Recommendation of Electric Vehicle Charging Stations in Driving Situations Based on a Preference Objective Function

Abstract

As the adoption of electric vehicles (EVs) rapidly increases, the expansion of charging infrastructure has become a critical issue. Unlike internal combustion engine vehicles, EV charging is sensitive to factors such as the time and location for charging, depending on the charging speed and capacity of the battery. Therefore, recommending an appropriate charging station that comprehensively considers not only the user’s preference but also the charging time, waiting time, charging fee rates, and power supply status is crucial for the user’s convenience. Currently, charging station recommendation services suggest suitable charging stations near a designated location and provide information on charging capacity, fee rates, and availability of chargers. Furthermore, research is being conducted on EV charging station recommendations that take into account various charging environments, such as power grid and renewable energy conditions. To solve these optimization problems, a large amount of information about the user’s history and conditions is required. In this paper, we propose a real-time charging station recommendation method based on minimal and simple current information while driving to the destination. We first propose a preference objective function that considers the factors of distance, time, and fees, and then analyze the recommendation results based on both synthetic and real-world charging environments. We also observe the recommendation results for different combinations of the weights for these factors. If we set all the weights equally, we can obtain appropriate recommendations for charging stations that reflect driving distance, trip time, and charging fees in a balanced way. On the other hand, as the number of charging stations in a given area increases, it has been found that gradually increasing the weighting of charging fees is necessary to alleviate the phenomenon of rising fee rates and provide balanced recommendations.

1. Introduction

Environmental pollution caused by the large-scale use of internal combustion engine vehicles in modern society has emerged as a serious issue. To address this problem, many countries are promoting the adoption of electric vehicles (EVs). According to a report by the International Energy Agency (IEA) [1], approximately 14 million new EVs were registered worldwide in 2023, bringing the total number on the roads to 40 million. It is estimated that EV sales could reach around 17 million in 2024. This increase in EV adoption plays a crucial role in reducing emissions and enhancing environmental sustainability.

As EV adoption accelerates, the expansion of EV charging infrastructure has become a significant challenge [1]. Unlike refueling for internal combustion engine vehicles, which is relatively quick and unrestricted by time, EV charging is highly sensitive to factors such as battery charging speed and capacity, influencing the choice of charging time and location. Therefore, considering not only user preferences but also charging time, waiting time, charging fee rates, and EV charger conditions, providing appropriate recommendations for charging stations is essential to enhance user convenience [2,3,4,5,6,7,8,9,10,11].

Home charging using private chargers is currently the most common way to charge EVs. EV owners with access to a private parking space equipped for charging can charge their vehicles overnight, which is not only convenient but also allows them to take advantage of lower electricity prices during periods of relatively low demand. However, in densely populated cities where most residents live in multi-unit homes, access to home charging is limited, leading EV owners to rely more on public charging. This is most evident in the Republic of Korea, one of the most densely populated countries in the world, which has the highest share of public charging capacity for EVs [1].

The current charging station recommendation service suggests suitable charging stations near a designated location and provides information on charging capacity, charging fee rates, and charging availability. Recent studies on EV charging station recommendations for public chargers include multi-criteria decision-making (MCDM) methods [12], user charging pattern analysis based on historical data [3,9,10,13,14], and collaborative filtering techniques that optimize recommendations based on past usage information [3,15]. Furthermore, studies on Internet of Things (IoT)-based real-time charging station recommendation systems [2] include methods for predicting future usage rates using machine learning models trained on historical usage data [5,9,10,13] and approaches applying genetic algorithms for multi-objective optimization [16]. Beyond basic considerations such as cost, time, and charging speed, additional factors and preference-optimized recommendation methods have also been explored. These include the following: federated learning, which processes data locally and shares only model updates to prioritize user privacy [17]; machine learning-based predictions of future usage rates to improve the operational efficiency of electric taxis and distribute charging station utilization more evenly [13]; and recommendation methods based on the analysis of EV users’ price sensitivity data [18]. Recommendations for charging stations based on charging speed, price, payment method, or preferences inferred from past usage history or directly entered by the user have also been conducted, focusing on explicitly or implicitly identifying the user’s preferences [8]. Fuzzy logic has been used to recommend the most suitable charging station considering multiple criteria simultaneously [11]. In a study aimed at minimizing total travel time, an optimization of driver’s route selection was conducted based on the congestion game problem [19]. Current recommendation research is summarized in

Table 1. Research on EV charging station recommendations.

