Prediction of the Charging Probability of Electric Vehicles with Different Power Levels

Abstract

As the market share of electric vehicles (EVs) increases year by year, their charging load forecasting has become a research hotspot and a difficulty. Aiming at the shortcomings of the current research on the charging probability prediction of EVs with different power levels, this paper proposes a multi-power-level EV charging probability prediction method. Firstly, based on the characteristics of electric vehicles, the power of charging facilities, and the travel habits of owners, the SOC mathematical models of charging start time, as well as the start and end state of charge, are established, and the different charging power selection models are established in combination with the parking time. Then, the Monte Carlo simulation method is used to predict the charging probability of electric vehicles with different power levels on typical dates such as working days, weekends, and holidays.

1. Introduction

As the world’s attention to environmental protection and sustainable energy utilization continues to increase, the market share of EVs as an important alternative to traditional fuel vehicles is showing a trend that is increasing year by year [1]. This growth not only helps to reduce greenhouse gas emissions and alleviate the energy crisis, but also largely promotes the green revolution in the transportation sector. The authors of [2] provide a review of the development of electric vehicle technology, focusing on the technological progress in aspects such as batteries, electric motors, charging technologies, and charging facilities. They point out that the development of electric vehicles is not only dependent on technological innovation but also constrained by factors such as battery costs, charging speed, range, and the popularity of charging networks. Improving the energy efficiency and usability of electric vehicles will be the key to promoting their widespread application, which not only helps reduce reliance on fossil fuels but also significantly reduces carbon emissions. The authors of [3] explore the scenario analysis of the future development of electric vehicles and propose to predict the evolution of electric vehicle technologies and markets based on the influence of multiple factors such as society, economy, and politics. Their research shows that the popularization of electric vehicles will not only rely on technological progress but will also be affected by factors such as policy support, public acceptance, and the construction of power grid infrastructure. Through these future scenario analyses, the research provides a new perspective on the role of electric vehicles in the global energy transition. The authors of [4] further emphasize the application of artificial intelligence (AI) in sustainable energy systems, particularly in the potential of optimizing the interaction between electric vehicles and smart grids. AI technology can assist in optimizing the charging behavior of electric vehicles, predicting charging demands, and regulating energy flow through the collaboration between smart grids and electric vehicles, thereby reducing energy waste and enhancing grid stability. The application of AI helps to promote the role of electric vehicles in energy management, making them a key component in smart grids.

However, the widespread popularity of electric vehicles has also brought a series of new challenges, among which the forecasting of electric vehicle charging load has become a research hotspot and a practical difficulty. Accurately forecasting the charging load of electric vehicles is of crucial significance for optimizing the operation of the power grid, rationally planning the layout of charging facilities, and ensuring the stable supply of the power system [5]. There are still some limitations in the current research on the probability prediction of electric vehicle charging at different power levels. Some studies fail to fully consider the comprehensive effects of various factors such as the characteristics of electric vehicles, the power of charging facilities, and the travel habits of owners when building charging load prediction models [6,7,8]. For example, some studies have analyzed the charging laws of electric vehicles, but they are not detailed enough in classification modeling and do not fully cover the behavioral differences of different types of electric vehicles on different dates and charging scenarios. In addition, the modeling of key factors such as charging start time, as well as charging start and end state of charge (SOC), also needs to be improved [9]. Some studies lack sufficient data support or reasonable theoretical assumptions when determining these parameters, resulting in the accuracy and reliability of the model being affected. In terms of charging power selection, existing studies often do not fully combine the parking time of electric vehicles to conduct in-depth analyses and fail to fully consider the actual demand of vehicle owners for charging power under different parking hours. Only a few scholars have carried out relevant research on the probability prediction of EV charging at different power levels [10]. The authors of [11] consider the differences in charging laws of different types of EVs, classify electric vehicles according to their uses, and construct EV models. At the same time, the effects of time-of-use electricity price and the diversity of charging methods on the charging load distribution are comprehensively considered, and a method for forecasting the charging load of different types of EVs is proposed. The authors of [12] point out that the slow-charging methods mostly occur at night, and the fast-charging and slow-charging methods of vehicles can be determined according to the corresponding proportions. In [13], aiming at the private car charging scenario of a typical work or community parking lot, the charging power of EVs is calibrated as 60 kW and 7 kW, respectively, so that the charging probability prediction method of different charging power levels is proposed based on the charging time obeying the normal distribution.

