Electric Power Systems Research 238 (2025) 111062

Contents lists available at ScienceDirect

Electric Power Systems Research

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

Energy management system for a workplace PV-EV charging station with

active and reactive power dispatch of EVs considering transformer aging
and uncertainties
Rayaprolu.M. Raghuveer * , Bhavesh Bhalja , Pramod Agarwal
Department of Electrical Engineering, Indian Institute of Technology Roorkee, Uttarakhand, India

A R T I C L E I N F O A B S T R A C T

Keywords: This study presents a new energy management system (EMS) for a grid-tied photovoltaic (PV) – electric vehicle
Energy management system (EMS) (EV) integrated workplace charging station. The proposed EMS is developed as a convex stochastic mixed-integer
Electric vehicles (EVs) quadratically constrained problem (MIQCP) to minimize the expected apparent power demand while limiting the
Loss-of-life (LoL)
distribution transformer’s accelerated aging and satisfying the EV driving needs. This is accomplished by
Photovoltaic generation (PV)
Vehicle-to-grid (V2G)
scheduling real and reactive powers of EVs in real-time through Vehicle-to-grid (V2G) mode. The inherent
diurnal uncertainties and seasonal variations associated with the workplace non-EV load, PV generation are
incorporated using probabilistic and hierarchical clustering techniques, respectively. The implementation of the
developed EMS under receding horizon model enables real-time operation by adapting to dynamic arrivals of
EVs. The effectiveness of the proposed EMS is validated through numerous simulations from the view point of the
distribution system operator (DSO), charging station owner (CSO), and EV prosumer. The results indicate a
substantial reduction in the peak demand, minimized transformer’s loss-of-life (LoL), and operating cost saving
while satisfying the EV driving needs in comparison to uncoordinated charging method.

suggested in [4] and [5] to reduce total charging cost, there is no

1. Introduction consideration of renewable energy sources, which renders EV usage
unsustainable, being powered by conventional sources. Then, EVs are
Due to continued depletion of oil resources and an increase in integrated into PV coupled microgrids, considering the uncertainties
greenhouse gas emissions worldwide, the utilization of Electric Vehicles associated with both EVs and PV to reduce the overall cost [6]. Never-
(EVs) has witnessed a significant upsurge. According to the International theless, this does not account for transformer aging and seasonal vari-
Energy Agency, the global EV fleet is projected to reach 145 million by ations, leading to unreliable operation by the grid operators as well as
2030 [1]. EVs could be charged at various charging infrastructures, EV aggregators. Thereafter, charge controller based on fuzzy logic
including residential, workplace, public, and commercial places. Among technique is employed for EV charging in [7] and [8]. However, trans-
all, workplace EV charging is more vital due to better flexibility in terms former aging is considered in [7], while the same is not incorporated in
of magnitude and direction (G2V & V2G) because of longer parking [8]. Moreover, both of these aforementioned works may not assure a
times. Moreover, the EV charging demand is in line with the rooftop PV globally optimal solution. Then, a stochastic control strategy using a
generation during the day at the workplace [2]. Conversely, uncon- two-layered EMS is implemented in a microgrid integrated with wind
trolled EV charging poses significant challenges to the grid in terms of generation and EVs to decrease the grid dependency [9]. Nonetheless,
increased peak demand, higher power losses and deteriorated power transformer aging has not been taken into account in the EMS, thereby
quality which in turn may cause overloading of the grid’s assets, spe- deteriorating the transformer’s life and leading to premature
cifically transformers and cables [3]. Hence, to alleviate these adverse replacement.
impacts, workplaces can employ an energy management system (EMS) EMS utilizing mixed-integer linear programming (MILP) formulation
for smart charging of EVs from PV with V2G to achieve technical, is presented for workplace EV charging to reduce charging cost in [10]
financial, and environmental benefits. and [11]. However, challenges pertaining to the transformer aging and
Though optimal charge and discharge scheduling for EVs is reactive power dispatch of EVs in V2G mode have not been included in

* Corresponding author.
E-mail address: [email protected] (Rayaprolu.M. Raghuveer).

https://doi.org/10.1016/j.epsr.2024.111062
Received 16 May 2024; Received in revised form 15 August 2024; Accepted 7 September 2024
Available online 15 October 2024
0378-7796/© 2024 Elsevier B.V. All rights are reserved, including those for text and data mining, AI training, and similar technologies.
Rayaprolu.M. Raghuveer et al. Electric Power Systems Research 238 (2025) 111062

Nomenclature REV
m Apparent rating of mth EV charger.
AMB AMB
θt /θτ,s Ambient temperature at current time slot t/ at time slot τ
Indices and Sets and scenario s.
d/D Index for seasonal day/Set of seasonal days.
ΔθTA Transformer’s top-oil temperature rise over ambient
m/M Index for EV/Set of total EVs.
temperature at rated load.
n/N Index for current EV/Set of current EVs.
ΔθHT Transformer’s hot-spot temperature rise over top-oil
s/S Index for scenario/Set of all scenarios.
temperature at rated load.
t/T Index for current time slot/Set of current time slots.
L Transformer’s rated load loss to no load loss ratio.
τ/H Index for time slot in current prediction horizon/Set of
time slots in current prediction horizon. RatedTrafo Transformer’s rated capacity in kVA.
ILN Transformer’s insulation life corresponding to normal
Parameters aging in hours.
PnonEV
t /QnonEV
t Active/reactive non-EV power demand at current Δt Duration of one-time slot.
time slot t. πs Probability of scenario s.
PnonEV /QnonEV Active/reactive non-EV power demand at time slot τ
τ,s τ,s
Variables
and scenario s. Trafo Trafo
Pt /Qt Active/reactive power demand of transformer at
PPV PV
t /Pτ,s PV generation at current time slot t/ at time slot τ and
current time slot t.
scenario s. Trafo Trafo
PVfc nonEVfc Pτ,s /Qτ,s Active/reactive power demand of transformer at time
Pt /Pt Forecasted PV generation/non-EV power demand at
slot τ and scenario s.
current time slot t.
PEVcha
n,t /PEVdis
n,t Active charging/discharging power of nth EV at current
EEVinitial /EEVdep
m Initial energy at arrival/final energy at departure
m time slot t.
time of mth EV battery.
PEVcha EVdis
n,τ,s /Pn,τ,s Active charging/discharging power of nth EV at time
arr dep
tm /tm Arrival /departure time of mth EV. slot τ and scenario s.
EVmax EVmin
Em Maximum/minimum instantaneous energy of mth EV QEV EV
n,t /Qn,τ,s Reactive power at time slot t/ at time slot τ and scenario s
/Em
battery.
of nth EV.
QEVmax /QEVmin Maximum/minimum reactive power of mth EV
m m γ EV
n,τ,s Binary variable corresponding to charging/discharging
charger.
of nth EV at time slot τ and scenario s.
ηEVcha
m /ηEVdis
m Charging/discharging efficiency of mth EV charger.
EEV EV
n,t /En,τ,s Instantaneous energy at time slot t/ at time slot τ and
PEVmax
m Maximum active charging/discharging power of mth EV
charger. scenario s of nth EV.
BCEVm Battery capacity of mth EV.

the above said works. Furthermore, stochastic nature of PV, EVs, load is these studies neglect other crucial factors such as uncertainties in
also overlooked in [10]. Thereafter, the linear programming (LP) Renewable Energy Sources (RES) and real-time EV dynamics. Later, in
approach is suggested for EV charging station integrated with solar and [15–17] voltage regulation is achieved by coordinating active and
battery energy storage to minimize both operational expenditure and reactive power dispatch of EVs. To be more precise, robust optimization
the transformer aging [12]. Although the variability of the system inputs technique is applied in [15], while the hierarchical coordination
would be mitigated by battery storage, it may incur additional cost to framework [16] and the parallel consensus algorithm [17] are utilized in
the system as V2G can be utilized to harness the energy stored in EV a decentralized manner. Nevertheless, these approaches do not account
batteries. for transformer aging and EV dynamics.
Based on the fact that EVs could generate/consume reactive power at An overview of various existing EMS algorithms is summarized in
any level of state of charge (SOC) without degrading the battery’s life, Table 1. It is evident that majority of the works [4–11] except [7], focus
many studies [13–16] and [17] address various aspects of EV coordi- on active power dispatch only through V2G technology. Then, some
nation through joint active and reactive power scheduling of EVs. More studies [13–17] incorporate both active and reactive power manage-
specifically, [13] and [14] have considered transformer aging. However, ment of EVs, but [13,14], and [15–17] overlook uncertainties and

