You are on page 1of 12

This article has been accepted for publication in IEEE Transactions on Industry Applications.

This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/TIA.2024.3384343

An Optimal Hybrid Management of Electric Vehicle


Fleet Charging and Load Scheduling in Active
Electric Distribution System
Ali Ihsan Aygun, Member, IEEE, Md Shamim Hasan, Member, IEEE, Aniket Joshi, Member, IEEE, and
Sukumar Kamalasadan, Senior Member, IEEE

Abstract—This paper introduces a novel scheduling framework Pi,t The charging profile of ith EV, at time t
designed to manage the charging of electric vehicles (EVs) in Pi Injected active power at bus, i
a way that considers its effects on the power grid. Leveraging Pi The charging profile of ith EV
the Alternating Direction Method of Multipliers (ADMM), the
methodology offers a significant advantage by enabling decen- Qi,G Total reactive power generation at bus, i
tralized sub-problems, allowing for efficient and rapid solu- Qi Injected reactive power at bus, i
tions. The methodology developed as an algorithmic framework SoC The state of charge of an EV
incorporates various scheduling approaches for EV charging, SoCi,dep The state of charge of ith EV, at departure
including demand management techniques like valley filling and SoCi,req The desired state of charge of ith EV
peak shaving, along with real-time pricing (RTP) considerations.
These strategies aim to modify individual electricity consumption Ta , Td Represents the arrival and departure time of EV
patterns to reduce peak demand, ultimately enhancing energy Vi Voltage magnitude of bus, i
efficiency and ensuring the stability of the power system. The x1 , x2 Conditionally separable variables
results of the study highlight the crucial role of distributed opti- y An element common in both x1 , and x2
mization in improving both demand management strategies and
cost objectives. The results indicated that the proposed method
shows significant improvement in overall energy efficiency when I. I NTRODUCTION
compared to the state-of-the-art centralized convex optimization
HE growing adoption of electric vehicles (EVs) has
framework.
Index Terms—Electric Vehicle Integration, Energy Manage-
ment System, Alternating Direction Method of Multipliers
T brought about significant challenges and opportunities for
the electric distribution network [1]. The integration of EVs
(ADMM), Load Balancing, Demand Side Management. into the grid introduces a substantial increase in electricity
demand, particularly during peak hours, thereby placing ad-
ditional strain on the distribution infrastructure [2]. However,
this rise in EV usage also presents an opportunity for intelli-
N OMENCLATURE
gent management and optimization of the charging process to
α Penalty co-efficient for degradation cost , C ensure grid stability and efficient utilization of resources [3].
∆t Computational time step A smart charging algorithm is a crucial component of elec-
λ The Lagrange multipliers tric vehicle (EV) charging infrastructure, aiming to optimize
θi Bus Voltage angle of bus i charging efficiency and minimize grid stress. Smart charging
(.), (.) Represent the min. and max. of a variable (.) applications offer numerous advantages in EV charging. First,
a Update step size in each iteration step of ADMM these applications provide users with real-time monitoring and
B Selected bus load control of charging sessions, allowing them to optimize their
Bii Represents susceptance values of bus, i charging schedules and avoid peak demand periods, ultimately
Bij Represents the susceptance values associated reducing costs and enhancing grid stability [4]. Second, smart
with branch connecting between bus i and j charging applications often integrate smart grid technologies,
C Degradation cost enabling bidirectional communication, especially between EVs
Gii Represents conductance values of bus, i and the power grid, facilitating demand response programs
Gij Represents the conductance values associated and load balancing [5]. By participating in such programs,
with branch connecting between bus i and j EV owners can take advantage of financial incentives, such as
m1 , m2 , m3 Weightage factors for COP reduced electricity rates or monetary rewards, while supporting
p Denotes the electricity price $/kW h the integration of renewable energy sources [6]. Overall,
Pagg The load profiles of the aggregators smart charging applications empower EV owners by providing
Pi,G Total real power generation at bus, i greater control, cost savings, environmental benefits, and a
seamless charging experience.
Ali Ihsan Aygun is affiliated with the Electrical and Electronic Engineering
Department at Duzce University, Turkey. Aniket Joshi is with the US Research However, despite the advantages offered by the smart
Center, ABB Inc., Raleigh, NC, 27606 USA. Md Shamim Hasan and Sukumar charging algorithm, certain drawbacks need to be discussed.
Kamalasadan are associated with the Energy Production and Infrastructure One drawback is the potential increase in the complexity of
Center and the Department of Electrical and Computer Engineering at the
University of North Carolina at Charlotte, Charlotte, NC 28223, USA (email: implementation, particularly in large-scale deployments [7].
skamalas@uncc.edu). Additionally, the algorithm relies on accurate information

Authorized licensed use limited to: University of North Carolina at Charlotte. Downloaded on April 03,2024 at 04:51:53 UTC from IEEE Xplore. Restrictions apply.
© 2024 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.See https://www.ieee.org/publications/rights/index.html for more information.
This article has been accepted for publication in IEEE Transactions on Industry Applications. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/TIA.2024.3384343

