A gcn-rollout-based electric vehicle charging load prediction method

Abstract

This invention belongs to the field of electric vehicle charging load prediction technology, and discloses an electric vehicle charging load prediction method based on GCN-Rollout, including the following steps: real-time acquisition of multi-source data including road network topology data, traffic flow status data, and individual electric vehicle status data; based on graph convolutional network (GCN), extracting charging load impact features reflecting spatial dependencies from the multi-source data, and preprocessing them to obtain the current state vector representing the traffic flow and charging status at the road network nodes at the current moment; defining the charging demand ratio as a decision variable, and generating a discrete set of decision variables based on the current state vector; performing simulation evaluation for each candidate decision variable in the decision variable set to construct the corresponding long-term cost evaluation function; effectively solving the problem in the prior art where the prediction model has insufficient ability to fuse multi-source dynamic information, resulting in poor prediction accuracy and disconnection from scheduling decisions.

Description

Technical Field

[0001] This invention belongs to the field of electric vehicle charging load prediction technology, specifically relating to an electric vehicle charging load prediction method based on GCN-Rollout. Background Technology

[0002] With the rapid popularization of electric vehicles (EVs), their large-scale and randomized charging behavior poses a severe challenge to the stable operation and optimal scheduling of the power grid. Accurate forecasting of EV charging load has become one of the key technologies for realizing smart energy management and power grid security control. Currently, charging load forecasting methods mainly rely on historical data-driven time-series forecasting models and machine learning algorithms, which can capture the macroscopic patterns of load changes to a certain extent. However, current research on electric vehicle charging load forecasting is still in its infancy. Fully leveraging the spatiotemporal dynamics and user behavior patterns of the electric vehicle population is the core challenge in achieving high-precision, online load forecasting. Electric vehicle charging load forecasting methods face several main challenges. First, the factors influencing load are complex and diverse. To improve forecast accuracy, forecasting models need to comprehensively consider multi-source heterogeneous data such as road network topology, real-time traffic flow, vehicle movement trajectories, and user charging habits. These factors exhibit strong coupling and nonlinear relationships, which traditional models struggle to effectively characterize. Second, the system's state is highly time-varying. Vehicle movement and the resulting charging demand evolve rapidly in both spatial and temporal dimensions, making the system state information used for forecasting highly dynamic. Prediction models based on offline training or periodic updates struggle to capture these instantaneous changes in a timely manner, resulting in delayed forecasts that fail to meet the real-time requirements of online grid dispatching. Finally, the coordination between forecasting and decision-making is crucial. Simple load forecasting is disconnected from subsequent grid dispatching decisions. The forecasting process does not consider the impact of these decisions, and this open-loop forecasting model struggles to guarantee overall operational optimality in dynamic scenarios, making it difficult to efficiently solve forecasting and coordinated control problems in complex environments. Therefore, existing technologies suffer from insufficient ability to fuse multi-source dynamic information, resulting in poor forecast accuracy and a disconnect from dispatching decisions. Summary of the Invention

[0003] To address the shortcomings of existing technologies, the present invention aims to provide a method for predicting electric vehicle charging load based on GCN-Rollout, which solves the problem that existing prediction models lack the ability to fuse multi-source dynamic information, resulting in poor prediction accuracy and disconnection from scheduling decisions.

