Electric vehicle charging demand prediction method, device, terminal and medium

Abstract

The present application discloses a method, device, terminal and medium for predicting charging demand of electric vehicles. The solution provided by the present application constructs a dynamic heat map reflecting the traffic flow status of each road network node based on traffic charging network data through kernel density estimation, uses the dynamic heat map to capture traffic flow changes in real time, combines with graph convolutional networks to characterize road network topology constraints, and then, based on time and space dimensions, performs feature alignment on the dynamic heat map and external factor data. Through multimodal attention processing, cross-modal attention fusion is performed on the features of the dynamic heat map and the features of the external factor data to obtain a fusion vector. Based on the fusion vector, combined with a preset spatiotemporal graph neural network model, the charging demand prediction value of each road network node is determined, which overcomes the problem of spatiotemporal granularity differences of multi-source data, significantly improves the robustness of demand prediction in emergencies and high-volatility scenarios, and improves the accuracy of charging demand prediction in extreme scenarios.

Description

Technical Field

[0001] The present application relates to the technical field of load forecasting, and in particular to a method, device, terminal, and medium for forecasting charging demand of electric vehicles. Background Art

[0002] With the rapid development of electric vehicles, charging demand forecasting has become a key technology for optimizing charging station layout and grid scheduling. Traditional methods rely primarily on historical charging data, such as charging station usage records and vehicle driving trajectories. However, this approach struggles with dynamic traffic scenarios like emergencies and holiday peaks, as well as complex road network spatial relationships. Consequently, existing charging load forecasting technologies suffer from inaccurate predictions in extreme scenarios characterized by dynamic traffic and complex road network spatial relationships. Summary of the Invention

[0003] The present application provides a method, device, terminal and medium for predicting charging demand of electric vehicles, which are used to solve the technical problem that existing charging load prediction technology has inaccurate prediction when facing some extreme scenarios with dynamic traffic scenes and complex road network spatial relationships.

[0004] To solve the above technical problems, the first aspect of the present application provides a method for predicting electric vehicle charging demand, comprising:

[0005] Collect transportation charging network data and external factor data;

[0006] Based on the traffic charging network data, a dynamic heat map is constructed by kernel density estimation, wherein the dynamic heat map is used to reflect the traffic flow status of each road network node;

[0007] Based on the time and space dimensions, the dynamic heat map and the external factor data are feature aligned, and the features of the dynamic heat map and the features of the external factor data are cross-modally fused through multimodal attention processing to obtain a fusion vector;

[0008] Based on the fusion vector and in combination with a preset spatiotemporal graph neural network model, the charging demand prediction value of each road network node is determined.

[0009] Preferably, the transportation charging network data includes: vehicle GPS trajectory, traffic flow data, charging pile status and road network topology.

[0010] Preferably, the external factor data includes weather data, electricity price data and historical demand data.

[0011] Preferably, the determining of the charging demand prediction value of each road network node based on the fusion vector and in combination with a preset spatiotemporal graph neural network model includes:

[0012] Based on the fusion vector and combined with the dynamic adjacency matrix, the fusion vector is subjected to neighborhood aggregation processing by a spatial graph convolution unit to obtain spatial dependency features between the road network nodes, wherein the dynamic critical matrix is ​​used to reflect the similarity between different road network nodes;

[0013] Based on the spatial dependency features, a time convolution process is performed through a time gating unit to obtain the spatiotemporal dependency features of each road network node;

[0014] The charging demand prediction value of each road network node is determined based on the time-space dependence characteristics.

[0015] Preferably, the calculation formula of the dynamic critical matrix is ​​specifically:

[0016]

[0017] Where, The dynamic adjacency matrix value between network node i and network node j at time t, and are the heat map values ​​of network node i and network node j at time t, is the scaling factor for the difference in thermal values.

[0018] Preferably, the calculation formula for generating the dynamic heat map is specifically:

[0019]

[0020] Where, is the thermal value of the network node i at time t, is the topological distance between network node i and vehicle j, is the dynamic traffic bandwidth coefficient at time t, is the Gaussian kernel function, and n is the number of vehicles.

[0021] Preferably, after constructing the dynamic heat map, the following steps are further included:

[0022] The dynamic heat map is optimized through a graph convolutional neural network.

