Electric Vehicle Charging: A Business Intelligence Model

Abstract

The adoption of electric vehicles (EVs) has grown substantially in recent years, offering a cleaner and highly promising pathway toward the decarbonization of urban environments. However, this trend introduces new challenges in charging infrastructure and management. This paper proposes a synergistic integration of Business Intelligence (BI) and Artificial Intelligence (AI) techniques—including machine learning and data analytics—for solving the EV charging problem. We begin with an in-depth analysis of charging behaviors, leveraging extensive datasets from EVs, charging stations (CSs), and auxiliary sources. Based on this analysis, we introduce a BI framework utilizing advanced data mining methods to utilize large-scale data effectively. We then present a BI-based decision-making model that enables comprehensive analysis and optimized solutions for EV charge scheduling and the cooperation among different CS owners. The model is validated across multiple real-world scenarios and case studies, demonstrating significant improvements in charging efficiency, utilization, and reliability. By showcasing the practical applications of BI-driven analytics, our findings underscore the transformative impact of data-informed methodologies on EV charging operations. This paper concludes with a discussion of open research opportunities in AI- and BI-driven intelligent transportation—specifically in EV charging optimization, grid integration, and predictive analytics.

1. Introduction

In recent decades, there have been tremendous advancements in wireless and communications technology, which have led to the development of the Internet of Things (IoT). IoT includes the interconnection of smart devices and the exchange of information among them. The analysis of the collected data through Artificial Intelligence (AI) technology and machine learning techniques enables faster and more efficient decisions. The use of IoT is extended and affects all the aspects of everyday life; however, in this paper, our focus is on electric vehicle (EV) technology [1,2].

1.1. Motivation

The market of EVs is growing at a fast pace due to their advantages. Electric vehicles contribute to the decarbonization of road transport, noise-free traffic, the decrease in operational and maintenance costs [3,4], and the convenience of charging [5]. In addition to being pollution-free and ecologically efficient, EVs frequently use renewable resources and positively contribute to the transportation system of smart cities. However, EV charging poses significant challenges to the power grid network. In order to charge the increasing number of EVs, huge amounts of energy are needed, especially during particular times of the day, when large numbers of EV owners want to charge their vehicles. If charging occurs without control and scheduling, various problems appear, including a limited number of charging stations (CSs), slow charging, and long queues and waiting times. Real-world evidence highlights that unmanaged EV charging can lead to local grid congestion, voltage instability, and transformer overloading. For example, in Norway, rapid EV adoption in urban centers has necessitated urgent upgrades to distribution substations to prevent service interruptions [6]. The authors in [7] examine peak hourly loads for each month in California, and patterns of charging are investigated. The collaboration of many vehicles and charging stations might result in notable energy savings, particularly during periods of peak load. The cooperation concept is one of the main contributions of this paper. The primary problem is to create an incentive-compatible and energy-efficient market that maximizes the benefits of the market participants and allocates energy efficiently in a market model while taking into account individual interests of EVs and CSs, and the overall market gains.

The implementation of the proposed BI-driven energy trading and EV charging model is closely linked to the regulatory landscape and energy market structures in place. In the European Union, ongoing initiatives such as the Clean Energy for All Europeans [8] package and the EU Green Deal [9] actively support the development of energy communities, local energy markets, and peer-to-peer energy trading. According to EU Directive (EU) 2019/944 [10] on common rules for the internal electricity market, Citizen Energy Communities are legally recognized and encouraged to operate generation, consumption, storage, and sharing of energy. These frameworks create legal space for the kind of cooperative, decentralized model proposed in this paper, where EVs and CSs could function as active market participants, exchanging energy services based on local demand and digital intelligence. Several EU member states (e.g., the Netherlands, Italy, Spain, Greece, etc.) have launched pilot projects and living labs (https://www.wisegrid.eu/) exploring similar models, often supported by blockchain platforms, smart meters, and AI-based scheduling tools. Thus, the proposed method could be well aligned with the emerging structure of local energy systems in the EU, particularly when embedded in energy communities or city-scale mobility platforms. Future work should explore regulatory integration and incentive-compatible pricing models to enable smooth adoption across different jurisdictions.

1.2. Contributions

In this paper, we explore an energy trading market where a number of EVs are in motion in a geographical area, where a number of CSs, owned by different stakeholders, are deployed. The EVs need to charge their batteries with varying demands, and the stakeholders offer energy to sell at different prices. Thus, all the participants of the market are independent and have competitive interests. Additionally, a great challenge of the market includes the ignorance of the preferences, needs, decisions, and actions of each of the participants. More specifically, this market characteristic is very crucial during peak demand time periods, since imbalances are created and even delays and queues at the charging stations. These problems are energy inefficient, and create unnecessary costs and time waste for all the participants of the market. Thus, if we assume that the electric vehicles could be motivated to truthfully disclose their demands, a decision-making policy could be employed to optimize the operation of the market. An operator is responsible for gathering the large volume of information. The abundance of information has to be processed in order to be useful and to be utilized in the decision-making process. With the gathered information, an energy model is proposed to deal with the different desires and needs of the participants, and an equilibrium can be reached.

Given the aforementioned issues and the economic nature of the market, the tools of AI, namely machine learning and data analytics, are exploited, and a Business Intelligence (BI) model is proposed. Machine learning and data analytics are appropriate tools when a large amount of information is in hand. The conflicting interests of the participants involved in the market make the proposal of a BI model suitable for the problem. In this paper, a BI energy trading strategy is proposed. The system layout, time-varying energy characteristics (e.g., peak and off-peak periods), and the unique features of market participants (e.g., electric vehicles and charging stations) are all taken into consideration in the strategy.

The main contributions of this paper are as follows:

• We present an innovative system layout, where EVs, owned by distinctive individuals, compete with one another to buy energy from different CSs that are deployed by different stakeholders.
• We summarize the data that are employed in an energy trading market. It is important to identify the sources of spatial and time data that are processed in the BI model.
• We propose a Business Intelligence model. More specifically, we analyze the different layers that the BI is composed of, including the data layer, the BI layer, and the decision layer.
• We present optimal solutions for addressing a number of energy charging problems, including the charging schedule recommendation, the cooperation among different CS stakeholders, and optimal infrastructure planning. Various strategies are investigated, including double auction strategies and iterative approaches.

The remainder of this paper is as follows. In Section 2, we provide an overview of different charging strategies that are investigated in the literature. In Section 3, we describe the system model that we use in our work along with its operation. Additionally, the data sources needed in the charging algorithm are presented. Section 4 focuses on the Business Intelligence model that we propose. More specifically, a number of challenges and solutions are proposed and explained. The performance evaluation of the proposed solutions is included in Section 5, and a discussion of the results is given in Section 6. Finally, Section 7 concludes this paper.