[0004] The objective of this invention can be achieved through the following technical solutions: A method for predicting electric vehicle charging load based on GCN-Rollout includes the following steps: Real-time acquisition of multi-source data, including road network topology data, traffic flow status data, and individual electric vehicle status data; Based on Graph Convolutional Network (GCN), charging load impact features reflecting spatial dependencies are extracted from multi-source data and preprocessed to obtain the current state vector representing the vehicle flow and charging status at road network nodes at the current moment. ; Define the charging demand ratio as the decision variable, and based on the current state vector... Generate a discrete set of decision variables; For each candidate decision variable in the decision variable set, a simulation evaluation is performed to construct the corresponding long-term cost evaluation function, specifically including: 1) Based on actual charging demand and decision variables, calculate the immediate cost generated by adopting candidate decision variables; 2) Based on the current state vector Using the candidate decision variables adopted, the dynamic evolution of the system is simulated through state transitions to obtain the state vector at the next time step. ; 3) Based on the predefined base policy, the state vector at the next time step is... Estimate future accumulated costs; 4) Construct a long-term cost assessment function by combining immediate costs and estimated future cumulative costs; Based on the long-term cost evaluation function, the Rollout strategy is adopted to select the candidate decision variable with the smallest long-term cost evaluation value from the set of decision variables, and use it as the optimal decision variable. By combining the optimal decision variables and the total battery capacity of all electric vehicles in the region, the optimal charging load forecast for the current moment is output.

[0005] Furthermore, the network topology data is presented in a graph structure. It means that, among them, Represents a set of nodes at intersections or road segments; It is a set of edges, representing road connections; Average vehicle speed and total number of electric vehicles at traffic flow status data nodes; Individual status data for electric vehicles includes the vehicle's current location, remaining battery charge (SOC), and destination information.

[0006] Furthermore, based on the Graph Convolutional Network (GCN), the charging load impact features are extracted and preprocessed to obtain the current state vector. Specifically, it includes the following steps: The road network topology data is constructed into an adjacency matrix. A, among which, element Used to represent nodes and nodes The connection relationship is defined as follows: in, , This represents the total number of road network nodes. For each node in the road network topology, construct a five-dimensional feature vector. The feature vectors of each node together form a feature rectangle. X The specific expression is as follows: in, Indicates the degree centrality of a node; Indicates the moment of decision-making Located at node The total number of electric vehicles in the region; Indicates average driving speed; This represents the average state of charge of the electric vehicle's battery. Indicates the percentage of available charging stations; Degree centrality of nodes The expression used to indicate the connectivity importance of a node in the road network topology is as follows: For the characteristic rectangle X Normalize each element by its maximum and minimum values ​​to obtain the normalized feature matrix. Among them, normalized eigenvalues The specific calculation formula is as follows: in, It is a node The The original eigenvalues ​​of the dimension; These represent the positions of all nodes at the [number]th [time]. k Minimum and maximum values ​​in the dimension; For adjacency matrix A Adding self-loops and performing symmetric normalization yields a renormalized symmetric normalized adjacency matrix. ; Normalize the adjacency matrix With normalized characteristic matrix The input is fed into a graph convolutional network with at least two layers, and inter-layer feature propagation is performed through the graph convolutional network to obtain node embedding representations containing spatial dependencies; Global pooling is performed on the node embedding representation to obtain continuous feature vectors representing the road network state. Discretize the continuous feature vector to obtain a fixed-length discrete state vector, which serves as the current state vector. .

[0007] Furthermore, based on actual charging demand and decision variables, the immediate cost generated by adopting candidate decision variables is calculated, as shown in the following formula: In the formula, The proportion of charging demand represented by the currently selected decision variable; The total battery capacity of all electric vehicles in the region; Represents the current state vector Next, candidate decision variables are adopted. The immediate costs incurred; This indicates the actual charging demand at the current moment.

[0008] Furthermore, based on the current state vector Using the candidate decision variables adopted, the dynamic evolution of the system is simulated through state transitions to obtain the state vector at the next time step. The specific expression is as follows: In the formula, This is the state transition function. For parameter set; Let the random noise term satisfy a Gaussian distribution. .

