Electric vehicle charging load spatio-temporal distribution prediction method

Abstract

The invention discloses an electric vehicle charging load spatio-temporal distribution prediction method, and particularly relates to the field of charging load prediction. The method comprises the following steps: acquiring state-of-charge data of an electric vehicle battery under different charging modes and charging duration conditions in real time to obtain state-of-charge time sequence data; through frequency spectrum feature extraction, outputting resonance frequency feature data and resonance amplitude value feature data; performing spatial region clustering analysis on the resonant frequency characteristic data to output spatial resonance region division data, and performing time domain fluctuation rule analysis on the harmonic amplitude value characteristic data to obtain a load oscillation time domain characteristic curve; establishing an association mapping model between charging load space regional distribution and time domain distribution features, and generating fused spatial-temporal feature data; and finally, predicting the spatial-temporal distribution of the charging load of the electric vehicle at the future moment based on the fused spatial-temporal characteristic data. The method effectively improves the prediction precision of the time-space distribution of the charging load of the electric vehicle.

Description

Technical Field

[0001] This invention relates to the field of charging load prediction technology, and more specifically, to a method for predicting the spatiotemporal distribution of electric vehicle charging load. Background Technology

[0002] With the increasing popularity of electric vehicles and the expansion of charging infrastructure, the proportion of electric vehicle charging load in the power system load is gradually increasing. To ensure the safe and stable operation of the power grid and improve the utilization efficiency of charging facilities, it is necessary to accurately predict the spatiotemporal distribution of electric vehicle charging load.

[0003] Existing technologies for predicting electric vehicle charging load neglect the load fluctuation characteristics caused by the interaction between battery state of charge and different charging modes and charging durations. This makes it difficult for the prediction model to accurately depict the spatiotemporal distribution of charging load, resulting in a large error between the prediction results and the actual load, which makes it difficult to meet the actual needs of grid dispatch and charging service optimization. Summary of the Invention

[0004] In order to overcome the above-mentioned defects of the prior art, embodiments of the present invention provide a method for predicting the spatiotemporal distribution of electric vehicle charging load to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: A method for predicting the spatiotemporal distribution of electric vehicle charging load includes the following steps: Collect state-of-charge (POC) data of electric vehicle batteries under different charging modes and charging durations, denoise and calibrate the POC data, and output POC time-series data. Spectral features are extracted from the state-of-charge time series data to identify the high-frequency resonant fluctuation characteristics of the battery's state of charge, and the resonant frequency characteristic data and resonant amplitude characteristic data are output. Spatial region clustering analysis is performed on the resonant frequency characteristic data to determine the spatial region distribution that generates high-frequency oscillations in the charging load, and output spatial resonant region division data. By analyzing the time-domain fluctuation law of the resonance amplitude characteristic data, the time-domain distribution characteristics of the high-frequency oscillation of the charging load are obtained, and the time-domain characteristic curve of the load oscillation is output. Based on the spatial resonance region division data and the load oscillation time-domain characteristic curve, a correlation mapping model between the spatial regional distribution and time-domain distribution characteristics of charging load is established to generate fused spatiotemporal characteristic data. Based on the fusion of spatiotemporal feature data, the spatiotemporal distribution of electric vehicle charging load at future moments is predicted, and the prediction results of the spatiotemporal distribution of charging load are output.

[0006] In a preferred embodiment, state-of-charge (SOC) data of the electric vehicle battery under different charging modes and charging durations are collected. The SOC data is then denoised and calibrated to output time-series SOC data. Specifically: Real-time acquisition of battery state-of-charge data for electric vehicles under constant power charging mode, constant voltage charging mode, and constant current charging mode, as well as under different charging duration conditions. Wavelet denoising, anomaly identification, and data correction are performed on the collected battery state of charge data to obtain battery state of charge time series data.

[0007] In a preferred embodiment, the battery state-of-charge data consists of battery terminal voltage, battery terminal current, and sampling timestamp.

[0008] In a preferred embodiment, the battery state of charge timing data consists of a sampling timestamp, battery voltage, battery current, charging mode label, and charging duration label.

[0009] In a preferred embodiment, spatial region clustering analysis is performed on the resonant frequency characteristic data to determine the spatial region distribution that generates high-frequency oscillations in the charging load, and spatial resonant region division data is output, specifically as follows: Perform a Fourier transform on the battery state-of-charge time series data to obtain the spectral distribution of the battery state-of-charge time series data; The resonant frequency and resonant amplitude characteristic parameters of the battery state of charge under constant power charging mode, constant voltage charging mode, constant current charging mode, and different charging duration conditions are extracted from the spectrum distribution. Based on the resonant frequency characteristic parameters and the resonant amplitude characteristic parameters, the characteristic data of high-frequency resonant fluctuations generated by the battery state of charge are identified and determined, and the resonant frequency characteristic data and the resonant amplitude characteristic data are output respectively.

[0010] In a preferred embodiment, spatial region clustering analysis is performed on the resonant frequency characteristic data to determine the spatial region distribution that generates high-frequency oscillations in the charging load, and spatial resonant region division data is output, specifically as follows: Using the geographical coordinates, service radius, and grid connection node identifier of the charging equipment as spatial indexes, the resonant frequency characteristic data is mapped to the location of the charging equipment. Spatial region clustering analysis is performed on the mapped resonant frequency characteristic data on a preset spatial grid to obtain a set of spatial cluster labels and a set of cluster center coordinates; Perform boundary trimming and connectivity verification on the spatial clusters according to the administrative division boundaries or power grid feeder boundaries to generate a set of spatial resonant regions; Generate spatial resonance region partitioning data for each spatial resonance region.

[0011] In a preferred embodiment, the spatial resonance region division data includes the coordinates of the vertices of the region polygon, the number of charging devices in the region, and the resonant frequency statistics in the region.

[0012] In a preferred embodiment, the time-domain fluctuation law of the resonance amplitude characteristic data is analyzed to obtain the time-domain distribution characteristics of the high-frequency oscillation of the charging load, and the time-domain characteristic curve of the load oscillation is output, specifically: Based on the resonance amplitude characteristic data, multiple charging time intervals are divided according to different charging modes of the charging equipment; For the resonance amplitude characteristic data within each charging time interval, time-domain statistical analysis methods are used to calculate and generate the corresponding mean curve, variance curve, and fluctuation amplitude range. Based on the mean curve, variance curve, and fluctuation amplitude range, the distribution characteristics of high-frequency oscillation of charging load with charging duration are determined, and the time-domain characteristic curve of load oscillation corresponding to each charging duration interval is formed. Summarize the load oscillation time-domain characteristic curves for each charging time interval and output the load oscillation time-domain characteristic curve data.