Authors	Methodology	Description
Savari et al. [2]	IoT and load forecasting	Real-time EV charging data, traffic, and weather
Zhao et al. [3]	Collaborative filtering	EV users’ charging history, time, and amount
Ibrahim et al. [6]	RBM, Water wheel plant algorithm	Personalized charging station recommendations
Li [5]	Deep learning	User and charging station characteristics
Li et al. [7]	User preference and comment data analysis	Sentiment and preferred charging station types
Habbal and Alrifaie [8]	User preference	User-entered preferences or inferred preferences
Fan et al. [9]	Deep reinforcement learning and spatio-temporal analysis	User movement patterns and charging habits
Lin et al. [10]	PDQN (Deep Q-Network)	Learning user preferences
Alrifaie et al. [11]	Fuzzy multi-attribute decision making	Evaluation of factors using fuzzy logic
Sonmez et al. [19]	Congestion game problem	Minimization of the travel time

.

However, these existing studies primarily focus on stationary users or commercial vehicles, making them unsuitable for personal EV users or recommendations for driving situations. Additionally, methods based on machine learning for analyzing charging patterns are effective for reflecting user preferences and improving accuracy, but require extensive prior information and training processes on existing usage patterns. Therefore, these approaches are not suitable for ongoing charging recommendations, such as those for driving situations, where only current information is available, especially for new users with no prior data.

In this paper, we propose a method to recommend appropriate charging stations for EV users while driving to their destination. For EV charging station recommendations while driving, accurate recommendations that take into account user preferences and diverse scenarios require the inclusion of multiple variables such as distance, time, and fee rate. To recommend optimal charging stations by considering all these variables, a multivariable optimization process is necessary. However, multivariable optimization has the drawback of being complex and time-consuming [20]. To simplify this process while optimizing for various variables, we first suggest a new preference objective function, which combines all variables into a single term by assigning weighted values to each variable. Using this objective function, we then propose a recommendation method suitable for real-time charging while driving, based solely on current information, by applying the suggested objective function. To evaluate the performance of this method, we build synthetic environments and conduct simulations to analyze the recommendation results based on the proposed objective function, which reflects diverse user preferences. Additionally, we present discussions on appropriate weighting values.

According to the Society of Automobile Engineers (SAE) [21], EV chargers connected to the grid are categorized into AC and DC levels [22,23,24]. AC chargers use an onboard charger that converts AC power to DC power for charging the EV battery. They typically require a long charging time due to their low supply power of 3 kW to 11 kW, making them suitable for use at home or in workplaces. Three-phase AC chargers with higher power outputs (>20 kW) are also available (SAE J3068). AC slow chargers are mainly installed and used in places where EVs remain for extended periods due to their long charging times. DC fast chargers, on the other hand, supply DC power directly to EVs, enabling fast charging with high power. They are typically used in situations requiring quick charging, such as on city roads or highways [25]. Therefore, in the scenario described in this paper, where charging is performed while driving to the destination, only the use of DC fast chargers is considered. For EV charger capacities, we consider DC fast chargers with capacities ranging from 50 kW to 200 kW [24,26]. Although high-speed chargers with charging powers exceeding 200 kW are currently available, they were excluded from this study because of their relatively limited availability in the environment considered in this paper.

The structure of this paper is as follows. In Section 2, we define and explain the preference objective function, and propose the optimal charging station recommendation method. In Section 3, simulation results in synthetic environments along with discussions on these results are presented. Real charging environments of an area of Seoul, Republic of Korea, are considered in Section 4 to demonstrate recommendation examples. Finally, conclusions are provided in Section 5.

2. Recommendation Based on the Preference Objective Function

In this section, we first introduce a preference objective function. Next, we propose a method for recommending EV charging stations in driving situations using the proposed preference objective function.

