A multi-source data-based vehicle and pile network state sensing scheduling method

Abstract

The application discloses a kind of based on multi-source data's car pile net state perception scheduling method, this method includes: the multi-source data of real-time acquisition vehicle, charging pile and power grid, wherein the vehicle, charging pile is set to the node of graph neural network, vehicle and charging pile relationship is abstracted as the edge of graph neural network, node and edge are attached with time sequence characteristics;The multi-source data of vehicle, charging pile and power grid collected is input to pre-trained graph neural network model, the hierarchical message passing and aggregation of the model are passed through, generate the uniform state perception vector of each node, and the charging pile state is cooperatively perceived;Based on the result that charging pile state is cooperatively perceived, construct the collaborative optimization problem with optimal operation efficiency as the goal, generate the distributed collaborative control instruction set for vehicle, charging facility and power grid by the distributed decision mechanism based on graph structure.The present application can solve the key technical problems that the existing scheduling scheme is mostly one-way or local optimization, for example, only from the power grid angle load management is carried out, or only from the charging station angle operation optimization is carried out, lack of global collaborative perception and decision-making ability.

Description

Technical Field

[0001] This invention relates to the technical field of electric vehicle scheduling, and specifically to a vehicle-charging-network state-aware scheduling method based on multi-source data. Background Technology

[0002] In the integrated vehicle-station-network operation system, EVs are a key coupling element in the future smart city transportation-power-information coupling network. Accurately predicting EV charging load is the foundation for analyzing the impact of EVs on the distribution network and a necessary prerequisite for EV-distribution network interaction. As the energy hub unit of the integrated vehicle-station-network operation system, charging stations optimize their layout based on the spatiotemporal distribution prediction results of EV load, achieving optimized layout at the "station-network" level; they guide orderly charging considering the distribution network demand, achieving collaborative management at the "station-vehicle" level; they receive grid dispatch signals and send them to EVs, feeding back EV charging information to the distribution network, and using V2G technology to achieve interactive response at the "vehicle-network" level.

[0003] Existing scheduling schemes are mostly unidirectional or localized optimizations, such as load management from the perspective of the power grid or operation optimization from the perspective of charging stations, lacking global collaborative perception and decision-making capabilities. Summary of the Invention

[0004] The purpose of this invention is to overcome the defects and shortcomings of existing microscopic spectroscopy technology and propose a vehicle-charging-network state perception and scheduling method based on multi-source data. This method can solve the key technical problem that existing scheduling schemes are mostly unidirectional or local optimizations, such as load management only from the perspective of the power grid or operation optimization only from the perspective of charging stations, lacking global collaborative perception and decision-making capabilities.

[0005] This invention discloses a vehicle-charging-pile network state awareness scheduling method based on multi-source data. The method includes: real-time collection of multi-source data on vehicles, charging piles and power grid, wherein the vehicles and charging piles are set as nodes of a graph neural network, the relationship between vehicles and charging piles is abstracted as edges of the graph neural network, and nodes and edges are attached with time sequence features.

[0006] The collected multi-source data from vehicles, charging piles, and the power grid are input into a pre-trained graph neural network model. Through the hierarchical message passing and aggregation of the model, a unified state perception vector for each node is generated, and the state of the charging piles is collaboratively perceived.

[0007] Based on the results of collaborative perception of the charging pile status, a collaborative optimization problem with the goal of maximizing operational efficiency is constructed. Through a distributed decision-making mechanism based on graph structure, a distributed collaborative control instruction set for vehicles, charging facilities, and the power grid is generated.

[0008] Preferably, the multi-source data collected in real time from vehicles, charging piles, and the power grid includes:

[0009] Vehicle node data: real-time location, battery state of charge (SOC), driving speed, destination, planned stay time, vehicle type, and user charging preferences;

[0010] Charging pile node data: real-time power, voltage / current, operating status, affiliated charging station, rated power, service fee rate, and historical fault records;

[0011] Grid-related data: real-time load factor of the local transformer area, marginal electricity price at nodes, grid dispatch instructions, and renewable energy output forecasts.