[0023] At the same time, the second aspect of the present application provides an electric vehicle charging demand prediction device, comprising:

[0024] Data collection unit, used to collect traffic charging network data and external factor data;

[0025] a heat map construction unit, configured to construct a dynamic heat map based on the traffic charging network data by using a kernel density estimation method, wherein the dynamic heat map is used to reflect the traffic flow status of each road network node;

[0026] a feature fusion unit, configured to perform feature alignment on the dynamic heat map and the external factor data based on time and space dimensions, and perform cross-modal attention fusion of the features of the dynamic heat map and the features of the external factor data through multimodal attention processing to obtain a fusion vector;

[0027] The charging demand prediction unit is used to determine the charging demand prediction value of each road network node based on the fusion vector and in combination with a preset spatiotemporal graph neural network model.

[0028] A third aspect of the present application provides an electric vehicle charging demand prediction terminal, comprising: a memory and a processor;

[0029] The memory is used to store program code, and the program code is used to implement a method for predicting charging demand of an electric vehicle as provided in the first aspect of the present application;

[0030] The processor is configured to read and execute the program code.

[0031] The fourth aspect of the present application provides a computer-readable storage medium, in which program code is stored. The program code is used to be read and executed by a processor to implement an electric vehicle charging demand prediction method as provided in the first aspect of the present application.

[0032] It can be seen from the above technical solutions that this application has the following advantages:

[0033] The solution provided in this application constructs a dynamic heat map reflecting the traffic flow status of each road network node based on traffic charging network data through kernel density estimation, uses the dynamic heat map to capture traffic flow changes in real time, combines the graph convolution network to characterize the road network topology constraints, and then aligns the features of the dynamic heat map and the external factor data based on the time and space dimensions. Through multimodal attention processing, the features of the dynamic heat map and the features of the external factor data are cross-modally fused with attention, so as to effectively integrate the contextual information of heterogeneous data and obtain a fusion vector. Based on the fusion vector, combined with the preset spatiotemporal graph neural network model, the charging demand forecast value of each road network node is determined, which overcomes the problem of spatiotemporal granularity differences of multi-source data, significantly improves the robustness of demand forecasting in emergencies and high volatility scenarios, and thus improves the accuracy of charging demand forecasting in extreme scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0035] Figure 1 This is a flow chart of an embodiment of a method for predicting charging demand of electric vehicles provided in this application.

[0036] Figure 2 This is an overall framework diagram of an embodiment of an electric vehicle charging demand prediction method provided in this application.

[0037] Figure 3 This is a detailed flow chart of step 104 of the electric vehicle charging demand prediction method provided in this application.

[0038] Figure 4 This is a logical framework diagram of the dynamic spatiotemporal graph neural network prediction model in the electric vehicle charging demand prediction method provided in this application.

[0039] Figure 5 This is a structural diagram of an embodiment of an electric vehicle charging demand prediction device provided in this application.

[0040] Figure 6 This is a structural diagram of an embodiment of an electric vehicle charging demand prediction terminal provided in this application. DETAILED DESCRIPTION

[0041] The embodiments of the present application provide a method, device, terminal and medium for predicting electric vehicle charging demand, which are used to solve the technical problem that existing charging load prediction technology has inaccurate prediction when facing some extreme scenarios with dynamic traffic scenes and complex road network spatial relationships.

[0042] In order to make the purpose, features, and advantages of the invention of this application more obvious and easy to understand, the technical solutions in the embodiments of this application will be clearly and completely described below in conjunction with the drawings in the embodiments of this application. Obviously, the embodiments described below are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0043] First, a detailed description of an embodiment of a method for predicting charging demand of an electric vehicle provided by this application is as follows:

[0044] See also Figure 1 and Figure 2 , this application provides a method for predicting electric vehicle charging demand, comprising:

[0045] Step 101: Collecting traffic charging network data and external factor data;

[0046] It should be noted that the first is dynamic or static data such as transportation charging network data and external factor data. Among them, transportation charging network data includes: vehicle GPS trajectory, traffic flow data, charging pile status and road network topology. External factor data includes: weather data, electricity price data and historical demand data. The complete data collection information is shown in Table 1:

[0047]

[0048] By mapping GPS trajectories to specific road network nodes through Map-Matching, we can obtain more realistic vehicle distribution and road occupancy information; charging piles and traffic cameras can provide near-real-time status data, which, combined with factors such as weather and electricity prices, can reflect the impact of the external environment on charging demand. After this step, we can obtain an initial multi-source spatiotemporal input matrix and the adjacency matrix , where N represents the number of nodes in the road network and D represents the feature dimension of a single node.

