Electric vehicle charging management method, device and storage medium

Abstract

The application relates to the field of new energy automobile power supply, and provides an electric automobile charging management method, equipment and a storage medium. The method comprises the following steps: receiving real-time operation data from multiple charging terminals at an edge computing node; generating an initial charging strategy for each charging terminal based on the real-time operation data received by the edge computing node by using a first artificial intelligence model on the edge computing node; performing local optimization on the initial charging strategy by using a second artificial intelligence model deployed locally at the charging terminal, generating a final charging strategy and executing the final charging strategy; when it is monitored that power grid parameter fluctuation exceeds a predetermined safety threshold or the total load of a charging terminal cluster in a region approaches an upper limit of transformer capacity, starting a distributed collaborative decision mechanism among associated multiple edge computing nodes, dynamically negotiating and reallocating charging power of each charging terminal; and uploading key operation logs and decision data to a cloud end block chain node through an encrypted channel for safe notarization.

Description

Technical Field

[0001] This application relates to the field of power supply for new energy vehicles, and in particular to a method, device and storage medium for electric vehicle charging management. Background Technology

[0002] With the rapid development of the new energy vehicle industry, the number of electric vehicles on the road continues to grow, leading to an ever-increasing demand for charging infrastructure. Large-scale, unregulated charging of electric vehicles can put enormous pressure on regional power distribution networks, potentially causing a series of problems such as localized grid overload, voltage fluctuations, and degraded power quality. Therefore, how to efficiently and intelligently manage large-scale electric vehicle charging to meet user charging needs while ensuring the safe and stable operation of the power grid has become an important research direction in the fields of smart grids and the Internet of Things (IoT).

[0003] Currently, most existing electric vehicle charging management solutions employ centralized control architectures or simple local control strategies. Centralized control architectures typically upload operational data from all charging terminals to a cloud center for unified decision-making before issuing scheduling commands. While this architecture enables global optimization, the remote transmission and centralized processing of massive amounts of data introduces significant communication and computational delays, making it difficult to handle millisecond-level rapid changes in grid conditions and posing risks of decision lag and single points of failure. On the other hand, purely local control strategies (e.g., start-stop control based on fixed thresholds), while responsive, lack group collaboration and a global perspective. Each charging terminal operates independently, failing to create a synergistic effect to address global constraints such as transformer overload. Their optimization objectives are singular, making it difficult to achieve a balance between multiple objectives such as vehicle battery health, economic costs, and grid demand.

[0004] Therefore, the existing technology has the following main drawbacks: First, the response speed of centralized decision-making is slow, making it difficult to meet the timeliness requirements of real-time power grid regulation; second, the control strategy fails to effectively integrate multiple factors such as power grid operating status, user-specific needs (e.g., vehicle battery life protection) and real-time electricity prices, making it impossible to achieve refined multi-objective optimization; third, there is a lack of a fast, decentralized collaborative decision-making mechanism at the edge to deal with local power grid emergencies (e.g., power fluctuations, capacity approaching the limit, etc.). Summary of the Invention

[0005] This application provides an electric vehicle charging management method, device, and storage medium, which can achieve millisecond-level response, multi-objective collaborative optimization, and rapid collaborative handling of grid faults during electric vehicle charging.

[0006] On the one hand, this application provides an electric vehicle charging management method, the method comprising:

[0007] The edge computing node receives real-time operational data from multiple charging terminals.

[0008] Using a first artificial intelligence model deployed on the edge computing node, an initial charging strategy for each charging terminal is generated within a millisecond response time based on the real-time running data. The first artificial intelligence model is a prediction model based on the spatiotemporal graph convolutional network ST-GNN.

[0009] The initial charging strategy is optimized locally using a second artificial intelligence model deployed on the charging terminal to generate and execute the final charging strategy. The second artificial intelligence model is a decision model built with a deep reinforcement learning (DRL) algorithm.

[0010] When the power grid parameter fluctuations exceed the predetermined safety threshold or the total load of the charging terminal cluster in the area approaches the upper limit of the transformer capacity, a distributed collaborative decision-making mechanism is initiated among multiple associated edge computing nodes to dynamically negotiate and redistribute the charging power of each charging terminal.

[0011] Key operation logs and decision data during the charging process are uploaded to a cloud blockchain node via an encrypted channel for secure storage.