[0013] In a preferred embodiment, based on the spatial resonance region segmentation data and the load oscillation time-domain characteristic curve, a correlation mapping model between the spatial regional distribution and time-domain distribution characteristics of the charging load is established to generate fused spatiotemporal characteristic data, specifically: Based on the set of spatial cluster labels, the set of cluster center coordinates, and the coordinates of the vertex of the region polygon in the spatial resonance region partitioning data, the spatial feature vector of each spatial resonance region is determined. Based on the load oscillation time-domain characteristic curves of each charging time interval in the load oscillation time-domain characteristic curve data, the time-domain characteristic vector of each charging time interval is determined. Based on the spatial and temporal feature vectors, a mapping relationship is constructed between the spatial feature vector and the temporal feature vector of the charging load through correlation mapping, thereby obtaining the fused spatiotemporal feature data of the charging load.

[0014] In a preferred embodiment, based on fused spatiotemporal feature data, the spatiotemporal distribution of electric vehicle charging load at future times is predicted, and the predicted spatiotemporal distribution of charging load is output, specifically as follows: A spatiotemporal distribution prediction model for electric vehicle charging load was established using a spatiotemporal convolutional neural network. The fused spatiotemporal feature data is input into a spatiotemporal convolutional neural network, and the local spatial dependency features and local temporal dependency features of the fused spatiotemporal feature data are extracted through convolution operations. Multi-level convolution and pooling operations are performed on local spatial dependency features and local temporal dependency features to generate deep feature vectors; A fully connected network layer is used to perform nonlinear mapping on deep feature vectors, and the spatiotemporal distribution prediction results of electric vehicle charging load at future time are output.

[0015] The technical effects and advantages of the present invention's method for predicting the spatiotemporal distribution of electric vehicle charging load are as follows: By performing high-precision denoising and calibration on electric vehicle battery state-of-charge (POC) data, interference from acquisition noise and outliers can be eliminated. Spectral feature extraction from POC time-series data allows for accurate identification of high-frequency resonant fluctuations in battery POC under different charging modes and durations. Spatial regional clustering analysis of resonant frequency characteristic data accurately delineates the spatial distribution regions of high-frequency charging load oscillations, providing spatial decision-making basis for grid load scheduling and charging facility deployment. Time-domain statistical analysis of resonant amplitude characteristic data yields the distribution curve of high-frequency charging load oscillations with charging duration, providing a time-dimensional fluctuation reference for charging strategy optimization and dynamic pricing. Establishing a correlation mapping model between the spatial regional distribution and time-domain distribution characteristics of charging load enhances the ability to express complex spatiotemporal coupling relationships. Based on the fusion of spatiotemporal feature data, the spatiotemporal distribution of electric vehicle charging load at future moments can be predicted, effectively supporting refined grid scheduling and charging service optimization. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of a method for predicting the spatiotemporal distribution of electric vehicle charging load according to the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention. Example

[0018] Figure 1 This invention presents a method for predicting the spatiotemporal distribution of electric vehicle charging load, which includes the following steps: Collect state-of-charge (POC) data of electric vehicle batteries under different charging modes and charging durations, denoise and calibrate the POC data, and output POC time-series data. Spectral features are extracted from the state-of-charge time series data to identify the high-frequency resonant fluctuation characteristics of the battery's state of charge, and the resonant frequency characteristic data and resonant amplitude characteristic data are output. Spatial region clustering analysis is performed on the resonant frequency characteristic data to determine the spatial region distribution that generates high-frequency oscillations in the charging load, and output spatial resonant region division data. By analyzing the time-domain fluctuation law of the resonance amplitude characteristic data, the time-domain distribution characteristics of the high-frequency oscillation of the charging load are obtained, and the time-domain characteristic curve of the load oscillation is output. Based on the spatial resonance region division data and the load oscillation time-domain characteristic curve, a correlation mapping model between the spatial regional distribution and time-domain distribution characteristics of charging load is established to generate fused spatiotemporal characteristic data. Based on the fusion of spatiotemporal feature data, the spatiotemporal distribution of electric vehicle charging load at future moments is predicted, and the prediction results of the spatiotemporal distribution of charging load are output.

[0019] The system collects state-of-charge (POC) data of electric vehicle batteries under different charging modes and charging durations, performs noise reduction and data calibration on the POC data, and outputs POC time-series data, including: Real-time acquisition of battery state-of-charge data for electric vehicles under constant power charging mode, constant voltage charging mode, and constant current charging mode, as well as under different charging duration conditions. Battery state-of-charge (SOC) data is collected by a battery SOC sensor installed inside the electric vehicle charging interface or charging equipment under constant power charging, constant voltage charging, and constant current charging modes, as well as under different charging durations. The SOC sensor can be a combination of a voltage sensor and a current sensor. The voltage sensor's accuracy can be set to 0.01 volts; for example, a voltage sensor with a range of 0 to 500 volts can be selected to ensure that the accuracy of the battery SOC data acquisition meets the requirements. The current sensor's accuracy can be set to 0.1 amperes; for example, a Hall effect current sensor with a range of 0 to 200 amperes can be selected to ensure that the real-time current data collected matches the actual current variation range required during battery charging.

[0020] The sampling frequency of the battery state of charge (SOC) sensor for battery SOC data is set according to changes in the charging mode and charging duration. For example, in constant power charging mode, the sampling frequency can be set to once per second; in constant voltage charging mode, it can be set to once every two seconds; and in constant current charging mode, it can be set to once every 0.5 seconds. For different charging durations, the sampling frequency is dynamically adjusted based on the charging time to meet the requirements of different charging stages. For example, in the initial charging stage, the sampling frequency is set to once every 0.5 seconds, and in the final charging stage, it is reduced to once every 2 seconds. The real-time acquired battery SOC data consists of battery terminal voltage, battery terminal current, and a sampling timestamp.

