 11, when both EVG and DR are introduced, the
is 25.31 MW, the valley load is 16.39 MW, and the peak-valley load demand is smoothed to the maximum extent. In Case 4, the
ratio is 1.54. The optimization result of VPP scheduling in Case 3 peak load and the valley load are 24.85 MW and 17.02 MW, respec-
is shown in Fig. 10 (Schedule 4). tively. Compared with Case 1, the peak-valley ratio decreased from
Compared with Fig. 5, the load demand curve considering DR is 1.74 to 1.46, reaching the minimum point, which indicates that the
smoother. The WPP grid connection power increased by EVG and DR have a coordinated optimization effect. The EVG pro-
3.46 MW h, the PV grid connection power increased by vides reserve service for the VPP on the generation side. DR
1.67 MW h, the CPP output decreased by 4.45 MW h. According smoothes the load demand on the demand side to improve the
to the IBDR load distribution, customers provide positive output VPP grid connection. In this manner, a chain reaction between
(load) to consume the VPP power output during the valley load the generation side and demand side is created. The VPP schedul-
periods. In addition, customers provide negative output (curtail- ing optimization result in Case 4 is shown in Fig. 12 (Schedule 6).
ment load) to reduce the CPP output to provide more reserve ser- Compared with the other cases, the VPP grid connection power
vice capacity for the WPP and PV during the peak load periods. reaches a maximum in Case 4. The outputs of the WPP, PV and CPP
are 96.32 MW h, 46.16 MW h and 314.79 MW h, respectively.
Compared with Case 1, the WPP output increased by 5.43 MW h,
5.3.4. Case 4: VPP scheduling with EVG and DR and the PV output increased by 2.6 MW h, whereas the CPP output
This section discusses the optimization effect with both EVG decreased by 4.76 MW h. The EVG could substitute the CPP to pro-
and DR. The parameter values are the same as in the above scenar- vide reserve service for the VPP by charging or discharging power.
ios. The simulation result shows that the system operation cost, In addition, the IBDR output would be reduced, which indicates
abandoned energy cost and VPP operation income are that the EVG can reduce the system reserve cost. The VPP power
6990.46 Yuan, 5691.58 Yuan and 188307.53 Yuan, respectively. output in the four cases is shown in Fig. 13 (Schedule 7).
Therefore, when both EVG and DR are introduced, the system oper- According to Fig. 13, the VPP outputs in the four scenarios are
ation result is optimized, which not only makes use of the EVG 453.525 MW h, 455.65 MW h, 456.68 MW h and 458.83 MW h,
charging-discharging capacity to provide reserve service for the respectively. Compared with Case 1, the EVG can output in the
WPP and PV but also smoothes the load demand curve to expand peak load periods and consume power in the valley load periods
the VPP grid connection space. Compared with Case 3, the EVG in Case 2, which makes the peak-valley ratio of the VPP output
alleviates the uncertainty influence of the VPP output on system curve larger than the original curve, e.g., 1:00–5:00 and 9:00–
scheduling and reduces the system operation cost. Compared with 15:00. In addition, compared with Case 1, DR could smooth the
Case 2, PBDR could smooth the load demand curve, and IBDR pro- load demand curve in Case 3. The load in the peak load period
vides reserve service for the VPP from the demand side, which

Fig. 12. VPP scheduling optimization result in Case 4.

Fig. 10. the optimization result of VPP scheduling in Case 3.

Fig. 11. Load demand distribution in the four cases. Fig. 13. VPP power output in the four cases.
174 L. Ju et al. / Energy Conversion and Management 128 (2016) 160–177

Table 6
The optimized system operation results for the four cases.

Object function value VPP power output/MW h DR/MW Peak-valley ratio

f
f
f WPP PV CPP EVG IBDR PBDR
1 2 3

Charge Discharge Up Down Peak Flat Valley

Case1 7132 15,806 182,081 90.89 43.56 319.55 – – – – – – – 1.74
Case2 6600 12,222 184,563 93.86 44.50 316.28 15.29 13.76 – – – – – 1.59
Case3 7836 7416 186,368 94.35 45.23 315.10 – – 16.33 10.53 7.90 0.12 2.96 1.54
Case4 6990 5692 188,307 96.32 46.16 314.79 12.23 11.01 14.83 8.59 3.27 0.75 4.27 1.46

