Expressway charging demand prediction method and system based on space-time residual multi-graph convolutional network, terminal equipment and medium

Abstract

The invention discloses an expressway charging demand prediction method and system based on a space-time residual multi-graph convolutional network, terminal equipment and a medium, and relates to the technical field of intelligent transportation, and the method comprises the steps: obtaining the equipment data, geographic position data, historical charging order data and corresponding external environment data of an expressway charging station; based on the equipment and the geographic position data, a multi-graph comprising a geographic graph and a semantic graph is constructed with the charging stations as nodes, and the geographic graph and the semantic graph represent the geographic position relation and the internal attribute relation between the charging stations respectively; and inputting the historical charging order data, the external environment data and the multi-graph into a pre-trained space-time residual multi-graph convolutional network to obtain a highway charging demand prediction result. According to the method, multi-dimensional information such as road network topology, demand time sequence, weather and price is fully mined and fused, so that the prediction capability of the model is higher than that of a model formed by splicing a single data source or simple data, and accurate prediction of charging demand fluctuation is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent transportation, and in particular to a highway charging demand prediction method and system based on a spatio-temporal residual multi-graph convolution network, a terminal device and a medium. BACKGROUND

[0002] With the popularity of electric vehicles, its impact on power grid stability and transportation system is increasing. The highway charging network is the key to supporting long-distance travel of electric vehicles, and the large amount of charging data collected provides a basis for demand analysis. Highway charging demand is a dynamic and complex system, and its peak formation mechanism is influenced by multiple factors such as geography, time, weather, and holidays.

[0003] Although existing charging demand prediction research has widely applied traditional statistical models, machine learning, and deep learning methods, there are still significant limitations. First, most models lack the ability to capture the complex spatio-temporal dependencies of charging demand at different service areas and different time scales. Second, existing models face challenges in effectively integrating multi-source heterogeneous data such as road network topology and weather conditions, resulting in prediction accuracy that cannot meet the needs of fine-grained operation and management of power grids and transportation systems.

[0004] Therefore, there is an urgent need for a precise prediction method that can integrate multi-dimensional information to fill the gap in existing technology. SUMMARY

[0005] The technical problem to be solved by the present application is that in the existing technology in the field of intelligent transportation, there are two major problems in highway charging demand prediction. First, most models cannot fully capture the complex spatio-temporal dependencies of charging demand at different service areas and different time scales. Second, they cannot effectively integrate multi-source heterogeneous data such as road network topology and weather conditions, resulting in prediction accuracy that cannot meet the needs of fine-grained operation and management of power grids and transportation systems. Therefore, there is an urgent need for an effective solution to solve the above technical problems.

[0006] To solve the above technical problems, the technical solution adopted by the present application is as follows: In a first aspect, the present application provides a highway charging demand prediction method based on a spatio-temporal residual multi-graph convolution network, comprising: obtaining device data, geographic location data, and historical charging order data of charging stations on a highway, and obtaining external environmental data corresponding to the geographic location of the charging stations and the order time; based on the device data and the geographic location data, constructing a multi-graph with the charging stations as nodes, wherein the multi-graph includes a geographic graph and a semantic graph, the geographic graph is used to represent the geographic location relationship between the charging stations, and the semantic graph is used to represent the non-geographic internal attribute relationship between the charging stations; The historical charging order data, the external environment data, and the multi-graph input are used to pre-train a highway charging demand prediction model to obtain highway charging demand prediction results. The highway charging demand prediction model is a spatiotemporal residual multi-graph convolutional network.

[0007] In one implementation, constructing a multigraph based on the device data and the geographic location data, with the charging station as a node, includes: A geographic map is constructed based on the geographic location data. The geographic map includes at least one of a neighborhood correlation map, a distance correlation map, and a travel direction correlation map. The neighborhood correlation map is used to represent the adjacency relationship between the charging stations. The distance correlation map is used to represent the distance relationship between the charging stations. The travel direction correlation map is used to indicate whether the charging stations are charging stations in the same location but in different travel directions. A semantic graph is constructed based on the device data. The semantic graph includes at least one of a device correlation graph and a price correlation graph. The device correlation graph is used to represent the similarity of the device configuration of the charging station, and the price correlation graph is used to represent the high and low relationship of the charging price of the charging station.

[0008] In one implementation, before obtaining the highway charging demand prediction result by inputting the historical charging order data, the external environment data, and the multi-graph input pre-trained highway charging demand prediction model, the following steps are included: The external environment data is digitally encoded and normalized to obtain external environment characteristics, wherein the external environment data includes at least meteorological data; The historical charging order data is cleaned and aggregated and feature-constructed according to a preset time unit to obtain charging demand time series features, wherein the charging demand time series features are used to characterize the time-varying pattern of charging demand at each charging station.

[0009] In one implementation, the highway charging demand prediction model includes a decoder and an encoder with three parallel branches. The process of inputting the historical charging order data, the external environment data, and the multi-graph data into the pre-trained highway charging demand prediction model to obtain the highway charging demand prediction result includes: The charging demand time-series features, the external environment features, and the multigraph are respectively input into the encoder of the corresponding branch to obtain the corresponding time feature vector, environmental feature vector, and spatial feature vector; The temporal feature vector, environmental feature vector, and spatial feature vector are concatenated and fused to obtain the latent vector. The decoder recovers the latent vectors to obtain a shared feature tensor, and based on the shared feature tensor, the predicted charging demand is obtained.

[0010] In one implementation, the step of inputting the charging demand time-series features, the external environment features, and the multigraph into the encoder of the corresponding branch to obtain the corresponding time feature vector, environmental feature vector, and spatial feature vector includes: The multigraph is input into the first branch of the encoder to obtain a spatial feature vector. The first branch of the encoder is a spatial feature residual multigraph convolution module. The output of each layer of the spatial feature residual multigraph convolution module is the sum of the output of the previous layer and the feature matrix of the multigraph after graph convolution transformation. The charging demand time-series features are input into the second branch of the encoder to obtain a time feature vector, wherein the second branch of the encoder is a gated loop unit module, and the time feature vector is used to characterize the time dependency and periodicity of the charging demand. The external environment features are input into the third branch of the encoder to obtain the environment feature vector. The third branch of the encoder consists of two gating loop units and a fully connected layer.