[0012] Spatiotemporal environmental data: local traffic congestion index, weather conditions, temperature affecting charging efficiency, and time period.

[0013] Preferably, the heterogeneous graph convolution module uses a relation-specific weight matrix to transmit differentiated information for different types of edges. For node vi, its representation calculation at layer l+1 is as follows:

[0014]

[0015] Where, N r i W represents the set of neighbors of node vi under relation r. (l) r Let σ be the learnable weight matrix corresponding to relation r, and σ be the activation function. The spatiotemporal attention module includes two sub-modules: spatial attention and temporal attention. Spatial attention calculates the correlation weights between node pairs, and temporal attention captures the temporal dependencies of node features. The state decoding module maps the node representation after multi-layer aggregation into a specific state perception vector. For a charging pile node, this vector includes at least: real-time load rate, health score, competition intensity index, and demand prediction value.

[0016] Preferably, a unified state perception vector generated based on the state collaborative perception of heterogeneous graph neural networks is used as input to construct an optimization problem with the goal of minimizing the overall system operating cost, and a collaborative scheduling instruction for each entity is generated through a distributed decision-making mechanism based on graph structure.

[0017] Define the system optimization objective function; define system operation constraints, including: charging pile power upper and lower limit constraints, vehicle battery SOC constraints, grid power flow security constraints, and user latest departure time constraints; adopt a distributed solution method based on multi-agent deep reinforcement learning, treating each vehicle and charging pile as an agent, whose policy network takes the corresponding state perception vector as input and outputs local control actions; through a graph-level collaborative training algorithm, the policies of each agent evolve collaboratively while satisfying the global optimization objective, ultimately generating a scheduling instruction set, including: the power setpoint Pset of each charging pile, the dynamic service fee rate λ, and the charging station selection suggestions and charging / discharging power plans P for each vehicle. V2G .

[0018] Preferably,

[0019]

[0020] Preferably, the distributed decision-making mechanism based on graph structure specifically involves: defining local policy networks or decision functions for the power grid nodes, charging station nodes, and vehicle nodes in the dynamic heterogeneous spatiotemporal graph; unifying the state perception vector as the decision input for each node; and using a graph-level collaborative training algorithm to jointly optimize the local policies of all nodes under the global objective of the collaborative optimization problem, thereby achieving distributed collaborative decision-making.

[0021] The graph-level collaborative training algorithm is a multi-agent graph reinforcement learning algorithm, in which each decision node is regarded as an agent, its state is the unified state perception vector, its action is the local control command, the environment model is simulated by the heterogeneous spatiotemporal graph neural network, and the reward function is associated with the objective function of the collaborative optimization problem.

[0022] The generated collaborative control instruction set includes: real-time recommended power setting values ​​and dynamic electricity price signals for each charging pile, charging and discharging power plans for each electric vehicle supporting V2G function, and grid adjustment demand signals for the charging station cluster.

[0023] This invention also provides a large-scale, seamless aggregated control system for electric vehicles oriented towards virtual power plants, characterized in that it includes:

[0024] The data acquisition module is used to collect multi-source data from vehicles, charging piles and power grids in real time. The vehicles and charging piles are set as nodes of a graph neural network, and the relationship between vehicles and charging piles is abstracted as edges of the graph neural network. The nodes and edges are attached with time sequence features.

[0025] The perception module is used to input multi-source data collected from vehicles, charging piles and the power grid into a pre-trained graph neural network model. Through the hierarchical message passing and aggregation of the model, a unified state perception vector for each node is generated, and the charging pile status is collaboratively perceived.

[0026] The decision-making module is used to construct a collaborative optimization problem with the goal of maximizing operational efficiency based on the results of collaborative perception of the charging pile status. Through a graph-based distributed decision-making mechanism, it generates a distributed collaborative control instruction set for vehicles, charging facilities, and the power grid.