2. Related Work

The rapid growth of the EV market has directed the attention of the research community towards the advancement of the CS networks and the energy grid, in general. Since the emergence of EVs, charging is among the most discussed issues. For the purposes of this paper, we selected a set of works of the literature based on the following criteria:

• Relevance to the integration of BI and AI for EV charging, particularly methods that leverage data analytics, predictive modeling, and optimization.
• Recency, with preference for works published within the last 5–7 years, to reflect current technological and policy contexts.
• Methodological significance, including works that introduced or refined concepts directly applicable to our approach, such as dynamic pricing, congestion mitigation, and multi-agent coordination.
• Impact and credibility, prioritizing studies published in high-impact journals, reputable conference proceedings, or influential technical reports.

A number of works focused on the different charging technologies [11,12,13,14,15,16,17]. The authors portray the technologies, standards, topologies, and methodologies for charging EVs and discuss the challenges, advancements, limitations, and open issues of the energy grid.

However, the emerging use of EVs has created a great deal of challenges concerning EV charging and CS infrastructure deployment. These two topics have been thoroughly investigated in the literature. As EV traffic has increased through time, charge scheduling has become a difficult problem to be solved, as delays and queues were observed at CSs. In ref. [18], the authors addressed the problem of charge scheduling and classified the algorithms into unidirectional or bidirectional and centralized or distributed. The economic aspect of the different approaches was examined in [19], and different approaches were investigated in [20]. Due to the complexity of the charging problem, the researchers explored a number of optimization solutions. An overview of optimization scheduling techniques is discussed in [21]. A heuristic linear programming method is discussed in [22], and the authors in [23] formulate a dynamic programming algorithm for addressing the charge scheduling problem. Continuously, works [24,25,26] propose an optimal charge scheduling scheme, where different optimization objectives are picked, such as charging costs, charging duration, and charging demand balancing.

Other works focus on identifying the optimal deployment of charging station infrastructure. A comprehensive review of the strategies examining the optimal sites of CSs is provided in [27]. Mixed linear programming is used for CS planning [28]. A number of optimization methods for determining the best location and capacity of charging stations [29,30,31]. Fewer state-of-the-art works exploit the benefits of AI in order to specify the optimal locations of CSs [32,33]. Clearly, there are not many works that use AI techniques. Thus, this is the focus of our paper.

Additionally, in our prior approach [34], we proposed a charging solution. This work introduces several key advances. First, we incorporate multi-modal behavioral data analytics from EV users, derived from GPS and telematics streams, into the charging assignment process. Second, our model unifies BI and AI techniques, enabling both long-term demand forecasting and real-time, heuristic-based optimization. Third, we extend the framework to enable energy trading between charging stations via a double auction mechanism. Finally, we provide a computational performance comparison with traditional exact optimization methods, demonstrating the scalability of the proposed approach for real-time deployment. These enhancements position our work as an operationally viable and data-rich extension of earlier models.

3. System Model

In this section, we introduce the system model that we use in our work. Additionally, we explore the variable types of data characteristics involved in EV charging.

3.1. System Model and Operation

The system model of our work is depicted in Figure 1. We assume an urban geographical area, where a number of CSs are installed by multiple stakeholders and energy providers, and a number of EVs are moving within the range of the area. More specifically, we define the participants with the following characteristics:

• EVs: We assume that there are N vehicles in the geographical area, denoted by $E{V}_n$, where $n\in \mathcal{N}=\{1,\dots ,N\}$. The EVs move around the area and are in search of CSs in order to charge their batteries. The vehicles are equipped with telematic systems and sensors to make their locations visible and provide useful data in the system.
• CSs: We assume that there is a set of M CSs, denoted by $C{S}_m$, where $m\in \mathcal{M}=\{1,\dots ,M\}$. Each CS acts as an aggregator in order to provide the ability for EV owners to communicate with the power grid. Additionally, fundamental data are provided to the system through the CSs, including locations, energy levels, and availability, in order to be exploited in the BI model.

The communication between EVs and CSs facilitates the vehicles that have individual interests to participate in the energy trading market. EVs declare their needs for energy, while CSs state their availability.

3.2. Spatial Data

Spatial data plays a critical role in the charging of EVs. Understanding the geographical distribution of both EVs and CSs allows the identification of high-demand areas and the reduction in charging congestion. Moreover, spatial data enables the strategic placement of CSs based on their density, the traffic patterns, and the accessibility of locations [35]. The integration of spatial data into BI models allows balanced energy distribution, efficient infrastructure utilization, and enhanced user satisfaction.

3.2.1. Location Data of Electric Vehicles

The knowledge of the exact location of EVs is important in the decision-making process and predictive analysis. EVs are equipped with GPS and telematics systems to generate continuous data streams. The data are exploited to understand the driving behavior, the travel patterns, and the real-time location of EVs [36]. A BI model uses the data to recommend the proximal CSs, to predict energy demand, and to identify the charging hotspots. Thus, tracking EV locations provides valuable insights into charge scheduling. Beyond simply tracking location, GPS data streams collected from EVs are processed to derive time-series information on speed, acceleration, and route patterns. These are used to estimate behavioral indicators such as driving aggressiveness (e.g., frequent hard accelerations or decelerations), route regularity (e.g., predictable commutes vs. exploratory driving), stop-and-go frequency, particularly in urban traffic, and average trip duration and speed variance. These behavioral metrics are then used as input features in energy consumption forecasting models, allowing the BI system to personalize charging recommendations based on historical driving style. For instance, more aggressive drivers may receive earlier or more frequent charging prompts due to faster battery depletion.

3.2.2. Selection of Charging Station Location

Decisions regarding CS infrastructure are critical. The use of EVs is increasing at a fast pace, and among the disadvantages of their use is the problem of CS location. If the CSs are not located in the best locations, EV owners may be disappointed, and this problem may affect their motivation to use their EVs. Surely, the CSs cannot be placed randomly. Many works in the literature have investigated the problem of selecting the optimal location of the CSs. Consequently, CS site election is a multiple-criteria problem. Recent empirical work [37] illustrates how user behavior and spatial demand patterns can guide smart charging infrastructure planning, especially when integrated with grid-aware clustering strategies (e.g., supporting local energy district optimization). In a comprehensive data-driven study, the authors in [38] analyze EV charging behaviors across Ireland, revealing substantial peak evening charging at home and frequent fast-charger use. These findings underscore the potential grid stress during peak demand periods and the importance of integrating BI-aware incentives to reshape user behavior.