about the EV fleet, real-time pricing data, and grid conditions, presented. In this case, the DSO deals with Optimal Power
which may pose challenges in practice [8]. Moreover, the Flow (OPF) issues in scenarios with a high concentration
frequent on-off charging events associated with smart charging of EVs. The main challenge involves managing a complex
algorithms can lead to increased battery degradation and objective function with interconnected power flow constraints.
reduced battery life, impacting the overall longevity and cost- In [20], a novel distributed real-time ADMM technique is
effectiveness of EVs [9]. In the past, it was shown that the real- introduced for voltage regulation through controlling EVs and
time modeling and control of loads with renewable generation BESS, focusing on maximizing their utility function [21].
needs to be studied for the balance of the power grid with The authors have devised a single-loop iterative algorithm to
variable renewable generation [10], [11]. Also, the EVs acting address the problem of scheduling EV charging by integrating
as a load and generation provides a different challenge in the ADMM with the distribution line power flow model [22].
sense that the charging of EVs should coordinate with existing None of the above-mentioned state-of-the-art research
loads in the electric grid. works considered formulating the EV management framework
The solution procedures to address the challenges of smart considering multiple functionalities, which include real-time
fleet charging and load scheduling strategies can be catego- pricing, demand management, grid impact, cost-effectiveness,
rized as centralized or distributed. Centralized approaches rely stability, and efficient utilization of resources [23]. Having
on a central authority, often a distribution system operator multiple functionalities is critical for EV management due
(DSO), to collect and manage all the data about the network, to the fact that it can provide supply-side and demand-side
[12]. Since all decisions are consolidated at a central hub, balances, which is important when dealing with fleet-wide
centralized algorithms are known for their improved accuracy EV charging. Moreover, there is a critical need to address
and controllability [13]. Nevertheless, the substantial workload the stability, maximum utilization of resources, and economics
associated with data collection, processing, and dispatching using an aggregator framework, especially in the wake of
control signals can impede efficiency, especially when deal- recent grid mandates such as FERC order no. 2222 [24]. This
ing with large EV fleets. To address this, various control paper proposes an innovative approach for managing EV fleet
methods have been developed to improve the efficiency of charting and load scheduling by utilizing convex optimization
centralized charging strategies, often focusing on simplified techniques and integrating real-time pricing information. This
EV aggregation models [14]. However, it is worth noting ADMM-based approach aims to minimize the overall charging
that in these models, the EV Aggregator is an intermediary cost for EV fleets while mitigating the impact on the power
connecting DSOs and EV fleets. While these approaches grid [25]. The framework facilitates faster and more efficient
effectively enhance the efficiency of centralized EV charging decision-making by dividing the problem into decentralized
scheduling, they are not without their challenges, mainly sub-problems and iteratively optimizing the solution that re-
related to security. These issues encompass vulnerabilities to duces peak demand, minimizes costs, and ensures the stability
single-point failures and concerns about the privacy of end- and reliability of the power system.
user data [15]. In our earlier work, a design on a centralized framework
In contrast to centralized methods for scheduling EV charg- for EV charging [26] was developed where it was proved that
ing, distributed approaches involve breaking down the pri- the centralized optimization framework is: a) scalable and grid
mary problem into smaller sub-problems and distributing them aware, b) considers both the grid side and c) capable of shaving
among various entities, including EVs, EV aggregators, and peak demand, thus addressing few multiple functionalities
DSOs. There may be some degree of interconnection at the for EV management. In this paper, a distributed approach is
level of constraints or objectives, as pointed out in [16]. Each designed that can be integrated with multiple aggregators in
entity autonomously and concurrently addresses these smaller the electric distribution system and thus can provide scalable
sub-problems through a distributed iterative process. As a multiple objective management of EVs considering flexible
result, distributed EV charging strategies exhibit robustness loads. Compared to the state-of-the-art and our earlier work
against single-point failures and uphold the privacy of EV [26], the main contributions and the novelty of the proposed
users. This is possible because, unlike centralized approaches, approach are as follows.
distributed approaches do not require sharing sensitive infor-
mation about EVs with a higher-level entity. • The architecture integrates multiple functionalities to
Among the distributed approaches, the Alternating Direction manage EV fleets, such as real-time pricing, demand
Method of Multipliers (ADMM) [17] has recently attracted management, priority-based charging, cost-efficiency, sta-
significant attention. This is primarily due to its improved bility, and the optimal utilization of resources using
convergence property and the ability to achieve reasonably convex optimization and McCormick relaxation, and the
accurate results in distributed fleet charging and load man- results are illustrated compared to the state-of-the-art.
agement. Ref. [18] introduces an energy management system • This approach offers Vehicle-to-Grid (V2G) capabili-
that relies on the ADMM framework and incorporates chance ties with the dual objectives of influencing consumers’
constraints to address the stochastic behavior of EVs. The au- electricity usage patterns to reduce peak demand and
thors tackled the EV charging schedules through optimization. prioritizing the needs of drivers who may require faster
However, the research should provide more insights into how charging compared to others by using a privilege co-
the proposed charging solution affects the distribution grid. In efficient, improving energy efficiency, and strengthening
[19], a distributed multi-period problem based on ADMM is the stability of the electric grid.

Authorized licensed use limited to: University of North Carolina at Charlotte. Downloaded on April 03,2024 at 04:51:53 UTC from IEEE Xplore. Restrictions apply.
© 2024 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.See https://www.ieee.org/publications/rights/index.html for more information.
This article has been accepted for publication in IEEE Transactions on Industry Applications. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/TIA.2024.3384343

The paper is organized as follows. Section II discusses the where λ is a vector of Lagrangian variables. The problem can
mathematical preliminaries, and III outlines the design frame- be divided into two groups and minimize (x1 , y1 ) and (x2 , y2 )
work and theoretical approach. In Section IV, case studies are separately (7):
presented, and the simulation results are discussed, followed
sub-problem 1: g1 (λ) = inf (F1 (x1 , y1 ) + λT y1 )
by conclusions in Section V. x1 ,y1
(7)
II. P RELIMINARIES : D ISTRIBUTED O PTIMIZATION sub-problem 2: g2 (λ) = inf (F2 (x2 , y2 ) + λT y2 )
x2 ,y2
The objective of distributed optimization is to tackle a com- The following steps are involved in employing this method:
plex optimization problem by breaking it down into smaller
1) Concurrently solve the dual sub-problems: find x1 , y1
sub-problems, along with a master problem that gradually con-
that minimize f1 (x1 , y1 ) + λT y1 and determine x2 , y2
verges towards the optimal solution of the original problem.
that minimize f2 (x2 , y2 ) + λT y2 .
By ensuring rapid convergence of the master method and
2) Update the dual variables and λ.
simplicity in solving the sub-problems, computational time
savings can be achieved compared to centralized optimization. With each iteration, the solution progresses closer to obtaining
The following optimization problem is considered: the optimal solution for the original problem. For a more
comprehensive analysis, detailed information on dual decom-
min f (x1 ) + g(x2 ) position is analyzed as discussed in [27].
(1)
subject to x1 ∈ C1 , x2 ∈ C2
A. Alternating Direction Method of Multipliers (ADMM)
The subsystems f (x1 ) and g(x2 ) consist of conditionally Based Optimal EV Management
separable variables x1 and x2 , respectively in (1). When a Primal and dual decomposition methods suffer from a
common element exists between f (x1 ) and g(x2 ), the problem drawback where strict convexity of the objective function is
is not separable. In practical situations, such as: required to achieve an optimal solution. To overcome this
min f (x) = f1 (x1 , y) + f2 (x2 , y) (2) drawback, the ADMM introduces a regularization parameter to
the dual decomposition process, enhancing its stability. In this
The two sub-problems in (2) can be separately minimized, section, an overview of the key components of ADMM, which
followed by an iterative procedure to minimize the main is a numerical algorithm specifically designed for optimization
problem. This process can be described as follows: problems, is provided. ADMM combines the advantages of
decomposed dual ascent methods with the superior conver-
sub-problem 1: minimize f1 (x1 , y)
(3) gence properties of the method of multipliers. The ADMM
sub-problem 2: minimize f2 (x2 , y) formulation is expressed as:
with optimal values of δ1 (y) and δ2 (y). Consequently, the min f (x) + g(z)
original problem of (3) is transformed into (4): (8)
subject to Ax + Bz = c
master problem: minimize δ1 (y) + δ2 (y) (4) In this formulation (8), the optimization variables are repre-
sented as x ∈ Rn and z ∈ Rm . The parameters x ∈ Rp×n ,
with variables y. This approach can be formulated in the B ∈ Rp×m , and c ∈ Rp are considered as inputs. The values
form of primal decomposition. The sub-problems are solved of x and z are computed iteratively. During each iteration,
in parallel by a) finding x1 that minimizes f1 (x1 , y) with a the value of x is initially computed while keeping z constant,
sub-gradient g1 ∈ δ1 (y), and b) finding x2 that minimizes and subsequently, the value of z is solved using the updated x
f2 (x2 , y) with a sub-gradient g2 ∈ δ1 (y). The connecting from the preceding step. The updated equations for the state
variables are then updated as y = y − a(g1 + g2 ), where a and decision variables are formulated as follows:
represents the step size. Each iteration of the algorithm brings n ρ 2
o
the solution closer to the optimal solution of the main problem. xk+1 := argmin f (x) + Ax + Bz k − c + uk 2 (9)
z 2
Conversely, dual decomposition involves a different ap- n ρ k+1 2
o
proach where the emphasis is on optimizing the sub-problems z k+1 := argmin g(z) + x + Bz − c + uk 2 (10)
initially and subsequently optimizing a group of dual variables x 2
associated with the main problem using an iterative process. uk+1 := uk + Axk+1 + Bz k+1 − c (11)
In the context of the given problem, the modified variables y1
and y2 are considered as follows: In (9-11), the parameter ρ, which is a positive value, represents
the penalty parameter, while the vector u represents the
minimize f (x) = F1 (x1 , y) + F2 (x2 , y) Lagrangian multipliers.
(5)
subject to y1 = y2 III. P ROPOSED EV S MART C HARGING F RAMEWORK
where y1 and y2 are local versions of the variable y, the The growing adoption of EVs will impose additional stress
Lagrangian of the modified problem (5) can be formulated on the electric grid infrastructure. The grid experiences its
as (6): highest demand during a short duration, necessitating the
activation of high-cost generation units to meet this peak
L(x1 , y1 , x2 , y2 ) = F1 (x1 , y1 ) + F2 (x2 , y2 ) + λT (y1 − y2 ) demand. Historically, the conventional approach to addressing
(6) increased energy demand has involved adding more generation