The VSG technology and CAVI strategy in reference [14] provide a new application direction for the electric vehicle energy storage system, enabling electric vehicles not only to serve as energy storage units but also to regulate the grid frequency through virtual inertia. On the other hand, reference [15] delves deeply into the CC-VSG control strategy, focusing on the optimization of microgrid frequency regulation. By precisely controlling the energy storage system, it reduces frequency fluctuations and enhances the stability of the grid. These two technologies have great potential in the interaction between electric vehicles and microgrids, effectively enhancing the collaborative effect of electric vehicles in the future smart grid and promoting the further development of new energy and smart grids. These two papers provide important theoretical and technical support for the combined application of electric vehicles and microgrids.

In view of the above research shortcomings, this paper proposes a multi-power-level EV charging probability prediction method. This method deeply analyzes the characteristics of electric vehicles, the power of charging facilities, and the travel habits of owners and builds a more complete mathematical model of charging start time, as well as charging start and end SOC (state of charge), and combines the parking time to establish a mathematical model of different charging power selection. Finally, based on the Monte Carlo simulation method, an accurate prediction of the charging probability of electric vehicles with different power levels is realized for typical dates, including working days, weekends, and large holidays, in order to provide more valuable references for the research and practice of electric vehicle charging.

The developed algorithm possesses high-precision prediction capabilities, adaptability to multiple power levels, and the ability to handle complex factors. Firstly, by comprehensively considering various factors such as the charging facility power, vehicle type, and the travel habits of the vehicle owner of electric vehicles, the algorithm can provide accurate charging load prediction, avoiding the limitations of traditional methods and ensuring the reliability of the prediction results. Secondly, the algorithm can flexibly adapt to different power levels of charging facilities (such as 7 kW, 21 kW, 60 kW, 120 kW, etc.) and precisely predict the load changes in different charging scenarios, thereby improving the utilization efficiency of electric vehicle charging facilities. Finally, the algorithm can take into account the time-varying nature of charging demands and the changes in vehicle travel plans and still possess strong adaptability and accuracy in diverse practical scenarios, providing effective support for power grid management and the layout of charging facilities.

2. EV Charging Probability Prediction

In this study, the workdays, weekends, and holidays are selected as typical days for charging load forecasting. This is mainly because the charging demands of electric vehicles on these three types of days show significant differences. The charging demands on workdays usually exhibit regular fluctuations, especially during the commuting hours; weekends show different charging patterns due to leisure travel and long-distance trips; and holidays may have peak demands due to activities such as long-distance travel and family gatherings. By forecasting these three typical days, the changing trends of charging demands of electric vehicles can be comprehensively reflected, providing strong data support for load management of the power grid and planning of charging facilities. EV charging load is affected by various factors such as its own characteristics, charging facility power, and the traveling habits of vehicle owners. In this paper, we first establish the EV charging start moment, as well as the charging start and end SOC mathematical model, then establish the mathematical model of different charging power selections by combining the parking duration, and finally predict the charging probability of different power levels for a typical day containing weekdays, double days off, and large holidays based on the Monte Carlo simulation method.

3. EV Charging Load Prediction Model

This research is based on the analysis of the charging behavior data of electric vehicles in a residential area in Tianjin. The dataset used covers the charging records of a large number of electric vehicles within the area during the period of 2022. The dataset includes information such as the parking duration of the vehicles, the starting and ending values of SOC (state of charge) for each charging session, charging periods, peak charging hours, and user charging preferences.

To better predict the charging load of electric vehicles, this research collected and integrated various data related to electric vehicle charging, including actual vehicle travel conditions, charging demands on different dates and time periods, usage frequency of charging facilities, and selection of different charging point power levels. The data came from the local government’s traffic monitoring system, charging station operators’ charging records of electric vehicles, and questionnaires and interviews conducted by electric vehicle users. Through in-depth analysis of these data, we constructed a prediction model and made predictions on the charging load. The model not only takes into account the battery capacity of the vehicles, the power of charging facilities, and parking duration, but also comprehensively considers the changes in charging demands under different typical dates such as weekdays, weekends, and holidays [16].

