A regional charging pile demand prediction method and system based on big data analysis

Abstract

The application discloses a regional charging pile demand prediction method and system based on big data analysis, relates to the technical field of big data analysis, and comprises the following steps: S1, collecting multi-source charging demand correlation data to generate an original data set; S2, standardizing and extracting demand influence characteristics; S3, constructing a grid charging demand digital model; S4, constructing a charging demand space-time atlas; S5, inputting an improved Graph WaveNet network for prediction to generate a basic demand prediction result; S6, performing demand elasticity correction to generate a charging pile demand prediction result; and S7, performing supply-demand gap analysis, priority sorting and access checking to output decision information. The application adopts a regional charging demand space-time atlas, an improved Graph WaveNet network and demand elasticity correction, realizes accurate prediction of regional charging pile demand and optimization of construction decisions.

Description

Technical Field

[0001] This invention relates to the field of big data analytics, and in particular to a method and system for predicting regional charging pile demand based on big data analytics. Background Technology

[0002] With the continuous growth of new energy vehicle ownership and urban public energy replenishment demand, forecasting technologies for the scale, spatial layout, and time-of-day demand changes of regional charging piles have attracted widespread attention. Existing regional charging pile demand forecasting methods mainly rely on historical charging order statistics, vehicle ownership estimation, manual experience judgment, or single time series models to extrapolate demand, and combine existing charging facility utilization rates and some traffic flow data to determine the location of charging piles. However, in practical applications, the following problems commonly exist: Regional charging demand is influenced by multiple factors, including vehicle travel intensity, parking resource supply, existing charging facility service capacity, regional functional attributes, and grid access conditions. Existing methods typically only focus on historical charging frequency or charging volume, making it difficult to accurately reflect the spatial correlation and service substitution relationships between different regions. This leads to under-prediction in high-demand areas such as commercial districts, hospitals, and transportation hubs, while low-density areas may experience redundant construction. Furthermore, the dispersed nature of multi-source data, with a lack of unified identification, spatial calibration, and temporal alignment among vehicle travel data, parking resource data, existing charging facility operation data, and grid access data, results in inconsistent prediction samples, temporal misalignments, and unclear spatial attributions. Traditional statistical models or ordinary time series models struggle to simultaneously model short-term fluctuations, intraday variations, and weekly patterns of charging demand. They also fail to incorporate parking capacity, existing charging facility service saturation, and grid access capacity into the prediction correction process, leading to a disconnect between prediction results and actual construction feasibility. This affects the accuracy of prioritizing regional charging pile construction and verifying access constraints.

[0003] Therefore, how to provide a method and system for predicting regional charging pile demand based on big data analysis is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0004] One objective of this invention is to propose a method and system for predicting regional charging pile demand based on big data analysis. This invention employs a method combining multi-source charging demand correlation data fusion, a regional charging demand spatiotemporal map, and an improved GraphWaveNet network. It constructs a gridded regional charging demand digital model by uniformly identifying, spatiotemporally calibrating, and gridding vehicle travel data, parking resource data, existing charging facility operation data, and grid access correlation data. Based on this, it introduces a multi-relationship graph structure learning layer, a multi-scale temporal convolutional layer, and a resource constraint fusion layer to jointly model the spatial correlations, charging service substitution relationships, and temporal variation patterns between demand prediction grids, achieving basic regional charging demand prediction. Furthermore, it incorporates electric vehicle growth trends, existing charging facility service utilization status, and regional resource carrying capacity status to perform demand elasticity correction, and outputs supply-demand gap analysis, construction priority ranking, and access constraint verification results. This method has the advantages of high prediction accuracy, strong resource constraint adaptability, reliable construction recommendation results, and applicability to regional charging infrastructure planning decisions.

[0005] A method for predicting regional charging pile demand based on big data analysis according to an embodiment of the present invention includes the following steps: S1. Collect multi-source charging demand correlation data within the target area, perform unified identification and spatiotemporal calibration on the multi-source charging demand correlation data, and generate the original dataset of regional charging demand. S2. Standardize the original dataset of regional charging demand and extract the regional charging demand influencing features to generate a large data feature set of regional charging demand. S3. Based on the regional spatial basic data and regional resource constraint data, the big data feature set of regional charging demand is mapped to the demand prediction grid to generate a gridded regional charging demand digital model. S4. Extract the regional demand association nodes based on the gridded regional charging demand digital model, and construct a spatiotemporal map of regional charging demand. S5. Input the spatiotemporal map of regional charging demand into the improved Graph WaveNet network to jointly predict the spatial correlation and temporal variation of regional charging demand and generate the prediction result of regional basic charging demand. S6. Based on the regional basic charging demand forecast results and regional demand correction constraint information, perform demand elasticity correction to generate regional charging pile demand forecast results. S7. Based on the regional charging pile demand forecast results, perform supply and demand gap analysis, construction priority ranking, and access constraint verification to generate regional charging pile demand forecast decision information.

[0006] Optionally, S1 specifically includes: Acquire regional basic data, vehicle travel data, parking resource data, existing charging facility operation data, and grid connection correlation data within the target area to generate multi-source charging demand correlation data; Based on the regional affiliation information, facility affiliation information, and power supply affiliation information in the multi-source charging demand association data, a unified regional identification result is generated. Based on the location description information and spatial coordinate information in the multi-source charging demand association data, the spatial location of the charging demand occurrence location, parking resource location, charging facility location and grid connection location within the target area is spatially calibrated, and spatial calibration results are generated. Based on the trip occurrence time, parking record time, charging operation time and power grid load collection time in the multi-source charging demand association data, the time base is unified to generate time stamp results; The unified regional identification result, the spatial calibration result, and the time stamp result are written into the multi-source charging demand association data to generate the original dataset of regional charging demand.

[0007] Optionally, S2 specifically includes: Read the regional unified identification result, spatial calibration result, time stamp result and demand association field from the original dataset of regional charging demand to generate a dataset of charging demand to be processed; The dataset of charging demands to be processed is subjected to field integrity verification, duplicate record merging, and abnormal record filtering to generate a valid dataset of charging demands. Based on a unified data format and a unified time granularity, the effective charging demand dataset is standardized and transformed, and spatiotemporally aligned according to the unified regional identification result, the spatial calibration result, and the time stamp result to generate a standardized charging demand dataset. Based on the standardized charging demand dataset, regional travel activity characteristics, parking resource supply characteristics, existing charging facility utilization characteristics, and grid connection capacity characteristics are extracted to generate regional charging demand impact characteristics. The characteristics affecting regional charging demand are normalized and their features are correlated to generate a big data feature set of regional charging demand.

[0008] Optionally, S3 specifically includes: Read the big data feature set of the charging demand in the region, and obtain the basic spatial data and resource constraint data of the target region to generate grid mapping basic data; Based on the regional spatial basic data, the target area is divided into grids to generate a demand prediction grid, and the grid number, regional functional attributes and spatial boundary attributes are configured for the demand prediction grid. Based on the regional resource constraint data, determine the parking capacity conditions, existing charging facility service conditions, and grid access conditions corresponding to the demand prediction grid, and generate grid resource constraint results; Based on the spatial calibration results and the unified regional identification results, the big data feature set of regional charging demand is mapped to the corresponding demand prediction grid to generate gridded charging demand features. Based on the gridded charging demand characteristics and the gridded resource constraints, a gridded model structure including a demand grid layer, a resource constraint layer, and a charging demand characteristic layer is constructed to generate a digital model of charging demand in the gridded area.