[0027] This invention collects multi-source data from vehicles, charging piles, and the power grid in real time. The vehicles and charging piles are set as nodes in a graph neural network, and the relationship between them is abstracted as edges. Nodes and edges are assigned time-series features. The collected multi-source data is input into a pre-trained graph neural network model. Through hierarchical message passing and aggregation, a unified state perception vector is generated for each node, and the state of charging piles is collaboratively perceived. Based on the results of the collaborative perception of charging pile states, a collaborative optimization problem with optimal operational efficiency is constructed. Through a distributed decision-making mechanism based on a graph structure, a distributed collaborative control instruction set for vehicles, charging facilities, and the power grid is generated. By using a unified heterogeneous graph model, heterogeneous data sources such as vehicles, charging piles, and the power grid are organically integrated, achieving effective utilization of cross-domain information and providing a data foundation for global optimization. State perception accuracy is improved: utilizing the message passing mechanism of the heterogeneous graph neural network, each node can aggregate information from multi-hop neighbors, forming a deep perception of the overall system state. Its perception accuracy is more than 30% higher than that of traditional independent prediction models (verified by simulation). Achieving true collaborative scheduling: The proposed graph-based distributed decision-making mechanism ensures the autonomy of each entity while achieving global collaboration through graph learning, resolving the contradiction between the computational complexity of centralized optimization and the difficulty in coordinating distributed optimization. Enhancing system adaptability: The closed-loop feedback mechanism enables the system to respond to dynamic changes in real time, maintaining efficient operation even under emergencies such as power grid failures and traffic congestion, significantly improving system resilience. Possessing good scalability: The graph-based architecture naturally supports system scaling; adding new vehicles or charging piles only requires adding them as new nodes to the graph, without needing to reconstruct the entire system. Attached Figure Description

[0028] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0029] 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.

[0030] Figure 1This is a flowchart of a vehicle-charging station network state awareness and scheduling method based on multi-source data proposed in this invention;

[0031] Figure 2 This is a framework diagram of a large-scale, sensorless aggregate control system for electric vehicles oriented towards virtual power plants, as proposed in this invention. Detailed Implementation

[0032] 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 a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0033] It should be noted that all directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present invention are only used to explain the relative positional relationship and movement of each component in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indication will also change accordingly.

[0034] Furthermore, the use of terms such as "first" and "second" in this invention is for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, features defined with "first" and "second" may explicitly or implicitly include at least one of those features. Additionally, the technical solutions of the various embodiments can be combined with each other, but only on the basis of being achievable by those skilled in the art. When the combination of technical solutions is contradictory or impossible to implement, such a combination of technical solutions should be considered non-existent and not within the scope of protection claimed by this invention.

[0035] Example 1, Reference Figure 1 This invention discloses a vehicle-charging-network state perception scheduling method based on multi-source data. The method includes: Step 100, real-time collection of multi-source data of vehicles, charging piles and power grid, wherein the vehicles and charging piles are set as nodes of a graph neural network, the relationship between vehicles and charging piles is abstracted as edges of the graph neural network, and nodes and edges are attached with time sequence features.

[0036] Step 200: Input the collected multi-source data of vehicles, charging piles and power grid into the pre-trained graph neural network model. Through the hierarchical message passing and aggregation of the model, generate a unified state perception vector for each node and perform collaborative perception of the charging pile status.

[0037] Step 300: Based on the results of collaborative perception of the charging pile status, construct a collaborative optimization problem with the goal of optimizing operational efficiency, and generate a distributed collaborative control instruction set for vehicles, charging facilities and power grid through a distributed decision-making mechanism based on graph structure.