To begin with, the first feature that is considered when choosing the location for the CS deployment is the demographic profile of a geographical area. The demographic profile includes the population of an area and its density, the number of EV owners, and the number of CS owners that compete in the same geographical area. A second factor that should be considered in the selection of CS location is the daily activity profiles of EV owners. More specifically, it is important to examine the daily needs and visits around the geographical areas. Locations such as shopping malls, supermarkets, restaurants, university campuses, public service locations, and industrial and business locations are identified as suitable through the integration of demographic analytics and mobility data, which help to highlight areas of high dwell time and frequent visits. Third, the traffic profile of EV owners is important to be investigated (time variance of the traffic is examined in the next section). CSs can be installed in areas where there is large footfall. For example, charging stations need to be deployed in highways, busy intersections, and commuter routes, since EVs are highly concerned with charging their batteries while traveling long distances. Last but not least, an important upgrade will contemplate how delays can be avoided, especially in highly congested areas.

In addition to accessibility, visibility, and proximity to demand centers, the suitability of a location also depends on the charging technology deployed. For example, low-power AC chargers (3–22 kW) are well suited for long-dwell contexts such as office buildings, residential complexes, and park-and-ride facilities, where vehicles remain parked for extended periods. Conversely, high-power DC fast chargers (50–150 kW) or ultra-fast chargers (>150 kW) are essential for short-dwell, high-throughput locations such as highway rest stops, logistics hubs, or major transit corridors. Including technology–location compatibility in the site selection criteria ensures that infrastructure deployment meets both user needs and grid efficiency objectives.

3.2.3. Charging Station Density

Along with the selection of CS location, their density is crucial. The investigation of both number and proximity of charging stations is thoroughly studied in the state-of-the-art works. For example, even if we have a good number of CSs in a geographical area, if they are apart from each other, problems might appear. Similarly, it is not convenient to gather a small or a large number of CSs in a limited geographical area and have other areas with no CSs.

First and foremost, there is a critical factor that must be considered in the CS density decision. In specific geographical areas and at certain times of the day, EV owners may experience high congestion and long waiting times when they want to charge their EVs. This may lead to lower satisfaction, disappointment, and frustration. Thus, it is important to be able to predict the charging demand in urban and rural areas at different periods of the day. Over and above that, the accessibility of the areas should be studied thoroughly in order to understand how many CSs could be installed so as to fulfill the energy demands of EV owners. Even in areas with optimal CS placement, inadequate density can lead to congestion or underutilization. A balance must be struck between accessibility and distribution equity, informed by spatial analytics and real-time usage patterns. Summarizing, the charging duration of an electric vehicle is, moreover, a determining factor that affects the deployment of CS infrastructure. When delays can be avoided, if there are adequate CSs in an area, the life and experience of EV owners is facilitated.

The determination of optimal station density is also influenced by the charging speed and service profile of each CS. High-power DC hubs generally require fewer sites with higher throughput, strategically positioned to serve large catchment areas and minimize detour distances. In contrast, slower AC chargers are often deployed in greater numbers within dense urban or suburban areas to serve localized, long-duration charging needs. By coupling the demand density model with technology-specific suitability rules from Section 3.2.2, the BI framework can recommend not only where to place stations but also what type of technology each site should host.

3.3. Time Data

Time data is crucial for understanding and predicting fluctuations in EV charging demand throughout the day, week, and year. It enables the identification of peak and off-peak periods and supports load balancing. Incorporating temporal patterns into BI models helps optimize energy distribution and reduce grid congestion during high-demand time periods.

The charging demands vary dynamically over time. The charging profiles and patterns have been investigated in depth in the literature. A large number of researchers investigate how the charging demands differentiate according to different times of the day. Peak and off-peak times of CS use are very important in CS infrastructure planning. Based on the reports, it is observed that EV owners prefer to start charging their vehicles in the afternoon or early evening. In most cases, charging increases at midnight and thereafter decreases in the morning. Similar behavior is observed during workdays and weekends. However, during the weekends, the charging demand is slightly lower than in workdays. When comparing the charging energy demand during the different periods of the year, we notice that monthly consumption is larger in winter than in summer; however, the charging profile is similar, meaning that most of the EVs are charged in the afternoon and in the evening [39].

4. Business Intelligence Model

In this section, we analyze the Business Intelligence model that is studied in this paper. We describe the different aspects of the BI model concerning data, and we make a comprehensive analysis of EV charging problems, and we demonstrate the various solutions that can be proposed.

4.1. System Overview

In the last decades, an explosion of data has been observed. Large-scale datasets are available. In practice, companies and researchers have a large amount of structured or unstructured data in diverse formats. However, the available data are useless if they are not organized, analyzed, and reported. In this direction, data mining and analytics tools are employed in order for the diverse data to gain value and be transformed into information. Techniques such as artificial intelligence, machine learning, and statistics are used to create patterns and profiles for the categorization and clustering of data. BI models include these techniques. The reported data lead to improved data-driven decisions, and the solutions that are presented are designed to provide improved performance for the users. Finally, BI models offer data visualization tools, which convert data into charts or graphs to make it easier for any key stakeholders or decision-makers to better understand the solutions and to implemented them. The aforementioned procedures are divided into three layers: data layer, BI model layer, and decision layer, as shown in Figure 2. The layers are described in detail in the next subsections.

4.2. Data Layer

The electric vehicle network produces a vast amount of data. Following, we discuss the sources of the data:

• Data from EVs: The electric vehicles are often equipped with multiple sensors. The data that are gathered include the type of vehicle, the location of the EVs, the direction of the route, the levels of battery health, the energy consumption, the urgency or not of energy charging, and the vehicle and driver performance. Additionally, historical data are gathered, e.g., typical charging times, preferred charging locations, preferred routes, or planned trips. As technology advances and AI is used, information about predicted routes based on EV behavior can be provided to help the BI model. The raw data need to be filtered, categorized in order to be easily accessed, and employed.
• Data from CSs: Data are available from the CSs as well. Important information include the type of CSs, the location of the CSs, the parking locations, their availability for charging, the real-time waiting queues, the potential use of renewable energy sources, charging speed, and the cost of energy charging (e.g., flat rates, dynamic pricing). Information about ownership can also be used since the management of CSs through third parties affects both pricing and accessibility. These data are crucial for decisions about charging schedules and infrastructure planning, if they are classified and integrated correctly.
• Data from Geographic Information Systems (GISs) and other applications: GISs are hardware and software systems that collect, manage, analyze, and visualize geographical data. Both spatial and temporal data are available. The amount and the variety of data from GISs are enormous; nonetheless, we are interested in data that concern the location of roads and railroads, traffic zones, buildings locations, and even temperature. The analysis of the data provides valuable information about the mobility patterns of EVs and infrastructure planning of the CSs.
• Other data: Other sources of data include surveys and questionnaires that are addressed to EV owners and CS stakeholders. Additionally, the use of applications by the involved users and parties could provide real-time information and statistics. Other information are gathered concerning the weather conditions, since they affect energy demand, traffic peak, off-peak conditions, and renewable energy generation. Regulation and law constraints are also important. Based on the data, we are able to understand the performance of our system and adjust its operation to the needs and demands of EV users and CS owners.