[0012] On the other hand, this application provides an electric vehicle charging management device, the device comprising:

[0013] The receiving module is used to receive real-time operating data from multiple charging terminals at the edge computing node;

[0014] The first strategy generation module is used to generate an initial charging strategy for each charging terminal within a millisecond-level response time by using a first artificial intelligence model deployed on the edge computing node and based on the real-time running data. The first artificial intelligence model is a prediction model based on the spatiotemporal graph convolutional network ST-GNN.

[0015] The second strategy generation module is used to optimize the initial charging strategy locally using a second artificial intelligence model deployed on the charging terminal, generate the final charging strategy and execute it. The second artificial intelligence model is a decision model built with a deep reinforcement learning (DRL) algorithm.

[0016] The collaboration module is used to initiate a distributed collaborative decision-making mechanism among multiple associated edge computing nodes when the power grid parameter fluctuation exceeds the predetermined safety threshold or the total load of the charging terminal cluster in the area approaches the upper limit of the transformer capacity, so as to dynamically negotiate and redistribute the charging power of each charging terminal.

[0017] The upload module is used to upload key operation logs and decision data during the charging process to the cloud blockchain node for secure storage via an encrypted channel.

[0018] Thirdly, this application provides an electronic device, the device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the technical solution of the electric vehicle charging management method described above.

[0019] Fourthly, this application provides a storage medium storing a computer program, which, when executed by a processor, implements the steps of the above-described electric vehicle charging management method.

[0020] As can be seen from the technical solution provided in this application, on the one hand, by deploying a spatiotemporal graph convolutional network model based on millisecond-level policy generation at edge computing nodes close to the data source, the latency of remote data transmission to the cloud can be effectively reduced, enabling the system to react instantly to rapid changes in the power grid state, greatly improving the timeliness of control and ensuring the real-time safety and stability of the power grid; on the other hand, the spatiotemporal graph convolutional network model at the edge nodes focuses on the global preliminary optimization of regional power grid safety and efficiency, while the local decision-making model of the charging terminal further integrates personalized factors such as vehicle battery health status and real-time electricity price for in-depth optimization, so that the final executed strategy is no longer... It is not a compromise of a single objective, but a truly multi-objective collaborative optimal solution, thereby simultaneously ensuring grid security, user economy, and vehicle battery life. Thirdly, by constructing a distributed collaborative decision-making mechanism among edge computing nodes, when grid parameters exceed limits or load approaches capacity, multiple edge nodes can autonomously and quickly exchange and negotiate information without relying on the cloud center, ultimately reaching a consistent power allocation scheme that satisfies global constraints (e.g., total transformer capacity) and maximizes overall satisfaction. This decentralized collaborative approach not only avoids the risk of single-point failures but also forms a kind of "collective intelligence," jointly maintaining the stable operation of the regional power grid. In summary, the technical solution of this application achieves millisecond-level response for electric vehicle charging, multi-objective collaborative optimization, and rapid collaborative handling of grid faults. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a flowchart of the electric vehicle charging management method provided in the embodiments of this application;

[0023] Figure 2This is a schematic diagram of the structure of the electric vehicle charging management device provided in the embodiments of this application;

[0024] Figure 3 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0025] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0026] In this specification, adjectives such as "first" and "second" are used only to distinguish one element or action from another, without necessarily requiring or implying any actual such relationship or order. Where circumstances permit, reference to an element or component or step (etc.) should not be construed as being limited to only one of the elements, components, or steps, but may be one or more of the elements, components, or steps, etc.

[0027] For ease of description, the dimensions of the various parts shown in the accompanying drawings are not drawn to actual scale.