[0038] This invention collects multi-source data from vehicles, charging piles, and the power grid in real time. The vehicles and charging piles are set as nodes in a graph neural network, and the relationship between them is abstracted as edges. Nodes and edges are assigned time-series features. The collected multi-source data is input into a pre-trained graph neural network model. Through hierarchical message passing and aggregation, a unified state perception vector is generated for each node, and the state of charging piles is collaboratively perceived. Based on the results of the collaborative perception of charging pile states, a collaborative optimization problem with optimal operational efficiency is constructed. Through a distributed decision-making mechanism based on a graph structure, a distributed collaborative control instruction set for vehicles, charging facilities, and the power grid is generated. By using a unified heterogeneous graph model, heterogeneous data sources such as vehicles, charging piles, and the power grid are organically integrated, achieving effective utilization of cross-domain information and providing a data foundation for global optimization. State perception accuracy is improved: utilizing the message passing mechanism of the heterogeneous graph neural network, each node can aggregate information from multi-hop neighbors, forming a deep perception of the overall system state. Its perception accuracy is more than 30% higher than that of traditional independent prediction models (verified by simulation). Achieving true collaborative scheduling: The proposed graph-based distributed decision-making mechanism ensures the autonomy of each entity while achieving global collaboration through graph learning, resolving the contradiction between the computational complexity of centralized optimization and the difficulty in coordinating distributed optimization. Enhancing system adaptability: The closed-loop feedback mechanism enables the system to respond to dynamic changes in real time, maintaining efficient operation even under emergencies such as power grid failures and traffic congestion, significantly improving system resilience. Possessing good scalability: The graph-based architecture naturally supports system scaling; adding new vehicles or charging piles only requires adding them as new nodes to the graph, without needing to reconstruct the entire system.

[0039] Preferably, the multi-source data collected in real time from vehicles, charging piles, and the power grid includes:

[0040] Vehicle node data: real-time location, battery state of charge (SOC), driving speed, destination, planned stay time, vehicle type, and user charging preferences;

[0041] Charging pile node data: real-time power, voltage / current, operating status, affiliated charging station, rated power, service fee rate, and historical fault records;

[0042] Grid-related data: real-time load factor of the local transformer area, marginal electricity price at nodes, grid dispatch instructions, and renewable energy output forecasts.

[0043] Spatiotemporal environmental data: local traffic congestion index, weather conditions, temperature affecting charging efficiency, and time period.

[0044] Preferably, the heterogeneous graph convolution module uses a relation-specific weight matrix to transmit differentiated information for different types of edges. For node vi, its representation calculation at layer l+1 is as follows:

[0045]

[0046] Where, N r i W represents the set of neighbors of node vi under relation r. (l) r Let σ be the learnable weight matrix corresponding to relation r, and σ be the activation function. The spatiotemporal attention module includes two sub-modules: spatial attention and temporal attention. Spatial attention calculates the correlation weights between node pairs, and temporal attention captures the temporal dependencies of node features. The state decoding module maps the node representation after multi-layer aggregation into a specific state perception vector. For a charging pile node, this vector includes at least: real-time load rate, health score, competition intensity index, and demand prediction value.

[0047] Graph construction process: Create a vehicle node for each online vehicle, with the feature vector: [SOC, latitude, longitude, speed, destination_eta]. Create a charging station node for each charging station, with the feature vector: [power, voltage, status_code, queue_length, price]. Divide the city into 1km×1km grids, and create a region node for each grid with a charging station, with the feature vector: [traffic_index, vehicle_count, weather_code].

[0048] Establish edge relationships:

[0049] Spatial proximity edge: The vehicle establishes an edge with all charging stations within 3km, with a weight of 1 / (1+distance).

[0050] Service connection edge: An edge is established between the vehicle that is charging and the corresponding charging pile, with a weight of 1.0.

[0051] Power grid supply side: Based on the power distribution topology, establish a connection side between the charging station and the upstream transformer.

[0052] Traffic impact edge: An edge is established between the road congestion node and the downstream charging station area node, with weight equal to the congestion index.