[0049] Step 102: construct a dynamic heat map based on the transportation charging network data through kernel density estimation;

[0050] Among them, the dynamic heat map is used to reflect the traffic flow status of each road network node;

[0051] It should be noted that based on the data obtained in the previous step, a dynamic heat map is generated on the basis of these data to preliminarily depict the traffic flow, congestion level, and potential charging demand distribution of road network nodes. The dynamic heat map of this embodiment can be generated using a heat map generation algorithm based on adaptive kernel density estimation (AKDE).

[0052]

[0053] Where, is the thermal value of node i at time t; is the topological distance between node i and vehicle j. This embodiment uses topological distance, which is different from the simple Euclidean distance and is more consistent with the constraints of the actual road network: if two nodes are connected by multiple roads with high-level sections, the distance may be small; if a detour is required or congestion is severe, the distance is increased to reflect the actual traffic flow path and traffic conditions; is the dynamic bandwidth coefficient, and its value can be adjusted according to the actual traffic flow changes. During peak hours, the traffic flow variance is large, and the bandwidth can be appropriately increased to improve the smoothness of the kernel function; during off-peak hours, the bandwidth is reduced to focus on local hot spots and highlight local differences; is a Gaussian kernel function used to map distance to density contribution.

[0054] Step 103: Based on the time and space dimensions, the dynamic heat map and the external factor data are feature aligned. Through multimodal attention processing, the features of the dynamic heat map and the features of the external factor data are cross-modally fused to obtain a fusion vector.

[0055] It should be noted that the dynamic heat map obtained in the previous step is aligned with the external factor data in the temporal and spatial dimensions; and the most critical external factors are highlighted through the attention mechanism to enrich the input information of the prediction model. After obtaining the optimized heat map, the spatiotemporal feature extraction and alignment strategy is shown in Table 2:

[0056]

[0057] After resampling to the same time step, the model can be processed in parallel ; Environmental feature normalization (such as electricity prices) is used to prevent the magnitude of the numbers from being too large or too small, affecting training stability.

[0058] Afterwards, a multimodal attention mechanism was introduced to highlight the relative importance of different modalities at different times. Dynamic heat map features (road traffic information) were fused with external factors such as weather, electricity prices, and historical demand data through cross-modal attention. This allowed the most relevant features to be dynamically highlighted and redundant information to be suppressed during the model input phase. The details are as follows:

[0059] 1. Multimodal input: For node i at time t, several modal feature vectors unified to the same time step and node index are obtained through the processing in the previous steps. Among them, some modal feature vectors involved in this embodiment are represented as follows:

[0060]

[0061] Among them, among them, is the traffic flow characteristic of node i at time t; is the weather characteristics of node i at time t, such as temperature, humidity, etc.; represents the electricity price information of node i at time t; It represents the historical demand characteristics of node i at time t and the time series information of past demand.

[0062] 2. Mapping to Q / K / V space: To enable the attention mechanism to identify the importance of each modality, we use a trainable linear transformation for each modality. The specific calculation formula is as follows:

[0063]

[0064] Where m represents the modality type (such as traffic, weather, price, etc.), The trainable linear transformation parameters of the modality m are trainable parameters; is the eigenvector of the mth mode of node i at time t.

[0065] 3. Attention calculation: Combine Q, K, and V of different modalities to obtain their weighted output. The weighted calculation formula is as follows:

[0066]

[0067] Based on the weighted output, the most influential modal feature at that moment / node is given a higher weight, where Q, K, and V are the query, key, and value matrices from different modalities, respectively, representing the information of different modalities; It is a feature dimension scaling factor used to stabilize the size of the inner product and avoid excessively large or small gradients; is the standard softmax function used to normalize the output.

[0068] Then, the above weighted outputs are combined to obtain , represents the multimodal feature vector of node i after fusion at time t. Summarizing all N nodes, we can get ,in is the dimension obtained after multimodal fusion.

[0069] Step 104: Based on the fusion vector and in combination with a preset spatiotemporal graph neural network model, the charging demand prediction value of each road network node is determined.

[0070] It should be noted that, based on the fusion vector obtained in step 103, a dynamic spatiotemporal neural network model is used to predict the charging load demand to determine the charging demand prediction value of each road network node, so as to reasonably dispatch power resources according to the charging demand prediction result.