3.1. The Starting Moment of Charging

Based on the different travel characteristics and charging modes of EV owners on weekdays, double holidays, and large holidays, the EV charging probability is categorized into three typical days, weekdays, weekends, and large holidays, and the charging type is categorized into AC slow-charging pile charging and DC fast-charging pile charging.

Taking the charging facilities in Tianjin as an example, the data collected from the slow-charging piles show that most of the charging time of the charging owners with charging time within two hours is during working hours. Considering that vehicle owners are forced to choose slow-charging piles due to the lack of fast-charging piles, these types of data are excluded from the original data, and then the starting moments of charging at AC slow-charging piles on weekdays, weekends, and large holidays are obtained. From the collected data of fast chargers, it can be seen that the charging time is guided by price and other mechanisms, and almost all charging times are within two hours and the charging start time is similar, so these kinds of unreasonable data are eliminated to obtain the charging start time of fast chargers on weekdays, weekends, and holidays.

Considering the Gaussian kernel function as shown in Equation (1), the probability of travel moments on typical days for EVs is fitted using the nonparametric kernel density estimation algorithm; the results are shown in Figure 1, and the results of the fitting test are shown in Table 1. From a horizontal perspective, it can be observed that the charging probability and the fluctuation trend of charging probability at each moment between slow charging and fast charging for typical days in this city vary significantly; from a vertical perspective, it can be seen that the charging probability trend is similar between typical days in the city, and there is a slight difference in the peak charging probability.

$$f_h(x)={\displaystyle \frac{1}{nh}}{\displaystyle \sum _{i=1}^n\frac{1}{\sqrt{2\pi }}}\exp [-{\displaystyle \frac{{(x-X_i)}^2}{2h^2}}]$$

(1)

Here, $X_i$ is the sample, n is the sample capacity, and h is the bandwidth (h > 0).

Figure 1. Distribution of charging start times of slow-/fast-charging piles: (a) weekday slow charge; (b) weekday fast charge; (c) weekend slow charge; (d) weekend fast charge; (e) holiday slow charge; (f) holiday fast charge.

Figure 1. Distribution of charging start times of slow-/fast-charging piles: (a) weekday slow charge; (b) weekday fast charge; (c) weekend slow charge; (d) weekend fast charge; (e) holiday slow charge; (f) holiday fast charge.

Table 1. Bandwidth h and chi-square calibration results.

Type	Bandwidth h	Chi-Square Statistic	Threshold Value	Result
a	1.1	1.4	35.1	Pass
b	1.5	2.3	35.1	Pass
c	1.1	2.3	35.1	Pass
d	1.5	2.2	35.1	Pass
e	1.1	2.1	35.1	Pass
f	1.5	2.2	35.1	Pass

Table 1. Bandwidth h and chi-square calibration results.

Type	Bandwidth h	Chi-Square Statistic	Threshold Value	Result
a	1.1	1.4	35.1	Pass
b	1.5	2.3	35.1	Pass
c	1.1	2.3	35.1	Pass
d	1.5	2.2	35.1	Pass
e	1.1	2.1	35.1	Pass
f	1.5	2.2	35.1	Pass

3.2. Charging Start and End SOC

Based on the Chinese statistical data [17] from a survey study of private EV owners’ charging habits, it can be seen that more than 50% of owners will choose to charge when the remaining SOC is below 40%, of which 1% choose to charge when the remaining SOC is between 1 and 20%, and 37% choose to charge when the remaining SOC is between 21 and 40%. Meanwhile, the percentage of owners who charge as they stop is 22%, and the percentage of owners whose SOC is above 80% at the end of charging is 90%. The results show that most owners will charge their EVs to a more saturated state.