[0009] Optionally, S4 specifically includes: Read the demand grid layer, resource constraint layer and charging demand feature layer in the digital model of charging demand in the gridded area, extract the demand prediction grid nodes that represent the distribution of charging demand in the area, and generate a set of demand grid node attributes based on the regional functional attributes, charging demand impact characteristics, grid resource constraint results and spatiotemporal calibration information corresponding to the demand prediction grid nodes. Based on the traffic attraction information, parking resource information, existing charging facility information and power grid access information in the gridded regional charging demand digital model, traffic attraction nodes, parking resource nodes, existing charging facility nodes and power grid access nodes that are associated with the demand prediction grid nodes are extracted respectively to generate a set of regional demand associated nodes. Based on the spatial coverage relationship, travel flow relationship, parking supply and demand relationship, charging service substitution relationship and access constraint relationship between the demand prediction grid nodes and the regional demand associated nodes in the regional demand associated node set, regional demand associated edges are established, and the relationship type of the regional demand associated edges is labeled to generate a set of regional demand associated edges. Based on the time stamping results, the set of demand grid node attributes, the set of regional demand associated nodes, and the set of regional demand associated edges are sequentially bound according to a preset time granularity to form a spatiotemporal association graph data that includes changes in node state, changes in edge relationships, and changes in demand intensity. Based on the spatiotemporal correlation graph data, the demand prediction grid nodes, the regional demand correlation nodes in the set of regional demand correlation nodes, the regional demand correlation edges in the set of regional demand correlation edges, the node temporal attributes, and the edge temporal attributes are organized into a graph structure to generate a spatiotemporal graph of regional charging demand.

[0010] Optionally, S5 specifically includes: Read the demand prediction grid nodes, regional demand association nodes, regional demand association edges, node temporal attributes, and edge temporal attributes from the spatiotemporal graph of regional charging demand, and generate graph prediction input data. Based on the map prediction input data, demand node status data, regional correlation data, historical demand sequence data, and resource constraint sequence data are extracted to generate regional charging demand prediction samples. The regional charging demand prediction samples are input into the improved Graph WaveNet network. The improved Graph WaveNet network is used to jointly model the spatial correlation between the demand prediction grids and the temporal variation law of charging demand to generate grid-level demand prediction features. Based on the grid-level demand prediction characteristics, the charging demand intensity, number of charging services, charging power demand and charging duration demand of each demand prediction grid during the prediction period are predicted to generate grid-level charging demand prediction values. Based on the resource constraint sequence data, the grid-level charging demand prediction value is initially constrained and checked, and the grid-level charging demand prediction value that exceeds the regional resource carrying capacity is constrained and initially corrected to generate the checked grid-level charging demand prediction value. The verified grid-level charging demand forecast values ​​are summarized according to the demand forecast grid and forecast period to generate the regional basic charging demand forecast results.

[0011] Optionally, the step of jointly modeling the spatial correlation between demand forecasting grids and the temporal variation of charging demand by improving the Graph WaveNet network to generate grid-level demand forecasting features specifically involves: The improved Graph WaveNet network includes a graph input layer, a multi-relationship graph structure learning layer, a multi-scale temporal convolutional layer, a resource constraint fusion layer, and a demand prediction output layer; The region charging demand prediction sample is received through the graph input layer, and the demand node status data, the region correlation data, the historical demand sequence data and the resource constraint sequence data are converted into node input vector, edge relationship input vector and time series input vector. The multi-relationship graph structure learning layer constructs a regional charging demand multi-relationship adjacency matrix based on the regional charging demand spatiotemporal graph, and fuses the regional charging demand multi-relationship adjacency matrix with the adaptive adjacency matrix in the improved Graph WaveNet network to generate a dynamic graph structure representation that includes explicit and implicit associations between demand prediction grids. The historical demand sequence data is processed by dilated causal convolution through the multi-scale temporal convolutional layer to extract the temporal variation features of charging demand under short-term fluctuation cycle, intraday variation cycle and intraweek variation cycle, and generate a multi-scale demand temporal representation. The resource constraint sequence data is embedded into the dynamic graph structure representation and the multi-scale demand time representation through the resource constraint fusion layer, and the demand prediction grid limited by parking capacity, existing charging facility service capacity and grid access capacity is constrained and corrected to generate a constraint fusion spatiotemporal representation. The demand prediction output layer maps the constraint fusion spatiotemporal representation to output the grid-level demand prediction features corresponding to each demand prediction grid.

[0012] Optionally, S6 specifically includes: Read the basic charging demand forecast results for the region, extract the basic demand intensity, basic charging service frequency, basic charging power demand and basic charging duration demand for each demand forecast grid within the forecast period, and generate a basic demand forecast matrix. The regional demand adjustment constraint information is obtained, and electric vehicle growth trend information, regional development planning information, time-period travel fluctuation information, existing charging facility service utilization status and regional resource carrying status are extracted from the regional demand adjustment constraint information to generate demand elasticity adjustment input data. Based on the electric vehicle growth trend information and the regional development planning information, calculate the long-term demand growth correction coefficient corresponding to each demand forecast grid, and calculate the time elasticity correction coefficient corresponding to each forecast period based on the travel fluctuation information of the time period. Calculate the charging service saturation correction coefficient based on the existing charging facility utilization rate, queuing status and service radius, and write the charging service saturation correction coefficient into the demand elasticity correction input data; Calculate the resource constraint correction coefficient based on parking capacity and grid access capacity, and write the resource constraint correction coefficient into the demand elasticity correction input data; Based on the basic demand forecast matrix, the long-term demand growth correction coefficient, the time elasticity correction coefficient, the charging service saturation correction coefficient, and the resource constraint correction coefficient, a demand elasticity correction model is constructed to correct the regional basic charging demand forecast results and generate a corrected grid-level charging demand forecast value. The corrected grid-level charging demand forecast is subjected to demand boundary verification. When the corrected grid-level charging demand forecast exceeds the resource carrying capacity limit of the corresponding demand forecast grid, demand transfer correction is performed based on the service substitution capability of the adjacent demand forecast grid to generate the grid-level charging demand forecast after boundary verification. The grid-level charging demand prediction values ​​after boundary verification are summarized according to the demand prediction grid and prediction time period to generate regional charging pile demand prediction results.

[0013] Optionally, S7 specifically includes: Read the demand forecast results of the charging piles in the region, extract the predicted charging demand intensity, predicted charging service frequency, predicted charging power demand and predicted charging duration demand of each demand forecast grid within the forecast period, and generate demand decision input data. Obtain the existing charging facility supply capacity and regional resource access constraint information within the target area, calculate the serviceable charging capacity corresponding to each demand prediction grid based on the existing charging facility supply capacity, and calculate the new construction capacity corresponding to each demand prediction grid based on the regional resource access constraint information. Based on the demand decision input data and the available charging capacity, a supply and demand gap analysis is performed to determine the charging service gap value, charging power gap value and peak hour service gap value corresponding to each demand prediction grid, and to generate regional charging supply and demand gap results. Based on the regional charging supply and demand gap results, regional functional attributes, demand growth trend, resource access constraint degree and service substitution capability of adjacent demand prediction grids, the construction priority score corresponding to each demand prediction grid is calculated, and the grids are sorted from high to low according to the construction priority scores to generate construction priority ranking results. Based on the new construction capacity, the construction priority ranking result is checked for access constraints to determine whether each demand prediction grid meets the parking carrying capacity conditions, charging service coverage conditions and grid access conditions, and an access constraint check result is generated. Based on the regional charging supply and demand gap results, the construction priority ranking results, and the access constraint verification results, the recommended construction quantity, recommended construction type, recommended construction location range, and recommended access level for each demand prediction grid are determined, and regional charging pile construction recommendation results are generated. The regional charging supply and demand gap results, the construction priority ranking results, the access constraint verification results, and the regional charging pile construction recommendation results are combined to generate regional charging pile demand prediction decision information.

