A user behavior-based electric vehicle charging load prediction method

Abstract

The application provides a kind of electric vehicle charging load prediction method based on user behavior, comprising the following steps: A, obtaining data, including electric vehicle user's vehicle data, time data, location data, trajectory data, POI data, SOC data;B, data preprocessing;C, the travel data of electric vehicle is mined, including map matching, generating regenerative data, spatial modeling is carried out through grid division, POI identification, functional area space flow feature mining;D, based on the data mining result of step C, the travel process simulation and charging decision simulation of electric vehicle user are carried out, the charging demand of user is calculated based on travel process simulation and charging decision simulation, and the theoretical charging demand of each region is obtained.The application is based on the real behavior data of electric vehicle user, and the actual demand of user is more fully analyzed and mined, and the charging load prediction carried out by this will be closer to the real demand of user.

Description

Technical Field

[0001] This invention relates to the field of big data technology, specifically a method for predicting electric vehicle charging load based on user behavior. Background Technology

[0002] According to statistics from the International Energy Agency, transportation systems account for approximately 29% of global energy consumption and 23% of related carbon emissions, making them a significant contributor to the current energy crisis and environmental pollution. To achieve energy conservation and emission reduction in transportation systems, my country has invested heavily in promoting the development of the electric vehicle industry. Against this backdrop, it is necessary to research data-driven electric vehicle charging load forecasting technology to guide related electric vehicle services and provide theoretical and technical support for improving the user experience of electric vehicles.

[0003] Current research primarily relies on charging data to predict electric vehicle charging load. For example, patent application number 201710351396.8, titled "Electric Vehicle Charging Demand Prediction Method and System," proposes a method for predicting electric vehicle charging demand. This method mainly determines the probability of charging at the current time based on battery level, estimated parking time at the current location, and historical data, and calculates the charging demand probability for each electric vehicle, without considering users' specific travel behaviors. Patent application number CN201911075075.5, titled "Charging Station Load Prediction Method, Device, Computer Equipment, and Storage Medium," discloses a charging station load prediction method. However, this method mainly focuses on the charging station's perspective, predicting based on historical data such as the number of vehicles arriving, their type ratio, and arrival time, without fully considering user behavior. Therefore, it cannot fully uncover the true needs of electric vehicle users. Application (Patent) No.: 202011142389.5, Title: Method, System and Device for Predicting Electric Vehicle Charging Demand Based on Neural Network. This patent mainly uses historical charging pile information and charging demand as model training data, and weather forecast information and holiday information as model auxiliary training data. However, it does not analyze and mine the specific travel behavior of electric vehicle users, so the predicted demand cannot fully reflect the real needs of users.

[0004] Some studies have incorporated electric vehicle (EV) operating data for prediction. For example, patent application number 201811108495.4, entitled "A Method and Apparatus for Predicting Electric Vehicle Charging Demand," discloses a method and apparatus for predicting EV charging demand. This method primarily uses historical traffic environment information and corresponding historical EV driving speeds to establish a neural network model to predict EV driving speed, thereby determining the EV's energy consumption. Finally, it determines the EV charging demand based on energy consumption and battery data. However, this approach still fails to predict demand from the user's perspective.

[0005] In addition, some studies have incorporated electric vehicle travel data for demand forecasting, but these studies typically lack comprehensive data coverage and fail to fully capture user needs. For example, patent application number 202110427920.1, entitled "Method and System for Forecasting Urban Electric Vehicle Charging Demand Based on Intelligent Agent Simulation," primarily forecasts demand based on traveler behavior parameters determined by probability density functions and electric vehicle trajectory data, without considering geographical factors or other factors, thus lacking comprehensiveness.

[0006] In summary, existing research rarely takes the perspective of electric vehicle users and uses rich user behavior data to predict charging load. Summary of the Invention

[0007] To address the aforementioned shortcomings of existing technologies, this invention provides a method for predicting charging load for electric vehicles based on user behavior. The main advantage lies in its ability to more fully analyze and uncover actual user needs based on real behavioral data of electric vehicle users. This results in charging load predictions that more closely reflect actual user demands. Furthermore, this invention considers data dimensions including trajectory data, geographical data such as road network data, POI (Point of Interest) data, and battery data, making it more comprehensive than existing technologies.