Table 1
Taxonomy of existing EMS algorithms.
Reference V2G Transformer’s aging constraints Uncertainties Seasonal variations Real-time EV dynamics

P Q RES Non-EV load

[4,5] ✓ ⨯ ⨯ ⨯ ⨯ ⨯ ✓
[6] ✓ ⨯ ⨯ ✓ ⨯ ⨯ ⨯
[7] ⨯ ⨯ ✓ ⨯ ⨯ ⨯ ✓
[8,9] ✓ ⨯ ⨯ ✓ ⨯ ⨯ ✓
[10] ✓ ⨯ ⨯ ⨯ ⨯ ⨯ ⨯
[11] ✓ ⨯ ⨯ ✓ ⨯ ✓ ✓
[12] ⨯ ⨯ ✓ ⨯ ⨯ ✓ ⨯
[13,14] ✓ ✓ ✓ ⨯ ⨯ ⨯ ⨯
[15] ✓ ✓ ⨯ ⨯ ✓ ⨯ ⨯
[16,17] ✓ ✓ ⨯ ⨯ ⨯ ⨯ ⨯
This Work ✓ ✓ ✓ ✓ ✓ ✓ ✓

✓ = considered, ⨯ = not considered, P = Active power, Q = Reactive power, RES = Renewable Energy Sources.

2
Rayaprolu.M. Raghuveer et al. Electric Power Systems Research 238 (2025) 111062

transformer aging constraints, respectively. Furthermore, deterministic uncertainties related to forecasts of the PV generation, workplace non-
techniques excluding [5,9], rely heavily on the day-ahead predicted EV load, and EV parameters, which include arrival time, departure
data, which might result in inaccurate scheduling of EVs, rendering time, and initial energy. Notably, this study does not require forecasting
them unreliable. While, studies apart from [11,12] incorporating un- of EV parameters due to the implementation of the EMS utilizing
certainties of RES, only account for diurnal variations, neglecting sea- receding horizon framework as explained in the forthcoming Section
sonal fluctuations. Few works [4,5,7–9] and [11] consider EV dynamics 3.2. Furthermore, seasonal variations of the said parameters are
in real-time leading to realistic optimal solutions. However, most of the captured from three representative seasonal days spanning a year
works [12–17] overlook this aspect, resulting in sub-optimal solutions. through the hierarchical clustering method employing Ward’s linkage
To the best of authors’ knowledge, none of the existing works compre- [18]. These three seasonal days corresponds to (i) spring (ii) summer
hensively considers all the mentioned applications, comprising joint and (iii) autumn, which exhibit unique net load profiles (difference
active and reactive power dispatch of EVs, transformer aging, diurnal as between workplace non-EV load and PV generation). Consequently, the
well as seasonal uncertainties of RES and real-time EV dynamics within a simulations conducted considering both daily forecast uncertainties and
single EMS formulation. Therefore, this study aims to overcome these seasonal variations make the system pragmatic. It is worth mentioning
limitations by providing the comprehensive EMS framework with the that day-ahead forecasting is beyond the scope of study. The detailed
following noteworthy contributions: forecast uncertainty modeling is presented below.

• In contrast to the prevailing approaches, the proposed EMS optimizes 2.1. PV generation forecast uncertainty
the power flows between the grid, EVs, and PV system by scheduling
both active and reactive power dispatch of EVs in real-time that Beta probability density function (PDF) effectively models the un-
( )
simultaneously benefits all the three entities including distribution PVfc
certainty due to PV generation forecast Pt for a given seasonal day,
system operator (DSO), charging station owner (CSO) and EV
which is expressed by (1) [19].
prosumer.
• Furthermore, linearized transformer’s aging constraints are incor- Γ(αt + βt )
fPPVfc (y) = (y)αt − 1 .(1 − y)βt − 1 . (1)
porated into the proposed EMS formulation to avoid overloading and t Γ(αt )Γ(βt )
premature aging of the transformer caused by additional EV charging
demand as the existing transformer is not designed to cater the extra In (1), ‘y’ represents the occurrence of PPVfc
t , which takes the values in
EV demand, which is lacking in the previous works. the range of [0 to 1]. These values are obtained by normalizing PPVfc
t with
• A stochastic optimization framework is utilized in the proposed EMS respect to its maximum value PPVmax and the normalized value is denoted
to incorporate multiple uncertainties associated with forecasts of PV as PPVfcN
t . For a given forecasting value PPVfc , the statistical parameters
generation and non-EV load by the respective probability distribu- ( t )
PV(fc)
such as mean and standard deviation μt , σPV(fc)
t are calculated using
tions. Moreover, seasonal variations throughout the year have also
been captured by employing hierarchical clustering, which is rarely (2)-(3). Subsequently, the shape and scale parameters (αt , βt ) of beta
found in prior techniques. PDF are computed according to (4)-(5).
• Unlike the existing methods, the suggested EMS can handle inherent
PPVfc
t
uncertainties associated with EVs in real-time, thus eliminating the μPVfc = PPVfcN = (2)
t t
PPVmax
need for forecasting EV information that includes arrival time, de-
parture time, and initial energy. This is because of the implementa- σ PVfc = 0.2 μPVfc + 0.21 (3)
t t
tion of the formulated EMS within the receding horizon framework.
( )
PVfc
1− μ t 1 2
2. Modeling of system uncertainties αt = − μPVfc
t (4)
σ PVfc2
t μPVfc
t

In this study, an EMS is developed for a workplace with a V2G- ( )

enabled EV charging station tied to the grid through a distribution 1
β t = αt PVfc
− 1 (5)
transformer. As depicted in Fig. 1, this workplace is seamlessly inte- μt
grated with a PV source. The system, shown in Fig. 1, has multiple
2.2. Uncertainty related to workplace non-EV load forecast

Workplace non-EV load refers to the base load excluding EV charging

demand. The Gaussian PDF is used to model the non-EV load forecast
( )
nonEVfc
Pt uncertainty for a given seasonal day as described in (6) [20].
nonEVfc
In (6), ʹxʹ is the occurrence of Pt within the range of [0 to 1]. This is
nonEVfc
achieved by normalizing Pt with respect to its maximum value
nonEVfcN
PnonEVmax and it is denoted by Pt . For a given forecasting value
nonEVfc
P , the statistical parameters like mean and standard deviation
(t )
nonEVfc nonEVfc
μt , σt are obtained using (7)-(8).

( ( )2 )
1 x− μnonEVfc
t
fPnonEVfc (x) = √̅̅̅̅̅̅.exp − 2
(6)
t
σnonEVfc
t . 2π σnonEVfc
t

PnonEVfc
t
μnonEVfc = PnonEVfcN = (7)
t t
PnonEVmax

σ PnonEVfc
t = 0.1 μnonEVfc
t (8)
Fig. 1. Demonstration of the system under study.