[0014] A regional charging pile demand forecasting system based on big data analysis according to an embodiment of the present invention includes the following modules: The multi-source data acquisition module is used to collect multi-source charging demand correlation data within the target area, perform unified identification and spatiotemporal calibration on the multi-source charging demand correlation data, and generate the original dataset of regional charging demand. The standardized feature processing module is used to standardize the original dataset of regional charging demand and extract the features that affect regional charging demand to generate a big data feature set of regional charging demand. The gridded modeling module is used to map the big data feature set of regional charging demand to the demand prediction grid based on regional spatial basic data and regional resource constraint data, and generate a gridded regional charging demand digital model. The spatiotemporal graph construction module is used to extract regional demand-related nodes based on the gridded regional charging demand digital model and construct a spatiotemporal graph of regional charging demand. The demand forecasting module is used to input the spatiotemporal map of regional charging demand into the improved Graph WaveNet network, jointly predict the spatial correlation and temporal variation of regional charging demand, and generate the basic demand forecast results for regional charging. The demand elasticity correction module is used to perform demand elasticity correction based on the regional basic charging demand forecast results and regional demand correction constraint information, and generate regional charging pile demand forecast results. The prediction and decision output module is used to perform supply and demand gap analysis, construction priority ranking, and access constraint verification based on the regional charging pile demand prediction results, and generate regional charging pile demand prediction and decision information.

[0015] The beneficial effects of this invention are as follows: This invention addresses the problems of scattered data sources, unclear spatial attribution, inconsistent temporal caliber, and insufficient consideration of resource constraints in existing regional charging pile demand forecasting by integrating multi-source charging demand correlation data fusion, gridded regional charging demand digital model, regional charging demand spatiotemporal map, and improved Graph WaveNet network. It adopts unified identification, spatial calibration, and unified temporal benchmark, mapping vehicle travel data, parking resource data, existing charging facility operation data, and grid access correlation data to the demand forecasting grid, achieving unified modeling of data from different sources, with different granularities, and with different business calibers. In the spatial modeling stage, a regional charging demand spatiotemporal map is constructed, including demand forecasting grid nodes, traffic attraction nodes, parking resource nodes, existing charging facility nodes, and distribution network access nodes. Complex inter-regional correlations are represented through spatial coverage relationships, travel flow relationships, parking supply and demand relationships, charging service substitution relationships, and access constraint relationships. In the forecasting stage, an improved Graph WaveNet network is introduced. The WaveNet network integrates multi-relation graph structure learning, multi-scale temporal convolution, and resource constraints to simultaneously extract short-term fluctuations, intraday variations, and weekly variations in charging demand. It also supplements potential spatial dependencies through adaptive adjacency relationships. In the correction and decision-making stages, it incorporates electric vehicle growth trends, the utilization status of existing charging facilities, and regional resource carrying capacity to adjust demand elasticity, further outputting supply-demand gap analysis, construction priority ranking, and access constraint verification results. Ultimately, it achieves accurate prediction of regional charging pile demand and optimized construction decisions. Attached Figure Description

[0016] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1This is an overall flowchart of a regional charging pile demand prediction method based on big data analysis proposed in this invention; Figure 2 This is a schematic diagram of the structure of a regional charging pile demand prediction system based on big data analysis proposed in this invention; Detailed Implementation The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0017] refer to Figure 1 A method for predicting regional charging pile demand based on big data analysis includes the following steps: S1. Collect multi-source charging demand correlation data within the target area, perform unified identification and spatiotemporal calibration on the multi-source charging demand correlation data, and generate the original dataset of regional charging demand. S2. Standardize the original dataset of regional charging demand and extract the features that affect regional charging demand to generate a big data feature set of regional charging demand. S3. Based on regional spatial basic data and regional resource constraint data, map the big data feature set of regional charging demand to the demand prediction grid to generate a gridded regional charging demand digital model. S4. Extract the regional demand-related nodes based on the gridded regional charging demand digital model, and construct a spatiotemporal map of regional charging demand. S5. Input the spatiotemporal map of regional charging demand into the improved Graph WaveNet network to jointly predict the spatial correlation and temporal variation of regional charging demand and generate the prediction results of regional basic charging demand. S6. Based on the regional charging infrastructure demand forecast results and regional demand correction constraint information, perform demand elasticity correction to generate regional charging pile demand forecast results. S7. Based on the regional charging pile demand forecast results, conduct supply and demand gap analysis, construction priority ranking, and access constraint verification to generate regional charging pile demand forecast decision information.

[0018] In this embodiment, S1 specifically refers to: Acquire regional basic data, vehicle travel data, parking resource data, existing charging facility operation data, and grid connection correlation data within the target area to generate multi-source charging demand correlation data; Based on the regional affiliation information, facility affiliation information, and power supply affiliation information in the multi-source charging demand association data, a unified regional affiliation result is generated. Based on the location description and spatial coordinate information in the multi-source charging demand association data, the spatial location of the charging demand occurrence location, parking resource location, charging facility location and grid connection location within the target area is spatially calibrated, and spatial calibration results are generated. Based on the trip occurrence time, parking record time, charging operation time and grid load collection time in the multi-source charging demand correlation data, the time base is unified to generate time stamp results; The regional unified identification results, spatial calibration results, and time stamping results are written into the multi-source charging demand association data to generate the original dataset of regional charging demand. In the specific implementation process, the data collection boundary is the coverage area of ​​streets, communities, business districts, industrial parks and transportation hubs within the target area. The basic regional data includes regional area, functional attributes, population density, employment density and public service facility distribution data. Vehicle travel data includes vehicle passage records, parking entry and exit records, travel origin and destination records and peak hour traffic records. Parking resource data includes parking lot number, number of parking spaces, parking space turnover rate and opening hours data. Existing charging facility operation data includes charging facility number, rated power, charging start time, charging end time, charging amount, number of charging times, fault status and queuing record data. Grid access related data includes power grid number, distribution transformer capacity, line load rate, available access capacity and voltage deviation data. The collected multi-source charging demand correlation data were organized according to a unified data template. The unified regional identifier adopted a three-level coding method of "regional number + functional zone number + grid number". The regional number was set to 6 digits, the functional zone number to 2 digits, and the grid number to 4 digits. Existing charging facility numbers and parking resource numbers were each indexed uniquely. Duplicate numbers were merged according to the priority of the collection source and the update time. The spatial location calibration adopted a unified coordinate benchmark, and the spatial coordinate accuracy was controlled within 5m. Data with missing coordinates were matched according to address text, facility name, and regional affiliation information. Data with a matching confidence of less than 0.85 was put into a manual review list. The time benchmark adopted a unified minute-level timestamp. Vehicle travel data, parking resource data, and existing charging facility operation data were aggregated according to a 15-minute time granularity. Grid load data was converted according to a 5-minute to 15-minute time granularity. Data with cross-system time deviations exceeding 30 seconds were corrected for time deviations. After completing the unified identifier, spatial calibration, and time stamping, all types of data were associated and stored according to the unified regional identifier results to form the original dataset of regional charging demand.

[0019] In this embodiment, S2 specifically refers to: Read the regional unified identification results, spatial calibration results, time stamp results and demand association fields from the original regional charging demand dataset to generate a charging demand dataset to be processed; Perform field integrity checks, duplicate record merging, and abnormal record filtering on the charging demand dataset to be processed to generate a valid charging demand dataset. Based on a unified data format and unified time granularity, the effective charging demand dataset is standardized and transformed, and spatiotemporally aligned according to the unified regional identification results, spatial calibration results, and time stamp results to generate a standardized charging demand dataset. Based on the standardized charging demand dataset, regional travel activity characteristics, parking resource supply characteristics, existing charging facility utilization characteristics, and grid connection capacity characteristics are extracted to generate regional charging demand impact characteristics. The characteristics of regional charging demand are normalized and their correlations are organized to generate a big data feature set of regional charging demand. In the specific implementation process, a data index table is first established according to the unified regional identification results. Data records from the same demand forecasting grid, the same existing charging facilities, the same parking resources, and the same time period are merged into the same processing batch. When verifying the integrity of fields, the threshold for the integrity rate of required fields is set to 95%. Key fields such as area number, spatial coordinates, timestamp, charging power, charging duration, number of parking spaces, and grid access capacity are verified item by item. Records with a key field missing rate of more than 5% are entered into the abnormal data list. Non-key fields missing are filled in by the average of the nearest time periods in the same demand forecasting grid. When merging duplicate records, the unified regional identifier, spatial calibration result, time stamp result, and data source number are used as the joint primary key. Data records with the same charging facility number and a time difference of less than 60 seconds and a spatial distance of less than 10 meters are identified as duplicate records and are retained and merged according to data source priority, collection time integrity, and field integrity rate. When screening abnormal records, records with charging capacity less than 0 kWh, single charging time exceeding 12 hours, existing charging facility utilization rate exceeding 100%, parking space turnover rate less than 0 or greater than 1.5, and grid load rate exceeding 120% are marked as abnormal and are verified in conjunction with three adjacent time windows. During the standardization process, vehicle travel data, parking resource data, and existing charging facility operation data are uniformly converted to a 15-minute time granularity. Grid access-related data are aggregated according to the same time granularity, and spatial data is uniformly mapped to the demand forecast grid. The normalization process adopts the maximum-minimum normalization method to convert regional travel activity characteristics, parking resource supply characteristics, existing charging facility utilization characteristics, and grid access capacity characteristics to the range of 0 to 1. Then, the features are associated and organized according to the unified regional identification results and time stamp results to generate a big data feature set of regional charging demand.