[0008] The technical solution adopted in this invention is as follows:

[0009] A method for predicting electric vehicle charging load based on user behavior includes the following steps:

[0010] A. Acquire data, including vehicle data, time data, location data, trajectory data, POI data, and SOC data of electric vehicle users;

[0011] B. Preprocess the data obtained in step A;

[0012] C. Mining the data after preprocessing in step B, including map matching, generating regenerated data, spatial modeling through grid division, POI identification, and mining of the spatiotemporal flow characteristics of functional areas;

[0013] D. Based on the data mining results of step C, simulate the travel process and charging decision of electric vehicle users. Calculate the charging demand of users based on the travel process simulation and charging decision simulation to obtain the theoretical charging demand of each region.

[0014] Furthermore, the vehicle data mentioned in step A includes the vehicle ID; the time data includes the journey start time, arrival / departure time of each destination along the route, and stop time; the location data includes the start / end point location and the stop location; and the trajectory data includes the trajectory GPS latitude, trajectory GPS longitude, speed data, and direction data.

[0015] Furthermore, step B involves preprocessing the data obtained in step A, specifically as follows:

[0016] First, the data is divided into several sets based on daily data;

[0017] Then, the bad data is deleted. The bad data includes: data that is not within the study area; multiple duplicate data of the same vehicle within a short distance in the same time period; data with abnormal speed; and data with abnormal offset.

[0018] Finally, based on specific needs, the POI data is classified into: residential areas, commercial areas, industrial areas, and public service areas.

[0019] Furthermore, step C involves mining the data preprocessed in step B, specifically as follows:

[0020] First, map matching is performed, which involves matching the GPS trajectory with the road network model;

[0021] Secondly, generate regenerative feature data: OD trip set, road network peer speed set, and actual driving path set;

[0022] Then, spatial modeling is performed, and the study area is divided into grids at an appropriate scale. The division method is as follows: ① The spatial range of the study is divided at different scales, and the grids are numbered: A u (j)={A u (1),A u (2),...A u (m)}, where u is the partitioning scale, u∈ [50,500], and the interval is 50m; A u (j) represents the index of the j-th grid at scale u; m represents the total number of grids; ② Calculate the effective data volume within the grid at different spatial scales: Among them, E u(i) represents the number of valid OD points in the i-th grid. The optimal u value is obtained through steps ① and ②, and the grid is divided based on this value.

[0023] Furthermore, POI identification is performed: POIs are classified based on the actual situation. Then, functional areas are identified based on the POI data, i.e., the functional nature of the grid is determined by calculating the density ratio of each type of POI within the grid, thus determining the functional type of the grid: ① Calculate the number of each type of POI in the OD point set of each grid: Where, N i k n is the number of POIs of type k in the i-th grid; o (i,m,k) represents the starting point of the m-th type k element in the i-th grid; n d (i,n,k) represents the endpoint of the nth k-th type in the i-th grid; ② Calculate the POI type in each grid: N i =∑n(i,k), where N i ① Let n(i,k) be the total number of POIs in the i-th grid, and n(i,k) be the number of POIs of type k in the i-th grid; ③ Aggregate the grids as needed;

[0024] Finally, the spatiotemporal flow characteristics of the functional areas are mined, and the spatiotemporal flow characteristics of the functional areas are mined by combining OD set POI identification.

[0025] Furthermore, step D specifically includes:

[0026] First, simulate the electric vehicle travel process: Based on the data mining results in step C, determine the probability distribution of each travel characteristic of the electric vehicle. For each characteristic, use the Monte Carlo simulation method to extract random numbers that conform to its probability distribution to simulate the travel process of each electric vehicle.

[0027] Then, a charging decision simulation for electric vehicle users is conducted: based on three factors—the remaining battery power of the electric vehicle, the next travel distance, and the length of stay at the current location—it is determined whether the electric vehicle will be charged after arriving at a certain location and the charging time. The Monte Carlo simulation method is used to extract random numbers that follow the probability distribution of these three features and simulate them on a computer to obtain data for each charging of each electric vehicle. Based on this, a decision is made.

[0028] Furthermore, by integrating the simulation of electric vehicle travel process and the simulation of charging decision process, we can obtain daily charging demand data for electric vehicles, including the location, time and duration of charging demand.

[0029] Finally, charging demand calculations were performed: the study area was divided into grids; demand within the same small grid area was considered to occur at the same location; the center of each small grid area was called the theoretical charging demand point for that area; the day was divided into 24 time periods per hour, and the fast charging demand FQ for each small area during each time period was calculated. need-t And slow charging demand SQ need-t The theoretical charging demand Q for the region is selected from the period with the highest demand. need Then we can obtain the following formula:

[0030] Q need =max{FQ need-t +SQ need-t}(t=1,2,...24).