[0053] The final heterogeneous graph contains approximately 5,350 nodes and 120,000 edges, with the graph structure updated every 30 seconds.

[0054] State collaborative perception employs a heterogeneous spatiotemporal graph neural network model, with the following configuration: Input layer dimensions: 15-dimensional for vehicle nodes, 10-dimensional for charging pile nodes, and 8-dimensional for region nodes. Heterogeneous graph convolutional layers: 2 layers, with a hidden layer dimension of 128, using the ReLU activation function.

[0055] Spatiotemporal attention heads: 4 heads of attention, time window length T=6 (corresponding to 30 minutes of history).

[0056] Output layer: Each node generates a 256-dimensional unified state-aware vector.

[0057] Model Training: Supervised learning training was performed using data from the past three months. The loss function was: L = λ1Lload + λ2Lqueue + λ3Lprice, where Lload is the mean squared error of charging pile load prediction, Lqueue is the cross-entropy loss of queue length prediction, and Lprice is the KL divergence of price sensitivity. After training, the model's mean absolute percentage error in load prediction on the test set was 8.3%, significantly better than the traditional ARIMA model (15.7%). Online Inference: Online inference was performed every 5 minutes, inputting the latest dynamic heterogeneous graph and outputting the perception vector for each charging pile. For example, for pile number 3 of a fast charging station A, its perception vector is shown as follows:

[0058] Real-time load factor: 0.85 (high load)

[0059] Health score: 0.92 (Excellent)

[0060] Competition intensity: 0.45 (moderate competition)

[0061] Demand forecast in 15 minutes: 0.95 (approaching full load); Grid support potential: 0.75 (can participate in peak shaving); Distributed coordinated scheduling; Scheduling optimization model parameters: Objective function weights: α=0.4, β=0.2, γ=0.3, δ=0.1

[0062] Time range: the next 2 hours, in 15-minute intervals.

[0063] Constraints:

[0064] Charging pile power: 0-150kW continuously adjustable

[0065] Vehicle SOC: Charge stops after reaching 80%.

[0066] Power grid safety: Node voltage deviation not exceeding ±5%.

[0067] User constraint: The reserved vehicle will be fully charged before the agreed time.

[0068] Distributed decision-making mechanism:

[0069] A local policy network (3-layer fully connected neural network) is deployed for each charging station. The input is a 256-dimensional perception vector, and the output is the suggested power Prec and the dynamic price adjustment Δλ.

[0070] A decision network is deployed for each vehicle seeking charging, with its own state vector and the perception vector of candidate charging stations as inputs, and the output being the probability distribution of charging station selection.

[0071] A multi-agent proximal policy optimization algorithm is used for collaborative training, with centralized training and distributed execution.

[0072] Scheduling instruction generation: Taking a certain scheduling cycle as an example, the system generates the following instructions:

[0073] For charging stations with a load rate exceeding 90% (a total of 8 stations), the service fee rate will be increased by 15%-25%.

[0074] For charging stations with a load rate of less than 30% (a total of 12), the service fee rate will be reduced by 10%, and preferential information will be pushed to nearby vehicles.

[0075] Arrange 15 vehicles with SOC>70% and supporting V2G to discharge to the grid at a power of 30kW during peak grid hours (18:00-20:00).

[0076] Twenty vehicles with an SOC of less than 20% in urgent need will be directed to charging stations with moderate load and no need to queue.

[0077] Command execution and effect evaluation; Commands are issued through the following channels: Charging pile control commands: issued to the charging pile controller through the charging operation platform to adjust power and rate.

[0078] Vehicle guidance instructions: pushed to the driver via the vehicle's infotainment system or mobile app. V2G dispatch instructions: communicate with the vehicle's BMS via the aggregator platform.

[0079] Execution effect monitoring: Grid side: The highest load rate of distribution transformers in the dispatch area decreased from 95% to 82%, and the voltage qualification rate increased from 96.5% to 99.2%.