In terms of the charging probability of stop-and-go charging, this paper considers two aspects: charging with slow-charging piles and charging with fast-charging piles. For the owners of vehicles using slow-charging piles, EVs show the charging phenomenon of “a small number of times”, in which the owners carry out charging behaviors when they travel to the charging destinations. In this paper, the capacity, remaining SOC, and charging interval of each EV charging behavior are marked. The remaining SOC is calculated by subtracting the power consumed at the end of the current trip from the power consumed at the end of the previous trip or charging. Combined with Monte Carlo model sampling, it is charged for each charging cycle. For owners who use fast-charging piles to charge their vehicles, although they can charge their EVs when they get home every day, the average daily mileage of such owners is about 50 km, and at this time, the battery SOC of most EVs is above 70%, resulting in a high probability that the owners will not engage in EV charging behaviors. Therefore, this paper reconsiders the starting SOC of charging for stop-and-go EVs by adding the proportion of stop-and-go charging probability to the part of the starting SOC probability between 10 and 20%.

Relevant transportation statistics [18] show that the probability density distribution of miles traveled for a single trip for EVs is lognormal, and its probability density function is denoted as

$$f_D(d)={\displaystyle \frac{1}{d{\sigma }_D\sqrt{2\pi }}}{e}^{-{\displaystyle \frac{{(\ln d-{\mu }_D)}^2}{2{\sigma }_D^2}}}$$

(2)

where d is the mileage traveled for a single trip in km; ${\mu }_D$ is the expectation of ln d; and ${\sigma }_D$ is the standard deviation of ln d.

Due to geographic location, the mean and standard deviation of miles traveled on a typical day will vary across cities and regions. Therefore, the mean and variance of the mileage traveled can be adjusted within a reasonable range according to the local conditions. Typically, the charging start SOC of EVs and the amount of cumulative travel mileage show a negative correlation. In this paper, we set the probability distribution of the cumulative number of trips of EVs to obey the uniform distribution; then, the charging start SOC of EVs can be expressed as

$$\left\{\begin{array}{c}SO{C}_j=(1-{\displaystyle \sum _{k=1}^n\frac{d_k}{d_{full}}})\times 100\%\\ d_{full}=E\times E_{per}\end{array}\right.$$

(3)

where $SO{C}_j$ is the charging start SOC; $d_k$ is the driving range of the kth trip in km; n is the total number of trips made by EVs; $d_{full}$ is the maximum driving range of the EV at the time of charging to the end SOC in km; E is the rated capacity of EV; and $E_{per}$ is the energy consumption per unit mileage of EV.

3.3. EV Charging Power Selection

With the increasing battery capacity of EVs and the rapid growth of ownership, the 7 kW charging pile has shown difficulty in meeting the charging needs of vehicle owners. In order to better solve the charging problems of vehicle owners, this paper sets the charging modes of EVs as 7 kW and 21 kW AC slow-charging modes and 60 kW and 120 kW DC fast-charging modes based on the existing conventional charging pile power levels. In addition, the user’s charging pile power selection tendency is judged based on the EV’s parking duration [19]. By fitting the parking duration data of a region, the probability distribution of parking duration is obtained as shown in Figure 2. The fitting test results are shown in

Table 2. Bandwidth h and chi-square results.

Type	Bandwidth/h	Chi-Square Statistic	Threshold Value	Result
parking duration	1.5	2.4	35.1	pass

.

In this paper, EVs with parking times less than 1 h are charged by 120 kW DC fast-charging chargers for charging, especially when the SOC is low. This helps with rapid charging and ensures sufficient driving range. For electric vehicles with parking times ranging from 1 to 2 h, after calculations based on Formula (5), it is advisable to choose 60 kW DC fast charging for charging. This choice takes into account the initial SOC value of the vehicle and the charging time requirement to ensure an efficient charging process and meet the travel needs of the owners. For electric vehicles with parking times exceeding 3 h, based on the combination of SOC and parking time, 7 kW or 21 kW AC chargers are selected for charging. For vehicles with parking times exceeding 14 h, considering the SOC at the start of charging and the SOC at the end of charging, 7 kW chargers are sufficient to meet the charging needs of the vehicles. Therefore, vehicles with parking times exceeding 14 h only use 7 kW chargers for charging to avoid excessive charging time.