[0021] Wavelet denoising, anomaly identification, and data correction are performed on the collected battery state of charge data to obtain battery state of charge time series data. Wavelet denoising is performed on the acquired battery state-of-charge (POC) data to eliminate high-frequency noise interference that may occur during real-time acquisition. A suitable wavelet basis function is selected, such as the db4 wavelet basis function, to ensure that the original feature information of the battery POC data is not lost during denoising. The scale of wavelet decomposition is determined, and the appropriate number of wavelet decomposition levels is verified through multiple experiments; for example, four levels are set. Based on the selected wavelet basis function and the number of decomposition levels, wavelet decomposition is performed on the battery POC data to obtain low-frequency approximation coefficients and high-frequency detail coefficients. A thresholding method is used to threshold the high-frequency detail coefficients. A soft thresholding method can be used, determined by the VisuShrink thresholding algorithm; for example, the threshold value is the square root of the product of the standard deviation of the high-frequency detail coefficients and the natural logarithm of the data length. Subsequently, the thresholded high-frequency detail coefficients and the unprocessed low-frequency approximation coefficients are reconstructed using wavelet denoising to complete wavelet denoising, thus obtaining the denoised battery POC data.

[0022] Anomaly identification and correction are performed on the denoised battery state-of-charge (POC) data. Anomaly identification is achieved through statistical methods. Statistical analysis is performed on the denoised battery POC data, such as calculating the mean and standard deviation of data points. Anomaly thresholds are then used to identify data points by comparing them point-by-point. The 3σ principle can be used to determine the anomaly threshold; for example, data points deviating from the mean by more than three times the standard deviation are identified as anomalies. After identifying anomalies, interpolation correction is performed using cubic spline interpolation. An interpolation function is constructed using adjacent data points of the anomaly to correct the anomaly, resulting in corrected battery POC data. The corrected battery POC data more accurately reflects the actual POC changes during the battery's charging process.

[0023] The processed battery state-of-charge (POC) data is organized into a battery POC time-series data format. This data is stored in a table format, including but not limited to sampling timestamps, battery voltage, battery current, charging mode labels, and charging duration labels. Each data item is indexed by a timestamp. The charging mode labels in the table include three types: constant power charging mode, constant voltage charging mode, and constant current charging mode. The charging duration labels are based on the actual duration of the charging process, for example, divided into multiple charging duration intervals such as 0-30 minutes, 30-60 minutes, and 60-90 minutes.

[0024] Spectral feature extraction is performed on the state-of-charge time-series data to identify the high-frequency resonant fluctuation characteristics of the battery's state of charge, and the resonant frequency characteristic data and resonant amplitude characteristic data are output, including: Perform a Fourier transform on the battery state-of-charge time series data to obtain the spectral distribution of the battery state-of-charge time series data; The battery terminal voltage and battery terminal current in the battery state-of-charge (POC) time-series data are constructed as independent time-series signals, and Fourier transforms are performed on each to obtain their respective spectral distributions. To prevent signal spectral leakage, a window function, such as the Hanning window function, needs to be applied to each segment of the battery POC time-series data before performing the Fourier transform on the battery terminal voltage and battery terminal current. The length of each segment of the battery POC time-series data is set according to the charging mode and charging duration conditions. For example, in constant power charging mode, each segment is set to 600 data points; in constant voltage charging mode, each segment is set to 300 data points; and in constant current charging mode, each segment is set to 1200 data points, to adapt to the timing variation characteristics of each charging mode.

[0025] The resonant frequency and resonant amplitude characteristic parameters of the battery state of charge under constant power charging mode, constant voltage charging mode, constant current charging mode, and different charging duration conditions are extracted from the spectrum distribution. In the battery terminal voltage spectrum and battery terminal current spectrum, specific frequency ranges are selected. For example, using a spectral peak search method, the frequency positions corresponding to characteristic peaks with amplitudes higher than the average spectral amplitude are identified and determined. The frequency corresponding to the characteristic peak with an amplitude higher than the average spectral amplitude is defined as the resonant frequency characteristic parameter. The resonant amplitude characteristic parameter is extracted as follows: after determining the characteristic peak, the difference between the amplitude height of the characteristic peak and the average amplitude of its neighborhood is used as the resonant amplitude characteristic parameter. To ensure the objectivity and accuracy of the extracted resonant frequency and resonant amplitude characteristic parameters, the characteristic peak judgment threshold is determined using the spectral signal-to-noise ratio criterion. For example, a characteristic peak amplitude must be 5 times or more higher than the average amplitude of its neighborhood to be considered a valid resonant characteristic peak.

[0026] Based on the resonant frequency characteristic parameters and the resonant amplitude characteristic parameters, the characteristic data of high-frequency resonant fluctuations generated by the battery state of charge are identified and determined, and the resonant frequency characteristic data and the resonant amplitude characteristic data are output respectively. The resonant frequency characteristic data includes the resonant frequency characteristic parameters corresponding to different charging duration conditions under each charging mode, as well as the corresponding charging mode label and charging duration interval label. Each resonant frequency characteristic data item consists of information such as characteristic frequency, charging mode type, and charging duration interval range. The resonant amplitude characteristic data includes the resonant amplitude characteristic parameters corresponding to different charging duration conditions under each charging mode, and is also equipped with a charging mode label and charging duration interval label. Each resonant amplitude characteristic data item consists of information such as characteristic amplitude, charging mode type, and charging duration interval range.

[0027] To identify the characteristic data of high-frequency resonant fluctuations in the battery's state of charge (SOC), these fluctuations are determined based on resonant frequency and amplitude characteristic data. The frequency and distribution of resonant frequency and amplitude characteristic data are calculated under various charging modes and durations. For example, the number of occurrences of a specific resonant frequency, the average value of the corresponding amplitude parameter, and its distribution range are statistically analyzed. A joint distribution matrix of resonant frequency and amplitude is established based on the statistical results, using resonant frequency as the row index and resonant amplitude as the column index. Each matrix cell represents the frequency of occurrence of the corresponding frequency and amplitude combination. Representative high-frequency resonant fluctuation characteristic data of the SOC are identified and determined based on data combinations in the joint distribution matrix whose frequencies exceed a predetermined threshold. The frequency threshold in the joint distribution matrix is ​​set to be greater than 1.5 times the average value of all frequency combinations in the matrix. For example, when the frequency in a matrix cell exceeds 1.5 times the average frequency of the entire matrix, it is considered valid high-frequency resonant fluctuation characteristic data. The identified high-frequency resonant fluctuation characteristic data of the SOC are output separately in the form of resonant frequency and amplitude characteristic data.

