Electric vehicle charging load prediction method and system based on traffic balance theory and deep neural network, medium, equipment and product

Abstract

The invention discloses an electric vehicle charging load prediction method and system based on a traffic equilibrium theory and a deep neural network, a medium, equipment and a product in the technical field of traffic power system coupling analysis. The method comprises the following steps: constructing a traffic user balance model and an electric vehicle feasible path generation model, and calculating the total charging load of each charging station node in multiple scenes to obtain a charging load sample set; and carrying out feature screening on the charging load sample set by using correlation analysis and an SHAP method based on a Shapley value, training a multilayer full-connection deep neural network model by using the screened charging load sample set, and carrying out charging load prediction on each charging station node after training. According to the method, the coupling rule between the traffic behavior and the charging load can be accurately revealed, the interpretability and generalization ability of the model are considered while the prediction precision is improved, and efficient and intelligent technical support is provided for electric vehicle charging facility planning and power distribution system operation optimization.

Description

Technical Field

[0001] This invention relates to the field of electric vehicle charging load prediction and traffic power system coupling analysis technology, and in particular to an electric vehicle charging load prediction method, system, medium, equipment and product based on traffic equilibrium theory and deep neural network. Background Technology

[0002] With the widespread application of electric vehicles in urban transportation systems, the spatiotemporal distribution characteristics of charging load have become a crucial factor affecting the operational safety and planning optimization of power distribution networks. Existing research largely relies on historical charging data or single traffic scenarios for predictive analysis, lacking systematic modeling of electric vehicle travel behavior and energy constraints, making it difficult to accurately reflect the mechanism by which traffic flow changes shape charging load. Furthermore, traditional data-driven prediction methods show a significant decrease in accuracy when high-quality samples are lacking or when applied across regions, while process-driven models are computationally complex and lack real-time performance, making it difficult to balance prediction accuracy and application efficiency. Due to the lack of a unified framework capable of simultaneously characterizing traffic travel mechanisms and load variation patterns, existing technologies struggle to achieve accurate and interpretable predictions of electric vehicle charging load. Summary of the Invention

[0003] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a method, system, medium, equipment, and product for predicting electric vehicle charging load based on traffic equilibrium theory and deep neural networks. By introducing a traffic user equilibrium modeling mechanism that considers energy constraints, and combining multi-scenario path allocation simulation, feature correlation analysis, and deep learning prediction models, a quantitative coupling relationship between traffic behavior and electric vehicle charging load is established. This enables high-precision and interpretable prediction of charging station node load, reveals the impact of traffic flow changes on charging demand distribution, and provides scientific data support and decision-making basis for the planning and layout of electric vehicle charging facilities and the optimization of power distribution system operation.

[0004] To solve the above-mentioned technical problems, the present invention is implemented using the following technical solution:

[0005] In a first aspect, the present invention provides a method for predicting electric vehicle charging load based on traffic equilibrium theory and deep neural networks, comprising:

[0006] Based on the traffic network structure, a traffic user equilibrium model and an electric vehicle feasible path generation model are constructed.

[0007] Based on the traffic user equilibrium model and the electric vehicle feasible path generation model, the total charging load of each charging station node under multiple scenarios is calculated to obtain a charging load sample set.

[0008] The charging load sample set was subjected to feature filtering using correlation analysis and the SHAP method based on Shapley values ​​to obtain a filtered charging load sample set.

[0009] Using the filtered charging load sample set, the multilayer fully connected deep neural network model is trained end-to-end, and the trained multilayer fully connected deep neural network model is used to predict the charging load of each charging station node to obtain the prediction results.

[0010] Optionally, constructing a traffic user equilibrium model based on the traffic network structure includes:

[0011] Establish a transportation network based on the transportation network structure. ;in, Represents a set of road segments. Represents the set of nodes in a transportation network;

[0012] Based on the node set of the transportation network Get the set of OD pairs ; where each OD pair Including the origin node and the destination node, ;

[0013] According to each OD pair Path traffic With broad travel costs Define the traffic user equilibrium conditions and construct a traffic user equilibrium model; where, Indicates a path. Indicates OD pair path Traffic flow Indicates OD pair path The generalized cost of travel.

[0014] Optionally, the traffic user equilibrium condition is as follows:

[0015] ,

[0016] in, Indicates OD pair The minimum travel cost;

[0017] The objective function of the traffic user equilibrium model is as follows:

[0018] ,

[0019] in, Represents the path flow vector. Represents the traffic flow vector of a road segment. This represents the cost coefficient per unit time. Indicates road segment, Indicates road segment Traffic flow Indicates road segment The passage time, Represents the set of charging station nodes. , Indicates charging station node Traffic flow Indicates that electric vehicles are at charging station nodes Total charging time Indicates OD pair The set of paths Indicates the selection of a path Electric vehicles at charging station nodes The amount of charge, Indicates charging station node Electricity price;

[0020] The traffic user equilibrium model is subject to the following constraints:

[0021]

[0022]

[0023]

[0024]

[0025]

[0026]

[0027] in, Indicates OD pair The actual travel needs Indicates road segment Free passage time, Indicates road segment Traffic capacity, Indicates charging station node Average waiting time in queues Indicates charging station node capacity, Indicates charging station node The charging power, For binary variables, Indicates the selection of a path Electric vehicles at charging station nodes Charge it. Indicates no charging; For binary variables, Representing a path Passing section , Representing a path No passing section .