[0009] Furthermore, based on the predefined base policy, the state vector at the next time step is... Estimating future accumulated costs involves the following steps: Using the base policy to obtain the state vector at the next time step Begin estimating future accumulated costs, base strategy The mathematical expression is: base strategy An optimal prediction network trained offline using historical data Achieve optimal prediction network The state vector is taken as input, and the output is a decision variable representing the corresponding proportion of charging demand. The mathematical expression for estimating future cumulative costs is as follows: in, Represents the state vector from the next time step. The initial future accumulated cost; Represents the discount factor, and ; Represents the optimal prediction network Based on the state vector The output decision vector; State vector Next, select the decision vector. The immediate costs incurred Furthermore, by combining immediate costs and estimated future cumulative costs, a long-term cost assessment function is constructed, the specific expression of which is as follows: In the formula, This indicates the long-term cost assessment value.

[0010] Furthermore, by combining the optimal decision variables and the total battery capacity of all electric vehicles in the region, the optimal charging load prediction value for the current moment is output, and the specific calculation formula is as follows: in, This represents the optimal charging load forecast value at the current moment; Represents the optimal decision variable; This represents the sum of the total battery capacity of all electric vehicles in the region that are connected to the power grid or can be immediately scheduled for charging at the current moment; Total battery capacity The calculation formula is as follows: in, Indicates in A collection of electric vehicles that can be charged at any time. It is an electric car Battery capacity.

[0011] The beneficial effects of this invention are: This invention employs Graph Convolutional Networks (GCNs) for feature extraction and fusion of multi-source data, effectively characterizing the complex spatial dependencies between road network topology, vehicle flow status data, and individual electric vehicle states. This improves the accuracy of state representation and the ability to fuse multi-source data. By transforming the charging load prediction problem into an optimization problem with the charging demand ratio as the decision variable, and based on the Rollout strategy, the long-term cost assessment values ​​corresponding to each decision variable, including immediate costs and future cumulative costs, are evaluated. The decision variable corresponding to the minimum long-term cost assessment value is selected as the optimal decision variable. This achieves a comprehensive consideration of the real-time response and long-term impact of the prediction process on the dynamic evolution of the system, thereby significantly improving the dynamic adaptability of the prediction while ensuring prediction accuracy. Combining the optimal decision variable and the total battery capacity of all electric vehicles in the region, the optimal charging load prediction value at the current moment is output, achieving a close connection between the prediction results and scheduling decisions. This effectively solves the problem in existing technologies where the prediction model lacks the ability to fuse multi-source dynamic information, resulting in poor prediction accuracy and a disconnect from scheduling decisions. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 This is a schematic diagram of the overall process of the present invention; Figure 2 This is a schematic diagram of the experimental verification results of the present invention. Detailed Implementation

[0014] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0015] like Figures 1 to 2 As shown, a method for predicting electric vehicle charging load based on GCN-Rollout includes the following steps: Real-time acquisition of multi-source data, including road network topology data, traffic flow status data, and individual electric vehicle status data; Based on Graph Convolutional Network (GCN), charging load impact features reflecting spatial dependencies are extracted from multi-source data and preprocessed to obtain the current state vector representing the vehicle flow and charging status at the road network nodes at the current moment. ; Define the charging demand ratio as the decision variable, and based on the current state vector... Generate a discrete set of decision variables; For each candidate decision variable in the decision variable set, a simulation evaluation is performed to construct the corresponding long-term cost evaluation function, specifically including: 5) Based on actual charging demand and decision variables, calculate the immediate cost generated by adopting candidate decision variables; 6) Based on the current state vector Using the candidate decision variables adopted, the dynamic evolution of the system is simulated through state transitions to obtain the state vector at the next time step. ; 7) Based on the predefined base policy, process the state vector at the next time step. Estimate future accumulated costs; 8) Construct a long-term cost assessment function by combining immediate costs and estimated future cumulative costs; Based on the long-term cost evaluation function, the Rollout strategy is adopted to select the candidate decision variable with the smallest long-term cost evaluation value from the set of decision variables, and use it as the optimal decision variable. By combining the optimal decision variables and the total battery capacity of all electric vehicles in the region, the optimal charging load forecast for the current moment is output.

