Electrical Power and Energy Systems 158 (2024) 109946

Contents lists available at ScienceDirect

International Journal of Electrical Power and Energy Systems

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

Dynamic pricing strategy for electric vehicle charging stations to distribute

the congestion and maximize the revenue
Abed Kazemtarghi, Ayan Mallik ∗, Yan Chen
The Polytechnic School, Arizona State University, 6075 Innovation Way W, Mesa, 85212, AZ, USA

ARTICLE INFO ABSTRACT

Keywords: Electric vehicle (EV) charging station (CS) congestion is highly dependent on the EV owner’s behavior and
Charging station pricing their selected CS as charging choice. A fixed pricing strategy causes some CSs to be congested with EVs
Dynamic pricing strategy waiting in line to charge, while there are some other CSs with available electric plugs, which adversely
Electric vehicles charging
affects both charging station operator (CSO) revenue and EV users’ welfare. To solve this problem, this paper
Scenario-based stochastic optimization
presents a dynamic pricing strategy aimed at conducting EVs from congested CSs to the uncongested ones
through controlling the charging prices of CSs at different times. The problem is formulated as a scenario-
based stochastic optimization with the objective of maximizing overall revenue of the CSO. Moreover, an
attraction function model is developed to quantify the charging choice of the EV owners by considering the
effective parameters of CSs in EV charging choice decisions. QGIS software is used in this work to formulate a
realistic modeling for CS locations and EV routes to calculate the distances from EVs to CSs. Three scenarios
are designed to evaluate the performance of the proposed framework and to compare the results with the fixed
pricing approach. The results indicate that the proposed dynamic pricing strategy mitigates the congestion of
CSs while facilitating an increased number of charged EVs up to 48% as well as increasing the overall revenue
of the CSO.

1. Introduction CSs. For this purpose, various dynamic charging pricing methods are
proposed [6–9].
Nowadays, transportation electrification is welcomed by both con- There are several works in the literature that study charging pricing
sumers and governments due to lower fuel cost and supporting green strategy as competition between CSOs and model the problem using
energy resources. Electric vehicles (EVs) usage is growing rapidly be- game theory frameworks [10–17]. The authors in [10] investigate the
cause of remarkable advancement in EV manufacturing technology. price competition among CSs with renewable power generators using
Owing to the possibility of working in both grid-to-vehicle (G2V) and the Stackelberg game method. The price elasticity of EVs, the effect of
vehicle-to-grid (V2G) modes of operation, EVs can be considered as the distance between an EV and the CSs, and the impact of the number
flexible resources moving around the network from power systems and type of charging outlets are considered in this work; however, this
operation perspective. High penetration level of the EVs affects the work does not consider a realistic model of EV routes to the CSs and
normal operation of the power grid, which may pose serious challenges calculates the distances according to a simple geometric model which
to the distribution system operator (DSO) [1–3].
affects the accuracy of the results. The work in [11] presents a strategic
Charging management of the EVs distributed over the grid and
charging pricing scheme for CSOs based on a non-cooperative Stack-
charging stations (CSs) is one of the most demanding problems in power
elberg game framework. The Stackelberg equilibrium investigates the
system studies [4,5]. For example, depending on the location of CSs,
pricing competition among multiple CSOs while a soft actor-critic based
quantity and location of EVs, and at a specific time slot, let us consider
multi-agent deep reinforcement learning algorithm is developed to
a scenario where some CSs are congested with EV users waiting for
solve the proposed equilibrium framework while considering privacy-
charging while there are plenty of available charging plugs in the
other CSs. This situation has adverse impacts on the load profile of the conservation constraints among CSOs. Nevertheless, this work does not
grid and decreases the charging station operator (CSO) profits through present a closed form expression to show how vehicles are attracted to
selling electricity to the EV users. In this case, the electricity price of the different CSs. The authors in [12] analyze price competition among
the CS is the main variable that can attract EVs to the uncongested CSs with service capacity constraints and use an ordinal potential

∗ Corresponding author.
E-mail address: [email protected] (A. Mallik).

https://doi.org/10.1016/j.ijepes.2024.109946
Received 15 August 2023; Received in revised form 18 February 2024; Accepted 13 March 2024
Available online 20 March 2024
0142-0615/© 2024 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY-NC license (http://creativecommons.org/licenses/by-
nc/4.0/).
A. Kazemtarghi et al. International Journal of Electrical Power and Energy Systems 158 (2024) 109946

the charging plug capacity and charger types existing in the CS, which
Nomenclature affects the competition between the CSs. The work in [13] proposes a
non-cooperative game pricing strategy framework by the approach of
Abbreviations profit-sharing and user equilibrium principles to maximize the social
𝐶𝑆 Charging station welfare of the electrified transportation system stakeholders consisting
of electricity wholesalers, fast charging stations, and EV users, but this
𝐶𝑆𝑂 Charging station operator
work also does not propose a closed form function to show how EVs
𝐷𝑆𝑂 Distribution system operator
select their desired CS for charging. The work in [14] is focused on
𝐸𝑉 Electric vehicle the EV public charging market with heterogeneous CSs under the time-
𝐺2𝑉 Grid-to-vehicle based billing model. In the proposed hierarchical game, each CS sets
𝐾𝐾𝑇 Karush-Kuhn–Tucker the charging price to maximize its own revenue first, then the EVs
𝑃 𝐸𝑉 Plug-in electric vehicles choose their desired CSs and determine the charging time; however, the
𝑄𝐺𝐼𝑆 Quantum geographic information system power flow constraints, which affect the pricing strategy and relevant
𝑆𝑂𝐶 State of charge results, are not considered in this work. The work in [15] studies
𝑉 2𝐺 Vehicle-to-grid dynamic pricing policy for maximizing the long-term profits of CSOs
using multi-agent reinforcement learning and Markov game to model
Parameters, Variables, and Indices the layer of CSOs as a competitive market. However, the proposed
𝜔𝑟 he worst-case scenario in the ambiguity set model does not reflect the impact of charging level and number of
electric plugs in each CS, which regulates the competitive market.
𝑃𝑟 Upper limit of the charging price
The work in [16] is focused on maximizing the sum of the revenues
𝑃𝑟 Lower limit of the charging price
of all CSs managed by one corporation, assuming other corporations’
𝐴𝑇 𝑇𝑖𝑗𝑡 Attraction of charging station 𝑖 to EV pricing strategies are fixed. The EV charging and CS pricing problems
customer 𝑗 at time 𝑡 in this work are modeled as a hierarchical Stackelberg game with the
𝐶𝑏𝑎𝑡𝑡,𝑘 Battery capacity of the EV connected to the corporation at the upper layer as the leader and EV flows at the lower
plug 𝑘 layer as followers. Nevertheless, a simple linear assumption for the
𝑑𝑖𝑗𝑡 Distance from EV customer 𝑗 to charging queuing cost is considered in this work that does not reflect the realistic
station 𝑖 at time 𝑡 model of EV sections.
𝑓 Total revenue of CSO over all time horizon Moreover, there are multi-level optimization methods proposed in
under study the literature aimed at solving competitive charging pricing prob-
𝑓𝑖𝑡 Revenue of the charging station 𝑖 at time 𝑡 lems [18–21]. The work in [18] pro-poses a tri-level pricing framework
𝑗 Index for EV customers where CSs set charging prices at the upper level, EVs make route
𝑘 Index for charging plugs and charging options at the middle level, and the electricity price is
𝑙𝐴 ∕𝑙𝐵 Distance of centers 𝐴/𝐵 from the interme- cleared in the power distribution network at the lower level. Then,
diate location the single-level optimal pricing model is established with traffic flow
assigned, power generation scheduled, and electricity price explicitly
𝑀 Total number of charging stations indexed
contained via Karush–Kuhn–Tucker (KKT) conditions. A tri-level opti-
by 𝑖
𝑡 mization model is proposed in [19], where the upper and middle levels
𝑁𝑎𝑣,𝑖 Number of available charging plugs in
capture the profit maximization problems of this CS and its competitor
charging station 𝑖 at time 𝑡
𝑡
respectively, while the lower level represents the optimal power flow
𝑁𝑜𝑐𝑐,𝑖 Number of occupied plugs of charging problem of the distribution network. In addition, the authors in [20]
station 𝑖 at time 𝑡 propose a bi-level programming model where the upper level is the
𝑃𝑖 Average charging power of charging station load management problem, and the lower level is the charging station
𝑖 selection problem for EVs. The optimal solution of the lower-level
𝑃 𝑂𝑃𝐴 ∕𝑃 𝑂𝑃𝐵 Populations of centers 𝐴/𝐵 problem is a function in the charging price that is determined in the
𝑃 𝑟𝑡𝑖 Charging price of charging station 𝑖 at time upper-level problem. Nevertheless, the multi-level formulation of the
𝑡 problem in these works increases the complexity of the model and
𝑆𝑖 Number of electric charging plugs in charg- needs further considerations to reformulate the problem into a single
ing station 𝑖 level to overcome the computational challenges.
𝑆𝑂𝐶𝑒𝑛𝑑,𝑘 SOC of the EV connected to the plug 𝑘 at There are also other works aimed at optimizing the EV charging
the end of charging pricing problem as well as other technical factors. The work in [22]
𝑆𝑂𝐶𝑠𝑡𝑟,𝑘 SOC of the EV connected to the plug 𝑘 at designs an optimal pricing scheme to minimize the service dropping
the start of charging rate of the charging station considering dual-mode charging station.
𝑇 Total number of time slots indexed by 𝑡 Proposed pricing scheme minimizes the number of EVs that leave the
charging station without being charged. Although this work improves
𝑇𝐴 ∕𝑇𝐵 Trade from the intermediate location at-
the service quality, it does not optimize the queue length and total
tracted by centers 𝐴/𝐵
profit of CSs. A stochastic scheduling framework embedded with an
𝑡𝑒𝑛𝑑,𝑖𝑘 Charging end time of EV connected to the
incentive charging strategy for EVs is proposed in [23], in which DSO
plug 𝑘 at charging station 𝑖
combines the scheduling of the power grid and EV charging by flexible
𝑊 Ambiguity set of scenarios price approaches to contribute to system operation, but this work does
not propose a detailed model on EV and CS location and distance
calculation, which highly impacts the selected CS by EVs. Moreover,
a price-based transfer model is proposed to describe the correlation
game framework to investigate the structure of the competition. More- between charging price and EV transfer. The work in [24] classifies
over, they propose a decentralized algorithm to enable effective price EVs into four categories according to distinct driving characteristics
coordination to achieve equilibrium with maximized social welfare; and charging requirements and then, determines optimum charging
however, the price competition model in this work does not consider price and energy management of EV parking lots by considering price