Authorized licensed use limited to: University of North Carolina at Charlotte. Downloaded on April 03,2024 at 04:51:53 UTC from IEEE Xplore. Restrictions apply.
© 2024 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.See https://www.ieee.org/publications/rights/index.html for more information.
This article has been accepted for publication in IEEE Transactions on Industry Applications. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/TIA.2024.3384343

where Ta and Td represent the arrival and departure times,


respectively.
Furthermore, it is essential to consider power flow con-
straints in order to gain a comprehensive understanding of the
grid’s operating conditions. The injected active power at bus
i is denoted as Pi and the injected reactive power as Qi , with
N representing the total number of buses in the distribution
network. The power balance equation can be expressed as [28]:

X X
Pi,G − Pi,L = Vi2 Gii
n
X
+ Vi Vj [Gij cos(θi − θj ) + Bij sin(θi − θj )] (17a)
Fig. 1: System’s communication infrastructure. j=1

X X
capacity. However, the escalation in demand has negative Qi,G − Qi,L = −Vi2 Bii
consequences for the system’s overall operation. n
X
In EV smart charging, the aggregator plays a crucial role + Vi Vj [Gij sin(θi − θj ) − Bij cos(θi − θj )] (17b)
as an intermediary between the utility and the EVs (Fig. j=1
1). Aggregators receive data regarding the status of electric Here, the variables Vi and Vj correspond to the voltage
vehicles, including information such as state-of-charge (SoC), magnitudes at buses i and j, respectively. The parameters
arrival and departure times, and battery capacity. Based on Gii and Bii represent the conductance and susceptance values
the data, the aggregator optimizes and sends the optimal associated with bus i. In contrast, Gij and Bij denote the
solution to the central energy market. Based on the information conductance and susceptance of the branch connecting buses i
exchange between the energy market and EVs, the aggregator and j. The solution variables for the optimization problem is a
finds the optimal charging scheduling (CS) for the respected set of power flow parameters, denoted as [θi , Vi , Pi,G , Qi,G ]T .
EVs. Then, the aggregator sends charging and discharging Each power flow variable is subject to minimum and maximum
signals to the EVs. The sum of the electric power at the limits, which restrict their values to a specified range as
aggregator is determined by the sum of the charging power follows:
for all EVs, which can be expressed as: 
K
X M
X


 θi ≤ θi ≤ θi
Paggi (t) = Pi,n (t) (12) V ≤ V ≤ V

i i i
i=1 i=1 (18)
P i,G ≤ Pi,G ≤ P i,G


In (12), K signifies the overall count of aggregators, M rep- Q ≤ Qi,G ≤ Q

i,G
i,G
resents the number of EVs under the control of the aggregator
connected to the bus, and Pi,n (t) represents the total power The (.) and (.) represent the minimum and maximum of
required for charging EV, i at time t. The SoC of an EV at a variables, respectively.
given time t can be formulated as (13).
A. Formulation of Combined Problem, and Convex Relaxation
P i,t ∗ ∆t
SoC i,t = SoC i,t-1 + (13) This architecture brings together various features to effi-
Ci ciently manage electric vehicle fleets. These features include
Here, Ci represents the capacity of the EV battery, Pi,t stands real-time pricing, demand management, priority-based charg-
for the power output of the charger, and ∆t denotes the ing, cost-effectiveness, stability, and the most effective use of
time interval. The boundary conditions of the SoC can be available resources. This also takes into account Vehicle-to-
determined as Grid (V2G) capabilities and customers’ usage patterns. Based
on these, the combined problem can be formed as follows:
SoC min < SoC i,t < SoC max (14)
min m1 pT (Pagg + α ∗ C)∆t + m2 pT (B + P agg + α ∗ C)∆t
The (SoC) and the battery voltage level are regulated within a +m3 (Dmin + Pagg )∆t + (1 − m1 − m2 − m3 )pT (P agg )∆t
predetermined range. Given the lifespan of cars, it is important  subject to (17) − (18) and,
to avoid excessive charging or discharging of batteries. 
 0 ≤ P agg ≤ P agg,max

0 ≤ P i,t ≤ P i,max

SoC i,dep >= SoC i,req

(15) 


min ≤ C agg ≤ C max
 C
SoCi,dep and SoCi,req denote the SoC level of EV(i) at the  SoCi,t ≤ SoC max
time of departure and the desired SoC level, respectively. The


 SoC min ≤ SoC i,t ≤ SoC max


relation can be expressed as follows: 

SoC i,req. ≤ SoC i,dep

t ∈ (Ta ∼ Td ) (16) (19)

