Sustainable Cities and Society 41 (2018) 332–340

Contents lists available at ScienceDirect

Sustainable Cities and Society

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

A two stage hierarchical control approach for the optimal energy T

management in commercial building microgrids based on local wind power
and PEVs
⁎
Mehdi Tavakolia, Fatemeh Shokridehakia, Mousa Marzbandb, Radu Godinac, Edris Pouresmaeild,
a
Babol (Noshirvani) University of Technology, Babol, Iran
b
Faculty of Engineering and Environment, Department of Maths, Physics and Electrical Engineering, Northumbria University Newcastle, Newcastle upon Tyne, NE1 8ST,
UK
c
C-MAST, University of Beira Interior, R. Fonte do Lameiro, Covilhã, 6201-001, Portugal
d
Department of Electrical Engineering and Automation, Aalto University, 02150, Espoo, Finland

A R T I C LE I N FO A B S T R A C T

Keywords: The inclusion of plug-in electrical vehicles (PEVs) in microgrids not only could bring benefits by reducing the on-
Demand response (DR) peak demand, but could also improve the economic efficiency and increase the environmental sustainability.
Model predictive control (MPC) Therefore, in this paper a two stage energy management strategy for the contribution of PEVs in demand re-
Conditional value at risk (CVaR) sponse (DR) programs of commercial building microgrids is addressed. The main contribution of this work is the
Plug-in electric vehicles (PEV)
incorporation of the uncertainty of electricity prices in a model predictive control (MPC) based plan for energy
Wind power
Commercial building microgrids
management optimization. First, the optimization problem considers the operation of PEVs and wind power in
order to optimize the energy management in the commercial building. Second, the total charged power reference
which is computed for PEVs in this stage is sent to the PEVs control section so that it could be allocated to each
PEV. Therefore, the power balance can be achieved between the power supply and the load in the proposed
microgrid building while the operational cost is minimized. The predicted values for load demand, wind power,
and electricity price are forecasted by a seasonal autoregressive integrated moving average (SARIMA) model. In
addition, the conditional value at risk (CVaR) is used for the uncertainty in the electricity prices. In the end, the
results confirm that the PEVs can effectively contribute in the DR programs for the proposed microgrid model.

1. Introduction most attractable sources of RERs which has shown a rapid growth due
to its proficiency and accessibility (Zeng, Zhao, & Yang, 2013). How-
The power system has been facing various technical challenges with ever, its unpredictable nature makes it an uncertain source of energy
the increasing number of distributed generation units mainly based on which aggravates the conditions of reaching the power balance. Con-
RERs such as wind and photovoltaic which are utilized especially close sequently, what is desirable in a microgrid with RERs is to compensate
to the load centers in order to benefit the grid in reducing the on-peak the variation in their power generation. On the other hand, the use of
demand and decreasing the congestion of generation units (Ashabani & PEVs has recently increased significantly due to its advantages over the
Mohamed, 2012; Pouresmaeil, Miguel-Espinar, Massot-Campos, internal combustion engine vehicles due to environmental concerns and
Montesinos-Miracle, & Gomis-Bellmunt, 2013). To tackle these pro- fossil fuel consumption. Nevertheless, PEVs are regarded as high power
blems, a solution could be using microgrids, which are capable of al- consumers since they consume a remarkable amount of power
lowing a bidirectional power flow between generation units and load throughout charging procedure which may lead to power shortages,
demands. They facilitate the interconnection of distributed power increase on-peak demand, and difficulties related to the reliability of
generation units, energy storage systems and different kinds of load in a the power grid (Qian, Zhou, Allan, & Yuan, 2011). Challenges related to
restricted zone (Olken, 2013). Microgrids are expected to operate in using PEVs such as power losses, voltage fluctuations and the power
two modes, i.e., grid connected mode and islanded mode. It is necessary quality deterioration in the current distribution systems are stated in
for each microgrid to have a balanced power between generators and Clement-Nyns, Haesen, and Driesen (2011). Thus, the use of PEVs along
consumers (Sortomme & El-Sharkawi, 2011). Wind power is one the with the presence of wind power in microgrids in an uncoordinated

⁎
Corresponding author.
E-mail address: edris.pouresmaeil@aalto.fi (E. Pouresmaeil).

https://doi.org/10.1016/j.scs.2018.05.035
Received 14 January 2018; Received in revised form 21 February 2018; Accepted 21 May 2018
Available online 26 May 2018
2210-6707/ © 2018 Elsevier Ltd. All rights reserved.
M. Tavakoli et al. Sustainable Cities and Society 41 (2018) 332–340

Nomenclature NPEV,ava Number of plugged-in PEVs

SOCt State of the charge at the tth step time
Indices V2G Vehicle to grid
DR Demand response
i Index of the PEV PEVs Plug-in electric vehicles
k Index of time step in the first stage controller RERs Renewable energy resources
CVaR Conditional value at risk
Abbreviations
Variables
SOC State of the charge
SARIMA Seasonal autoregressive integrated moving average CtGD The electricity prices transferred from grid to load at the
Costt (Xtij , Ct ) Total operational cost of the commerial bulding mi- tth step time
crogrid CtGEV The electricity prices transferred from grid to PEV at the
Dt The base load of commercial building microgrid at the tth tth step time
step time CtEVG The electricity prices transferred from PEV to grid at the
XtW Total generated power by wind at the tth step time tth step time
Xtij The power transfered from unit i to unit j at the tth step CtWG The electricity prices transferred from wind to grid at the
time tth step time
PPEV , i Charging/discharging power of the ith PEV ηC Charging efficiency of the PEVs battery
ref
PPEV , agg Optimum charge/discharge power reference for ag- ηD Discharging efficiency of the PEVs battery
gregated PEVs w Weighting factor for the price risk consideration
PPEV,agg The charging power of aggregated PEVs. For PPEV,agg> 0,
PEVs are in charging mode and for PPEV,agg< 0, PEVs are Parameters
in discharging mode
EPEV,agg Energy stored in the PEVs aggregator c Power to energy conversion factor
EPEV,arr Energy stored in the arriving PEVs Pkm Power consumed per kilometer
EPEV,arr,i Energy stored in the ith PEVs when it arrives CBattery Battery capacity of each PEV
EPEV,dep Energy stored in the departing PEVs SOCmin Minimum state of the charge of a PEV
NPEV,arr Number of PEV arriving commercial building in each step SOCmax Maximum state of the charge of a PEV
time PPEV,max Maximum charging rate of each PEV
di Daily driving distance of the ith PEV Ct The electricity price at the tth step time