2
A. Kazemtarghi et al. International Journal of Electrical Power and Energy Systems 158 (2024) 109946

elasticity of EV charging demand. However, this work does not consider of EVs and CSs, number of charging plugs in each CS, and average
the effective parameters in the price elasticity model of EVs to reflect charging power of CSs. The geographical information and measurement
how EVs are guided toward the CSs. The authors in [25] proposed are accomplished using the quantum geographic information system
a smart dynamic pricing approach based on a fuzzy logic controller (QGIS) to have a realistic modeling of distances incorporated in the
based decision-making structure for EV charging in a distribution sys- charging process modding and dynamic pricing scheme.
tem; however, the proposed algorithm does not reflect the effective The major contributions of this paper are as follows:
parameters in EV charging process like location, number of plugs,
1. Development and all-inclusive formulation of an attraction func-
and rated power of the chargers for CSs. The presented work in [26]
tion that explains the charging choice of EVs depending on the
considers uncertainties related to the charging demand volatility, in-
charging price of the CS, distance from EV to the CS, number of
herent intermittency of renewable energy generation, and wholesale
charging plugs in CS, and average charging power of the CS.
electricity price fluctuation within the stochastic dynamic pricing for
2. Formulation of the proposed dynamic pricing strategy in the
EV charging service providers, but this work does not provide a model
form of a scenario-based stochastic optimization problem with
for CS selections by EVs that affects the EV charging problem. The
the objective to direct EV owners from congested CSs toward the
authors in [27] present a new coordinated dynamic pricing model to
uncongested ones and consequently maximize the gross profit of
reduce the overlaps between residential and CS loads by inspiring the
all CSs.
temporal Plug-in electric vehicles (PEVs) load shifting during evening
3. Utilizing a realistic geographical model of CS locations using
peak load hours. The idea in this work is to dynamically adjust the
QGIS software and helpful tools to compute the fastest routes
price incentives to drift PEVs toward less underutilized CSs. Although
from EVs to CSs in order to form the distance matrix used in the
the proposed method improves the total load profile during the whole
charging process modeling and dynamic pricing scheme.
day by guiding EVs toward the uncongested CSs, the solution does
not necessarily maximize the revenue of the CSs. The work in [28] The reason for focus on the congestion of the CSs in this manuscript
studies the travel patterns of electric vehicles to predict the controllable is that the congestion of the CSs directly represents the distribution of
capacity of electric vehicles. Then, the charging preferences of different EVs between different CSs, which impacts the number of charging EVs,
types of users are studied and a pricing strategy is developed, which ac- total sold energy, and consequently the CSO revenue. Furthermore, the
counts for the dispatching requirements of the microgrid and dispatch waiting time of the EVs can be measured by monitoring the congestion
of electric vehicle charging load based on price signals. Modeling the of the CSs and thereby enhancing the EV owners satisfaction through a
EV charging problem with focus on EV types and their status, instead more uniform distribution of EVs among CSs by the proposed dynamic
of considering important parameters from power system perspective, pricing strategy.
gives a micro understanding of the problem with lots of variables and In addition, Table 1 compares the pricing strategy, objectives, opti-
uncertainties that reduces the accuracy of the results. An online pricing mization method, and the technical considerations in this work with the
strategy is established in [29] upon a reward-based model to prevent most relevant previous works in the literature. According to the table,
network instability and power outages. The utility in this work provides the online pricing strategy is proposed in [29], which can potentially
incentives to the charging stations for their contributions in the EVs cause inconvenience to the EV users in terms of frequent price updates.
charging load shifting. Then, a constrained optimization problem is On the other hand, TOU pricing in [28,32] with hourly or three-level
developed to minimize the total charging demand of the EVs during price update cannot fully utilize the advantage of dynamic pricing
peak hours. This work presents an interesting solution to enhance the to maximize the CSO revenue. Therefore, the dynamic pricing with
power system resiliency, but it is not necessarily the optimum solution 5-minutes intervals is applied in this work to efficiently update the
from the revenue perspective. charging prices while considering the EV users’ satisfaction. Moreover,
In addition, the work in [30] proposes a dynamic pricing and the dynamic pricing problem is formulated as a scenario-based stochas-
control of a solar-powered EV CS though bi-level optimization ap- tic optimization program in this work to accurately account for the
proach, where pricing tariffs ensure an economic and price responsive uncertainty of the EV load and their individual locations. The rest of the
operation, then EV charging schedules are computed for energy bidding paper is structured as follows: Section 2 elaborates on the system model
capacity to provide balancing services. However, the decision making that comprises EV charging process modeling, attraction function to
process of the EV users on the CS selection is not modeled in this work, show how EV users select the desired CS for charging, and relevant
which potentially impacts the proposed time-of-use (TOU) dynamic assumptions. Section 3 proposes the dynamic pricing strategy with
pricing scheme. The authors in [31] investigated the charging behavior formulation in the form of the scenario-based stochastic optimization
of EVs according to the price signals to maintain grid balance. They problem. Afterwards, performance evaluation of the developed pricing
used travel simulation to establish the relationship between travel de- strategy and related simulation results are discussed in Section 4.
mand and electricity prices. Nevertheless, the formulated optimization Finally, Section 5 puts forward relevant conclusions of the work.
model in this work is only with the objective to minimize the grid
loss and voltage drop. Although technical performance enhancement 2. EV charging system model
can be considered as an objective in the model, the main goal in the
dynamic pricing problem is to maximize the revenue of the CS, which This section elaborates on the proposed modeling of the EV charging
is not considered in this work. The work in [32] developed a genetic system used in this study. Assumptions on how and when EVs select
algorithm-based multi-objective optimization model to generate hourly their desired CS to charge are explained. Attraction of a CS to the EVs
dynamic TOU electricity tariffs and facilitate the decision making in is formulated and coverage zone of CSs is defined. Then, EV charging
load scheduling. However, the problem with this work is that it studies process and distribution of EVs between CSs are modeled.
the pricing method for a single CS, while the dynamic pricing definitely
impacts the decision of the EV users on the selected CS, but this aspect 2.1. Assumptions
of the dynamic pricing method is not investigated in this work.
To address limitations related to the previous works in the liter- There are several influence factors identified from studying behav-
ature, and with the purpose of maximizing the CSO revenue as well ioral attributes of different kinds of people which affects EV charging
as a uniform distribution of EVs over CSs, a dynamic pricing strategy behavior. These factors are roughly divided into socio-demographic
is developed in this work considering a detailed EV charging process attributes and alternative attributes i.e., scenario factors [33]. The
modeling. Furthermore, the charging choice of EVs is modeled in this socio-demographic factors have been extensively investigated by some
work mathematically using the important parameters such as location other prior works in the literature [34–38]. Different travelers may