3
Rayaprolu.M. Raghuveer et al. Electric Power Systems Research 238 (2025) 111062

( )
3. Framework of the proposed EMS algorithm 15000 15000
,s = exp
FτAA − HST , ∀τ, s (20)
110 + 273 θτ,s + 273
This section presents a comprehensive mathematical modeling of the
proposed stochastic mixed-integer quadratically constrained program-
FτAA
,s ≤ 2.7, ∀τ, s (21)
ming (MIQCP) problem. It then delves into the operation of the devel-
oped EMS algorithm within the receding horizon model. Additionally, it
θHST AMB TO HST
τ,s = θτ,s + Δθτ,s + Δθτ,s , ∀τ, s (22)
discusses the practical implementation of the developed approach,
highlighting its potential applications and integration into real-world
(( Trafo )2 )n
scenarios. Kτ,s L+1
ΔθTO = Δθ TA
, ∀τ, s (23)
τ,s
L+1
3.1. Mathematical formulation of the proposed EMS
(( )2 )m
The primary objective of the proposed EMS is to flatten the daily load ΔθHST
τ,s = Δθ
HT
KTrafo
τ,s , ∀τ , s (24)
profile, including the additional EV demand, while considering the
afore-mentioned uncertainties. Hence, a stochastic objective function
( )2 1 (( )2 ( )2 )
KTrafo = PTrafo + QTrafo , ∀τ , s (25)
(f), as per (9), is formulated that minimizes apparent demand of the τ,s
(RatingTrafo )
2 τ,s τ,s

transformer over the entire current prediction horizon ∀τ∈[t, h] which is

Equality constraints, as given by (10) and (11), represent the active
fragmented into two parts. The first part corresponds to the apparent
and reactive power balances, respectively. It is worth mentioning that a
demand of the current time slot t (real-time). Conversely, the second part
is the expected apparent power for the following time slots in the current binary variable γEV
n,τ,s restricts the simultaneous charge and discharge of
( )
prediction horizon τ∈[t + 1, h] incorporating all the scenarios ‘s’ with active power scheduling PEVcha n,τ,s /PEVdis
n,τ,s of nth EV during τth time slot
their corresponding probability of occurrences (πs ). The cost function ‘f’ during scenario ‘s’. The value of γEV
n,τ,s = 1 is considered for charging
can be minimized by scheduling all EVs’ active and reactive powers in
whereas γ EV = 0 is meant for discharging. Similarly, reactive power
real-time to mitigate the variations of real-time and the expected net (n,τ,s )
real/reactive demands. For instance, if the expected net active/reactive schedule QEV n,τ,s is positive/negative for absorbing/injecting reactive
demand is high, the aggregate active/reactive power of EV shall be low power. The Eqs. (12)–(15) are related to the EV battery energy storage
and vice-versa. Consequently, utilization of the proposed EMS not only constraints in which (12) determines the instantaneous energy of the EV
improves life of the transformer but also reduces peak demand along ( )
battery EEVn,τ,s , ∀ n ∈ N, τ ∈ H, s ∈ S. Moreover, all the EVs reach the
with operating cost. The mathematical formulation of the developed
stochastic real-time mixed-integer quadratically constrained problem charging station with initial energy EEVinitial
n as shown in (13). Later, EVs
(MIQCP) based EMS is given below. shall be charged to the desired final energy EEVdep
n to satisfy the EV user’s
(( S
(
h ((
) driving needs before departure, as represented by (14). Also, EEV n,τ,s is
)2 ( Trafo )2 ) ∑ ∑ )2 ( )2 )
minf = PTrafo
t + Q t + π s P Trafo
τ,s + QTrafo
τ,s
constrained between the minimum and maximum limits to protect the
s=1 τ=t+1 battery life of EV, as shown in (15). The active and reactive powers of
( )
(9) each EV shall not exceed the apparent rating of charger REV , as given
(n
The above objective function as per (9) is subjected to various con- by (16). Further, active and reactive power schedules Pn,τ,s /PEVdisEVcha
n,τ,s ,
straints as depicted by (10)-(25). )
( ) ∑( ) Qn,τ,s must be within operational limits, as shown in (17)-(19). Further,
EV

PTrafo
τ,s = PnonEV
τ,s − PPV
τ,s + PEVcha EVdis
n,τ,s − Pn,τ,s , ∀τ, s (10) it is assumed that all EV chargers are bidirectional which can operate in
n∈N
all the four quadrants of the PQ plane at a non-unity power factor. The
∑ minimum and maximum operating limits of reactive powers
QTrafo = QnonEV + QEV
n,τ,s , ∀τ, s (11) ( EVmin EVmax
τ,s τ,s
n∈N
Qn /Qn ) of all EVs are computed for a given REV n and PEVmax
n ,
respectively [17].
( )
PEVdis
n,τ,s
Finally, Eqs. (20)–(25) are related to transformer’s aging constraints,
EEV
n,τ,s = EEV
n,τ− 1,s + η EVcha EVcha
n .Pn,τ,s − .Δt, ∀n, τ, s (12) which are determined in accordance with IEEE C57.91-2011 standard
ηEVdis
n
[21]. The aging of a transformer can be estimated by its insulation
degradation, which depends on the maximum temperature rise i.e., the
EEV
n,tnarr ,s = En
EVinitial
, ∀n, s (13) ( )
hot-spot temperature θHST τ,s . It is obvious from (20) that the accelerated
( )
EEV ≥ EEVdep , ∀n, s (14) aging factor FτAA varies exponentially with respect to θHST
n τ,s , and is
dep
n,t ,s
n ,s

linearized through piecewise linear approximation technique [13]. The

EEVmin ≤ EEV
n,τ,s ≤ En
EVmax
, ∀n, τ, s (15) reference θHST
τ,s is considered to be 110 C in this study, that gives rise to
n ◦

( )2 ( )2 ( )2 ⎫ FτAA AA
,s = 1, indicating normal aging. It shall be noted that, if Fτ,s > 1, leads
PEVcha + QEV ≤ REV to accelerated aging causing loss-of-life. Conversely, if FτAA
,s < 1, results
n,τ,s n,τ,s n
⎬
( )2 ( )2 ( , ∀n, τ, s (16)
PEVdis + QEV ≤ REV
) 2 ⎭ in life extension. However, owing to practical operating conditions, as
n,τ,s n,τ,s n
the transformer is subjected to long durations of light loading and lesser
ambient temperatures, θHSTτ,s shall not exceed 120 C for normal aging
◦
0 ≤ PEVcha EV EVmax
n,τ,s ≤ γ n,τ,s .Pn , ∀n, τ, s (17)
[14]. Consequently, this results in FτAA
,s = 2.7 and hence the constraint
( ) (21) is imposed to preserve the transformer’s life. Moreover, loading
0 ≤ PEVdis
n,τ,s ≤ 1 − γ EV EVmax
n,τ,s .Pn , ∀n, τ, s (18) (
Trafo
)
factor Kτ,s , calculated according to (25) is a quadratic constraint,
QEVmin ≤ QEV and further linearized with a set of line segments representing the circle
n,τ,s ≤ Qn , ∀n, τ, s (19)
EVmax
n
[22].

4
Rayaprolu.M. Raghuveer et al. Electric Power Systems Research 238 (2025) 111062

3.2. Proposed EMS under receding horizon framework

( )
In this study, EVs’ active and reactive powers PEVcha
n,t /PEVdis EV
n,t , Qn,t are