Assume that the area of interest contains S charging stations, with each charging station denoted by $s(\in \{1,\cdots ,S\left\}\right)$. For a given charging station of s, let the distance-related term be represented as $T_d\left(s\right)$, the time-related term as $T_t\left(s\right)$, and the charging fee rate-related term as $T_r\left(s\right)$. The characteristics of the parameters used in each term are summarized in

Table 2. Parameters used for each term of the preference objective function.

Term of the Objective Function	Parameter
Distance $T_d$	Total driving distance
Time $T_t$	Total driving time
	Charging time for a 50% increase of SOC
	Waiting time at a charging station
Rate $T_r$	Charging fee rate (USD/kWh)

, while the information associated with each charging station is summarized in

Table 3. Parameters of information about charging stations.

Charging Station Information	Parameter
Basic information	Charging station name
	Charger capacity (50 kW, 100 kW, and 200 kW)
	Charging fee rate (USD/kWh)
	Charging station location
	Operating company
Time-related information	Average vehicle speed near the charging station
	Charging time for a 50% increase of SOC
	Waiting time at the charging station

.

A more detailed explanation of the terms for distance, time, and charging rates, which take into account the parameters presented in Table 2 and Table 3, is as follows. For an area of interesting, let a and b denote the current and destination locations in ${\mathbb{R}}^2$, respectively. When one drives via the charging station of s, of which location is $c\left(s\right)(\in {\mathbb{R}}^2)$, the term of distance is defined as

$$T_d\left(s\right):=|a-c\left(s\right)|+|b-c\left(s\right)| \left(\mathrm{k}\mathrm{m}\right),$$

(1)

where $| · |$ implies the Euclidean distance. In Equation (1), we considered a straight line between each point to simplify the distance calculation. The term $T_d$ represents the information about the total driving distance in the recommendation method.

We now define the time-related term $T_t$. When the battery capacity of an EV is $C_B$ (kWh) and the charging capacities of the EV chargers at charging station s are equal to $p\left(s\right)$ (kW), we consider a scenario where the state of charge (SOC) increases from 30% to 80%, representing a 50% charge in constant current charging mode. Because fully charging an EV battery takes a long time and accelerates battery aging, it is reasonable to assume that the SOC is increased up to 80% to ensure both fast charging and extended battery life [27,28].

For simplification in the mathematical formulation, we assume that both the charger efficiency and the battery efficiency are equal to 1. Under these assumptions, the charging time can be calculated using the following simplified relationship [24,26]:

$${\displaystyle \frac{C_B}{2p\left(s\right)}} \left(\mathrm{h}\right).$$

(2)

Next, the total driving time is calculated based on the average vehicle speed in the area of interest. Let v (km/h) represent the average speed of the vehicle along the route. When synthesizing the simulation environment, the average speed in the Seoul area in 2024 is set as $v=22.7$ km/h, based on the data from Seoul Transport Operation & Information Service (TOPIS) [29]. The total driving time reflecting traffic conditions can then be expressed as

$${\displaystyle \frac{T_d\left(s\right)}{v}} \left(\mathrm{h}\right).$$

(3)

In addition to two terms in Equations (2) and (3), the time term $T_t\left(s\right)$, which represents total trip time to the destination including charging, needs to take into account the waiting time $\tau$ at an arbitrarily chosen charging station. Therefore, the time term $T_t\left(s\right)$ is defined as

$$T_t\left(s\right):={\displaystyle \frac{C_B}{2p\left(s\right)}}+{\displaystyle \frac{T_d\left(s\right)}{v}}+\tau \left(\mathrm{h}\right).$$

(4)

In Equation (4), $\tau$ is a random variable representing the waiting time before charging at a charging station in the event that all chargers are in use and can have an exponential distribution. A charging station can have multiple chargers. In this paper, to simplify the analysis, we assume that the random variable $\tau$, which represents the waiting time, accounts for the presence of multiple chargers [30,31,32]. We can give different weights to emphasize certain components among the three components that make up the time term in Equation (4).