[0028] Existing electric vehicle charging management solutions mostly employ centralized control architectures or simple local control strategies. Centralized control architectures typically upload operational data from all charging terminals to a cloud center for unified decision-making before issuing scheduling commands. While this architecture enables global optimization, the remote transmission and centralized processing of massive amounts of data introduces significant communication and computational delays, making it difficult to cope with millisecond-level rapid changes in grid conditions and posing risks of decision lag and single points of failure. On the other hand, purely local control strategies (e.g., start-stop control based on fixed thresholds), while responsive, lack group collaboration and a global perspective. Each charging terminal operates independently, failing to create a synergistic effect to address global constraints such as transformer overload. Their optimization objectives are singular, making it difficult to achieve a balance between multiple objectives such as vehicle battery health, economic costs, and grid demand. Therefore, the existing technology has the following main drawbacks: First, the response speed of centralized decision-making is slow, making it difficult to meet the timeliness requirements of real-time power grid regulation; second, the control strategy fails to effectively integrate multiple factors such as power grid operating status, user-specific needs (e.g., vehicle battery life protection) and real-time electricity prices, making it impossible to achieve refined multi-objective optimization; third, there is a lack of a fast, decentralized collaborative decision-making mechanism at the edge to deal with local power grid emergencies (e.g., power fluctuations, capacity approaching the limit, etc.).

[0029] To address the aforementioned problems in the prior art, this application proposes an electric vehicle charging management method, the flowchart of which is attached. Figure 1 As shown, the main steps include S101 to S105, which are detailed below:

[0030] Step S101: Receive real-time operating data from multiple charging terminals at the edge computing node.

[0031] If the real-time operating data of the charging terminals is directly uploaded to the cloud center for processing, not only will the data transmission link be long, but network latency and jitter will also be uncontrollable, failing to meet the real-time requirements of millisecond-level power grid control. Furthermore, aggregating all the data will place enormous pressure on the core network bandwidth. To reduce data transmission distance and latency, laying the foundation for subsequent millisecond-level response, and thus resolving the contradiction between the high latency and high bandwidth consumption caused by centralized cloud processing of massive terminal data and the real-time requirements of charging control, this application allows edge computing nodes to receive real-time operating data from multiple charging terminals. This real-time operating data includes at least vehicle battery status information, power grid operating parameters, and environmental status information, etc. Environmental status information can include environmental data such as light intensity and wind speed, while vehicle battery status information includes the battery temperature of the vehicle being charged using the charging terminal. and the current loop count It should be noted that the vehicles mentioned in this application refer to vehicles that are charged using a charging terminal.

[0032] Step S102: Using the first artificial intelligence model deployed on the edge computing node, an initial charging strategy for each charging terminal is generated within a millisecond-level response time based on the real-time running data received by the edge computing node. The first artificial intelligence model is a prediction model based on the spatiotemporal graph convolutional network ST-GNN.

[0033] If traditional mathematical programming models (e.g., linear programming) are used at edge nodes as in existing technologies, the computation is complex, the solution is time-consuming, and it is difficult to complete in a short time, making it unable to effectively handle uncertainty and predict future states. If simple rules or single-terminal models are used, the mutual influence between terminals through the power grid cannot be captured (e.g., an increase in the charging power of charging port A leads to a decrease in the voltage of charging port B), resulting in a lack of coordination in decision-making and easily leading to suboptimal solutions where each terminal acts independently. In other words, traditional simple control strategies or cloud optimization models cannot model complex spatiotemporal relationships, while charging behavior is a strongly spatiotemporally coupled behavior. To address this contradiction, the solution of this application is to utilize a first artificial intelligence model deployed on edge computing nodes. Based on the real-time running data received by the edge computing nodes, it generates an initial charging strategy for each charging terminal within a millisecond response time. The first artificial intelligence model is a prediction model based on a Spatial-Temporal Graph Convolutional Network (ST-GNN). The ST-GNN-based prediction model is the core computing engine for achieving regional optimization and millisecond-level response. It is specifically designed to process spatiotemporal data and can dynamically capture the spatial correlation and temporal evolution patterns between terminals, thereby generating collaborative and forward-looking optimization strategies.