[0031] Furthermore, the decision-making judgment specifically includes:

[0032] First, determine whether to charge: when SOC < SOC min +Q next When SOC is greater than or equal to SOC, charging is selected; when SOC is greater than or equal to SOC, charging is selected min +Q next If the charging time is not specified, then no charging will be selected. Here, SOC refers to the current state of charge of the electric vehicle. min Q is the lowest acceptable state of charge for the user. next This is the amount of electricity needed for the next trip.

[0033] Second, determine which charging mode to choose: when SOC+Q 慢 ≥SOC min +Q next When SOC+Q is active, choose slow charging; when SOC+Q is active, choose slow charging. 慢 <SOC min +Q next When charging, select fast charging. Among them, Q... 慢 The amount of slow charge during the dwell time;

[0034] Third, determine the charging amount and charging time: if it can be fully charged within the specified time, then Q 充 =SOC max -SOC, If the container cannot be filled within the specified time, then t = t 停留 Q 充 = t*P. Where Q 充 For charging power, SOC max The value is the charge level when fully charged, t is the charging time, and P is the charging power. 停留 This refers to the duration of stay.

[0035] Furthermore, the travel characteristics include travel location, travel time, travel destination, arrival time at destination, travel distance, and duration of stay at destination.

[0036] Furthermore, the expected value E(ε) of the random variable ε in the Monte Carlo simulation is determined by performing N repeated samplings of the random variable ε to obtain n observations, and then calculating it according to the following formula:

[0037]

[0038] This invention, based on real-world behavioral data of electric vehicle users, more thoroughly analyzes and uncovers users' actual needs. The charging load predictions made through this analysis are closer to the users' true requirements. Furthermore, this invention considers data dimensions including trajectory data, geographical data such as road network data and POI data, and battery data, making it more comprehensive than existing technologies. The prediction method proposed in this invention can provide a valid basis for further electric vehicle charging station site selection and other charging-related services. Attached Figure Description

[0039] Figure 1 This is a flowchart of one embodiment of the electric vehicle charging load prediction method based on user behavior according to the present invention;

[0040] Figure 2 This is a flowchart of step D of the present invention;

[0041] Figure 3 This is a flowchart of the electric vehicle user charging decision simulation of the present invention;

[0042] Figure 4 This is a time distribution map of actual charging demand and predicted charging demand within the study area of ​​this invention embodiment. Detailed Implementation

[0043] To make the features and advantages of this patent more apparent and understandable, the invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0044] like Figure 1 The present invention provides a method for predicting electric vehicle charging load based on user behavior, specifically including the following steps:

[0045] A. Obtain the following data from the data source: vehicle data (vehicle ID); time data (trip start time, arrival / departure time of each destination along the route, and stop time); location data (start / end point location, and stop location); trajectory data (track GPS latitude, track GPS longitude, speed data, and direction data); POI (Point of Interest) data; and SOC (State of Charge) data.

[0046] B. Preprocess the obtained raw data, specifically as follows:

[0047] First, the data is divided into several sets based on daily data.

[0048] Then, bad data is deleted, mainly including: data outside the study area; multiple duplicate data of the same vehicle within the same time period over a short distance; data with abnormal speed, such as trajectory data with instantaneous speed exceeding 120km / h; and data with abnormal offset.

[0049] Finally, based on specific needs, the POI data was categorized as follows: Residential Areas, Commercial Areas, Industrial Areas, and Public Service Areas.

[0050] C. Mining is performed based on the data preprocessed in step B, specifically as follows:

[0051] First, map matching is performed, which involves matching the GPS trajectory with the road network model.

[0052] Secondly, regenerated feature data is generated. This mainly includes the following regenerated feature data: OD travel set, road network travel speed set, and actual travel path set.

[0053] Next, spatial modeling is performed. The study area is divided into grids at an appropriate scale.

[0054] Furthermore, POI identification is performed. Based on the actual situation, POIs are classified. On this basis, functional areas are identified based on the POI data (the functional nature of the grid is determined by calculating the density ratio of various POIs within the grid), the functional type of the grid is determined, and the grids are aggregated as needed.

[0055] Finally, the spatiotemporal flow characteristics of the functional areas are mined, and the spatiotemporal flow characteristics of the functional areas are mined by combining OD set POI identification.