3
A. Kazemtarghi et al. International Journal of Electrical Power and Energy Systems 158 (2024) 109946

Table 1
Comparison of the proposed dynamic pricing strategy with the previous works in the literature.
Work Pricing strategy Objective(s) Optimization method Technical considerations
[26] Stochastic dynamic Max profit Stochastic dynamic Active and reactive power flow
pricing Max customer satisfaction programming (SDP)
Min the impact on the
grid
[27] Coordinated dynamic Min the overlap between Rule-based heuristic Temporal EV load shifting
Pricing the
EV load and residential
load
[28] Dynamic TOU pricing Min total charging Linear programming EV travel patterns
costs of EV users (LP) User satisfaction degree
[29] Online pricing Min the EVs’ charging Online reinforcement Load shifting
loads at peak hours learning
[32] Hourly dynamic Min electricity cost Genetic algorithm Price elasticity of demand
TOU tariff Min peak-to-valley Demand response program
difference
This work Dynamic pricing Maximize the CSO Scenario-based Modeling EV decision on the CS
with 5-min frequency revenue stochastic optimization selection
Modeling EV charging process
Geographical model of EV
distances to CSs

make different charging choices when facing the same situation, and target CS. From Reilly’s law of retail gravity [40], two cities attract
socio-demographic variables are considered to reflect the changes of retail trade from any intermediate city or town in the vicinity of the
the travelers in behavioral studies. Strong relationships exist among breaking point, approximately in direct proportion to the population of
socio-demographics, activity engagement, and travel behavior [34]. the two cities and in inverse proportion to the square of the distances
The travel patterns of low-income people and high-income people from these two cities to the intermediate town. This idea can be
are significantly different, as well as age, gender, driving experience, quantified through the following function:
and education level. In addition, the work in [35] considers the het- ( )
𝑇𝐴 𝑃 𝑂𝑃𝐴 𝑙𝐵 2
erogeneity of individuals and take these influencing factors as the = (1)
𝑇𝐵 𝑃 𝑂𝑃𝐵 𝑙𝐴
latent variables that are unique to individuals, such as risk aversion
attitude and positive vehicle maintenance attitude. The heterogeneity where 𝑇𝐴 and 𝑇𝐵 represent the trade from the intermediate location
of individuals is included in this work by studying the socioeconomic attracted by centers 𝐴 and 𝐵, 𝑃 𝑂𝑃𝐴 and 𝑃 𝑂𝑃𝐵 are populations of
characteristics of users, measurement of individual attitudes and a centers, and 𝑙𝐴 and 𝑙𝐵 represent distance of centers 𝐴 and 𝐵 from the
discrete choice experiment through the conducted face-to-face survey. intermediate location. In this modeling, attraction of a specific center
The work in [36] studies a comparative and mixed methods assessment depends on its population and distance from the intermediate location.
of the influence of gender, education, occupation, age, and household By applying Reilly’s law of retail gravity to the EV charging context,
size on the importance of EV charging time, EV public chargers, and CSs can be considered as the centers and EVs are the customers located
fuel economy preferences, drawing primarily on a survey across the in the intermediate points. Population of the centers in (1) should be
Nordic region. The authors in [37] investigated the importance of socio- replaced by the parameters of EV charging context which affects the
demographic attributes, geographic conditions, car interest, personal decision of the EV owners in choosing their desired CS. To have a
and social norms, and environmental concerns. Logistic regression proper consideration of effective parameters in attraction of a CS, the
analyses on the collected data through survey show how age, education basic attraction function which will be used for the proposed dynamic
level, and more importantly the individual’s personal norms impact pricing framework in this work is formulated in a modified form as
the EV travel behavior. Additionally, the work in [38] establishes follows.
the relationship between the public charging infrastructure and 21 𝑆𝑖 𝑃𝑖
𝐴𝑇 𝑇𝑖𝑗𝑡 = ( )2 (2)
socio-demographic, technical, and economic factors using mixed-effect 𝑡
𝑃 𝑟𝑖 𝑑𝑖𝑗𝑡
regression method and according to the governmental sources and press
articles data. This equation represents 𝐴𝑇 𝑇𝑖𝑗𝑡 , which is the attraction of 𝐶𝑆𝑖 to
On the other hand, scenario factors include the battery’s state the 𝑗th EV at time 𝑡. 𝑆𝑖 in this equation is the total number of electric
of charge (SOC), charging cost, range anxiety, etc. SOC reflects the charging plugs in the 𝐶𝑆𝑖 , 𝑃𝑖 is the rating power of the chargers in
vehicle’s remaining energy and is considered to be one of the most im- 𝐶𝑆𝑖 corresponding to the charging level of the CS, 𝑃 𝑟𝑡𝑖 is the electricity
portant factors. Price is another critical factor that EV travelers are con- price of 𝐶𝑆𝑖 at time slot 𝑡, and 𝑑𝑖𝑗𝑡 is the distance between the 𝑗th EV
cerned about. In normal situation, people prefer a low price [39]; how- and 𝐶𝑆𝑖 at time 𝑡. Here, it is assumed that the CS parameters in (2)
ever, some EV owners are time-conscious who are willing to pay more are shared with the EV users through navigation systems, mobile apps,
and save time by visiting the uncongested CSs. This paper is aimed at or online maps. In addition, it is assumed in this study that the CS is
specifically developing a quantitative charging behavior model, which the final destination of the EV at the time of decision-making, and no
includes the effective parameters of CSs in EVs charging choice, as changes are made in the CS selection in the proposed model after an EV
discussed in the next section. user selects a CS candidate. Also, the time and energy cost related to
the traveling from the current location to the selected CS is reflected in
2.2. CS attraction function and coverage zone the attraction function by parameter 𝑑𝑖𝑗𝑡 , which is the distance between
the 𝑗th EV and 𝐶𝑆𝑖 at time 𝑡. This distance is calculated by the QGIS
In order to encourage EVs to use the more uncongested CSs for that provides a realistic distance according to the fastest realistic routes
charging and to consequently increase the CSOs revenue, it is necessary rather than considering the direct distance between the two points
to have a proper understanding of how EVs individually select their which may not be a realistic estimate. The parameters included in the

4
A. Kazemtarghi et al. International Journal of Electrical Power and Energy Systems 158 (2024) 109946

Fig. 1. Competition of two CSs to attract an EV for charging.