Authorized licensed use limited to: University of North Carolina at Charlotte. Downloaded on April 03,2024 at 04:51:53 UTC from IEEE Xplore. Restrictions apply.
© 2024 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.See https://www.ieee.org/publications/rights/index.html for more information.
This article has been accepted for publication in IEEE Transactions on Industry Applications. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/TIA.2024.3384343

where ∆t represents time step, p denotes the price, C repre-


sents the degradation cost, α is the coefficient associated with
the degradation cost, and B is equal to the selected bus load. In
addition, m1 , m2 , and m3 are weightage factors for different
approaches. Moreover, the first part (pT (Pagg + α ∗ C)∆t),
the second part (pT (B + P agg + α ∗ C)∆t), the third part
((Dmin + Pagg )∆t) and the fourth part (pT (P agg )∆t) cor-
responds to minimizing charging cost, vehicle to grid ap-
plication, valley filling approach and priority-based charging
approach, respectively. These approaches are described in
detail in Section IV.
For convexification, the square of the voltage and current
variable Vi2 and |Iij |2 are relaxed as = ui and lij . If Fig. 2: Modified IEEE 123-bus with three aggregators.
we consider the weightage factors as mi and the different
approaches as yi , the one single approach can be expressed The communication between the aggregator and energy market
as: is performed through macro-iterations of distributed ADMM
w = mi yi (20) defined in (22)-(24).
Now, as m1 , m2 , and m3 , weightage factors, are optimization
variables and (20) is non-convex, we have to update these
IV. C ASE S TUDIES AND R ESULTS
through McCormick envelopes for convexification [29].
The tests are conducted on a real-life system mapped based
w ≥ mi yi + mi yi − mi yi
on the IEEE 123 bus to Charlotte, an eastern US city. The
w ≥ mi yi + mi yi − mi yi IEEE 123-bus load is multiplied by the load multiplication fac-
w ≤ mi yi + mi yi − mi yi (21) tor. The details of the aggregator in the IEEE 123-bus system
w ≤ mi yi + mi yi − mi yi are illustrated in Fig. 2 [30]. The base voltage of the system
is 4.16 kV and comprises 127 power lines. Additionally, it
B. ADMM-based Smart Charging Framework
features three voltage regulators positioned between Bus 10
The proposed distributed smart charging framework is de- and Bus 15, Bus 26 and Bus 27, and Bus 118 and Bus 68. In
signed using ADMM. The problem at hand involves a multi- the current study, three aggregators are placed at three different
criteria optimization between the aggregator, which manages locations named buses 11, 21, and 31. Three aggregators
the charging stations, and individual EVs within a specified can have different numbers of EV charging stations, and one
time horizon t ∈ {1, . . . , T } with a fixed number of EVs. aggregator will act as one area of the ADMM algorithm along
Leveraging the ADMM technique (9)-(11), a smart charging with nearby buses. The IEEE 123-bus has been divided into
algorithm can be defined as follows: three areas based on these three aggregators’ positions. Two
sets of EVs (50 and 250) are considered for the studies that are
( )
k+1 ρ X 2
Pagg := argmin fobj + IN + Paggi − 1T PVk i + uk 2 distributed on these three buses. An example of the locations
i
P aggi 2 i
of EVs on each bus based on the city map is illustrated in Fig.
n ρ 2
o (22) 3. The Charlotte-rated load has been considered as the rated
Pik+1 := arg min Ii + P k+1 + 1T Pi + uk 2 (23) load of the IEEE 123-bus system, while the 1440 minutes (24
Pi 2 aggi
X hours) load profile factor has been given in Fig. 4. The load on
uk+1 := uk + k+1
(Paggi
− 1T Pik+1 ) (24) each bus is the spot load multiplied by the load profile factor.
i Five case studies are conducted with real-life data and field
In this method, the aggregator aims to optimize the total power models that represent a local utility feeder modified using the
load profiles, represented by the vector Paggi ∈ R1×T , where IEEE 123 bus system.
2
∥.∥2 denotes the Euclidean norm. The vector 1 represents a The charging or discharging power rate of the charger needs
column vector of ones with a length of M . On the other to fall within a predefined range. The specific values of the
hand, each electric vehicle (EV) seeks to optimize its charging parameters utilized for the study are provided in Table I. The
profile, denoted by the vector Pi = [P1 , ...., PM ]T , where following assumptions were made for the illustration of cases.
M represents the total number of plug-in electric vehicles • The charging duration for electric vehicles is assumed to
(PEVs). The vector Pi ∈ R1×T describes the charging profile be 24 hours, divided into intervals of 5 minutes.
for an individual EV over time (T). The indicator indicates • Aggregators have multiple objectives depending on the
the constraints imposed by the network and the aggregated specific service they aim to provide.
vehicles charging functions IN and IVi . • The charging needs of electric vehicles must be fulfilled

0, Paggi by the end of the day.
IN = Certain individuals prefer their electric cars to be charged
∞, otherwise •
 (25) as rapidly as possible.
0, Pi
IVi = • It is assumed that all parameters are known.
∞, otherwise

Authorized licensed use limited to: University of North Carolina at Charlotte. Downloaded on April 03,2024 at 04:51:53 UTC from IEEE Xplore. Restrictions apply.
© 2024 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.See https://www.ieee.org/publications/rights/index.html for more information.
This article has been accepted for publication in IEEE Transactions on Industry Applications. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/TIA.2024.3384343

Algorithm 1: Proposed Smart Charging Algorithm


1: Require
1: Initialize all the variables;
2: Obtain
1: EVs’ arrival/departure time;
2: EVs’ initial and desired level of SoC;
3: DSO obtains non-EV load profile and market price;
3: while true do
1: Update base demand and EV populations demand at t;
2: Run Algorithm based on (26) or (29) or (30) or (35)
based on the appropriate approach by considering
(12)-(16);
Fig. 3: Location of aggregator buses on the GIS map. 3: Update variables through (9)-(11) and (22)-(24);
4: check optimality condition;
5: If the optimal condition is satisfied, Compute the
optimal charging schedule and go to step 6, or if not
satisfied, return to step 2;
end while
4: Send charging and discharging signal

TABLE II
C HARGING C OST FOR C ASE 1
No. of Before proposed After proposed Saving
EVs approach approach (%)
Fig. 4: Load profile factor of each time (minute). Agg. 1 752.4 345.07 54 %
50
Agg. 2 752.4 345.07 54 %
TABLE I EVs
Agg. 3 752.4 345.07 54 %
C HARACTERISTICS OF C ASE S TUDIES Agg. 1 4750 3562.5 25 %
250
Agg. 2 4750 3562.5 25 %
EVs
Vehicle Number 50 250 Agg. 3 4750 3562.5 25 %
Vehicle Battery Capacity 30 kWh 30 kWh
Maximum EV Space 50 50
Initial Charge Level 9kWh 9kWh
Final Charge Level 30kWh 30kWh The utility supplies the main source of power, while an aggre-
Maximum Power Output 7 kW 7 kW gator takes on the role of a central controller responsible for
∆t (Time Step) 5 min. 5 min. executing the suggested optimal billing method. For this case,
the optimization model is illustrated below, with outcomes
presented for each EV at the charging station.
All the simulations are done for 1440 daily minutes, which
has been divided into 288 times steps, each with 5 minutes. min pT P agg ∆t
The comparisons are with the state-of-the-art convex optimal subject to (17) − (18) and,

power flow-based centralized model adopted from [28] rep-  0 ≤ P agg ≤ P agg,max
(26)


 0 ≤P ≤P
resented as "without proposed approach". All the simulations i,t i,max
are done in MATLAB platform in a Windows computer with 
 SoCi,t ≤ SoC max
a 2.8 GHz Intel Core-i5 processor and 16 GB of memory