[0028] Spatial region clustering analysis is performed on the resonant frequency characteristic data to determine the spatial region distribution that generates high-frequency oscillations in the charging load, and the spatial resonant region division data is output, including: Using the geographical coordinates, service radius, and grid connection node identifier of the charging equipment as spatial indexes, the resonant frequency characteristic data is mapped to the location of the charging equipment. Each charging device's geographic coordinates are uniquely determined using longitude and latitude. The service radius represents the spatial coverage area served by the charging device, while the grid node identifier characterizes the connection relationship between the charging device and the grid node. Geographic coordinates are determined using a Global Positioning System (GPS) receiver; for example, the measurement accuracy can be set to an error range of less than 5 meters. The service radius is determined based on the charging device's rated power and the user density of the coverage area; for example, a charging device with a rated power of 120 kW can have a service radius of 500 meters. The grid node identifier is a unique identifier provided by the power system to identify the grid node to which the charging device is connected.

[0029] Spatial region clustering analysis is performed on the mapped resonant frequency characteristic data on a preset spatial grid to obtain a set of spatial cluster labels and a set of cluster center coordinates; Using the minimum latitude and longitude coordinates of the charging equipment distribution area as the origin and the maximum latitude and longitude coordinates as the boundary, a regular rectangular grid is generated. The size of each grid cell is determined according to the actual analysis requirements. For example, the longitude span and latitude span of each grid cell can be set to 0.01 degrees.

[0030] The mapped resonant frequency characteristic data are placed in a pre-established spatial grid for spatial region clustering analysis. The spatial region clustering analysis employs a density clustering algorithm. The density clustering algorithm clusters data by setting a core density threshold and a radius threshold, calculating the number of resonant frequency characteristic data points of charging devices within each grid cell, which is then used as the density value of the grid cell. The density threshold is set by statistically analyzing the distribution of density values ​​across all grid cells, selecting grid cells with a density value higher than the average density value as core grid cells; for example, it can be set to twice the overall density average. A radius threshold is set to determine spatial connectivity; for example, it can be set to the span of two grid cells, i.e., 0.02 degrees. Based on the density and radius thresholds, the core grid cell is determined by traversing all grid cells. Starting from the core grid cell, the clustering region is expanded, and adjacent grid cells that meet the density requirements and are within the radius threshold range are clustered, ultimately forming multiple spatial clusters. After clustering, the output is a set of spatial cluster labels and a set of cluster center coordinates. The set of cluster labels is represented in integer form, for example, denoted as cluster label 1, cluster label 2, etc.; the cluster center coordinates are obtained by taking the average of the geographic coordinates of all grid cells in each spatial cluster.

[0031] Perform boundary trimming and connectivity verification on the spatial clusters according to the administrative division boundaries or power grid feeder boundaries to generate a set of spatial resonant regions; Administrative division boundaries or power grid feeder boundaries are imported from a Geographic Information System (GIS) and stored in polygon vector data format. Boundary clipping processing includes: overlaying the resulting spatial clusters with the administrative division boundaries or power grid feeder boundaries; using spatial overlay analysis, geometric clipping is performed on each spatial cluster to confine its polygonal region to the actual boundary area; connectivity verification is performed on the clipped polygonal regions of the spatial clusters to remove isolated regions not connected to the main region. After boundary clipping and connectivity verification, the final set of spatial resonance regions is obtained. Each spatial resonance region is stored as a geographic polygon, containing geographic coordinate data in the format of a sequence of vertex coordinates of the geographic polygon. For example, the number of vertex coordinates for each region can be set to no less than three, with coordinate precision to six decimal places for latitude and longitude, to determine the accurate range of each spatial resonance region.

[0032] For each spatial resonance region, a record is generated containing the coordinates of the region's polygon vertices, the number of charging devices in the region, and the statistics of the resonant frequency in the region, forming spatial resonance region division data. The number of charging devices within a region is determined by counting the number of charging device location data points in each spatial resonant region. The resonant frequency statistics within a region include the mean, median, maximum, and minimum resonant frequencies. The calculation method for the resonant frequency statistics is as follows: average the resonant frequency characteristic data corresponding to all charging devices within the region; sort the data by resonant frequency values ​​and take the middle value to obtain the median; and take the maximum and minimum frequency data points to obtain the maximum and minimum values.

[0033] The coordinates of the polygon vertices, the number of charging devices, and the statistical data of the resonant frequency for each spatial resonant region are stored in the form of structured data records to form spatial resonant region division data. The spatial resonant region division data is stored in the form of standardized data tables. Each spatial resonant region division data item consists of a region number, a list of polygon vertex coordinates, the total number of charging devices, the mean resonant frequency, the median resonant frequency, the maximum resonant frequency, and the resonant frequency.

[0034] Time-domain fluctuation analysis was performed on the resonance amplitude characteristic data to obtain the time-domain distribution characteristics of the high-frequency oscillation of the charging load, and the time-domain characteristic curve of the load oscillation was output, including: Based on the resonance amplitude characteristic data, multiple charging time intervals are divided according to different charging modes of the charging equipment; Charging modes include constant power charging, constant voltage charging, and constant current charging. For each charging mode, the resonant amplitude characteristic data is divided into multiple charging time intervals based on the actual charging time of the charging device. The method for dividing the charging time intervals is as follows: based on the actual charging needs of the battery and the typical characteristics of the charging process, combined with the charging time distribution characteristics obtained through statistical analysis, the charging time can be determined. For example, the charging time can be divided into multiple fixed continuous time intervals such as 0-30 minutes, 30-60 minutes, 60-90 minutes, and 90-120 minutes. The number of data points contained in each interval must meet the minimum sample size requirement of the time-domain analysis statistical method; for example, the number of data points in a single interval should be no less than 50 data samples.

[0035] For the resonance amplitude characteristic data within each charging time interval, time-domain statistical analysis methods are used to calculate and generate the corresponding mean curve, variance curve, and fluctuation amplitude range. Statistical analysis methods include, but are not limited to, mean analysis, variance analysis, and fluctuation amplitude range analysis. The mean analysis method involves summing all data points of the resonance amplitude characteristic data within each charging time interval and dividing by the total number of data points to obtain the average value of the resonance amplitude characteristic data within that charging time interval, thus generating a mean curve. The variance analysis method involves squaring the difference between each data point and the average value within the corresponding charging time interval, summing all the squares, and then dividing by the total number of data points to determine the variance value of the resonance amplitude characteristic data within each charging time interval, thus generating a variance curve. The fluctuation amplitude range analysis method involves determining the difference between the maximum and minimum values ​​of the resonance amplitude characteristic data points within each charging time interval, using this difference as the fluctuation amplitude range, thereby determining the range of data fluctuation within the corresponding charging time interval.