The selection of charging power here combines the SOC status of the electric vehicle, reasonably adjusts the power of the charger, optimizes the charging process, and ensures the balance between charging efficiency and the travel needs of the owners.

In this paper, we select the top 50 Chinese EVs in terms of sales from January 2014 to the end of June 2023 as the object of study and extract the capacity of vehicles based on the probability of the percentage of vehicles in the capacity interval, as shown in Figure 3.

Since most of the EVs on the market with a battery capacity of less than 25 kWh can only support charging at charging piles with a capacity of less than 7 kW, this paper assumes that only 7 kW charging piles are used for small-capacity EVs. For large-capacity EVs with a battery capacity greater than 65 kWh, considering the charging efficiency of the charging pile, the charging start SOC and the charging end SOC, if a 7 kW charging pile is used to meet the travel needs of a large-capacity EV owner, the charging duration will be at least 7 h, and in the future, the EV will have to participate in orderly charging as well as the demand for the frequency and peak adjustment of the virtual plant, so it is necessary to leave margins for the charging duration of the EV. At the same time, according to the “2022 China Electric Vehicle User Charging Behavior White Paper” [20], EV owners prefer to use charging piles with higher charging power for charging, so it is set that large-capacity EV charging will use 21 kW AC charging piles or 60 kW and 120 kW DC charging piles. At this point, the constraints are

$$\left\{\begin{array}{cc}P=7 \mathrm{kW}& E\le 25 \mathrm{kWh}\\ P=21 \mathrm{kW}/60 \mathrm{kW}/120 \mathrm{kW}& E\ge 65 \mathrm{kWh}\end{array}\right.$$

(4)

where P is EV charging power and E is EV battery capacity.

The conditions for selecting the charging pile power for EVs with a battery capacity above 25 kWh based on the probabilistic model of the parking duration for which the sample parking duration is taken are

$${\displaystyle \frac{P_A\eta T_P}{60E}}+SO{C}_j\le SO{C}_c$$

(5)

where $T_P$ is the parking duration; $SO{C}_c$ is the charging end SOC; $P_A$ is the charging power; and η is the charging efficiency of the charging pile, for which this paper chose the value of 0.9.

3.4. EV Charging Duration

The main influencing factors of EV charging duration are charging start SOC, charging end SOC, charging power, and EV battery capacity; then, the formula of EV charging duration is

$$t_c={\displaystyle \frac{(SO{C}_c-SO{C}_j)E}{P_c}}$$

(6)

where $t_c$ is the EV charging duration in h and $P_c$ is the charging power in kW.

3.5. Monte Carlo Simulation Charging Probability Prediction

The Monte Carlo method is a stochastic simulation technique based on probability statistics, which is particularly suitable for problems involving uncertainty and complexity. Due to the fact that the charging behavior of electric vehicles is influenced by various factors, including parking time, SOC, battery capacity, and other changes in multiple aspects, the Monte Carlo method can simulate various possible charging scenarios through a large number of random samplings, thereby estimating the probability distribution of charging. Compared with other optimization methods, the Monte Carlo method can better handle nonlinear relationships and the interaction of multiple factors, providing more accurate and flexible charging load prediction results. The prediction block diagram is shown in Figure 4. When conducting the Monte Carlo simulation, this paper employed the MATLAB(2018b) software for random sampling, data processing, and simulation calculation.

In this paper, we simulate the charging probability model of EVs with different charging power levels on different typical days of weekdays, double holidays, and large holidays based on Monte Carlo simulation and the behavioral characteristics of EV owners as shown in the following steps.

Step 1: Based on the established probabilistic model of charging start moment, charging start and end SOCs, and different charging power selections, extract the EV charging start moment $T_s$, the corresponding charging start and end SOCs, and the stopping time.

Step 2: Using Equation (5), calculate the charging power for selecting the EV for each typical day, and then calculate the charging duration of the EV using Equation (6) for $T_c$.

Step 3: Finally, record the charging status of each sample according to Equation (7), and then, according to Equation (8), calculate the average value of EV charging probability.