decreased, and the load in the valley load period increased. The (1) If considering uncertainties, the system will reduce the VPP
VPP output power distribution curve is identical to the load power generation and decrease the power shortage cost.
demand curve. In Case 4, both the EVG and DR are incorporated. This could control the influence of the uncertainty of the
The VPP output combined its characteristics in Case 2 and Case WPP and PV on system operation. The grid connection
3, namely, the VPP output curve is smoother, and the grid connec- power of the VPP will decreases more with a higher confi-
tion power reaches the maximum. Table 6 shows the optimization dence level. That is, stochastic chance constraint planning
scheduling results of the four scenarios. can be applied to overcome VPP uncertainties and provide
According to Table 6, compared with the other three cases, the risk control tools for decision-makers.
scheduling scheme is optimal in Case 4. For the objective values, (2) The proposed three-stage hybrid intelligent solution algo-
compared with Case 2, the system operation cost increased, the rithm could obtain the optimal decision solution of the
system abandoned energy cost decreased, and the VPP operation VPP scheduling model. The cloud evolve thought is intro-
income was optimized, which indicate that the system operation duced into the traditional particle swarm algorithm by using
result reached the optimal level. The VPP power generation the cloud evolve model to update the excellent particles.
reached the maximum of 458.83 MW h, which is more than the Stochastic simulation technology could prevent the early
453.525 MW h in Case 1. For the peak-valley ratio, the EVG, IBDR maturation problem and obtain the global optimal particles.
and PBDR have the same effect on peak-valley shifting. The peak The entropy weight method and the fuzzy satisfaction the-
load decreased by 1.36 MW, the valley load increased by ory could solve the weight coefficients of the objective func-
1.93 MW, and the peak-valley ratio reached the minimum of tions and obtain the optimal decision solution of the multi-
1.46. For the VPP output structure, the EVG replaced the CPP to objective model.
provide reserve service. Compared with Case 1 and Case 2, the (3) The system operation cost and abandoned energy cost will
CPP output in Case 4 decreased by 1.49 MW and 4.76 MW, respec- reach minimum values and the VPP operation income will
tively. For the DR implementation effect, the EVG helps overcome reach a maximum value when the EVG and DR are both
the VPP output uncertainty, which alleviates the pressure on DR. introduced. The EVG could replace the CPP output to provide
Therefore, the PBDR and IBDR outputs were less than those in spinning reserve service for the WPP and PV and ensure the
Case 2. stability of VPP output. DR can smooth the demand load
curve, expand the VPP grid connection space and achieve
linkage optimization of the generation side and demand
side.
6. Conclusions

In this study, a WPP, PV, EVG and CPP are aggregated in a VPP.
Stochastic chance constraint planning is introduced to overcome Acknowledgments
the uncertainties of VPP scheduling. Then, a stochastic scheduling
model for the VPP is proposed with the objective functions of min- This study was supported by the National Science Foundation of
imum system compensation cost, minimum abandoned energy China (Grant Nos. 71273090 and 71573084) and the Fundamental
cost and maximum VPP operation income. Then, a three-stage Research Funds for the Central Universities (2015XS29).
hybrid intelligent solution algorithm is developed to solve the pro-
posed model using the particle swarm algorithm, the entropy Appendix A
weight method and the fuzzy satisfaction theory. The simulation
results show: See Schedules 1–7.

Schedule 1
The scheduling result comparison of VPP in Case 1 (unit: MW h) (Fig. 5).

With uncertainty Without uncertainty With uncertainty Without uncertainty

WPP PV CPP WPP PV CPP WPP PV CPP WPP PV CPP
1 4.199 0 13.79 4.199 0 13.79 13 3.9 4.008 12.582 3.9 5 11.536
2 3.325 0 13.79 5.499 0 11.494 14 3.302 4.401 13.79 3.302 4.401 13.79
3 2.795 0 13.79 4.998 0 11.463 15 3.497 4 13.79 3.497 4 13.79
4 4.372 0 11.29 4.467 0 11.19 16 3.302 3.402 13.79 3.302 3.402 13.79
5 4.467 0 11.19 4.659 0 10.987 17 2.697 2.3 13.79 2.697 2.3 13.79
6 4.727 0 11.19 4.797 1.318 9.725 18 3.003 2.3 13.79 3.003 2.3 13.79
7 3.926 1.506 11.29 4.199 1.741 10.754 19 3.601 1.399 13.79 3.601 1.399 13.79
8 2.6 2.241 13.79 2.6 2.241 13.79 20 3.998 0.999 13.79 3.998 0.999 13.79
9 3.9 3.501 13.79 3.9 3.501 13.79 21 4.699 0 13.79 4.699 0 13.79
10 3.626 4.198 13.79 4.199 4.198 13.185 22 3.9 0 13.79 3.9 0 13.79
11 3.802 4.5 13.79 3.802 4.5 13.79 23 4.498 0 13.79 4.498 0 13.79
12 3.553 4.801 13.79 3.9 4.801 13.424 24 5.2 0 13.79 5.2 0 13.79
 L. Ju et al. / Energy Conversion and Management 128 (2016) 160–177 175

Schedule 2
The scheduling result comparison of VPP in Case 1 (Fig. 7).