manner could bring some practical challenges for the demand side attention from the research community (Clement-Nyns et al., 2011;
management due to their unpredictable nature. This incoordination Qian et al., 2011; Wang et al., 2012).
also increases on-peak demand since the PEV owners tend to connect Several studies from the literature have been focusing on the con-
the PEV to the charging station in the evening when they arrive home tribution of the local power generations (i.e., wind) in different load
or in the morning when they arrive at the work place. Besides, the PEVs sectors (Ansari, Al-Awami, Sortomme, & Abidoeric, 2015; Bai & Qiao,
remain in stand-by mode up to 96 percent of the time in a day and this 2015; Kennel, Gorges, & Liu, 2013; Li, Ahn, Peng, & Sun, 2013;
time is more than enough for their battery to be fully charged (Wang, Meegahapola, Robinson, Agalgaonkar, Perera, & Ciufo, 2014; Rassaei,
Wang, Dounis, & Yang, 2012). In addition, in contrast to other electrical Soh, & Chua, 2015; Soares, Morais, Sousa, Vale, & Faria, 2013; Su &
appliances, the integration of several PEVs simultaneously connected in Chow, 2012a, 2012b; Tan, Yang, & Nehorai, 2014; Wang et al., 2011;
parking station could be considered as a potential source for the DR in Wu, Yang, Bao, & Yan, 2013; Xu, Hu, Song, Zhao, & Zhang, 2014). In
the power system. This is justified by the amount of the aggregated Wu et al. (2013), the coordination between wind power and PEVs is
PEVs charging load by being regarded as a bulk load demand which presented by three power dispatching methods in order to create a
could be shifted or interrupted during while it is connected to the balance between the power generation and the consumers in the mi-
charging station. So, a coordinated charging schedule for PEVs with crogrid. In Kennel et al. (2013) and Li et al. (2013) a control scheme is
power system DR could have a significant improvement on DR through presented based on integrating PEVs and wind power in order to pre-
an appropriate management system. In this way, all the PEVs could be serve the grid frequency in a desired range. In Wang et al. (2011) a
entirely charged while the DR is satisfied by shifting or interrupting the model is employed with the aim of providing DR by considering the
charging of PEVs. For this purpose, the aggregated PEVs could be cooperative performance between PEVs and wind generators. Four
considered as a battery energy storage device in the microgrid and they settings are recommended for PEVs and it is concluded that the whole
could be used for reducing the on-peak demand and also for compen- operational cost of the system has been reduced through an optimized
sating the oscillation of RERs. PEVs could discharge their battery with charging. In Tan et al. (2014) and Rassaei et al. (2015) PEV charging is
vehicle-to-grid (V2G) technology when they are parked in the parking analyzed under a DR program in the upcoming smart grids in order to
station. For this to function, the owners can choose whether they allow reduce the peak demand of the grid. In Tan et al. (2014) the PEVs
their vehicle to participate in this program or not. The benefit is that it charging pattern and the stochastic nature of the initial SOC is modelled
allows for the microgrid to discharge their battery when a power im- by a normal distribution function. However, the final requested level of
balance occurs or when the electricity price is high. In this way, a SOC is not considered in that study. In addition, the uncertainty of PEVs
portion of the load dedicated to the charging PEVs could be curtailed so in Rassaei et al. (2015) is taken into account based on real-life practice.
that they can be charged when the electricity price is cheaper or the In Soares et al. (2013) different kinds of methods comprising V2G, DR
load is lower instead of charging immediately after connecting to the programs, uncontrolled charging and smart charging of PEVs are
charging station. As a result, using the PEVs for DR purposes and also compared for the day-ahead energy management. In Bai and Qiao
coordinating between PEVs and renewable generation in order to im- (2015) a robust optimization model is presented for a large-scale de-
prove the power balance in microgrids has been getting an increasing ployment of PEVs which are considered as a dispatchable energy

