Download as pdf or txt
Download as pdf or txt
You are on page 1of 19

International Journal of Computers and Applications

ISSN: (Print) (Online) Journal homepage: https://www.tandfonline.com/loi/tjca20

Optimal scheduling of vehicle-to-Grid power


exchange using particle swarm optimization
technique

Arkan Mulla & H. T. Jadhav

To cite this article: Arkan Mulla & H. T. Jadhav (2021): Optimal scheduling of vehicle-to-Grid
power exchange using particle swarm optimization technique, International Journal of Computers
and Applications, DOI: 10.1080/1206212X.2021.1903707

To link to this article: https://doi.org/10.1080/1206212X.2021.1903707

Published online: 01 Apr 2021.

Submit your article to this journal

Article views: 37

View related articles

View Crossmark data

Full Terms & Conditions of access and use can be found at


https://www.tandfonline.com/action/journalInformation?journalCode=tjca20
INTERNATIONAL JOURNAL OF COMPUTERS AND APPLICATIONS
https://doi.org/10.1080/1206212X.2021.1903707

RESEARCH ARTICLE

Optimal scheduling of vehicle-to-Grid power exchange using particle swarm


optimization technique
Arkan Mulla and H. T. Jadhav
Department of Electrical Engineering, Rajarambapu Institute of Technology, Islampur, Maharashtra, India

ABSTRACT ARTICLE HISTORY


Electric vehicles (EV) are the inevitable future of the transportation industry considering the current energy Received 17 October 2020
scenario and all the incentives and benefits they provide. Nonetheless, EVs exhibit a novel characteristic of Accepted 11 March 2021
a distributed energy storage device (ESD) owing to their large onboard batteries. This feature of EV can be KEYWORDS
utilized to provide many ancillary services by the virtue of vehicle-to-grid (V2G) operation. One such service Vehicle-to-Grid (V2G); valley
is the minimization of load variation on the grid. Two key challenges, concerning the practical execution filling; peak shaving; optimal
of the V2G operation, are the availability of EVs for V2G operation and their ever-changing State of Charge scheduling; particle swarm
(SOC). Due to the mobility of EVs as a transportation medium, their availability is highly uncertain. This paper optimization; flattening the
addresses these issues by introducing a scheduling scheme for V2G power exchange by considering the load curve
stochastic nature of EV grid connectivity. This is done by first developing an optimization algorithm using
the particle swarm optimization technique, with input data that best represents the stochastic nature of EV
availability. Then, the performance of the algorithm is evaluated by conducting several case studies. The
results obtained for various case studies by performing simulations are represented and elaborated. Finally,
the statistical analysis of the results signifies that the proposed V2G scheduling scheme can substantially
flatten the load profile. Apart from the algorithm itself, another novel nature of the paper lies within the
wide range of analyses carried out to study the effect of such scheduling schemes. These studies include
the effect on the SOC level of different EVs, the effect on price fluctuation, time complexity analysis of the
algorithm, etc.

List of Symbols and Abbreviations Pi comparative parameter


Pd,after the difference between peak and valley point in
EV Electric Vehicle
load curve after optimization
ESD Energy Storage Device
Pd,before the difference between peak and valley point of the
V2G Vehicle- to-Grid
load curve before optimization
SOC State of Charge
ICE Internal Combustion Engine
PHEV Plug-in Hybrid Electric Vehicle
PEV Plug-in Electric Vehicle 1. Introduction
GA Genetic Algorithm Human history has seen many revolutionary events that have
PSO Particle Swarm Optimization shaped our future for better or worse. At present, global warm-
Pt Target Load ing and climate change happen to be such events that we, among
Pl Actual Load all other species living on this planet, are going through. These
P Load Difference issues, along with the depleting fuel reserves are posing a grave
PEV total power exchanged by all EVs threat to energy security and the sustainable development of
K number of EVs connected to the grid that at the society. According to a recent report on global temperatures, the
same time fulfill SOC limits combined land, and ocean surface temperatures has increased
Pevi power exchanged by the ith EV at an average rate of 0.07˚C (0.13˚F) per decade since 1880,
SOCi State of Charge of the ith EV however, this rate has only gone up to 0.13˚C (0.32˚F) since
SOCmin minimum SOC limit 1981 [1]. Fossil fuels on the other hand are on the verge of
SOCmax maximum SOC limit depletion owing to the rapid exploitation of these resources in
Pev,max maximum power exchange rate of the EV the industrial and transportation sectors [2]. The transporta-
Pgrid power injected into the grid by generation stations tion sector accounts for the largest share of fuel, and within
Eb energy capacity of the battery it, passenger transportation (light-duty vehicles such as cars,
SOCnewi new updated SOC of the ith EV buses, 3-wheelers, 4-wheelers, etc.) accounts for about 61% of
Plnew new load after optimization
SOCi Initial SOC

CONTACT H. T. Jadhav htj.rit@gmail.com Rajarambapu Institute of Technology, Rajaramnagar, Islampur, Maharashtra, India
© 2021 Informa UK Limited, trading as Taylor & Francis Group
2 A. MULLA AND H. T. JADHAV