[0071] More specifically, if Figure 3 As shown, the steps 104 provided in this embodiment specifically include:

[0072] Step 1041: Based on the fused vector and combined with the dynamic adjacency matrix, the fused vector is subjected to neighborhood aggregation processing by a spatial graph convolution unit to obtain spatial dependency features between road network nodes.

[0073] Among them, the dynamic critical matrix is ​​used to reflect the similarity between different road network nodes;

[0074] Step 1042: Based on the spatial dependency features, a time convolution process is performed by a time gating unit to obtain the spatiotemporal dependency features of each road network node.

[0075] Step 1043: Determine the charging demand prediction value of each road network node based on the spatiotemporal dependency characteristics.

[0076] It should be noted that if Figure 4 First, based on the fusion vector obtained in step 103 and the dynamic adjacency matrix constructed based on the dynamic heat map, the fusion vector is subjected to neighborhood aggregation processing through the spatial graph convolution unit to obtain the spatial dependency characteristics between the road network nodes. The construction expression of the dynamic adjacency matrix in this embodiment is specifically as follows:

[0077]

[0078] Where, The dynamic adjacency matrix value between network node i and network node j at time t, and are the heat map values ​​of network node i and network node j at time t, It is the scaling factor of the thermal value difference, which is used to control the impact of the thermal difference between nodes on the connection weight. It changes in the time dimension, allowing the association between nodes to be dynamically adjusted with the thermal difference of the traffic flow.

[0079] After completing the dynamic adjacency matrix, spatial graph convolution is used to utilize right Perform neighborhood aggregation and extract local spatial dependencies. The specific operation expression is as follows:

[0080]

[0081] in, is the dynamic adjacency matrix at time t, representing the spatial dependence between nodes, is the fusion vector matrix at time t, It is a graph convolution operation used to aggregate information of adjacent nodes.

[0082] After completing the spatial graph convolution, time gating processing is performed. The specific operation expression is as follows:

[0083]

[0084] in, is the spatial convolution result at time t, i.e., the spatial dependency feature, is the weight of the convolution kernel, is the expansion coefficient, K is the convolution kernel size, which indicates how many time steps of data need to be integrated in each convolution operation. This operation can capture long sequence dependencies and take into account parallel computing efficiency.

[0085] Then, through a layer of perceptron mapping, we can get the The charging demand forecast value at the moment:

[0086]

[0087] in, is the predicted charging demand, indicating the time The corresponding demand forecast value of each node; is the result of time gating processing at time t, that is, the spatiotemporal dependency feature, is a mapping function used to map the spatiotemporal features to the final predicted values, wherein the mapping relationship between the spatiotemporal features and the predicted values ​​can be specifically determined through model training.

[0088] Among them, the training examples of graph convolutional networks and time-gated networks are as follows:

[0089] a. Pre-training GCN part, first fully combine the heat map with the road network information.

[0090] b. During the end-to-end fine-tuning phase, multimodal attention, dynamic adjacency, and TCN are trained together to streamline the overall data flow.

[0091] Through the preset loss function, the GCN pre-training can ensure that the road network information is reasonably captured; the final loss considers both numerical accuracy and spatial coherence, making the prediction results more accurate and reasonable;

[0092] The expression of the loss function is as follows:

[0093]

[0094] in, represents the predicted charging demand vector output by the model, is the actual observed charging demand vector, MAE is the mean absolute error, which measures the prediction accuracy, Spatial-Smoothness is a spatial smoothness constraint that encourages the predicted values ​​between adjacent nodes to avoid sudden jumps, and α and β are the weights for controlling different objectives.

[0095] The real-time update mechanism can adopt the edge-cloud collaborative architecture and deploy the pruned or quantized DSTGNN model at the edge (RSU) to reduce the number of parameters and computational overhead; when encountering a sudden change in the heat map ( ) timely alarm and local execution of short-term forecasts, where The heat map mutation threshold is used. If vehicle distribution or demand exceeds a preset change range within a short period of time, it is considered an anomaly or emergency, triggering an edge response strategy. The cloud receives edge monitoring results and local features, performs incremental learning to update the master model weights, and sends the latest weights to the edge for synchronization. This collaborative model enables fast minute-level predictions and localized warnings at the edge, while the cloud performs batch or incremental training on larger-scale historical data, allowing the model to continuously adapt to evolving traffic and demand distributions.

[0096]

[0097] in is the updated model weight; is the current model weight; is the learning rate, which controls the step size of weight update; is the gradient of the loss function, which is used to guide the model on how to adjust the weights; It is the mutation area data from the edge, which serves as the input of incremental learning.