[0011] In one implementation, the multiple methods recover the latent vectors through the decoder to obtain a shared feature tensor, and based on the shared feature tensor, obtain a predicted charging demand value, including: The latent vector is linearly transformed and reshaped through a fully connected layer of the decoder to obtain a two-dimensional shared feature tensor. The shared feature tensor is input into the regression prediction residual multi-graph convolutional sub-network, and after multi-layer graph convolution processing, the specific charging demand prediction value of each station at future time is obtained.

[0012] In one implementation, the training process of the highway charging demand prediction model includes: The training dataset is input into the highway charging demand prediction model to be trained, and the combination of mean squared error loss and regularization term is used as the optimization objective. The model is iteratively trained until the optimization objective converges. The performance of the trained model was verified using root mean square error and mean absolute error, resulting in the pre-trained highway charging demand prediction model.

[0013] Secondly, embodiments of the present invention also provide a highway charging demand prediction system based on a spatiotemporal residual multi-graph convolutional network, the system comprising: The input data acquisition module is used to acquire equipment data, geographical location data, and historical charging order data of charging stations on highways, and to acquire external environmental data corresponding to the geographical location and order time of the charging station. A multi-graph construction module is used to construct a multi-graph based on the device data and the geographic location data, with the charging station as the node. The multi-graph includes a geographic graph and a semantic graph. The geographic graph is used to represent the geographic location relationship between the charging stations, and the semantic graph is used to represent the non-geographical intrinsic attribute relationship between the charging stations. The model prediction module is used to input the historical charging order data, the external environment data and the multi-graph into a pre-trained highway charging demand prediction model to obtain highway charging demand prediction results. The highway charging demand prediction model is a spatiotemporal residual multi-graph convolutional network.

[0014] Thirdly, embodiments of the present invention also provide a terminal device, the terminal device including a memory, a processor, and a highway charging demand prediction program based on a spatiotemporal residual multi-graph convolutional network stored in the memory and executable on the processor. When the processor executes the highway charging demand prediction program based on a spatiotemporal residual multi-graph convolutional network, it implements the steps of the highway charging demand prediction method based on a spatiotemporal residual multi-graph convolutional network as described in any of the above schemes.

[0015] Fourthly, embodiments of the present invention also provide a computer-readable storage medium storing a highway charging demand prediction program based on a spatiotemporal residual multi-graph convolutional network. When the highway charging demand prediction program based on a spatiotemporal residual multi-graph convolutional network is executed by a processor, it implements the steps of the highway charging demand prediction method based on a spatiotemporal residual multi-graph convolutional network as described in any of the above schemes.

[0016] Beneficial Effects: This invention discloses a method, system, terminal device, and medium for predicting highway charging demand based on a spatiotemporal residual multi-graph convolutional network, relating to the field of intelligent transportation technology. The method first acquires equipment data, geographical location data, and historical charging order data of highway charging stations, and acquires external environmental data corresponding to the geographical location and order time of the charging stations. Then, based on the equipment data and the geographical location data, a multi-graph is constructed with the charging stations as nodes. The multi-graph includes a geographic graph and a semantic graph. The geographic graph represents the geographical location relationships between the charging stations, and the semantic graph represents the non-geographical intrinsic attribute relationships between the charging stations. Finally, the historical charging order data, the external environmental data, and the multi-graph are input into a pre-trained highway charging demand prediction model to obtain the highway charging demand prediction result. The highway charging demand prediction model is a spatiotemporal residual multi-graph convolutional network. This invention constructs a multi-graph system containing geographic and semantic graphs, combined with a spatiotemporal residual multi-graph convolutional network. It fully mines and integrates multi-dimensional information such as road network topology, demand time series, weather, and price, accurately capturing the complex spatiotemporal dependence and multi-source data correlation of charging demand. This makes the model's predictive ability higher than that of models with a single data source or simple data splicing, achieving accurate prediction of charging demand fluctuations. Furthermore, it provides support for the layout and operation of charging facilities, power planning, and traffic management, alleviating charging congestion and improving highway service and operational efficiency. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating a specific implementation method for a highway charging demand prediction method based on a spatiotemporal residual multi-graph convolutional network, as provided in an embodiment of the present invention.

[0018] Figure 2 This is a schematic diagram of the process of a highway charging demand prediction method based on a spatiotemporal residual multi-graph convolutional network provided in an embodiment of the present invention.

[0019] Figure 3 This is a schematic diagram of the prediction model flow in the highway charging demand prediction method based on spatiotemporal residual multi-graph convolutional network provided in the embodiments of the present invention.

[0020] Figure 4 This is a schematic diagram of the principle of a highway charging demand prediction device based on a spatiotemporal residual multi-graph convolutional network provided in an embodiment of the present invention.

[0021] Figure 5 This is a block diagram illustrating the internal structure of the terminal device provided in an embodiment of the present invention. Detailed Implementation

[0022] To make the objectives, technical solutions, and effects of this invention clearer and more explicit, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0023] The flowchart shown in the attached diagram is for illustrative purposes only and does not necessarily include all content, operations, or steps, nor does it require execution in the described order. For example, some operations or steps can be broken down, combined, or partially merged, so the actual execution order may change depending on the actual situation.

[0024] It should be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0025] It should be understood that, in order to clearly describe the technical solutions of the embodiments of the present invention, the terms "first" and "second" are used in the embodiments of the present invention to distinguish identical or similar items with essentially the same function and effect. For example, "first control information" and "second control information" are only used to distinguish different control information and do not limit their order.

[0026] Those skilled in the art will understand that the words "first" and "second" do not limit the quantity or the order of execution, and that the words "first" and "second" do not necessarily imply that they are different.