transportation energy [3]. Apart from energy consumption, the the proposed algorithm is demonstrated and justified using time
internal combustion engine (ICE) vehicles that are powered by complexity analysis, which, to the best of the authors’ knowl-
the combustion of fossil fuels, pose an even greater threat to edge, hasn’t been performed in any other method proposed in
human health due to the increasing emission rate of pollutants the literature. The authors have also included a study of the effect
and hazardous gases [4]. The transportation sector accounts for of the proposed scheduling on hourly price fluctuations as part
25% of global CO2 emissions and about 14% of greenhouse of sensitivity analysis.
gas emissions [5]. Electrification of the transportation indus- While V2G operation can be beneficial on many fronts,
try, including the re-introduction of electric vehicles has been its effectiveness and practical implementation comprise many
deemed as a prominent solution to these issues [6]. Due to a challenges and constraints as discussed by authors of [25–27].
huge boost in the popularity of EVs and their integration with The major challenge associated with effective V2G operation
advanced renewable energy technology, the future of energy and is the stochastic nature of the EVs when it comes to their
human health can be secured. availability and connectivity with the grid. Since EVs have a
There are different types of electric vehicles introduced, such prime function as a transportation tool, their availability for
as hybrid EV (HEV), plug-in EV (PEV), plug-in hybrid EV V2G operation becomes highly uncertain. We cannot say for
(PHEV), etc. Plug-in EV (PEV) has the advantage of large certain when a vehicle will be connected to the grid or get
capacity onboard batteries making them more effective in V2G disconnected. Therefore it is important to incorporate this ran-
operations. PEVs are the main focus of the proposed work, dom nature of the EV connectivity while developing the V2G
hence, hereon they are simply referred to as EVs. Day by day, algorithm [28]. Another key challenge is to maintain the bat-
the penetration of EVs on the power grid is only going to tery SOC within permissible limits with the intent that the
increase which also increases the number of issues and sever- batteries should have sufficient range at any time. EV batter-
ity of these problems [7]. The impact of EVs on the distribution ies go through frequent charge and discharge cycles during
grid may include wearing of transformers, power quality issues, V2G operation which considerably affects their lifespan. There-
grid congestion, etc [8]. But these issues and their impact can fore, the SOC levels must be regulated to prevent any dras-
be nullified by the proper management and strategic integration tic effect on battery life. The satisfaction of the EV’s owner
of EVs into the grid, and one such strategy is V2G opera- and their demands are important issues that should be con-
tion. As discussed, EVs have relatively large-capacity batteries sidered, and they are primarily concerned with previously dis-
that can be recharged by extracting power when plugged into cussed arguments. Hence, all these issues must be addressed
the grid. With bi-directional power flow, they can also inject while developing an algorithm for V2G operation concerning
power back into the grid. V2G operation is feasible with a any objective.
large number of EVs participating in the power exchange and The work in this paper deals with the development of an
an aggregator managing the operation. This V2G operation of optimal scheduling algorithm for V2G power exchange. The
EVs, along with the novel characteristic of EV as a distributed objective of this algorithm is to minimize the fluctuations in
energy storage device, can be utilized for many applications and the load profile around a predefined target load and effectively
ancillary services. These services include frequency regulation flatten the load curve, while at the same time comply with all
[9,10], voltage regulation [11,12], minimization of operating the constraints. First, we need to clearly define the objective,
cost [13–16], valley filling in load curve[17,18], peak shaving which in this case is the minimization of the load variations.
in load curve [19,20], or both called as flattening the load curve Then, a mathematical form of this objective function is cre-
[21–24] etc. ated for the optimization. Then, all the constraints are defined
Speaking of the V2G scheme for load leveling objective (i.e. and modelled. These constraints are of two types, equality con-
peak shaving and valley filling), many authors have presented straints, and non-equality constraints. These constraints are EV
different algorithms primarily using a genetic algorithm. The availability, EV charging/discharging rate, battery SOC limit,
authors of [21] have introduced a GA based V2G algorithm with and power balance. There are various optimization techniques
a vehicle sample size of 1800. A primary drawback of this study available for V2G optimization. Two of the most popular opti-
is that authors have taken the time step of one hour which is mization techniques are particle swarm optimization (PSO) and
clearly not ideal. Meanwhile, authors of [24] have introduced the genetic algorithm (GA). Both techniques have their fair
a time step of fifteen minutes which is relatively better con- share of pros and cons, but for our objective, we selected the
sidering the computational limits but suffers from a smaller PSO technique [29]. All these points are further elaborated in
sample size. Similarly, [22] and [23] have also introduced the section 2.
V2G algorithm with the same drawbacks. One more issue with Apart from the algorithm itself, a profound amount of
all the methods proposed in the literature so far is that all the thought is given in preparing data for simulation purposes. The
vehicles in the sample size used for simulations have the same random and uncertain nature of EV availability was considered
specifications (i.e. battery capacity, initial SOC, power exchange while developing EV availability data. A relatively large sam-
rate, etc.) which is not ideal, as the EV market in the actual ple size of about 2000 EVs is considered, with different types
world is much more diverse. This paper aims at addressing these of EVs having different specifications, such as battery capacity
issues and proposing a possible solution. Apart from this, there and charging/discharging rate. In the real-world, the status of
are other novelties of the study carried out in this paper. The EV connectivity to the grid can change at any given moment in
scheduling method used in the algorithm handles each EV indi- time. However, to best represent this case, a sample time of 15
vidually, i.e. connection disconnection, and desired SOC level of min is chosen. This means, once an EV is connected or discon-
each EV is accounted separately. The performance superiority of nected from the grid, its connectivity status will remain as such
INTERNATIONAL JOURNAL OF COMPUTERS AND APPLICATIONS 3

Figure 1. V2G Power Exchange Scheme.

for 15 min. And once 15 min have passed, their connectivity sta- 2. Problem formulation
tus will either change or remain the same. Even though choosing
The optimal scheduling of V2G power exchange involves find-
a shorter sample time is ideal, it increases computational stress
ing an optimal solution to the defined objective function. How-
and memory requirements for the hardware. Speaking of EVs,
ever, the achievement of the best solution is limited by multiple
authors of previous work done on a similar subject [21–24]
conflicting constraints. Therefore, to optimally schedule and
have assumed a small sample size of EVs, all of them having
execute the V2G operation, the particle swarm optimization
the same battery capacity and charging/discharging rate. This
technique is used. This section discusses the mathematical mod-
assumption is not accurate in the real-world where EVs come
elling of the objective function and the various constraints of the
in different categories and specifications. Contrary to this, the
problem.
authors of this paper have considered 5 different types of EVs, all
having a different share in sample size and specifications based
on the EVs available in the market. 2.1. Objective function
The paper is organized section wise with a total of six
sections. Problem formulation containing objective function The objective of the V2G optimization problem in this paper
and all the constraints are elaborated in section 2. Section 3 is to minimize the load variation occurring on the grid by
deals with the development and description of the optimization effectively utilizing the charging and discharging power of the
algorithm. Preparation of input data such as vehicle availability, grid-connected EVs. This can be achieved by performing func-
load curve, EV specification, etc. is discussed in section 4. The tions called valley filling (while charging) and peak load shaving
algorithm is then tested for different scenarios, and with such (while discharging) depicted in Figure 2. These functions are
thorough input data, it further helps to validate the practicality performed by defining a target load (Pt) around which, the
of the algorithm in the real world. The different case studies and algorithm will try to flatten the load curve. Whenever the actual
their results are discussed in section 5. A performance index is load (Pl) exceeds the target load, the algorithm will instruct all
introduced to evaluate the effectiveness of the algorithm. Finally, available EVs to inject power into the grid. Similarly, when the
a conclusion is drawn in section 6. A graphical depiction of the actual load falls below the target load, the EVs are instructed to
proposed V2G scheme is shown in Figure 1. extract power from the grid by charging the onboard batteries.
Before diving into the algorithm, there are some assumptions
made that are listed below.