[0098] Through the above steps, this patented solution successfully combines dynamic heat maps with graph neural networks, achieves a deep integration of vehicle distribution and road network topology, and finely depicts the dynamic evolution of charging demand in space; and through multimodal attention and TCN and other time series technologies, overcomes the problem of spatiotemporal granularity differences in multi-source data, significantly improving the robustness of demand forecasting in emergencies and high-volatility scenarios; finally, cloud-edge collaboration and lightweight models are used to achieve real-time monitoring and emergency response of charging demand, effectively shortening the closed-loop cycle of perception-decision-update. The entire system architecture is as follows: Figure 2 In practical applications, this solution can provide minute-level charging demand forecasting and dynamic hotspot identification, helping power grid operators to more efficiently deploy charging piles and balance power loads.

[0099] Furthermore, the solution provided in this embodiment may further include the following steps after step 102 and before step 103:

[0100] Step 1021: Optimize the dynamic heat map through a graph convolutional neural network.

[0101] It should be noted that after obtaining the dynamic heat map in step 102, the initial heat map can be optimized through a graph convolutional network (GCN). As shown in the optimization expression mentioned below, GCN can optimize the accuracy and adaptability of the heat map by mining the complex topological relationships in spatial data. The optimization expression is as follows:

[0102] .

[0103] in, is the thermal value set or matrix of the dynamic thermal map at time t before optimization, is the optimized heat value set or matrix of the dynamic heat map at time t. L is the road network adjacency matrix, which expresses the actual road connections and weights (such as highways / ordinary roads) between nodes. A higher value indicates a closer connection between nodes. It is a graph convolution operation used to fuse the spatial relationships between nodes and can generally be set to 3 layers.

[0104] It is understood that if the method of this embodiment includes step 1021, then the subsequent steps involve thermal values. The steps are optimized Replace the original For example, the expression of the dynamic adjacency matrix mentioned in step 1041 can be adjusted to:

[0105]

[0106] The above is a detailed description of an embodiment of an electric vehicle charging demand prediction method provided by the present application. The following is a detailed description of an embodiment of an electric vehicle charging demand prediction device provided by the present application.

[0107] See also Figure 5 , the present application provides an embodiment of an electric vehicle charging demand prediction device, comprising:

[0108] Data collection unit 201, used to collect traffic charging network data and external factor data;

[0109] A heat map construction unit 202 is used to construct a dynamic heat map based on the traffic charging network data by using a kernel density estimation method, wherein the dynamic heat map is used to reflect the traffic flow status of each road network node;

[0110] The feature fusion unit 203 is used to align the features of the dynamic heat map and the external factor data based on the time and space dimensions, and perform cross-modal attention fusion on the features of the dynamic heat map and the features of the external factor data through multimodal attention processing to obtain a fusion vector;

[0111] The charging demand prediction unit 204 is used to determine the charging demand prediction value of each road network node based on the fusion vector and in combination with a preset spatiotemporal graph neural network model.

[0112] In addition, the present application also provides a detailed description of related embodiments such as an electric vehicle charging demand prediction terminal and a computer-readable storage medium.

[0113] like Figure 6 As shown, the present application provides an embodiment of an electric vehicle charging demand prediction terminal, the implementation types of which include but are not limited to: personal computers, industrial computers, servers and embedded intelligent devices, and the main components of the terminal include: memory 33 and processor 31, and the memory 33 and processor 31 can be connected via a communication bus 34;

[0114] The memory 33 is used to store program codes, and the program codes are used to implement a method for predicting charging demand of an electric vehicle as described in the above embodiment;

[0115] The processor 31 is used to read and execute program codes.

[0116] In a fourth aspect, the present application provides a computer-readable storage medium, in which program code is stored. The program code is used to be read and executed by a processor to implement a method for predicting electric vehicle charging demand as described in the above embodiment.

[0117] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the terminals, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0118] In the several embodiments provided in this application, it should be understood that the disclosed terminals, devices, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interface, device or unit, which can be electrical, mechanical or other forms.

[0119] The terms "first," "second," "third," "fourth," and the like (if any) in the specification of the present application and the accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a particular order or precedence. It should be understood that the terms used in this manner are interchangeable where appropriate, so that the embodiments of the present application described herein, for example, can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having," and any variations thereof, are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or apparatus comprising a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such process, method, product, or apparatus.