[0027] It should also be understood that the term “and / or” as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0028] With the increasing prevalence of electric vehicles, their impact on power grid stability and transportation systems is deepening. Highway charging networks are crucial for supporting long-distance electric vehicle travel, and the vast amounts of charging data collected provide a foundation for demand analysis. Highway charging demand is a dynamic and complex system, with peak formation mechanisms influenced by multiple factors including geography, time, weather, and holidays.

[0029] Although existing research on charging demand forecasting has widely applied traditional statistical models, machine learning, and deep learning methods, significant limitations remain. First, most models are insufficient in capturing the complex spatiotemporal dependencies of charging demand across different service areas and time scales. Second, existing models face challenges in effectively integrating multi-source heterogeneous data such as road network topology and meteorological conditions, resulting in forecast accuracy that cannot meet the needs of refined operation and management of power grids and transportation systems.

[0030] Therefore, in the existing technologies in the field of intelligent transportation, there are two major problems in the prediction of highway charging demand. First, most models are unable to fully capture the complex spatiotemporal dependence of charging demand in different service areas and at different time scales. Second, they cannot effectively integrate multi-source heterogeneous data such as road network topology and meteorological conditions, resulting in the prediction accuracy failing to meet the needs of refined operation and management of power grid and transportation systems.

[0031] To address the aforementioned technical challenges, this invention proposes a method, system, terminal device, and medium for predicting highway charging demand based on a spatiotemporal residual multi-graph convolutional network. The method first acquires equipment data, geographic location data, historical charging order data, and corresponding external environment data for highway charging stations. Then, based on the equipment and geographic location data, a multi-graph is constructed with charging stations as nodes, comprising a geographic graph and a semantic graph. The geographic graph and semantic graph represent the geographic location relationships and intrinsic attribute relationships between charging stations, respectively. Finally, the historical charging order data, external environment data, and the multi-graph are input into a pre-trained spatiotemporal residual multi-graph convolutional network to obtain the highway charging demand prediction result. This invention fully mines and integrates multi-dimensional information such as road network topology, demand time series, weather, and prices, making the model's predictive ability higher than models based on a single data source or simple data splicing, achieving accurate prediction of charging demand fluctuations.

[0032] Figure 2 This paper demonstrates several steps in the proposed highway charging demand prediction method based on a spatiotemporal residual multi-graph convolutional network. Specifically, it first acquires a dataset and performs data preprocessing and temporal transformation. Then, it constructs a multi-graph association structure and defines the temporal features. Based on this, a spatiotemporal residual multi-graph convolutional network prediction model is built. Multi-source features are extracted in the parallel encoder of this model, and the multi-source features are concatenated and fused to obtain a latent vector. The latent vector is decoded by the decoder to reshape the spatial features, and the prediction result is generated through the residual network of the decoder. Furthermore, for this spatiotemporal residual multi-graph convolutional network prediction model, an optimization objective function needs to be defined for training and optimization.

[0033] This embodiment provides a method for predicting highway charging demand based on spatiotemporal residual multi-graph convolutional networks, such as... Figure 1 As shown, the specific steps include the following: Step S100: Obtain equipment data, geographical location data and historical charging order data of the charging stations on the highway, and obtain external environment data corresponding to the geographical location and order time of the charging stations.

[0034] In this embodiment, the layout information of charging stations along the highway is first obtained based on existing map data, including the geographical location data of each charging station. The geographical location data consists of the spatial location information of each charging station, associated with its location on the highway, service area, and specific coordinates, and is used to subsequently construct a geographic map to represent the geographical relationships between the stations.

[0035] Next, the core data for the task to be predicted is obtained. Specifically, endogenous charging order data is collected, namely historical charging orders from charging stations in various highway service areas. Taking one data sample as an example, this data sample includes fields such as the name of the charging station, its geographical location, the highway and highway section it belongs to, the model of the charging equipment, the start time, the end time, the charging amount, the electricity cost, and the service fee.

[0036] After obtaining the geographical distribution information of charging stations along the highway and historical charging order data, exogenous environmental data is collected. This exogenous environmental data may include data determined by non-geographical factors such as meteorological data and whether the date is a holiday. Specifically, external environmental data corresponding to the location of the charging station and the order time is acquired. This data, as an exogenous variable affecting charging demand, may include characteristics across multiple dimensions such as weather type, temperature, humidity, cloud cover, wind speed, and whether the date is a holiday in the area.

[0037] Step S200: Based on the device data and the geographic location data, construct a multi-graph with the charging station as the node. The multi-graph includes a geographic graph and a semantic graph. The geographic graph is used to represent the geographic location relationship between the charging stations, and the semantic graph is used to represent the non-geographical intrinsic attribute relationship between the charging stations.

[0038] In this embodiment, based on the definitions of graph and adjacency matrix, the association graph of charging stations in different service areas and different directions of highway is defined as a directed graph, where each node of the graph refers to a charging station, which is a set of charging stations, and the edges and adjacency matrix of the graph represent the relationships between charging stations.

[0039] Based on this, a multi-graph structure is constructed to represent the complex spatiotemporal relationships between charging stations. The multi-graph mainly includes a geographic graph and a semantic graph, where the geographic graph is used to represent the geographical location relationships between charging stations, while the semantic graph is used to represent the non-geographical intrinsic attribute relationships between the charging stations.

[0040] In one implementation, constructing a multigraph based on the device data and the geographic location data, with the charging station as a node, specifically includes the following steps: Step S210: Construct a geographic map based on the geographic location data. The geographic map includes at least one of a neighborhood correlation map, a distance correlation map, and a travel direction correlation map. The neighborhood correlation map is used to represent the adjacency relationship between the charging stations. The distance correlation map is used to represent the distance relationship between the charging stations. The travel direction correlation map is used to represent whether the charging stations are charging stations in the same location but in different travel directions. Step S220: Construct a semantic graph based on the device data. The semantic graph includes at least one of a device correlation graph and a price correlation graph. The device correlation graph is used to represent the similarity of the device configuration of the charging station, and the price correlation graph is used to represent the high and low relationship of the charging price of the charging station.