• The car parks have a sufficient amount of space and are


equipped with V2G chargers.
• Even though current EVs don’t have the feature of bi-
directional power flow, it is assumed that the external V2G
charger control circuit can bypass this limitation.
• The aggregator has access to vehicle information such as
current SOC status and battery capacity.
• The aggregator is given full authority over the amount and
duration of power to be exchanged as long as the vehicle is
connected to the grid. Figure 2. Depiction of Valley Filling and Peak Load Shaving.
4 A. MULLA AND H. T. JADHAV

In the situation when the actual load is equal to the target load, must not exceed this limit.
no power is shared between EV and the grid. This objective
function is mathematically expressed in equation number (1). Pev, charging ≤ Pev, max (6)
Pev, discharging ≥ −Pev, max (7)
minP = Pt(t) − Pl(t) − PEV(t) (1)

N In equation (6) and (7), Pev,max is the maximum allowable
PEV(t) = Pevi (t) (2) power exchange rate. Since the EV fleet developed in this paper
i=1 includes five different models of EV, they will have different
power exchange limits. This is further discussed in section 4.3.
In equation (1), P is the load variation in the grid with
respect to the target load, and PEV is the power exchanged by
2.2.3. Power balance
all (N = 2000) EVs, which is depicted in equation (2). Mean-
Power balance is an important constraint from the power grid’s
while, the target load (Pt), actual load (Pl), and total EV power
point of view since its violation may result in the fluctuation of
exchanged (PEV) are all time (t) dependant. Deconstructing the
system parameters. Therefore, the algorithm must comply with
equation (1) on the level of an individual vehicle, we get
this constraint by equating injected power and load on the grid.
Pt(t) Pl(t)
minp = − − Pevi (t) (3) Pgrid(t) = Pl(t) + PEV(t) (8)
K K
In equation (3), parameter K is introduced, which is nothing but Here, Pgrid is the power injected by the generation plants.
the total number of EVs connected to the grid that at the same
time satisfies the SOC limit. The parameter K is introduced to 2.2.4. EV availability
equally distribute the total power to be exchanged amongst all The availability of EV for V2G operation is one of the constraints
the EVs that are eligible for V2G operation. It should be noted and sort of a challenge when it comes to effective implementa-
that equal distribution of power to be exchanged isn’t always tion of the V2G power exchange. Since it will act as input data
ideal since each EV will have a different V2G capability. The to the algorithm for simulation purposes, more on this topic is
term Pevi , which is power shared by the ith vehicle, is positive discussed in section 4.2.
when power is extracted from the grid and negative when power
is injected into the grid. 3. Optimization algorithm
The proposed algorithm has many time iteration loops and has
2.2. Optimization constraints a large number of EVs to take into account, hence an effective
The best possible solution to the optimization problem is limited optimization technique is required for good performance. In
by several constraints on both the EV and the grid side. And the this paper, the particle swarm optimization (PSO) technique has
algorithm has to comply with all these constraints in order to be been selected for the optimization of the V2G power exchange.
compatible. PSO is a memory-based computational algorithm that searches
for a globally optimal solution within a population for a given
2.2.1. EV SOC limit problem. The population is a randomly generated solution,
There are two primary reasons behind keeping the battery SOC found by updating the generations. In PSO, random potential
of the vehicle within the limit. The first is to reduce the possi- solutions named particles, are moving in free sample space with
ble degradation of the battery pack. As V2G operation requires a specific velocity. Each particle can interact with other particles,
frequent charging and discharging, it is not feasible to fully and all of them tend to move towards a global-optima. This is
discharge and charge batteries since it imposes stress on the elec- achieved by constantly updating the velocity and the direction of
trochemical elements of the battery which causes a reduction a particle which depends upon the personal best solution of that
in their lifespan. Another reason for not fully discharging the particle and the global best solution. This process is repeated
battery is to keep a sufficient amount of reserve available at any until the stopping criteria is reached, which is generally a spe-
instance for the vehicle’s next commute. By taking both reasons cific number of iterations. For the algorithm, a population size
into account, the EV battery SOC for integration in V2G oper- of 100 and stopping criteria of 100 iterations are chosen.
ation is limited between 50% to 90%. The upper limit of SOC is The proposed V2G algorithm can be divided into four stages.
set at 90% considering the overcharging issues. Now, the grid- In the first stage, input data is gathered for optimization and the
connected vehicle will be allowed to discharge only if its SOC is target load is defined. In the second stage, the time loop is initi-
greater than SOCmin, similarly, it will be allowed to charge only ated and the target load and actual load are compared. Based on
if its SOC is less than SOCmax. the comparison, vehicles are instructed to either charge or dis-
charge. No action is taken if the target load and actual load are
Allow discharging only if SOC > SOCmin (4) equal. In the third stage, PSO is initiated to optimally distribute
Allow charging only if SOC < SOCmax (5) total power and calculate the power to be shared by individual
vehicles. And in the final stage, battery SOCs are updated for
2.2.2. EV Power Exchange rate the next time iteration. A detailed step-by-step operation of the
The maximum rate at which EVs can either charge or discharge algorithm can be understood by observing the flow chart given
is generally limited by the battery and charging station hard- in Figure 3. A pseudo-code of the proposed algorithm is given
ware. Therefore, the power exchange rate between EV and grid below.
INTERNATIONAL JOURNAL OF COMPUTERS AND APPLICATIONS 5

Figure 3. Flowchart of the Proposed V2G Optimization Algorithm.

A pseudo-code of the proposed algorithm is given below. 7. Initialize the vehicle iteration loop.
1. Start the algorithm 8. Check for the SOC limit given below. If not, then return
2. Initiate by defining target load (Pt) and PSO parameters the power to be exchanged by that vehicle equal to zero (i.e.
such as population size, stopping criteria, and constants. Pev = 0) and jump to step 21. If SOC limits are satisfied, then
3. Load input data, e.g. vehicle availability, battery capacity proceed to the next step.
(Eb), and SOC limits. Allow discharging only if SOC > SOCmin
4. Initiate the time iteration loop. Allow charging only if SOC < SOCmax
5. Compare target load (Pt) and actual load (Pl) for the given 9. Check whether the vehicle is connected to the grid. If not,
time and decide necessary action to be taken. i.e. return the power to be exchanged by that vehicle equal to zero
Charge if (Pl) < (Pt) (i.e. Pev = 0) and jump to step 21. If connected proceed to the
Discharge if (Pl) > (Pt) next step.
No action if (Pl) = (Pt) 10. Initialize PSO by generating the initial population of the
6. Update the value of K, which is the number of EVs con- solutions (i.e. power to be shared by the vehicle).
nected to the grid which at the same time satisfies the SOC 11. Evaluate fitness function, which is
limit. minp = Pt(t) Pl(t)
K − K − Pev (t)
i
6 A. MULLA AND H. T. JADHAV