[0016] This invention employs Graph Convolutional Networks (GCNs) for feature extraction and fusion of multi-source data, effectively characterizing the complex spatial dependencies between road network topology, vehicle flow status data, and individual electric vehicle states. This improves the accuracy of state representation and the ability to fuse multi-source data. By transforming the charging load prediction problem into an optimization problem with the charging demand ratio as the decision variable, and based on the Rollout strategy, the long-term cost assessment values ​​corresponding to each decision variable, including immediate costs and future cumulative costs, are evaluated. The decision variable corresponding to the minimum long-term cost assessment value is selected as the optimal decision variable. This achieves a comprehensive consideration of the real-time response and long-term impact of the prediction process on the dynamic evolution of the system, thereby significantly improving the dynamic adaptability of the prediction while ensuring prediction accuracy. Combining the optimal decision variable and the total battery capacity of all electric vehicles in the region, the optimal charging load prediction value at the current moment is output, achieving a close connection between the prediction results and scheduling decisions. This effectively solves the problem in existing technologies where the prediction model lacks the ability to fuse multi-source dynamic information, resulting in poor prediction accuracy and a disconnect from scheduling decisions.

[0017] Define the charging demand ratio as the decision variable, and based on the current state vector... Generating a discrete set of decision variables involves the following steps: Define Action The charging demand ratio, theoretically, takes the value within the continuous interval [0, 1]. This continuous interval [0, 1] is uniformly discretized as follows: Each level generates a finite, discrete set of optional decision variables as follows: in, The discretization level is a positive integer greater than 1. Representing the There are candidate decision variables; when When the set of decision variables is .

[0018] Network topology data in graph structure It means that, among them, Represents a set of nodes at intersections or road segments; It is a set of edges, representing road connections; Average vehicle speed and total number of electric vehicles at traffic flow status data nodes; Individual status data for electric vehicles includes the vehicle's current location, remaining battery charge (SOC), and destination information.

[0019] Based on the Graph Convolutional Network (GCN), the charging load impact features are extracted and preprocessed to obtain the current state vector. Specifically, it includes the following steps: The road network topology data is constructed into an adjacency matrix. A , among which, element Used to represent nodes and nodes The connection relationship is defined as follows: in, , This represents the total number of road network nodes. For each node in the road network topology, construct a five-dimensional feature vector. The feature vectors of each node together form a feature rectangle. X The specific expression is as follows: in, Indicates the degree centrality of a node; Indicates the moment of decision-making Located at node The total number of electric vehicles in the region; Indicates average driving speed; This represents the average state of charge of the electric vehicle's battery. Indicates the percentage of available charging stations; Degree centrality of nodes The expression used to indicate the connectivity importance of a node in the road network topology is as follows: For the characteristic rectangle X Normalize each element by its maximum and minimum values ​​to obtain the normalized feature matrix. Among them, normalized eigenvalues The specific calculation formula is as follows: in, It is a node The The original eigenvalues ​​of the dimension; These represent the positions of all nodes at the [number]th node. k Minimum and maximum values ​​in the dimension; For adjacency matrix A Adding self-loops and performing symmetric normalization yields a renormalized symmetric normalized adjacency matrix. ; Normalize the adjacency matrix With normalized characteristic matrix The input is fed into a graph convolutional network with at least two layers, and inter-layer feature propagation is performed through the graph convolutional network to obtain node embedding representations containing spatial dependencies; Global pooling is performed on the node embedding representation to obtain continuous feature vectors representing the road network state. Discretize the continuous feature vector to obtain a fixed-length discrete state vector, which serves as the current state vector. ; Among them, the normalized feature matrix and adjacency matrix The input is fed into a graph convolutional network model, where neighbor node information is aggregated through a graph structure defined by an adjacency matrix. The first layer uses an activation function to output node embeddings. The process can be described by the following model: in, It is a renormalized symmetric normalized adjacency matrix, where (Add self-loop) yes The degree matrix (diagonal matrix) ). It is the trainable weight matrix of the first layer. It is the dimension of the first hidden unit.