[0028] Optionally, the objective function of the electric vehicle feasible path generation model is as follows:

[0029] ,

[0030] The feasible path generation model for electric vehicles is subject to the following constraints:

[0031] ,

[0032] ,

[0033] ,

[0034] ,

[0035] ,

[0036] ,

[0037] ,

[0038] ,

[0039] in, This represents the adjacency matrix of nodes and road segments in a transportation network. Let represent the node access decision vector for path p. Indicates OD pair The node demand vector, Indicates the starting node of road segment a. The termination node of road segment a, Indicates the point reached. The state of charge of the battery at that location. Indicates the point reached. The state of charge of the battery at that location. Indicates the section of road Electricity consumed while driving. Indicates at node The amount of charge; For the node The amount of charge; As an auxiliary variable, Representing a path Passing section , Representing a path No passing section , As a preset positive number, Range anxiety parameter for electric vehicles represents the minimum battery capacity threshold that users can tolerate. Indicates the maximum energy storage capacity of the battery. Indicates the initial charge stored in the battery. Represents a node Charging capacity limitations at the location, Represents a node There are no charging stations. Represents a node There are charging stations available; For path The charge of the starting node.

[0040] Optionally, the step of calculating the total charging load of each charging station node under multiple scenarios based on the traffic user equilibrium model and the electric vehicle feasible path generation model to obtain a charging load sample set includes:

[0041] Initialize traffic flow for all road segments. Traffic flow at charging station nodes The available paths for each OD pair are solved using the electric vehicle feasible path generation model to obtain an initial path set. ;

[0042] Iteratively execute the following steps until no new path is available to update any OD pair:

[0043] According to the set of paths Update traffic flow for all road segments using a traffic user equilibrium model. Traffic flow at charging station nodes And calculate the travel cost for each route. ;

[0044] Based on the updated traffic flow of all road sections Traffic flow at charging station nodes The available new paths for each OD pair are updated using the electric vehicle feasible path generation model, and the travel cost of each available new path is calculated. ;

[0045] For each OD pair, the travel cost if a new route is available... Lower than the travel cost of each route Then add the available new path to the path set. ;

[0046] After the iteration terminates, the traffic flow and charging amount at charging station nodes for each path of all OD pairs are counted, the total charging load of each charging station node is calculated, and a charging load sample set is obtained.

[0047] The total charging load of each charging station node is obtained by the following formula:

[0048]

[0049] in, Indicates charging station node Total charging load.

[0050] Optionally, the step of using correlation analysis and the SHAP method based on Shapley values ​​to perform feature filtering on the charging load sample set to obtain a filtered charging load sample set includes:

[0051] Based on the charging load sample set, construct a sample feature matrix. With the output target vector ;in, , This represents the total number of charging load samples. Representing feature dimension, Indicates the first The first charging load sample One characteristic, , Indicates transpose. Indicates the first Charging load labels for each charging load sample;

[0052] Using Pearson correlation analysis on the sample feature matrix With the output target vector The degree of linear correlation between them is measured to obtain the correlation coefficient;

[0053] Remove sample feature matrix The feature matrix of candidate samples is obtained by considering features whose absolute correlation coefficient is less than a set threshold. ;

[0054] The candidate sample feature matrix was processed using the SHAP method based on Shapley values. Perform feature importance assessment to obtain the importance value of each feature;

[0055] Based on the absolute value of the importance values ​​of each feature, the candidate sample feature matrix is... Sort the features in descending order and retain the top ones. The key sample feature matrix is ​​obtained from these features. ;in, This is the default value;

[0056] Joint key sample feature matrix and output target vector The filtered charging load sample set is obtained.

[0057] Optionally, the correlation coefficient is obtained by the following formula:

[0058] ,

[0059] in, Indicates the first The correlation coefficient of each feature , Represents the sample feature matrix The Middle The average of the features, Indicates the output target vector The average value;

[0060] The importance values ​​of each feature are obtained using the following formula:

[0061] ,

[0062] in, Represents the feature matrix of candidate samples The Middle The importance value of each feature, Indicates that it does not include the first A subset of features of each feature Represents the feature matrix of candidate samples The feature set in Represents the feature matrix of candidate samples The total number of features in Indicates the feature matrix of candidate samples Selecting a feature subset The resulting sample feature vector This indicates that a multi-layer fully connected deep neural network model can operate based solely on a subset of features. Under the condition of output target vector The prediction results In the sample feature vector Based on the introduction of the first The expanded sample feature vector obtained after considering each feature Indicating in the feature subset Based on the introduction of the first After considering each feature, the multi-layer fully connected deep neural network model outputs the target vector. The prediction results.

[0063] Optionally, the multi-layer fully connected deep neural network model includes an input layer, Hidden layers and output layers; among which, This is the default value;

[0064] The expression for the multilayer fully connected deep neural network model is:

[0065] ,

[0066] in, Indicates the prediction result. Indicates the number of charging stations. Represent real numbers, This represents a nonlinear mapping function for a multilayer fully connected deep neural network. The parameters represent those of a multi-layer fully connected deep neural network. This represents the sample feature vector of a multilayer fully connected deep neural network model. The dimension of the sample feature vector;

[0067] The expression for the hidden layer is:

[0068] ,

[0069] in, Indicates the first The output of the hidden layer, Indicates the first The weight matrix of the hidden layer, Indicates the first The bias vector of the hidden layer. This represents the activation function.