Figure 3. Continued

12. Find the personal best (Pbest) and global best (Gbest) 19. Now we have found power to be shared by the
solution. vehicle which is equal to the global best solution, i.e.
13. Initiate the iteration count. Pev = Gbest.
14. Calculate velocity and a subsequent new set of solutions. 20. Update the SOC of the vehicle by using the equation given
15. Check if the new solutions are within the bounds that are below
given below. If not, bring them within bounds and proceed to
the next step. (SOCi ∗ Ebi ) + (0.25 ∗ Pevi )
SOCnewi =
Ebi
Pev, charging ≤ Pev, max Pev, charging ≥ −Pev, max
Here, Pevi is multiplied by 0.25 to find the energy shared in
16. Evaluate objective function using the new set of solutions. 15 min of sample time, and Ebi the battery energy capacity of
17. Now check for Pbest and Gbest solution. If the new Pbest the ith vehicle.
and Gbest solution is found, update them and proceed. 21. Check if all the vehicles have been accounted for. If no,
18. Advance the iteration count and return to step (14) until advance the vehicle count and return to step (8). If yes, proceed
the stopping criteria is reached which is 100 iterations. to the next step.
INTERNATIONAL JOURNAL OF COMPUTERS AND APPLICATIONS 7

22. Now calculate the total power shared by all the vehi- every 15 min, which means we have a total of 96 load curve
cles (PEV) by taking the summation of power shared by all the data points. In the real-world, load curve forecasting can be
vehicles. used for a day-ahead or real-time scheduling of the V2G power
23. Calculate new load (Plnew) using the following formula exchange [30,31]. Meanwhile, our algorithm and simulation
Plnew = Pl + PEV studies depict real-time scheduling.
24. Check if the time iteration has ended. If no, advance time
iteration and return to step (5). If yes, then print the results of 4.2. EV availability
all the time iterations.
25. END The V2G power exchange is possible only if there are EVs con-
nected to the grid. However, the availability of each EV cannot
be predicted due to its mobility and stochastic nature. The driv-
4. Preparation of input data ing pattern of the individual vehicle is influenced by many fac-
In order to validate the effectiveness of the proposed schedul- tors that are not feasible to collect and study [32]. But, a rough,
ing scheme, several simulation studies are required to be carried collective driving pattern of EVs can be prepared for the sim-
out using some input data. This input data should be pre- ulation study of the proposed V2G algorithm which is shown
pared in a reasonable and indiscriminative way to best represent in Figure 5. [21]. Consider a generic town, having about 2000
real-life scenarios. The SOC limits are set at 0.5–0.9 consider- EVs, and all of these EVs have access to V2G bidirectional charg-
ing all the reasons and arguments presented in section 2.2.1. ers in car parks. The car parks are divided into two categories,
Meanwhile, the initial SOC of each EV is randomly distributed residential car parks, and commercial car parks. In residential
around a mean value. Other inputs used for simulation studies car parks, the majority of EVs are available during night time
are discussed in the following sections. up until 6.00 am After that, the number of EVs in residential
car parks starts to decrease as people leave for offices, schools,
and their respective workplaces. Exactly opposite to this, the
4.1. Load curve
number of EVs in commercial car parks is highest during office
A 24-hour load curve of a regional grid shown in Figure 4. is hours, i.e. from 9.00 am up to 6.00 pm And on the other side, at
prepared for simulation purposes. The load curve is sampled night time, commercial car parks are sparsely populated. Even

Figure 4. Load Curve.

Figure 5. EV Grid Connection Probability.


8 A. MULLA AND H. T. JADHAV

Figure 6. Optimized Load Curve with 8 MW Target Load and Average SOCi of (a) 40%, (b) 50%, (c) 60%, (d) 70%, and (e) 80%.

though the EV availability, shown in Figure 5., is sampled at Table 1. EV Models Used for Simulation.
every hour, the connectivity status of each EV changes randomly Battery Power
after 15-minute. Capacity Exchange Range Percentage
EV Model (kWh) rate (kW) (km) Share
Model 1 (Tesla Model 3) [34] 50 11 354 30%
4.3. EV specification Model 2 (Chevrolet Bolt) [35] 60 7.2 383 24%
Model 3 (Nissan Leaf) [36] 40 6.6 243 20%
As discussed earlier, there are 5 different EV models used for the Model 4 (Volkswagen e-Golf) [37] 35.8 7.2 233 12%
simulation study. These EVs and their specifications such as bat- Model 5 (Other) 25 3.3 150 14%
tery capacity, power exchange rate, range of vehicle along with
their share in the vehicle fleet have been listed in Table 1. The
specifications and the share in the sample size of all these vehi- 5.1. Simulation of algorithm
cles have been referred from actual EVs available in the market
[33]. A total of 20 scenarios were created by different values of tar-
get load (Pt) and average initial SOC (SOCi) of the EVs. Target
loads were set at 8, 8.4, 8.8, and 9.2 MW, while average initial
5. Results and discussion
SOCs were set at 40%, 50%, 60%, 70%, and 80% for each target
In this section, the compatibility and effectiveness of the pro- load. The comparison between the actual load curve and opti-
posed V2G algorithm are investigated. The investigation is done mized load curves achieved using the proposed V2G algorithm
by conducting different case studies by simulating different sce- for different scenarios is shown in Figures 6–9.
narios. Analysis of the different results is done to validate the Figure 6. (a) shows the optimized load curve for a target
feasibility of the algorithm. load of 8 MW and an average SOCi of 40%. If we observe the
INTERNATIONAL JOURNAL OF COMPUTERS AND APPLICATIONS 9