The data layer integrates diverse information sources—real-time sensor data from EVs and CSs, geospatial data from GIS systems, and user-generated data from surveys or applications, using a multi-stage data pipeline. Key components include the following:

• ETL (Extract, Transform, Load) procedures to ingest data from APIs, files, and streams.
• Schema harmonization, where raw inputs are mapped to a unified data model, resolving naming conflicts, missing values, and unit differences (e.g., converting GPS coordinates, normalizing battery level formats).
• Time-series synchronization, aligning event logs from different sources using timestamps to ensure coherence.
• Geospatial alignment, using GIS metadata to assign spatial attributes and enable spatial joins between EV traces and CS locations.
• Data validation and imputation, including handling outliers, filling missing values using interpolation or statistical models, and removing duplicated records.

This integration process enables the BI model to operate on a unified, clean, and semantically consistent dataset, which supports real-time analytics, predictive modeling, and optimization tasks across the charging ecosystem.

The effectiveness of the proposed BI framework depends on the integration of multiple heterogeneous data streams, including EV telemetry, CS operational data, geographic information system layers, and grid status indicators. In practical deployment scenarios, however, access to all such datasets by a single stakeholder is often constrained by commercial confidentiality, privacy regulations, and institutional data silos. For instance, CS operators typically possess detailed operational statistics for their own infrastructure, yet cannot directly access proprietary EV telemetry from competing networks. Similarly, EV manufacturers may collect telematics and locational data from their user base, but generally lack comprehensive visibility of charging infrastructure outside their own service network. Municipalities and grid operators, while having access to certain traffic and network datasets, may have limited or no access to private CS usage records.

To address these limitations, the framework is designed to function in environments with partial data availability, leveraging a combination of open, aggregated, and synthetic datasets to complement proprietary information. Relevant open data resources include public charging infrastructure datasets, such as Open Charge Map, crowdsourced charging platform data, including PlugShare and ChargeMap, open traffic and GIS datasets, such as OpenStreetMap, which provide road network topology, traffic congestion patterns, and points of interest relevant for siting analysis and synthetic or simulated datasets, generated using established demand models and behavioral assumptions, to represent realistic charging demand patterns in the absence of empirical records.

4.3. BI Model Layer

The raw data that are gathered are then processed in order to find solutions to a number of problems concerning the energy charging of EVs. Following, we describe the challenging issues that the BI layer deals with:

• Demand balancing: One crucial challenge in electric vehicle charging is the problem of balancing the charging load across the multiple charging stations that are scattered across the geographical area. Charging load balancing is connected to the peak demand management with the ultimate goal of minimizing the congestion at certain CSs and to divert charging demand away from the peak hour of the day. Another objective is to exploit renewable energy resources. The most common approaches in the literature are used to route EVs individually in order to balance the demand. However, these solutions are not mature, and the demand balancing has limited prospects. Another set of state-of-the-art works explores the load balancing demand from the perspective of EV routing and choosing the most suitable parking lot for charging the vehicle [40,41,42,43]. Other works focus on the prediction of EV demand patterns and the analysis of energy demand fluctuations [44,45,46,47]. The proposition of a BI system enables the exploitation of the wide range of information that is available and from different resources in order to provide improved solutions. To begin with, BI employs the techniques of data analysis and mining in order to organize raw data, classifies them, and finds useful information. Using this data, charging patterns are identified and peak time demands and spatial congestion points are recognized. The time and spatial behavior of EVs is then evaluated and effectively analyzed. The main objective of demand prediction is to be able to identify the high-demand time windows and anticipate the CSs use in order to avoid overcrowding. Moreover, understanding charging preferences and habits motivates the recommendation for charging schedules based on EV behavior. The BI model is able to predict the charging demand in a specific urban region based on EV density and nearby charging stations. ARIMA and Long Short-Term Memory can be used for energy charging forecasting.
• Energy scheduling and control management: It is common that the owners of electric vehicles choose the time that they charge their vehicles based on their work and duties or based on the battery levels of their car. Therefore, there is no coordination between EV owners and the CSs to manage convenient charging. However, uncoordinated charging has negative impacts on the energy grid that feeds the CS infrastructure (e.g., energy deficiencies and fluctuations) and on the satisfaction of the vehicle owners (e.g., long waiting times). A number of works in the literature deal with the charge scheduling problem. In the majority of the works, the problem is considered an optimization problem, where a central controller manages the system and decides the optimal schedule for the CS stakeholder [48,49,50,51,52,53,54,55,56]. Fewer works investigate the problem as a decentralized technique [57,58,59,60,61]. Concerning the objective goals, the time and load fluctuations are considered. Additionally, there is a need for balancing energy charging so as to eliminate the peaks. Finally, the benefits of the CS stakeholder are in focus. More specifically, the maximization of the utilization of CSs, the reduction in queuing times, and the efficient distribution of EVs are the main objectives of the CS owners. From our point of view, there is a great need to take into account the demands and needs for all the involved parties (e.g., CS owners, EV owners, energy providers). Additionally, the demand patterns and the mobility of the vehicles could be exploited. Thus, a BI model considers the offered data at their best. Next, a holistic technique identifies the objectives that may be contradicting and tries to reach the optimal solution for everyone involved. Clustering models, linear regression, and optimization tools can be used.
• Cost optimization: Following the aforementioned challenge, it is important to investigate the charging cost for both the EV and CS owners. Several works have addressed the specific challenge and proposed solutions whose focus was charging cost [62,63,64,65,66,67]. It is important to use the BI tools to proposed strategies for dynamic pricing that will motivate the owners of EVs to cooperate and charge their vehicles at specific time periods so as not to create congestion and delays. Additionally, the research of optimal solutions is enabled through the use of a Business Intelligence model. Machine learning and data mining can be explored. In our framework, cost optimization refers to the minimization of the total service cost for both EV owners and CS stakeholders, which includes the following:

  −

  Energy cost based on charging pricing models (e.g., dynamic or time-of-use pricing);

  −

  Travel cost, proportional to distance between EVs and CSs;

  −

  Delay cost, measured by waiting time due to congestion.

  The multi-objective optimization problem is inherently nonlinear and NP-hard, especially under real-time constraints and heterogeneous preferences. To solve this practically, we integrate BI and AI as follows:

  −

  BI layer identifies behavioral patterns, demand forecasts, and historical congestion data to inform the optimization.

  −

  Decision layer applies heuristic optimization (e.g., greedy algorithms, local search, rule-based systems) that can produce near-optimal results quickly.