[0070] Optionally, the step of using the filtered charging load sample set to perform end-to-end training on the multilayer fully connected deep neural network model includes:

[0071] The filtered charging load sample set is preprocessed by normalization and divided into training set, validation set and test set according to proportion;

[0072] For each training batch, the multilayer fully connected deep neural network model is trained using the training set, the loss of the training set is calculated using the mean squared error loss function, and the parameters of the multilayer fully connected deep neural network model are updated using the Adam optimization algorithm.

[0073] After each training batch, the multilayer fully connected deep neural network model is validated using a validation set. The loss of the validation set is calculated using the mean squared error loss function. When the loss of the validation set no longer decreases or an overfitting trend appears, an early stopping strategy is adopted to terminate the training.

[0074] After all training is completed, the prediction performance of the multilayer fully connected deep neural network model is evaluated using the test set. If the prediction performance reaches the preset accuracy threshold, the trained multilayer fully connected deep neural network model is output.

[0075] The expression for the mean squared error loss function is as follows:

[0076]

[0077] in, Indicates loss, The total number of training / validation samples, Indicates the first Prediction results for each training / validation sample. Indicates the first The actual charging load of each training / validation sample;

[0078] The prediction performance includes at least one of mean absolute error, root mean square error, and coefficient of determination.

[0079] The mean absolute error is obtained by the following formula:

[0080] ,

[0081] in, This represents the total number of test samples. Indicates the first The actual charging load of each test sample Indicates the first Prediction results for each test sample Indicates the mean absolute error;

[0082] The root mean square error is obtained by the following formula:

[0083] ,

[0084] in, Indicates the root mean square error;

[0085] The coefficient of determination is obtained by the following formula:

[0086]

[0087] in, The coefficient of determination is represented by the coefficient of determination. This represents the average of the predicted results for all test samples.

[0088] Secondly, the present invention provides an electric vehicle charging load prediction system based on traffic equilibrium theory and deep neural networks, comprising:

[0089] The model building module is used to: build a traffic user equilibrium model and an electric vehicle feasible path generation model based on the traffic network structure;

[0090] The sample generation module is used to: calculate the total charging load of each charging station node under multiple scenarios based on the traffic user equilibrium model and the electric vehicle feasible path generation model, and obtain a charging load sample set.

[0091] The feature filtering module is used to: perform feature filtering on the charging load sample set using correlation analysis and the SHAP method based on Shapley values, to obtain a filtered charging load sample set.

[0092] The charging load prediction module is used to: train a multilayer fully connected deep neural network model end-to-end using the filtered charging load sample set, and use the trained multilayer fully connected deep neural network model to predict the charging load of each charging station node, thereby obtaining the prediction results.

[0093] Thirdly, the present invention provides a computer-readable storage medium having computer instructions stored thereon, which, when executed by a processor, implement the steps of the electric vehicle charging load prediction method based on traffic equilibrium theory and deep neural networks as described in any of the first aspects.

[0094] Fourthly, the present invention provides a computer device, comprising:

[0095] Memory, used to store computer instructions;

[0096] A processor for executing the computer instructions to implement the steps of the electric vehicle charging load prediction method based on traffic equilibrium theory and deep neural networks as described in any of the first aspects.

[0097] Fifthly, the present invention provides a computer program product, including computer instructions, characterized in that, when the computer instructions are executed by a processor, they implement the steps of the electric vehicle charging load prediction method based on traffic equilibrium theory and deep neural networks as described in any of the first aspects.

[0098] Compared with existing technologies, the beneficial effects achieved by this invention are as follows:

[0099] 1. By introducing a traffic user equilibrium model that considers energy constraints, and integrating road capacity, travel demand and charging behavior factors, an integrated modeling mechanism for traffic-load allocation of electric vehicles was established. Compared with traditional methods that rely on single historical data, it can more realistically reflect the characteristics of travel path selection and charging demand distribution under different traffic scenarios.

[0100] 2. A key feature screening mechanism based on correlation analysis and SHAP method is proposed, which can quantitatively identify the main factors affecting the change of charging load and improve the interpretability and robustness of the model;

[0101] 3. By inputting the selected key features into a multi-layer fully connected deep neural network model, end-to-end prediction of charging station node load was achieved. The model exhibits good convergence, strong generalization ability, and high prediction accuracy. It integrates traffic behavior mechanism modeling with the nonlinear mapping capabilities of deep learning, effectively revealing the coupling law between traffic flow changes and electric vehicle charging load. It achieves high-precision and interpretable charging load prediction results, providing a scientific basis and technical support for the planning of electric vehicle charging infrastructure and the optimization of power distribution system operation. It has good versatility and promotion value. Attached Figure Description

[0102] Figure 1 A flowchart of an electric vehicle charging load prediction method based on traffic equilibrium theory and deep neural networks provided in an embodiment of the present invention;

[0103] Figure 2 This is a schematic diagram of a traffic network structure provided according to an embodiment of the present invention;

[0104] Figure 3 A graph showing the correlation analysis results provided according to an embodiment of the present invention;

[0105] Figure 4 This is a graph showing the ranking results of feature importance assessment according to an embodiment of the present invention;

[0106] Figure 5 The diagram shows the loss curves for the training and validation sets provided according to an embodiment of the present invention.

[0107] Figure 6 This is a comparison chart of the predicted and actual charging load values ​​provided in an embodiment of the present invention. Detailed Implementation

[0108] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments and specific features in the embodiments are detailed descriptions of the technical solution of the present application, rather than limitations thereof. In the absence of conflict, the embodiments and technical features in the embodiments can be combined with each other.