attraction function in (2) are aligned with the EV user’s welfare, while
a higher rating power of the charger corresponds to the lower charging
time and a higher number of charging plugs in the CS introduces higher
chance of achieving a faster charging. Competition between two CSs
to attract an EV is illustrated in Fig. 1. The attraction boundary point
in this figure is the intermediate location where the attractions of the
two CSs to the EV at this point are equal. The boundary point can be
calculated by (3) as presented below:
𝑙𝐴 + 𝑙𝐵
𝑙𝐴 = √ (3)
𝑆𝐵 𝑃𝐵 𝑃 𝑟𝐴
1+ 𝑆𝐴 𝑃𝐴 𝑃 𝑟𝐵

where 𝑙𝐴 and 𝑙𝐵 are the distance from the attraction boundary point Fig. 2. EV charging process modeling flowchart.

to the 𝐶𝑆𝐴 and 𝐶𝑆𝐵 , respectively. The boundary points determine the
coverage zone of CSs. Larger coverage zone of CS increases the chance
of CS to attract more EVs. Considering the fixed number of charging their desired CS according to the attraction function in (2) and select
plugs and charger rating power for CSs, Eq. (3) demonstrates that the the CS with the highest attraction as their charging choice. If there is
coverage zone of CS depends on the electricity price. any available plug at the desired CS, the EV is assigned to a plug and
In a specific scenario in Section 5, it is assumed that the CSs are the charging end time is calculated by (6); otherwise, the EV should
equipped with an online crowd meter that means the EV owners can wait in line for an occupied plug to be available. If there is not any
check the availability of the electric plugs at different CSs and make available plug until the maximum waiting time 𝑡𝑤,𝑚𝑎𝑥 , the EV leaves
decisions based on this additional information. In this case, proposed the CS.
attraction function is modified as expressed below: At the end of each time slot, the total sold energy and corresponding
𝑡
𝑆𝑖 𝑃𝑖 𝑁𝑎𝑣,𝑖 CSO revenue is calculated based on the number of occupied plugs,
𝐴𝑇 𝑇𝑖𝑗𝑡 = ( )2 (4) electricity price, and charging power level of the CSs. The process
𝑃 𝑟𝑡𝑖 𝑑𝑖𝑗𝑡 iterates for all the time slots under study, and finally the overall results
are published.
𝑡
where 𝑁𝑎𝑣,𝑖 is the number of available charging plugs of 𝐶𝑆𝑖 at time 𝑡. The dynamic pricing strategy formulated in Section 3 is aimed
Moreover, the CS boundary point in this case is calculated by at managing the EV routes with the presented charging process to
𝑙𝐴 + 𝑙𝐵 minimize the number of EVs waiting at the CSs and conduct EVs from
𝑙𝐴 = √ (5) congested CSs toward the uncongested ones, which eventually increases
𝑆𝐵 𝑃𝐵 𝑁𝑎𝑣,𝐵 𝑃 𝑟𝐴
1+ 𝑆𝐴 𝑃𝐴 𝑁𝑎𝑣,𝐴 𝑃 𝑟𝐵 the total revenue of CSO.

2.3. EV charging process model 3. Dynamic pricing strategy to maximize total CSO revenue

Charging process modeling of the EVs during the time horizon under With the goal of utilizing potential flexibility of the charging price
study is proposed in this section as presented in Fig. 2. The model starts of CSs, this section presents the dynamic pricing strategy formulation,
with initializing CSs data, status of plugs, maximum acceptable waiting developed objective function, applied optimization method, and rele-
time, and time horizon under study. Then, all the plugs in CSs are vant constraints. The main purpose is to optimally control the charging
first checked at the current time slot 𝑡∗ to see if any EV has completed price of CSs to encourage EV owners to use uncongested CSs, which
charging. This is done by comparing the calculated charging end time increases both EV customers’ welfare by minimizing the waiting time
(𝑡𝑒𝑛𝑑,𝑖𝑘 ) in (6) with the current time slot (𝑡∗ ) for all the plugs indexed and CSO total revenue by maximizing total sold energy. This study
by 𝑘 at all the CSs indexed by 𝑖. considers the case that all the CSs are governed by one single entity,
say a CSO and the objective is to maximize the net profit sum obtained
(𝑆𝑂𝐶𝑒𝑛𝑑,𝑘 − 𝑆𝑂𝐶𝑠𝑡𝑟,𝑘 ) ∗ 𝐶𝑏𝑎𝑡𝑡,𝑘
𝑡𝑒𝑛𝑑,𝑖𝑘 = (6) from all the CSs under that CSO. So, we have a single player and there
𝑃𝑖 is not any competition; hence, the Nash equilibrium is not applicable
In this equation, 𝑆𝑂𝐶𝑒𝑛𝑑,𝑘 and 𝑆𝑂𝐶𝑠𝑡𝑟,𝑘 represent the end and start in the considered study. Due to the random nature of EV models,
SOC of the EV connected in the charging plug 𝑘, respectively. 𝐶𝑏𝑎𝑡𝑡,𝑘 is start and end SOC, arrival times, and location of EVs when they want
the capacity of the EV battery, and 𝑃𝑖 is the average charging power to choose the target CS to charge, the problem is formulated by a
of the 𝑖th CS. If the EV is charged completely, the corresponding plug scenario-based stochastic optimization method, where the worst-case
status will be updated to ‘‘available’’. Next, the model checks if any scenario is considered in each case study in order to have a conservative
EV needs to be charged at the current time 𝑡. The arriving EVs choose approach toward the problem and results. This consideration leads to

5
A. Kazemtarghi et al. International Journal of Electrical Power and Energy Systems 158 (2024) 109946

Fig. 3. Communications in the proposed dynamic pricing strategy.

robust scenario-based results where the minimum performance of the where 𝜔𝑟 is the worst-case scenario existing in the ambiguity set 𝑊 . The
proposed dynamic pricing strategy corresponding to the worst scenario objective function in (8) is subject to the EV charging process model
is compared with the fixed pricing strategy. Also, it is possible to use presented Section 2.3 and charging price limits as follows.
a probability distribution for scenario selection; however, due to high
uncertainties related to the EVs arrival time, start and end SOC, and 𝑃 𝑟 ≤ 𝑃 𝑟𝑡𝑖 ≤ 𝑃 𝑟 ∀𝑖, 𝑡 (9)
EV battery capacity, it is not guaranteed to find such an accurate The minimum charging price is typically set to cover the costs
distribution. Therefore, a robust scenario-based approach is adopted in of providing the service including electricity cost, infrastructure cost,
this work. maintenance cost, taxes, and profits. On the other hand, the maximum
Fig. 3 shows the communications between CSs, EVs, and CSO within charging price is mainly determined by electricity market competition
the dynamic pricing strategy. According to this figure, CSs share their and consumer demand. In some cases, market regulators may set a cap
information on the charging power, total number of charging plugs, on charging prices to safeguard consumers against price gouging.
number of available charging plugs at time 𝑡, and the charging price at
time 𝑡 with the EV owners. EV owners select their desired CS according
3.2. Optimization method
to the developed attraction function in (2). On the other hand, CSO
monitors the occupancy of the CSs at time 𝑡 and sets the charging prices
The constrained optimization problem in (8) is formulated as fol-
for the following time slot (𝑃 𝑟𝑡+1 ∀𝑖) accordingly in an optimal fashion
𝑖 lows by replacing the variables 𝑃 𝑟𝑡𝑖 with 𝑥.
to maximize the revenue and increase the EV owners’ satisfaction by
reducing their waiting time and offering cheaper charging prices. min −𝑓 (𝑥) (10a)
𝑥