333
M. Tavakoli et al. Sustainable Cities and Society 41 (2018) 332–340

storage device for DR purposes in order to allow them to supply the 2. System under study
load demand and contribute in ancillary services. Overall, with the aim
of shifting the bulk demand of PEVs to off-peak times the most of PEVs The system studied in this paper is a local wind power-based com-
charging management is based on the different kinds of electricity mercial microgrid with several PEVs and load demand (which is the
pricing proposed by the utility, such as setting the price based on real base load without PEV load) and can be observed in Fig. 1. This mi-
time (Su & Chow, 2012a, 2012b), day-ahead (Sortomme & El-Sharkawi, crogrid is connected to the main grid in order to have the capability of
2011; Ansari et al., 2015), or the consumption time (Xu et al., 2014). operation in both islanded and grid connected modes.
Nevertheless, in most of these publications, the uncertainties related to In this study the PEVs are considered as a load or a power supplier,
electricity price were not taken into account which may lead to in- depending if they’re charging or discharging, respectively. Only pri-
efficient solutions, especially when the price prediction is not very ac- vately owned PEVs are considered for this purpose. It is assumed that
curate. In addition, the majority of these works only focus on the only the PEV owners that are the commercial building’s employees
charging process of PEVs, but less attention is paid to the coordination connect their PEV to the charging station in the morning when they
between PEVs and wind power generation for the purposes of DR. arrive at work. In addition, all of them disconnect their PEV at the
Due the reasons mentioned above, this paper presents a two stage 9 P.M. For this purpose the arrival time is a stochastic variable but the
control approach for DR management based on predictive control. departure time is deterministic one. The PEVs can be used as an energy
Initially, the PEVs and wind power unit coordination is made for the storage system to inject power to the microgrid during on-peak demand
microgrid in order to create a power balance between power generation periods and absorb the excess power of wind power source during off-
and demand. Then, the calculation of the aggregated PEVs power peak periods. The time horizon for the modelled microgrid is assumed
command is sent to the PEVs for analysis. The PEVs controller section to be between 9:00 A.M. to 9:00 P.M. which often contains one period
optimally allots the computed power for charging or discharging to of on-peak and one period of off-peak load. In addition, in order to
each individual PEV. To solve this problem an online predictive control entice the PEVs owner to participate in this program, an incentives
problem is proposed with some physical constraints for the microgrid must be dedicated to them. Furthermore, some of the PEV owners
(such as working hours, minimum and maximum charge level for bat- (employees) register themselves as dispatchable PEVs and allowing to
teries, charging and discharging rate limits and so on). The considered be demand response candidates. In this way, they commit to be
microgrid in this paper is a small to medium size commercial building. plugged-in during the day in order to be utilized by the commercial
Such a microgrid typically uses low voltage distribution system and has building if required. Understandably, despite the dedicated incentives,
approximately up to 1 MW of consumed power under its peak demand some PEVs owner disconnect their PEVs from the charging station to
(Meegahapola et al., 2014). Finally, the key contribution of this work is depart from the commercial building. When this occurs it is represented
to utilize PEVs as a potential DR tool in the commercial building mi- as white Gaussian noise in the formulation. In addition, the amount of
crogrid while incorporating the uncertainty of the electricity price in energy stored in the PEVs which arrive at the commercial building at
the optimization model by using the CVaR. time step k, is a disturbance to the model and is represented as a sto-
The remaining of this paper is organized as follows. The considered chastic variable EPEV,arr(k). The total number of PEVs is considered to be
case study and its model are presented in Section III. The proposed 100 in this paper. The amount of EPEV,arr(k) depends on the number of
control approach design is explained in Section IV. Simulation results PEVs, NPEV,arr(k) and their initial energy (EPEV,arr,i) while connecting to
are presented and discussed in Section V in order to verify the suggested the charging station. Furthermore, regarding the stochastic character-
plan. Finally, the paper is concluded in Section VI. istic of the PEVs driving pattern, a statistical model is used for the ar-
rival times and driving distances by considering the statistics of the

Fig. 1. The schematic of the proposed under study model.

334
M. Tavakoli et al. Sustainable Cities and Society 41 (2018) 332–340

internal combustion engine vehicles as an approximate reference Lee, paper in approach, but their model is more inclusive and they consider
Adornato, and Filipi (2012). By using such data and by using the other objects such as different kinds of PEVs, the speed, acceleration
normal distribution, the maximum likelihood estimation and their and the travelling time of PEVs.
curve-fitting, NPEV,arr(k), can be attained. Generally, most of the PEVs
arrive before 1 P.M. and thus NPEV,arr(k) can be modelled by a truncated
normal distribution between 9 A.M. and 1 P.M. according to Eq. (1). In 3. The proposed control approach design
addition, from 1 P.M. to 9 P.M., NPEV , arr (k ) is assessed to be zero.
Uncoordinated use of PEVs with wind power in the proposed mi-
1 (NPEV , arr (k ) − μN , k )2 crogrid exacerbate the power mismatch between power generation and
p (NPEV , arr (k )) = × exp(− )
σNPEV , arr , k 2π 2σN2PEV , arr , k (1) load demand due to the stochastic charactristic of the wind power
generation and PEVs. This is unfavorable for both microgrid and main
In which NPEV,arr predicts the number of the PEVs arriving at the grid and creates difficulties for the demand side management. As a
commercial building while NPEV,ava is the accumulation of NPEV,arr for result, with the aim of achieving an optminal operation with the
each time step and is obtained as follows: minimum possible cost for the commercial building, the charging pro-
NPEV , ava (k + 1) = NPEV , ava (k ) + NPEV , arr (k ) (2) cedure of PEVs must be done in a way to enhance the power balance
between the generation and load in order to help the demand side
Also, NPEV,arr is an inherently uncertain variable and this makes the control. In addition, it is expected the PEVs to be fully charged before
NPEV,ava a stochastic variable according to Eq. (2). In order to simplify disconnecting from the charging station. Also, it is essential to make
the overall system, the uncertainty related to NPEV,ava is considered in sure that PEVs are fully charged at the anticipated plug-out time. In
the first stage of the proposed approach and so in the second stage it is order to achieve these goals, a two stage hierarchical control approach
only taken into account the deterministic one. For this reason, the PEVs is presented in this paper. In the first stage the power reference for PEVs
arriving at the commercial building at the kth time step, are allowed to aggregator is specified by an optimization problem which will be dis-
be charged or discharged from the next step time that is (k+1)th step cussed in section IV-C. In the second stage, the aggregated charging
time. Thus, the number of PEVs in the kth step time which is required to power is allocated to each PEVs by its control section. In this way, the
be managed by the PEV control section is NPEV,ava(k – 1) which is a first control set is applied to the proposed microgrid. In the next stage,
deterministic quantity. the actual values of the system are measured to repeat the process
In addition, the daily distance covered by each PEV can be modelled above. Thus, the feedback controller is used to update the values for the
approximately by a logarithmic normal distribution with the mean optimization process in the next iteration since it depends on the new
μd = 3.20 and variance σd = 0.88 as presented in (3). measured states. However, for the expansion of the proposed control
1 (ln di − μd )2 strategy, a prediction technique is needed for the load, electricity price,
p (di ) = × exp(− ) and wind power.
dσd 2π 2σd2 (3)
Therefore, the EPEV,arr,i can be obtained as follows:
3.1. Prediction of the wind power, load demand, and electricity price
EPEV , arr , i = CBattery − di Pkm (4)
In order to describe the uncertain characteristics of the electricity
Now it is possible to estimate the EPEV,arr,i(k) based on the previous
prices, of the time-varying load demand and wind power, a prediction
equations. Initially, μ NPEV , arr (k ) is computed, which is the number of
model is required. The probabilistic planning of each stochastic process
PEVs to be added in the time step k. Then the EPEV,arr(k) can be pre-
can be defined by finding the joint distribution of its random variables
dicted as:
which describes both the probabilistic manner of each random variable
μN
PEV , arr , k on its marginal distributions and the interrelations which existed
EPEV , arr (k ) = ∑ EPEV , arr , i (k ) among all of variables (statistical dependencies). In this way, auto-
i=1 (5)
regressive moving average (ARMA) models which rely on these two
Finally, by observing Fig. 1, the discrete time model for the pro- principles is a fitting choice. However, numerous of events periodically
posed commercial microgrid can be expressed by: (e.g., daily) show a seasonal trend, that means there is a relationship
between the observations made during the similar period in successive
E , agg (k + 1) = [1] EPEV , agg (k ) + [ηc ] PPEV , agg (k ) + [1] EPEV , arr (k )
PEV
    periods. Besides the seasonal link, there is also a relationship between
X (k + 1)
⏟A ⏟ B
X (k )
⏟b/
U (k ) ZE
the observations made during sequential periods. This fact can be seen
− [1] EPEV , dep (k ) 
P (k ) = [1] PPEV , agg (k ) in the load demand which shows a similar behavior every day and every
⏟  E ⏟ 
c/ Zw Y (k ) C
U (k ) (6) week, establishing an instance in both daily and weekly seasonality. In
these situations a seasonal autoregressive integrated moving average
where, EPEV,agg is modelled as the space state variable, PPEV,agg is re- model is needed and is known as SARIMA (an extension of the ARMA
garded as the control input, EPEV,arr is a stochastic disturbance, and model) which considers seasonality and potential seasonal unit (http://
EPEV,dep is considered to be white Gaussian noise. Yet, it is assumed that www.nyiso.com/public/markets_operations/market_data/load_data/
most of the PEVs remain plugged-in throughout the day due to the index.jsp). Suppose y as a stochastic process with a seasonality of order
influence of several incentives designed for this purpose. Hence, the S. The general expression of a seasonal ARIMA model with parameters
value of EPEV,dep is much smaller than EPEV,arr. For this reason, EPEV,dep is (p , d, q) × (P , D, Q)S can be expressed by:
not modelled in details. However, in order to consider the unmeasured
p P q Q
disturbances in the system, a feedback controller is needed for com-
(1 − ∑ ϕj B j)(1 − ∑ φj B jS )(1 − B )d (1 − B S ) Dyt =(1 − ∑ θj B j)(1 − ∑ ϑj B jS ) εt
pensation. So, a simple solution is to create a model predictive control j=1 j=1 j=1 j=1
(MPC) controller for this purpose which is later explained in detail.
(7)
Furthermore, it should be emphasized that other methods of modelling
exist regarding the stochastic nature of the initial energy of PEVs With a seasonal component of P autoregressive parameters
(Alizadeh, Scaglione, Davies, & Kurani, 2014; Bashash & Fathy, 2012; φ1, φ2 , ..., φP , Q moving average parameters ϑ1, ϑ2 , ..., ϑQ and a differ-
Huang, Jia, Qiu, Guan, & Geconinck, 2015; Lee et al., 2012; Qian, Zhou, entiation order D. In this paper the wind power generation, load de-
& Yuan, 2015; Sechilariu, Wang, & Locment, 2013; Shahidinejad, mand, and electricity prices are characterized and predicted by a sea-
Bibeau, & Filizadeh, 2010; Tan & Wang, 2014). They are similar to this sonal ARIMA model.