[0034] In one embodiment of this application, the first artificial intelligence model (i.e., the prediction model based on ST-GNN) treats the charging terminal as a node in a graph and dynamically constructs a graph structure with grid parameters and environmental conditions as edge features. This allows for the simultaneous prediction of the optimal charging power probability distribution of all charging terminals within a short-term time window in the future. Specifically, the dynamic construction of the graph structure with grid parameters and environmental conditions as edge features is based on the physical distance between charging terminals and the grid connection topology to construct a static spatial adjacency matrix. The line power flow value in the real-time collected grid operation parameters and the temperature value in the environmental condition information are used as dynamic edge weights to weight and update the static spatial adjacency matrix, forming a dynamic spatiotemporal graph structure. Accordingly, utilizing the first artificial intelligence model deployed on edge computing nodes, and based on the real-time operational data received by the edge computing nodes, the initial charging strategy for each charging terminal can be generated within a millisecond response time as follows: The dynamic spatiotemporal graph structure generated during the dynamic graph structure construction is input into the spatial graph convolution module, which integrates a multi-head attention mechanism to extract and output the spatial feature sequence formed by the nonlinear energy flow interaction between charging terminals; the spatial feature sequence formed by the nonlinear energy flow interaction is input into the temporal convolutional network (TCN) module to extract and output its long-term dependency features in the time dimension; the long-term dependency features are input into the output layer of the hybrid density network (MDN), which calculates and outputs the Gaussian mixture model parameters characterizing the probability distribution of future charging power for each charging terminal.

[0035] Step S103: Using a second artificial intelligence model deployed locally on the charging terminal, the initial charging strategy generated by the first artificial intelligence model is locally optimized to generate the final charging strategy and execute it. The second artificial intelligence model is a decision model constructed using the deep reinforcement learning (DRL) algorithm.

[0036] If, as in existing technologies, the charging terminal unconditionally executes the instructions of the edge computing node, it cannot adapt to individual differences among charging terminals (e.g., vehicle battery health, user cost preferences), resulting in a lack of flexibility in the strategy and potentially harming the battery life of the vehicle or the user's economic benefits. To resolve the contradiction between the unified strategy issued by the edge computing node and the personalized conditions and needs of each charging terminal, in one embodiment of this application, the reward function R of the second artificial intelligence model (i.e., the decision model constructed using the deep reinforcement learning (DRL) algorithm) is designed as follows: ,in, and These represent the health status of the vehicle battery at time t and time t-1, respectively. The initial policy power issued to the edge computing nodes. This is the actual adjusted power. For real-time electricity price costs, , and These are weighting coefficients; localized optimization must at least consider the vehicle's battery health status characteristics. It should be noted that these weighting coefficients... , and The settings can be dynamically adjusted. Specifically, the adjustment process involves receiving user preference selection instructions submitted through the charging terminal's human-machine interface. These instructions may include "vehicle battery life priority," "economy priority," or "fast charging priority" modes. Based on the user's preference selection instructions submitted through the charging terminal's human-machine interface, a preset weight configuration template is invoked to adjust the settings accordingly. , and The values ​​can be set once or dynamically adjusted periodically. The decision model built with the deep reinforcement learning (DRL) algorithm, through its reward function, finely balances multiple objectives such as vehicle battery health, execution bias, and economic cost locally, and performs personalized fine-tuning of the global strategy, thereby solving the adaptation problem of the "last mile".

[0037] Furthermore, during localization optimization, the vehicle battery health state term (SOH) in the reward function R of the second AI model is dynamically corrected using the following formula: ,in, and The attenuation coefficient is denoted by , and the optimal operating temperature is denoted by , thus enabling refined lifespan management based on real-time vehicle battery conditions.

[0038] As mentioned earlier, the second artificial intelligence model deployed locally on the charging terminal can perform localized optimization of the initial charging strategy generated by the first artificial intelligence model. This optimization process can be as follows: anonymized user preference selection data and corresponding charging strategy execution effect data are uploaded to edge computing nodes; the edge computing nodes aggregate preference selection data from multiple users within the region and train a user macro-preference transfer model locally using a federated learning framework; the model gradients obtained during the training of the user macro-preference transfer model are uploaded to a cloud server; the cloud server securely aggregates and updates the model gradients from multiple edge computing nodes to generate global model parameters; and the updated global model parameters of the user macro-preference transfer model are distributed to each edge computing node to optimize the initial strategy generation logic of the first artificial intelligence model.

[0039] It should be noted that, in the above embodiments, the user macro-preference transfer model is an encoder-decoder architecture based on a deep neural network, specifically including the following layers and modules:

[0040] 1) Input layer: Receives anonymized vectorized input data, which consists of two parts: the first part is a one-hot encoding vector of the user preference selection instruction (e.g., [1,0,0] represents "battery life priority"); the second part is a normalized feature vector corresponding to the execution effect of the charging strategy, which includes at least the quantified values ​​of the change in battery health status, actual charging cost, and deviation from the expected charging time.