[0109] It should be noted that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0110] Example 1

[0111] This invention discloses a method for predicting electric vehicle charging load based on traffic equilibrium theory and deep neural networks, with reference to... Figure 1 As shown, the specific steps include the following:

[0112] S1. Based on the traffic network structure, construct a traffic user equilibrium model and an electric vehicle feasible path generation model.

[0113] S2, based on the traffic user equilibrium model and the electric vehicle feasible path generation model, calculate the total charging load of each charging station node under multiple scenarios to obtain a charging load sample set;

[0114] S3, use correlation analysis and the SHAP method based on Shapley values ​​to perform feature filtering on the charging load sample set to obtain the filtered charging load sample set.

[0115] S4. Using the filtered charging load sample set, the multilayer fully connected deep neural network model is trained end-to-end, and the trained multilayer fully connected deep neural network model is used to predict the charging load of each charging station node to obtain the prediction result.

[0116] Specifically, in step S1, this embodiment constructs a traffic user equilibrium model that considers energy constraints, inputting parameters such as road capacity, length, free-flow time, and electric vehicle travel demand, to simulate the path selection and traffic flow distribution of electric vehicles under different travel scenarios:

[0117] Establish a transportation network based on the transportation network structure. ;in, Represents a set of road segments. This represents the set of nodes in the transportation network; in this embodiment, a transportation network consisting of 20 nodes and 8 charging stations is selected as the research object, and its network structure is as follows: Figure 2 As shown in Table 1, to reflect different traffic operation conditions, roads are divided into four categories, each with different capacities, lengths, and free-flow travel times.

[0118] Based on the node set of the transportation network Get the set of OD pairs ; where each OD pair Including the origin node and the destination node, This embodiment sets 10 basic OD travel demand pairs, as shown in Table 2, to reflect typical urban travel demand patterns.

[0119] According to each OD pair Path traffic With broad travel costs Define the traffic user equilibrium condition:

[0120] ,

[0121] in, Indicates a path. Indicates OD pair path Traffic flow Indicates OD pair path The generalized cost of travel Indicates OD pair The minimum travel cost;

[0122] The objective function for establishing a traffic equilibrium model that considers energy constraints is as follows:

[0123] ,

[0124] in, Represents the path flow vector. Represents the traffic flow vector of a road segment. This represents the cost coefficient per unit time. Indicates road segment, Indicates road segment Traffic flow Indicates road segment The passage time, Represents the set of charging station nodes. , Indicates charging station node Traffic flow Indicates that electric vehicles are at charging station nodes Total charging time Indicates OD pair The set of paths Indicates the selection of a path Electric vehicles at charging station nodes The amount of charge, Indicates charging station node Electricity price;

[0125] The traffic user equilibrium model is subject to the following constraints:

[0126]

[0127]

[0128]

[0129]

[0130]

[0131]

[0132] in, Indicates OD pair The actual travel needs Indicates road segment Free passage time, Indicates road segment Traffic capacity, Indicates charging station node Average waiting time in queues Indicates charging station node capacity, Indicates charging station node The charging power, For binary variables, Indicates the selection of a path Electric vehicles at charging station nodes Charge it. Indicates no charging; For binary variables, Representing a path Passing section , Representing a path No passing section .

[0133] By solving the traffic user equilibrium model described above, the equilibrium flow distribution results of each road segment and charging station node can be obtained, providing basic input data for the subsequent generation of charging load samples.

[0134] Table 1 Road Parameter Settings

[0135]

[0136] Table 2 Basic OD Travel Demand

[0137]

[0138] Based on the traffic network and charging station node set established in step S1, a feasible path generation model for electric vehicles is constructed to determine the set of paths that meet energy constraints and have the optimal travel cost. The objective function of the electric vehicle feasible path generation model is as follows:

[0139] ,

[0140] The feasible path generation model for electric vehicles is subject to the following constraints:

[0141] ,

[0142] ,

[0143] ,

[0144] ,

[0145] ,

[0146] ,

[0147] ,

[0148] ,

[0149] in, This represents the adjacency matrix of nodes and road segments in a transportation network. Let represent the node access decision vector for path p. Indicates OD pair The node demand vector, whose elements represent the node access demand relationship of electric vehicles from the starting node to the destination node under path constraints. This indicates that road segment a starts from the node. With the termination node constitute, Indicates the point reached. The state of charge of the battery at that location. Indicates the point reached. The state of charge of the battery at that location. Indicates the section of road Electricity consumed while driving. Indicates at node The amount of charge; For the node The amount of charge; As an auxiliary variable, Representing a path Passing section , Representing a path No passing section , This represents a sufficiently large positive number used to linearize logical conditions within constraints. Range anxiety parameter for electric vehicles represents the minimum battery capacity threshold that users can tolerate. Indicates the maximum energy storage capacity of the battery. Indicates the initial charge stored in the battery. Represents a node Charging capacity limitations at the location, Represents a node There are no charging stations. Represents a node There are charging stations available; For path The charge of the starting node.