[0020] In this embodiment, S3 specifically refers to: Read the big data feature set of regional charging demand, and obtain the regional spatial basic data and regional resource constraint data corresponding to the target area to generate grid mapping basic data; Based on the regional spatial basic data, the target area is divided into grids to generate demand prediction grids, and grid numbers, regional functional attributes and spatial boundary attributes are configured for the demand prediction grids. Based on the regional resource constraint data, determine the parking capacity conditions, existing charging facility service conditions, and grid access conditions corresponding to the demand forecast grid, and generate grid resource constraint results; Based on the spatial calibration results and the unified regional identification results, the big data feature set of regional charging demand is mapped to the corresponding demand prediction grid to generate gridded charging demand features. Based on the gridded charging demand characteristics and gridded resource constraints, a gridded model structure including a demand grid layer, a resource constraint layer, and a charging demand characteristic layer is constructed to generate a digital model of charging demand in the gridded area. In the specific implementation process, the regional spatial basic data is stored in GIS vector data format, and the spatial coordinates are uniformly converted to the same coordinate benchmark, with the point position error controlled within 5m; the demand prediction grid is divided according to the functional density of the target area, with 250m×250m grids used for commercial concentrated areas and transportation hub areas, 500m×500m grids used for residential areas and office areas, and 1km×1km grids used for low-density areas, and each demand prediction grid is configured with a 12-digit grid number; Regional resource constraint data is aggregated according to the demand forecast grid. Parking capacity conditions include the total number of parking spaces, the number of available parking spaces, and the average turnover rate. Existing charging facility service conditions include the number of charging facilities, rated power, average utilization rate, failure rate, and service radius. Grid access conditions include distribution transformer capacity, line load rate, available access capacity, and voltage deviation. The regional charging demand big data feature set is mapped to the corresponding demand forecast grid according to the spatial calibration results and the regional unified identification results, generating a 64-dimensional gridded charging demand feature vector. This vector is stored according to the demand grid layer, resource constraint layer, and charging demand feature layer, forming a gridded regional charging demand digital model.

[0021] In this embodiment, S4 specifically refers to: Read the demand grid layer, resource constraint layer and charging demand feature layer from the digital model of charging demand in the gridded area, extract the demand prediction grid nodes that represent the distribution of charging demand in the area, and generate a set of demand grid node attributes based on the regional functional attributes, charging demand impact characteristics, grid resource constraint results and spatiotemporal calibration information corresponding to the demand prediction grid nodes. Based on the traffic attraction information, parking resource information, existing charging facility information and power grid access information in the gridded regional charging demand digital model, traffic attraction nodes, parking resource nodes, existing charging facility nodes and power grid access nodes that are associated with the demand prediction grid nodes are extracted respectively to generate a set of regional demand-related nodes. Based on the spatial coverage relationship, travel flow relationship, parking supply and demand relationship, charging service substitution relationship and access constraint relationship between the demand forecast grid nodes and the regional demand associated nodes in the regional demand associated node set, regional demand associated edges are established, and the relationship type of the regional demand associated edges is labeled to generate a set of regional demand associated edges. Based on the time stamping results, the set of demand grid node attributes, the set of regional demand associated nodes, and the set of regional demand associated edges are time-series bound according to a preset time granularity to form a spatiotemporal association graph data that includes changes in node state, changes in edge relationships, and changes in demand intensity. Based on the spatiotemporal correlation graph data, the demand prediction grid nodes, the regional demand correlation nodes in the set of regional demand correlation nodes, the regional demand correlation edges in the set of regional demand correlation edges, the node temporal attributes, and the edge temporal attributes are organized into a graph structure to generate a spatiotemporal graph of regional charging demand. In the specific implementation process, the demand forecasting grid nodes take a single demand forecasting grid as the basic node unit. Each demand forecasting grid node is configured with node attributes such as grid number, regional functional attributes, grid center coordinates, population density, vehicle travel intensity, parking resource supply, utilization rate of existing charging facilities, and grid access margin. The node attribute vector dimension is set to 64 dimensions. Regional demand-related nodes are constructed according to four categories: traffic attraction nodes, parking resource nodes, existing charging facility nodes, and distribution network access nodes. Each type of node is configured with a node number, node type identifier, and spatial location identifier. The node type identifier adopts a 4-digit code. Regional demand-related edges are categorized and labeled according to spatial coverage, travel flow, parking supply and demand, charging service substitution, and access constraints. The number of relationship types is set to 5. Edge weights are calculated based on spatial distance, travel volume, parking supply and demand matching degree, service radius of existing charging facilities, and grid access margin, and are normalized to the range of 0 to 1. When binding time series, node time series attributes and edge time series attributes are generated at a 15-minute time granularity. Four consecutive time slices form a short-term change window, and 96 consecutive time slices form a daily change window. Finally, the demand prediction grid nodes, regional demand-related nodes, and regional demand-related edges are organized into a spatiotemporal graph structure and stored in the form of a node attribute matrix, edge relationship matrix, edge weight matrix, and time slice index table to generate a spatiotemporal graph of regional charging demand.

[0022] In this embodiment, S5 specifically refers to: Read the demand prediction grid nodes, regional demand association nodes, regional demand association edges, node temporal attributes, and edge temporal attributes from the spatiotemporal graph of regional charging demand, and generate graph prediction input data. Based on the input data of the map prediction, demand node status data, regional correlation data, historical demand sequence data and resource constraint sequence data are extracted to generate regional charging demand prediction samples; The regional charging demand prediction samples are input into the improved Graph WaveNet network. The improved Graph WaveNet network is used to jointly model the spatial relationship between the demand prediction grids and the temporal variation of charging demand, generating grid-level demand prediction features. Based on the grid-level demand forecasting characteristics, the charging demand intensity, number of charging services, charging power demand and charging duration demand of each demand forecasting grid are predicted during the forecast period, generating grid-level charging demand forecast values. Based on the resource constraint sequence data, the grid-level charging demand forecast is initially checked for constraints, and the grid-level charging demand forecast that exceeds the regional resource carrying capacity is constrained and initially corrected to generate the checked grid-level charging demand forecast. The verified grid-level charging demand forecast values ​​are summarized according to the demand forecast grid and forecast period to generate the regional basic charging demand forecast results. In the specific implementation process, the input data for the graph prediction is organized according to the demand prediction grid and time slices. The node status data includes regional functional attributes, vehicle travel intensity, parking resource supply, utilization rate of existing charging facilities, and grid access capacity. The regional correlation data includes spatial adjacency relationships, travel flow relationships, parking supply and demand relationships, charging service substitution relationships, and access constraint relationships. The historical demand sequence data includes historical charging times, historical charging volume, historical charging duration, and historical demand intensity. The resource constraint sequence data includes parking capacity limits, the service capacity of existing charging facilities, and the grid access capacity. The prediction sample is constructed using a sliding window method, with the input window length set to 96 time slices, each time slice being 15 minutes. The prediction step size is set to 4 time slices, 16 time slices, and 96 time slices, corresponding to short-term prediction, half-day prediction, and intraday prediction, respectively. After inputting the regional charging demand forecast samples into the improved Graph WaveNet network, the network outputs grid-level demand forecast features for each demand forecast grid, with the feature dimension set to 128. Based on the grid-level demand forecast features, the charging demand intensity, number of charging services, charging power demand, and charging duration demand are predicted respectively. The number of charging services is measured in times, the charging power demand is measured in kWh, and the charging duration demand is measured in minutes. During the initial constraint verification, the grid-level charging demand forecast values ​​are compared with the parking capacity limit, the service capacity of existing charging facilities, and the grid access capacity. When the forecast value exceeds 110% of the corresponding resource capacity limit, a constraint mark is made, and an initial correction is made according to the resource capacity limit. Finally, the grid-level charging demand forecast values ​​after verification are summarized according to the demand forecast grid number and forecast period to generate the regional basic charging demand forecast results.