[0036] Based on the mean curve, variance curve, and fluctuation amplitude range, the distribution characteristics of high-frequency oscillation of charging load with charging duration are determined, and the time-domain characteristic curve of load oscillation corresponding to each charging duration interval is formed. The mean curve reflects the central tendency of the resonant amplitude within each charging time interval; the variance curve reflects the dispersion of the resonant amplitude characteristic data, i.e., the stability of the resonant amplitude variation with charging time interval; and the fluctuation amplitude range reflects the range of the resonant amplitude characteristic data variation with charging time interval. The time-domain distribution characteristics of load oscillations are determined jointly by the mean curve, variance curve, and fluctuation amplitude range. For example, when the mean curve shows a trend of first increasing and then decreasing with increasing charging time, the variance curve shows a trend of gradually decreasing with increasing charging time, and the fluctuation amplitude range shows a gradual narrowing with increasing charging time, the time-domain distribution characteristics of the load oscillations corresponding to the charging time interval are determined to be a process of gradually weakening oscillation intensity.

[0037] For each charging duration interval, a corresponding load oscillation time-domain characteristic curve is generated. The method for generating the load oscillation time-domain characteristic curve is as follows: Plot the charging duration interval as the x-axis and the mean resonance amplitude of the corresponding charging duration interval as the y-axis to show the change of the mean amplitude with charging duration; then plot the corresponding error ranges with the upper and lower limits of the mean and variance of the resonance amplitude to show the fluctuation of the resonance amplitude over time; finally, mark the corresponding numerical range on the y-axis to provide an indication of the oscillation amplitude range. Therefore, the load oscillation time-domain characteristic curve corresponding to each charging duration interval can reflect the change law of the resonance amplitude characteristic data with charging duration, and simultaneously reflect the changing trend of oscillation intensity.

[0038] Summarize the load oscillation time-domain characteristic curves for each charging time interval and output the load oscillation time-domain characteristic curve data; The data of all load oscillation time-domain characteristic curves are grouped and organized according to charging mode type, forming a data structure with charging mode as the primary category and charging duration interval as the secondary category. Each data structure represents the oscillation characteristics of the corresponding charging mode and charging duration interval, including calculated time-domain characteristic parameters such as mean, variance, and fluctuation amplitude range. The summarized load oscillation time-domain characteristic curve data is organized in the form of a structured data table. Each data record includes a charging mode label, charging duration interval, mean resonance amplitude, variance of resonance amplitude, fluctuation amplitude range of resonance amplitude, and the corresponding characteristic curve data point set.

[0039] Based on spatial resonance region segmentation data and load oscillation time-domain characteristic curves, a correlation mapping model between the spatial regional distribution and time-domain distribution characteristics of charging load is established, generating fused spatiotemporal characteristic data, including: Based on the set of spatial cluster labels, the set of cluster center coordinates, and the coordinates of the vertex of the region polygon in the spatial resonance region partitioning data, the spatial feature vector of each spatial resonance region is determined. The determination of spatial feature vectors includes geometric feature extraction and statistical feature calculation. Geometric feature extraction is based on the vertex coordinates of the region polygon, which are stored with six decimal places of latitude and longitude, and the sequence length is no less than three vertices. The area of ​​each spatial resonance region is calculated using the spherical polygon area formula. After converting the latitude and longitude coordinates to planar coordinates, the shoelace formula is applied. For example, the coordinate transformation uses the universal transverse Mercator projection, with the region center as the origin, ensuring that the area calculation error is less than 1 square meter. The perimeter of each spatial resonance region is calculated by summing the spherical distances between adjacent vertices. The spherical distance is calculated using the Haversine formula, inputting the latitude and longitude values ​​of the vertex coordinates and outputting the distance value in meters. The shape index is calculated, defined as the ratio of the square of the perimeter to 4π multiplied by the area, used to quantify the complexity of the region's shape; a larger ratio indicates a more irregular shape. Statistical feature calculation is based on the number of charging devices and the resonant frequency statistics within the region. The charging device density is obtained by dividing the total number of charging devices by the region area. The resonant frequency characteristics include the mean resonant frequency, the variance of the resonant frequency, and the resonant frequency range. The variance of the resonant frequency is obtained by calculating the squared standard deviation of the resonant frequency characteristic data corresponding to all charging devices in the region. The resonant frequency range is obtained by subtracting the minimum resonant frequency from the maximum resonant frequency. The construction of the spatial feature vector combines the above features into a fixed-dimensional vector, with the vector dimension set to 6 dimensions, including area, perimeter, shape index, charging device density, mean resonant frequency, and variance of the resonant frequency. Each feature value is normalized before combination. The normalization method uses min-max scaling, mapping the original value to the interval between 0 and 1. The scaling parameter is determined based on historical data statistics. For example, when normalizing the area, the minimum value is the minimum area of ​​all regions, and the maximum value is the maximum area of ​​all regions.