[0150] In step S2, based on the traffic user equilibrium model and the electric vehicle feasible path generation model, the total charging load of each charging station node under multiple scenarios is calculated to obtain a charging load sample set, including:

[0151] (1) Initialize the traffic flow of all road segments. Traffic flow at charging station nodes The available paths for each OD pair are solved using the electric vehicle feasible path generation model to obtain an initial path set. ;

[0152] (2) Based on the set of paths Update traffic flow for all road segments using a traffic user equilibrium model. Traffic flow at charging station nodes And calculate the travel cost for each route. ;

[0153] (3) Based on the updated traffic flow of all road sections Traffic flow at charging station nodes The available new paths for each OD pair are updated using the electric vehicle feasible path generation model, and the travel cost of each available new path is calculated. ;

[0154] (4) For each OD pair, if the travel cost of the new route is available. Lower than the travel cost of each route Then add the available new path to the path set. , and repeat (3); if no path update is available for any OD pair, the iteration terminates;

[0155] (5) After the iteration terminates, count the traffic flow of each path for all OD pairs and the charging amount at the charging station nodes, calculate the total charging load of each charging station node, and obtain the charging load sample set.

[0156] The total charging load of each charging station node is obtained by the following formula:

[0157]

[0158] in, Indicates charging station node Total charging load.

[0159] Through the above calculations, charging load samples of each charging station node under multiple scenarios are obtained, providing a training dataset for subsequent feature selection and multi-layer fully connected deep neural network models.

[0160] In step S3, feature screening is performed on the charging load samples. Correlation analysis and the SHAP method based on Shapley values ​​are used to identify the key features that have the most significant impact on the charging load. The specific implementation steps are as follows:

[0161] S3.1 Based on the multi-scenario charging load samples generated in step S2, this embodiment generates 2000 diverse traffic operation scenarios by applying 5% to 20% random perturbation to road capacity, length and free-flow travel time, and scaling the basic OD travel demand within the range of 0.5 to 1.5 times; each perturbation scenario is solved by a traffic equilibrium model to obtain the total charging load of each charging station;

[0162] S3.2. Based on the solution results, a dataset for feature selection is constructed. The sample feature vector is composed of two parts: first, a 10-dimensional OD travel demand, corresponding to the demand of 10 OD pairs under different perturbations in the basic scenario; second, road parameter features, extracting three types of parameters—capacity, length, and free-flow travel time—from all 56 arc segments in the network, and expanding them in the order of capacity-length-free-flow time. The two types of features are concatenated sequentially to form a complete sample feature vector. The sample feature matrix is ​​obtained. With the output target vector :

[0163]

[0164] Among them, the sample feature vector The total dimensions are 178. The 10th parameter representing OD travel demand. This indicates the capacity of the 56th arc segment of the network. This represents the length of the 56th arc segment in the network. This represents the free-flow travel time of the 56th arc segment of the network; the output target y is composed of the total charging load of each charging station under different scenarios, forming a multi-dimensional output corresponding to the number of station nodes; , This represents the total number of charging load samples. Representing feature dimension, Indicates the first The first charging load sample One characteristic, , Indicates transpose. Indicates the first Charging load labels for each charging load sample;

[0165] S3.3, Using Pearson correlation analysis on the sample feature matrix With the output target vector The degree of linear correlation between them is measured to obtain the correlation coefficient:

[0166] ,

[0167] in, Indicates the first The correlation coefficient of each feature The symbol indicates a positive or negative correlation. The larger the absolute value, the stronger the correlation, and the greater the impact of the feature on the target; the closer to 0, the weaker the correlation. Represents the sample feature matrix The Middle The average of the features, Indicates the output target vector The average value;

[0168] S3.4, Removing Sample Feature Matrix The feature matrix of candidate samples is obtained by considering features whose absolute correlation coefficient is less than a set threshold. The ranking results of the Pearson correlation coefficient are as follows: Figure 3 As shown;

[0169] S3.5. The candidate sample feature matrix is ​​processed using the SHAP method based on Shapley values. Perform feature importance assessment to obtain the importance value of each feature:

[0170] ,

[0171] in, Represents the feature matrix of candidate samples The Middle The importance value of each feature, Indicates that it does not include the first A subset of features of each feature Represents the feature matrix of candidate samples The feature set in Represents the feature matrix of candidate samples The total number of features in Indicates the feature matrix of candidate samples Selecting a feature subset The resulting sample feature vector This indicates that a multi-layer fully connected deep neural network model can operate based solely on a subset of features. Under the condition of output target vector The prediction results In the sample feature vector Based on the introduction of the first The expanded sample feature vector obtained after considering each feature Indicating in the feature subset Based on the introduction of the first After considering each feature, the multi-layer fully connected deep neural network model outputs the target vector. The prediction results;

[0172] S3.6. Based on the absolute values ​​of the importance values ​​of each feature, analyze the feature matrix of the candidate samples. Sort the features in descending order and retain the top ones. The key sample feature matrix is ​​obtained from these features. ,in, The preset values ​​are used; the ranking results of the importance assessment of each feature are as follows: Figure 4 As shown;

[0173] S3.7, Combine key sample feature matrices and output target vector The filtered charging load sample set is obtained.

[0174] Based on the above analysis, both results indicate that OD (Original Demand) travel demand dominates the explanation of electric vehicle charging load, while road parameters have a weaker role and are prone to introducing redundancy. Therefore, this embodiment ultimately retains only OD travel demand features as input to the neural network model. This significantly reduces feature dimensionality while preserving the main influencing factors, improving the model's training efficiency and interpretability.