SoC i,req. ≤ SoC i,dep

and implemented using the CVX solver [31]. In the cases
following, we will show how each function can be formulated For the case studies, we consider 50 and 250 electric vehicles,
in the form of a modified ADMM framework and then, respectively. The initial and desired SoC demands are assumed
finally, how they can be combined. The overall framework to be known, and the algorithm generates a charging plan for
is illustrated using Algorithm 1. each vehicle based on the energy price. 50 charging spaces are
presumed to be accessible, enabling simultaneous charging of
A. Case 1-Minimizing Charging Costs up to 50 cars. The details pertaining to the case study can be
found in Table I.
The purpose of this objective is to shift high-demand loads
from peak hours to off-peak hours without affecting overall
TABLE III
electricity consumption. To achieve this, a convex optimization
C HARGING C OST C OMPARISON WITH D ISTRIBUTED
model is introduced to plan the charging of electric vehicles
O PTIMIZATION A PPROACH FOR C ASE 1
(EVs) while considering real-time pricing (RTP) forecasts. The
goal is to optimize power procurement costs while also pe- No. of Distributed optimi- Proposed distibuted Improvement
nalizing degradation resulting from frequent on-off switching. EVs zation approach [32] optimization approach (%)
50 81.686 79.74 2.38

Authorized licensed use limited to: University of North Carolina at Charlotte. Downloaded on April 03,2024 at 04:51:53 UTC from IEEE Xplore. Restrictions apply.
© 2024 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.See https://www.ieee.org/publications/rights/index.html for more information.
This article has been accepted for publication in IEEE Transactions on Industry Applications. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/TIA.2024.3384343

Fig. 5: Schedule for charging 50 EVs with and without the Fig. 7: A comparison of busbar demand without EV load,
proposed method. busbar demand with EV load with and without the proposed
approach.

Fig. 6: Schedule for charging 250 electric vehicles with the


proposed method. Fig. 8: The difference of busbar demand of average daily load
(including 50 EVs) with and without the proposed approach.
By taking into account the demand from the aggregator, the
algorithm computes the lowest charging cost and transmits in Table II, which shows around 54% savings for 50 EVs and
signals to the chargers to regulate the charging schedule. 25% savings for 250 EVs.
If necessary, electric vehicles can undergo multiple charging To further illustrate the impact of the distributed opti-
cycles. The outcomes of the ADMM iteration with 50 vehicles mization approach, the charging cost is compared with the
are depicted in Fig. 5. Fig. 6 illustrates the impact of ADMM distributed approach (Method-3) described in [32]. The load
with 250 EVs, where the number of available charging stations profile and characteristics of 50 EVs are also taken from [32]
is limited to 50, resulting in some vehicles having to wait. to get a similar result, and the result comparison is illustrated
In such cases, the algorithm determines the optimal charging in Table III. From the comparison, it can be seen that 2.38%
period for the remaining vehicles once the 50 charging spots more savings are attainable from the proposed approach. The
are occupied. A comparison of actual busbar demand without execution of the algorithm reveals the occurrence of multiple
the EV load, average bus bar demand, busbar demand with the switching charging positions. As the battery is a vital and
EV load without the proposed approach, and busbar demand expensive component in EVs, frequent on-off events can
with EV load with the proposed approach, is illustrated in negatively affect its lifespan. To address this limitation, the
Fig. 7, which shows that the proposed methodology finds an objective function is modified as follows:
optimal charging schedule in the low-price electricity time
steps. The difference in busbar demand of average daily load min pT (Pagg + α ∗ C)∆t (27)
(including 50 EVs) with and without the proposed approach
is illustrated in Fig. 8, which shows significant improvement Here, C represents the degradation cost, which is defined as:
in load variations with the proposed approach.
P1440
Additionally, to demonstrate the effectiveness of the algo- C= t=1 abs[P n,t − P n,t-1 ] (28)
rithm, the cost of EV charging without the proposed method
is calculated based on a convex optimization-based centralized The results obtained with different penalty coefficients for
algorithm adopted from [28], and the cost of EV charging is 50 EVs, represented as α, are displayed in Fig. 9. It can be
calculated after applying the proposed algorithm. The cost and seen that the load variations can be reduced with coefficients,
the savings achieved from the proposed algorithm are tabulated which also optimizes the output power.

Authorized licensed use limited to: University of North Carolina at Charlotte. Downloaded on April 03,2024 at 04:51:53 UTC from IEEE Xplore. Restrictions apply.
© 2024 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.See https://www.ieee.org/publications/rights/index.html for more information.
This article has been accepted for publication in IEEE Transactions on Industry Applications. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/TIA.2024.3384343

Fig. 9: EV load (KW) for different values of α for the proposed


approach.
Fig. 11: Profile of the bus load with and without the imple-
B. Case 2-Vehicle to Grid (V2G) Application mentation of V2G technology for α = 0.25.
In contrast to conventional peak-shaving and valley-filling
applications, the integration of V2G presents a more cost-
effective and efficient alternative. It also offers the advantage
of quickly responding to changes in grid demand, as discussed
in [33]. Various sources, including [34]–[36], suggest that
EV drivers can be incentivized by charging their vehicles
and actively help reduce the negative effects of charging
on the grid. The optimization problem for the V2G case is
represented as:
min pT (B + P agg + α ∗ C)∆t
subject to (17) − (18) and,

 P agg,min ≤ P agg. ≤ P agg,max
(29)


i,min ≤ P i,t ≤ P i,max
 P

 SoC min ≤ SoC i,t ≤ SoC max



 Fig. 12: Profile of the bus load with and without the imple-
SoC i,req. ≤ SoC i,dep

mentation of V2G technology for α = 0.5.
where ∆t is the time step, B is equal to the selected bus load,
C is equal to the cost of degradation, and α is the coefficient can be seen that the algorithm finds the charging/discharging
associated with the degradation cost. schedule based on peak hours and also tries to utilize the off-
peak load demand for α = 0.5 while α = [0.05, 0.25] uses
the off-peak demand period, but after time steps 230, both
these have some peak demands. So, it can help to shave peak
demand hours and fill up the off-peak demand periods.
Moreover, the voltage profile for aggregators (Agg.) 1, 2,
and 3 (bus 11, 21, and 31) with EV load for the state-of-the-art
centralized approach and proposed ADMM-based distributed
approach are plotted in Fig. 13 for the case with α = 0.5. From
Fig. 13, it can be seen that the voltage goes down to 0.865
p.u. with the state-of-the-art centralized approach, while with
the proposed distributed charging approach, the voltage goes
down to 0.92 p.u. So, the voltage profile has been improved
significantly. It can be noted that when the value of α is
raised more than 0.5, the algorithm no longer provides any
advantages to the vehicles and instead concentrates solely on
meeting the charging needs. The choice of the α value can be
Fig. 10: Profile of the bus load with and without the imple- adjusted according to the desired gain level to be achieved.
mentation of V2G technology for α = 0.05. C. Case 3-Valley Filling Approach
Fig. 10, 11, and 12 demonstrate the outcomes obtained from The objective of the valley-filling function is to enhance
the ADMM iteration with 50 vehicles, utilizing various values electricity consumption during periods of low demand, re-
of α where the parameters are as discussed in Section IV-A. It ferred to as "valleys." These functions primarily focus on

Authorized licensed use limited to: University of North Carolina at Charlotte. Downloaded on April 03,2024 at 04:51:53 UTC from IEEE Xplore. Restrictions apply.
© 2024 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.See https://www.ieee.org/publications/rights/index.html for more information.
This article has been accepted for publication in IEEE Transactions on Industry Applications. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/TIA.2024.3384343

(a)
Fig. 14: Profile of the bus load with and without the utilization
of 50 electric vehicles for valley filling approach.