Figure 7. Optimized Load Curve with 8.4 MW Target Load and Average SOCi of (a) 40%, (b) 50%, (c) 60%, (d) 70%, and (e) 80%

curve, we notice that the target load is not reached through- SOCi and different target load, we observe that even the selec-
out the day. While valley filling is done effectively by charging tion of target load has a significant impact on the best achievable
the vehicle, peak shaving operation is failed. This is because result.
the average SOCi is not high enough and it is lower than the We will now introduce a performance index to statistically
lower SOC limit (i.e. SOCmin = 50%). Hence, while vehicle dis- analyze and compare the results of all scenarios [21]. The
charging is allowed, peak shaving is prohibited. However, if we performance index is defined as the percentage of success-
increase the average SOCi, keeping the target load the same, we ful operations achieved using the proposed V2G algorithm.
can observe that the performance of the algorithm is improved, The performance index varies from 0 to 100%, with a higher
shown in Figure 6. (b), (c), (d), and (e). Hence, from these five percentage indicating a higher success rate. The performance
scenarios, we can say that, as we increase the average SOCi, index of all twenty scenarios for individual operation (valley
the performance of the algorithm gets better. But this statement filling and peak shaving) and overall performance is shown
contradicts the results of the other fifteen scenarios shown in in Table 2.
Figures 7–9. Consider the two cases with a target load of 8.4 The performance index shown in the above table further
MW and average SOCi of 70% and 80%, shown in Figure 7. (d) helps us understand the performance of the algorithm with
and (e) respectively. The former shows almost 100% achieve- different target loads and average SOCi. If we observe the perfor-
ment of the target load curve, but when we increase the average mance index of valley filing operation, we see that performance
SOCi, the achievement of the target load drops by a significant degrades as the average SOCi increases for each target load.
amount. This is because the initial valley filling (charging) oper- Meanwhile, the degradation in performance only gets worse
ation pushes the already high value of SOC beyond the upper with increasing target load. However, exactly the opposite hap-
limit of SOC (i.e. SOCmax = 90%). Therefore, EVs that exceed pens with a performance index of peak shaving operation. In
this limit are prohibited from further power exchange. The same the case of peak shaving, the performance index gets better
argument applies to the similar anomalies seen in Figures 8 and as we increase the target load and average SOCi. The effect
9. Now, if we compare the results obtained with the same average of the performance of these two operations is reflected in the
10 A. MULLA AND H. T. JADHAV

Figure 8. Optimized Load Curve with 8.8 MW Target Load and Average SOCi of (a) 40%, (b) 50%, (c) 60%, (d) 70%, and (e) 80%.

Table 2. Performance Index. representation of the performance index shown in Figure 10.
Average Initial SOC (SOCi) These arguments further support the reasons and statements
presented earlier.
Operation Target Load 40% 50% 60% 70% 80%
From all the discussions presented till now, we can say that
Valley Filing 8 100 100 100 100 100
8.4 100 100 100 100 75.51
the performance of the proposed V2G algorithm depends on
8.8 100 100 100 87.50 66.07 both, target load and average SOCi. Hence, care should be taken
9.2 100 100 95 75.40 47.54 while selecting target load based on EVs available and their aver-
Peak Shaving 8 4.91 18.03 68.85 81.96 83.60 age SOCi to get the best performance out of the V2G operation.
8.4 17.02 63.82 78.72 87.23 87.23 It should be noted that the algorithm itself gives the best perfor-
8.8 62.5 80 90 90 90 mance in each scenario; it is the external factors that limit the
9.2 84.37 100 100 100 100
achievable results. Meanwhile, for all the scenarios presented,
Combined 8 39.5 47.91 80.20 88.54 89.58 the best optimization is achieved for the case with a target load
8.4 58.33 82.29 89.58 93.75 81.25
8.8 84.37 91.66 95.83 88.54 76.04 of 9.2 MW and an average SOCi of 50%. This is demonstrated in
9.2 94.79 100 96.87 84.37 63.54 the load curve in Figure 9. (b) and with the performance index
of 100%.
A detailed analysis of the best-case scenario is represented in
Figure 11. With a target load of 9.2 MW and an average SOCi
performance index of the combined operation. We see that of 50%, the optimized load curve has achieved the target load
better performances are achieved somewhere in the middle throughout the day. Figure 12 (a) and (b) help present a clearer
ground. This can be effectively illustrated using a graphical picture of the power exchange happing between EVs and the
INTERNATIONAL JOURNAL OF COMPUTERS AND APPLICATIONS 11

Figure 9. Optimized Load Curve with 9.2 MW Target Load and Average SOCi of (a) 40%, (b) 50%, (c) 60%, (d) 70%, and (e) 80%.

grid. Figure 12. (a) shows the power available from EVs (for some EVs (i.e. Model 1 and Model 2) has increased far drasti-
charging and discharging) and the actual power that is shared cally compared to the other three vehicles (i.e. Model 3, 4, and
with the grid throughout the day. Meanwhile, Figure 12. (b) 5). This is because of the battery energy capacity of these vehi-
depicts the relation between power injected/extracted by EVs cles. The SOC of the EVs with smaller battery energy capacity
and the actual load curve. These two figures help us under- will increase at higher rates than the SOC of those EVs that have
stand that the proposed algorithm can successfully track the batteries with large energy capacity with the same rate of power
load curve and exchange the required power to meet the target. exchange.

5.2. Performance of different EV models 5.3. Effect on price fluctuation


Figure 13. shows the SOC status of randomly selected EVs The effect of a load-leveling scheduling scheme such as the one
from each of the five models used, for the best-optimized sce- proposed in the paper is important to consider in smart grids
nario. The figure also includes the average SOC of all 2000 EVs where the price of electricity is fluctuating depending on the
throughout the day. It is clear that all EVs have higher SOC time of the day and load the grid. Since the proposed algorithm
than their initial SOC at the end of the optimization. The aver- changes the load profile of a locality, it will also change the price
age SOC curve also supports this claim. Meanwhile, the SOC of fluctuations of the grid. Figure 14. shows the hourly price of
12 A. MULLA AND H. T. JADHAV

Figure 10. Overall Performance Index.

Figure 11. Optimized Load Curve of Best Optimized Scenario.