335
M. Tavakoli et al. Sustainable Cities and Society 41 (2018) 332–340

3.2. Uncertainty consideration priority at each step time to supply the base load. In addition, the ag-
gregated power stored in the PEVs could be used to meet the demand
In this study f (x,y) is the loss related to a set of decision variables x, power. Finally, if the wind power and PEVs can not supply the demand,
and is selected from a certain subset X of Rn and the random variable y then the main grid is responsible for injecting the shortage power to the
of Rm. The vector x can be taken as the set of available portfolios, commercial building. It assumed that the considered commercial
whereas the vector y points out the uncertainty set. The aim is to building is open from 9 A.M. to 9 P.M. and this period could be dis-
achieve a value for the decision vector x that can minimize the cost cretized into N N intervals of length Δt . Each time step is considered to
function, conditioned by the uncertainty in vector y. Value at risk (VaR) be 5 min in this paper. As a result, 144 points are obtained from 9 A.M.
is one of the most applicable risk measurements and is particularly to 9 P.M. for t = 1, 2, ..., 144 , there are seven decision variable as fol-
fitting for loss distribution functions with fat tail behaviors. For a spe- lows:
cified confidence level β, the VaR is the smallest loss over the rolling
Xt = (XtGD , XtGEV , XtWD , XtWEV , XtWG , XtEVD , XtEVG ) (13)
horizon time which is exceeded with probability 1–β according to fol-
lowing equation: The superscript G, W, EV, and D denotes the grid, wind, PEVs and
base load demand (without PEVs) of the commercial building, respec-
VaRβ = min{α ∈ R: P {f (x , y ) ≤ α } ≥ β } for 0 ≤ β ≤ 1 (8)
tively. With this assumption and the fact that the intent is to minimize
Even though VaR is a familiar risk measure used in economic pro- the operational cost of the commercial building, the total cost of ex-
blems, it is a non-coherent risk measure which has negatives points changed power in each step time can be obtained as:
such as non-convexity, non-smoothness, subadditivity etc., which make
Costt (Xt , Ct ) = (XtGD CtGD + XtGEV CtGEV ) − (η DXtEVG CtEVG + XtWG CtWG )
it an unattractive option in the optimization programming. With the
purpose of avoiding this difficulty, there is a desirable alternative risk (14)
measure known as conditional value at risk (CVaR) also recognized as For the sake of simplicity, identical values are considered for prices
average value at risk or mean shortfall. For a given confidence level β, i.e. CtGD = CtGEV = CtEVG = CtWG = Ct , which is acceptable in most en-
CVaR is defined as: ergy markets. Thus, the cost function will be equal to:
β − CVaR = Ey (f (x , y )|f (x , y ) ≥ β − VaR) (9) Costt (Xt , Ct ) = Ct . [(XtGD + XtGEV ) − (η DXtEVG + XtWG )]. (15)