[0080] On the charging station side: the load balance of each station (standard deviation / mean) improved from 0.38 to 0.21, and the overall utilization rate increased by 15%.

[0081] On the user side: the average waiting time was reduced from 25 minutes to 12 minutes, and the user satisfaction survey score improved from 3.8 / 5 to 4.3 / 5.

[0082] The system performs closed-loop optimization every 15 minutes and updates the scheduling strategy based on the latest data.

[0083] Preferably, a unified state perception vector generated based on the state collaborative perception of heterogeneous graph neural networks is used as input to construct an optimization problem with the goal of minimizing the overall system operating cost, and a collaborative scheduling instruction for each entity is generated through a distributed decision-making mechanism based on graph structure.

[0084] Step 210: Define the system optimization objective function; define system operation constraints, including: charging pile power upper and lower limit constraints, vehicle battery SOC constraints, grid power flow security constraints, and user latest departure time constraints. Step 220: Employ a distributed solution method based on multi-agent deep reinforcement learning, treating each vehicle and charging pile as an agent. Its policy network takes the corresponding state-aware vector as input and outputs local control actions. Step 230: Through a graph-level collaborative training algorithm, enable the policies of each agent to co-evolve while satisfying the global optimization objective, ultimately generating a scheduling instruction set, including: the power setpoint Pset for each charging pile, the dynamic service fee rate λ, and the charging station selection suggestions and charging / discharging power plans P for each vehicle. V2G .

[0085] Preferably,

[0086]

[0087] Preferably, the distributed decision-making mechanism based on graph structure specifically involves: defining local policy networks or decision functions for the power grid nodes, charging station nodes, and vehicle nodes in the dynamic heterogeneous spatiotemporal graph; unifying the state perception vector as the decision input for each node; and using a graph-level collaborative training algorithm to jointly optimize the local policies of all nodes under the global objective of the collaborative optimization problem, thereby achieving distributed collaborative decision-making.

[0088] The graph-level collaborative training algorithm is a multi-agent graph reinforcement learning algorithm, in which each decision node is regarded as an agent, its state is the unified state perception vector, its action is the local control command, the environment model is simulated by the heterogeneous spatiotemporal graph neural network, and the reward function is associated with the objective function of the collaborative optimization problem.

[0089] The generated collaborative control instruction set includes: real-time recommended power setting values ​​and dynamic electricity price signals for each charging pile, charging and discharging power plans for each electric vehicle supporting V2G function, and grid adjustment demand signals for the charging station cluster.

[0090] Example 2, Reference Figure 2 The present invention also provides a large-scale, sensorless aggregate control system for electric vehicles for virtual power plants, characterized in that it includes:

[0091] The data acquisition module is used to collect multi-source data from vehicles, charging piles and power grids in real time. The vehicles and charging piles are set as nodes of a graph neural network, and the relationship between vehicles and charging piles is abstracted as edges of the graph neural network. The nodes and edges are attached with time sequence features.

[0092] The perception module is used to input multi-source data collected from vehicles, charging piles and the power grid into a pre-trained graph neural network model. Through the hierarchical message passing and aggregation of the model, a unified state perception vector for each node is generated, and the charging pile status is collaboratively perceived.

[0093] The decision-making module is used to construct a collaborative optimization problem with the goal of maximizing operational efficiency based on the results of collaborative perception of the charging pile status. Through a graph-based distributed decision-making mechanism, it generates a distributed collaborative control instruction set for vehicles, charging facilities, and the power grid.