Charge power Discharge power Charge power Discharge power Charge power Discharge power
1 1000 0 9 0 1000 17 497.2 502.8
2 1000 0 10 0 1000 18 400 600
3 1000 0 11 0 1000 19 0 1000
4 1000 0 12 0 1000 20 0 1000
5 1000 0 13 0 1000 21 0 1000
6 700 300 14 260 740 22 800 200
7 600 400 15 400 600 23 1000 0
8 400 600 16 600 400 24 1000 0

Schedule 3
Sensitive analysis results for electric vehicle number in Case 3 (unit: MW h) (Fig. 8).

None 1000 1200 1400 1600 1800 2000 None 1000 1200 1400 1600 1800 2000
1 18 19.8 20.16 20.52 20.88 21.24 21.6 13 19.85 19.24 19.05 18.85 18.45 18.23 18.05
2 16.41 18.21 18.57 18.93 19.29 19.65 20.01 14 20.91 20.12 19.962 19.804 19.646 19.488 19.33
3 15.88 17.68 18.04 18.4 18.76 19.12 19.48 15 20.65 20.35 20.29 20.23 20.17 20.11 20.05
4 15.09 16.89 17.25 17.61 17.97 18.33 18.69 16 20.91 21.31 21.39 21.47 21.55 21.63 21.71
5 15.09 16.89 17.25 17.61 17.97 18.33 18.69 17 23.82 23.86 23.87 23.88 23.88 23.89 23.9
6 15.35 16.1 16.25 16.4 16.55 16.7 16.85 18 25.12 24.82 24.76 24.7 24.64 24.58 24.52
7 16.15 16.55 16.63 16.71 16.79 16.87 16.95 19 24.88 23.18 22.84 22.5 22.16 21.82 21.48
8 19.59 19.29 19.23 19.17 19.11 19.05 18.99 20 25.94 24.24 23.9 23.56 23.22 22.88 22.54
9 21.97 20.27 19.93 19.59 19.25 18.91 18.57 21 26.21 24.51 24.17 23.83 23.49 23.15 22.81
10 20.91 19.21 18.87 18.53 18.19 17.85 17.51 22 25.15 24.85 24.79 24.73 24.67 24.61 24.55
11 22.24 20.54 20.2 19.86 19.52 19.18 18.84 23 21.71 23.51 23.87 24.23 24.59 24.95 25.31
12 21.44 19.74 19.4 19.06 18.72 18.38 18.04 24 20.65 22.45 22.81 23.17 23.53 23.89 24.25

Schedule 4
The optimization result of VPP scheduling in Case 3 (unit: MW h) (Fig. 10).

WPP PV CPP IBDR WPP PV CPP IBDR WPP PV CPP IBDR

1 4.199 0 13.79 2.65 9 3.9 3.501 13.79 2.5 17 2.697 2.3 13.79 0
2 4.135 0 13.79 2.87 10 3.535 4.198 13.2 2.5 18 3.003 2.3 13.79 0.08
3 3.709 0 13.69 2.9 11 3.802 4.5 13.79 2.6 19 3.601 1.399 13.79 2.43
4 5.417 0 11.19 3.16 12 3.9 4.801 12.69 2.6 20 3.998 0.999 13.79 3.42
5 4.557 0 11.19 1.44 13 3.9 5 12.26 1.3 21 4.699 0 13.79 2.33
6 4.797 1.095 11.19 1.1 14 3.302 4.401 13.55 1.26 22 3.9 0 13.79 0.11
7 4.199 1.093 11.29 0 15 3.497 4 13.79 0 23 4.498 0 13.79 3.14
8 2.6 2.241 13.79 0.2 16 3.302 3.402 13.79 0 24 5.2 0 13.79 3.07

Schedule 5
Load demand distribution in four cases (unit: MW h) (Fig. 11).