  −

  For more complex scenarios (e.g., large urban areas), metaheuristic techniques such as genetic algorithms or multi-agent reinforcement learning can be employed to balance objectives under constraints.

  These approaches are computationally efficient, make use of real-time BI insights, and are capable of adapting dynamically to shifting demand and pricing conditions.

4.4. Decision Layer

Following, we propose the BI models to address different problems related to EVs and their charging. We provide solutions for both electric vehicle owners and CSs. In order to fully understand the context of our work, we analyze the scenario that is explored. Referring to Figure 1, we consider a scenario with a large number (N) of EVs and a large number (M) of CSs. EVs are moving through a wide geographical area where CSs are located. Both EVs and CSs are equipped with sensors and communication systems in order to share information about battery levels, energy demand, location, availability, and charges. We further assume that EV and CS owners agree to cooperate using a BI model with the goal of finding an overall optimal solution, without denigrating their own demands and desires, however.

• Charging schedule recommendation: Deciding the charging schedule is very important for the improvement of our system performance. More specifically, it is important for the EVs to decide dynamically when and where to charge and not decide randomly simply based on their proximity to a CS and the battery level. Additionally, CSs can play a crucial role when they cooperate with each other and with EVs. This is where a BI model is employed. A dynamic schedule using optimization tools can minimize the cost and waiting times of the involved counterparts while at the same time load balance is achieved across the CS by taking into account the proximity of EVs and CSs, charging speed, and energy cost. An application can notify EV owners of the optimal time and location of charging. The objectives of the optimization problem are, among others, the following: (a) the minimization of waiting time and the travel distance of EVs, (b) the maximization of CS utilization, and (c) the minimization of overloading the energy grid during peak hours. Linear programming, multi-agent simulations, and game theory are employed. In our previous work, we proposed an innovative market formulation in which autonomous EVs and CSs are motivated to cooperate dynamically with changing roles. We adopted a multi-objective strategy that is repeated in steps [34]. Another formulation of the optimization problem considers the minimization of the cost of both EVs and CSs, as follows:

  $$\begin{array}{c}\hfill \mathbf{P}\mathbf{1}:min\sum _{i=1}^N\sum _{j=1}^M\left(\alpha ·d_{i,j}+\beta ·t_{i,j}+\gamma ·c_j\right)·x_{i,j}\end{array}$$

  (1)

  where $d_{i,j}$ is the distance between $E{V}_i$ and $C{S}_j$, $t_{i,j}$ is the estimated waiting time of $E{V}_i$ at $C{S}_j$, and $c_j$ is the charging cost per energy unit at $C{S}_j$. The binary variable $x_{i,j}$ indicates the assignment of the $E{V}_n$ at $C{S}_j$ (value 1 indicates the assignment, 0 is when $E{V}_i$ is not assigned). The coefficients $\alpha$, $\beta$, and $\gamma$ are weight parameters for distance, waiting time, and cost, respectively. The optimization problem is minimized under some constraints as follows:
  Each EV should be assigned to one CS,

  $$\sum _{j=1}^M{x}_{i,j}=1,\forall i\in \mathcal{N}$$

  (2)
  Each CS has a limited number of available charging slots, $q_j$, for EVs to charge,

  $$\sum _{i=1}^N{x}_{i,j}\le q_j,\forall j\in \mathcal{M}$$

  (3)
  The total energy charged to EVs at the CSs ($b_{i,j}$ is the charged energy for $E{V}_i$ at $C{S}_j$) should not exceed the maximum grid energy level (namely P),

  $$\sum _{j=1}^M\sum _{i=1}^N{x}_{i,j}·b_{i,j}\le P,\forall i\in \mathcal{N},\forall j\in \mathcal{M}$$

  (4)
  The optimization problem mentioned above is a mixed integer linear programming approach and can be easily solved for medium-sized instances like 100 EVs and CSs. However, solving the charging schedule recommendation problem becomes computationally difficult in large-scale and real-time scenarios. Thus, we adopt an efficient greedy weighted matching approach that approximates the optimal assignment with significantly lower computational complexity. For the greedy approach, described in Algorithm 1, we compute the cost for each EV–CS pair, based in the weighted sum of travel distance, expected wait time, and energy cost. Next, each EV is assigned to a CS (the one with the lowest score), among those with available capacity. The algorithm is iterative for all EVs. The complexity of the algorithm is $O(N·M·\log M)$, making it suitable for real-time applications.

  Algorithm 1 Greedy charging schedule recommendation algorithm
  1:	Set of EVs: $\{E{V}_1,\dots ,E{V}_N\}$	▹ Input parameters
  2:	Set of CSs: $\{C{S}_1,\dots ,C{S}_M\}$
  3:	Distance from $E{V}_i$ to $C{S}_j$: $d_{i,j}$
  4:	Estimated waiting time at $C{S}_j$: $t_{i,j}$
  5:	Cost per kWh at $C{S}_j$: $c_j$
  6:	Available charging slots at $C{S}_j$: $q_j$
  7:	Weight parameters $\alpha$, $\beta$, $\gamma$
  8:	For each $E{V}_i$ and $C{S}_j$, compute a score:	▹ Cost matrix computation
  9:	$Scor{e}_{i,j}=\alpha ·d_{i,j}+\beta ·t_{i,j}+\gamma ·c_j$
  10:	For each $E{V}_i$	▹ Greedy assignment
  11:	Sort CSs by $Scor{e}_{i,j}$ ascending
  12:	Assign to the first $C{S}_j$ with available slot $q_j>0$
  13:	Update $q_j=q_j-1$
  14:	A mapping of EVs to CSs	▹ Output
  15:	Total system score ${\sum }_{i}Scor{e}_{i,j}$