scheduled in real-time through the implementation of the proposed EMS

under receding horizon framework, which can seamlessly incorporate
future system uncertainties while utilizing the updated input data. This
method is specifically employed to handle the uncertainties of EVs’ in
real-time, thereby eliminating the need for forecasting EV information
( )
arr dep
tm , tm , EEVinitial
m which eventually leads to reliable and accurate EV
scheduling. The proposed real-time stochastic EMS under the receding Fig. 2. Illustration for determining current EV and current prediction hori-
horizon framework is summarized in Algorithm-1. zon sets.
The first step is to import the day-ahead forecast data of workplace
non-EV load and PV generation for all seasonal days. Then, for a given (
seasonal day (d), the real-time information at the beginning of the cur- at the current time slot ‘t’, the real-time data PnonEV t , QnonEV
t , PPV arr
t , tm ,
)
rent time slot ‘t’ is updated. Afterwards, both the dynamic sets i.e. cur- dep
tm and EEVinitial
m is applied. For the following time slots τ ∈ [t + 1,h], the
rent EV {N} and the current prediction horizon {H} need to be ( )
determined. This is the crucial step of the proposed algorithm, which is forecasted data, incorporating uncertain scenarios PnonEV τ,s , QnonEV
τ,s , PPV
τ,s ,
explained here. The arrival and departure times of EVs are denoted by are utilized. Then, the optimization problem is solved, which results in
arr
tm dep
and tm , respectively. If mth EV satisfies the condition. t ≤ tm
arr
,t ≤ optimal EV active and reactive power schedules over the entire current
dep
tm and EEV EVdep
m,t < Em then, mth EV is included in the current EV set N prediction horizon [t, h]. However, the optimal schedule of the current
( )
with index ‘n’. Thereafter, at the commencement of each time slot, the time slot PEVcha
n,t /PEVdis EV
n,t , Qn,t is only executed and the remaining
current prediction horizon is determined, which ranges from the current ( )
EVcha EVdis EV
schedules Pn,τ,s /Pn,τ,s , Qn,τ,s ∀τ ∈ [t +1, h] are discarded, which are
time slot ‘t’ to the most recent departure EV time slot ‘h’, where h =
{ } revised in the upcoming time slots by repeatedly solving the same
max tndep . Therefore, the current prediction horizon set is denoted by H problem again after the realization of uncertainties. Finally, real-time
= {t, t +1, t +2, …h} and indexed by ‘τ’. The above process is demon- active and reactive power schedules of all EVs are ascertained for all
strated in Fig. 2, which shows the computation of dynamic sets N and H the time slots t ∈ T over a period of 24 h. This process is repeated for all
at the beginning of the current time slot t = 9. Four EVs having different three representative seasonal days d ∈ D.
( )
arr dep
arrival times (tm ), departure times, tm and instantaneous energies
( ) 3.3. Practical Implementation of the proposed EMS
EV
Em,t of 100 %, 60 %, 40 %, 20 %, respectively, have been considered.
As noticed from Fig. 2, EV1 is fully charged whereas 2nd, 3rd and 4th In this study, all EVs are assumed to participate in V2G due to their
EVs satisfy all the conditions, as mentioned above. Therefore, the cur- longer parking times at workplace [2], thereby allowing EMS to leverage
rent EV set N = {2, 3, 4}, end most departure EV slot h = 18 and the the flexibility of charging/discharging EVs. EV users update the
current prediction horizon set H = {9, 10, 11, 12, …, 18}. real-time EV information upon arrival at the charging station through
Now, the optimization problem, formulated in Section 3.1 with the the user-friendly interface (web/mobile app), including arrival time
( )
objective function as depicted in (9), is subjected to the constraints arr
(tm ), initial energy at arrival EEVinitial
m , EV battery capacity (BCEV
m ),
mentioned in (10)-(25). However, the constraints that pertain to the dep
departure time (tm ) and desired final energy at departure time (EEVdep
m ).
current time slot are indexed by ‘t’ without an index ‘s’. In contrast, the
Moreover, smart chargers also provide EV charger information such as
constraints associated with subsequent time slots within the current [ EV EVmax EVmax EVmin ]
Rm , Pm , Qm , Qm in real-time to the EMS through Open
prediction horizon are indexed with both ‘τ’ and ‘s’. It is to be noted that,
Charge Point Protocol (OCPP), an open source communication protocol

Algorithm 1
Proposed EMS under receding horizon model.
Input: Import the day-ahead forecast data of the workplace non-EV load and PV generation for all seasonal days.

1. for d = 1 to D do
2. for t = 1 to T do
dep
3. Input the real- time data at the beginning of the current time slot ‘t’ PnonEV
t , QnonEV
t , PPV arr
t , tm , tm , Em
EVinitial
.
4. Determine the dynamic sets N and H, as explained in Section 3.2.
5. Generate scenarios by considering forecast uncertainties ∀τ ∈ [t +1, h] PPVτ,s , Pτ,s
nonEV
, QnonEV
τ,s using Beta and normal distributions according to (1)-(8).
6. for τ = t to h do
7. Solve MIQCP for minimizing ‘f’ using (9) subject to (10)-(25) and obtain PEVcha EVdis EV
n,τ,s , Pn,τ,s , Qn,τ,s ∀τ ∈ [t, h].
8. Execute the control variables PEVcha
n,t , PEVdis EV
n,t , Qn,t for the current time slot t.
9. Update state variable EEV
n,t using (12).
10. end for
11. t = t + 1
12. end for
13. d = d + 1
14. end for
Output: Real-time scheduling of EVs’ active and reactive powers PEVcha
n,t , PEVdis EV
n,t , Qn,t ∀ t ∈ T, d ∈ D.

5
Rayaprolu.M. Raghuveer et al. Electric Power Systems Research 238 (2025) 111062

[23]. The real-time data received from both EV users and smart chargers Table 3
are integrated with the EMS backend data repository. Now utilizing this Statistical distributions of EV parameters.
data, real-time EV schedules are determined by running the proposed Parameter Mean Standard deviation
EMS algorithm. Subsequently, the obtained EV charge/discharge
EEVinitial 23 Kwh 2 Kwh
schedules are sent to the corresponding smart chargers via OCPP. Then, tm
m
arr
9 a.m. 1h
EV chargers controls the charging/discharging current of respective EVs dep
tm 5 p.m. 1h
through ISO 15118 communication protocol [24]. Thus, the proposed
EMS can be implemented practically, which enhances both accuracy of
20 scenarios are reported in Table 2. The real-time PV generation
scheduling of EVs and user comfort while mitigating uncertainties. ( nonEV )
(PPV
t ) and non-EV demand Pt are obtained from PV Watts software
[30] and load data recorded from smart meters, respectively.
4. Case study and results
In this work, EV parameters such as initial energy (EEVinitial
m ), arrival
( arr ) ( )
dep
The formulated convex MIQCP optimization problem using Eqs. (9)– time tm and departure time tm are dynamically updated in real-
(25) is modeled in the Generic Algebraic Modeling System (GAMS) time upon reaching the workplace charging station and randomly
software and solved using the "CPLEX" solver to ensure a globally generated from the normal PDFs with the mean and standard deviation
optimal solution [25]. The case study data used to validate the proposed as reported in Table 3. The statistical parameters of these PDFs are
EMS is mentioned below. All simulations of the proposed method, along estimated from the historical driving patterns [31]. For simplicity, all
with conventional EV charging methods, are conducted on a desktop 100 EVs are assumed to be homogenous i.e. the Nissan Leaf EV model
computer equipped with a 10-core Intel i7 processor having 16 GB RAM, ( ) ( )
with a battery capacity BCEV m of 40 kWh [32]. The minimum EEVmin m
and a clock speed of 3.2 GHz. The results obtained are presented and and maximum (EEVmax ) limits of the instantaneous energy of EV battery
m
thoroughly discussed in the upcoming numerical results subsection.
are taken as 8 kWh and 40 kWh, respectively. The EVs’ desired final
( ) ( )
energy before departure EEVdepm is set to 32 kWh 0.8.BCEV m . The
( EVcha EVdis )
4.1. Simulation setup charging/discharging efficiency ηm /ηm is considered as 95 %.
( )
The apparent rating of the EV charger REV m is considered as 15 kVA
The case study is applied to a workplace integrated with PV and EV ( )
with its maximum active power PEVmax m being 15 kW while the mini-
charging station with 100 charging piles supporting V2G technology. (
mum and maximum limits of reactive power QEVmin /QEVmax ) are
The maximum PV on-site generation and non-EV load capacities are 600 m m

kW and 1500 kW, respectively. Here, the power factor of the non-EV ±13 kVAr.
demand is taken as 0.9 [26]. The normalized forecasts of PV genera-
PVfcN nonEVfcN
tion (Pt ) and non-EV demand (Pt ) for three representative 4.2. Numerical results
seasonal days are shown in Fig. 3(a) and Fig. 3(b), respectively. It can be
seen in Fig. 3 that forecasted non-EV load and PV generation exhibit Simulations are performed for two conventional methods of EV
both the diurnal and seasonal variations, which are obtained from charging along with the proposed EMS. The first method, known as
Electrical Engineering Department, IIT Roorkee campus, India [27]. To uncoordinated charging/Method-1, where EVs charge without any delay
model the forecast uncertainties of PV generation and non-EV demand, after arriving at the workplace maintaining a consistent charging rate
initially 10,000 scenarios are generated using (1) to (8) through Monte until they reach desired final energy before departing, serves as the
Carlo sampling [28]. However, this leads to a huge computational baseline method. The second method/Method-2 is the smart charging
burden. Therefore, K-means clustering approach [29] is employed for method considering only active power dispatch of EVs but ignoring
reducing the scenarios to 20 while maintaining a reasonably accurate transformer’s aging constraints, which is considered to prove the effi-
representation of the underlying uncertainties and the weights of these cacy of the proposed method. Therefore, Method-2 is simulated without

1.2 0.9
Normalized PV power in p.u.