[0056] D. Based on the data mining results from step C, construct an electric vehicle charging demand model, specifically as follows (e.g.) Figure 2 As shown):

[0057] First, the electric vehicle travel process is simulated. Based on the travel data mined in step C, the probability distributions of various travel characteristics of electric vehicles (departure location, departure time, departure destination, arrival time, travel distance, and duration of stay at the destination) are determined. For each characteristic, random numbers conforming to its probability distribution are extracted using the Monte Carlo simulation method, thus simulating the travel process of each electric vehicle. Specifically, the expected value of the random variable ε in the Monte Carlo simulation is determined by repeatedly sampling the random variable ε N times to obtain n observations, and then further calculated using the following formula:

[0058]

[0059] Next, a simulation of electric vehicle users' charging decisions is conducted. This simulation primarily uses three factors—the remaining battery power, the next travel distance, and the duration of stay at the current destination—to determine whether an electric vehicle will charge upon arrival at a given location and the charging duration. A Monte Carlo simulation method is used to extract random numbers following a probability distribution of these three characteristics and the simulation is performed on a computer to obtain data for each electric vehicle's charging cycle. The flowchart for the electric vehicle user charging decision simulation is shown below. Figure 3 As shown, three main charging decision judgments are made:

[0060] First, determine if charging is required. When SOC < SOC min +Q next When SOC is greater than or equal to SOC, charging is selected; when SOC is greater than or equal to SOC, charging is selected min +Q next If the charging time is not specified, then charging will not be performed. Here, SOC refers to the current state of charge of the electric vehicle. min Q is the lowest acceptable state of charge for the user. next This is the amount of electricity needed for the next trip.

[0061] Second, determine which charging mode to choose. When SOC+Q 慢 ≥SOC min +Q next When SOC+Q is active, choose slow charging; when SOC+Q is active, choose slow charging. 慢 <SOC min +Q next When charging, select fast charging. Among them, Q... 慢 This represents the amount of charge generated during the dwell time.

[0062] Third, determine the charging amount and charging time. If it can be fully charged within the specified time, then Q... 充 =SOC max -SOC, If the container cannot be filled within the specified time, then t = t 停留 Q 充 = t*P. Where Q充 For charging power, SOC max The value is the charge level when fully charged, t is the charging time, and P is the charging power. 停留 This refers to the duration of stay.

[0063] Furthermore, by integrating the simulation of electric vehicle travel process and the simulation of charging decision process, we can obtain daily charging demand data for electric vehicles, including the location, time, and duration of charging demand.

[0064] Finally, charging demand is calculated. The study area is divided into grids; demand within the same small grid area is considered to occur at the same location; the center of each small grid area is called the theoretical charging demand occurrence point for that area; the day is divided into 24 time periods per hour, and the fast charging demand FQ for each small area in each time period is calculated. need-t And slow charging demand SQ need-t The theoretical charging demand Q for the region is selected from the period with the highest demand. need Then we can obtain the following formula:

[0065] Q need =max{FQ need-t +SQ need-t}(t=1,2,...24)

[0066] The predicted charging demand for the study area was ultimately obtained. Based on this, the prediction results were validated in both the time and spatial domains: in the time domain, the effectiveness was validated by comparing the difference between the predicted charging demand and the actual charging demand at different times of the day, such as... Figure 4 As shown in Table 1, the verification was conducted by comparing actual charging demand with predicted charging demand within 1km in the spatial domain.

[0067] Table 1

[0068]

[0069] The results show that the present invention has high accuracy in both the time and spatial domains, and can provide a good solution for electric vehicle charging load prediction to a certain extent.

[0070] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Anyone can still modify or make equivalent substitutions to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the claims of the present invention.