[0020] The second layer output contains the final node embeddings of spatial dependencies. The process is as follows: in, It is the dimension of the second-layer hidden unit, and also the embedding dimension of each node in the end. This is the node embedding output from the second layer, containing 2-hop neighborhood information. Next, a global average pooling operation is performed on the node embedding matrix to obtain a graph-level continuous state vector, as follows: in, yes The Okay, what you get It is a continuous state vector. The number of rows in the vector.

[0021] Finally, the continuous state vector is discretized by binning, as follows: in, It is the floor function. It is in Integers within the range. For a graph-level continuous state vector in the th order... The original values ​​in each dimension For the first Preset lower limits for each feature dimension For the first Preset upper limit values ​​for each feature dimension Indicates the first The number of bins in each feature dimension.

[0022] Based on actual charging demand and decision variables, the immediate cost generated by adopting candidate decision variables is calculated as follows: In the formula, The proportion of charging demand represented by the currently selected decision variable; The total battery capacity of all electric vehicles in the region; Represents the current state vector Next, candidate decision variables are adopted. The immediate costs incurred; This indicates the actual charging demand as of the current moment. Real-time costs serve as a reference for improving the parameters of graph convolutional networks (GCNs).

[0023] Based on the current state vector Using the candidate decision variables adopted, the dynamic evolution of the system is simulated through state transitions to obtain the state vector at the next time step. The specific expression is as follows: In the formula, This is the state transition function, with parameters... Obtained through offline training using historical data, used to determine the current state vector. and decision variables Predict the core dynamics of the system; It is a random noise term used to simulate environmental uncertainty, and it follows a mean of 0 and a covariance of . The Gaussian distribution, i.e. ; State transition function By one Implemented using a layered feedforward neural network, the input of which is the state. and action spliced ​​vector The forward propagation process of this network is established as follows: in, , This represents the activation function. and For the weights and bias parameters of each layer, and These are the output layer parameters.

[0024] Based on a predefined base policy, the state vector at the next time step... Estimating future accumulated costs involves the following steps: Using the base policy to obtain the state vector at the next time step We begin estimating future accumulated costs. The mathematical expression for the base policy is: base strategy An optimal prediction network trained offline using historical data Achieve optimal prediction network The state vector is taken as input, and the output is a decision variable representing the corresponding proportion of charging demand. The mathematical expression for estimating future cumulative costs is as follows: in, Represents the state vector from the next time step. The initial future accumulated cost; Represents the discount factor, and ; Represents the optimal prediction network Based on the state vector The output decision vector; State vector Next, select the decision vector. The immediate costs incurred.

[0025] By combining immediate costs and estimated future cumulative costs, a long-term cost assessment function is constructed, the specific expression of which is as follows: In the formula, This indicates the long-term cost assessment value.

[0026] Combining the optimal decision variables and the total battery capacity of all electric vehicles in the region, the optimal charging load prediction value for the current moment is output, and the specific calculation formula is as follows: in, This represents the optimal charging load forecast value at the current moment; Represents the optimal decision variable; This represents the sum of the total battery capacity of all electric vehicles in the region that are connected to the power grid or can be immediately scheduled for charging at the current moment; Total battery capacity The calculation formula is as follows: in, Indicates in A collection of electric vehicles that can be charged at any time. It is an electric car Battery capacity.