0.8
Normalized load in p.u.

1 Spring Spring
Autumn 0.7 Autumn
Summer Summer
0.8 0.6
0.5
0.6
0.4
0.4 0.3
0.2
0.2
0.1
0 0
0 5 10 15 20 25 0 5 10 15 25
Time (Hour) 20
Time (Hour)
(a) (b)

Fig. 3. Normalized forecasts (p.u.) for representative days of three seasons. (a) Workplace non-EV demand. (b) PV generation.

Table 2
Weights of 20 scenarios.
s 1 2 3 4 5 6 7 8 9 10

πs 0.05 0.05 0.03 0.08 0.06 0.09 0.07 0.07 0.04 0.02
s 11 12 13 14 15 16 17 18 19 20
πs 0.03 0.07 0.07 0.03 0.03 0.05 0.06 0.08 0.01 0.01

6
Rayaprolu.M. Raghuveer et al. Electric Power Systems Research 238 (2025) 111062

taking into account reactive power dispatch i.e., QEVn,t = 0 and the opti-

EV reactive power(kVAr)
0
mization problem comprises only (9)-(19) by excluding aging con-
straints (20)-(25). However, the proposed EMS incorporates both active Method-1
and reactive power dispatch of EVs while limiting accelerated aging of -500 Method-2
Proposed
the transformer. Besides, all these simulations are conducted for a
duration of 24 h (T = 24) with a time resolution of 1 h (Δt = 1) across all -1000
the representative seasonal days (D = 3). Lastly, the simulation results 0 5 10 (a) 15 20 25
prove the efficacy of the proposed EMS from the viewpoint of the dis-

EV reactive power(kVAr)
0
tribution system operator (DSO), charging station owner, and EV pro-
sumer, respectively, as elaborated below.
Method-1
-500 Method-2
4.2.1. Aggregate active and reactive power schedules of EVs Proposed
Fig. 4 displays the cumulative EV active power for all methods during
all seasonal days. As mentioned earlier, the proposed EMS schedules the -1000
0 5 10 15 20 25
(b)
active and reactive powers of all EVs in real-time to mitigate the vari-

EV reactive power(kVAr)
ations of both real-time and the expected net real/reactive demand.
0

Thus, total EV power is higher when the expected net apparent demand
is lower and vice-versa to flatten the overall load profile. For instance, Method-1
during summer, EVs discharge active power during the 8th, 9th, 17th,
-500 Method-2

and 18th hours as the net active load is higher (lower PV generation) and
Proposed

charge during the remaining hours (10–16) as the net active load is -1000
lower (higher PV generation), as shown in Fig. 4(b). However, the total 0 5 10 (c) 15 20 25

EV energy during the day remains the same for all the seasonal days, Time (Hour)
irrespective of the differences in their net load profiles. Additionally, it is
Fig. 5. Aggregate EV reactive power. (a) Spring day. (b) Summer day. (c)
evident that aggregate EV active power for Method-2 is similar to the
Autumn day.
proposed EMS as both these techniques account for active power
dispatch of EVs. Also, total EV power is the highest in Method-1 across
hour, thus leading to the maximum total reactive dispatch of 750 kVAr,
all seasonal days owing to uncoordinated charging during 8 a.m.− 10 a.
as shown in Fig. 5(a). Even though EVs could operate in any quadrant, it
m., where 90 % of the EVs start charging without any delay after
is essential to note that the proposed EMS restricts it to 3rd and 4th
arriving.
quadrants by injecting the reactive power to mitigate the expected
Fig. 5 shows the aggregate EV reactive power for all methods. It is
building reactive demand locally, thereby minimizing the total apparent
noticed from Fig. 5 that total reactive power remains zero for both
demand drawn by the distribution transformer.
Method-1 and Method-2, as neither of these methods takes reactive
power dispatch into account. In contrast, with the proposed EMS, it
4.2.2. Transformer loading and peak demand to average ratio
increases with an increase in building reactive demand and the number
This study assumes that the existing distribution transformer is not
of connected EVs during that time slot and vice-versa. For example, on a
designed to cater to the additional EV demand. Consequently, the
spring day, the expected reactive demand is highest (750 kVAr) at 15th
transformer’s rating is chosen to be 1200 kVA, considering the peak

1000
EV active power(kW)

2000
Transformer load(kVA)

Method-1 Method-1
Method-2 Rated capacity of
Method-2
Proposed Transformer
500 1500 Proposed

1000
0
-100
0 5 10 15 20 25
(a) 500
0 5 10 (a) 15 20 25
1000
EV active power(kW)

Method-1 2000 Method-1

Transformer load(kVA)

Method-2 Rated capacity of Method-2

Proposed Transformer Proposed
500 1500

0 1000
-100
0 5 10 15 20 25
(b) 500
1000 0 5 10 (b) 15 20 25
EV active power(kW)

Method-1 2000
Transformer load(kVA)

Method-2 Method-1
Proposed Method-2
500 Rated capacity of Proposed
Transformer
1500

0
-1000 5 10 20 25 1000
(c) 15
Time (Hour) 0 5 10 (c) 15 20 25
Time (Hour)
Fig. 4. Aggregate EV active power. (a) Spring day. (b) Summer day. (c)
Autumn day. Fig. 6. Transformer loading. (a) Spring day. (b) Summer day. (c) Autumn day.

7
Rayaprolu.M. Raghuveer et al. Electric Power Systems Research 238 (2025) 111062

Table 4
Results of methods for representative seasonal days.
Season Method FdEQA LoLd
Trafo
(%) PAR Cost (USD)

Spring Method-1 10.6 0.0725 2.303 2165.2

Method-2 0.458 0.0031 1.8269 1752.6
Proposed 0.034 0.000229 1.749 1434.6
Summer Method-1 66.8 0.458 2.0126 2799.8
Method-2 6.46 0.042 1.7812 2213.3
Proposed 0.383 0.0026 1.7095 1883.2
Autumn Method-1 147 1.001 1.8788 3097.6
Method-2 19.6053 0.134 1.7817 2480.6
Proposed 0.412 0.0028 1.6977 2074.6