[0041] In this embodiment, the basic network graph structure of the multigraph is first defined. Specifically, the highway charging network is abstracted into the basic network graph structure. Each charging station is a node. Connections between nodes Through the adjacency matrix express.

[0042] Subsequently, as Figure 3 As shown, a geographic map representing spatial correlations determined by geographical location is constructed based on the basic network graph structure. The geographic map must include at least one of the following: a neighborhood correlation map, a distance correlation map, or a row-direction correlation map.

[0043] Neighborhood correlation graph Used to indicate whether service area stations are geographically adjacent, defined as ,in, The adjacency matrix represents the neighborhood correlation graph, indicating whether charging stations are geographically adjacent. Each element is expressed as:

[0044] in, The service area refers to the neighboring matrix of adjacent regions with neighborhood relationships. Line number The elements in the column.

[0045] Distance Correlation Plot To quantify the impact of distance between different charging stations on demand, a Gaussian kernel function is used to calculate distance similarity. Distance correlation plot. Defined as ,in, The adjacency matrix represents the distance similarity of charging stations in the distance correlation graph, where each element is expressed as:

[0046] in, Similarity in distance between charging stations Line number Elements in the column This is the weighted distance value. , Represents the Euclidean distance, based on the weighting parameters. Obtain the weighted distance value. Spatial weights represent location. Position The influence of Gaussian kernel weights This is the maximum value in the distance matrix.

[0047] Row correlation diagram This is used to represent the association between two charging stations in the same service area but in different directions (such as eastbound and westbound), defined as follows: ,in, The adjacency matrix represents the correlation graph of the travel directions, used to characterize whether charging stations are charging stations in the same service area of ​​the highway but in different travel directions. The expression for each element is:

[0048] in, The first row in the relation matrix of different rows within the same service area Line number The elements in the column.

[0049] Based on the basic network graph structure, a semantic graph is constructed to express the intrinsic relationships determined by non-geographical factors. The semantic graph should include at least one of the following: an equipment correlation graph or a price correlation graph.

[0050] Equipment correlation diagram Based on a set of equipment models at charging stations, the similarity of equipment configurations between stations is calculated using the Jaccard similarity coefficient. The equipment correlation graph is defined as follows: ,in, The adjacency matrix represents the device correlation graph, used to characterize device configuration similarity. Each element is expressed as:

[0051] in, The first point refers to the similarity of equipment at different sites. Line number Elements in the column and Sites and A collection of device models, Indicates the number of shared equipment models. The Jaccard similarity coefficient represents the union size of the device models. Specifically, it is the ratio of the intersection to the union of different charging station equipment model sets, used to quantify the similarity of equipment configuration between charging stations. The constructed matrix is ​​symmetric and has a diagonal value of 1.

[0052] Price Correlation Chart This is used to represent the relative price levels of charging at different charging stations. The charging price may include a service fee. A price correlation graph is defined as follows: ,in, The adjacency matrix representing the price correlation graph is used to characterize the relative high and low relationships of charging prices. The expression for each element is:

[0053] in, The first point refers to the similarity of prices across different websites. Line number The elements in the column. Suppose there are several ranking metrics used to represent the relative ranking of charging stations on price-related indicators. Indicates charging station In terms of indicators The ranking value on the screen Indicates charging station In terms of indicators The ranking value on the website. Indicates station In terms of indicators The ranking on the site is better than the site hour. Indicates station On which indicators does it rank higher than the field? That is, how many indicators are there? ,satisfy . The number of indicators used to represent the price evaluation includes the unit price of electricity, the unit price of service fees, and the total price. Used for normalization.

[0054] Multiple graphs were constructed to represent the relationships between charging stations. For location relationship representation, a combination test was conducted using neighborhood graphs, Euclidean distance graphs, geographic weighted regression correlation graphs, and row-direction relationship graphs. The results demonstrated that combining neighborhood graphs with row-direction relationship graphs most effectively represents the physical location associations between charging stations, providing the optimal spatial feature input for the model, and outperforming single distance or geographic correlation measures.

[0055] In one implementation, before obtaining the highway charging demand prediction result by inputting the historical charging order data, the external environment data, and the multi-graph input to the pre-trained highway charging demand prediction model, the following steps are specifically included: Step S230: Digitally encode and normalize the external environment data to obtain external environment characteristics, wherein the external environment data includes at least meteorological data; Step S240: Clean the historical charging order data and aggregate and construct features according to a preset time unit to obtain charging demand time series features, wherein the charging demand time series features are used to characterize the time-varying pattern of charging demand for each charging station.

[0056] In this embodiment, the external environmental data, including meteorological data, is first preprocessed. Categorical data such as weather types are digitally encoded, while numerical data such as temperature and humidity are processed using the minimum-maximum normalization method to ensure data scale consistency. Specifically, weather types and cloud cover in the external environmental data are categorical data, whose form cannot be directly used by the model and therefore require digital encoding. For example, label encoding is used to map different weather types such as sunny, cloudy, and rainy to unique integers, such as mapping sunny to 0, cloudy to 1, and rainy to 2, thus converting categorical data into numerical form for easier subsequent analysis. For numerical data such as temperature, humidity, and wind speed in the external environmental data, since the original data have different scales and dimensions, direct use may cause the model to learn biased towards features with larger values, affecting prediction accuracy. Therefore, the minimum-maximum normalization method is used to normalize the data to the (0,1) interval, ensuring data scale consistency. This processing unifies numerical data of different dimensions and value ranges to the same scale, improving the accuracy and stability of subsequent models in learning data features and providing a data foundation for accurately predicting highway charging demand.

[0057] Subsequently, as Figure 3 As shown, the data cleaning process specifically involves removing invalid records from the original order data, including records where key fields such as station name, start time, end time, and charging amount are empty, records where the charging time is too long or too short, abnormal records where the payment amount is zero, and abnormal data points where the service fee is too high or too low and deviates from the reasonable range.