[0027] To verify the method of this application, the following experiments were conducted: The hardware environment consisted of a CPU i7-11700H @ 2.50 GHz, 24GB of memory, and MATLAB as the runtime environment. A 24-hour online prediction of the electric vehicle charging load at 32 road network traffic nodes was performed. The electric vehicle battery capacity was considered to be uniformly distributed between 50-80 kWh, with an electric vehicle charging power of 7 kW, and the initial simulated SOC range was 20%-80%. The total prediction time was 24 hours, with a discrete time step of 1 hour, and a future discount factor was used. The value was set to 0.95, and the total number of vehicles was 1000; the experimental results are shown below. Figure 2 The figure shows the predicted total charging load of the system. Figure 2It can be seen that the predicted charging load exhibits diurnal fluctuations, which is due to the typical behavior of electric vehicle users who drive during the day and charge at night. It can also be noted that the charging load peaks at night, which is consistent with the actual grid load curve, verifying the effectiveness of the proposed prediction method.

[0028] The established GCN-Rollout prediction framework effectively integrates multi-source spatiotemporal data while considering the real-time and dynamic requirements of charging load prediction, thus improving the prediction system's adaptability to the dynamic behavior of electric vehicles. The feature extraction method based on GCN graph convolutional networks can deeply mine the spatiotemporal correlation features of road network structure and traffic flow status, improving the accuracy of state representation. Furthermore, this method, combined with the improved features of the Rollout online strategy, achieves synergy between prediction and decision-making through forward-looking simulation optimization while maintaining prediction accuracy, significantly improving the solution speed and efficiency of charging load prediction in complex environments.

[0029] The charging load prediction method of this invention can effectively characterize the charging behavior patterns of electric vehicle users and solve the difficulty of predicting the spatiotemporal distribution of load under the coupling of multiple factors. During online prediction, the algorithm is improved by using the Rollout strategy to enhance the foresight of the decision-making process. Simultaneously, the traditional prediction method is improved by combining GCN feature extraction, which ensures the rationality of the load distribution during the prediction process and improves the efficiency of problem solving.

[0030] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0031] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the present invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention.

Claims

1. A method for predicting electric vehicle charging load based on GCN-Rollout, characterized in that, Includes the following steps: Real-time acquisition of multi-source data, including road network topology data, traffic flow status data, and individual electric vehicle status data; Based on Graph Convolutional Network (GCN), charging load impact features reflecting spatial dependencies are extracted from multi-source data and preprocessed to obtain the current state vector representing the vehicle flow and charging status at road network nodes at the current moment. ; Define the charging demand ratio as the decision variable, and based on the current state vector... Generate a discrete set of decision variables; For each candidate decision variable in the decision variable set, a simulation evaluation is performed to construct the corresponding long-term cost evaluation function, specifically including: 1) Based on actual charging demand and decision variables, calculate the immediate cost generated by adopting candidate decision variables; 2) Based on the current state vector Using the candidate decision variables adopted, the dynamic evolution of the system is simulated through state transitions to obtain the state vector at the next time step. ; 3) Based on the predefined base policy, the state vector at the next time step is... Estimate future accumulated costs; 4) Construct a long-term cost assessment function by combining immediate costs and estimated future cumulative costs; Based on the long-term cost evaluation function, the Rollout strategy is adopted to select the candidate decision variable with the smallest long-term cost evaluation value from the set of decision variables, and use it as the optimal decision variable. By combining the optimal decision variables and the total battery capacity of all electric vehicles in the region, the optimal charging load forecast for the current moment is output.

2. The electric vehicle charging load prediction method based on GCN-Rollout according to claim 1, characterized in that, Network topology data in graph structure It means that, among them, Represents a set of nodes at intersections or road segments; It is a set of edges, representing road connections; Average vehicle speed and total number of electric vehicles at traffic flow status data nodes; Individual status data for electric vehicles includes the vehicle's current location, remaining battery charge (SOC), and destination information.