Finally, we formulate the term $T_r$ for the charging fee rate. When the rates of the chargers at the charging station of s are equal to $r\left(s\right)$ (USD/kWh) and a general DC flat-rate system without a basic rate is considered, we simply set this term as

$$T_r\left(s\right):=r\left(s\right) (\mathrm{USD}\mathrm{/}\mathrm{kWh}).$$

(5)

The charging fee can be obtained from $C_{B}r\left(s\right)/2$ [26]. Thus, the term of Equation (5) can be replaced by the charging fee. Considering membership plans or subscription-based pricing, charging fees can be reduced depending on the operational distance of the EV [26]. However, for simplicity in this study, we primarily focused on the DC flat-rate pricing system. The method presented in this paper can be further extended to develop charging station recommendation methods that take membership or subscription-based pricing systems into account.

We now define the objective function by considering EV user preferences. Let $w_d$, $w_t$, and $w_r$ represent the weights for distance, time, and rate, respectively, with the sum of these weights equal to 1. The preference objective function, denoted as $L\left(s\right)$, for the optimization is defined as

$$L\left(s\right):=w_d{T}_d\left(s\right)+w_t{T}_t\left(s\right)+w_r{T}_r\left(s\right).$$

(6)

In Equation (6), each term is normalized by its maximum value before multiplying and adding the weights of the three terms. This ensures that each term has an equal influence on the overall objective function. Letting $s_0$ denote the final recommended charging station point, $s_0$ is obtained by minimizing the preference objective function of Equation (6) as

$$s_0:=arg\underset{s}{min}L\left(s\right),$$

(7)

which can be conducted by computing the objective function $L\left(s\right)$ for all charging stations s in the region of interest and selecting the station $s_0$ that yields the minimum value of $L\left(s\right)$. The proposed recommendation method is illustrated in Figure 1. From Equation (7), an optimal charging station is recommended as the one that minimizes the objective function value based on the goal of minimizing the terms in Table 2. Here, to accurately recommend appropriate charging stations through optimization, research on optimal weight settings is necessary by analyzing the recommendation results according to the changes in the weights of the preference objective function. In the following simulation sections, recommendation results are demonstrated with different combinations of the weights in detail.

3. Simulation Results

In this section, we conducted simulations to examine the characteristics and performances of the charging station recommendation method proposed in this paper. First, we synthesized various charging environments and applied the recommendation method in the synthesized charging environments. We then analyzed the recommendation results for different combinations of the weight in the preference objective function of Equation (6) and various synthetic charging environments. Simulations of the recommended method were performed using Python [33].

3.1. Synthetic Charging Environments

In order to verify the performance of the proposed recommendation method through simulations, we first synthesized various charging environments. The synthetic charging environments had dimensions of 50 km in both width and length, with the total number of charging stations ranging from $S=20$ to $S=100$. Additionally, the composition ratio of each charging capacity in the synthesized environments was set to match the actual composition ratio. The actual composition ratio used was determined using statistical data on the charger capacity distribution of EV charging stations in each region of the Republic of Korea in 2024. Detailed statistical data are summarized in

Table 4. EV chargers installed by region in the Republic of Korea as of 2024.

Charging Capacity p (kW)	Seoul	Gyeonggi	Incheon	Gangwon	Jeolla	Gyeongsang	Jeju
50	1432	2270	438	426	1110	2472	1222
100	1839	3880	829	1334	3197	6071	699
200	497	1514	261	362	867	1324	198

. For simulations, we used the distribution of the Seoul region, where the charger capacity ratios in the synthetic environment were 0.38, 0.49, and 0.13, for 50 kW, 100 kW, and 200 kW, respectively.

The battery capacity $C_B$ of EVs was set based on the average battery capacity of representative EV models sold in the Republic of Korea. According to a document released by the Korea Energy Agency (KEA) [34] in December 2023, the battery capacities of EVs officially measured and certified for EV efficiency can be categorized into two types: standard and long-range EVs, depending on the distance achievable on a single charge. These categories are summarized in

Table 5. Classification of representative EV models sold in the Republic of Korea by battery capacity (Korea Energy Agency Automobile Fuel Efficiency Labeling System as of December 2023 [34]).