(b)
Fig. 13: Voltage profile for agg. 1, agg. 2, and agg. 3 with
EV load for (a) state-of-the-art centralized approach and (b)
ADMM-based proposed distributed approach. Fig. 15: Profile of the bus load with and without the utilization
of 250 electric vehicles for valley filling approach.
optimizing the charging schedule of EVs to coincide with
these low-demand valleys. By doing so, they aim to prevent Section IV-A apply to this specific case study also, except for
additional power usage peaks. The optimization problem for space limitations for charging more than 50 EVs in a single
the current scenario is defined as follows: aggregator. Once the EVs are included as additional loads, the
ADMM algorithm leverages the charging demand power to
max (Dmin + Pagg )∆t fill the low-demand valley. The area under the curve before
subject to (17) − (18) and, and after the proposed approach’s load line and average load

 0 ≤ P agg ≤ P agg,max line curves are determined and tabulated in Table IV. From
(30)


 0 ≤ P i,t ≤ P i,max Table IV, it is evident that the valley filling approach is very

 SoC i,min ≤ SoC i,t ≤ SoC i,max effective in reducing peak loads and filling up the off-peak


SoC i,req. ≤ SoC i,dep periods.

In this scenario, the calculation time step is denoted as ∆t. D. Case 4-Priority Based Charging Approach
It assumes the known fixed demand profile B ∈ Rn . By Until now, our primary focus has been on prioritizing the
utilizing the ADMM algorithm, the aim is to identify a utility requirements and ensuring the long-term health and
period with lower demand compared to other time intervals, durability of the utility. However, it is equally important to
enabling the charging schedules to fill the valley and achieve a
smoother power demand profile. This approach helps prevent
the creation of new peak points. The optimization problem TABLE IV
in (30) does not consider the frequent on-off (charging or A REA UNDER THE AVERAGE LOAD LINE CURVE AND
discharging) cost degradation factor C as the goal of the BEFORE AND AFTER THE PROPOSED APPROACH ’ S L OAD
algorithm is to find the optimal scheduling which fills the L INE FOR C ASE 3
valleys most efficiently. Before proposed After proposed
Case 3 Improvement
Fig. 14 and 15 show the outcomes of the ADMM algorithm approach approach
for 50 and 250 EVs, respectively, both with and without the 50 EVs 61.32 44.89 16.43
250 EVs 91.88 13.76 78.12
implementation of the algorithm. The EV parameters from

Authorized licensed use limited to: University of North Carolina at Charlotte. Downloaded on April 03,2024 at 04:51:53 UTC from IEEE Xplore. Restrictions apply.
© 2024 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.See https://www.ieee.org/publications/rights/index.html for more information.
This article has been accepted for publication in IEEE Transactions on Industry Applications. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/TIA.2024.3384343

10

consider the specific needs of individual drivers. In certain sit-


uations, certain drivers may require faster charging compared
to others. To tackle this issue, we have devised a charging
strategy that prioritizes the needs of individual customers. This
approach takes into account a privilege coefficient, which is
determined based on the requests made by drivers, particularly
in situations where available charging points are scarce.
In this approach, we consider three fleets consisting of 25
EVs each that require recharging within a 1440-minute time-
frame (288-time steps of 5 minutes each). It is crucial to ensure
that the total charging energy for all vehicles remains below
the maximum supply limit, denoted as Smax , as specified by
the utility company. Therefore, the energy distribution should
be planned by taking into account the urgency of charging and
the required charging duration for each vehicle.
Fig. 16: Schedule for charging based on customer’s request.
PM
n=1 P n,t ≤ S max
(31)

The constraint equation (31) ensures that the sum of the power
demands P i,t for all vehicles n at each time period t does not
exceed Smax . Here, M is the number of vehicles in a fleet.
The vehicle groups exhibit varying preferences regarding the
desired level of charge over time. The target minimum charge
profiles are defined as follows:
 δn
tar t
SoCt,n = SoCndes (32)
T +1
The vehicle groups exhibit varying preferences for the desired
charge level over time. The target minimum charge profiles,
tar
represented by SoCt,n , are determined based on the equation
provided. The parameter δn represents the urgency level for
charging, where smaller values indicate a higher sense of
Fig. 17: The correlation between charging power and electric-
urgency. The target minimum charge level for vehicle n is
ity price.
influenced by a combination of δn and SoCndes . The charging
demand si,t for EV i during period t is calculated as the
tar The optimization problem in (35) does not also consider
difference between the target minimum charge SoCi,t and
the frequent on-off (charging or discharging) cost degradation
the actual charge SoCi,t . Additionally, the average value of
factor C as the goal of the algorithm is to prioritize the
the overall power demand S is determined by calculating the
charging of specific vehicles. Case study parameters of each
mean of the squared charging demands si,t across all time
vehicle are given in Table V.
periods t and vehicles i. The charging demand si,t for EV i
during period t is calculated as:
TABLE V
tar PARAMETERS USED FOR C ASE 4

si,t = SoCi,t − SoCi,t , t = 1, . . . , T + 1 (33)

In addition, the average value of the overall power require- Fleet 1 Fleet 2 Fleet 3
Vehicle Number 25 25 25
ment S is characterized as: Battery Capacity 30 kwh 30 kwh 30 kWh
T +1 X
N Initial Charge Level 6kwh 0 9kwh
1 X Ultimate Charge Level 18kwh 30kwh 22.5kwh
S= s2 (34)
(T + 1)N t=1 n=1 i,t δ 0.2 0.4 0.9
Smax 75 kw
Total Charging Period 1440 minute
The objective of this problem is to minimize the average total
demand S while ensuring the maximum supply limit.
min pT (P agg )∆t The outcome of the charging schedules is depicted in Fig.
subject to (17) − (18) and, 16. It can be observed that fleet 2, characterized by the lowest
 SoC and δ values, initiates its charging period promptly,
C min ≤ C agg ≤ C max
followed by fleet 1 and fleet 3. Fig. 17 shows the relation

(35)


 P i,min ≤ P i,t ≤ P i,max between the total charging power and electricity price, which

 SoC min ≤ SoC i,t ≤ SoC max illustrates that at low price time steps, the charging power is

SoC i,req. ≤ SoC i,dep higher.

Authorized licensed use limited to: University of North Carolina at Charlotte. Downloaded on April 03,2024 at 04:51:53 UTC from IEEE Xplore. Restrictions apply.
© 2024 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.See https://www.ieee.org/publications/rights/index.html for more information.
This article has been accepted for publication in IEEE Transactions on Industry Applications. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/TIA.2024.3384343

11

consumption costs. While our project primarily concentrated


on four specific instances, these scenarios have the potential
for further expansion and application to diverse load models
in subsequent research.