[0023] In this embodiment, the spatial relationships between demand forecasting grids and the temporal variation of charging demand are jointly modeled by improving the Graph WaveNet network to generate grid-level demand forecasting features, specifically: The improved Graph WaveNet network includes a graph input layer, a multi-relationship graph structure learning layer, a multi-scale temporal convolutional layer, a resource constraint fusion layer, and a demand prediction output layer; The graph input layer receives regional charging demand prediction samples and converts demand node status data, regional correlation data, historical demand sequence data and resource constraint sequence data into node input vectors, edge relationship input vectors and time series input vectors. The multi-relationship adjacency matrix of regional charging demand is constructed based on the spatiotemporal graph of regional charging demand through a multi-relationship graph structure learning layer. The multi-relationship adjacency matrix of regional charging demand is then fused with the adaptive adjacency matrix in the improved Graph WaveNet network to generate a dynamic graph structure representation that includes explicit and implicit associations between demand prediction grids. By performing dilated causal convolution processing on historical demand sequence data through multi-scale temporal convolutional layers, the temporal variation features of charging demand under short-term fluctuation cycles, intraday variation cycles, and intraweek variation cycles are extracted to generate multi-scale demand temporal representations. By embedding resource-constrained sequence data into dynamic graph structure representation and multi-scale demand time representation through the resource-constrained fusion layer, and performing constraint correction on the demand prediction grid limited by parking capacity, existing charging facility service capacity and grid access capacity, a constrained fusion spatiotemporal representation is generated. The constraint fusion spatiotemporal representation is mapped through the demand forecast output layer, and the grid-level demand forecast features corresponding to each demand forecast grid are output. In its implementation, the improved Graph WaveNet network inherits the joint graph convolution and temporal convolution modeling framework of the traditional GraphWaveNet network in its overall structure, including graph structure input, temporal feature extraction, and prediction output processes. However, it introduces a multi-relation graph structure learning layer and a resource constraint fusion layer in the intermediate structure to replace the traditional GraphWaveNet network's processing method that mainly relies on a single adjacency matrix and historical sequence features for prediction. First, the demand node state data in the regional charging demand prediction samples are mapped to a node input vector with a dimension of 128, and the regional association data are mapped to an edge relationship input vector with a dimension of 64. The historical demand sequence data is used to construct a temporal input vector according to 96 time slices, with each time slice corresponding to 15 minutes. The multi-relation graph structure learning layer constructs five types of regional charging demand multi-relation adjacency matrices based on spatial coverage relationships, travel flow relationships, parking supply and demand relationships, charging service substitution relationships, and access constraint relationships. These matrices are then fused with an adaptive adjacency matrix with a weight of 0.7:0.3 to obtain a dynamic graph structure representation. Subsequently, a multi-scale temporal convolutional layer employs three sets of dilated causal convolutions with dilation rates of 1, 2, and 4 to extract short-term fluctuations, intraday variations, and intraweekly variations. The resource-constrained fusion layer embeds parking capacity, existing charging facility service capacity, and grid access capacity into the spatiotemporal representation, constraining and correcting predicted features that exceed resource carrying capacity. Compared to the traditional Graph WaveNet network, this invention not only considers the implicit spatiotemporal correlations between demand prediction grids but also explicitly introduces charging service substitution relationships and grid access constraints. This reduces local overfitting caused by solely relying on historical demand sequences and improves the interpretability and feasibility of regional charging demand prediction results in construction decision-making scenarios.

[0024] In this embodiment, S6 specifically refers to: Read the regional basic charging demand forecast results, extract the basic demand intensity, basic charging service frequency, basic charging power demand and basic charging duration demand of each demand forecast grid within the forecast period, and generate a basic demand forecast matrix. Obtain regional demand adjustment constraint information, and extract electric vehicle growth trend information, regional development planning information, time-period travel fluctuation information, existing charging facility service utilization status and regional resource carrying status from the regional demand adjustment constraint information to generate demand elasticity adjustment input data; Based on information on electric vehicle growth trends and regional development plans, the long-term demand growth correction coefficients for each demand forecast grid are calculated, and the time elasticity correction coefficients for each forecast period are calculated based on travel fluctuation information during different time periods. Calculate the charging service saturation correction coefficient based on the existing charging facility utilization rate, queuing status and service radius, and write the charging service saturation correction coefficient into the demand elasticity correction input data. Calculate the resource constraint correction coefficient based on parking capacity and grid access capacity, and write the resource constraint correction coefficient into the demand elasticity correction input data; Based on the basic demand forecast matrix, long-term demand growth correction coefficient, time elasticity correction coefficient, charging service saturation correction coefficient, and resource constraint correction coefficient, a demand elasticity correction model is constructed to correct the regional basic charging demand forecast results and generate the corrected grid-level charging demand forecast value. Demand boundary verification is performed on the corrected grid-level charging demand forecast. When the corrected grid-level charging demand forecast exceeds the resource carrying capacity limit of the corresponding demand forecast grid, demand transfer correction is performed based on the service substitution capability of the adjacent demand forecast grid to generate the grid-level charging demand forecast after boundary verification. The grid-level charging demand forecast values ​​after boundary verification are summarized according to the demand forecast grid and forecast period to generate the regional charging pile demand forecast results. In the specific implementation process, the basic demand forecasting matrix is ​​organized in two dimensions according to the demand forecasting grid number and the forecasting period. Each row corresponds to a demand forecasting grid, and each column corresponds to a 15-minute forecasting time slice. The matrix fields include basic demand intensity, basic charging service frequency, basic charging power demand, and basic charging duration demand. The long-term demand growth correction coefficient is determined based on electric vehicle growth trend information and regional development planning information, with a value range of 0.90 to 1.30. Higher coefficients are used for grids in commercial areas, industrial parks, and transportation hubs with higher regional development intensity, and lower coefficients are used for low-density areas. The time elasticity correction coefficient is determined based on the travel fluctuation information of different time periods, with a value range of 0.80 to 1.25. The coefficient corresponding to peak travel periods is higher than that of ordinary periods. The charging service saturation correction coefficient is calculated based on the utilization rate of existing charging facilities, queuing status, and service radius. When the utilization rate of existing charging facilities exceeds 85% or the average queuing time exceeds 20 minutes, the demand correction intensity of the corresponding demand forecast grid is increased. The resource constraint correction coefficient is determined based on parking capacity and grid access capacity, with a value range of 0.70 to 1.00. When the availability of parking spaces is lower than 15% or the grid access capacity is lower than the predicted new load, the corrected grid-level charging demand forecast is compressed. When the corrected grid-level charging demand forecast exceeds 110% of the resource carrying capacity limit, the excess portion is transferred according to the service substitution capacity of adjacent demand forecast grids. The service substitution capacity threshold is set at 0.60, prioritizing transfer to adjacent demand forecast grids that are closer, have higher existing charging facility service margins, and sufficient grid access margins. Through this demand elasticity correction process, the problem of overestimating or underestimating construction demand caused by relying solely on historical demand forecasts can be avoided, making the regional charging pile demand forecast results more consistent with actual resource carrying capacity conditions.