3.1. Scenario-based stochastic optimization for dynamic pricing strategy ℎ(𝑥) = 0 (10b)
𝑔(𝑥) ≤ 0 (10c)
EVs are attracted to the CSs according to the CS parameters pro-
posed in attraction function in (2) and based on the EV charging process where 𝑓 , ℎ, and 𝑔 are twice continuously differentiable functions. Using
model, presented in Section 2.3. Here, the objective is to determine an interior method, the nonlinear program in (10) can be replaced by
the optimum set of CS charging prices (𝑃 𝑟𝑡𝑖 ) that leads to the CSs to a sequence of barrier subproblems in the form of (11) as follows.
support the maximum EVs during the whole time under study. In other
∑
𝑚
words, dynamic pricing of CS changes the attraction of CSs over time min 𝜑𝜇 (𝑧) ≡ −𝑓 (𝑥) + 𝜇 ln 𝑠𝑖 (11a)
so that EVs are encouraged to use cheaper CSs, which are probably 𝑧
𝑖=1
uncongested CSs as well; hence, CSO can sell more electricity and its
ℎ(𝑥) = 0 (11b)
revenue is maximized. The revenue of 𝐶𝑆𝑖 at time 𝑡 is calculated based
on the electricity price, number of occupied charging plugs, and the 𝑔(𝑥) + 𝑠 = 0 (11c)
charging power, as expressed in (7), and the objective function of the
Here, 𝑠 > 0 is a vector of slack variables, 𝑧 = (𝑥, 𝑠), and 𝜇 > 0 is
scenario-based stochastic optimization problem is defined in (8) for all
the barrier parameter. The Lagrangian function associated with (11) is
CSs and time slots accordingly as follows.
defined by
𝑓𝑖𝑡 = 𝑃 𝑟𝑡𝑖 𝑁𝑜𝑐𝑐,𝑖
𝑡
𝑃𝑖 (7)
(𝑧, 𝜆; 𝜇) = 𝜑𝜇 (𝑧) + 𝜆𝑇ℎ ℎ(𝑥) + 𝜆𝑇𝑔 (𝑔(𝑥) + 𝑠) (12)
∑
𝑇 ∑
𝑀
𝑡,𝜔𝑟 𝑡,𝜔 where 𝜆ℎ and 𝜆𝑔 are Lagrange multipliers and 𝜆 = (𝜆ℎ , 𝜆𝑔 ). The first-
max 𝑓= 𝑃 𝑟𝑖 𝑁𝑜𝑐𝑐,𝑖𝑟 𝑃𝑖 ∀𝜔𝑟 ∈ 𝑊 (8)
𝑃 𝑟𝑡𝑖 order optimality conditions for the barrier problem in (13) can be
𝑡=1 𝑖=1

6
A. Kazemtarghi et al. International Journal of Electrical Power and Energy Systems 158 (2024) 109946

Table 2 that the primary component that the holidays/weekdays related test-
Pricing Strategies Under Study.
case distinctively takes care of is the aspect of different EV charging
Case Pricing strategy
demand profiles in the analysis, which would not alter any funda-
number
mental part of the CS selection strategy and the charging process
I Fixed pricing
model developed in Section 2. Furthermore, the peak and off-peak
II Peak and off-peak pricing
III Proposed dynamic pricing framework pricing strategy defined as case II investigates the similar impact as the
IV Proposed dynamic pricing + CSs equipped with crowd meter holidays/weekdays analysis has on the EV charging demand profile and
consequently on the charging pricing, number of charged EVs, total sold
energy, and finally total revenue of the CSO.
The EV charging process model is simulated according to the
expressed by
[ ] [ ] flowchart presented in Fig. 2 over a 1-day time horizon with 5-minute
∇𝑓 (𝑥) + 𝐴ℎ (𝑥)𝑇 𝜆ℎ + 𝐴𝑔 (𝑥) 𝜆𝑔 0 steps corresponding to 288 time slots. The frequently variations of
= (13)
𝑆𝛬𝑔 𝑒 − 𝜇𝑒 0 charging price will increase the operation difficulty of the CS operator
together with (11b), (11c) and restriction that 𝑠 and 𝜆𝑔 be nonnegative. and also may cause inconvenience to the EV owners. On the other
Here 𝑆 and 𝛬𝑔 denote diagonal matrices with the entries given by hand, lower frequency of the price updating prevents the CS operator
vectors 𝑠 and 𝜆𝑔 , respectively, and 𝐴ℎ , 𝐴𝑔 are the Jacobian matrices to take advantage of the dynamic pricing strategy fully to maximize the
of ℎ and 𝑔. revenue by more uniform distribution of the EVs among different CSs.
Applying Newton’s method to the system (13), (11b), and (11c), In this study, the frequency of the price updating (i.e., every 5 min)
from the current iterate (𝑧, 𝜆) results in the primal–dual system pre- is selected according to the farthest possible CS selection of the area
sented below: under study. It means that if an EV selects the farthest CS which has
[ ][ ] [ ] the highest attraction (through offering very low charging price, high
𝑊 (𝑧, 𝜆; 𝜇) 𝐴(𝑥)𝑇 𝑑𝑧 ∇ (𝑧, 𝜆; 𝜇)
=− 𝑧 (14) charging power, or a greater number of available charging plugs), the
𝐴(𝑥) 0 𝑑𝜆 𝑐(𝑧)
EV selects that corresponding location as the final destination without
where any change in price on the way. Therefore, the time interval of price
[ ] [ ] [ ]
𝑑 𝑑ℎ ℎ(𝑥) updates is selected based on the maximum time required for an EV to
𝑑𝑧 = 𝑥 , 𝑑𝜆 = , 𝑐(𝑧) = (15)
𝑑𝑠 𝑑𝑔 𝑔(𝑥) + 𝑠 travel to the farthest CS with higher attraction, which is 5 min in this
case study. Once a CS selection is made, the choice is locked in and
[ ] cannot be altered along the way.
𝐴ℎ (𝑥) 0 Number of 12 CS equipped with Level-2 chargers with an average
𝐴(𝑥) = (16a)
𝐴𝑔 (𝑥) 𝐼 charging power of 7.2 kW located in Tempe, Arizona, USA is considered
[ 2 ]
∇𝑥𝑥 (𝑧, 𝜆; 𝜇) 0 with the number of charging plugs presented in [41]. Further, to
𝑊 (𝑧, 𝜆; 𝜇) = (16b) investigate the impact of charging level in this study, 4 more DC-fast
0 𝑆 −1 𝛬𝑔
CSs with the average charging power of 50 kW are included. To have a
The new iterate is given below using step lengths 𝛼𝑧 and 𝛼𝜆 .
realistic understanding of CS locations and EV routes, it is necessary to
𝑧+ = 𝑧 + 𝛼𝑧 𝑑𝑧 , 𝜆+ = 𝜆 + 𝛼𝜆 𝑑𝜆 (17) use a geographical tool. QGIS software with high capabilities is used
for this purpose. Based on the location of the CSs on the map and
To validate the efficacy of the proposed dynamic pricing scheme, according to the number and random location of the EVs at time 𝑡,
performance evaluation is presented in the next section that includes the fastest routes of EVs to the CSs are obtained, as shown in Fig. 4,
the case study, simulation results, and related discussion.
and finally distance matrix 𝐷𝑡 is formed that includes the distance of
individual EVs to each CS. Then, these distances are used to calculate
4. Performance evaluation of developed dynamic pricing strategy
the attraction of each CS to the EV users according to the developed
model in (2), which subsequently is used in the charging process model
This section presents the case studies to validate the efficacy of the
and developed dynamic pricing framework. The EVs demand profile
proposed dynamic pricing strategy. In addition, a realistic model of
pattern used in this work is according to the realistic profile presented
CS locations and EV routings is presented using geographical modeling
in [42], as illustrated in Fig. 5. According to the EV demand profile,
tools. Further, simulation results and relevant discussion are presented.
hours 8 to 15 are assigned to the peak hours and the other hours
are considered as the off-peak hours in the peak and off-peak pricing
4.1. Scenarios and simulation setting
strategy defined in case II. Furthermore, To have a realistic modeling of
To assess the performance of the proposed CS dynamic pricing EV battery capacities, ten common EV models along with their battery
strategy, four pricing strategies are studied in this work according to capacity and corresponding full charging time for level-2 and DC-fast
Table 2. Case I introduces a fixed pricing strategy with the electricity charging systems are presented in Table 3.
price equal to the average charging price of the dynamic pricing in Incorporating traffic situations into the dynamic pricing strategy
case III. Case II represents the peak and off-peak pricing as the most offers the potential for more accurate estimations of charging demand
fundamental dynamic pricing method, which shows how the EV users and user behavior, but it comes with several challenges. The complexity
respond to the changes in charging prices, but in only two price levels of the model increases significantly due to the need to account for vari-
defined according to the EV charging load demand. Case III investigates ous factors influencing traffic conditions, such as time of day, weather,
the basic form of the developed dynamic pricing strategy where the and special events. Obtaining and managing the extensive data required
charging price of CSs are varying dynamically over the time with the for accurate modeling can be resource-intensive, and frequent updates
objective of maximizing total revenue of the CSO, as presented in (8). may strain computational resources, which is a challenge in real-time
Finally, case IV considers an additional assumption of the CSs equipped implementation. Additionally, uncertainty in traffic patterns poses a
with an online crowd meter, which affects the attraction function of the challenge, as unexpected events can disrupt travel times and route
CS and consequently the decision of the EV users on the selected CS to choices. Implementing dynamic pricing based on traffic conditions may
visit, as expressed in (4). also raise regulatory, ethical, and competitive concerns, necessitating
Another impactful factor in the dynamic charging pricing problem is careful consideration of the trade-offs involved in adopting such a
the difference between weekday and holidays. It is worth mentioning pricing strategy.