( )
building capacity and the maximum PV capacities as previously indi- factor
Trafo
Kt . It is also seen in Fig. 7(c), during autumn day, the
cated. The loading of the transformer for all seasonal days covering all
three methods is shown in Fig. 6. It is noticed from Fig. 6 that across all maximum value of FtAA = 1723.3, 74.3 & 2.45, occurs at 10th, 17th and
seasonal days, transformer loading depends on the total demand 18th hours, using Method-1, Method-2 and the proposed EMS, respec-
Trafo
including EV charging load. Using the proposed EMS, during the autumn tively. This is due to higher average θAMB
t and Kt for an autumn day.
and summer seasons, the transformer is loaded near to its rated capacity, Further, the peak value of FtAA is far greater than 2.7 for Method-1 and
which is attributed to the higher total demand. Besides, it exceeds the Method-2 as the constraint (21) is not incorporated, thereby leading to
rated capacity from 17th-20th hours as the total demand is higher, accelerated aging. However, it is observed from Fig. 7 that using the
which is in turn because of large non-EV demand and zero PV genera- proposed method, during a given seasonal day d, at any time t,
tion. Conversely, it stays well below the rated capacity for the spring day FtAA < 2.7, thereby satisfying the constraint (21). Practically, the
due to lower total demand. Furthermore, in Method-2, the transformer transformer is subjected to varying load and ambient temperatures
( )
experiences higher loading when compared to the proposed method
through-out the day. Therefore, equivalent aging factor FdEQA is
across all seasons, due to non-consideration of reactive power dispatch
of EVs. Also, the highest transformer overloading during any seasonal calculated using (26) by taking the average of FtAA . Subsequently,
( )
Trafo
day occurs in Method-1, followed by Method-2 and then the proposed transformer’s loss-of-life over a day LoLd is determined according
EMS. As a result, the Peak-to-Average Ratio (PAR) is the least for the
to (27). Considering 40 years as the normal insulation life of transformer
proposed EMS, exhibiting an average reduction of up to 16.76 % when Trafo
(ILN), LoLd shall not exceed 0.0068 % during any representative day
compared to Method-1, as detailed in Table 4.
Trafo
[33]. Further, FdEQA & LoLd for all methods across all the seasonal days
4.2.3. Effect on transformer aging and loss-of-life are reported in Table 4. It can be inferred from Table 4, when employing
( ) Trafo
As previously mentioned in Section 3.1, hot-spot temperature θHST the proposed EMS, both FdEQA < 1 andLoLd < 0.0068%, which in-
( AA ) t
can be used to estimate the transformer accelerated aging Ft , that dicates life extension. However, this is violated in case of Method-1 and
( ) Method-2, causing loss-of-life. The results show a substantial average
depends on both ambient temperature θAMB t and transformer’s loading
reduction of 99.63 % in LoL using the proposed EMS in comparison to
the uncoordinated charging method, and hence, prolonging the trans-
former’s life.
Accelerated Aging Factor FtAA

200
2
Method-1
1.5 ( T )
∑ AA
Ft × Δt
150 Method-2
1
Proposed
0.5
FdEQA = t=1 (26)
100 0
0 5 10 15 20 25 T
50 ( EQA )
Fd × T
LoLTrafo = × 100 (27)
0
25
d
ILN
0 5 10 15 20
(a)
4.2.4. Operating cost
Accelerated Aging Factor FtAA Accelerated Aging Factor FtAA

900 25
Method-1 20 Under grid pricing scheme based on both active power ($/kWh) and
reactive power ($/kVArh) adopted from [34], the operational cost is the
Method-2 15
600 Proposed 10
5 least resulting in an average annual saving of 33.12 % using the pro-
300 0
0 10 15 20 25 posed EMS in contrast to Method-1 as demonstrated in Table 4. This can
5
be attributed to the significant reduction in transformer loading through
0 active and reactive dispatch of EVs. The operating cost on any given
seasonal day is higher for Method-2 than the cost associated with the
0 5 10 (b) 15 20 25

1800 80 proposed EMS, primarily because Method-2 lacks EV reactive power

Method-1 60 dispatch. In contrast, the operating cost reaches its peak in Method-1,
predominantly due to increased transformer loading from 8 a.m. − 10
Method-2
1200 40
Proposed
20
0 a.m.
0 5 10 15 20 25
600
4.2.5. EVs final energy fulfillment
00 The primary purpose of the developed EMS is to meet the trans-
25
portation requirements of EVs for alleviating range anxiety. Moreover,
5 10 (c) 15 20
Time (Hour)
all EVs reach their intended final energy levels before departing, as
Fig. 7. Transformer accelerated aging factor FtAA . (a) Spring day. (b) Summer illustrated in Fig. 8(a), using the designed EMS. Although it is exem-
day. (c) Autumn day. plified for an autumn day, it also holds true for all other seasons. As an

8
Rayaprolu.M. Raghuveer et al. Electric Power Systems Research 238 (2025) 111062

38 8 32
36

EV active power(kW)
6 31

EV Energy (kWh)
34
4 EV Power 30

EV Energy (kWh)
32 EV Energy 29
2
30
Initial Energy 0 28
28
Final Energy
26 -2 27
24 -4 26
22 -6 25
20 -8 24
0 10 20 30 40 50 60 70 80 90 100 9 10 11 12 13 14 15 16 17
Number of EVs Time (Hour)
(a) (b)
Fig. 8. Energy of EVs (a) plot of initial and final energy of all EVs (b) energy progression of EV17.

illustration, the energy progression of EV17, which arrives with an

Solution time (sec)

80 Connected EVs
Solution time 3
initial energy of 23 kWh at 9 a.m. and attains the final energy of 32 kWh

Connected EVs
60
prior to its departure at 5 p.m., is presented in Fig. 8(b). 2
40
20
4.3. Resiliency to dynamic EV arrivals with proposed EMS employing 1
0
receding horizon framework 6 8 10 12 14 16 18 20 24
Time (Hour)
The proposed EMS can adapt to the EVs’ dynamic arrivals, de-
Fig. 10. Computational time of the proposed EMS for a summer day.
partures, and initial energy, where any update is accompanied by re-
optimization. It shall be noted that the number of connected EVs in
the parking lot at each time slot changes, necessitating a dynamic proposed EMS to provide the optimal EV schedules, will allow it to
rescheduling of the connected EVs to mitigate the expected apparent operate in real-time, even with time slots of less than a minute.
demand. To confirm this, an EV7, which arrives with an initial energy of
23 kWh at 9 a.m. and departing at 17 p.m., for a typical summer day has
4.5. Effect of PV generation on the transformer life
been considered. The original active and reactive demand schedules for
this EV during 9th-12th hours are updated to new schedules after every
To demonstrate the impact of PV generation on the transformer’s
hour of rescheduling, as illustrated in Fig. 9(a) and (b), respectively.
life, three scenarios are considered for an autumn day including (1) with
Although rescheduling for this EV happens until the 17th hour, it has
adequate PV generation corresponding to 100 % irradiance, (2) with 50
only been shown for the first four hours (9 a.m.− 12 p.m.) for better
% PV generation, and (3) with sudden cloud cover resulting in 25 % PV
understanding.
generation. The transformer loading under the said scenarios is shown in
Fig. 11. It can be witnessed from Fig. 11 that the proposed EMS responds
4.4. Feasibility of the proposed EMS for real-time application to change in PV generation by scheduling the EVs such that they will
discharge during periods of low PV generation and vice-versa. It is also
The solution time of the receding horizon optimization framework observed that reduced PV generation leads to higher loading of the
and the connected EVs for each time slot during a typical summer day is transformer. Consequently, transformer’s life is determined using (28)
shown in Fig. 10. The short computation time of 3 s, required by the [36] and found to be 97.84 years, 41.51 years, and 23.61 years for

8 -7.8
EV reactive power(kVAr)

6 9th hour
EV active power(kW)

-8
4 10thhour
2 -8.2 11thhour
12thhour
0 -8.4
-2
-4 -8.6
9th hour
-6 10thhour -8.8
-8 11thhour
12thhour -9
-10 9 10 11 12 13 14 15 16 17
9 10 11 12 13 14 15 16 17
Time (Hour) Time (Hour)
(a) (b)

Fig. 9. Dynamics of EV 7 (a) active power schedules (b) reactive power schedules.