R EFERENCES
[1] S. Rivera, S. M. Goetz, S. Kouro, P. W. Lehn, M. Pathmanathan,
P. Bauer, and R. A. Mastromauro, “Charging infrastructure and grid
integration for electromobility,” Proceedings of the IEEE, vol. 111, no. 4,
pp. 371–396, 2023.
[2] C. S. Ioakimidis, D. Thomas, P. Rycerski, and K. N. Genikomsakis,
“Peak shaving and valley filling of power consumption profile in non-
residential buildings using an electric vehicle parking lot,” Energy, vol.
148, pp. 148–158, 2018.
[3] N. M. Manousakis, P. S. Karagiannopoulos, G. J. Tsekouras, and F. D.
Kanellos, “Integration of renewable energy and electric vehicles in power
systems: A review,” Processes, vol. 11, no. 5, 2023.
[4] M. S. Mastoi, S. Zhuang, H. M. Munir, M. Haris, M. Hassan, M. Usman,
Fig. 18: A comparison for combining the four cases with m1 = S. S. H. Bukhari, and J.-S. Ro, “An in-depth analysis of electric vehicle
0.2, m2 = 0.1 and m3 = 0.5. charging station infrastructure, policy implications, and future trends,”
Energy Reports, vol. 8, pp. 11 504–11 529, 2022.
TABLE VI [5] R. Mehta, D. Srinivasan, A. Khambadkone, J. Yang, and A. Trivedi,
“Smart charging strategies for optimal integration of plug-in electric
C OMPARISON A MONG D IFFERENT A PPROACHES vehicles within existing distribution system infrastructure,” IEEE Trans-
actions on Smart Grid, vol. PP, pp. 1–1, 04 2016.
Approach 50 EVs Imp. 250 EVs Imp. [6] D. Fischer, A. Harbrecht, A. Surmann, and R. McKenna, “Electric
Miniming vehicles’ impacts on residential electric local profiles – a stochastic
345.07 54 % 3562.5 25%
charging cost ($) modelling approach considering socio-economic, behavioural and spatial
Vehicle to Min. V improve factors,” Applied Energy, vol. 233-234, pp. 644–658, 2019.
NA NA
grid (V2G) up to 5.98 % for α = 0.5 [7] O. Sadeghian, A. Oshnoei, B. Mohammadi-ivatloo, V. Vahidinasab, and
Valley filling A. Anvari-Moghaddam, “A comprehensive review on electric vehicles
(area under avg. 44.89 16.43% 13.76 78.12% smart charging: Solutions, strategies, technologies, and challenges,”
load curve) Journal of Energy Storage, vol. 54, p. 105241, 2022.
Priority based [8] O. Frendo, N. Gärtner, and H. Stuckenschmidt, “Real-time smart charg-
The lowest SoC will start charging promptly
charging ing based on precomputed schedules,” IEEE Transactions on Smart Grid,
Combined The combined approach improves the overall vol. PP, pp. 1–1, 05 2019.
approach load profile [9] S. Pelletier, O. Jabali, G. Laporte, and M. Veneroni, “Battery degradation
and behaviour for electric vehicles: Review and numerical analyses of
several models,” Transportation Research Part B: Methodological, vol.
103, pp. 158–187, 2017, green Urban Transportation.
E. Case 5: Combined Approach [10] S. Kamalasadan and A. A. Ghandakly, “A neural network parallel adap-
tive controller for fighter aircraft pitch-rate tracking,” IEEE Transactions
The four approaches described above are combined in on Instrumentation and Measurement, vol. 60, no. 1, pp. 258–267, 2010.
a single optimization problem, and their convexification is [11] M. S. Hasan, S. J. Hossain, and S. Kamalasadan, “An optimization-
described in Section III-A. The comparison of without the based volt/var control approach with increased penetration of inverter-
based resources (ibrs) in the bulk and distribution grid,” in 2023
proposed approach and with the proposed approach are given IEEE International Conference on Power Electronics, Smart Grid, and
in Fig. 18 for the optimal value of α = 0.05, m1 = 0.2, m2 = Renewable Energy (PESGRE), 2023, pp. 1–6.
0.1 and m3 = 0.5, which shows that the proposed method [12] M. Moeini-Aghtaie, P. Dehghanian, and M. Davoudi, “Energy man-
agement of plug-in hybrid electric vehicles in renewable-based energy
can find the optimal scheduling even with the combination hubs,” Sustainable Energy, Grids and Networks, vol. 32, p. 100932,
of above four cases. A comparative comparison among the 2022.
different approaches is illustrated in Table VI, which illustrates [13] P. Richardson, D. Flynn, and A. Keane, “Optimal charging of electric
vehicles in low-voltage distribution systems,” IEEE Transactions on
that the combined approach effectively manages the peak load Power Systems, vol. 27, no. 1, pp. 268–279, 2012.
points and improves the load profile. [14] S. Kiani, K. Sheshyekani, and H. Dagdougui, “An extended state space
model for aggregation of large-scale evs considering fast charging,”
IEEE Transactions on Transportation Electrification, vol. 9, no. 1, pp.
V. C ONCLUSION 1238–1251, 2023.
In this paper, we aimed to develop a smart charging schedul- [15] M. Wang, Y. Mu, Q. Shi, H. Jia, and F. Li, “Electric vehicle aggregator
modeling and control for frequency regulation considering progressive
ing algorithm based on a distributed ADMM algorithm. The state recovery,” IEEE Transactions on Smart Grid, vol. 11, no. 5, pp.
results of our study demonstrate that the proposed control 4176–4189, 2020.
processes effectively reduce the maximum peak load without [16] N. I. Nimalsiri, C. P. Mediwaththe, E. L. Ratnam, M. Shaw, D. B.
Smith, and S. K. Halgamuge, “A survey of algorithms for distributed
compromising customer comfort. These findings suggest that charging control of electric vehicles in smart grid,” IEEE Transactions
the utilization of these algorithms can promote the wider on Intelligent Transportation Systems, vol. 21, no. 11, pp. 4497–4515,
adoption of EVs, improve the power grid system, and enhance 2020.
[17] S. Boyd, N. Parikh, and E. Chu, Distributed optimization and statistical
the usage of smart devices. Furthermore, we have observed learning via the alternating direction method of multipliers. Now
that the implementation of dynamic pricing proves to be Publishers Inc, 2011.
beneficial for both the grid and customers. The outcomes of [18] B. Wang, P. Dehghanian, and D. Zhao, “Chance-constrained energy man-
agement system for power grids with high proliferation of renewables
our research reveal that these approaches provide an alterna- and electric vehicles,” IEEE Transactions on Smart Grid, vol. 11, no. 3,
tive solution for customers to minimize their overall energy pp. 2324–2336, 2020.