7
A. Kazemtarghi et al. International Journal of Electrical Power and Energy Systems 158 (2024) 109946

Table 3
EV Models Characteristics.
No. EV Model Capacity Level-2 DC-fast
(kWh) full charging full charging
time (h) time (h)
1 Tesla Model S 100 13.8 2
2 Tesla Model X 100 13.8 2
3 Tesla Model 3 50 6.9 1
4 Tesla Model Y 68 9.4 1.36
5 Chevrolet Bolt EV 65 9 1.3
6 Nissan Leaf 40 5.5 0.8
7 Ford Mach-E 70 9.7 1.4
8 Audi e-tron 95 13.2 1.9
9 Porsche Taycan 79 10.9 1.58
10 Hyundai Kona 64 8.9 1.28

slots that leads to the total revenue of $2,061. The charging price in
case I is fixed for all CSs and during all time slots. In case II, the number
of charged EVs is increased to 864, which results in 14% more sold
energy in comparison with the case I. The CSO revenue in this case is
$2,371, and the average waiting time of the EVs is reduced to 15 min
that is 4 min less the case I. It is observed that the proposed dynamic
pricing method in case III can increase the number of charged EVs to
1,046 through changing the charging price of the CSs and consequently
manipulating the attractions of the CSs to the customers to encourage
them to use the uncongested CSs. The total sold energy is increased
from 23.56 MWh in case II to 28.64 MWh in case III corresponding
to the total revenue of $2,917. Additionally, case IV is defined such
that CSs share more information with EV users through the online
crowd meter that shows the online number of available electric plugs
at each CS. This additional information causes more EVs to visit the
uncongested CSs and increases the number of charged EVs to 1,128
corresponding to the 29.57 MWh sold energy and $3,069 revenue for
the CSO.
From the CS revenue perspective, the high attraction CSs (e.g., CSs
equipped with DC fast chargers or CSs with a high number of charging
plugs) can keep a high charging price with fully occupied charging
plugs to maximize the revenue, while the other CSs need to reduce
their charging price to attract more EVs. The added revenue in this
case is due to the increased number of charged EVs, as presented in
Fig. 4. CS locations and EV routes modeling using QGIS; (a) CS and EV locations at
Table 4. This leads to both more sold energy and more revenue for the
a specific time; (b) fastest EV routes to 𝐶𝑆1 that is used to form the distance matrix.
CS operator and enhanced satisfaction of the EV owners resulted by
lower waiting times.
The other benefit of the proposed dynamic pricing strategy is en-
hanced satisfaction of EV owners through reducing the average waiting
times resulted by distributing the congestion of the CSs, as illustrated
in Table 4. A maximum waiting time is considered in the EV charging
process modeling that shows EV owners will leave the CS after this
waiting time. The number of EVs that leave the CSs after the defined
maximum waiting time reflects the customer satisfaction level of the
charging systems infrastructure. From Table 4, the average waiting time
of the EVs is reduced from 19 min for case I to 9 min for case IV,
reflecting the great satisfaction of EV owners thanks to the proposed
dynamic pricing strategy. Furthermore, Fig. 6 shows the number of
EVs left the CSs during the peak time slots between 𝑡 = 84 to 𝑡 = 180
corresponding to the hours 8 to 15. It is observed that the proposed
dynamic pricing strategy in cases III and IV can significantly reduce
Fig. 5. EVs demand profile pattern.
the number of departed EVs by changing the charging prices and
consequently distributing the EVs toward the uncongested centers.
The visual distribution of EVs at the specific time slot 𝑡 = 137
4.2. Results and discussion
for different case studies is illustrated in Fig. 7, which demonstrates
the impact of the proposed dynamic pricing strategy in terms of in-
To evaluate the performance of the proposed dynamic pricing strat- creased number of charging EVs and a more uniform distribution of
egy, detailed results and discussion are presented in this section. Ta- EVs among different CSs. This is a scaled figure where the bigger EV
ble 4 shows the overall simulation results of the four pricing strategies. icon corresponds to a greater number of charging EVs at the relevant
In the fixed pricing scenario (case I), 758 EVs are charged over all time CS. Comparison of all cases gives the result that each case leads to an

8
A. Kazemtarghi et al. International Journal of Electrical Power and Energy Systems 158 (2024) 109946

Fig. 6. Number of EVs departed the CSs after maximum waiting time during the peak load correspondent time slots.

Fig. 7. Scaled distribution of charging EVs at 𝑡 = 137 in different case studies: (a) Case I; (b) Case II; (c) Case III; (d) Case IV.

Table 4 Moreover, Table 5 demonstrates details on the impact of dynamic

Overall Simulation Results.
pricing method on distribution of EVs between CSs. In case III, CS
Case No. of Waiting Sold energy Revenue of
charging prices are in the range of 5 to 15 ¢∕𝑘𝑊 ℎ at 𝑡 = 137 while
no. charged EVs time (min) (MWh) the CSO ($)
high-attraction CSs corresponding to the higher number of charging
I 758 19 20.61 2,061
plugs and higher average charging power of chargers have the highest
II 864 15 23.56 2,371
III 1,046 11 28.64 2,917 price that increases the revenue of the CSO. On the other hand, keeping
IV 1,128 9 29.57 3,069 the price at a higher level decreases the attraction of the CS to EVs that
decide to charge at the next time slots, as observed for the CSs 12 and
13. Further, comparing the prices at 𝑡 = 137 and 𝑡 = 138 shows that
CSs 1, 2, 6, 8, 9, 10, and 11 have attracted more EVs by reducing their
enhanced EV distribution and number of charging EVs than the pre-
electricity price and consequently increasing the attractiveness of their
vious case. The more uniform distribution of EVs in cases III and IV is
service. Moreover, CSs 14, 15, and 16 that provide the DC-fast charging
achieved thanks to the proposed dynamic pricing strategy and changing service with lower charging times are among high-attraction CSs that
the charging prices to conduct the EVs toward the uncongested CSs can keep higher prices while the EV users are still satisfied and select
and thereby enhancing satisfaction level of EV owners by lower waiting these centers as their charging choice. Regarding the case I, because
times. of the fixed pricing approach, the changes in attractiveness of the CSs