electricity for a given load profile (Figure 4.) of a smart grid. To And the respective curve fitted for the polynomial is shown in
evaluate the modified hourly prices after the execution of the Figure 16. This gives us a polynomial equation for the price as
proposed scheduling scheme, we must first establish a relation- a function of load that we now can use to evaluate new hourly
ship between load and electricity price [38]. This is done by the prices for the Levelized load curve.
polynomial regression method, utilizing the data from the load New hourly prices for Levelized load for different target loads
curve(Figure 4.) and price fluctuation curve (Figure 14.). are illustrated in Figure 17. The average price of electricity
before load leveling is about 0.0379 USD/kWh. By observation
ρi = a1 .Pli + a2 Pli2 + a3 .Pli3 + . . . . . . . . . + am .Plim + am+1 (9) we can see that the average price of electricity is higher, i.e.
Equation (9) represents price as a function of load. Here, ρi is the 0.0384, 0.0387, 0.0392, and 0.0398 USD/kWh for target load of
price at the ith hour of the day, Pli is the load at the ith hour of the 8, 8.4, 8.8, 9.2 MW respectively. So we can say that although
day, a1,2, ... .m are the regression coefficients, and m is the degree the price fluctuation was minimized, the economic benefits
of the polynomial. The hourly load from the original load curve were constrained by the selection of the target load. So, as an
(Pli ) is used to estimate the regression coefficients, and then extension of the present work, a multi-objective optimization
new prices are evaluated by substituting Levelized load (Plli ) in algorithm can be developed to solve this issue.
equation (9). The degree of the polynomial (m) is selected based
on the adjusted R2 index. This index depicts the quality of the
5.4. Time complexity analysis
model, when the model fits the data the index approaches 1.
So, the degree of polynomial should be selected such that the Time complexity is defined as the execution time of an
adjusted R2 index is maximum or when the change in its value algorithm as a function of input size. As the size of input
becomes small (Figure 15). data directly affects the number of steps or instructions in an
The degree of the polynomial is selected as 3 by observing the algorithm, this further affects the execution time. This gives
adjusted R2 index curve for different degrees of the polynomial. information about variations in execution time as the size of
INTERNATIONAL JOURNAL OF COMPUTERS AND APPLICATIONS 13

Figure 12. (a) Available EV Power (b) EV Power Exchanged with Grid.

Figure 13. SOC Status.

input data increases. It helps us to compare different algorithms involved. So, the execution time for different number of EVs
developed to execute the same objective. Since these algorithms (from 100 to 2000 EVs) is evaluated which gives us a relationship
are not performed with the same input data or on the same work between execution time and input size which can be demon-
station, they cannot be compared by execution time only. In strated graphically as shown in Figure 19. We can observe that
computer science, the time complexity is generally represented the graph is almost linear, which means time would vary lin-
by Big O notation. It is written in the form O(n), where O order early with variation in the input size. So the Big O notation for
of growth and n stands for input data. There are different types the proposed algorithm is O(n).
of time complexities, such as Figure 18. As discussed previously, time complexity analysis helps us
The time complexity of the proposed algorithm is analyzed compare the performance of different algorithms developed to
by changing the input size by means of the number of EVs perform a similar function. So now we will utilize this tool to
14 A. MULLA AND H. T. JADHAV

Figure 14. Price of Electricity.

Figure 15. Adjusted R2 index.

Figure 16. Curve Fitted for Price Function.

compare the performance of the proposed algorithm with the shown maximum and minimum time taken per time step (i.e.
ones proposed in the literature by other authors. As mentioned each 15 min time step) by the algorithm to execute for five differ-
before, no other author has performed time complexity analysis ent input sizes. The simulation was performed for 1000 iterative
on their algorithm to the best of the authors’ knowledge. But, experiments. Similarly, the proposed algorithm was simulated
the authors of [24] have done a similar analysis where they have with identical conditions, and a comparison of maximum time
INTERNATIONAL JOURNAL OF COMPUTERS AND APPLICATIONS 15

Figure 17. New Hourly Price for Levelised Load at 50% SOCi and Target Load of (a) 8 MW, (b) 8.4 MW, (c) 8.8 MW, (d) 9.2 MW.

Figure 18. Big O Notation.

Figure 19. Execution Time V Number of EVs.


16 A. MULLA AND H. T. JADHAV

Figure 20. Execution Time Comparison.

Table 3. Performance Evaluation of the Algorithm. scenarios is mainly due to the variation in the number of equa-
Scenario tions that are to be solved by the algorithm. More equations
Execution
Average Initial Target Load Time would have to be solved for the case where there are more eli-
Sr. No. SOC (%) (MW) (second) gible vehicles for the V2G power exchange, relative to the case
1 40 8 28.6938 with fewer eligible vehicles. This can be easily understood by
2 50 8 37.8337 comparing the execution time the performance index shown
3 60 8 64.4005 in Figure 10. We can observe that the performance for a par-
4 70 8 82.1742
5 80 8 80.2059 ticular target load improves, the execution time for the same
6 40 8.4 48.6484 also increases (as the number of the eligible vehicles taking part
7 50 8.4 63.4225 in the V2G power exchange increases) The algorithm consists
8 60 8.4 83.2209
9 70 8.4 85.2231 of 100 PSO iterations, and 100 PSO particle sample size for
10 80 8.4 76.8313 each one of 2000 vehicles sample size for each one of 96-time
11 40 8.8 61.1645 steps. From these numbers, the reader can imagine the enor-
12 50 8.8 74.6292
13 60 8.8 82.8197 mous number of equations this algorithm solves. It should also
14 70 8.8 79.4325 be noted that the execution time (mentioned in Table 3) is taken
15 80 8.8 64.7634 by all the time steps and not by the individual time step, which
16 40 9.2 72.5441
17 50 9.2 79.3448
means that the execution time per time step is far shorter. These
18 60 9.2 75.9330 arguments help to validate the practicability of the proposed
19 70 9.2 65.2840 algorithm for the real-time scheduling of V2G power exchange.
20 80 9.2 49.4116
The performance of the best-case scenario of the proposed
algorithm was compared with the best performance of algo-
rithms proposed in the previous literature on a similar subject.
taken per time step was compared as shown in Figure 20. The For this purpose, a comparative parameter (Pi) was introduced
execution time of the proposed algorithm is greater than that [21]. This parameter can be defined as the percentage improve-
of the reference initially. But, the rate at which the execution ment of peak and valley load differences. The comparative
time increases as input size is increased is far less than that of parameter is formulated in equation (9).
the reference algorithm. We can observe in Figure 20. that the  
 Pd, after − Pd, before 
execution time taken by the proposed algorithm for 300 EVs is Pi =   ∗ 100% (9)
Pd, before 
almost half of the time taken by the reference algorithm.
Speaking of execution time taken by the algorithm for each Here, Pd,before, and Pd,after are the difference between the peak
scenario explained previously in section 5.1. which can fur- load point and the valley load before and after the optimiza-
ther be used to validate the functioning of the algorithm for tion, respectively. The comparison done using this parameter
real-time scheduling purposes. The algorithm was simulated on is shown in Table 4. The reader should note that the reason
MATLAB/Simulink (R2019a) on a workstation (CPU 2.11 GHz, behind this comparative study is not to draw any conclusion that
RAM 8GB, 64bit Operating System) in a lab. Table 3 shows the proposed algorithm has superiority in performance over
the execution time taken by the algorithm for each of the 20 other algorithms represented. Since all these algorithms, includ-
scenarios. Among these 20 scenarios, the minimum execution ing the one proposed in this paper, have been simulated with
time taken is 28.6938 s by the 1st scenario and the maximum input data that is completely random in nature, direct statistical
execution time taken is 85.2231 s by the 9th scenario. This vari- comparison cannot be carried out unless all these algorithms
ation in the execution time of the algorithm across different are simulated using the same input data. The intent behind this
INTERNATIONAL JOURNAL OF COMPUTERS AND APPLICATIONS 17

Table 4. Comparative Analysis of Proposed V2G Algorithm. Systems, International Journal of Electrical Power and Energy Systems
(Elsevier), IET Generation, Transmission & Distribution, Taylor & Francis
Proposed
Parameter Ref. [19] Ref. [21] Ref. [24] Algorithm Journals, and International Journal of Green Energy.