[0058] After cleaning, the order data is aggregated according to a preset hourly time unit, and the number of charging demand orders for each charging station within each hourly time period is counted. The aggregated data is then arranged in date and time order, forming a dataset with charging stations as columns and time as rows. Based on this dataset, feature construction is further performed to obtain temporal features of charging demand. Specifically, based on the temporal variation pattern of charging demand, an input feature sequence is constructed for each prediction time of the model. This input feature sequence is used as the input to the prediction model. The input feature sequence specifically includes the demand from the two time periods preceding the prediction time. and The requirement for the same time interval the day before the prediction time. The demand for the same time interval one week prior to the prediction time. .set up Indicates the first Charging station at The demand for a time interval can be specifically represented as the number of charging orders, whereby... ,make Indicates the first The time interval of a day The demand for charging stations is analyzed. This time-series characteristic of charging demand reflects the changes in charging demand at each charging station at different time points, and is used to extract time features such as proximity, trend, and periodicity, providing a data foundation for the model to capture the time dynamics of charging demand.

[0059] Step S300: Input the historical charging order data, the external environment data and the multi-graph into the pre-trained highway charging demand prediction model to obtain the highway charging demand prediction result, wherein the highway charging demand prediction model is a spatiotemporal residual multi-graph convolutional network.

[0060] In this embodiment, a Gated Recurrent Unit-Residual Multi-Graph Convolutional Network (GR-RMGCN) based on spatiotemporal residuals is used as the highway charging demand prediction model. This model extracts and fuses multi-graph structure features, time series features and external factor features through an encoder to generate a unified latent vector, which is then decoded by a decoder to output the charging demand prediction results for future periods.

[0061] To verify the overall performance of GR-RMGCN used in this embodiment, several benchmark models were selected for horizontal comparison, covering the traditional time series model HA (Historical Average), the traditional machine learning model XGBoost (Extreme Gradient Boosting), the deep learning model SP-LSTM (Spatio-Temporal Long Short-Term Memory), and the advanced graph neural network model MGCN (Multi-Graph Convolutional Network). The experimental evaluation metrics used were mean absolute error, root mean square error, and mean square error.

[0062] Experimental results show that the GR-RMGCN used in this embodiment comprehensively outperforms traditional models. Compared to the traditional time-series model HA, which cannot capture periodicity and drastic changes in demand, and the machine learning model XGBoost, which ignores the spatial relationships between charging stations, the method of this invention outperforms by an average of 10%-20%. For example, compared to the better-performing XGBoost model, the model of this invention reduces the mean absolute error by 10.95%. Compared to the suboptimal model MGCN, which is also based on graph convolutional networks, the method of this invention demonstrates stronger performance. Experimental data show that the model of this invention achieves a performance gain of 6.15% in mean absolute error and 5.52% in root mean square error. Therefore, it can be understood that the GR-RMGCN used in this embodiment, by constructing and fusing multi-dimensional graph structures, can more profoundly capture the periodicity, volatility, spatial dependence, and semantic relationships in charging demand, thereby achieving better prediction accuracy.

[0063] In one implementation, the highway charging demand prediction model includes a decoder and an encoder with three parallel branches. The process of inputting the historical charging order data, the external environment data, and the multi-graph data into the pre-trained highway charging demand prediction model to obtain the highway charging demand prediction result specifically includes the following steps: Step S310: Input the charging demand time-series features, the external environment features, and the multigraph into the encoder of the corresponding branch to obtain the corresponding time feature vector, environmental feature vector, and spatial feature vector; Step S320: Concatenate and fuse the time feature vector, environmental feature vector, and spatial feature vector to obtain the latent vector; Step S330: The latent vector is recovered through the decoder to obtain the shared feature tensor, and the charging demand prediction value is obtained based on the shared feature tensor.

[0064] In this embodiment, the charging demand time-series features extracted based on the time-series prediction characteristics are first input into the prediction model. These features include those based on proximity (demand between two previous times), trend (demand at the same interval as the previous day), and periodicity (demand at the same interval as the previous week), resulting in a time feature vector. Residual multi-graph convolution is then performed based on the multi-graph construction results. This residual multi-graph convolutional network captures the spatial and semantic dependencies at each time step, yielding a spatial feature vector. Weather features are then encoded and concatenated according to the input data structure requirements of the time-series prediction. Specifically, to incorporate external factors such as weather into the model, environmental feature vectors representing multiple dimensions of raw meteorological data, such as temperature, humidity, and wind speed, are obtained in the above steps and undergo specialized encoding processing. This process is implemented through a weather feature encoder, which captures the temporal dynamics of the weather data itself and transforms and compresses it into an information-dense embedding vector, resulting in the environmental feature vector, providing comprehensive input for the decoder. Finally, the time feature vector, environmental feature vector, and spatial feature vector are concatenated and fused to obtain the latent vector.

[0065] Subsequently, the prediction result is output based on the decoding result of the decoder. The decoder first constructs a shared feature tensor based on the aforementioned latent vectors, and then generates the final prediction through further processing.

[0066] In one implementation, the highway charging demand prediction model includes a decoder and an encoder with three parallel branches. The step of inputting the charging demand time-series features, the external environment features, and the multigraph into the encoders of the corresponding branches to obtain the corresponding time feature vector, environmental feature vector, and spatial feature vector specifically includes the following steps: Step S311: Input the multigraph into the first branch of the encoder to obtain a spatial feature vector, wherein the first branch of the encoder is a spatial feature residual multigraph convolution module, and the output of each layer of the spatial feature residual multigraph convolution module is the sum of the output of the previous layer and the feature matrix of the multigraph after graph convolution transformation. Step S312: Input the charging demand time sequence features into the second branch of the encoder to obtain a time feature vector, wherein the second branch of the encoder is a gated loop unit module, and the time feature vector is used to characterize the time dependency and periodicity of the charging demand. Step S313: Input the external environment features into the third branch of the encoder to obtain the environment feature vector, wherein the third branch of the encoder consists of two gating loop units and a fully connected layer.