9
A. Kazemtarghi et al. International Journal of Electrical Power and Energy Systems 158 (2024) 109946

Table 5
Impact of Dynamic Pricing on Distribution of EVs.
CS Case III Case I
no.
𝑃 𝑟137
𝑖
137
𝑁𝑜𝑐𝑐,𝑖 𝑃 𝑟138
𝑖
138
𝑁𝑜𝑐𝑐,𝑖 137
𝑁𝑜𝑐𝑐,𝑖 138
𝑁𝑜𝑐𝑐,𝑖
(¢/kWh) (¢/kWh)
1 9.07 0 7.77 1 0 0
2 15 3 13.56 4 3 3
3 14.99 7 15.00 8 8 8
4 5.01 0 5.00 0 0 0
5 14.98 4 14.99 4 0 0
6 7.72 1 5.38 7 0 0
7 5.00 0 5.03 0 0 0
8 15 16 14.51 27 16 17
9 10.15 2 8.28 3 0 1
10 14.99 0 7.03 1 0 0
11 15.00 2 5.01 3 0 0
12 13.40 6 11.40 4 0 0
13 14.51 2 9.07 1 0 0
14 15.00 2 15.00 2 2 2
15 15.00 4 15.00 4 3 3
16 15.00 3 15.00 5 2 2

during the time are only due to the varying distances of the EVs to
the CSs, which does not cause any remarkable change in the attraction
of the CSs, so it is observed that the EVs are distributed between the
same CSs at both time slots. In other words, EV owners do not see any
benefit to visit the other centers because of the fixed pricing approach,
which reduces both EV owners welfare and CSO revenue. Also, a more
uniform distribution of the EVs resulted by dynamic pricing framework
can be concluded from the total number of charged EVs at a specific
time slot for each case; at 𝑡 = 137 slot, 34 EVs are charged in all CSs
while this number increases to 52 EVs in case III.
Fig. 8 represents the impact of dynamic pricing strategy in case III
on competition between CSs to attract the EVs by comparing the CS
coverage zones and their boundary points. It can be observed from
Fig. 4 that 𝐶𝑆5 has 6 neighbors including CSs number 4, 6, 7, 8, 12, and
16. To have a better illustration, the boundary point (and consequently
coverage zone) shifts between 𝐶𝑆5 and its neighboring CSs are shown
in Fig. 8(a) in an hourly manner during the day and corresponding
dynamic charging prices along with the EV charging load demand are
illustrated in Fig. 8(b). For example, it is observed that the charging
price of 𝐶𝑆6 at ℎ = 4 is decreased, which increases the attraction and
as a result the coverage zone of this CS is widened. This means that
a greater number of EVs located in the routes between 𝐶𝑆5 and 𝐶𝑆6
during the ℎ = 4 are willing to visit 𝐶𝑆6 for charging rather than 𝐶𝑆5
due to the higher price of 𝐶𝑆5 . The similar results can be concluded
during ℎ = 4 for interactions between 𝐶𝑆5 and 𝐶𝑆12 . The charging
price for 𝐶𝑆12 is reduced during this hour to attract more EVs. Also,
it is observed from Fig. 8(a) that the impact of this price reduction is
reflected in the distance from 𝐶𝑆12 to the boundary point between 𝐶𝑆5
and 𝐶𝑆12 . In this situation, more EVs located between 𝐶𝑆5 and 𝐶𝑆12
are attracted by 𝐶𝑆12 in comparison to the previous time slot due to
the reduced charging price of this CS during the fourth hour.
Similarly, the impact of dynamic pricing strategy in case IV on the
competition between 𝐶𝑆5 and its neighboring CSs to attract more EV Fig. 8. Impact of dynamic pricing strategy in case III on CS boundary points; (a)
users are illustrated in Fig. 9(a). Also, the dynamic charging prices variations in distance from the 𝐶𝑆5 to the boundary points respect to its neighboring
during the day and the EV load demand are shown in Fig. 9(b). For CSs; (b) dynamic charging prices of 𝐶𝑆5 and its neighboring CSs along with the EV
instance, comparing the distance from 𝐶𝑆5 and 𝐶𝑆8 from the boundary charging demand.
point between these CSs gives the idea that 𝐶𝑆8 has a larger coverage
zone rather than 𝐶𝑆5 all over the day. The reason is that 𝐶𝑆8 in this
study has the greatest number of 32 charging plugs, which notably the hours is the 𝐶𝑆16 . The reason for this observation is that the 𝐶𝑆16
increases the attraction of this CS to the EV users according to the is equipped with the DC-fast chargers, which significantly reduces the
attraction function in (2). Therefore, this CS maintains a relatively
charging time, and this is an attractive feature to the EV users; hence,
high charging price during the day to result more revenue to the CSO;
this CS has a higher attraction, larger coverage zone, and consequently
although it is observed that this CS has reduced its charging price in
a few hours such as ℎ = 2 and ℎ = 8 to keep the attraction in a high a longer distance to the boundary point in comparison with the 𝐶𝑆5 all
level. The other CS that has a larger coverage zone than 𝐶𝑆5 during all over the day.

10
A. Kazemtarghi et al. International Journal of Electrical Power and Energy Systems 158 (2024) 109946

Fig. 10. Comparison of the level-2 and DC-fast charging occupancy percentage during
all time slots.

with level-2 charging is about 25% at the peak time, while more than
75% the DC-fast chargers are occupied over the peak demand time. This
study shows that DC-fast charging is a promising technology, which
leads to both high customer satisfaction and more revenue for the CSO.

5. Conclusions

This work studies the impact of dynamic EV charging pricing strat-

egy on distribution of EVs between CSs. A comprehensive mathematical
model is presented in this paper to show how the charging choice of EVs
is determined and impacted according to the number of electric plugs,
charger types, distances from CS, and charging prices of CS. Then, this
model is integrated into a dynamic pricing framework, formulated as a
scenario-based stochastic optimization to determine the optimum set of
charging prices for each CS over a range of time slots to maximize the
overall revenue of CSO. CS locations and EV routes are modeled using
QGIS software to consider the realistic routes and distances from EVs
to different CSs. Four scenarios of charging pricing strategies including
fixed pricing, peak and off-peak pricing, developed dynamic pricing
framework in this work, and the proposed dynamic pricing strategy
with CSs equipped with online crowd meters are defined and compared
in terms of total CSO revenue, customer satisfaction, and congestion
of the CSs. The results demonstrate that the proposed dynamic pricing
method can increase the total number of charged EVs up to 37.9% and
the total revenue of the CSO by 41.5% in comparison with the fixed
pricing strategy. Moreover, EV users adjust their selection differently
when the CSs are equipped with the online crowd meter, which leads
to ∼3.2% increase in the amount of energy sold and CSO revenue
based on the presented case study. Further, results show that the
proposed dynamic pricing strategy improves the user satisfaction level
by reducing waiting times at congested CSs. Also, it is observed that CSs
equipped with the DC-fast chargers are more attractive to EV users so
that they can increase the revenue of CSO by 79%, and reduce charging
times of EVs by 7 times on average.

Declaration of competing interest

Fig. 9. Impact of dynamic pricing strategy in case IV on CS boundary points; (a)
variations in distance from the 𝐶𝑆5 to the boundary points respect to its neighboring
CSs; (b) dynamic charging prices of 𝐶𝑆5 and its neighboring CSs along with the EV
The authors declare that they have no known competing finan-
charging demand. cial interests or personal relationships that could have appeared to
influence the work reported in this paper.

Data availability
To analyze the impact of EV charging level, Fig. 10 is presented
to compare the plugs occupancy percentage of CSs equipped with the The data that has been used is confidential.
level-2 and DC-fast chargers over all time slots for the dynamic pricing
strategy in case III. It is observed that although a limited number of Acknowledgments
DC-fast chargers are included in this study, CSs with DC-fast chargers
are more attractive to the EV users due to their lower charging times as This material is based upon research supported in part by the U. S.
an important factor to the customers. Maximum occupancy rate of CSs Office of Naval Research under award number N00014-23-1-2347.