[0175] In step S4, the input feature is set to a 10-dimensional OD (Original Demand) travel demand vector. The output is an 8-dimensional vector consisting of the charging loads of the 8 charging station nodes. , Indicates the first The charging load of each charging station node can be represented by the model as follows:

[0176] ,

[0177] in, Indicates the prediction result. Indicates the number of charging stations. Represent real numbers, This represents a nonlinear mapping function for a multilayer fully connected deep neural network. The parameters represent those of a multi-layer fully connected deep neural network. This represents the sample feature vector of a multilayer fully connected deep neural network model. The dimension of the sample feature vector;

[0178] A multi-layer fully connected deep neural network model consists of an input layer, two hidden layers, and an output layer; each hidden layer contains 128 neurons, the activation function is ReLU, and the expression for the hidden layer is:

[0179] ,

[0180] in, Indicates the first The output of the hidden layer, Indicates the first The weight matrix of the hidden layer, Indicates the first The bias vector of the hidden layer. This represents the activation function.

[0181] Using the filtered charging load sample set, end-to-end training of a multilayer fully connected deep neural network model is performed, including:

[0182] The filtered charging load sample set is preprocessed by normalization and divided into training set, validation set and test set according to the ratio; the preferred ratio is 6:2:2.

[0183] For each training batch, the multilayer fully connected deep neural network model is trained using the training set, the loss of the training set is calculated using the mean squared error loss function, and the parameters of the multilayer fully connected deep neural network model are updated using the Adam optimization algorithm.

[0184] After each training batch, the multilayer fully connected deep neural network model is validated using a validation set. The loss on the validation set is calculated using the mean squared error loss function. Training is terminated early when the loss on the validation set no longer decreases or shows an overfitting trend. The loss curves of the model on the training and validation sets are shown below. Figure 5 As shown;

[0185] After all training is completed, the prediction performance of the multilayer fully connected deep neural network model is evaluated using the test set. If the prediction performance reaches the preset accuracy threshold, the trained multilayer fully connected deep neural network model is output.

[0186] The expression for the mean squared error loss function is as follows:

[0187]

[0188] in, Indicates loss, The total number of training / validation samples, Indicates the first Prediction results for each training / validation sample. Indicates the first The actual charging load of each training / validation sample;

[0189] The prediction performance includes at least one of mean absolute error, root mean square error, and coefficient of determination. The prediction performance evaluation metrics of the multilayer fully connected deep neural network model are shown in Table 3.

[0190] The mean absolute error is obtained by the following formula:

[0191] ,

[0192] in, This represents the total number of test samples. Indicates the first The actual charging load of each test sample Indicates the first Prediction results for each test sample Indicates the mean absolute error;

[0193] The root mean square error is obtained by the following formula:

[0194] ,

[0195] in, Indicates the root mean square error;

[0196] The coefficient of determination is obtained using the following formula:

[0197]

[0198] in, The coefficient of determination is represented by the coefficient of determination. This represents the average of the predicted results for all test samples.

[0199] Table 3. Neural Network Performance Evaluation Indicators

[0200]

[0201] When the model reaches the preset accuracy threshold on the test set, it indicates that the model has converged and has good generalization performance; the final trained deep neural network prediction model is output, achieving accurate prediction of the charging load of each charging station node; the comparison results of the predicted values ​​and actual values ​​of all nodes on the test set are as follows: Figure 6 As shown, the data points are highly clustered in the vicinity, exhibiting a clear linear relationship, indicating that the model can accurately capture the trend of charging load changes.

[0202] Example 2

[0203] Based on the same inventive concept as Embodiment 1, this embodiment of the invention discloses an electric vehicle charging load prediction system based on traffic equilibrium theory and deep neural networks, comprising:

[0204] The model building module is used to: build a traffic user equilibrium model and an electric vehicle feasible path generation model based on the traffic network structure;

[0205] The sample generation module is used to: calculate the total charging load of each charging station node under multiple scenarios based on the traffic user equilibrium model and the electric vehicle feasible path generation model, and obtain a charging load sample set.

[0206] The feature filtering module is used to: perform feature filtering on the charging load sample set using correlation analysis and the SHAP method based on Shapley values, to obtain a filtered charging load sample set.

[0207] The charging load prediction module is used to: train a multilayer fully connected deep neural network model end-to-end using the filtered charging load sample set, and use the trained multilayer fully connected deep neural network model to predict the charging load of each charging station node, thereby obtaining the prediction results.

[0208] The specific functions of each module described above are explained in the relevant content of the method in Embodiment 1, and will not be repeated here.

[0209] Example 3:

[0210] This embodiment provides a computer-readable storage medium storing computer instructions that, when executed by a processor, implement the steps of the electric vehicle charging load prediction method based on traffic equilibrium theory and deep neural networks as described in any one of Embodiments 1.

[0211] Example 4:

[0212] This embodiment provides a computer device, including:

[0213] Memory, used to store computer instructions;

[0214] A processor is configured to execute the computer instructions to implement the steps of the electric vehicle charging load prediction method based on traffic equilibrium theory and deep neural networks as described in any one of Embodiments 1.

[0215] Example 5:

[0216] This embodiment provides a computer program product, including computer instructions, characterized in that, when the computer instructions are executed by a processor, they implement the steps of the electric vehicle charging load prediction method based on traffic equilibrium theory and deep neural networks as described in any one of Embodiment 1.

[0217] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0218] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0219] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0220] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0221] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the protection scope of the present invention.