Case 1 Case 2 Case 3 Case 4 Case 1 Case 2 Case 3 Case 4

1 18 19.8 20.5 20.84 13 19.85 18.15 18.84 18.14
2 16.41 18.56 19.02 19.85 14 20.91 20.12 19.77 18.94
3 15.88 17.71 18.5 19.21 15 20.65 20.35 20.46 20.16
4 15.09 16.91 17.84 19.08 16 20.91 21.31 21.17 21.57
5 15.09 16.89 16.98 18.12 17 23.82 23.86 23.85 23.89
6 15.35 16.1 16.39 17.14 18 25.12 24.82 24.89 24.31
7 16.15 16.55 16.41 17.02 19 24.88 23.18 22.56 22.86
8 19.59 19.29 19.3 19 20 25.94 24.24 23.62 23.12
9 21.97 20.27 19.62 18.56 21 26.21 24.51 23.94 23.24
10 20.91 19.21 18.56 18.45 22 25.15 25.65 25.31 23.93
11 22.24 20.54 19.84 18.86 23 21.71 23.73 24.21 24.85
12 21.44 19.74 19.04 19.12 24 20.65 22.45 23.36 23.68
176 L. Ju et al. / Energy Conversion and Management 128 (2016) 160–177

Schedule 6
VPP scheduling optimization result in Case 4(unit: MW h) (Fig. 12).

WPP PV CPP EVG IBDR WPP PV CPP EVG IBDR

1 4.2 0 13.79 1.8 1.4 13 3.9 5 11.53 1.7 0.35
2 5.5 0 13.23 1.8 2 14 3.3 4.4 12.67 0.79 1.34
3 4.33 0 13.78 1.8 1.89 15 3.5 4 13.66 0.3 0.26
4 4.2 0 13.79 1.8 2.55 16 3.3 3.4 13.79 0.4 0.34
5 5.1 0.45 11.35 1.8 1.59 17 2.7 2.3 13.79 0.04 0.03
6 4.8 0.97 11.19 0.75 1.19 18 3 2.3 13.79 0.3 0.57
7 4.2 1.7 11.29 0.4 0.55 19 3.6 1.4 13.79 1.7 0.66
8 2.6 2.24 13.79 0.3 0.35 20 4 1 13.79 1.7 1.46
9 3.9 3.5 13.55 1.7 2.05 21 4.7 0 13.79 1.7 1.61
10 4.2 4.2 12.38 1.7 1.1 22 3.9 0 13.79 1.1 2.1
11 3.8 4.5 12.92 1.7 2.02 23 4.5 0 13.79 1.8 1.7
12 3.9 4.8 12.77 1.7 0.96 24 5.2 0 13.79 1.8 1.59

Schedule 7
VPP power output in four cases (unit: MW h) (Fig. 13).

Case 1 Case 2 Case 3 Case 4 Case 1 Case 2 Case 3 Case 4

1 17.989 16.099 17.99 16.85 13 20.436 23.955 21.16 22.125
2 16.993 15.372 17.93 16.55 14 21.493 22.283 21.25 21.165
3 16.461 14.261 17.4 16.12 15 21.287 21.587 21.29 21.457
4 15.657 12.452 16.61 15.24 16 20.494 20.094 20.49 20.094
5 15.646 12.886 15.75 15.12 17 18.787 18.744 18.79 18.7468
6 15.84 14.437 17.08 15.78 18 19.093 19.393 19.09 19.593
7 16.694 15.961 16.58 16.792 19 18.79 20.49 18.79 20.49
8 18.631 18.931 18.63 18.931 20 18.787 20.487 18.79 20.487
9 21.191 22.891 21.19 22.652 21 18.489 20.189 18.49 20.189
10 21.582 23.887 20.93 22.48 22 17.69 17.59 17.69 16.81
11 22.092 23.792 22.09 22.918 23 18.288 17.488 18.29 17.188
12 22.125 24.191 21.39 23.17 24 18.99 18.19 18.96 17.89