[0040] Based on the load oscillation time-domain characteristic curves of each charging time interval in the load oscillation time-domain characteristic curve data, the time-domain characteristic vector of each charging time interval is determined. The determination of the time-domain feature vector includes curve feature extraction and statistical aggregation. Curve feature extraction is performed on the load oscillation time-domain feature curve for each charging time interval. The load oscillation time-domain feature curve is stored in the form of discrete data points, with the horizontal axis representing the midpoint value of the charging time interval and the vertical axis representing the mean resonance amplitude, along with error range data. Key statistics are extracted from the mean curve, including the mean, slope, curvature, and peak position. The mean is obtained by taking the arithmetic mean of the vertical coordinate values ​​of all data points on the curve; the slope is obtained by calculating the regression coefficients by fitting the curve data points using linear regression, with the least squares method as the objective function to minimize the sum of squared residuals; the curvature is obtained by calculating the approximate value of the second derivative of the curve using the central difference method, with the step size set to the interval between adjacent data points; the peak position is determined by finding local maximum points, which are defined as points whose vertical coordinate values ​​are greater than the vertical coordinate values ​​of the two data points before and after them. The variance mean and variance change rate are extracted from the variance curve. The variance mean is obtained by taking the arithmetic mean of the ordinate values ​​of the data points on the variance curve; the variance change rate is obtained by the ratio of the difference between the endpoints of the variance curve to the time span. The amplitude range mean and amplitude range range are extracted from the fluctuation amplitude range. The amplitude range mean is obtained by taking the arithmetic mean of the fluctuation amplitude range for each charging time interval; the amplitude range range is obtained by subtracting the minimum fluctuation amplitude range from the maximum fluctuation amplitude range. The construction of the time-domain feature vector combines the above features into a fixed-dimensional vector with 8 dimensions, including the mean curve mean, mean curve slope, mean curve curvature, number of mean curve peaks, variance curve mean, variance change rate, fluctuation amplitude range mean, and fluctuation amplitude range range. Each feature value is standardized before combination using Z-score standardization, which involves subtracting the mean of historical data and dividing by the standard deviation. The mean and standard deviation are calculated based on the feature values ​​of all charging time intervals.

[0041] Based on the spatial feature vector and the temporal feature vector, a mapping relationship is constructed between the spatial feature vector and the temporal feature vector of the charging load through correlation mapping, thereby obtaining the fused spatiotemporal feature data of the charging load; The association mapping employs a multiple linear regression model. The input to the multiple linear regression model is a spatial feature vector, and the output is a temporal feature vector. The mathematical form of the multiple linear regression model is Y = XW + B, where Y represents the temporal feature vector, X represents the spatial feature vector, W represents the weight matrix, and B represents the bias vector. The weight matrix W has a dimension of 6 rows and 8 columns, and the bias vector B has a dimension of 8. Model parameters are determined through training on historical data. The training data includes spatial feature vectors and corresponding temporal feature vectors from multiple historical time points, covering different dates and time periods; for example, data from the hourly times of each day within the past 30 days are selected. The training process uses the least squares method to solve for the weight matrix W and the bias vector B, with the objective function being the minimization of the sum of squared prediction errors. The least squares solution is obtained through the normal equation. To prevent overfitting, an L2 regularization term is added to the normal equation. The regularization parameter λ is determined through cross-validation; for example, λ is selected by performing a grid search with a step size of 0.01, ranging from 0.01 to 1.0, to minimize the mean squared error of the validation set. After the mapping relationship is established, for any new spatial feature vector, the trained multiple linear regression model is applied to predict the corresponding temporal feature vector. The predicted temporal feature vector is then concatenated with the original spatial feature vector to form fused spatiotemporal feature data. The fused spatiotemporal feature data is a 14-dimensional vector, with the first 6 dimensions representing the spatial feature vector and the last 8 dimensions representing the predicted temporal feature vector. The concatenated vector undergoes final normalization using a min-max scaling method, with parameters determined statistically based on the training data. The fused spatiotemporal feature data is represented in vector form.

[0042] Based on fused spatiotemporal feature data, the spatiotemporal distribution of electric vehicle charging load at future times is predicted, and the predicted spatiotemporal distribution of charging load is output, including: A spatiotemporal distribution prediction model for electric vehicle charging load was established using a spatiotemporal convolutional neural network. The spatiotemporal convolutional neural network comprises an input layer, multiple convolutional layers, pooling layers, fully connected layers, and an output layer. The input layer receives fused spatiotemporal feature data. Before input, this fused spatiotemporal feature data needs to be organized into spatial grid time series data. The spatial grid is based on the preset spatial grid defined in step S3. Each grid cell of the preset spatial grid has a fixed longitude and latitude span, for example, a longitude span of 0.01 degrees and a latitude span of 0.01 degrees. Each grid cell is assigned a fused spatiotemporal feature vector based on its spatial resonance region. The assignment method is as follows: if the grid cell is completely inside the polygon of a spatial resonance region, the fused spatiotemporal feature vector of that region is used directly; if the grid cell is covered by multiple spatial resonance regions, the fused spatiotemporal feature vector of the region with the largest coverage area is used; if the grid cell is not covered by any spatial resonance region, the fused spatiotemporal feature vectors of neighboring grid cells are used for filling through spatial interpolation. The spatial interpolation uses an inverse distance weighting method, with the weight inversely proportional to the distance to the center point of the grid cell. The distance threshold is set to twice the preset span of the spatial grid cell, for example, 0.02 degrees. The time series dimension is constructed using historical time points, covering multiple past moments. For example, a time point is selected every 30 minutes from the past 24 hours, for a total of 48 time steps. Each time step corresponds to a fused spatiotemporal feature data grid for a given moment. Therefore, the input data is a four-dimensional tensor, with dimensions equal to the number of grid rows multiplied by the number of grid columns multiplied by the number of time steps multiplied by the feature dimension. For example, the number of grid rows is 100, the number of grid columns is 100, the number of time steps is 48, and the feature dimension is 14. The tensor data is standardized before input using Z-score standardization. The mean and standard deviation of each feature dimension are calculated based on historical data, which includes input tensors from the same time period within the past 30 days.

[0043] The convolutional layers of the spatiotemporal convolutional neural network are designed to extract local spatial and temporal dependent features. Local spatial dependent features are extracted using two-dimensional convolutional layers. These layers slide their kernels along the spatial dimensions (grid rows and columns). The kernel size is set according to the spatial resolution, determined by analyzing the spatial autocorrelation function to define the local spatial range. For example, if the spatial autocorrelation function decays to 0.5 within a 5-grid distance, a 5x5 kernel size is chosen. The number of kernels is determined through feature importance analysis. This analysis uses a random forest algorithm to calculate the Gini importance of each feature dimension. The number of features with importance higher than the average is selected as the initial number of kernels. For example, with a feature dimension of 14, an average Gini importance of 0.07, and an importance threshold of 0.07, 8 kernels are chosen. The stride is set to 1, and the same padding method is used to maintain the spatial dimension. The ReLU activation function is chosen; ReLU is defined as the output being the maximum of the input and zero, introducing non-linearity. Local temporal dependent features are extracted using a one-dimensional convolutional layer. This layer slides the convolutional kernel along the temporal dimension. The kernel size is determined based on the autocorrelation function of the time series; for example, if the autocorrelation function decays to 0.6 over three time steps, a kernel size of 3 is chosen. The number of kernels is determined through temporal feature variance analysis. The variance of the features at each time step is calculated, and the number of time steps with a large rate of change in variance is selected. For example, if the rate of change in variance is greater than 1.5 times the overall variance, a kernel size of 6 is chosen. The convolution stride is set to 1, and the same padding method is used. The ReLU activation function is also selected. Spatial and temporal convolutions are performed sequentially, with spatial convolution performed first, followed by temporal convolution, to progressively extract spatial and temporal features.