Claims

1. A method for predicting electric vehicle charging load based on traffic equilibrium theory and deep neural networks, characterized in that, include: Based on the traffic network structure, a traffic user equilibrium model and an electric vehicle feasible path generation model are constructed. Based on the traffic user equilibrium model and the electric vehicle feasible path generation model, the total charging load of each charging station node under multiple scenarios is calculated to obtain a charging load sample set. The charging load sample set was subjected to feature filtering using correlation analysis and the SHAP method based on Shapley values ​​to obtain a filtered charging load sample set. Using the filtered charging load sample set, the multilayer fully connected deep neural network model is trained end-to-end, and the trained multilayer fully connected deep neural network model is used to predict the charging load of each charging station node to obtain the prediction results.

2. The electric vehicle charging load prediction method based on traffic equilibrium theory and deep neural networks according to claim 1, characterized in that, The step of constructing a traffic user equilibrium model based on the traffic network structure includes: Establish a transportation network based on the transportation network structure. ;in, Represents a set of road segments. Represents the set of nodes in a transportation network; Based on the node set of the transportation network Get the set of OD pairs ; where each OD pair Including the origin node and the destination node, ; According to each OD pair Path traffic With broad travel costs Define the traffic user equilibrium conditions and construct a traffic user equilibrium model; where, Indicates a path. Indicates OD pair path Traffic flow Indicates OD pair path The generalized cost of travel.

3. The electric vehicle charging load prediction method based on traffic equilibrium theory and deep neural networks according to claim 2, characterized in that, The traffic user equilibrium conditions are as follows: ， in, Indicates OD pair The minimum travel cost; The objective function of the traffic user equilibrium model is as follows: ， in, Represents the path flow vector. Represents the traffic flow vector of a road segment. This represents the cost coefficient per unit time. Indicates road segment, Indicates road segment Traffic flow Indicates road segment The passage time, Represents the set of charging station nodes. , Indicates charging station node Traffic flow Indicates that electric vehicles are at charging station nodes Total charging time Indicates OD pair The set of paths Indicates the selection of a path Electric vehicles at charging station nodes The amount of charge, Indicates charging station node Electricity price; The traffic user equilibrium model is subject to the following constraints: in, Indicates OD pair The actual travel needs Indicates road segment Free passage time, Indicates road segment Traffic capacity, Indicates charging station node Average waiting time in queues Indicates charging station node capacity, Indicates charging station node The charging power, For binary variables, Indicates the selection of a path Electric vehicles at charging station nodes Charge it. Indicates no charging; For binary variables, Representing a path Passing section , Representing a path No passing section .

4. The electric vehicle charging load prediction method based on traffic equilibrium theory and deep neural networks according to claim 3, characterized in that, The objective function of the electric vehicle feasible path generation model is as follows: ， The feasible path generation model for electric vehicles is subject to the following constraints: ， ， ， ， ， ， ， ， in, This represents the adjacency matrix of nodes and road segments in a transportation network. Let represent the node access decision vector for path p. Indicates OD pair The node demand vector, Indicates the starting node of road segment a. The termination node of road segment a, Indicates the point reached. The state of charge of the battery at that location. Indicates the point reached. The state of charge of the battery at that location. Indicates the section of road Electricity consumed while driving. Indicates at node The amount of charge; For the node The amount of charge; As an auxiliary variable, Representing a path Passing section , Representing a path No passing section , As a preset positive number, Range anxiety parameter for electric vehicles represents the minimum battery capacity threshold that users can tolerate. Indicates the maximum energy storage capacity of the battery. Indicates the initial charge stored in the battery. Represents a node Charging capacity limitations at the location, Represents a node There are no charging stations. Represents a node There are charging stations available; For path The charge of the starting node.

5. The electric vehicle charging load prediction method based on traffic equilibrium theory and deep neural networks according to claim 4, characterized in that, The process involves calculating the total charging load of each charging station node under multiple scenarios based on the traffic user equilibrium model and the electric vehicle feasible path generation model, resulting in a charging load sample set, including: Initialize traffic flow for all road segments. Traffic flow at charging station nodes The available paths for each OD pair are solved using the electric vehicle feasible path generation model to obtain an initial path set. ; Iteratively execute the following steps until no new path is available to update any OD pair: According to the set of paths Update traffic flow for all road segments using a traffic user equilibrium model. Traffic flow at charging station nodes And calculate the travel cost for each route. ; Based on the updated traffic flow of all road sections Traffic flow at charging station nodes The available new paths for each OD pair are updated using the electric vehicle feasible path generation model, and the travel cost of each available new path is calculated. ; For each OD pair, the travel cost if a new route is available... Lower than the travel cost of each route Then add the available new path to the path set. ; After the iteration terminates, the traffic flow and charging amount at charging station nodes for each path of all OD pairs are counted, the total charging load of each charging station node is calculated, and a charging load sample set is obtained. The total charging load of each charging station node is obtained by the following formula: in, Indicates charging station node Total charging load.

6. The electric vehicle charging load prediction method based on traffic equilibrium theory and deep neural networks according to claim 1, characterized in that, The charging load sample set is filtered using correlation analysis and the SHAP method based on Shapley values ​​to obtain a filtered charging load sample set, including: Based on the charging load sample set, construct a sample feature matrix. With the output target vector ;in, , This represents the total number of charging load samples. Representing feature dimension, Indicates the first The first charging load sample One characteristic, , Indicates transpose. Indicates the first Charging load labels for each charging load sample; Using Pearson correlation analysis on the sample feature matrix With the output target vector The degree of linear correlation between them is measured to obtain the correlation coefficient; Remove sample feature matrix The feature matrix of candidate samples is obtained by considering features whose absolute correlation coefficient is less than a set threshold. ; The candidate sample feature matrix was processed using the SHAP method based on Shapley values. Perform feature importance assessment to obtain the importance value of each feature; Based on the absolute value of the importance values ​​of each feature, the candidate sample feature matrix is... Sort the features in descending order and retain the top ones. The key sample feature matrix is ​​obtained from these features. ;in, This is the default value; Joint key sample feature matrix and output target vector The filtered charging load sample set is obtained.