Authorized licensed use limited to: University of North Carolina at Charlotte. Downloaded on April 03,2024 at 04:51:53 UTC from IEEE Xplore. Restrictions apply.
© 2024 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.See https://www.ieee.org/publications/rights/index.html for more information.
This article has been accepted for publication in IEEE Transactions on Industry Applications. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/TIA.2024.3384343

12

[19] X. Zhou, S. Zou, P. Wang, and Z. Ma, “Voltage regulation in constrained Md Shamim Hasan (M’24) received a B.Sc. degree
distribution networks by coordinating electric vehicle charging based in Electrical and Electronic Engineering from the
on hierarchical admm,” IET Generation, Transmission & Distribution, Bangladesh University of Engineering and Technol-
vol. 14, no. 17, pp. 3444–3457, 2020. ogy (BUET), Dhaka, Bangladesh, in 2016. He is
[20] T. Rahman, Y. Xu, and Z. Qu, “Continuous-domain real-time distributed a Ph.D. candidate in the Department of Electrical
admm algorithm for aggregator scheduling and voltage stability in and Computer Engineering at the University of
distribution network,” IEEE Transactions on Automation Science and North Carolina, Charlotte, NC, USA. He previously
Engineering, vol. 19, no. 1, 2022. worked as an Engineer at the Power Grid Company
[21] S. Patel, M. Ahmed, and S. Kamalasadan, “A novel energy storage- of Bangladesh (PGCB), Bangladesh. His research in-
based net-load smoothing and shifting architecture for high amount of terests include power system optimization, microgrid
photovoltaics integrated power distribution system,” IEEE Transactions modeling, power system dynamics, and control.
on Industry Applications, vol. 56, no. 3, pp. 3090–3099, 2020.
[22] S. Kiani, K. Sheshyekani, and H. Dagdougui, “Admm-based hierarchical
single-loop framework for ev charging scheduling considering power
flow constraints,” IEEE Transactions on Transportation Electrification,
pp. 1–10, 2023.
[23] C. P. Guzmán, J. C. Lápez, G. Sanchez, L. Z. Terada, M. J. Rider, Aniket Joshi (M’21) received the bachelor’s degree
and L. C. P. da Silva, “An admm-based distributed energy management in electrical engineering from the University of
system for microgrids,” in 2022 IEEE PES Innovative Smart Grid Pune, Pune, India, in 2013, and the M.S. and Ph.D.
Technologies Conference Europe (ISGT-Europe), 2022, pp. 1–5. degrees in electrical engineering from the University
[24] F. O. No, “Ferc order no. 2222: A new day for distributed energy of North Carolina, Charlotte, NC, USA, in 2016 and
resources,” Washington, DC: First street, pp. 2020–09, 2020. 2021, respectively. He is currently a Research Scien-
[25] X. Zhou, S. Zou, P. Wang, and Z. Ma, “Admm-based coordination of tist with the US Research Center, ABB Inc., Raleigh,
electric vehicles in constrained distribution networks considering fast NC. His research interests include microgrids, smart
charging and degradation,” IEEE Transactions on Intelligent Transporta- inverters, hybrid energy storage systems, and optimal
tion Systems, vol. PP, pp. 1–14, 08 2020. control.
[26] A. I. Aygun and S. Kamalasadan, “Centralized charging approach to
manage electric vehicle fleets for balanced grid,” in 2022 IEEE Inter-
national Conference on Power Electronics, Smart Grid, and Renewable
Energy (PESGRE). IEEE, 2022, pp. 1–6.
[27] S. Boyd, L. Xiao, A. Mutapic, J. Dattorro, and J. Mattingley, “Subgra-
dient methods, decomposition methods, alternating projections,” Notes S. Kamalasadan (SM’17) received the B.Tech. de-
for EE364b, Stanford University, 2007. gree in electrical and electronics engineering from
[28] M. S. Hasan, M. M.-U.-T. Chowdhury, B. D. Biswas, and S. Ka- the University of Calicut, India, in 1991, the M.Eng.
malasadan, “Semi-definite programming based scalable and accurate degree in electrical power systems management from
optimal power flow models for radial distribution networks,” IEEE the Asian Institute of Technology, Bangkok, Thai-
Transactions on Industry Applications, pp. 1–13, 2023. land, in 1999, and the Ph.D. degree in electrical
[29] B. Kocuk, S. S. Dey, and X. A. Sun, “Strong socp relaxations for the engineering from The University of Toledo, Toledo,
optimal power flow problem,” Operations Research, vol. 64, no. 6, pp. OH, USA, in 2004. He is a Professor in the Depart-
1177–1196, 2016. ment of Electrical and Computer Engineering at the
[30] M. S. Hasan, M. M.-U.-T. Chowdhury, and S. Kamalasadan, “Sequential University of North Carolina at Charlotte, Charlotte,
quadratic programming (sqp) based optimal power flow methodologies NC, USA.
for electric distribution system with high penetration of ders,” IEEE His research interests include intelligent and autonomous control, power
Transactions on Industry Applications, pp. 1–11, 2024. systems dynamics, stability and control, smart grid, microgrid, and real-time
[31] I. CVX Research, “CVX: Matlab software for disciplined convex optimization and control of power systems. He is a recipient of the National
programming, version 2.0,” http://cvxr.com/cvx, Aug. 2012. Science Foundation (NSF) CAREER Award and several IEEE Best Paper
[32] A. Alsabbagh, H. Yin, and C. Ma, “Distributed electric vehicles charging Awards. He has published more than 250 research articles and secured more
management with social contribution concept,” IEEE Transactions on than $ 12M in grants and contracts. He is the chair of the IEEE Power and
Industrial Informatics, vol. 16, no. 5, pp. 3483–3492, 2020. Energy Society (PES) Power and Energy Education Committee, a member of
[33] Z. Wang and S. Wang, “Grid power peak shaving and valley filling using the 2022 IEEE PES Leadership Team, and a Council member of the IEEE
vehicle-to-grid systems,” IEEE Transactions on power delivery, vol. 28, Industry Application Society (IAS)
no. 3, pp. 1822–1829, 2013.
[34] E. Sortomme and M. A. El-Sharkawi, “Optimal charging strategies for
unidirectional vehicle-to-grid,” IEEE Transactions on Smart Grid, vol. 2,
no. 1, pp. 131–138, March 2011.
[35] J. Tomić and W. Kempton, “Using fleets of electric-drive vehicles for
grid support,” Journal of Power Sources, vol. 168, no. 2, pp. 459 – 468,
2007.
[36] W. Kempton, V. Udo, K. Huber, K. Komara, S. Letendre, S. Baker,
D. Brunner, and N. Pearre, “A test of vehicle-to-grid (v2g) for energy
storage and frequency regulation in the pjm system,” 2008.

Ali Ihsan Aygun (M’23) received his Bachelor’s


degree in Electrical and Electronic Engineering from
Yıldız Technical University, Turkey, in 2014, the
M.S. degree in Electrical Engineering from George
Washington University in 2018 and Ph.D. degrees in
Electrical Engineering from the University of North
Carolina, Charlotte, NC, USA, in 2022. He is cur-
rently a lecturer at the Department of Electrical and
Electronics Engineering, Duzce University, Duzce,
Turkey. His research interests include smart grids,
electric vehicle optimization, and control.

Authorized licensed use limited to: University of North Carolina at Charlotte. Downloaded on April 03,2024 at 04:51:53 UTC from IEEE Xplore. Restrictions apply.
© 2024 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.See https://www.ieee.org/publications/rights/index.html for more information.

You might also like