	Standard	Long Range
EV model	HyundaiIoniq6, KonaElectric; Tesla3, Y; Kia NiroEV, EV3, RayEV; BenznewEQA; All-ElectricMini; Peugeote-208	HyundaiNewIoniq5, Ioniq6, KonaElectric; Tesla3, Y; KiaNiroEV, EV3, EV6; GenesisG60, G80electrified; AudiQ4e-torn, Q8e-torn; BenznewEQE; BMWi5; VolvoC40recharge; Polestar2; VolkswagenID.4
Battery capacity $C_B$ (kWh)	54.74	82.05

. The average battery capacities are 54.74 kW and 82.05 kWh for standard and long range EVs, respectively. Note that the weight of a long-range EV is greater than that of a standard EV due to the heavier battery. Figure 2 illustrates the driving distances per charge for the EVs listed in Table 5, showing that long-range EVs can provide longer driving distances than standard EVs. Note that high-performance EVs, such as the Audi Q8 e-tron, have short driving ranges even with large batteries. Additionally, among the EVs sold in the Republic of Korea, there are several EV models with battery capacities exceeding 100kWh, such as the Cadillac Lyriq, Mercedes-Benz EQS, BMW iX, Lotus Eletre, Maserati Grecale Folgore, Rolls-Royce Spectre, Porsche Taycan, and Tesla Cybertruck.

In Equation (5), the charging fee rate r was determined based on the rates provided by three companies, GSCHARGEV (www.gschargev.co.kr (accessed on 21 March 2025)), Everon (www.everon.co.kr (accessed on 21 March 2025)), and SKelectlink (www.skelectlink.co.kr (accessed on 21 March 2025)), as listed in

Table 6. Example of DC flat-rates (USD/kWh) offered by EV charging companies (November 2024).

Charging Capacity	Charging Fee Rate (USD/kWh)
(kW)	GSCHARGEV	Everon	SKelectlink
50	0.172	0.172	0.185
100	0.224	0.217	0.241
200	0.245	0.246	0.269

. Since the differences among the charging fee rates of the companies are not significant, it is not easy to observe the impact of the rate on the recommendation results when varying the weight of the charging fee rate $w_f$.

In Figure 3, we illustrate examples of synthetic charging environments, where $S=40$ and $S=80$ charging stations, represented as colored circles (◯), are randomly generated. Each charging station is assigned information about the parameters listed in Table 3. Charging stations with different charging capacities are indicated by different colors.

Several recommendation examples of the proposed method are illustrated in Figure 4, where $S=40$ charging stations are generated for the synthetic charging environment. As shown in Figure 4, the current EV location (◻) and destination (△) are randomly generated to evaluate the recommendation method for various routes within a synthetic charging environment. Figure 4 shows different EV routes and recommendation results based on the preference objective function $L\left(s\right)$. Note that the traffic conditions are reflected by the parameter v in Equation (3). In Figure 4a, an example of the recommendation result is presented, where all weights are set equally to consider all factors uniformly. On the right side of the figure, a color map for the objective function values is displayed with lower values represented in blue and higher values in red. Although there are several charging stations (a and b) closer than the recommended one (R), they are excluded from the recommendation due to higher preference objective function values. This occurs when the charging time or waiting time at a station is significantly higher, resulting in considerably longer trip time, which increases the objective function value and leads to exclusion from the recommendation results. On the other hand, the charging fee rates do not differ greatly from one charging station to another and are evenly distributed across the board. Hence, a suitable charging station can be found nearby rather than far away. For reference, the values of $L\left(s\right)$ are 1.25, 0.988, and 0.957 for station a, b, and R, respectively. In Figure 4b is shown an example of three recommended stations ($R_d$, $R_t$, and $R_f$), where only one weight is set to 1. When $w_d=1$, the distance factor is given top priority and the charging station closest to the driving straight path was selected ($R_d$). As an example of the other extreme, when $w_f=1$, the charging station with the lowest charging rate is selected even if the driving distance is quite long ($R_f$).