[0041] 2) Feature encoder: It consists of two fully connected layers and is used to map high-dimensional input data to a low-dimensional "macro-preference latent space". The latent space vector is intended to abstractly represent the comprehensive optimization tendency corresponding to a certain type of user preference pattern.

[0042] 3) Context fusion module: Receives the latent space vector output from the feature encoder and concatenates it with the regional real-time context features extracted from the edge computing nodes (e.g., current grid average load rate, regional real-time electricity price, time period encoding). This module learns the influence of context information on preference expression through a fully connected layer.

[0043] 4) Policy Influence Decoder: Consists of three fully connected layers. It takes the output of the context fusion module as input and aims to predict policy adjustment parameters to address macro-level preferences. The strategy modifies the parameters. It will be used to fine-tune the weights of the last layer of the first artificial intelligence model (ST-GNN), thereby affecting its initial policy generation logic.

[0044] 5) Output layer: Output strategy correction parameters .

[0045] The above user macro-preference transfer model is trained through the following steps S1 to S4:

[0046] Step S1: At the edge computing node, the collected triples (user preference data, policy execution effect data, and regional context data) are used to form a local training dataset.

[0047] Step S2: Input the data into the model to obtain the predicted policy correction parameters. ,Will A copy of the first artificial intelligence model (ST-GNN) is applied, and historical data is used to re-infer the initial charging strategy of the simulation, defining a composite loss function. ,in, and It's a hyperparameter. It is the task loss, calculated in the application. Subsequently, the mean squared error between the initial policy generated by the ST-GNN model and the ideal baseline policy recalculated offline based on the global optimal objective (without considering preferences) is... It is the L2 regularization loss, used to prevent policy adjustments during prediction. The value is too large in order to maintain model stability.

[0048] Step S3: Calculate the loss function L for all model parameters. gradient The Adam optimizer is used, and mini-batch stochastic gradient descent is performed on the local dataset to update the local model parameters. In order to minimize the loss L.

[0049] Step S4: After completing one round of local training, calculate the gradient of the model parameters. (or the difference between the locally updated parameters and the previous round's global parameters, i.e.) Encrypt it and prepare to upload.

[0050] In the above embodiments, the cloud server securely aggregates and updates model gradients from multiple edge computing nodes to generate global model parameters. Specifically, the cloud server securely receives homomorphically encrypted model gradient data packets (each packet contains encrypted gradient information, edge node digital signatures, timestamps, etc.) from multiple edge computing nodes, verifies the validity and timeliness of each packet's signature, and marks and isolates packets that fail verification. For verified encrypted gradient data packets, the cloud server performs batch decryption using the corresponding private key to restore the plaintext model gradients or model parameter updates obtained from local training on each edge computing node. It checks whether the dimensions of all decrypted gradient tensors are aligned with the parameter dimensions of the current global model, discarding any misaligned gradients. An aggregation weight is assigned to each edge computing node providing a valid gradient; this aggregation weight is dynamically calculated based on the size of its local training dataset and the stability of its historical gradient contributions. The weighted average global gradient update is calculated using the following formula: ,in, Let be the gradient or parameter update amount after decryption for the i-th edge node. The aggregation weights are assigned to these weights. During this aggregation process, a differential privacy noise mechanism is introduced to add small random noise conforming to a Gaussian distribution to the aggregation result, further enhancing privacy and security. The weighted average global gradient update is then applied. Using global learning rate Update the global parameters of the user macro-preference transfer model according to the following formula. : Update the global model parameters The data is re-encrypted and securely distributed to all edge computing nodes participating in this round of training. At the same time, instructions containing the configuration of the new round of training tasks (e.g., number of training rounds, batch size) are sent to each edge node to trigger the next round of federated learning iteration.