References [16] Shayeghi H, Sobhani B. Integrated offering strategy for profit enhancement of
distributed resources and demand response in microgrids considering system
uncertainties. Energy Convers Manage 2014;87:765–77.
[1] Mashhour M, Golkar MA, Moghaddas-Tafreshi SM. Extending market activities
[17] Riveros JZ, Bruninx K, Poncelet K. Bidding strategies for virtual power plants
for a distribution company in hourly-ahead energy and reserve markets – Part
considering CHPs and intermittent renewables. Energy Convers Manage
I: problem formulation. Energy Convers Manage 2011;52(1):477–86.
2015;103:408–18.
[2] David S, Arturs P, Ioulia P, et al. Benefits and cost implications from integrating
[18] Perfumo C, Kofman E, Braslavsky JH, et al. Load management: model-based
small flexible nuclear reactors with off-shore wind farms in a virtual power
control of aggregate power for populations of thermostatically controlled
plant. Energy Policy 2012;46:558–73.
loads. Energy Convers Manage 2012;55:36–48.
[3] Hrvoje P, Igor K, Tomislav C. Virtual power plant mid-term dispatch
[19] Zhang D, Shah N, Papageorgiou LG. Efficient energy consumption and
optimization. Appl Energy 2013;101:134–41.
operation management in a smart building with microgrid. Energy Convers
[4] Hrvoje P, Juan MM, Conejo Antonio J, et al. Offering model for a virtual power
Manage 2013;74:209–22.
plant based on stochastic programming. Appl Energy 2013;105:282–92.
[20] Dimopoulos GG, Georgopoulou CA, Stefanatos IC, et al. A general-purpose
[5] Sturm B, Meyers S, Zhang YJ, et al. Process intensification and integration of
process modelling framework for marine energy systems. Energy Convers
solar heat generation in the Chinese condiment sector – a case study of a
Manage 2014;86:325–39.
medium sized Beijing based factory. Energy Convers Manage
[21] Shafie-khah M, Moghaddam MP, Sheikh-El-Eslami MK. Development of a
2015;106:1295–308.
virtual power market model to investigate strategic and collusive behavior of
[6] Petrovic SN, Karlsson KB. Danish heat atlas as a support tool for energy system
market players. Energy Policy 2013;61:717–28.
models. Energy Convers Manage 2014;87:1063–76.
[22] Morais H, Kadar P, Faria P, Vale ZA, et al. Optimal scheduling of a renewable
[7] Steffe B. Prospects for pumped-hydro storage in Germany. Energy Policy
micro-grid in an isolated load area using mixed-integer linear programming.
2012;45:420–9.
Renew Energy 2010;35:151–6.
[8] Tan ZF, Zhang HJ, Shi QS, et al. Multi-objective operation optimization and
[23] Peik-Herfeh M, Seifi H, Sheikh-El-Eslami MK. Decision making of a virtual
evaluation of large-scale NG distributed energy system driven by gas-steam
power plant under uncertainties for bidding in a day-ahead market using point
combined cycle in China. Energy Build. 2014;76:572–87.
estimate method. Int J Electr Power Energy Syst 2013;44(1):88–98.
[9] Wu KH, Zhou H, An SC, Huang T. Optimal coordinate operation control for
[24] Yang HM, Yi DX, Zhao JH. Distributed optimal dispatch of virtual power plant
wind–photovoltaic–battery storage power-generation units. Energy Convers
based on ELM transformation. Management 2014;10(4):1297–318.
Manage 2015;90:466–75.
[25] Yang JJ, Zhao JH, Wen FS, et al. Development of bidding strategies for virtual
[10] Ju LW, Tan ZF, Yuan JY, et al. A bi-level stochastic scheduling optimization
power plants considering uncertain outputs from plug-in electric vehicles and
model for a virtual power plant connected to a wind–photovoltaic–energy
wind generators. Automat Electr Power Syst 2014;38(22):92–102.
storage system considering the uncertainty and demand response. Appl
[26] Braslavsky JH, Wall JR, Reedman LJ. Optimal distributed energy resources and
Energy 2016;171:184–99.
the cost of reduced greenhouse gas emissions in a large retail shopping center.
[11] Spyros SK, Evangelos R, Liana MC, et al. Implementing agent-based emissions
Appl Energy 2015;155:120–30.
trading for controlling Virtual Power Plant emissions. Electr Power Syst Res
[27] Morteza S, Mohammad KS, Mahmoud RH. The design of a risk-hedging tool for
2013;102:1–7.
virtual power plants via robust optimization approach. Appl Energy
[12] Moghaddam IG, Nick M, Fallahi F, et al. Risk-averse profit based optimal
2015;155:766–77.
operation strategy of a combined wind farm-cascade hydro system in an
[28] Bracco S, Delfino F, Pampararo F, et al. A dynamic optimization-based
electricity market. Renew Energy 2013;55:252–9 [Julio HB, Josh].
architecture for polygeneration microgrids with tri-generation, renewables,
[13] Omid P, Kimmo K, Josep MG. Microgrids in active network management—Part
storage systems and electrical vehicles. Energy Convers Manage
I: hierarchical control, energy storage, virtual power plants, and market
2015;96:511–20.
participation. Renew Sustain Energy Rev 2014;36:428–39.
[29] Zhang N, Hu ZG, Springer C, et al. A bi-level integrated generation-
[14] Pandzic H, Kuzle I, Capuder T. Virtual power plant mid-term dispatch
transmission planning model incorporating the impacts of demand response
optimization. Appl Energy 2013;101:134–41.
by operation simulation. Energy Convers Manage 2016;123:84–94.
[15] Tascikaraoglu A, Erdinc O, Uzunoglu M, et al. An adaptive load dispatching and
[30] Molderink A, Bakker V, Bosman MGC. Management and control of domestic
forecasting strategy for a virtual power plant including renewable energy
smart grid technology. IEEE Trans Smart Grid 2010;1(2):109–19.
conversion units. Appl Energy 2014;119:445–53.
 L. Ju et al. / Energy Conversion and Management 128 (2016) 160–177 177