3.2. Recommendations for Different Weight Combinations

We analyze the charging station recommendation results based on various weight configurations in the objective function of Equation (6). Within the same synthetic charging environment, we denote the weight of a reference term as $w_0$ and the weights of the other two terms as $w_1$ and $w_2$, respectively. By gradually increasing the reference weight $w_0$ from 0 to 1, we observe the average value of each term in the objective function under the optimal recommendation. During this process, the weights of the remaining terms satisfy the following relationship:

$$w_1=w_2={\displaystyle \frac{1-w_0}{2}},$$

(8)

which ensures the weights of the remaining terms are equal and that the sum of all weights equals 1.

For analysis on the weights, simulations were conducted based on the conditions of Equation (6), and the results are presented in Figure 5. In the simulations, recommendation results were obtained from 100 different routes in each of 100 synthetic charging environments, and the average values of distance, time, and rate for the final recommendations were calculated. The synthetic charging environments were generated based on the conditions described in Section 3.1. Figure 5a illustrates the case where the reference weight is assigned to the distance weight $w_d$. As $w_d$ increases, the distance to the recommended charging station consistently decreases as expected. It is clear that the average time term exhibits an increasing trend as the weight $w_t$ decreases. However, the average charging fee rate shows an unusual characteristic, in which it slightly decreases as $w_d$ increases.

Figure 5b shows the case where the reference weight is assigned to the time weight $w_t$. As $w_t$ increases, the average time decreases rapidly at relatively small values of $w_t$, while both the average charging fee rate and driving distance increase. This indicates that, as expected, prioritizing faster charging times leads to a trade-off with higher charging fee rates and longer driving distances. Specifically, the average fee rate increases because stations with faster charging speeds and larger charging capacities are recommended to save time.

Figure 5c illustrates the case where the reference weight is assigned to the weight of the charging fee rate $w_r$. As $w_r$ increases, the charging fee rate decreases, but both the driving distance and total trip time increase, as expected. This demonstrates that minimizing charging rates requires sacrifices in terms of distance and time. In other words, selecting a station solely to minimize charging rates ($w_r=1$) often results in excessively long distances, trip times, or both, creating a significant drawback, as shown in Figure 4b and Figure 5c. Therefore, instead of setting the weight of the charging fee rate to 1, it is recommended to adjust it to around 0.8 or 0.9. This ensures that the other factors are still minimally considered, thereby preventing overly adverse outcomes. In real-world charging environments, charging capacities within a given area often do not differ significantly. Therefore, in these cases, the charging fee rates do not show much difference. In fact, for the same charging capacity, the difference in fee rates between charging companies is less than 10%. If charging fee rates do not vary significantly from one charging station to another, then assigning equal weights and considering both trip time and distance simultaneously can be regarded as an appropriate recommendation method. The experimental characteristics from Figure 5 are summarized in

Table 7. Summary of the simulation results in Figure 5 (a single arrow indicates a change of less than 10%, while two arrows indicate a change of more than 10%).

Reference Weight	Average Distance	Average Time	Average Rate
Distance: $w_d\to 1$	minimum	↑	$\downarrow \downarrow$
Time: $w_t\to 1$	$\downarrow \downarrow$	minimum	$\uparrow \uparrow$
Rate: $w_r\to 1$	$\uparrow \uparrow$	$\uparrow \uparrow$	minimum
$w_d=w_t=w_r$	low	low	middle

.

Recommendation results with respect to the number of stations S is shown in Figure 6 for the equal weight case of $w_t=w_t=w_r$. It is clear that, as S increases, both distance and time decrease. However, a slight increase in the charging fee rate is observed. As the number of charging stations in a given area increases, constraints on time and travel distance decrease. Hence, to alleviate the phenomenon of charging fee rates actually increasing and provide balanced recommendations, it is necessary to gradually increase the weighting of charging fees.

4. Example of Charging Station Recommendations in a Real-World Environment

In this section, we applied the proposed recommendation method to real-world scenarios to verify its performance using the charging station distribution data from Seoul, Republic of Korea.