Claims

1. A method for predicting electric vehicle charging load based on user behavior, characterized in that... Includes the following steps: A. Acquire data, including vehicle data, time data, location data, trajectory data, POI data, and SOC data of electric vehicle users; B. Preprocess the data obtained in step A; C. Mining the data after preprocessing in step B, including map matching, generating regenerated data, spatial modeling through grid division, POI identification, and mining of the spatiotemporal flow characteristics of functional areas; D. Based on the data mining results of step C, simulate the travel process and charging decision of electric vehicle users. Calculate the charging demand of users based on the travel process simulation and charging decision simulation to obtain the theoretical charging demand of each region. The vehicle data in step A includes the vehicle ID; the time data includes the journey start time, arrival / departure times of each destination along the route, and stop times; the location data includes the start / end point location and the location of the stop; the trajectory data includes the trajectory GPS latitude, trajectory GPS longitude, speed data, and direction data. Step B involves preprocessing the data obtained in step A, specifically as follows: First, the data is divided into several sets based on daily data; Then, the bad data is deleted. The bad data includes: data that is not within the study area; multiple duplicate data of the same vehicle within a short distance in the same time period; data with abnormal speed; and data with abnormal offset. Finally, based on specific needs, the POI data is classified into: residential areas, commercial areas, industrial areas, and public service areas. Step C involves mining the data preprocessed in step B, specifically as follows: First, map matching is performed, which involves matching the GPS trajectory with the road network model; Secondly, generate regenerative feature data: OD trip set, road network peer speed set, and actual driving path set; Then, spatial modeling is performed, and the study area is divided into grids at an appropriate scale. The division method is as follows: ① The spatial range of the study is divided at different scales, and the grids are numbered: Where u is the dividing scale, u∈[50,500], and the interval is 50m; Let be the index of the j-th grid at scale u; m represents the total number of grids; ② Calculate the effective data volume within the grid at different spatial scales: ,in, Let u be the number of valid OD points in the i-th grid. Through steps ① and ②, the optimal u value is obtained, and the grid is divided based on this value. Furthermore, POI identification is performed: POIs are classified based on the actual situation. Then, functional areas are identified based on the POI data, i.e., the functional nature of the grid is determined by calculating the density ratio of each type of POI within the grid, thus determining the functional type of the grid: ① Calculate the number of each type of POI in the OD point set of each grid: ,in, The number of POIs of type k in the i-th grid; n is the starting point of the m-th type k in the i-th grid; d (i,n,k) represents the endpoint of the nth type k in the i-th grid; ② Calculate the POI type in each grid: ,in Let be the total number of POIs in the i-th grid. ③ The number of POIs of type k in the i-th grid; ③ Aggregate the grids as needed; Finally, the spatiotemporal flow characteristics of the functional areas are mined, and the spatiotemporal flow characteristics of the functional areas are mined by combining the OD set POI identification. Step D specifically includes: First, simulate the electric vehicle travel process: Based on the data mining results in step C, determine the probability distribution of each travel characteristic of the electric vehicle. For each characteristic, use the Monte Carlo simulation method to extract random numbers that conform to its probability distribution to simulate the travel process of each electric vehicle. Then, a charging decision simulation for electric vehicle users is conducted: based on three factors—the remaining battery power of the electric vehicle, the next travel distance, and the duration of stay at the current destination—it is determined whether the electric vehicle will be charged after arriving at a certain location and the charging time. The Monte Carlo simulation method is used to extract random numbers that follow the probability distribution of these three features and simulate them on a computer to obtain data for each charging of each electric vehicle. Based on this, a decision is made. Furthermore, by integrating the simulation of electric vehicle travel process and the simulation of charging decision process, we can obtain daily charging demand data for electric vehicles, including the location, time and duration of charging demand. Finally, charging demand calculations were performed: the study area was divided into grids; demand within the same small grid area was considered to occur at the same location; the center of each small grid area was called the theoretical charging demand point for that area; the day was divided into 24 time periods per hour, and the fast charging demand FQ for each small area during each time period was calculated. need-t And slow charging demand SQ need-t The theoretical charging demand Q for the region is selected from the period with the highest demand. need Then we can obtain the following formula: 。 2. The electric vehicle charging load prediction method based on user behavior as described in claim 1, Its features are: The decision-making judgment specifically includes: First, determine if charging is required: when When, select charging; when If the charging time is not specified, then charging will not be performed. Here, SOC refers to the current state of charge of the electric vehicle. The lowest acceptable state of charge for the user. This is the amount of electricity needed for the next trip. Second, determine which charging mode to choose: When When charging, choose slow charging; when charging is slow. When selecting fast charging, Q... 慢 The amount of slow charge during the dwell time; Third, determine the charging amount and charging time: if it can be fully charged within the specified time, then... If the container cannot be filled within the allotted time, then... , .in For charging power, The charge level is t, the charging time is P, and the charging power is P. 停留 This refers to the duration of stay.

3. The electric vehicle charging load prediction method based on user behavior as described in claim 1, Its features are: The travel characteristics include travel location, travel time, travel destination, arrival time at destination, travel distance, and length of stay at destination.

4. The electric vehicle charging load prediction method based on user behavior as described in claim 1, Its features are: Mathematical expectation of random variable ε in Monte Carlo simulation The value is determined by performing N repeated samplings on the random variable ε to obtain n observations, and then further calculating it according to the following formula: 。

Images