It specifies the expected conditional value of the cost function, Nevertheless, due to the volatile and unpredictable nature of elec-
conditioned by this value being larger than β-percentile. In contrast to tricity prices, the strategy above couldn’t be an appropriate manner for
the traditional robust optimization techniques, the minimization of optimal energy management since abandons the uncertainty of elec-
CVaR is a flexible selection for choosing the objective function. It is tricity prices. Therefore, it is essential to consider the risk and un-
capable to improve the optimization performance greatly since it uses certainty while seeking an optimal approach. For this reason, the final
distributional data on the uncertain parameter of y. In fact, minimizing cost function in (16) is obtained by incorporating the cost function
the CVaR of the cost leads to the minimization of the risk of the system represented in (15) into the minimization of the conditional value at
being exposed to high losses. In addition, for linear cost function pro- risk formulation in (10). In addition, the aggregated charging power of
blems minimizing CVaR can be expressed as a linear programming PEVs must be maintained in proper limits. If not, the PEV aggregated
problem which is an attractive choice in practical applications. The batteries will be overcharged when the wind density is high and it won’t
CVaR can be approximately formulated by: be fully charged when the wind density is low. If this occurs for a long
M
time, it will remarkably decrease the life time of the battery. Finally,
1 considering risk in electricity prices besides regarding battery limita-
CVaRβ = min(α +
M (1 − β )
∑ [f (x , yi ) − α]+)
i=1 (10) tions and constraints, leads to a linear programming problem as fol-
lows:
where, indicates the positive elements of z, α is the β-VaR, M is the
z+
N
number of Monte Carlo paths to estimate the expected value of
β − CVaR in the cost function, and yi indicates the ith generated path of
Min [ ∑ (Costt (Xt , Ct )) + (w × CVaRβ )]
T=1 (16)
the uncertain variable. In order to solve this problem, it is generally
proposed to substitute the 0+ with a set of constraints. Thus, the cor- Subject to:
responding equation for minimizing CVaR is formulated as follows: (XtGD CtGD + XtGEV CtGEV ) − (η DXtEVG CtEVG + XtWG CtWG ) ≤ z i + α (17)
M
1 zi ≥ 0 (18)
CVaRβ = min(α +
M (1 − β )
∑ z i) Subject to: z i ≥ 0, z i ≥ f (x , yi ) − α
i=1
XtWD = min{Dt , XtW } (19)
(11)
XtW = XtWEV + XtWD + XtWG (20)
3.3. Problem formulation of the proposed approach
Dt = XtGD + η DXtEVD + XtWD (21)
As it is mentioned above, the proposed control approach contains (XtEVD + XtEVG ) − ηC (XtGEV + XtWEV ) ≤ [SOC − SOCmin]. NPEV , ava. CBattery
two stages. In the first stage, an optimization problem is solved to in
order to minimize the operational cost related to the exchanged power (22)
between the main grid and microgrid. It is designed in a way to improve ηC (XtGEV + XtWEV ) − (XtEVD + XtEVG ) ≤ [SOCmax − SOC ]. NPEV , ava. CBattery
the demand side management by utilizing wind power and PEVs ag-
(23)
gregated power in a coordinated manner. In order to yield this purpose,
the following cost function has to be minimized: ηC (XtGEV + XtWEV ) ≤ ΔS C . CBattery (24)
N
Min ∑ Costt (Xtij , Ct ) (XtEVD + XtEVG ) ≤ ΔS D. CBattery (25)
T=1 (12)
Constraint (19) indicates that the whole power generated by the
Considering the prposed system in Fig. 1, wind power is the first wind is injected to the load demand when the wind power is lower than