Figure 7 illustrates a charging environment in the Seoul metropolitan area. The region spans approximately 10.2 km × 11.2 km and includes 30 charging stations with a 50 kW capacity (black, small rectangles), 24 stations with a 100 kW capacity (blue, medium rectangles), and 19 stations with a 200 kW capacity (red, large rectangles). Additionally, there are 29 charging stations with chargers exceeding 200 kW. Hyundai and Kia Motors, as well as Audi, BMW, Genesis, Lexus, Mercedes-Benz, Mini, Polestar, Porsche, Peugeot, Tesla, Volkswagen, and Volvo offer membership and subscription-based charging services in partnership with several EV charging station operators [26]. Note that there are more than 14 charging stations with superchargers in the area shown in Figure 7b for charging Tesla EVs. Additionally, Porsche Korea Ltd. operates three regular charging stations and two ultra-fast chargers, capable of charging at speeds exceeding 300 kW in the same area.

To simplify the simulation, the same experimental conditions as in the previous sections were applied. As in the previous sections, only DC fast chargers with power capacities of 50 kW, 100 kW, and 200 kW were considered in the simulations. AC slow chargers with capacities ranging from 3 kW to 11 kW were excluded from the recommendation process. Specifically, instead of setting the number of chargers at each station, the waiting time $\tau$ for charging was randomly assigned using an exponential distribution with a mean of 5 min in Equation (4). Additionally, it was assumed that all charging stations followed a DC flat-rate pricing scheme, excluding membership or subscription-based pricing plans [26].

Figure 8 shows the simulation results of applying the proposed recommendation method to four arbitrarily selected routes within the charging environment shown in Figure 7b. In this case, the weights in the objective function were set equally, based on the weight analysis results in the previous section, ensuring equal emphasis on the distance, time, and rate terms in the preference objective function of Equation (6). The figures show the optimally recommended charging stations and routes based on the preference objective function. The charging stations colored blue in each figure can be suggested as alternative options for charging the EV. Hence, we can select a more appropriate charging station among them while moving the EV.

To develop an application that recommends charging stations in a real-world charging environment, it is necessary to use an application programming interface (API) that provides maps, navigation information, and EV charging station data, which are available from platform companies such as Google LLC [35] or Naver Corp [36]. The recommendation method developed in this paper can be expanded and adapted into a practical application using such API information.

5. Conclusions

In this paper, we proposed a method for recommending charging stations during EV trips from the current position to the destination. To achieve this, we define a preference objective function that considers various factors important to EV users, such as total distance, time, and charging fees. The proposed method was validated in synthetic experimental environments, not only by evaluating its recommendation performance but also by conducting experiments with various weight configurations to suggest appropriate weight settings. It was confirmed that suitable charging stations, reflecting user preferences, could be recommended based solely on the given information, even without extensive prior data on user behavior patterns. Furthermore, the proposed weights effectively reflect user preferences while avoiding extreme biases, enabling well-balanced charging station recommendations. These weights are expected to be integrated into deep learning models as trainable parameters in the future, potentially simplifying the process of learning user behavior patterns. The results of the analysis through simulations on selecting weights in the proposed method are summarized as follows:

• If we want to recommend charging stations that reflect the driving distance, trip time, and charging fee rate in a balanced way, we can set all the weights equally.
• If distance or time is a priority, setting the weight to around 0.9 instead of 1 helps minimize the influence of other factors and prevents them from reaching undesirable states.
• Regarding charging fee rates, the power capacity of DC fast chargers ranges from 50 kW to 200 kW, and the differences in rates under the DC flat-rate system are not significant. As a result, optimizing for charging fees does not lead to substantial savings. Moreover, selecting a charging station solely based on minimizing charging fees can have the drawback of resulting in excessively long driving distance or trip time.
• If the differences in charging fee rates are not significant, it is beneficial to reduce the weight assigned to the charging fee rates and increase the weights for driving distance and trip time.

The recommendation method proposed in this paper is an analytical experiment that seeks the optimal solution by integrating various factors into a single objective function, and simplifies the details of each factor by using their probabilities. To commercialize the recommendation method, additional research is necessary, focusing on real-time traffic conditions, specific routes between the departure and destination points, and detailed analyses of waiting times at charging stations. Furthermore, research on the use of preference objective functions that take into account membership EV charging rates is also needed.