[0067] In this embodiment, the first branch of the encoder is a spatial feature residual multi-graph convolution module, whose expression is:

[0068] Among them, the current layer output Output from the previous layer After graph convolution transformation The results are added together to obtain, For activation function, For degree matrix, It is an adjacency matrix. This is a trainable weight matrix.

[0069] The multigraph is input into the first branch of the encoder to obtain the spatial feature vector.

[0070] The second branch of the encoder is the Gated Recurrent Unit (GRU) module, which processes the charging demand time-series features constructed in the previous steps to capture the long-term temporal dependence and periodic patterns of demand. Specifically, through the GRU module, the model captures the charging demand time-series features, compresses them into an information-dense embedding vector based on its own temporal dynamics, and then fuses them with other features in the future.

[0071] The charging demand time-series features are input into the second branch of the encoder to obtain the time feature vector.

[0072] The third branch of the encoder is a separate weather feature encoder, consisting of two layers of gated loop units and a fully connected layer, and its expression is:

[0073] in, It refers to the encoded weather characteristics, which condense the key information in the original weather time series data. It is the first-level gated loop unit, which directly processes the raw weather feature input. To capture basic time patterns. It is the second-level gated recurrent unit, which receives the output of the first-level gated recurrent unit and further extracts higher-level time-dependent features. This is the weight matrix of the fully connected layer. For the bias of the fully connected layer, For activation function, Input for weather characteristics. , This represents all the internal parameters of the GRU model.

[0074] The external environment features are input into the third branch of the encoder to obtain the environment feature vector.

[0075] In one implementation, the highway charging demand prediction model includes a decoder and an encoder with three parallel branches. The decoder recovers the latent vector to obtain a shared feature tensor, and based on the shared feature tensor, a predicted charging demand value is obtained. Specifically, the model includes the following steps: Step S331: Through a fully connected layer of the decoder, the latent vector is linearly transformed and reshaped to obtain a two-dimensional shared feature tensor; Step S332: Input the shared feature tensor into the regression prediction residual multi-graph convolutional sub-network, and after multi-layer graph convolution processing, obtain the specific charging demand prediction value of each station at future time.

[0076] In this embodiment, as Figure 3 As shown, the output vector spatial feature vector, temporal feature vector, and weather feature vector of the parallel encoder module are first concatenated to generate a latent vector. The latent vector is a longer vector containing comprehensive spatiotemporal and meteorological information, resulting from the concatenation operation. Subsequently, this latent vector is input into a fully connected layer for final refinement and integration, generating a highly condensed latent vector. This vector represents the encoder's comprehensive understanding of the current state. Specifically, the latent vector expression is:

[0077] in, Weight matrix The accompanying bias term, RMGC represents the residual multi-graph convolution, i.e., the first branch of the encoder. To flatten the output of the RMGC encoder, For the RMGC encoder weight set, This represents each input feature. The output of the second branch of the encoder, This represents all the internal parameters of the GRU model, which is responsible for extracting time-series features from the input. This refers to the encoded weather characteristics. The outputs of the three branches of the encoder are then concatenated.

[0078] Subsequently, the latent vectors are decoded and spatial features are reconstructed. The decoder receives the latent vectors generated by the encoder. This restores the one-dimensional, highly compressed information into multi-dimensional features with spatial structure. This is achieved through a fully connected layer. Perform a linear transformation, then use a reshape operation to restore the transformed one-dimensional vector into a two-dimensional shared feature tensor. The shared feature tensor expression is:

[0079] Among them, the shared feature tensor The dimensions correspond to the spatial structure. The weights are responsible for weighting the input latent vector. Perform a linear transformation. The bias vector corresponds to the weights. The activation function is called Reshape, which is a reshaping operation that takes a one-dimensional vector and reshapes it into a multi-dimensional tensor according to a preset dimension.

[0080] After obtaining the shared feature tensor, the prediction result is generated through the regression prediction residual multi-graph convolutional sub-network in the decoder. Specifically, the generated shared feature tensor with spatial structure is... As input, a residual multi-graph convolutional network specifically designed for regression tasks is given, named the Regression Prediction Residual Multi-Graph Convolutional Sub-Network. This network has independent weights specifically learned for the regression task. and .like Figure 3 As shown, after fine processing using multi-layer graph convolution, the network directly outputs predicted charging demands for each site at future times. Final prediction results. Represented as:

[0081] in, Let be the model's predicted final charging demand at time t on day d. In this expression... For residual multi-graph convolution operations, with For the weights of the shared feature tensor of the input Feature extraction is performed, and residual connections are used to alleviate gradient problems. At the same time, multi-graph convolution is used to fuse multi-dimensional graph information. For quadratic graph convolution, for Output results are The weights are then subjected to another graph convolution to further refine the features and integrate multi-scale and multi-type graph structure information.

[0082] In one implementation, the training process of the highway charging demand prediction model includes: Step S410: Input the training dataset into the highway charging demand prediction model to be trained, and use the combination of mean squared error loss and regularization term as the optimization objective to iteratively train the model until the optimization objective converges. Step S420: Use root mean square error and mean absolute error to verify the performance of the trained model, and obtain the pre-trained highway charging demand prediction model.

[0083] In this embodiment, the objective function and training optimization are first defined. Specifically, a prediction error term is defined, and the first part of the optimization objective is the mean squared error loss, which is calculated by plotting the predicted values ​​of all samples. Compared with the true value The sum of squares of the differences. The goal of this term is to make the model's predictions as close to reality as possible. The second part of the optimization objective is the L2 regularization term. This term limits the model's complexity by penalizing the magnitude of all weights in the model. Its main function is to prevent the model from overfitting on the training data, thereby improving its generalization ability on unknown data. The expression for the optimization objective is:

[0084] in, To optimize the objective, all weights and bias Find the minimum value for all days. and time step Summation, The square of the L2 norm of all weights, For the predicted demand, For real needs, This is the activation function.