At any $t_0$ time, if the EV is charging, then γ = 1; otherwise, γ = 0. The judgment condition that the drawn random sample is in a charging state at time $t_0$ can be expressed as follows:

$$\left\{\begin{array}{ll}{t}_{\mathrm{s}}>t_0,t_{\mathrm{s}}+t_{\mathrm{c}}<t_0+1440\hfill & \gamma =0\hfill \\ t_{\mathrm{s}}+t_{\mathrm{c}}<t_0\hfill & \gamma =0\hfill \\ other\hfill & \gamma =1\hfill \end{array}\right.$$

(7)

For the randomly selected sample arrays $T_s$ and $T_c$, based on Equation (6), the charging status of each random sample in 1440 min of a day can be calculated judiciously. In order to record the charging status of EV samples in 1440 moments of a day, an array T with capacity N × 1440 is established, where N denotes the number of samples; then, the charging probability of EVs in 1440 moments of a day can be expressed as

$$D(i)={\displaystyle \frac{n_i}{N}}\times 100\%$$

(8)

where D(i) is the charging probability at moment I and $n_i$ is the number of random samples of EVs that are in the charging state at moment i.

4. Analysis of Calculation Examples

We took a residential area in Tianjin city as an example to conduct an analysis of the method proposed in this paper. In 2022, the population of this city was 13.63 million, and the number of private cars in use was 2.242 million. Therefore, the number of private cars per thousand people in this city was calculated to be 165. Meanwhile, the number of electric vehicles in this city in 2022 was 375,000, accounting for 16.72%. The relevant parameters of the community are shown in

Table 3. Relevant parameters of the residential area.

Number of Households	Charging Power of Charging Facilities/kW	Resident Population
10,000	7, 21, 60, 120	30,000

. Based on the number of private cars per thousand people and the proportion of EVs in this city, the number of private cars in this community in 2022 was calculated to be 4950, and the number of EVs in use was 827.

The vehicle travel information of the target city is shown in

Table 4. Vehicle travel information of the target city.

Year	Rate of Vehicles Dispatched		Driving Mileage per Trip		Number of Trips
	Weekday	Weekend	Weekday	Weekend	Weekday	Weekend
2017	74%	65%	13.9	16.6	3.32	3.25
2018	70.2%	62.0%	14.7	16.3	3.37	3.38
2019	77%	70.4%	14.9	16.6	3.36	3.28
2020	49.8%	42.9%	15.7	16.7	3.05	2.99
2021	51.3%	47.2%	16.0	17.2	3.16	3.08
2022	47.2%	43.7%	15.8	16.5	3.14	3.08

. Due to the impact of the epidemic from 2020 to 2022, the travel of car owners was greatly restricted, so this period is not taken as a reference. According to the vehicle travel information of Tianjin city shown in Table 4, the rate of vehicles dispatched on weekdays is set to 0.74, the rate of vehicles dispatched on weekends is set to 0.66, and the rate of vehicles dispatched on major holidays is set to 0.7.

In this paper, the sample size N of EVs is set to 100,000, the number of repetitions is set to 1000, and the power consumption per kilometer is set to 0.13 kWh. The charging probabilities of community EVs on weekdays, weekends, and major holidays are predicted, respectively, and the probability prediction results of different charging power levels on each typical day are obtained, as shown in Figure 5.

Table 5. Average charging time of electric vehicles with different power levels.

Average Charging Time/Min	Type of Typical Day			Coincidence Rate
	Weekday	Weekend	Holiday
7 kW	128	134	135	0.34
21 kW	52	55	55	0.16
60 kW	37	37	39	0.06
120 kW	21	21	22	0.02

shows the average charging time of EVs of community residents. It can be seen that the average charging time of different charging power levels on major holidays is the longest. However, overall, the average charging times of different power levels on the three typical days do not vary much. The charging times of 7 kW, 21 kW, 60 kW, and 120 kW are approximately 180 min, 90 min, 40 min, and 20 min, respectively.

4.1. Forecast Results of Electric Vehicle Charging Load

The charging load curves of electric vehicles in the region on different typical days predicted by the Monte Carlo method are shown in Figure 6.