9
Rayaprolu.M. Raghuveer et al. Electric Power Systems Research 238 (2025) 111062

100% PV 50% PV 25% PV

1400
Transformer loading (kVA)

1200

1000

800

600

400

200

0
6 8 10 12 14 16 18 20
Time (Hour)

Fig. 11. Transformer loading for different scenarios demonstrating the effect of
PV generation.

scenario-1, scenario-2, and scenario-3, respectively. Hence, the trans-

former’s life is significantly reduced for scenario-3 with poor PV gen-
eration when compared to scenario-1 having adequate PV generation.
( )
1
Life Trafo
= × 100 (28)
365.LoLTrafo
d

4.6. Impact of initial energy of EVs

( )
To evaluate the impact of EVs initial energy EEVinitial
m on the pro-
posed algorithm, a sensitivity analysis is conducted by varying the mean
value of EEVinitial
m from 20 % to 70 % of the EV battery capacity in steps of
10 %. The corresponding variations in transformer’s loss of life (LoL),
peak-to-average ratio (PAR), and cost with respect to the initial energy
of EVs are shown in Figs. 12(a), (b) and (c), respectively. As depicted in Fig. 13. Effect of variation in PV capacity on (a) transformer LoL (b) PAR
(c) cost.
Fig. 12, an increase in the initial energy of EVs results in reduced
transformer’s LoL, increased PAR, and decreased cost. This is because
higher initial energy levels of EVs reduce the aggregate EV demand respectively. The variations in transformer’s loss of life (LoL), peak-to-
according to (29), thereby lowering the average load while having average ratio (PAR), and cost with respect to the PV capacity and non-
minimal impact on the peak load, which typically occurs between EV load capacity are illustrated in Figs. 13 and 14, respectively. As
18:00–20:00 h when most of the EVs depart the charging station. It is shown in Fig. 13, an increase in PV generation results in a reduction in
( ) the transformer’s LoL, an increase in PAR, and a decrease in cost. This is
important to note that variation in final energy at departure EEVdep
m will
( EVinitial ) due to the fact that higher PV generation reduces the average load but
exactly have an opposite effect to that of EVs initial energy Em .
has minimal impact on the peak load, which occurs during the hours of
Therefore, it is observed from Fig. 12 that the best and worst values of 18:00–20:00 when PV generation is absent. Conversely, it is noticed
initial EV energy are 70 % and 20 %, respectively with EEVdep
m being 80 %. from Fig. 14 that an increase in non-EV load capacity results in higher
m
∑ transformer LoL, decreased PAR, and increased costs. This is because of
[ EVdep ]
TotalEVdemand = Em − EEVinitial
m (29) the increase in average load with an increase of non-EV demand.
m=1

4.8. Comparison with existing works

4.7. Effect of PV generation and non-EV load capacities
The proposed EMS has been compared with several existing state-of-
To evaluate the impact of PV generation and non-EV load capacities art algorithms and their performance in terms of three metrics are
on the proposed algorithm, a sensitivity analysis is conducted for a highlighted in Table 5. It is worthwhile to mention that the optimization
spring day by varying their capacities from 75 % to 125 % of their problem formulations of the existing algorithms are used in the same
maximum capacities in steps of 12.5 %. The maximum capacity (100 %) way as mentioned in the respective work. To ensure a fair comparison,
of PV generation and non-EV load correspond to 600 kW, and 1500 kW, the same PV generation and non-EV demand profiles utilized in the

Fig. 12. Effect of EVs initial energy on (a) transformer LoL (b) PAR (c) cost.

10
Rayaprolu.M. Raghuveer et al. Electric Power Systems Research 238 (2025) 111062

proposed method are applied to all existing schemes. Additionally, all

studies consider 100 EVs and are implemented under receding horizon
approach to account for real-time EV dynamics. The scheme presented
in [12] seeks to minimize the energy cost besides stationary battery
energy storage degradation cost. However, this work has overlooked
both active and reactive power dispatch of EVs in V2G mode. Further,
the use of stationary battery increases the demand during low price
hours as it tries to charge during low price hours and discharge during
high price hours. In [35], the objective function tries to flatten the total
active load demand including EV demand, whereas the algorithm
described in [36] optimizes both loss-of-life of transformer and energy
arbitrage cost. However, reactive power support of EVs is not taken into
account in the above-mentioned approaches. The scheme presented in
[37] minimizes the load factor without considering either the active or
reactive power contributions from EVs. It is evident from Table 5 that
the proposed method has outperformed the prevailing approaches in
terms of lower transformer loss-of-life, improved peak to average ratio
(PAR) and reduced cost. This is because proposed method schedules
both active and reactive power from EVs to offset the net active load and
non-EV reactive demand, respectively. Consequently, apparent loading
on the transformer is reduced, leading to an improvement in transformer
life and a decrease in PAR and reduced cost. Although, cost optimization
is not the main focus of this work, cost reduction is also realized by
lowering the apparent loading of the transformer. Therefore, the cost
using the proposed method is comparable to those in [12] and [36] that
focus only on optimizing active power costs but overlooking its reactive
counterpart.

4.9. Scalability

Fig. 14. Effect of variation in non-EV load capacity on (a) transformer LoL (b)
The proposed EMS algorithm is applied at the local node where the
PAR (c) cost. workplace PV-EV charging station is installed. Consequently, EVs, PV,
and transformer constraints of this particular node are decoupled from
the large scale power system network. Additionally, the receding hori-
Table 5 zon implementation restricts the size of proposed MIQCP problem to the
Comparison of the proposed scheme with several existing state-of-art dimension of EVs at a specific time slot, usually involving few hundred
algorithms. EVs. To demonstrate scalability, the proposed EMS has been tested by
Scheme presented in Trafo
LoLd (%) PAR Cost (USD) varying the number of EVs in charging station to 150, 200, 250, and 300,
− 4
respectively. For all these cases, the computational time at each time slot
[12] 32 × 10 2.228 1472.8
is illustrated in Fig. 15. It can be witnessed from Fig. 15 that the
[35] 31 × 10− 4 1.826 1752.6
[36] 4.58 × 10− 4 1.795 1500.3 maximum computational time taken is 854.56 s at 11th hour in case of
[37] 10 × 10− 4 1.78 1648.5 300 EVs, thus making the proposed EMS feasible for real-time operation
Proposed scheme 2.1 × 10− 4 1.749 1434.6 and scalable for large EV fleet.

1000
150 EVs
200 EVs
800 250 EVs
Computational time (s)

300 EVs
600

400

200

0 6 8 10 12 14 16 18 20
Time (Hour)
Fig. 15. Computational time for varying number of EVs.