[0085] Combining the two optimization objectives above forms the final optimization objective function. The goal of model training is to find a set of parameters that minimizes the total value of this function.

[0086] Next, model training and evaluation were performed. The dataset was divided into training and test sets in chronological order. Adam was used as the optimizer, and hyperparameters such as the learning rate were set. The model was trained iteratively on the training set for multiple rounds, and the loss curve was monitored until convergence. Then, performance evaluation was conducted. The predictive performance of the trained model was verified on the test set, and root mean square error and mean absolute error were selected as the final evaluation metrics.

[0087] In summary, under the technical solution of the above embodiments, by constructing a multi-graph system containing geographic and semantic graphs, and combining it with a spatiotemporal residual multi-graph convolutional network, the system fully mines and integrates multi-dimensional information such as road network topology, demand time series, weather, and price. This accurately captures the complex spatiotemporal dependence of charging demand and the correlation of multi-source data, making the model's predictive ability higher than that of models with a single data source or simple data splicing. This achieves accurate prediction of charging demand fluctuations and provides support for the layout and operation of charging facilities, power planning, and traffic management, alleviating charging congestion and improving highway service and operational efficiency.

[0088] like Figure 4As shown in the figure, this embodiment of the invention provides a highway charging demand prediction system based on a spatiotemporal residual multigraph convolutional network. The system includes: an input data acquisition module 10, a multigraph construction module 20, and a model prediction module 30.

[0089] Specifically, the input data acquisition module 10 is used to acquire equipment data, geographical location data, and historical charging order data of charging stations on highways, and to acquire external environment data corresponding to the geographical location and order time of the charging stations; the multigraph construction module 20 is used to construct a multigraph based on the equipment data and the geographical location data, with the charging stations as nodes, wherein the multigraph includes a geographic graph and a semantic graph, the geographic graph is used to represent the geographical location relationship between the charging stations, and the semantic graph is used to represent the non-geographical intrinsic attribute relationship between the charging stations; the model prediction module 30 is used to input the historical charging order data, the external environment data, and the multigraph into a pre-trained highway charging demand prediction model to obtain highway charging demand prediction results, wherein the highway charging demand prediction model is a spatiotemporal residual multigraph convolutional network.

[0090] Based on the above embodiments, the present invention also provides a terminal device, the principle block diagram of which can be as follows: Figure 5 As shown, the terminal device includes a processor, memory, network interface, display screen, and temperature sensor connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The network interface is used to communicate with external terminals via a network connection. When executed by the processor, the computer program implements a highway charging demand prediction method based on a spatiotemporal residual multi-graph convolutional network. The display screen can be an LCD screen or an e-ink screen. The temperature sensor is pre-installed inside the terminal device to detect the operating temperature of the internal components.

[0091] Those skilled in the art will understand that Figure 5 The schematic diagram shown is only a partial structural diagram related to the present invention and does not constitute a limitation on the terminal device to which the present invention is applied. The specific terminal device may include more or fewer components than shown in the figure, or combine certain components, or have different component arrangements.

[0092] In one embodiment, a terminal device is provided, including a memory and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by one or more processors. The one or more programs include instructions for performing the following operations: Acquire equipment data, geographical location data, and historical charging order data of charging stations along highways, and acquire external environmental data corresponding to the geographical location and order time of the charging stations; Based on the device data and the geographic location data, a multi-graph is constructed with the charging station as the node. The multi-graph includes a geographic graph and a semantic graph. The geographic graph is used to represent the geographic location relationship between the charging stations, and the semantic graph is used to represent the non-geographical intrinsic attribute relationship between the charging stations. The historical charging order data, the external environment data, and the multi-graph input are used to pre-train a highway charging demand prediction model to obtain highway charging demand prediction results. The highway charging demand prediction model is a spatiotemporal residual multi-graph convolutional network.

[0093] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided by this invention can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0094] In summary, this invention discloses a method, system, terminal device, and medium for predicting highway charging demand based on a spatiotemporal residual multi-graph convolutional network, relating to the field of intelligent transportation technology. The method first acquires equipment data, geographical location data, and historical charging order data of highway charging stations, and acquires external environment data corresponding to the geographical location and order time of the charging stations. Then, based on the equipment data and the geographical location data, a multi-graph is constructed with the charging stations as nodes. The multi-graph includes a geographic graph and a semantic graph. The geographic graph represents the geographical location relationships between the charging stations, and the semantic graph represents the non-geographical intrinsic attribute relationships between the charging stations. Finally, the historical charging order data, the external environment data, and the multi-graph are input into a pre-trained highway charging demand prediction model to obtain the highway charging demand prediction result. The highway charging demand prediction model is a spatiotemporal residual multi-graph convolutional network. This invention constructs a multi-graph system containing geographic and semantic graphs, combined with a spatiotemporal residual multi-graph convolutional network. It fully mines and integrates multi-dimensional information such as road network topology, demand time series, weather, and price, accurately capturing the complex spatiotemporal dependence and multi-source data correlation of charging demand. This makes the model's predictive ability higher than that of models with a single data source or simple data splicing, achieving accurate prediction of charging demand fluctuations. Furthermore, it provides support for the layout and operation of charging facilities, power planning, and traffic management, alleviating charging congestion and improving highway service and operational efficiency.

[0095] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0096] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for predicting highway charging demand based on spatiotemporal residual multi-graph convolutional networks, characterized in that, The method includes: Acquire equipment data, geographical location data, and historical charging order data of charging stations along highways, and acquire external environmental data corresponding to the geographical location and order time of the charging stations; Based on the device data and the geographic location data, a multi-graph is constructed with the charging station as the node. The multi-graph includes a geographic graph and a semantic graph. The geographic graph is used to represent the geographic location relationship between the charging stations, and the semantic graph is used to represent the non-geographical intrinsic attribute relationship between the charging stations. The historical charging order data, the external environment data, and the multi-graph input are used to pre-train a highway charging demand prediction model to obtain highway charging demand prediction results. The highway charging demand prediction model is a spatiotemporal residual multi-graph convolutional network.