[0025] In this embodiment, S7 specifically refers to: Read the regional charging pile demand forecast results, extract the predicted charging demand intensity, predicted charging service frequency, predicted charging power demand and predicted charging duration demand for each demand forecast grid within the forecast period, and generate demand decision input data. Obtain the existing charging facility supply capacity and regional resource access constraint information within the target area, calculate the serviceable charging capacity corresponding to each demand forecast grid based on the existing charging facility supply capacity, and calculate the new construction capacity corresponding to each demand forecast grid based on the regional resource access constraint information. Based on the demand decision input data and available charging capacity, a supply and demand gap analysis is performed to determine the charging service gap value, charging power gap value and peak-hour service gap value corresponding to each demand forecast grid, and generate regional charging supply and demand gap results. Based on the regional charging supply and demand gap results, regional functional attributes, demand growth trends, resource access constraints, and service substitution capabilities of adjacent demand forecast grids, the construction priority score corresponding to each demand forecast grid is calculated, and the grids are sorted from high to low according to their construction priority scores to generate construction priority ranking results. Based on the available construction capacity, the construction priority ranking results are checked for access constraints to determine whether each demand prediction grid meets the parking capacity conditions, charging service coverage conditions, and grid access conditions, and to generate access constraint check results. Based on the results of regional charging supply and demand gap, construction priority ranking, and access constraint verification, the recommended number of constructions, recommended construction types, recommended construction location range, and recommended access level for each demand prediction grid are determined, and regional charging pile construction recommendation results are generated. The results of regional charging supply and demand gap, construction priority ranking, access constraint verification, and regional charging pile construction recommendation are combined to generate regional charging pile demand forecasting decision information. In the specific implementation process, the demand decision input data is summarized according to the demand forecast grid number. The forecast time slice still adopts a 15-minute granularity. The peak charging demand intensity, the average number of daily charging services, the cumulative charging power demand, and the charging duration demand during peak hours are calculated separately. The supply capacity of existing charging facilities is calculated based on the number of charging facilities, rated power, availability rate, and average service duration. The available charging capacity is represented by the number of services and the amount of service that existing charging facilities can provide within a single demand forecast grid. The regional resource access constraint information includes the number of available parking spaces, the number of new charging facilities that can be added, the grid access capacity, and the line load rate. Demand forecast grids with a line load rate exceeding 85% or a voltage deviation exceeding ±7% are marked as access-restricted grids. When analyzing the supply and demand gap, the difference between the predicted number of charging services and the available charging capacity is calculated. When the charging service gap value is greater than 20 times / day or the service gap value during peak hours exceeds 30% of the available capacity, the corresponding demand prediction grid is listed as a key candidate grid for construction. The construction priority scoring adopts a percentage system, with the weight of the regional charging supply and demand gap result set at 0.40, the weight of the demand growth trend set at 0.20, the weight of the regional functional attributes set at 0.15, the weight of the resource access constraint degree set at 0.15, and the weight of the service substitution capacity of adjacent demand prediction grids set at 0.10. When checking the access constraints, parking carrying capacity conditions and grid access conditions are judged first. Demand prediction grids that meet the verification requirements output recommended construction quantity, recommended construction type, recommended construction location range, and recommended access level. Demand prediction grids that do not meet the verification requirements output a postponed construction mark or alternative grid recommendation results, ultimately forming regional charging pile demand prediction decision information.

[0026] refer to Figure 2 A regional charging pile demand forecasting system based on big data analysis includes the following modules: The multi-source data acquisition module is used to collect multi-source charging demand correlation data within the target area, perform unified identification and spatiotemporal calibration on the multi-source charging demand correlation data, and generate the original dataset of regional charging demand. The standardized feature processing module is used to standardize the original dataset of regional charging demand and extract the features that affect regional charging demand to generate a big data feature set of regional charging demand. The gridded modeling module is used to map the big data feature set of regional charging demand to the demand prediction grid based on regional spatial basic data and regional resource constraint data, and generate a gridded regional charging demand digital model. The spatiotemporal graph construction module is used to extract regional demand-related nodes based on the gridded regional charging demand digital model and construct a spatiotemporal graph of regional charging demand. The demand forecasting module is used to input the spatiotemporal map of regional charging demand into the improved Graph WaveNet network, jointly predict the spatial correlation and temporal variation of regional charging demand, and generate the basic demand forecast results for regional charging. The demand elasticity correction module is used to perform demand elasticity correction based on the regional basic charging demand forecast results and regional demand correction constraint information, and generate regional charging pile demand forecast results. The prediction and decision output module is used to perform supply and demand gap analysis, construction priority ranking, and access constraint verification based on the regional charging pile demand prediction results, and generate regional charging pile demand prediction and decision information.

[0027] Example 1: To verify the feasibility of this invention in practice, it was applied to a scenario of regional charging pile demand forecasting and construction decision-making in the core area of ​​the eastern new district of a certain city. The test period was one month, and the application area covered typical vehicle usage scenarios such as large residential communities, commercial complexes, office parks, public transportation interchanges, parking lots around hospitals, and entrances and exits of urban expressways. The total area of ​​this area is approximately 42.6 km², with 38 existing public charging stations, 812 charging guns, approximately 28,600 parking spaces, approximately 58,000 registered electric vehicles, and an average daily vehicle traffic volume of approximately 314,000 trips. There are significant concentrated travel and charging phenomena during morning and evening peak hours. Before implementation, the area mainly relied on manual experience, historical charging order statistics, and single-station utilization reports to determine the demand for charging pile construction, which easily led to two problems: on the one hand, there were serious queues during peak hours near commercial complexes, hospitals, and transportation hubs, with the average queuing time at some charging stations exceeding 25 minutes; on the other hand, the charging facilities already built in some low-density residential areas were underutilized, with an average daily utilization rate of less than 18%, resulting in a mismatch between construction resources and actual demand.

[0028] In this scenario, the present invention collects multi-source charging demand correlation data from regional traffic management platforms, parking lot management platforms, existing charging facility operation platforms, and power grid load monitoring systems. It unifies vehicle travel records, parking space turnover records, charging order records, charging facility operating status, and power grid load data into the same temporal granularity and spatial grid. The target area is divided into 126 demand prediction grids, with different functional areas divided at three scales: 250m×250m, 500m×500m, and 1km×1km. Simultaneously, traffic attraction points, parking resource points, existing charging facility points, and power grid access points are constructed as regional demand correlation nodes. A spatiotemporal map of regional charging demand is formed through spatial coverage, travel flow, parking supply and demand, charging service substitution, and access constraints.

[0029] During the prediction process, this invention inputs a spatiotemporal map of regional charging demand into an improved Graph WaveNet network to jointly model the spatial relationships between different demand prediction grids and the temporal variation patterns of charging demand. Compared to methods that rely solely on extrapolating historical charging order trends, this invention further incorporates vehicle flow, changes in parking resources, the alternative service capabilities of adjacent grids, and grid access capacity. This enables the accurate identification of potentially high-demand areas such as hospitals, commercial complexes, and transportation hubs, while reducing the priority of new construction in low-density and resource-constrained areas to avoid blindly adding new charging piles.

[0030] To further demonstrate the beneficial effects of this invention, it was compared with traditional historical mean forecasting methods and ordinary time series forecasting methods under the same test period, data range, and evaluation criteria. The forecasting objects included the number of charging services, charging volume demand, peak-hour demand intensity, and new construction priority for each demand forecasting grid during future intraday periods. Actual charging orders, actual charging volume, actual queuing records, and manually verified construction demand locations were used as verification criteria. The statistical results are shown in Table 1.