7. The electric vehicle charging load prediction method based on traffic equilibrium theory and deep neural networks according to claim 6, characterized in that, The correlation coefficient is obtained using the following formula: ， in, Indicates the first The correlation coefficient of each feature , Represents the sample feature matrix The Middle The average of the features, Indicates the output target vector The average value; The importance values ​​of each feature are obtained using the following formula: ， in, Represents the feature matrix of candidate samples The Middle The importance value of each feature, Indicates that it does not include the first A subset of features of each feature Represents the feature matrix of candidate samples The feature set in Represents the feature matrix of candidate samples The total number of features in Indicates the feature matrix of candidate samples Selecting a feature subset The resulting sample feature vector This indicates that a multi-layer fully connected deep neural network model can operate based solely on a subset of features. Under the condition of output target vector The prediction results In the sample feature vector Based on the introduction of the first The expanded sample feature vector obtained after considering each feature Indicating in the feature subset Based on the introduction of the first After considering each feature, the multi-layer fully connected deep neural network model outputs the target vector. The prediction results.

8. The electric vehicle charging load prediction method based on traffic equilibrium theory and deep neural networks according to claim 1, characterized in that, The multi-layer fully connected deep neural network model includes an input layer, Hidden layers and output layers; among which, This is the default value; The expression for the multilayer fully connected deep neural network model is: ， in, Indicates the prediction result. Indicates the number of charging stations. Represent real numbers, This represents a nonlinear mapping function for a multilayer fully connected deep neural network. The parameters represent those of a multi-layer fully connected deep neural network. This represents the sample feature vector of a multilayer fully connected deep neural network model. The dimension of the sample feature vector; The expression for the hidden layer is: ， in, Indicates the first The output of the hidden layer, Indicates the first The weight matrix of the hidden layer, Indicates the first The bias vector of the hidden layer. This represents the activation function.

9. The electric vehicle charging load prediction method based on traffic equilibrium theory and deep neural networks according to claim 1, characterized in that, The step of using the filtered charging load sample set to perform end-to-end training on a multilayer fully connected deep neural network model includes: The filtered charging load sample set is preprocessed by normalization and divided into training set, validation set and test set according to proportion; For each training batch, the multilayer fully connected deep neural network model is trained using the training set, the loss of the training set is calculated using the mean squared error loss function, and the parameters of the multilayer fully connected deep neural network model are updated using the Adam optimization algorithm. After each training batch, the multilayer fully connected deep neural network model is validated using a validation set. The loss of the validation set is calculated using the mean squared error loss function. When the loss of the validation set no longer decreases or an overfitting trend appears, an early stopping strategy is adopted to terminate the training. After all training is completed, the prediction performance of the multilayer fully connected deep neural network model is evaluated using the test set. If the prediction performance reaches the preset accuracy threshold, the trained multilayer fully connected deep neural network model is output. The expression for the mean squared error loss function is as follows: in, Indicates loss, The total number of training / validation samples, Indicates the first Prediction results for each training / validation sample. Indicates the first The actual charging load of each training / validation sample; The prediction performance includes at least one of mean absolute error, root mean square error, and coefficient of determination. The mean absolute error is obtained by the following formula: ， in, This represents the total number of test samples. Indicates the first The actual charging load of each test sample Indicates the first Prediction results for each test sample Indicates the mean absolute error; The root mean square error is obtained by the following formula: ， in, Indicates the root mean square error; The coefficient of determination is obtained by the following formula: in, The coefficient of determination is represented by the coefficient of determination. This represents the average of the predicted results for all test samples.

10. An electric vehicle charging load prediction system based on traffic equilibrium theory and deep neural networks, characterized in that, include: The model building module is used to: build a traffic user equilibrium model and an electric vehicle feasible path generation model based on the traffic network structure; The sample generation module is used to: calculate the total charging load of each charging station node under multiple scenarios based on the traffic user equilibrium model and the electric vehicle feasible path generation model, and obtain a charging load sample set. The feature filtering module is used to: perform feature filtering on the charging load sample set using correlation analysis and the SHAP method based on Shapley values, to obtain a filtered charging load sample set. The charging load prediction module is used to: train a multilayer fully connected deep neural network model end-to-end using the filtered charging load sample set, and use the trained multilayer fully connected deep neural network model to predict the charging load of each charging station node, thereby obtaining the prediction results.

11. A computer-readable storage medium storing computer instructions thereon, characterized in that, When the computer instruction is executed by the processor, it implements the steps of the electric vehicle charging load prediction method based on traffic equilibrium theory and deep neural network as described in any one of claims 1-9.

12. A computer device, characterized in that, Memory, used to store computer instructions; A processor for executing the computer instructions to implement the steps of the electric vehicle charging load prediction method based on traffic equilibrium theory and deep neural networks as described in any one of claims 1-9.

13. A computer program product comprising computer instructions, characterized in that, When the computer instructions are executed by the processor, they implement the steps of the electric vehicle charging load prediction method based on traffic equilibrium theory and deep neural networks as described in any one of claims 1-9.

Images