2. The method for predicting highway charging demand based on spatiotemporal residual multi-graph convolutional networks according to claim 1, characterized in that, The construction of a multi-graph based on the device data and the geographic location data, with the charging station as a node, includes: A geographic map is constructed based on the geographic location data. The geographic map includes at least one of a neighborhood correlation map, a distance correlation map, and a travel direction correlation map. The neighborhood correlation map is used to represent the adjacency relationship between the charging stations. The distance correlation map is used to represent the distance relationship between the charging stations. The travel direction correlation map is used to indicate whether the charging stations are charging stations in the same location but in different travel directions. A semantic graph is constructed based on the device data. The semantic graph includes at least one of a device correlation graph and a price correlation graph. The device correlation graph is used to represent the similarity of the device configuration of the charging station, and the price correlation graph is used to represent the high and low relationship of the charging price of the charging station.

3. The method for predicting highway charging demand based on spatiotemporal residual multi-graph convolutional networks according to claim 1, characterized in that, Before obtaining the highway charging demand prediction result by inputting the historical charging order data, the external environment data, and the multi-graph input into the pre-trained highway charging demand prediction model, the following steps are included: The external environment data is digitally encoded and normalized to obtain external environment characteristics, wherein the external environment data includes at least meteorological data; The historical charging order data is cleaned and aggregated and feature-constructed according to a preset time unit to obtain charging demand time series features, wherein the charging demand time series features are used to characterize the time-varying pattern of charging demand at each charging station.

4. The method for predicting highway charging demand based on spatiotemporal residual multi-graph convolutional networks according to claim 3, characterized in that, The highway charging demand prediction model includes a decoder and an encoder with three parallel branches. The model inputs the historical charging order data, the external environment data, and the multi-graph data into a pre-trained highway charging demand prediction model to obtain highway charging demand prediction results, including: The charging demand time-series features, the external environment features, and the multigraph are respectively input into the encoder of the corresponding branch to obtain the corresponding time feature vector, environmental feature vector, and spatial feature vector; The temporal feature vector, environmental feature vector, and spatial feature vector are concatenated and fused to obtain the latent vector. The decoder recovers the latent vectors to obtain a shared feature tensor, and based on the shared feature tensor, the predicted charging demand is obtained.

5. The method for predicting highway charging demand based on spatiotemporal residual multi-graph convolutional networks according to claim 4, characterized in that, The step of inputting the charging demand time-series features, the external environment features, and the multigraph into the encoder of the corresponding branch to obtain the corresponding time feature vector, environmental feature vector, and spatial feature vector includes: The multigraph is input into the first branch of the encoder to obtain a spatial feature vector. The first branch of the encoder is a spatial feature residual multigraph convolution module. The output of each layer of the spatial feature residual multigraph convolution module is the sum of the output of the previous layer and the feature matrix of the multigraph after graph convolution transformation. The charging demand time-series features are input into the second branch of the encoder to obtain a time feature vector, wherein the second branch of the encoder is a gated loop unit module, and the time feature vector is used to characterize the time dependency and periodicity of the charging demand. The external environment features are input into the third branch of the encoder to obtain the environment feature vector. The third branch of the encoder consists of two gating loop units and a fully connected layer.

6. The method for predicting highway charging demand based on spatiotemporal residual multi-graph convolutional networks according to claim 4, characterized in that, The process of recovering the latent vectors through the decoder to obtain a shared feature tensor, and obtaining the predicted charging demand value based on the shared feature tensor, includes: The latent vector is linearly transformed and reshaped through a fully connected layer of the decoder to obtain a two-dimensional shared feature tensor. The shared feature tensor is input into the regression prediction residual multi-graph convolutional sub-network, and after multi-layer graph convolution processing, the specific charging demand prediction value of each station at future time is obtained.

7. The method for predicting highway charging demand based on spatiotemporal residual multi-graph convolutional networks according to claim 1, characterized in that, The training process of the highway charging demand prediction model includes: The training dataset is input into the highway charging demand prediction model to be trained, and the combination of mean squared error loss and regularization term is used as the optimization objective. The model is iteratively trained until the optimization objective converges. The performance of the trained model was verified using root mean square error and mean absolute error, resulting in the pre-trained highway charging demand prediction model.

8. A highway charging demand prediction system based on spatiotemporal residual multi-graph convolutional networks, characterized in that, The system includes: The input data acquisition module is used to acquire equipment data, geographical location data, and historical charging order data of charging stations on highways, and to acquire external environmental data corresponding to the geographical location and order time of the charging station. A multi-graph construction module is used to construct a multi-graph based on the device data and the geographic location data, with the charging station as the node. The multi-graph includes a geographic graph and a semantic graph. The geographic graph is used to represent the geographic location relationship between the charging stations, and the semantic graph is used to represent the non-geographical intrinsic attribute relationship between the charging stations. The model prediction module is used to input the historical charging order data, the external environment data and the multi-graph into a pre-trained highway charging demand prediction model to obtain highway charging demand prediction results. The highway charging demand prediction model is a spatiotemporal residual multi-graph convolutional network.

9. A terminal device, characterized in that, The terminal device includes a memory, a processor, and a highway charging demand prediction program based on a spatiotemporal residual multi-graph convolutional network stored in the memory and executable on the processor. When the processor executes the highway charging demand prediction program based on a spatiotemporal residual multi-graph convolutional network, it implements the steps of the highway charging demand prediction method based on a spatiotemporal residual multi-graph convolutional network as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a highway charging demand prediction program based on a spatiotemporal residual multi-graph convolutional network. When the highway charging demand prediction program based on a spatiotemporal residual multi-graph convolutional network is executed by a processor, it implements the steps of the highway charging demand prediction method based on a spatiotemporal residual multi-graph convolutional network as described in any one of claims 1-7.

Images