11
Rayaprolu.M. Raghuveer et al. Electric Power Systems Research 238 (2025) 111062

5. Conclusions [8] T. Ma, O. Mohammed, Optimal charging of plug-in electric vehicles for a car-park
infrastructure, IEEE Trans. Ind. Appl. 50 (4) (2014) 2323–2330.
[9] P. Kou, D. Liang, L. Gao, F. Gao, Stochastic coordination of plug-in electric vehicles
This work proposes a real-time stochastic convex MIQCP-based EMS and wind turbines in microgrid: a model predictive control approach, IEEE Trans.
designed for a workplace integrated with EVs and PV to minimize the Smart Grid 7 (3) (2016) 1537–1551.
expected apparent demand considering transformer’s aging, diurnal [10] D. Van Der Meer, G.R.C. Mouli, G.M.E. Mouli, L.R. Elizondo, P. Bauer, Energy
management system with PV power forecast to optimally charge EVs at the
uncertainties and seasonal variations associated with the workplace workplace, IEEE Trans. Ind. Informatics. 14 (1) (2018) 311–320.
non-EV load and PV generation. Moreover, the developed EMS is [11] G.R.C. Mouli, M. Kefayati, R. Baldick, P. Bauer, Integrated PV charging of EV fleet
operated under receding horizon fashion adapting to real-time EV dy- based on energy prices, V2G, and offer of reserves, IEEE Trans. Smart Grid 10 (2)
(2019) 1313–1325.
namics. The proposed approach is verified through simulations using a [12] C.M. Affonso, M. Kezunovic, Technical and economic impact of PV-BESS charging
practical case study, which is compared with two existing methods of EV station on transformer life: a case study, IEEE Trans. Smart Grid 10 (4) (2019)
charging, namely (i) uncoordinated charging and (ii) smart charging 4683–4692.
[13] P. Andrianesis, M. Caramanis, Distribution network marginal costs: enhanced AC
without reactive power dispatch. The findings reveal a substantial OPF including transformer degradation, IEEE Trans. Smart Grid 11 (5) (2020)
average annual reduction of 99.63 % in transformer loss of life (LoL) and 3910–3920.
a 16.76 % decrease in peak demand, using the developed EMS when [14] S.S. Karimi Madahi, et al., Co-optimization of energy losses and transformer
operating costs based on smart charging algorithm for plug-in electric vehicle
compared to an uncoordinated charging scheme. Furthermore, deploy- parking lots, IEEE Trans. Transport. Electrific. 7 (2) (2021) 527–541.
ment of the proposed EMS in contrast to an uncoordinated charging [15] S. Pirouzi, J. Aghaei, M.A. Latify, G.R. Yousefi, G. Mokryani, A robust optimization
approach increases an average annual profit of EV charging station approach for active and reactive power management in smart distribution
networks using electric vehicles, IEEE Syst. J. 12 (3) (2018) 2699–2710.
operator up to 33.12 %. Additionally, the final energy demand of all the
[16] J. Wang, G.R. Bharati, S. Paudyal, O. ceylan, B.P. Bhattarai, K.S. Myers, Reactive
EVs is also fulfilled before their departures. Further, the proposed EMS is power support to distribution grids, IEEE Trans. Ind. Informatics. 15 (1) (2019)
found to be scalable for large EV fleet and feasible for real-time opera- 54–63.
tion due to its faster computational time. This work can be extended by [17] Q. Hu, S. Bu, V. Terzija, A distributed P and Q provision based voltage regulation
scheme by incentivized EV fleet charging for resistive distribution networks, IEEE
deploying the proposed algorithm in a distribution network, including Trans. Transport. Electrific. 7 (4) (2021) 2376–2389.
both workplace and residential EV charging facilities to achieve [18] S. Pineda, J.M. Morales, Chronological time-period clustering for optimal capacity
continuous utilization throughout a 24 h period. Additionally, a chance- expansion planning with storage, IEEE Trans. Power Syst. 33 (6) (2018)
7162–7170.
constraint programming problem can be developed by incorporating [19] Y. Xu, Z.Y. Dong, R. Zhang, D.J. Hill, Multi-timescale coordinatedvoltage/var
power flow and voltage constraints for the energy management of an control of high renewable-penetrated distribution systems, IEEE Trans. Power Syst.
active distribution network with multiple EV charging stations consid- 32 (6) (2017) 4398–4408.
[20] T. Niknam, M. Zare, J. Aghaei, Scenario-based multiobjective volt/var control in
ering uncertainties. Further, incentives for EV battery degradation can distribution networks including renewable energy sources, IEEE Trans. Power Del.
be included in the energy management system to enhance V2G partici- 27 (4) (2012) 2004–2019.
pation while addressing concerns related to battery deterioration. [21] IEEE, IEEE Std C57.91-2011- IEEE guide for loading mineral- oil-immersed
transformers and step-voltage regulators, 2012.
[22] T. Akbari, M.T. Bina, A linearized formulation of AC multiyear transmission
CRediT authorship contribution statement expansion planning: a mixed-integer linear programming approach, Elect. Power
Syst. Res. 114 (2014) 93–100.
[23] Open Charge Alliance, “Open charge point protocol 2.0.1,” 2022. [Online].
Rayaprolu.M. Raghuveer: Writing – original draft, Methodology,
Available: https://www.openchargealliance.org/protocols/ocpp-201/.
Conceptualization. Bhavesh Bhalja: Writing – review & editing, Su- [24] Road Veh. Veh.-to-Grid Commun. Interface part 2: netw. and appl.protoc.
pervision. Pramod Agarwal: Writing – review & editing, Supervision. requirements, ISO/IEC 15118-2, 2014. [Online]. Available: https://www.iso.org/h
ome.html.
[25] IBM, "IBM ILOG CPLEX optimization studio," Tech. Rep., 2014. [Online].
Declaration of competing interest Available: http://pic.dhe.ibm.com.
[26] W. Kersting, Distribution System Modeling and Analysis, CRC, Boca Raton, FL,
The authors declare that they have no known competing financial 2002.
[27] T. Gangwar, N.P. Padhy, P. Jena, Energy management approach to battery energy
interests or personal relationships that could have appeared to influence storage in unbalanced distribution networks, IEEE Trans. Ind. Appl. 60 (1) (2024)
the work reported in this paper. 1345–1356.
[28] L. Wu, M. Shahidehpour, T. Li, Cost of reliability analysis based on stochastic unit
commitment, IEEE Trans. Power Syst. 23 (3) (2008) 1364–1374.
Data availability [29] S. Sannigrahi, S.R. Ghatak, P. Acharjee, Multi-scenario based bi-level coordinated
planning of active distribution system under uncertain environment, IEEE Trans.
Data will be made available on request. Ind. Appl. 56 (1) (2020) 850–863.
[30] NREL (National Renewable Energy Laboratory). PVWatts Calculator. [Online].
Available: https://pvwatts.nrel.gov/.
[31] FHWA, National household travel survey, 2017 national household travel survey
References data. 2017. [Online]. Available: https://nhts.ornl.gov/.
[32] R.M. Raghuveer, B.R. Bhalja, P. Agarwal, Real-time energy management of EVs in a
microgrid integrated with distributed energy resources considering uncertainties,
[1] IEA, Global EV Outlook 2021, IEA, Paris, 2021. https://www.iea.org/reports/glo
in: 3rd IEEE International Conference on Sustainable Energy and Future Electric
bal-ev-outlook-2021.
Transportation (SEFET), IEEE, 2023, pp. 1–6.
[2] G.R.C. Mouli, et al., Economic and CO2 emission benefits of a solar powered
[33] P.K. Sen, S. Pansuwan, Overloading and loss-of-life assessment guidelines of oil-
electric vehicle charging station for workplaces in The Netherlands, in: IEEE
cooled transformers, in: 45th Annual Rural Electric Power Conference, IEEE, 2001,
Transportation Electrification Conference Exposition (ITEC), IEEE, 2016, pp. 1–7.
pp. 1–8.
[3] K. Clement-Nyns, E. Haesen, J. Driesen, The impact of charging plug-in hybrid
[34] O. Gandhi, W. Zhang, C.D. Rodríguez-Gallegos, M. Bieri, T. Reindl, D. Srinivasan,
electric vehicles on a residential distribution grid, IEEE Trans. Power Syst. 25 (1)
Analytical approach to reactive power dispatch and energy arbitrage in
(2010) 371–380.
distribution systems with DERs, IEEE Trans.Power Syst. 33 (6) (2018) 6522–6533.
[4] Y. He, B. Venkatesh, L. Guan, Optimal scheduling for charging and discharging of
[35] B. Kandpal, A. Verma, Demand peak reduction of smart buildings using feedback-
electric vehicles, IEEE Trans. Smart Grid 3 (3) (2012) 1095–1105.
based real-time scheduling of EVs, IEEe Syst. J. 16 (3) (2022) 4279–4290.
[5] P. Sanchez-Martin, S. Lumbreras, A. Alberdi-Alen, Stochastic programming applied
[36] M.R. Sarker, D.J. Olsen, M.A. Ortega-Vazquez, Co-optimization of distribution
to EV charging points for energy and reserve service markets, IEEE Trans. Power
transformer aging and energy arbitrage using electric vehicles, IEEE Trans. Smart
Syst. 31 (1) (2016) 198–205.
Grid 8 (6) (2017) 2712–2722.
[6] C. Chen, S. Duan, Optimal integration of plug-in hybrid electric vehicles in
[37] İ. Şengör, O. Erdinç, B. Yener, A. Taşcıkaraoğlu, J.P.S. Catalão, Optimal energy
microgrids, IEEE Trans. Ind. Informat. 10 (3) (2014) 1917–1926.
management of EV parking lots under peak load reduction based DR programs
[7] M. Soleimani, M. Kezunovic, Mitigating transformer loss of life and reducing the
considering uncertainty, IEEe Trans. Sustain. Energy 10 (3) (2019) 1034–1043.
hazard of failure by the smart EV charging, IEEE Trans. Ind. Appl. 56 (5) (2020)
5974–5983.

12