[0044] The fused spatiotemporal feature data is input into a spatiotemporal convolutional neural network, and the local spatial dependency features and local temporal dependency features of the fused spatiotemporal feature data are extracted through convolution operations. Input data flows into the spatiotemporal convolutional neural network in the form of a four-dimensional tensor. It first passes through a spatial convolutional layer, which applies two-dimensional convolution operations. These two-dimensional convolution operations perform dot product and summation between the convolution kernel and local spatial regions of the input tensor, outputting a feature map. The feature map size is the same as the input spatial dimension; for example, if the input spatial dimension is 100 rows and 100 columns, the output feature map will be 100 rows and 100 columns with 8 channels. Subsequently, the feature map is passed to a temporal convolutional layer, which applies one-dimensional convolution operations. These one-dimensional convolution operations slide along the temporal dimension, convolving the time series at each spatial location, outputting a feature map with a dimension of 100 rows and 100 columns, 48 ​​time steps, and 6 channels. The weights in the convolution operation are initialized using the Xavier initialization method. Xavier initialization calculates the weight range based on the number of input and output channels, defined as a range from negative √6 divided by the sum of the number of input and output channels to positive √6 divided by the sum of the number of input and output channels. For example, if the number of input channels is 8 and the number of output channels is 6, the weight range is approximately -0.35 to +0.35. The bias parameters are initialized to 0. The output of the convolution operation undergoes a non-linear transformation using the ReLU activation function.

[0045] Multi-level convolution and pooling operations are performed on local spatial dependency features and local temporal dependency features to generate deep feature vectors; The multi-layer convolutional network consists of alternating spatial and temporal convolutional layers. The number of layers is determined based on data complexity and overfitting risk. The layer selection method involves analyzing the validation set error curve; the number of layers is stopped when the validation set error no longer significantly decreases. For example, three spatial convolutional layers and three temporal convolutional layers are set. The kernel size of the second spatial convolutional layer is set to 3 rows and 3 columns, and the number of kernels is determined by a multiple of the number of feature map channels. For example, the first layer outputs 8 channels, and the second layer is set to 16 channels. The kernel size of the third spatial convolutional layer is set to 3 rows and 3 columns, and the number of kernels is set to 32 channels. The corresponding temporal convolutional layers all have a kernel size of 3, and the number of kernels is set to 12, 24, and 48 channels respectively. A pooling layer is added after each convolutional layer. Max pooling is used, with a 2x2 window and a stride of 2 in the spatial dimension for dimensionality reduction and enhanced feature invariance; and max pooling with a window size of 2 and a stride of 2 is used in the temporal dimension. The size of the feature map after pooling is gradually reduced. For example, the spatial dimension is reduced from 100 rows and 100 columns to 50 rows and 50 columns, then 25 rows and 25 columns; the temporal dimension is reduced from 48 time steps to 24 time steps, then 12 time steps. After multiple convolutions and pooling, the output feature map is flattened into a one-dimensional vector. For example, the final feature map dimension is 25 rows and 25 columns, 12 time steps, and 48 channels. The flattened vector length is 25 x 25 x 12 x 48, which equals 360,000 dimensions. The flattened vector is regularized using a dropout layer. The dropout rate is determined by cross-validation, using 5-fold cross-validation. The dropout rate is tested from 0.2 to 0.5 with a step size of 0.1, and the value with the minimum loss on the validation set is selected. For example, the dropout rate is set to 0.3. The deep feature vector is output from the dropout layer.

[0046] A fully connected network layer is used to perform nonlinear mapping on deep feature vectors, and the spatiotemporal distribution prediction results of electric vehicle charging load at future time are output. A fully connected network layer consists of two hidden layers and one output layer. The number of neurons in the hidden layers is determined by the dimension of the feature vector and the complexity of the prediction task. The number of neurons in the first hidden layer is 1 / 10 of the dimension of the flattened vector; for example, if the dimension of the flattened vector is 360,000, then the number of neurons in the first hidden layer is 36,000. The number of neurons in the second hidden layer is half of the number in the first hidden layer, for example, 18,000. The ReLU activation function is used. The number of neurons in the output layer is consistent with the prediction target, which is the spatiotemporal distribution of charging load at future time points. For example, predicting the charging load value of each grid cell one hour later, therefore the number of neurons in the output layer equals the total number of grid cells; for example, in a 100x100 grid, the number of neurons in the output layer is 10,000. The output layer activation function uses a linear activation function to output continuous values. The weights of the fully connected layer are initialized using the He initialization method. He initialization calculates the weight range based on the number of input neurons, defined as from negative √6 divided by the number of input neurons to positive √6 divided by the number of input neurons. For example, if the number of input neurons is 360,000, the weight range is approximately -0.000128 to positive 0.000128. The bias parameters are initialized to 0. Historical data is used during training, including the input tensor and the corresponding actual charging load values, obtained from the power grid monitoring system. The loss function is mean squared error, calculated as the average of the squared differences between the predicted and actual values. The optimizer is the Adam optimizer, whose learning rate is determined through learning rate scheduling. The initial learning rate is set to 0.001, multiplied by 0.9 every 10 training epochs. The number of training epochs is determined using early stopping, which monitors the validation set loss; training stops when the validation set loss does not decrease for 10 consecutive epochs. The model outputs a prediction of the spatiotemporal distribution of charging load at future times. The results are represented in a grid format, with each grid cell containing a load value in kilowatts. The data is in a two-dimensional array format and can be visualized using a geographic information system. The prediction results are used for power grid dispatching and planning.

[0047] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. The semiconductor medium can be a solid-state drive.

[0048] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0049] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and modules described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0050] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or modules may be electrical, mechanical, or other forms.