• Cooperation among different CS owners: The second direction of the BI model concerns the motivation for the cooperation of the CS owners in order to optimally utilize the resources and increase the satisfaction of the electric vehicle owners. It is evident that the stakeholders that are in charge of deploying the CSs have contraindicative interests. However, a cooperative scheme ensures a fair profit distribution among all the participants of the charging market. After collecting the required data from the EVs (e.g., battery level, charging needs, location, direction of traveling), the CSs (e.g., location, availability, pricing, energy source), the traffic patterns (e.g., peak hours, travel paths), and other sources (e.g., surveys, applications), the data are compiled and processed in order to provide centralized and decentralized solutions, dynamic pricing approaches, load balancing models, and profit sharing mechanisms. More specifically, all the CS owners, or at least their majority, are motivated to share some of their data in a common platform or application. Through data processing and the use of AI algorithms, dynamic charging schedules are recommended. These schedules prevent overloading or under-utilization. By using machine learning strategies, we predict peak charging times, and we adjust pricing and availability accordingly by taking into account waiting times, costs, and locations. The BI model decides the charging schedule and then divides the profits fairly among the CS owners based on metrics and key indicators such as utilization and energy contribution. This scheme incentivizes the owners of the charging infrastructure to collaborate without losing their competitiveness. In a previous work, we proposed a double auction mechanism for a wireless network where network operators cooperate to share their traffic [68]. Similar solutions could be employed for the charging market.
  To facilitate the cooperation among independent CSs, we propose a BI-driven double auction mechanism that enables real-time, market-based coordination of charging demand. In this framework, CSs experiencing excess demand (i.e., overloaded stations) act as buyers (denoted by $\mathcal{B}$), submitting bids, $b_i,i\in \mathcal{B}$, to offload a portion of their incoming EV traffic. Let $d_i$ be the demand of buyer i, representing the number of EVs they wish to offload. Simultaneously, underutilized CSs act as sellers (denoted by $\mathcal{S}$), offering to accept redirected EVs in exchange for compensation, defined by their ask prices, $a_j,j\in \mathcal{S}$. We assume $s_j$ as the available capacity of seller j, which is the number of EVs they can accept. The BI system functions as a central auctioneer, matching bids and asks based on pricing compatibility and capacity availability. When a match is found, the participating CSs agree on a clearing price, typically the midpoint between the bid and ask, which ensures mutual benefit. This approach encourages load balancing, enhances overall infrastructure utilization, and creates economic incentives for collaboration, even in competitive environments.
  The use of a midpoint between bid and ask prices as the clearing price provides a tractable and incentive-compatible pricing strategy in two-sided markets. It balances the interests of both buyers and sellers by ensuring that trades occur within their acceptable value ranges. From a game-theoretic perspective, this approach satisfies individual rationality (i.e., no party trades at a loss), is budget-balanced, and encourages truthful bidding under mild assumptions of risk neutrality and quasi-linear utility functions. Although more advanced mechanisms like VCG auctions can ensure a dominant strategy, incentive compatibility, the midpoint rule remains widely used in practical multi-agent systems for its simplicity and approximation of Nash equilibrium behavior, especially in repeated or dynamic auction environments [69].
  The model is scalable, incentive-compatible, and adaptable to both energy and service-based cooperation scenarios. The allocation problem is to determine the optimal solution $x_{i,j},i\in \mathcal{B},j\in \mathcal{S}$ that maximizes the distinctive objectives of the involved parties in the auction, subject to constraints. The decision variable $x_{i,j}$ decides whether a buyer i is matched with a seller j ($x_{i,j}=1$) or not ($x_{i,j}=0$). We assume $q_{ij}$ is the number of EVs transferred from buyer i to seller j. The maximization problem, reflecting the gain from each matched trade, where $\left(b_i-a_j\right)$ is the net benefit and $q_{i,j}$ is the trade volume, is formulated as follows:

  $$\mathbf{P}\mathbf{2}:\max \sum _{i\in \mathcal{B}}\sum _{j\in \mathcal{S}}\left(b_i-a_j\right)·q_{i,j}$$

  (5)
              s.t.

  $$\begin{array}{cc}\hfill & \sum _{j\in \mathcal{S}}{q}_{i,j}\le d_i,\forall i\in \mathcal{B}\hfill \end{array}$$

  (6a)

  $$\begin{array}{cc}\hfill & \sum _{i\in \mathcal{B}}{q}_{i,j}\le s_j,\forall j\in \mathcal{S}\hfill \end{array}$$

  (6b)

  $$\begin{array}{cc}\hfill & x_{i,j}=0,\mathrm{if}{b}_i<a_j\hfill \end{array}$$

  (6c)

  $$\begin{array}{cc}\hfill & q_{i,j}\le M·x_{i,j},\forall i,j\hfill \end{array}$$

  (6d)

  $$\begin{array}{cc}\hfill & x_{i,j}\in \{0,1\},\forall i\in \mathcal{B},\forall j\in \mathcal{S}\hfill \end{array}$$

  (6e)

  $$\begin{array}{cc}\hfill & q_{i,j}\in {\mathbb{Z}}_{+},\forall i\in \mathcal{B},\forall j\in \mathcal{S}\hfill \end{array}$$

  (6f)
  Constraint (6a) ensures that each buyer CS i (who wants to offload EVs) cannot offload more EVs than it has in overflow, meaning that the total number of EVs redirected to all sellers from CS i must not exceed demand $d_i$. Constraint (6b) ensures that each seller CS j (who is willing to accept redirected EVs) does not accept more than its remaining charging capacity. The total number of EVs received from all buyers must be less than or equal to the station’s available slots $s_j$. With constraint (6c), a trade is disallowed between buyer i and seller j if the buyer’s bid is less than the seller’s ask. The constraint (6d) links the binary match variable $x_{i,j}$ with the number of EVs transferred $q_{i,j}$. If $x_{i,j}=0$ (no trade between i and j), then $q_{i,j}=0$. If $x_{i,j}=1$, then $q_{i,j}$ can be any value up to a large constant M. The constant M is a large positive integer used in the big-M formulation to enforce the coupling between the binary assignment variable $x_{i,j}$ and the continuous trade volume $q_{i,j}$. Practically, M should be set as an upper bound on the maximum number of EVs that could be traded between any buyer–seller pair. A safe choice is $M\ge \{ma{x}_{i\in \mathcal{B}}{d}_i,ma{x}_{j\in \mathcal{S}}{s}_j$}, where $d_i$ is the overflow demand of buyer i, and $s_j$ is the available capacity of seller j. This ensures that $q_{i,j}$ can only be positive when $x_{i,j}=1$, and zero otherwise. Choosing M too low may inadvertently restrict feasible trades, whereas excessively high values can weaken solver performance due to numerical instability. Therefore, M should be chosen conservatively based on known or bounded system parameters, such as the maximum expected EV demand or charging slot capacity in real-world deployment scenarios. Constraint (6e) is the binary decision variable, and constraint (6f) represents a non-negative integer representing the number of EVs traded. A greedy heuristic is a practical solution, which is fast, simple, and effective, to solve the maximization problem in Equation (5). The double auction approach is shown in Algorithm 2.