By collecting the data of charging piles in this area, the actual data of total charging load for different typical days in this area can be obtained, as shown in Figure 7a. Meanwhile, the root mean square error (RMSE) between the calculated result and the predicted value is calculated, and the result is presented in Figure 7b.

As can be seen from Figure 7, based on the calculated RMSE, it can be concluded that the model demonstrates high accuracy in the predictions of multiple typical days, with a small difference between the predicted and actual charging loads. Although the error is slightly larger on some days, the overall RMSE is relatively small, indicating that the model can effectively capture the changing trends of the data, possesses high reliability and stability, and is suitable for practical applications.

As can be seen from Figure 6, the charging load peaks are at 11:00, 14:00–18:00, and 23:00. During the daytime, the regional charging load fluctuates significantly. The charging load at night changes in a relatively simple pattern, showing a clear trend of first rising and then falling. The charging load gradually increases after 18:00, reaches its maximum value at 23:00, and then gradually decreases until early morning. By 6:00, the charging load is almost zero, indicating that users have gradually finished charging.

4.2. Future Charging Load Results of Electric Vehicles

With the rapid development of electric vehicles, many scholars have conducted research on the prediction of their penetration rate. By the end of 2022, the number of new energy vehicles in Tianjin city reached 374,700. According to the prediction of the Tianjin city New Energy Vehicle Public Data Collection and Monitoring Research Center, by 2025, the infiltration rate of electric vehicles into the new car sales volume in Tianjin city will reach 50–60%, and the number of new energy vehicles in Tianjin city will exceed 900,000. The number of new energy vehicles in the research area will reach 1874 by 2025. If the area maintains its existing charging patterns, the charging load in this area in 2025 will be as shown in the Figure 8. It can be seen from the figure that the peak charging load will reach 700 kW at 23:00 on weekdays.

5. Conclusions

This paper conducts research on the charging probability prediction of electric vehicles with different power levels. The proposed multi-power-level charging probability prediction method is innovative and practical to a certain extent. Through in-depth analysis of the characteristics of electric vehicles, the power of charging facilities, and the travel habits of car owners, models for the starting moment of charging, SOC, and charging power selection are established, and the Monte Carlo simulation is used to achieve the charging probability prediction for typical dates.

In the instance analysis section, the study collected the charging data of electric vehicles in a residential area in Tianjin city and conducted a detailed analysis of the charging behaviors of different types of electric vehicles. These electric vehicles were classified based on their charging power (7 kW, 21 kW, 60 kW, and 120 kW) and battery capacity (less than 25 kWh and greater than 65 kWh). The charging time varied depending on the power size. For instance, the 7 kW charging stations were suitable for vehicles parked for a long time and had a longer charging time, while the 120 kW fast charging stations were suitable for vehicles requiring quick charging and had a shorter charging time. Through the Monte Carlo simulation method, the study predicted the charging probabilities of electric vehicles on different working days, weekends, and holidays, and analyzed the changes in charging demands at peak times (such as 11:00, 14:00–18:00, and 23:00) for different power charging stations. These results provided valuable data support for grid management, helping to better plan charging facilities and optimize power supply.

There is still considerable expansion space for future research. Firstly, in terms of data, more data from different regions, longer time spans, and various types of electric vehicles can be collected to enhance the universality and accuracy of the model. Additionally, more influencing factors can be considered in model optimization, such as weather, seasonal changes, and sudden changes in the travel plans of vehicle owners, to make the model closer to the actual charging scenarios and enhance its explanatory and predictive capabilities for complex actual situations. Future research can also incorporate the impact of different energy sources on the prediction of electric vehicle charging loads and analyze the overall impact of electric vehicles on the environment and energy structure under different energy compositions, thereby providing a more comprehensive decision-making basis for policymakers. With the popularization of electric vehicles and the exploration of their energy storage potential, the role of electric vehicles in microgrids and smart grids will become increasingly important, becoming a key component in the development of smart grids. In the future, the energy storage and regulation functions of electric vehicles will not only solve the problem of grid frequency but also exert profound influences in energy management, environmental protection, and economic sustainability, promoting the realization of energy transformation and sustainable development.