[0094] This invention collects multi-source data from vehicles, charging piles, and the power grid in real time. The vehicles and charging piles are set as nodes in a graph neural network, and the relationship between them is abstracted as edges. Nodes and edges are assigned time-series features. The collected multi-source data is input into a pre-trained graph neural network model. Through hierarchical message passing and aggregation, a unified state perception vector is generated for each node, and the state of charging piles is collaboratively perceived. Based on the results of the collaborative perception of charging pile states, a collaborative optimization problem with optimal operational efficiency is constructed. Through a distributed decision-making mechanism based on a graph structure, a distributed collaborative control instruction set for vehicles, charging facilities, and the power grid is generated. By using a unified heterogeneous graph model, heterogeneous data sources such as vehicles, charging piles, and the power grid are organically integrated, achieving effective utilization of cross-domain information and providing a data foundation for global optimization. State perception accuracy is improved: utilizing the message passing mechanism of the heterogeneous graph neural network, each node can aggregate information from multi-hop neighbors, forming a deep perception of the overall system state. Its perception accuracy is more than 30% higher than that of traditional independent prediction models (verified by simulation). Achieving true collaborative scheduling: The proposed graph-based distributed decision-making mechanism ensures the autonomy of each entity while achieving global collaboration through graph learning, resolving the contradiction between the computational complexity of centralized optimization and the difficulty in coordinating distributed optimization. Enhancing system adaptability: The closed-loop feedback mechanism enables the system to respond to dynamic changes in real time, maintaining efficient operation even under emergencies such as power grid failures and traffic congestion, significantly improving system resilience. Possessing good scalability: The graph-based architecture naturally supports system scaling; adding new vehicles or charging piles only requires adding them as new nodes to the graph, without needing to reconstruct the entire system.

[0095] Example 3 provides an electronic device, including: a processor, a transmitting system, an input system, an output system, and a memory. The memory is used to store computer program code, which includes computer instructions. When the processor executes the computer instructions, the electronic device executes a vehicle-to-charging-network state perception and scheduling method based on multi-source data.

[0096] Example 4 provides a computer-readable storage medium storing a computer program, the computer program including program instructions, which, when executed by a processor of an electronic device, cause the processor to perform large-scale, sensorless aggregated control of electric vehicles oriented towards a virtual power plant.

[0097] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features claimed herein.

Claims

1. A vehicle-charging station network state-aware scheduling method based on multi-source data, characterized in that, The method includes: The system collects multi-source data on vehicles, charging piles, and the power grid in real time. The vehicles and charging piles are set as nodes in a graph neural network, and the relationship between vehicles and charging piles is abstracted as edges in the graph neural network. The nodes and edges are attached with time-series features. The collected multi-source data from vehicles, charging piles, and the power grid are input into a pre-trained graph neural network model. Through the hierarchical message passing and aggregation of the model, a unified state perception vector for each node is generated, and the state of the charging piles is collaboratively perceived. Based on the results of collaborative perception of the charging pile status, a collaborative optimization problem with the goal of maximizing operational efficiency is constructed. Through a distributed decision-making mechanism based on a graph structure, a distributed collaborative control instruction set for vehicles, charging facilities, and the power grid is generated.

2. The vehicle-charging station network state awareness and scheduling method based on multi-source data according to claim 1, characterized in that, The real-time collected multi-source data from vehicles, charging piles, and the power grid includes: Vehicle node data: real-time location, battery state of charge (SOC), driving speed, destination, planned stay time, vehicle type, and user charging preferences; Charging pile node data: real-time power, voltage / current, operating status, affiliated charging station, rated power, service fee rate, and historical fault records; Grid-related data: real-time load factor of the local transformer area, marginal electricity price at nodes, grid dispatch instructions, and renewable energy output forecasts; Spatiotemporal environmental data: local traffic congestion index, weather conditions, temperature affecting charging efficiency, and time period.

3. The vehicle-charging station network state awareness and scheduling method based on multi-source data according to claim 1, characterized in that, The heterogeneous graph convolution module uses a relation-specific weight matrix to transmit differentiated information for different types of edges. For node vi, its representation calculation at layer l+1 is as follows: ； Where, N r i W represents the set of neighbors of node vi under relation r. (l) r Let σ be the learnable weight matrix corresponding to relation r, and σ be the activation function. The spatiotemporal attention module includes two sub-modules: spatial attention and temporal attention. Spatial attention calculates the correlation weights between node pairs, while temporal attention captures the temporal dependencies of node features. The state decoding module maps the node representation after multi-layer aggregation into a specific state perception vector. For a charging pile node, the vector includes at least: real-time load rate, health score, competition intensity index, and demand forecast value.