  Algorithm 2 Greedy double auction algorithm
  1:	Buyers: Each has a bid price $b_i$ and overflow $d_i$, $\forall i\mathcal{B}$	▹ Input parameters
  2:	Sellers: Each has an ask price $a_j$ and available capacity $s_j$, $\forall j\mathcal{S}$
  3:	Initialize $q_{i,j}=0$, $\forall i,j$	▹ Initialization
  4:	for Each buyer i in sorted list do
  5:	Set $demand\leftarrow d_i$
  6:	for For each seller j in sorted list do
  7:	if $b_i\ge a_j$ and $s_j>0$ then
  8:	Allocate: $q=\min \left(demand,s_j\right)$	▹ Allocation phase
  9:	Update:	▹ Update phase
  10:	$q_{i,j}\leftarrow q$
  11:	$demand\leftarrow demand-q$
  12:	$s_j\leftarrow s_j-q$
  13:	if $demand=0$ then
  14:	break
  15.	Return $q_{i,j}$ and optionally compute the gain:	▹ Output
  16:	$Totalgain={\sum }_{i\in \mathcal{B}}{\sum }_{j\in \mathcal{S}}\left(b_i-a_j\right)·q_{i,j}$

5. Case Study Scenarios and Results

Having described in detail the BI model, in this section, we outline how the model can be practically applied in the EV market. The practical paradigms prove the benefits of the proposed BI model. In this section, we present analytical results derived from simulated or real-world datasets, evaluating the performance of the model in terms of scalability, decision support accuracy, and system responsiveness.

The BI model enables strategic decision-making for a wide range of case studies and scenarios. By integrating the data of EVs (e.g., temporal energy charging demand curves, customer preferences, and geospatial data), the data from CSs (e.g., information about CS infrastructure, real-time charging data), and other data (e.g., surveys and questionnaires), the BI model provides solutions concerning the energy charging problems. Next, we present several scenarios that illustrate the importance of the application of the proposed model to common challenges.

5.1. Charging Schedule Recommendation

To evaluate the performance of the proposed greedy charging schedule recommendation Algorithm 1, we assume a real-world urban simulation scenario. There are six CSs with varying capacities and service characteristics in the area. More specifically, each CS is located in a $10\times 10$ km urban region, has a limited charging capacity (number of EVs it can serve concurrently), and requires estimated waiting times and dynamic cost for energy charging that varies with congestion. Waiting time refers to the queuing delay prior to charger availability; specifically, the interval between an EV’s arrival at the charging station and the initiation of the physical plug-in process. It does not include post-connection system initialization delays, which are generally minor. The waiting times (ranging from 5 to 20 min) are synthetically generated values based on a simulated urban environment. These estimates reflect typical congestion levels at charging stations with varying service capacities and are informed by EV arrival rate assumptions used in the case study. The goal is to model heterogeneous load conditions that a real-world deployment might encounter. Additionally, he simulation scenario involves 50 EVs that move in the investigated geographical area, requiring energy charging from the CSs. The simulation parameters of the CSs are summarized in

Table 1. Simulation parameters.

CS	Capacity	Location Coordinates	Wait Time (min)	Cost (EUR/kWh)
$C{S}_1$	10	$(2,5)$	5	$0.30$
$C{S}_2$	8	$(5,8)$	12	$0.25$
$C{S}_3$	12	$(8,3)$	10	$0.28$
$C{S}_4$	5	$(3,1)$	15	$0.22$
$C{S}_5$	15	$(6,6)$	8	$0.35$
$C{S}_6$	7	$(1,9)$	20	$0.20$

.

EVs are randomly distributed across the urban region and assigned to CSs based on the score function, denoted in Algorithm 1. We assume $\alpha =1.0$, $\beta =0.2$, and $\gamma =5.0$, and thus the score is focused on the cost. This assumption is viable, since the minimization of the charging cost is of high importance for EV owners. The weights used in the scoring function ($\alpha =1.0$ (distance), $\beta =0.2$ (waiting time), and $\gamma =5.0$ (cost)) were selected to reflect realistic EV owner priorities. Prior studies and user feedback consistently highlight cost sensitivity as the dominant factor in charging decisions, followed by travel distance and then waiting time. While no formal optimization was performed to derive these weights, we conducted parametric sensitivity tests to verify that the allocation results are qualitatively robust to moderate changes in parameter values. The selected configuration also ensured meaningful trade-offs in the simulated assignment outcomes, highlighting how the BI model responds to multi-criteria decision-making inputs.

The machine learning components of our BI model, such as charging demand prediction and user behavior modeling, were trained using an $80/20$ train-test split on the simulated dataset, ensuring that test samples were not seen during training. In addition, we performed 5-fold cross-validation to evaluate generalization and prevent overfitting, averaging results across folds for robustness. For temporal features, we ensured chronological integrity in splits to avoid data leakage.

To evaluate computational efficiency, we compared our BI-based greedy weighted matching algorithm against a standard Mixed-Integer Linear Programming (MILP) solver.

Table 2. Performance summary of charging schedule recommendation strategy.

Method	Average Cost (EUR)	Runtime (ms)	Scalability	Optimality
MILP Solver	$4.78$	$\mathrm{12,500}$	Poor—$O\left(n^3\right)$	Exact
Greedy Heuristic (BI)	$4.93$	180	Excellent	Near optimal (3–5% gap)

summarizes the performance comparison between a traditional MILP solver and the proposed BI-based greedy heuristic for the EV charging schedule assignment problem. The Average Cost (EUR) column represents the mean total cost per EV assignment (including distance, wait time, and charging cost) across all simulation runs. The Runtime (ms) column reports the average time taken to compute a solution, highlighting the real-time capability of the BI model. Scalability is qualitatively assessed based on observed runtime behavior as the number of EVs and CSs increases. Optimality reflects the solution quality: the MILP solver finds exact optimal solutions but at high computational cost, while the greedy heuristic provides near-optimal solutions much faster (within 3–5%). These results confirm that while MILP can be useful for offline planning or small-scale cases, the proposed AI-based method is better suited for real-time, large-scale, and dynamic charging environments, where low-latency decisions are critical.

Figure 3a is a heatmap illustrating the EV-to-CS assignment. Each row in the heatmap represents an EV, and each column represents a CS. A black square at cell $(i,j)$ indicates that $E{V}_i$ is assigned to $C{S}_j$ for charging. The greedy algorithm assigns EVs based on the computed score (distance, waiting time, cost), while ensuring no CS exceeds its capacity limit. Therefore, multiple EVs can be assigned to the same CS, up to the station’s capacity. This heatmap captures the final assignment result of a single batch scheduling step; temporal dynamics such as staggered arrivals or queuing behavior are not modeled in this scenario. Specifically, each row represents an EV, and each column stands for a CS. When an EV is assigned to a CS, a black square is represented. From the figure, it is served that the proposed algorithm successfully allocates EVs to the available CSs, while avoiding overlaps and congestion. Additionally, it is noted that the distribution of EVs across the CSs is non-uniform, reflecting variations in price, distance, and waiting time.