336
M. Tavakoli et al. Sustainable Cities and Society 41 (2018) 332–340

commercial building base load. When the power generated by the wind Some of the arriving PEVs may have a SOC lower or higher than the
is higher than commercial building demand, the excess power will be specified limits. In this situations it is impossible to handle their SOC in
injected to the PEVs battery or the main grid. Constraint (21) shows one cycle in order to come in the desirable range [SOCmin, SOCmax ] be-
that the power demand of the commercial building is completely sup- cause of the maximum charge and discharge rate for PEVs. For this
plied by the grid, battery storage or wind power generation. The values reason, auxiliary variables λPEV , l, i , λPEV , u, i are introduced to soften these
of SOCmin and SOCmax are the minimum and maximum allowable charge limits. In addition, some PEVs connect to the charging station with low
level of the PEV battery and constraints (22) and (23) imply that the charging levels and for short periods of time. In this occasion, they
battery storage level in the next step time remains between two limits. cannot be fully charged before disconnecting from the charger. Thus, it
This occurs with the intent of preventing the battery to be damaged is impossible to make every PEV be completely charged before the plug-
during the charging process. In order to ensure that PEVs are fully out time. In this situation, it is assumed that all of the PEVs are almost
charged at the expected plug-out time that is 9 P.M. and the fact that fully charged before the use. This is attainable by considering the
they require a definite time for fully charging, it is a good decision to auxiliary variables in the constraints (30) and (31) which make the PEV
change the values of SOCmin and SOCmax according to the time. For this control section charge the PEVs battery as much as possible. Solving the
reason, in this paper, a typical setting for SOCmin and SOCmax of a li- problem in (27) with the constraints in (28)–(32) offers the optimum
thium-ion battery is considered in the Table 1. Moreover, the inequa- charging command for all of the PEVs connected to the charging sta-
tions 0 ≤ ΔS C ≤ 1 and 0 ≤ ΔS D ≤ 1 illustrate the maximum amount for tion.
charging and discharging rate and constraints (24) and (25) prevent the
battery to be charged or discharged faster than the acceptable rates 4. Simulation results
during each step time. After finding the decision variables above by
solving the optimization problem, the SOC of the aggregated PEVs is In order to investigate the accuracy of the proposed control ap-
obtained according to (6) as: proach in this paper, a simulation is carried out in this section. For this
EPEV , agg (t ) purpose, a system presented in section II (Fig. 1) is used to examine the
SOC (t ) = effectiveness of the suggested strategy. The considered microgrid which
NPEV , ava. CBattery (26)
consists of 1.5 MW wind power generation, 100 PEVs and base load that
Solving this problem leads to an optimal control action which is the has a bidirectional power flow with a main grid. The data of the elec-
first of them applied to the proposed microgrid. Then, this process is tricity prices is provided from (http://www.nyiso.com/public/markets_
repeated by the actual values of the system such as operations/market_data/load_data/index.jsp). The base load demand
EPEV , agg (t + 1), EPEV , arr (t + 1) and PPEV , agg (t + 1) . In addition, the belongs to a typical commercial microgrid. It is assumed that all of the
feedback controller is utilized to guarantee that the optimization pro- PEVs are identical. The battery storage parameters of the PEVs is dis-
blem uses the new measured values, and in this way, the impact of the played in Table 1. In order to have a reasonable approximation for the
noise EPEV , dep is compensated. base load of commercial building, electricity prices and wind power
generation, the SARIMA model is used for forecasting in case there is a
3.4. PEVs control section design seasonality effect in their modeling. In Fig. 2, the base load of com-
mercial building and wind power generation are shown over the time
At the second stage, the PEVs control section is responsible for al- horizon of 12 h between 9 A.M. and 9 P.M. The electricity price is il-
locating the charging power from the aggregated PEVs power PPEV , agg , lustrated in Fig. 3. As it is clear from Fig. 2, the wind power is occa-
which is specified by optimization problem in the first stage, to each sionally higher than the base load. This implies that the excess power
individual PEV. As discussed in section II, all of the PEVs arriving in the generated by wind cause a power imbalance in the commercial
tth step time, are allowed to be charged or discharged from the next step building. In addition, when there is a power shortage, the same problem
time i.e., (t + 1)th step time. For this reason, the number of PEVs con- occurs. Consequently, in order to improve the demand side manage-
sidered for management in the tth step time is equal to NPEV , ava (t −1) ment, the first objective is to create a power balance between the
which is a deterministic number. So, it is possible to design the PEV generation and the consumption in a way so that the operational cost
control section in a deterministic framework compared to the stochastic could be minimized. In addition, the second objective is to satisfy the
from of first stage, because the stochastic nature of arriving PEVs is PEV control section in order to charge PEVs as much as possible so that
considered in the first stage. Regarding these matters and the PEV they could have a satisfactory charging level at the plug-out time. On
model in (1)–(5), the PEV control section can be expressed as follows: the other hand, the forecasting electricity prices are generally accom-
panied with some uncertainty which cause a large effect on the energy
NPEV , ava (t − 1)
management strategy and make it more complex. Disregarding this
β [ ∑ PPEV , i (t )−PPEV , agg t ( )]2 uncertainty may lead to a non-optimal management. For these reasons,
i=1
NPEV , ava (t − 1) in this paper the CVAR is incorporated in the optimization problem as
+ (1 − β ) ∑ (λPEV , l, i (t )2+λPEV , u, i (t )2) in Eq. (16) in order to decrease the risk associated with electricity
i=1 (27) prices. In Fig. 4, SOC of the aggregated PEVs are depicted for three
Subject to: confidence level β = 0.90, β = 0.95 and β = 0.99. The minimum and
maximum value for SOC are also demonstrated in this figure. In addi-
|PPEV , i (t )| ≤ PPEV ,max i = 1, 2, ..., NPEV , ava (t − 1) (28) tion, Fig. 5 indicates the comparison of SOC in two different stats. The
first is the SOC of the aggregated PEV using the proposed control
0 ≤ EPEV , i (t ) + ηcPPEV , i (t ) ≤ CBattery i = 1, 2, ..., NPEV , ava (t − 1) (29)
Table 1
EPEV , i (t ) + ηcPPEV , i (t ) + λPEV , l, i (t ) i = 1, 2, ..., NPEV , ava (t − 1) Minimum and Maximum Values for the SOC of PEVs.
≥ SOCmin . CBattery (30) Time SOCmin SOCmax

EPEV , i (t ) + ηcPPEV , i (t ) − λPEV , u, i (t ) i = 1, 2, ..., NPEV , ava (t − 1) From 9:00 A.M. to 10:40 A.M. 20% 80%
From 10:40 A.M. to 12:20 P.M. 40% 90%
≤ SOCmax . CBattery (31) From 12:20 A.M. to 3:40 P.M. 50% 95%
From 3:40 A.M. to 7:00 P.M. 70% 100%
λPEV , l, i (t ), λPEV , u, i (t ) ≥ 0 i = 1, 2, ..., NPEV , ava (t − 1) (32) From 7:00 P.M. to 9:00 P.M. 90% 100%

337
M. Tavakoli et al. Sustainable Cities and Society 41 (2018) 332–340

Fig. 5. SOC of the PEVs with and without the proposed control, and
NPEV,ava*Cbattery.
Fig. 2. Base load and wind power.

Table 2
IParameters for PEVs Battery Energy Storage.