3. The electric vehicle charging load prediction method based on GCN-Rollout according to claim 2, characterized in that, Based on the Graph Convolutional Network (GCN), the charging load impact features are extracted and preprocessed to obtain the current state vector. Specifically, it includes the following steps: The road network topology data is constructed into an adjacency matrix. A , among which, element Used to represent nodes and nodes The connection relationship is defined as follows: in, , This represents the total number of road network nodes. For each node in the road network topology, construct a five-dimensional feature vector. The feature vectors of each node together form a feature rectangle. X The specific expression is as follows: in, Indicates the degree centrality of a node; Indicates the moment of decision-making Located at node The total number of electric vehicles in the region; Indicates average driving speed; This represents the average state of charge of the electric vehicle's battery. Indicates the percentage of available charging stations; Degree centrality of nodes The expression used to indicate the connectivity importance of a node in the road network topology is as follows: For the characteristic rectangle X Normalize each element by its maximum and minimum values ​​to obtain the normalized feature matrix. Among them, normalized eigenvalues The specific calculation formula is as follows: in, It is a node The The original eigenvalues ​​of the dimension; These represent the positions of all nodes at the [number]th [time]. k Minimum and maximum values ​​in the dimension; For adjacency matrix A Adding self-loops and performing symmetric normalization yields a renormalized symmetric normalized adjacency matrix. ; Normalize the adjacency matrix With normalized characteristic matrix The input is fed into a graph convolutional network with at least two layers, and inter-layer feature propagation is performed through the graph convolutional network to obtain node embedding representations containing spatial dependencies; Global pooling is performed on the node embedding representation to obtain continuous feature vectors representing the road network state. Discretize the continuous feature vector to obtain a fixed-length discrete state vector, which serves as the current state vector. .

4. The electric vehicle charging load prediction method based on GCN-Rollout according to claim 3, characterized in that, Based on actual charging demand and decision variables, the immediate cost generated by adopting candidate decision variables is calculated as follows: In the formula, The proportion of charging demand represented by the currently selected decision variable; The total battery capacity of all electric vehicles in the region; Represents the current state vector Next, candidate decision variables are adopted. The immediate costs incurred; This indicates the actual charging demand at the current moment.

5. The electric vehicle charging load prediction method based on GCN-Rollout according to claim 4, characterized in that, Based on the current state vector Using the candidate decision variables adopted, the dynamic evolution of the system is simulated through state transitions to obtain the state vector at the next time step. The specific expression is as follows: In the formula, This is the state transition function. For parameter set; Let the random noise term satisfy a Gaussian distribution. .

6. The electric vehicle charging load prediction method based on GCN-Rollout according to claim 5, characterized in that, Based on a predefined base policy, the state vector at the next time step... Estimating future accumulated costs involves the following steps: Using the base policy to obtain the state vector at the next time step Begin estimating future accumulated costs, base strategy The mathematical expression is: base strategy An optimal prediction network trained offline using historical data Achieve optimal prediction network The state vector is taken as input, and the output is a decision variable representing the corresponding proportion of charging demand. The mathematical expression for estimating future cumulative costs is as follows: in, Represents the state vector from the next time step. The initial future accumulated cost; Represents the discount factor, and ; Represents the optimal prediction network Based on the state vector The output decision vector; State vector Next, select the decision vector. The immediate costs incurred.

7. The electric vehicle charging load prediction method based on GCN-Rollout according to claim 6, characterized in that, By combining immediate costs and estimated future cumulative costs, a long-term cost assessment function is constructed, the specific expression of which is as follows: In the formula, This indicates the long-term cost assessment value.

8. The electric vehicle charging load prediction method based on GCN-Rollout according to claim 7, characterized in that, Combining the optimal decision variables and the total battery capacity of all electric vehicles in the region, the optimal charging load prediction value for the current moment is output, and the specific calculation formula is as follows: in, This represents the optimal charging load forecast value at the current moment; Represents the optimal decision variable; This represents the sum of the total battery capacity of all electric vehicles in the region that are connected to the power grid or can be immediately scheduled for charging at the current moment; Total battery capacity The calculation formula is as follows: in, Indicates in A collection of electric vehicles that can be charged at any time; It is an electric car Battery capacity.

Images