Pd,before (MW) 0.18 3.3 0.035 4.3 Arkan Mulla holds a bachelor’s in Electrical Engineering from ATS’s Sanjay
Pd,after (MW) 0.11 0.4 0.0019 0 Bhokare Group of Institutes, Miraj in 2019. He is currently pursuing a Mas-
Pi % 38.88 87.88 99.98 100 ter’s in Power System and Power Electronics at Rajarambapu Institute of
Technology, Sakhrale. His current research interests include Electric vehi-
cle Scheduling, V2G Technology, Optimization Techniques, Renewable
comparison is to relate the results and validate the proposed Energy Sources, Solar Micro-Inverter.
algorithm.
ORCID
6. Conclusion H. T. Jadhav http://orcid.org/0000-0003-4470-8978

In this paper, an optimal scheduling scheme for V2G operation


is proposed with the objective of minimizing the load variations References
in the power grid around a specified target load. There are var- [1] “Climate Change: Global Temperature NOAA Climate.gov”. https://
ious challenges related to the effective implementation of V2G www.climate.gov/news-features/understanding-climate/climate-cha
power exchange, such as the availability of EVs for V2G oper- nge-global-temperature (accessed Jul. 17, 2020).
ation, ever-changing SOC of EV batteries, etc. The proposed [2] Höök M, Tang X. Depletion of fossil fuels and anthropogenic cli-
mate change-A review. Energy Policy. 2013;52:797–809. doi:10.1016/
optimization algorithm responds to these challenges by taking j.enpol.2012.10.046.
into account the stochastic nature of EV-grid connectivity and [3] Eia. Transportation sector energy consumption Figure 8-1. Delivered
constantly updating the parameters at regular intervals. A thor- transportation energy consumption by country grouping, 2012-40
ough explanation of the algorithm is done using a flowchart and (quadrillion Btu), 2018. Accessed: Jul. 07, 2020. [Online]. Available:
pseudocode. The algorithm is validated by conducting various https://www.eia.gov/outlooks/ieo/pdf/transportation.pdf.
[4] WHO. Transport, Environment and Health. World Health Organiza-
case studies for various scenarios of target load, and average tion. Regional Office for Europe. 2008;2008(2):1–87. Accessed: Aug.
initial SOC. The results of all scenarios are analysed and it is 05, 2020. [Online]. Available: www.euro.who.int.
observed that the proposed algorithm can effectively and sig- [5] Leach F, Kalghatgi G, Stone R, et al. The scope for improving the
nificantly flatten the load curve while complying with all the efficiency and environmental impact of internal combustion engines.
constraints. This can considerably improve the efficiency and Transp. Eng. 2020;1:100005. doi:10.1016/j.treng.2020.100005.
[6] Nazaripouya H, Wang B, Black D. Electric vehicles and climate
performance of the power grid. It is also concluded that the change: additional contribution and improved Economic justifica-
achievement of the desired outcome is considerably affected by tion. IEEE Electrif. Mag. 2019;7(2):33–39. doi:10.1109/MELE.2019.
the selection of the target load and initial SOC of EVs partic- 2908792.
ipating in the V2G power exchange. Hence, these parameters [7] Yilmaz M, Krein PT. Review of the impact of vehicle-to-grid technolo-
have to be considered to achieve the best performance from gies on distribution systems and utility interfaces. IEEE Trans. Power
Electron. 2013;28(12):5673–5689. doi:10.1109/TPEL.2012.2227500.
the algorithm. The effect of the proposed load-leveling method [8] Richardson DB. Electric vehicles and the electric grid: A review of
on the price fluctuation of electricity is elaborated. The advan- modeling approaches, impacts, and renewable energy integration.
tages of the proposed algorithm in terms of the execution time Renew. Sustain. Energy Rev. 2013;19:247–254. doi:10.1016/j.rser.
were further validated by means of time complexity analysis and 2012.11.042.
comparing it with algorithms from previous literature on the [9] Liu H, Qi J, Wang J, et al. EV Dispatch Control for supplemen-
tary frequency regulation considering the expectation of EV owners.
same subject. For future development in this field, the compat- IEEE Trans Smart Grid. 2018;9(4):3763–3772. doi:10.1109/TSG.2016.
ibility of the V2G operation with the integration of renewable 2641481.
energy sources can be investigated considering the stochastic [10] Saiteja K, Krishnarayalu MS. Load frequency Control of Two-area
nature of these sources. smart grid. Int. J. Comput. Appl. 2015;117(14):1–9. doi:10.5120/
20619-3323.
[11] Lin J, Leung KC, Li VOK. Optimal scheduling with vehicle-to-
Disclosure statement grid regulation service. IEEE Internet Things J. 2014;1(6):556–569.
No potential conflict of interest was reported by the author(s). doi:10.1109/JIOT.2014.2361911.
[12] Wang X, He ZY, Yang JW. Unified strategy for electric vehicles
participate in voltage and frequency regulation with active power
Notes on contributors in city grid. IET Gener. Transm. Distrib. 2019;13(15):3281–3291.
doi:10.1049/iet-gtd.2018.7016.
Dr. H. T. Jadhav holds a bachelor’s in Electrical Engineering from Govt [13] Huang Q, Wang X, Fan J, et al. V2g optimal scheduling of
College of Engineering, Karad in 1995, Master’s in Power System from multiple EV aggregator based on TOU electricity price. Proc.
Walchand College of Engg, Sangli in 2003, and Ph.D. from S.V. National - 2019 IEEE Int. Conf. Environ. Electr. Eng. 2019 IEEE Ind.
Institute of Technology, Surat in March 2016. He is also certified as an Commer. Power Syst. Eur. EEEIC/I CPS Eur. 2019;2019(1):1–6.
Energy Auditor by the Bureau of Energy Efficiency (Govt. of India) in doi:10.1109/EEEIC.2019.8783654.
2006. He has teaching experience of over 25 years and is currently work- [14] Zhang W, Spence K, Shao R, et al. Optimal scheduling of spin-
ing as a professor and head of the program at the Electrical Engineer- ning reserve and user cost in vehicle-to-grid (V2G) systems. 2018
ing Department, RIT, Sakhrale. His primary research interests are Wind IEEE Energy Convers. Congr. Expo. ECCE. 2018;2018:1058–1064.
Energy, Economic Dispatch, Optimal Power Flow Studies, and Optimiza- doi:10.1109/ECCE.2018.8558391.
tion Techniques. He has many research articles to his name published in [15] He Y, Venkatesh B, Guan L. Optimal scheduling for charg-
international journals and major IEEE Conference Proceedings. He has ing and discharging of electric vehicles. IEEE Trans Smart Grid.
also served as a reviewer for IEEE CSEE Journal of Power and Energy 2012;3(3):1095–1105. doi:10.1109/TSG.2011.2173507.
18 A. MULLA AND H. T. JADHAV