[0051] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0052] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0053] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0054] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0055] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for predicting spatiotemporal distribution of electric vehicle charging load, characterized in that, Includes the following steps: The system collects state-of-charge (POC) data of electric vehicle batteries under different charging modes and charging durations, performs noise reduction and data calibration on the POC data, and outputs POC time-series data. It also extracts spectral features from the POC time-series data to identify the high-frequency resonant fluctuation characteristics of the battery's POC, and outputs resonant frequency characteristic data and resonant amplitude characteristic data. Spatial region clustering analysis is performed on the resonant frequency characteristic data to determine the spatial region distribution that generates high-frequency oscillations in the charging load, and output spatial resonant region division data. The time-domain fluctuation law of the resonance amplitude characteristic data is analyzed to obtain the time-domain distribution characteristics of the high-frequency oscillation of the charging load, and the time-domain characteristic curve of the load oscillation is output. Based on the spatial resonance region division data and the time-domain characteristic curve of the load oscillation, a correlation mapping model between the spatial regional distribution and the time-domain distribution characteristics of the charging load is established to generate fused spatiotemporal characteristic data. Based on the fused spatiotemporal characteristic data, the spatiotemporal distribution of the electric vehicle charging load at future times is predicted, and the spatiotemporal distribution prediction result of the charging load is output. 2.The method of claim 1, wherein, The system collects state-of-charge (SOC) data of electric vehicle batteries under different charging modes and charging durations, performs noise reduction and data calibration on the SOC data, and outputs SOC time-series data, specifically: Real-time acquisition of battery state-of-charge data for electric vehicles under constant power charging mode, constant voltage charging mode, and constant current charging mode, as well as under different charging duration conditions. Wavelet denoising, anomaly identification, and data correction are performed on the collected battery state of charge data to obtain battery state of charge time series data. 3.The method of claim 2, wherein, Battery state-of-charge data consists of battery terminal voltage, battery terminal current, and sampling timestamp. 4.The method of claim 3, wherein, The battery state of charge timing data consists of sampling timestamp, battery voltage, battery current, charging mode label, and charging duration label.

5. The method of claim 4, wherein, Spectral feature extraction is performed on the state-of-charge time-series data to identify the high-frequency resonant fluctuation characteristics of the battery's state of charge, and the resonant frequency characteristic data and resonant amplitude characteristic data are output, specifically: Perform a Fourier transform on the battery state-of-charge time series data to obtain the spectral distribution of the battery state-of-charge time series data; The resonant frequency and resonant amplitude characteristic parameters of the battery state of charge under constant power charging mode, constant voltage charging mode, constant current charging mode, and different charging duration conditions are extracted from the spectrum distribution. Based on the resonant frequency characteristic parameters and the resonant amplitude characteristic parameters, the characteristic data of high-frequency resonant fluctuations generated by the battery state of charge are identified and determined, and the resonant frequency characteristic data and the resonant amplitude characteristic data are output respectively.

6. The method of claim 5, wherein the method further comprises: Spatial region clustering analysis is performed on the resonant frequency characteristic data to determine the spatial region distribution that generates high-frequency oscillations in the charging load, and the spatial resonant region division data is output, specifically: Using the geographical coordinates, service radius, and grid connection node identifier of the charging equipment as spatial indexes, the resonant frequency characteristic data is mapped to the location of the charging equipment. Spatial region clustering analysis is performed on the mapped resonant frequency characteristic data on a preset spatial grid to obtain a set of spatial cluster labels and a set of cluster center coordinates; Perform boundary trimming and connectivity verification on the spatial clusters according to the administrative division boundaries or power grid feeder boundaries to generate a set of spatial resonant regions; Generate spatial resonance region partitioning data for each spatial resonance region.

7. The method of claim 6, wherein the method further comprises: The spatial resonance region division data includes the coordinates of the vertices of the region polygon, the number of charging devices in the region, and the statistics of the resonant frequency in the region.

8. The method for predicting the spatiotemporal distribution of electric vehicle charging load according to claim 7, characterized in that, Time-domain fluctuation analysis was performed on the resonance amplitude characteristic data to obtain the time-domain distribution characteristics of the high-frequency oscillation of the charging load, and the time-domain characteristic curve of the load oscillation was output, as follows: Based on the resonance amplitude characteristic data, multiple charging time intervals are divided according to different charging modes of the charging equipment; For the resonance amplitude characteristic data within each charging time interval, time-domain statistical analysis methods are used to calculate and generate the corresponding mean curve, variance curve, and fluctuation amplitude range. Based on the mean curve, variance curve, and fluctuation amplitude range, the distribution characteristics of high-frequency oscillation of charging load with charging duration are determined, and the time-domain characteristic curve of load oscillation corresponding to each charging duration interval is formed. Summarize the load oscillation time-domain characteristic curves for each charging time interval and output the load oscillation time-domain characteristic curve data.

9. The method for predicting the spatiotemporal distribution of electric vehicle charging load according to claim 8, characterized in that, Based on the spatial resonance region segmentation data and the load oscillation time-domain characteristic curve, a correlation mapping model between the spatial regional distribution and time-domain distribution characteristics of charging load is established to generate fused spatiotemporal characteristic data, specifically: Based on the set of spatial cluster labels, the set of cluster center coordinates, and the coordinates of the vertex of the region polygon in the spatial resonance region partitioning data, the spatial feature vector of each spatial resonance region is determined. Based on the load oscillation time-domain characteristic curves of each charging time interval in the load oscillation time-domain characteristic curve data, the time-domain characteristic vector of each charging time interval is determined. Based on the spatial and temporal feature vectors, a mapping relationship is constructed between the spatial feature vector and the temporal feature vector of the charging load through correlation mapping, thereby obtaining the fused spatiotemporal feature data of the charging load.

10. The method for predicting the spatiotemporal distribution of electric vehicle charging load according to claim 9, characterized in that, Based on fused spatiotemporal feature data, the spatiotemporal distribution of electric vehicle charging load at future times is predicted, and the predicted spatiotemporal distribution of charging load is output as follows: A spatiotemporal distribution prediction model for electric vehicle charging load was established using a spatiotemporal convolutional neural network. The fused spatiotemporal feature data is input into a spatiotemporal convolutional neural network, and the local spatial dependency features and local temporal dependency features of the fused spatiotemporal feature data are extracted through convolution operations. Multi-level convolution and pooling operations are performed on local spatial dependency features and local temporal dependency features to generate deep feature vectors; A fully connected network layer is used to perform nonlinear mapping on deep feature vectors, and the spatiotemporal distribution prediction results of electric vehicle charging load at future time are output.

Images