[0120] It should be understood that in this application, "at least one (item)" means one or more, and "plurality" means two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" can mean: only A exists, only B exists, and A and B exist at the same time, where A and B can be singular or plural. The character " / " generally indicates that the previous and next associated objects are in an "or" relationship. "At least one of the following items" or similar expressions refers to any combination of these items, including any combination of single items or plural items. For example, at least one of a, b or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.

[0121] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0122] In addition, the functional units in the various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0123] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes various media that can store program code, such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0124] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for predicting charging demand of electric vehicles, characterized in that: include: Collect transportation charging network data and external factor data; Based on the traffic charging network data, a dynamic heat map is constructed by kernel density estimation. The dynamic heat map is used to reflect the traffic flow status of each road network node. The calculation formula for generating the dynamic heat map is specifically as follows: ； Where, is the thermal value of the network node i at time t, is the topological distance between network node i and vehicle j, is the dynamic traffic bandwidth coefficient at time t, is the Gaussian kernel function, n is the number of vehicles; Based on the time and space dimensions, the dynamic heat map and the external factor data are feature aligned, and the features of the dynamic heat map and the features of the external factor data are cross-modally fused through multimodal attention processing to obtain a fusion vector; Based on the fused vector and in combination with a dynamic adjacency matrix, a neighborhood aggregation process is performed on the fused vector through a spatial graph convolution unit to obtain spatial dependency features between the road network nodes, wherein the dynamic adjacency matrix is ​​used to reflect the similarity between different road network nodes; Based on the spatial dependency features, a time convolution process is performed through a time gating unit to obtain the spatiotemporal dependency features of each road network node; Determining a charging demand prediction value for each road network node based on the spatiotemporal dependency characteristics; The calculation formula of the dynamic adjacency matrix is ​​specifically: ； Where, The dynamic adjacency matrix value between network node i and network node j at time t, and are the heat map values ​​of network node i and network node j at time t, is the scaling factor for the difference in thermal values.

2. The method for predicting charging demand of electric vehicles according to claim 1, characterized in that: The traffic charging network data includes: vehicle GPS trajectory, traffic flow data, charging pile status and road network topology.

3. The method for predicting charging demand of electric vehicles according to claim 1, characterized in that: The external factor data includes: weather data, electricity price data and historical demand data.

4. The method for predicting charging demand of electric vehicles according to claim 1, characterized in that: Building a dynamic heat map also includes: The dynamic heat map is optimized through a graph convolutional neural network.

5. An electric vehicle charging demand prediction device, characterized in that: include: Data collection unit, used to collect traffic charging network data and external factor data; The heat map construction unit is used to construct a dynamic heat map based on the traffic charging network data by kernel density estimation, wherein the dynamic heat map is used to reflect the traffic flow status of each road network node. The generation calculation formula of the dynamic heat map is specifically: ； Where, is the thermal value of the network node i at time t, is the topological distance between network node i and vehicle j, is the dynamic traffic bandwidth coefficient at time t, is the Gaussian kernel function, n is the number of vehicles; a feature fusion unit, configured to perform feature alignment on the dynamic heat map and the external factor data based on time and space dimensions, and perform cross-modal attention fusion of the features of the dynamic heat map and the features of the external factor data through multimodal attention processing to obtain a fusion vector; a charging demand prediction unit, configured to perform neighborhood aggregation processing on the fused vector using a spatial graph convolution unit based on the fused vector and a dynamic adjacency matrix to obtain spatial dependency characteristics between the road network nodes, wherein the dynamic adjacency matrix is ​​used to reflect the similarity between different road network nodes; perform temporal convolution processing based on the spatial dependency characteristics using a time gating unit to obtain spatiotemporal dependency characteristics of each road network node; and determine a charging demand prediction value for each road network node based on the spatiotemporal dependency characteristics; The calculation formula of the dynamic adjacency matrix is ​​specifically: ； Where, The dynamic adjacency matrix value between network node i and network node j at time t, and are the heat map values ​​of network node i and network node j at time t, is the scaling factor for the difference in thermal values.

6. An electric vehicle charging demand prediction terminal, characterized in that: include: memory and processor; The memory is used to store program code, and the program code is used to implement the electric vehicle charging demand prediction method according to any one of claims 1 to 4; The processor is configured to read and execute the program code.

7. A computer-readable storage medium, characterized in that The computer-readable storage medium stores program code, and the program code is used to be read and executed by a processor to implement the electric vehicle charging demand prediction method according to any one of claims 1 to 4.

Images