11
A. Kazemtarghi et al. International Journal of Electrical Power and Energy Systems 158 (2024) 109946

References [21] Lee C, Han J. Benders-and-Price approach for electric vehicle charging
station location problem under probabilistic travel range. Transp Res B
[1] Bayani R, Soofi AF, Waseem M, Manshadi SD. Impact of transportation elec- 2017;106:130–52.
trification on the electricity grid—A review. Vehicles 2022;4(4):1042–79. http: [22] Zhang Y, You P, Cai L. Optimal charging scheduling by pricing for EV
//dx.doi.org/10.3390/vehicles4040056. charging station with dual charging modes. IEEE Trans Intell Transp Syst
[2] Kazemtarghi A, Dey S, Mallik A. Optimal utilization of bidirectional evs 2018;20(9):3386–96.
for grid frequency support in power systems. IEEE Trans Power Deliv [23] Li Z, Sun Y, Yang H, Anvari-Moghaddam A. A consumer-oriented incentive strat-
2022;38(2):998–1010. egy for EVs charging in multi-areas under stochastic risk-constrained scheduling
[3] Rezaei O, Mirzapour O, Panahazari M, Gholami H. Hybrid AC/DC pro- framework. IEEE Trans Ind Appl 2022.
visional microgrid planning model considering converter aging. Electricity [24] Wu T, Li G, Bie Z. Charging price determination and energy management of
2022;3(2):236–50. EV parking lot considering price elasticity. In: 2019 IEEE 8th international
[4] Li J, Wang G, Wang X, Du Y. Smart charging strategy for electric vehicles based conference on advanced power system automation and protection. IEEE; 2019,
on marginal carbon emission factors and time-of-use price. Sustainable Cities Soc p. 1789–93.
2023;104708. [25] Erdinç O, Erenoğlu AK, Şengör İ, Taştan İC, Büyük AF, Catalão JP. A smart
[5] Gong L, Cao W, Liu K, Zhao J. Optimal charging strategy for electric vehicles dynamic pricing approach for electric vehicle charging in a distribution system.
in residential charging station under dynamic spike pricing policy. Sustainable In: 2020 9th international conference on power science and engineering. IEEE;
Cities Soc 2020;63:102474. 2020, p. 30–5.
[6] Limmer S. Dynamic pricing for electric vehicle charging—a literature review. [26] Luo C, Huang Y-F, Gupta V. Stochastic dynamic pricing for EV charging
Energies 2019;12(18):3574. stations with renewable integration and energy storage. IEEE Trans Smart Grid
[7] Li Y, Wang J, Wang W, Liu C, Li Y. Dynamic pricing based electric ve- 2017;9(2):1494–505.
hicle charging station location strategy using reinforcement learning. Energy [27] Moghaddam Z, Ahmad I, Habibi D, Masoum MA. A coordinated dynamic
2023;281:128284. pricing model for electric vehicle charging stations. IEEE Trans Transp Electr
[8] Lee S, Choi D-H. Dynamic pricing and energy management for profit maximiza- 2019;5(1):226–38.
tion in multiple smart electric vehicle charging stations: A privacy-preserving [28] Zhang Q, Hu Y, Tan W, Li C, Ding Z. Dynamic time-of-use pricing strategy
deep reinforcement learning approach. Appl Energy 2021;304:117754. for electric vehicle charging considering user satisfaction degree. Appl Sci
[9] Dai Y, Qi Y, Li L, Wang B, Gao H. A dynamic pricing scheme for electric vehicle 2020;10(9):3247.
in photovoltaic charging station based on Stackelberg game considering user [29] Moghaddam V, Yazdani A, Wang H, Parlevliet D, Shahnia F. An online reinforce-
satisfaction. Comput Ind Eng 2021;154:107117. ment learning approach for dynamic pricing of electric vehicle charging stations.
[10] Lee W, Schober R, Wong VW. An analysis of price competition in heterogeneous IEEE Access 2020;8:130305–13.
electric vehicle charging stations. IEEE Trans Smart Grid 2018;10(4):3990–4002. [30] Cedillo MH, Sun H, Jiang J, Cao Y. Dynamic pricing and control for EV charging
[11] Lu Y, Liang Y, Ding Z, Wu Q, Ding T, Lee W-J. Deep reinforcement learning-based stations with solar generation. Appl Energy 2022;326:119920.
charging pricing for autonomous mobility-on-demand system. IEEE Trans Smart [31] Tan J, Liu F, Xie N, Guo W, Wu W. Dynamic pricing strategy of charging station
Grid 2021;13(2):1412–26. based on traffic assignment simulation. Sustainability 2022;14(21):14476.
[12] Lu C, Wu J, Wu C. Privacy-preserving decentralized price coordination for EV [32] Liu Y, Zhu J, Sang Y, Sahraei-Ardakani M, Jing T, Zhao Y, et al. An aggregator-
charging stations. Electr Power Syst Res 2022;212:108355. based dynamic pricing mechanism and optimal scheduling scheme for the electric
[13] Lu Z, Shi L, Geng L, Zhang J, Li X, Guo X. Non-cooperative game pricing strategy vehicle charging. Front Energy Res 2023;10:1037253.
for maximizing social welfare in electrified transportation networks. Int J Electr [33] Wang Y, Yao E, Pan L. Electric vehicle drivers’ charging behavior analysis
Power Energy Syst 2021;130:106980. considering heterogeneity and satisfaction. J Clean Prod 2021;286:124982. http:
[14] Yu Y, Su C, Tang X, Kim B, Song T, Han Z. Hierarchical game for networked //dx.doi.org/10.1016/j.jclepro.2020.124982.
electric vehicle public charging under time-based billing model. IEEE Trans Intell [34] Cheng L, Chen X, Yang S, Wu J, Yang M. Structural equation models to
Transp Syst 2020;22(1):518–30. analyze activity participation, trip generation, and mode choice of low-income
[15] Han Y, Zhang X, Zhang J, Cui Q, Wang S, Han Z. Multi-agent reinforcement commuters. Transp Lett 2019;11(6):341–9.
learning enabling dynamic pricing policy for charging station operators. In: 2019 [35] Soto JJ, Márquez L, Macea LF. Accounting for attitudes on parking choice: An
IEEE global communications conference. IEEE; 2019, p. 1–6. integrated choice and latent variable approach. Transp Res A 2018;111:65–77.
[16] Wu B, Zhu X, Liu X, Jin J, Xiong R, Wu W. Revenue maximization of electric [36] Sovacool BK, Kester J, Noel L, de Rubens GZ. The demographics of decarbonizing
vehicle charging services with hierarchical game. In: International conference on transport: The influence of gender, education, occupation, age, and household
wireless algorithms, systems, and applications. Springer; 2021, p. 417–29. size on electric mobility preferences in the [nordic] region. Global Environ
[17] Dong X, Mu Y, Xu X, Jia H, Wu J, Yu X, et al. A charging pricing strategy Change 2018;52:86–100.
of electric vehicle fast charging stations for the voltage control of electricity [37] Westin K, Jansson J, Nordlund A. The importance of socio-demographic char-
distribution networks. Appl Energy 2018;225:857–68. acteristics, geographic setting, and attitudes for adoption of electric vehicles in
[18] Li K, Shao C, Zhang H, Wang X. Strategic pricing of electric vehicle charging Sweden. Travel Behav Soc 2018;13:118–27.
service providers in coupled power-transportation networks. IEEE Trans Smart [38] Haidar B, Rojas MTA. The relationship between public charging infrastructure
Grid 2022. deployment and other socio-economic factors and electric vehicle adoption in
[19] Chen S, Feng S, Guo Z, Yang Z. Trilevel optimization model for competitive France. Res Transp Econ 2022;95:101208.
pricing of electric vehicle charging station considering distribution locational [39] Daina N, Sivakumar A, Polak JW. Electric vehicle charging choices: Modelling
marginal price. IEEE Trans Smart Grid 2022. and implications for smart charging services. Transp Res C 2017;81:36–56.
[20] Kong W, Ye H, Wei N, Xing D, Chen W. Dynamic pricing based EV load [40] Wagner WB. An empirical test of Reilly’s law of retail gravitation. Growth Chang
management in distribution network. Energy Rep 2022;8:798–805. 1974;5(3):30–5.
[41] Alternative fuels data center, URL https://afdc.energy.gov/.
[42] Islam MS, Mithulananthan N. Daily EV load profile of an EV charging station at
business premises. In: 2016 IEEE innovative smart grid technologies-Asia. IEEE;
2016, p. 787–92.

12