ΔS D ΔS C ηC ηD

0.10 0.10 0.95 0.9

problem considers the forecasted energy price in each time step and the
associated uncertainty of it, in order to minimize the total operational
cost of the commercial building. As the confidence level rises, the op-
erational cost shows a trivial increase. But the system operation will be
more robust against price variability. In addition, it is revealed from the
Fig. 4 that the aggregated charging power of PEVs for all three con-
fidence levels are kept between 20% and 80% in most of the time in
order to increase the life time of the PEV batteries. Furthermore, as the
time was getting closer and closer to the plug-out time, the charging
level of the aggregated PEVs increases gradually so that the maximum
charge level could be reached which means that each PEV is nearly fully
Fig. 3. Electricity Price. charged before disconnecting from the charging station (Table 2).
When there isn’t any control of PEVs, they are charged directly after
connecting to the charging station until they are fully charged. This
may increase the on-peak demand and increase the system cost. In
addition, it may exacerbate the power mismatch in the microgrid.
However, by using the proposed approach in this paper it is possible to
minimize the power imbalance in the microgrid and help the demand
response management. In this way, the operation cost of the system is
also significantly reduced. By taking into consideration Figs. 4 and 5,
when the wind is lower than the base load, for example between 9 A.M.
and 1 P.M., the PEVs are not charged or they are even discharged since
they supply the base load in order to minimize the power injection from
the main grid into the commercial building in order to decrease its
operational cost. On the other hand, when the wind density is high and
its power generation is greater than the base load, for instance between
1 P.M. and 5 P.M., which is an off-peak load time, and the electricity
price is low, the PEVs are charged properly to maximize their charging
level. After 5 P.M. which is on-peak load time and the electricity price
goes up, the PEVs are properly discharged in order to supply the load
Fig. 4. Aggregated PEVs SOC, SOCmax and SOCmin.
and reduce the operational cost. Finally, during the last instances of the
connection to the charging station, when the base load is decreased, the
approach in this paper with the confidence level β = 0.99. The second PEVs are charged again so that the desired value of the charging level
is without the proposed control which is when the PEVs are charged could be reached before the plug-out time. As a result, the charging and
immediately after the connection to the charging station in order to discharging process of the PEVs depend on the base load, wind power,
reach the desired charging level. In the second state, there isn’t any the number of available PEVs and the electricity price. The proposed
control of the PEVs. The maximum capacity of the aggregated PEV control approach in this paper tried to create a power balance between
batteries is also illustrated in this figure. As it is clear from Figs. 4 and 5, production and consumption in a way to minimize the operational cost
SOC using the proposed approach varies during the simulation time of the commercial building while considering the uncertainty asso-
since it assists the demand response management. The optimization ciated to the electricity price. The power injected into the commercial

338
M. Tavakoli et al. Sustainable Cities and Society 41 (2018) 332–340

Fig. 8. The calculated aggregated charging reference for PEVs and the sum of
all individual PEV charging power.
Fig. 6. Power injected to the commercial building from main grid.

128.7805, respectively, and by using the proposed control with

β = 0.99, β = 0.95 and β = 0.90, respectively. Moreover, the total op-
erational cost is 206.6003 without the proposed control (when the PEVs
are charged immediately after connecting to the charging station until
reaching the full charge without any control). As a result, the opera-
tional cost of the commercial microgrid has remarkably decreased by
using the proposed control at circa 32.47%, 33.54% and 37.67% with
β = 0.99, β = 0.95 and β = 0.90, respectively. It becomes evident, as
the confidence level goes up, that the system will be more risk averse
which translates into higher costs. Finally, it should be mentioned that
it’s also possible to consider other objective functions for the com-
mercial building rather than minimizing its operational cost. Some
examples are: minimizing the power exchange between the main grid
and commercial microgrid, generating a pre-specified amount of power
by the commercial building or so on. In addition, in order to evaluate
the performance of the PEV control section for reference tracking, the
simulation result is depicted in Fig. 8. In this figure, the optimized
Fig. 7. Cost of power injected to the commercial building from main grid. power reference for the aggregated PEVs can be observed, calculated in
the first stage, and also the summation of all the individual PEV char-
ging power is illustrated. As it is clear from this figure, the aggregated
microgrid is from the main grid and its associated costs are depicted for
power reference is tracked by the summation of all the individual PEV
each time step in Figs. 6 and 7, respectively. Minus power in Fig. 6
charging power.
indicates the reverse direction of the power transfer i.e. from the
commercial microgrid to the main grid. When this occurs, the cost sign
is also minus which means the commercial building sells the power to 5. Conclusion
the main grid instead of buying it. In addition, as it is clear from Fig. 6,
that using the proposed control approach, the power transferred from Increasing the number of PEVs as a bulk load demand along with
main grid to the commercial building is varying all the time in order to the renewable energy resources which have an unpredictable nature
minimize the total operational cost of the microgrid based on the imposes some challenges for the future microgrids and the main grid.
variability of the wind power, base load, electricity price and PEVs One of the most important challenges is the power imbalance between
charging level by considering the uncertainty of the electricity price. the generation and the consumption. In order to improve the demand
However, in the uncontrolled approach of PEVs, the power is injected response in a commercial microgrid with the purpose of minimizing its
from the grid to the commercial building only when there is shortage in operational cost this paper presents a two stage predictive control ap-
the power supply for the base load and PEVs. The reason is that the proach for the energy management by considering the uncertainty in
uncontrolled PEVs are assumed as a bulk load for commercial building. electricity prices which is incorporated by the CVaR in the optimization
The total operational cost of the commercial microgrid throughout the problem. It is assumed that the employees arrive at the commercial
simulation is shown in Table 3 which is 139.5153, 1373.3132 and building in the morning stochastically and connect their PEVs to the
charging station and disconnect it at 9 P.M. when they depart. Using the
Table 3 proposed approach and considering the risk associated with the elec-
Total Cost of Injected Electricity to the Commercial Microgrid From the Main tricity price, the PEVs are mainly charged during the off-peak-demand
Grid. or when the electricity price is low. In this way, the total operational
Method Operational cost Improvement cost of the commercial building has noticeably decreased by 32.47%,
33.54% and 37.67% with β = 0.99, β = 0.95 and β = 0.90, respec-
Without the proposed control 206.6003 ($) – tively. In addition, by using the second stage of the proposed control
The proposed control with B = 0.99 139.5153 ($) 32.47%
approach, all the individual PEVs are charged in a way that the ag-
The proposed control with B = 0.95 137.3132 ($) 33.54%
The proposed control with B = 0.90 128.7805 ($) 37.67% gregated power reference is tracked appropriately. Any future work
could contain a design of fully stochastic control for the PEVs control

339
M. Tavakoli et al. Sustainable Cities and Society 41 (2018) 332–340

section, an analysis of different power sources and their associated full cooperation between the residential buildings and the commercial
costs, a more detailed model for the energy of the leaving PEVs, and a buildings.

Appendix A

CBattery = 20kWh, PPEV ,max = 4kW , Ekm = 0.2 kW km , c = 1/12.