In Figure 3b, the total score of the greedy charge scheduling algorithm, including distance, waiting time, and energy cost, is illustrated. Each bar of the histogram represents the number of EVs that are assigned to a CS with a total cost within a specific range. From the figure, it is observed that there is moderate concentration around medium costs (EUR 4–6). This shows that most EVs are assigned to CSs with total scores in this range, highlighting the balance that is achieved through our proposed strategy between distance, waiting time, and enegy cost. Very few CSs are willing to accept higher-cost assignments (up to EUR $6.5$); however, some EVs accept higher-cost assignments when they are far from CSs, or there is high congestion or limited capacity at low-cost CS. Few EVs are assigned to really low-cost CSs (<EUR $3.5$) and their low cost is due to the small distance, but still, they may suffer from waiting times. From the figure, it is concluded that the algorithm effectively balances load and cost, keeping most costs within a tolerable range. Even though the greedy algorithm does not reach a global optimal solution, it avoids congestion. The algorithm performs well under constraints and provides a viable solution, and the BI charge scheduling model realistically distributes the energy resources.

Table 3. Performance summary of charging schedule recommendation strategy.

CS	Assigned EVs	Capacity	Utilization (%)	Average Cost (EUR)
$C{S}_1$	10	10	$100.0\%$	$4.32$
$C{S}_2$	4	8	$50.0\%$	$5.62$
$C{S}_3$	10	12	$83.3\%$	$5.36$
$C{S}_4$	5	5	$100.0\%$	$4.93$
$C{S}_5$	14	15	$83.3\%$	$6.28$
$C{S}_6$	7	7	$100.0\%$	$5.87$

provides insight into how the charging schedule recommendation algorithm assigned EVs across available CSs during the simulation. The table includes four key metrics: assigned EVs, CS capacity, utilization, and average cost. It is observed that most CSs achieve high utilization (more than $80\%$), indicating the ability of the algorithm to maximize infrastructure usage. Additionally, the proposed approach is cost-aware, since the CSs that offer energy with higher assignment costs absorb more EVs if they had excess capacity and are nearer to EVs. A CS with low utilization might be located further from demand points or have less competitive pricing. This summary validates the effectiveness of the algorithm in spatially distributing EVs, optimizing CS utilization, and maintaining cost-efficiency. It also reveals how BI-based decision-making can balance demand across heterogeneous infrastructure, directly improving the quality of service and economic outcomes in EV environment.

5.2. Cooperation Among Different CS Owners

To evaluate the BI-based double auction mechanism for the cooperation among different CS owners, we assume 10 independently owned CSs in a high-demand urban area. In more detail, we assume that five CSs experience traffic overload, each one of them expecting to redirect 5 to 10 EVs that are waiting. Concurrently, five underutilized CSs have a spare capacity of 4 to 12 available charging slots. The overloaded CSs submit bids indicating the price it is willing to pay per redirected EV, while the underutilized CSs submit ask prices to accept incoming load. The BI model acts as an auctioneer, matching buyer–seller pairs based on bid–ask compatibility and gaining maximization. Simulation results show that the auction effectively balances the load, increases infrastructure utilization, and generates measurable economic gain, demonstrating the viability of cooperative load management among CS stakeholders.

Figure 4a illustrates the economic gains that are produced after a matching of overloaded buyer CSs with underutilized seller CSs has succeeded. The rows and columns represent the buyers and sellers, respectively. The values in each cell indicate the gains from each feasible matching. From the figure, it is observed that a number of buyer–seller pairs achieve high gains (>EUR $2.0$), indicating favorable trading conditions (high bid, low ask, and large quantity). The zero cells represent infeasible matches when the bids are too low or capacity and demand constraints prevent the trade. Thus, it is concluded that the BI model prioritizes gain-maximizing matches, validating greedy matching logic.

Continuously, Figure 4b shows how many EVs each seller CS agreed to accept from overloaded CSs as part of the double auction algorithm. First, one observation is that sellers with lower ask values and higher capacity accept more EVs, confirming the gain maximization. A second observation is noted from the figure. More specifically, one or two sellers may dominate the accepted load, indicating that some CSs are more strategically positioned to absorb redirected traffic, due to either pricing competitiveness or availability. Last, other sellers may accept fewer or no EVs, because their ask prices were too high relative to bids or their available capacity was exhausted in previous matches. In conclusion, the proposed model improves the utilization and revenue of the CSs through dynamic cooperation rather than isolated pricing strategies.

The histogram of Figure 4c shows the distribution of total gain values across all matched buyer–seller pairs in the double auction. First, as it is observed, most matches achieve moderate gains (EUR 1–3), while fewer achieve gains of EUR 4. The distribution of the gains highlights that the inefficient or marginally profitable trades are excluded by the greedy approach. The mechanism not only balances load but creates measurable financial value that could be shared among CS stakeholders. This supports the economic feasibility and incentive alignment of the cooperation model.

6. Discussion

The results of Algorithm 1 evaluated the performance in an urban scenario with spatial and temporal variations. The assignment approach confirmed that the algorithm successfully distributed EVs across CSs without violating capacity constraints. Additionally, the algorithm balanced user preferences for proximity, waiting time, and cost. Moreover, the cost distribution analysis revealed that the algorithm effectively adapted to spatial heterogeneity, allowing for fair and efficient allocation even in scenarios with uneven demand and congestion. These results confirmed that the strategy offers a scalable and responsive solution suitable for dynamic, urban charging environments.

The proposed double auction mechanism for the CSs’ cooperation was evaluated through a simulated scenario involving multiple overloaded and underutilized charging stations. The auction generated economically beneficial matches between CSs. Additionally, the cooperation was not uniformly distributed, meaning that CSs with competitive ask prices and sufficient spare capacity absorbed a higher share of redirected load, improving utilization. The mechanism offered a great potential for profit sharing among independent CS operators. Overall, the results validate the viability of using auction-based cooperation as a decentralized, market-compatible method to balance infrastructure loads, reduce customer wait times, and enhance grid-wide efficiency without requiring centralized ownership or regulation.

7. Conclusions

In this paper, we proposed a Business Intelligence model designed to enhance cooperation between electric vehicles and charging station owners. We began by reviewing the state-of-the-art literature, identifying both the innovations and limitations of existing approaches. Building on this foundation, we introduced a multi-layered BI architecture that integrates spatial and temporal data collection, predictive analytics, and decision-making components. The model addresses key challenges in EV charging ecosystems, including demand balancing, energy scheduling, cost optimization, and infrastructure planning.

Two core decision-layer strategies were developed: (i) a real-time, greedy charging schedule recommendation algorithm to efficiently assign EVs to CSs, and (ii) a cooperative double auction mechanism that enables CS operators to dynamically redistribute load and share surplus capacity. Simulation results demonstrated that our model improves charging coverage, energy utilization, infrastructure accessibility, operational efficiency, and revenue distribution among stakeholders. These findings highlight the potential of BI-driven coordination to support scalable, decentralized, and intelligent EV charging networks.

Future work will explore real-time deployment in live urban networks and integration with renewable energy forecasting for grid-aware optimization.