4. The vehicle-charging station network state awareness and scheduling method based on multi-source data according to claim 1, characterized in that, Using a unified state perception vector generated by state collaborative perception based on heterogeneous graph neural networks as input, an optimization problem is constructed with the goal of minimizing the overall system operating cost. Through a distributed decision-making mechanism based on graph structure, collaborative scheduling instructions for each entity are generated. Define the system optimization objective function; Define system operation constraints, including: upper and lower limits of charging pile power, vehicle battery SOC constraints, power grid flow security constraints, and user latest departure time constraints; A distributed solution method based on multi-agent deep reinforcement learning is adopted, in which each vehicle and charging pile is regarded as an agent, and its policy network takes the corresponding state perception vector as input and outputs local control actions. Through a graph-level collaborative training algorithm, the policies of each agent evolve collaboratively while satisfying the global optimization objective, ultimately generating a scheduling instruction set, including: the power setpoint Pset for each charging pile, the dynamic service fee rate λ, and the charging station selection suggestions and charging / discharging power plans P for each vehicle. V2G .

5. The vehicle-charging station network state awareness scheduling method based on multi-source data according to claim 4, characterized in that, 。 6. The vehicle-charging station network state awareness scheduling method based on multi-source data according to claim 1, characterized in that: The graph-based distributed decision-making mechanism specifically involves defining local policy networks or decision functions for the power grid nodes, charging station nodes, and vehicle nodes in the dynamic heterogeneous spatiotemporal graph; unifying the state perception vector as the decision input for each node; and using a graph-level collaborative training algorithm to jointly optimize the local policies of all nodes under the global objective of the collaborative optimization problem, thereby achieving distributed collaborative decision-making. The graph-level collaborative training algorithm is a multi-agent graph reinforcement learning algorithm, in which each decision node is regarded as an agent, its state is the unified state perception vector, its action is the local control command, the environment model is simulated by the heterogeneous spatiotemporal graph neural network, and the reward function is associated with the objective function of the collaborative optimization problem. The generated collaborative control instruction set includes: real-time recommended power setting values ​​and dynamic electricity price signals for each charging pile, charging and discharging power plans for each electric vehicle supporting V2G function, and grid adjustment demand signals for the charging station cluster.

7. A large-scale, sensorless aggregate control system for electric vehicles oriented towards virtual power plants, characterized in that, include: The data acquisition module is used to collect multi-source data from vehicles, charging piles and power grids in real time. The vehicles and charging piles are set as nodes of a graph neural network, and the relationship between vehicles and charging piles is abstracted as edges of the graph neural network. The nodes and edges are attached with time sequence features. The perception module is used to input multi-source data collected from vehicles, charging piles and the power grid into a pre-trained graph neural network model. Through the hierarchical message passing and aggregation of the model, a unified state perception vector for each node is generated, and the charging pile status is collaboratively perceived. The decision-making module is used to construct a collaborative optimization problem with the goal of maximizing operational efficiency based on the results of collaborative perception of the charging pile status. Through a graph-based distributed decision-making mechanism, it generates a distributed collaborative control instruction set for vehicles, charging facilities, and the power grid.

8. An electronic device, characterized in that, include: The electronic device comprises a processor, a transmitting system, an input system, an output system, and a memory, wherein the memory is used to store computer program code, the computer program code including computer instructions, and when the processor executes the computer instructions, the electronic device executes the vehicle-charging-network state perception and scheduling method based on multi-source data as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which includes program instructions that, when executed by a processor of an electronic device, cause the processor to perform the vehicle-charging-network state perception and scheduling method based on multi-source data as described in any one of claims 1 to 6.

Images