References vehicles charging loads on the thermal ageing of power transformers. International
Journal of Electrical Power & Energy Systems, 65, 102–112.
Rassaei, F., Soh, W. S., & Chua, K. C. (2015). Demand response for residential electric
Alizadeh, M., Scaglione, A., Davies, J., & Kurani, K. S. (2014). A scalable stochastic model vehicles with random usage patterns in smart grids. IEEE Transactions on Sustainable
for the electricity demand of electric and plug-in hybrid vehicles. IEEE Transactions on Energy, 6(4), 1367–1376.
Smart Grid, 5(2), 848–860. Sechilariu, M., Wang, B., & Locment, F. (2013). Building-integrated microgrid: Advanced
Ansari, M., Al-Awami, A. T., Sortomme, E., & Abidoeric, M. A. (2015). Coordinated local energy management for forthcoming smart power grid communication. Energy
bidding of ancillary services for vehicle-to-grid using fuzzy optimization. IEEE and Buildings, 59, 236–243.
Transactions on Smart Grid, 6(1), 261–270. Shahidinejad, S., Bibeau, E., & Filizadeh, S. (2010). Statistical development of a duty
Ashabani, S. M., & Mohamed, Y. A. R. I. (2012). A flexible control strategy for grid- cycle for plug-in vehicles in a North-American urban setting using fleet information.
connected and islanded microgrids with enhanced stability using nonlinear microgrid IEEE Transactions on Vehicular Technology, 59(8), 3710–3719.
stabilizer. IEEE Transactions on Smart Grid, 3(3), 1291–1301. Soares, J., Morais, H., Sousa, T., Vale, Z., & Faria, P. (2013). Day-ahead resource sche-
Bai, X., & Qiao, W. (2015). Robust optimization for bidirectional dispatch coordination of duling including demand response for electric vehicles. IEEE Transactions on Smart
large-scale V2G. IEEE Transactions on Smart Grid, 6(4), 1944–1954. Grid, 4(1), 596–605.
Bashash, S., & Fathy, H. (2012). Transport-based load modeling and sliding mode control Sortomme, E., & El-Sharkawi, M. A. (2011). Optimal charging strategies for unidirectional
of plug-in electric vehicles for robust renewable power tracking. IEEE Transactions on vehicle-to-grid. IEEE Transactions on Smart Grid, 3(1), 131–138.
Smart Grid, 3(1), 526–534. Su, W., & Chow, M. Y. (2012a). Computational intelligence-based energy management for
Clement-Nyns, K., Haesen, E., & Driesen, J. (2011). The impact of vehicle-to-grid on the a large-scale PHEV/PEV enabled municipal parking deck. Applied Energy, 96,
distribution grid. Electric Power Systems Research, 81(1), 185–192. 171–182.
Huang, Q., Jia, Q., Qiu, Z., Guan, X., & Geconinck, G. (2015). Matching EV charging load Su, W., & Chow, M. Y. (2012b). Performance evaluation of an EDA-based large-scale plug-
with uncertain wind: A simulation-based policy improvement approach. IEEE in hybrid electric vehicle charging algorithm. IEEE Transactions on Smart Grid, 3(1),
Transactions on Smart Grid, 6(3), 1425–1433. 308–315.
Kennel, F., Gorges, D., & Liu, S. (2013). Energy management for smart grids with electric Tan, J., & Wang, L. (2014). Integration of plug-in hybrid electric vehicles into residential
vehicles based on hierarchical MPC. IEEE Transactions on Industrial Informatics, 9(3), distribution grid based on two-layer intelligent optimization. IEEE Transactions on
1528–1537. Smart Grid, 5(4), 1774–1784.
Lee, T., Adornato, B., & Filipi, Z. (2012). Stochastic modeling for studies of real-world Tan, Z., Yang, P., & Nehorai, A. (2014). An optimal and distributed demand response
PHEV usage: Driving schedule and daily temporal distributions. IEEE Transactions on strategy with electric vehicles in the smart grid. IEEE Transactions on Smart Grid, 5(2),
Vehicular Technology, 61(4), 1493–1502. 861–869.
Li, C., Ahn, C., Peng, H., & Sun, J. (2013). Synergistic control of plug-in vehicle charging Wang, J. H., Liu, C., Ton, D., Zhou, Y., Kim, J., & Vyas, A. (2011). Impact of plug-in hybrid
and wind power scheduling. IEEE Transactions on Power Systems, 28(2), 1113–1121. electric vehicles on power systems with demand response and wind power. Energy
Meegahapola, L. G., Robinson, D., Agalgaonkar, A. P., Perera, S., & Ciufo, P. (2014). Policy, 39(7), 5482–5492.
Microgrids of commercial buildings: Strategies to manage Mode transfer from grid Wang, Z., Wang, L., Dounis, A., & Yang, R. (2012). Integration of plug-in hybrid electric
connected to islanded Mode. IEEE Transactions on Sustainable Energy, 5(4), vehicles into energy and comfort management for smart building. Energy and
1337–1347. Buildings, 47, 260–266.
Olken, M. (2013). The future of microgrids: Coming to a neighborhood near you. IEEE Wu, T., Yang, Q., Bao, Z., & Yan, W. (2013). Coordinated energy dispatching in microgrid
Power & Energy Magazine, 11(4), 4–6. with wind power generation and plug-in electri vehicles. IEEE Transactions on Smart
Pouresmaeil, E., Miguel-Espinar, C., Massot-Campos, M., Montesinos-Miracle, D., & Grid, 4(3), 1453–1463.
Gomis-Bellmunt, O. (2013). A control technique for integration of DG units to the Xu, Z., Hu, Z., Song, Y., Zhao, W., & Zhang, Y. (2014). Coordination of PEVs charging
electrical networks. IEEE Transactions on Industrial Electronics, 60(7), 2881–2893. across multiple aggregators. Applied Energy, 136, 582–589.
Qian, K., Zhou, C., Allan, M., & Yuan, Y. (2011). Modeling of load demand due to EV Zeng, Z., Zhao, R., & Yang, H. (2013). Micro-sources design of an intelligent building
battery charging in distribution systems. IEEE Transactions on Power Systems, 26(2), integrated with micro-grid. Energy and Buildings, 57, 261–267.
802–810.
Qian, K., Zhou, C., & Yuan, Y. (2015). Impacts of high penetration level of fully electric

340