[0031] Table 1. Comparison of Regional Charging Pile Demand Forecasting and Construction Decision-Making Effects: As shown in Table 1, this invention significantly outperforms traditional methods in both the accuracy of charging demand forecasting and the rationality of construction decisions. Traditional historical average forecasting methods primarily extrapolate from the number of charging sessions and the amount of electricity charged over a past period, failing to identify the impact of traffic flow changes, parking resource changes, and grid access constraints on regional charging demand. Therefore, the peak-hour demand intensity (MAPE) reaches 18.7%, and it easily misclassifies some grids with limited grid access or insufficient parking capacity as suitable construction areas. While ordinary time series forecasting methods can capture certain temporal trends, they do not adequately consider the service substitution relationship between adjacent demand forecasting grids, the service radius of existing charging facilities, and the available capacity of the distribution network. Consequently, the manual review pass rate for recommended construction sites is only 77.9%.

[0032] This invention establishes a joint prediction model involving multiple grids, multiple relationships, and multiple time periods by using a regional charging demand spatiotemporal map and an improved Graph WaveNet network. This allows vehicle travel intensity, parking resource supply, utilization status of existing charging facilities, and grid access capacity to participate in the calculation within the same prediction framework. Test results show that this invention reduces the prediction error of the average daily charging service frequency to 58.9 times / day, reduces the average error of charging demand to 826.6 kWh / day, and achieves a high-demand grid identification accuracy of 92.4%. Simultaneously, this invention incorporates the utilization status of existing charging facilities and the regional resource carrying capacity during the demand elasticity correction process. It performs constraint verification on resource-scarce areas and demand transfer correction on grids with adjacent alternative service capabilities, reducing the number of incorrectly recommended grids with limited access to 4 and increasing the manual review pass rate of recommended construction sites to 91.6%.

[0033] In terms of practical application effects, based on the regional charging pile demand forecasting decision information output by this invention, management prioritizes 18 demand forecasting grids as candidate areas for key construction or expansion. Among these, grids surrounding commercial complexes are recommended for adding fast charging piles, while grids surrounding hospitals and office parks are recommended for configuring a combination of fast and slow charging facilities. Low-density residential areas adopt strategies such as service substitution from adjacent grids and postponement of construction. After planning according to the recommendations of this invention, the average peak-hour queuing time in the region decreased by 31.7%, the average utilization rate of existing charging facilities increased by 22.8%, and the matching degree between charging facility construction sites and parking capacity and grid access conditions significantly improved. Therefore, this invention can solve the problems of single data sources, insufficient spatial correlation, inadequate consideration of resource constraints, and inaccurate judgment of construction priorities in traditional regional charging pile demand forecasting, and can provide more reliable forecasting basis and decision support for regional charging pile planning and construction.

[0034] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for predicting regional charging pile demand based on big data analysis, characterized in that, Includes the following steps: S1. Collect multi-source charging demand correlation data within the target area, perform unified identification and spatiotemporal calibration on the multi-source charging demand correlation data, and generate the original dataset of regional charging demand. S2. Standardize the original dataset of regional charging demand and extract the regional charging demand influencing features to generate a large data feature set of regional charging demand. S3. Based on the regional spatial basic data and regional resource constraint data, the big data feature set of regional charging demand is mapped to the demand prediction grid to generate a gridded regional charging demand digital model. S4. Extract the regional demand association nodes based on the gridded regional charging demand digital model, and construct a spatiotemporal map of regional charging demand. S5. Input the spatiotemporal map of regional charging demand into the improved Graph WaveNet network to jointly predict the spatial correlation and temporal variation of regional charging demand and generate the prediction result of regional basic charging demand. S6. Based on the regional basic charging demand forecast results and regional demand correction constraint information, perform demand elasticity correction to generate regional charging pile demand forecast results. S7. Based on the regional charging pile demand forecast results, perform supply and demand gap analysis, construction priority ranking, and access constraint verification to generate regional charging pile demand forecast decision information.

2. The method for predicting regional charging pile demand based on big data analysis according to claim 1, characterized in that, Specifically, S1 is: Acquire regional basic data, vehicle travel data, parking resource data, existing charging facility operation data, and grid connection correlation data within the target area to generate multi-source charging demand correlation data; Based on the regional affiliation information, facility affiliation information, and power supply affiliation information in the multi-source charging demand association data, a unified regional identification result is generated. Based on the location description information and spatial coordinate information in the multi-source charging demand association data, the spatial location of the charging demand occurrence location, parking resource location, charging facility location and grid connection location within the target area is spatially calibrated, and spatial calibration results are generated. Based on the trip occurrence time, parking record time, charging operation time and power grid load collection time in the multi-source charging demand association data, the time base is unified to generate time stamp results; The unified regional identification result, the spatial calibration result, and the time stamp result are written into the multi-source charging demand association data to generate the original dataset of regional charging demand.

3. The method for predicting regional charging pile demand based on big data analysis according to claim 1, characterized in that, Specifically, S2 is: Read the regional unified identification result, spatial calibration result, time stamp result and demand association field from the original dataset of regional charging demand to generate a dataset of charging demand to be processed; The dataset of charging demands to be processed is subjected to field integrity verification, duplicate record merging, and abnormal record filtering to generate a valid dataset of charging demands. Based on a unified data format and a unified time granularity, the effective charging demand dataset is standardized and transformed, and spatiotemporally aligned according to the unified regional identification result, the spatial calibration result, and the time stamp result to generate a standardized charging demand dataset. Based on the standardized charging demand dataset, regional travel activity characteristics, parking resource supply characteristics, existing charging facility utilization characteristics, and grid connection capacity characteristics are extracted to generate regional charging demand impact characteristics. The characteristics affecting regional charging demand are normalized and their features are correlated to generate a big data feature set of regional charging demand.

4. The method for predicting regional charging pile demand based on big data analysis according to claim 1, characterized in that, Specifically, S3 is: Read the big data feature set of the charging demand in the region, and obtain the basic spatial data and resource constraint data of the target region to generate grid mapping basic data; Based on the regional spatial basic data, the target area is divided into grids to generate a demand prediction grid, and the grid number, regional functional attributes and spatial boundary attributes are configured for the demand prediction grid. Based on the regional resource constraint data, determine the parking capacity conditions, existing charging facility service conditions, and grid access conditions corresponding to the demand prediction grid, and generate grid resource constraint results; Based on the spatial calibration results and the unified regional identification results, the big data feature set of regional charging demand is mapped to the corresponding demand prediction grid to generate gridded charging demand features. Based on the gridded charging demand characteristics and the gridded resource constraints, a gridded model structure including a demand grid layer, a resource constraint layer, and a charging demand characteristic layer is constructed to generate a digital model of charging demand in the gridded area.

5. The method for predicting regional charging pile demand based on big data analysis according to claim 1, characterized in that, Specifically, S4 is: Read the demand grid layer, resource constraint layer and charging demand feature layer in the digital model of charging demand in the gridded area, extract the demand prediction grid nodes that represent the distribution of charging demand in the area, and generate a set of demand grid node attributes based on the regional functional attributes, charging demand impact characteristics, grid resource constraint results and spatiotemporal calibration information corresponding to the demand prediction grid nodes. Based on the traffic attraction information, parking resource information, existing charging facility information and power grid access information in the gridded regional charging demand digital model, traffic attraction nodes, parking resource nodes, existing charging facility nodes and power grid access nodes that are associated with the demand prediction grid nodes are extracted respectively to generate a set of regional demand associated nodes. Based on the spatial coverage relationship, travel flow relationship, parking supply and demand relationship, charging service substitution relationship and access constraint relationship between the demand prediction grid nodes and the regional demand associated nodes in the regional demand associated node set, regional demand associated edges are established, and the relationship type of the regional demand associated edges is labeled to generate a set of regional demand associated edges. Based on the time stamping results, the set of demand grid node attributes, the set of regional demand associated nodes, and the set of regional demand associated edges are sequentially bound according to a preset time granularity to form a spatiotemporal association graph data that includes changes in node state, changes in edge relationships, and changes in demand intensity. Based on the spatiotemporal correlation graph data, the demand prediction grid nodes, the regional demand correlation nodes in the set of regional demand correlation nodes, the regional demand correlation edges in the set of regional demand correlation edges, the node temporal attributes, and the edge temporal attributes are organized into a graph structure to generate a spatiotemporal graph of regional charging demand.