[16] Wang B, Hu Y, Xiao Y, et al. An EV charging scheduling mech- [27] Sovacool B, Axsen J, Kempton W. Tempering the promise of electric
anism based on price negotiation. Futur. Internet. 2018;10(5). mobility? A sociotechnical review and research agenda for vehicle-
doi:10.3390/fi10050040. grid-integration (VGI) and vehicle-to-grid (V2G). Annu. Rev. Env-
[17] Jian L, Zheng Y, Shao Z. High efficient valley-filling strategy for cen- iron. Resour. 2017;42(August):16.1–16.30. doi:10.1146/annurev-envi
tralized coordinated charging of large-scale electric vehicles. Appl ron- 030117-020220.
Energy. 2017;186:46–55. doi:10.1016/j.apenergy.2016.10.117. [28] Sarabi S, Davigny A, Courtecuisse V, et al. Potential of vehicle-to-
[18] Liang H, Liu Y, Li F, et al. Dynamic economic/emission Dispatch grid ancillary services considering the uncertainties in plug-in electric
including PEVs for peak shaving and valley filling. IEEE Trans. Ind. vehicle availability and service/localization limitations in distribution
Electron. 2019;66(4):2880–2890. doi:10.1109/TIE.2018.2850030. grids. Appl Energy. 2016;171:523–540. doi:10.1016/j.apenergy.2016.
[19] Alam MJE, Muttaqi KM, Sutanto D. A controllable local peak-shaving 03.064.
strategy for effective utilization of PEV battery capacity for distribu- [29] Tan KM, Ramachandaramurthy VK, Yong JY. Integration of elec-
tion network support. IEEE Trans. Ind. Appl. 2015;51(3):2030–2037. tric vehicles in smart grid: A review on vehicle to grid tech-
doi:10.1109/TIA.2014.2369823. nologies and optimization techniques. Renew. Sustain. Energy Rev.
[20] Erden F, Kisacikoglu MC, Erdogan N. Adaptive V2G peak shaving and 2016;53:720–732. doi:10.1016/j.rser.2015.09.012.
smart charging Control for grid integration of PEVs. Electr. Power [30] Senjyu T, Mandal P, Uezato K, et al. Using hybrid correction method.
Components Syst. 2018;46(13):1494–1508. doi:10.1080/15325008. Power. 2005;20(1):102–109.
2018.1489435. [31] Cho H, Goude Y, Brossat X, et al. Modeling and forecasting daily
[21] Tan KM, Ramachandaramurthy VK, Yong JY, et al. Minimization electricity load curves: A hybrid approach. J. Am. Stat. Assoc.
of load variance in power grids-investigation on optimal vehicle- 2013;108(501):7–21. doi:10.1080/01621459.2012.722900.
to-grid scheduling. Energies. 2017;10(11):1–21. doi:10.3390/en1011 [32] Higgs B, Abbas M. Segmentation and clustering of car-following
1880. behavior: recognition of driving patterns. IEEE Trans. Intell.
[22] Ioakimidis CS, Thomas D, Rycerski P, et al. Peak shaving and val- Transp. Syst. 2015;16(1):81–90. doi:10.1109/TITS.2014.232
ley filling of power consumption profile in non-residential build- 6082.
ings using an electric vehicle parking lot. Energy. 2018;148:148–158. [33] The International Council on Clean Transportation. Global and U.S.
doi:10.1016/j.energy.2018.01.128. electric vehicle trends. Int. Counc. Clean Transp. 2019: 1–22.
[23] Jian L, Xue H, Xu G, et al. Regulated charging of plug-in [34] “Tesla Model 3 - Wikipedia.”. https://en.m.wikipedia.org/wiki/Tesla_
hybrid electric vehicles for minimizing load variance in household Model_3 (accessed Aug. 01, 2020).
smart microgrid. IEEE Trans. Ind. Electron. 2013;60(8):3218–3226. [35] “Chevrolet Bolt - Wikipedia.”. https://en.m.wikipedia.org/wiki/Chev
doi:10.1109/TIE.2012.2198037. rolet_Bolt#Specifications (accessed Aug. 01, 2020).
[24] Jian L, Zheng Y, Xiao X, et al. Optimal scheduling for vehicle-to-grid [36] “Nissan Leaf - Wikipedia.”. https://en.m.wikipedia.org/wiki/Nissan_
operation with stochastic connection of plug-in electric vehicles to Leaf (accessed Aug. 01, 2020).
smart grid. Appl Energy. 2015;146:150–161. doi:10.1016/j.apenergy. [37] “Volkswagen e-Golf price and specifications - EV Database.”.
2015.02.030. https://ev-database.org/car/1087/Volkswagen-e-Golf (accessed Aug.
[25] Liu BC, Ieee M, Chau KT, et al. Opportunities and challenges of 01, 2020).
vehicle-to-home, vehicle-to-grid technologies. Proceedings of the [38] Kim HT, Jin YG, Yoon YT. An Economic Analysis of Load Leveling
IEEE. 2013;101:2409–2427. doi:10.1109/JPROC.2013.2271951. with Battery Energy Storage Systems (BESS) in an Electricity Market
[26] Gough R, Dickerson C, Rowley P, et al. Vehicle-to-grid feasibility: A Environment: The Korean Case. 2019.
techno-economic analysis of EV-based energy storage. Appl Energy.
2017;192:12–23. doi:10.1016/j.apenergy.2017.01.102.

You might also like