[31] Zakariazadeh A, Jadid S, Siano P. Stochastic operational scheduling of smart [40] Jiang YW, Chen C, Wen BY. Application of stochastic simulation’s particle
distribution system considering wind generation and demand response swarm algorithm in the compensation of reactive power for wind farms. Proc
programs. Int J Electr Power Energy Syst 2014;63:218–25. CSEE 2008;28(13):47–52.
[32] Tan ZF, Ju LW, Brent R, et al. The optimization model for multi-type customers [41] Amjady N, Aghaei J, Shayanfar HA. Stochastic multiobjective market clearing of
assisting wind power consumptive considering uncertainty and demand joint energy and reserves auctions ensuring power system security. power
response based on robust stochastic theory. Energy Convers Manage systems. IEEE Trans Power Syst 2009;24(4):1841–54.
2015;105:1070–81. [42] Ju LW, Tan ZF, Li HH. Multi-objective synergistic scheduling optimization
[33] Ai X, Liu X. Dynamic economic dispatch for wind farms integrated power model for wind power and plug-in hybrid electric vehicles under different
system based on credibility theory. Proc CSEE 2011;31(S1):12–8. grid-connected modes. Math Probl Eng 2014:15.
[34] Hananeh F, Alireza Z, Shahram J. The role of demand response in single and [43] Zou YY, Yang L. Synergetic dispatch models of a Wind/PV/hydro virtual power
multi-objective wind-thermal generation scheduling: a stochastic plant based on representative scenario set. Power Syst Technol 2015;39
programming. Energy 2014;64:853–67. (7):1855–60.
[35] Wang Y, Ai X, Tan ZF, et al. Interactive dispatch modes and bidding strategy of [44] Yu S, Wei ZH, Sun GQ, et al. A bidding model for a virtual power plant
multiple virtual power plants based on demand response and game theory. considering uncertainties. Autom Electr Power Syst 2014;38(22):44–9.
IEEE Trans Smart Grid 2015;9:1–10. [45] Songli FAN, Qian AI, Xing HE. Risk analysis on dispatch of virtual power plant
[36] Rao R, Zhang XP, Xie J, Ju LW. Optimizing electric vehicle customers’ charging based on chance constrained programmingJ]. Proc CSEE 2013;35(6):4025–34.
behavior in battery swapping mode. Appl Energy 2015;155:547–59. [46] Amjady N, Aghaei J, Shayanfar HA. Stochastic multiobjective market clearing of
[37] Dai Z, Zheng XT. Design of close-loop supply chain network under uncertainty joint energy and reserves auctions ensuring power system security. Power
using hybrid genetic algorithm: a fuzzy and chance-constrained programming Syst, IEEE Trans Power Syst 2009;24(4):1841–54.
model. Comput Ind Eng 2015;88:444–57. [47] Yang FB, Zhang HG, Bei C, et al. Parametric optimization and performance
[38] Tan ZF, Ju LW, Li HH, Li JY, Zhang HJ. A two-stage scheduling optimization analysis of ORC (organic Rankine cycle) for diesel engine waste heat recovery
model and solution algorithm for wind power and energy storage system with a fin-and-tube evaporator. Energy 2015;91:128–41.
considering uncertainty and demand response. Int J Electr Power Energy Syst
2014:1057–69.
[39] Wang X, Li LX, Zheng YH, Xu QS. Reactive power optimization for wind power
system based on dynamic cloud evolutionary particle swarm optimization.
Power Syst Prot Control 2013;41(24):36–43.