6. The method for predicting regional charging pile demand based on big data analysis according to claim 1, characterized in that, Specifically, S5 is: Read the demand prediction grid nodes, regional demand association nodes, regional demand association edges, node temporal attributes, and edge temporal attributes from the spatiotemporal graph of regional charging demand, and generate graph prediction input data. Based on the map prediction input data, demand node status data, regional correlation data, historical demand sequence data, and resource constraint sequence data are extracted to generate regional charging demand prediction samples. The regional charging demand prediction samples are input into the improved Graph WaveNet network. The improved Graph WaveNet network is used to jointly model the spatial correlation between the demand prediction grids and the temporal variation law of charging demand to generate grid-level demand prediction features. Based on the grid-level demand prediction characteristics, the charging demand intensity, number of charging services, charging power demand and charging duration demand of each demand prediction grid during the prediction period are predicted to generate grid-level charging demand prediction values. Based on the resource constraint sequence data, the grid-level charging demand prediction value is initially constrained and checked, and the grid-level charging demand prediction value that exceeds the regional resource carrying capacity is constrained and initially corrected to generate the checked grid-level charging demand prediction value. The verified grid-level charging demand forecast values ​​are summarized according to the demand forecast grid and forecast period to generate the regional basic charging demand forecast results.

7. The method for predicting regional charging pile demand based on big data analysis according to claim 6, characterized in that, The method involves jointly modeling the spatial relationships between demand forecasting grids and the temporal variation patterns of charging demand using an improved GraphWaveNet network to generate grid-level demand forecasting features. Specifically: The improved Graph WaveNet network includes a graph input layer, a multi-relationship graph structure learning layer, a multi-scale temporal convolutional layer, a resource constraint fusion layer, and a demand prediction output layer; The region charging demand prediction sample is received through the graph input layer, and the demand node status data, the region correlation data, the historical demand sequence data and the resource constraint sequence data are converted into node input vector, edge relationship input vector and time series input vector. The multi-relationship graph structure learning layer constructs a regional charging demand multi-relationship adjacency matrix based on the regional charging demand spatiotemporal graph, and fuses the regional charging demand multi-relationship adjacency matrix with the adaptive adjacency matrix in the improved Graph WaveNet network to generate a dynamic graph structure representation that includes explicit and implicit associations between demand prediction grids. The historical demand sequence data is processed by dilated causal convolution through the multi-scale temporal convolutional layer to extract the temporal variation features of charging demand under short-term fluctuation cycle, intraday variation cycle and intraweek variation cycle, and generate a multi-scale demand temporal representation. The resource constraint sequence data is embedded into the dynamic graph structure representation and the multi-scale demand time representation through the resource constraint fusion layer, and the demand prediction grid limited by parking capacity, existing charging facility service capacity and grid access capacity is constrained and corrected to generate a constraint fusion spatiotemporal representation. The demand prediction output layer maps the constraint fusion spatiotemporal representation to output the grid-level demand prediction features corresponding to each demand prediction grid.

8. The method for predicting regional charging pile demand based on big data analysis according to claim 1, characterized in that, Specifically, S6 is: Read the basic charging demand forecast results for the region, extract the basic demand intensity, basic charging service frequency, basic charging power demand and basic charging duration demand for each demand forecast grid within the forecast period, and generate a basic demand forecast matrix. The regional demand adjustment constraint information is obtained, and electric vehicle growth trend information, regional development planning information, time-period travel fluctuation information, existing charging facility service utilization status and regional resource carrying status are extracted from the regional demand adjustment constraint information to generate demand elasticity adjustment input data. Based on the electric vehicle growth trend information and the regional development planning information, calculate the long-term demand growth correction coefficient corresponding to each demand forecast grid, and calculate the time elasticity correction coefficient corresponding to each forecast period based on the travel fluctuation information of the time period. Calculate the charging service saturation correction coefficient based on the existing charging facility utilization rate, queuing status and service radius, and write the charging service saturation correction coefficient into the demand elasticity correction input data; Calculate the resource constraint correction coefficient based on parking capacity and grid access capacity, and write the resource constraint correction coefficient into the demand elasticity correction input data; Based on the basic demand forecast matrix, the long-term demand growth correction coefficient, the time elasticity correction coefficient, the charging service saturation correction coefficient, and the resource constraint correction coefficient, a demand elasticity correction model is constructed to correct the regional basic charging demand forecast results and generate a corrected grid-level charging demand forecast value. The corrected grid-level charging demand forecast is subjected to demand boundary verification. When the corrected grid-level charging demand forecast exceeds the resource carrying capacity limit of the corresponding demand forecast grid, demand transfer correction is performed based on the service substitution capability of the adjacent demand forecast grid to generate the grid-level charging demand forecast after boundary verification. The grid-level charging demand prediction values ​​after boundary verification are summarized according to the demand prediction grid and prediction time period to generate regional charging pile demand prediction results.

9. The method for predicting regional charging pile demand based on big data analysis according to claim 1, characterized in that, Specifically, S7 is: Read the demand forecast results of the charging piles in the region, extract the predicted charging demand intensity, predicted charging service frequency, predicted charging power demand and predicted charging duration demand of each demand forecast grid within the forecast period, and generate demand decision input data. Obtain the existing charging facility supply capacity and regional resource access constraint information within the target area, calculate the serviceable charging capacity corresponding to each demand prediction grid based on the existing charging facility supply capacity, and calculate the new construction capacity corresponding to each demand prediction grid based on the regional resource access constraint information. Based on the demand decision input data and the available charging capacity, a supply and demand gap analysis is performed to determine the charging service gap value, charging power gap value and peak hour service gap value corresponding to each demand prediction grid, and to generate regional charging supply and demand gap results. Based on the regional charging supply and demand gap results, regional functional attributes, demand growth trend, resource access constraint degree and service substitution capability of adjacent demand prediction grids, the construction priority score corresponding to each demand prediction grid is calculated, and the grids are sorted from high to low according to the construction priority scores to generate construction priority ranking results. Based on the new construction capacity, the construction priority ranking result is checked for access constraints to determine whether each demand prediction grid meets the parking carrying capacity conditions, charging service coverage conditions and grid access conditions, and an access constraint check result is generated. Based on the regional charging supply and demand gap results, the construction priority ranking results, and the access constraint verification results, the recommended construction quantity, recommended construction type, recommended construction location range, and recommended access level for each demand prediction grid are determined, and regional charging pile construction recommendation results are generated. The regional charging supply and demand gap results, the construction priority ranking results, the access constraint verification results, and the regional charging pile construction recommendation results are combined to generate regional charging pile demand prediction decision information.

10. A regional charging pile demand forecasting system based on big data analysis, comprising executing the regional charging pile demand forecasting method based on big data analysis as described in any one of claims 1 to 9, characterized in that, Includes the following modules: The multi-source data acquisition module is used to collect multi-source charging demand correlation data within the target area, perform unified identification and spatiotemporal calibration on the multi-source charging demand correlation data, and generate the original dataset of regional charging demand. The standardized feature processing module is used to standardize the original dataset of regional charging demand and extract the features that affect regional charging demand to generate a big data feature set of regional charging demand. The gridded modeling module is used to map the big data feature set of regional charging demand to the demand prediction grid based on regional spatial basic data and regional resource constraint data, and generate a gridded regional charging demand digital model. The spatiotemporal graph construction module is used to extract regional demand-related nodes based on the gridded regional charging demand digital model and construct a spatiotemporal graph of regional charging demand. The demand forecasting module is used to input the spatiotemporal map of regional charging demand into the improved Graph WaveNet network, jointly predict the spatial correlation and temporal variation of regional charging demand, and generate the basic demand forecast results for regional charging. The demand elasticity correction module is used to perform demand elasticity correction based on the regional basic charging demand forecast results and regional demand correction constraint information, and generate regional charging pile demand forecast results. The prediction and decision output module is used to perform supply and demand gap analysis, construction priority ranking, and access constraint verification based on the regional charging pile demand prediction results, and generate regional charging pile demand prediction and decision information.

Images