[0051] On the one hand, considering that the second artificial intelligence model (i.e., the decision-making model built with the deep reinforcement learning (DRL) algorithm) makes decisions based on historical data and simulated environments during training after deployment, however, in actual operation, the degradation characteristics of vehicle batteries, usage environments (e.g., temperature), and user habits are all unique and time-varying. A fixed model cannot accurately predict the state of health (SOH) changes of all vehicle batteries under future real-world conditions, causing its optimization decisions (adjusting power to protect the vehicle battery) to gradually deviate from the optimal, or even produce counterproductive effects. On the other hand, existing solutions typically update the model by periodically and centrally retraining it, but this approach is slow, costly, and unable to adapt to the subtle, personalized changes of each charging terminal. In other words, the second artificial intelligence model may experience model prediction drift and mismatch with reality after long-term operation. Therefore, to address the challenge of enabling each charging terminal to autonomously and online detect its own decision-making biases and immediately adjust itself, allowing the second artificial intelligence model to continuously align with the actual state of the specific vehicle battery it serves, the above embodiment further includes the following online adaptive step after executing the final charging strategy: the charging terminal's local monitor continuously collects data on voltage, current, and internal resistance changes in the vehicle battery, inputs this data into a vehicle battery state observer based on an extended Kalman filter (EKF), and estimates the actual change in the vehicle battery's state of health (SOH) in real time with high accuracy. ;Will The amount of change predicted by the second artificial intelligence model when making decisions The system compares and calculates the policy execution deviation. Based on this deviation, a policy gradient algorithm is used to fine-tune the decision network parameters of the second AI model online, narrowing the gap between prediction and reality. By using an EKF-based observer to estimate the actual SOH change in real time with high precision and comparing it with the predictions of the second AI model, the model can continuously self-correct. Its charging power decisions become increasingly aligned with the actual health status of the specific vehicle's battery, thus more effectively extending battery life and achieving truly personalized and refined vehicle battery management. Furthermore, this online adaptive mechanism transforms each charging terminal from a static policy executor into an intelligent agent with continuous learning capabilities. It can proactively adapt to battery aging and environmental changes, ensuring the long-term effectiveness of the optimization strategy and significantly reducing maintenance costs associated with model expiration requiring manual intervention or global retraining.

[0052] Furthermore, while the aforementioned online adaptive mechanism makes each charging terminal an independent learning unit, accumulating valuable personalized experience, if this experience remains only local, the first AI model deployed on edge computing nodes, responsible for generating the initial strategy, cannot benefit from this scattered knowledge, cannot perceive the overall changing trend of the group of vehicle batteries, and its generated global strategy may gradually become outdated. In addition, directly uploading users' raw vehicle battery data to retrain the first AI model poses a serious risk of privacy leakage and data transmission burden. This means that how to enable the global model to learn and evolve from the individual terminal experiences while protecting user privacy is a key challenge. To address these issues, the aforementioned online adaptive mechanism can also introduce the following federated learning mechanism: each charging terminal increments the model parameters generated by this online fine-tuning. Encryption and desensitization are performed; the model parameters after encryption and desensitization are incremented. The data is uploaded to an edge computing node. The edge computing node uses a secure aggregation algorithm to aggregate model parameter increments from multiple charging terminals, which are then used to update the first artificial intelligence model deployed on that edge computing node, forming a terminal-edge co-evolution mechanism. As can be seen from the above embodiment, by introducing a federated learning mechanism, the local learning results (model parameter increments) of numerous charging terminals are securely aggregated without directly sharing original sensitive data, and used to update the first artificial intelligence model. This allows the global model to learn the latest knowledge about vehicle battery health dynamics scattered across various terminals, thereby generating a more accurate and forward-looking initial charging strategy. This forms a knowledge uplink channel from the charging terminal to the edge computing node. Through encryption, desensitization, and secure aggregation technologies, knowledge sharing is achieved while strictly protecting the privacy of users' vehicle battery data, meeting data compliance requirements, and clearing obstacles for the commercial deployment of the technology.

[0053] In the above embodiments, the edge computing node uses a secure aggregation algorithm to aggregate model parameter increments from multiple charging terminals to update the first artificial intelligence model deployed on the edge computing node. The terminal-edge co-evolution mechanism can be implemented through steps S1031 to S1035, as detailed below:

[0054] Step S1031: The edge computing node receives encrypted data packets uploaded from multiple charging terminals within its jurisdiction, verifies the source legitimacy and data integrity of each data packet, and isolates data packets that fail verification. Each data packet contains a second artificial intelligence model parameter increment generated by the corresponding terminal and homomorphically encrypted. And the associated integrity check code.

[0055] Step S1032: For all verified encrypted data packets, the edge computing node executes a secure aggregation algorithm in ciphertext state, which increments each valid parameter. Assign an aggregate weight Calculate the aggregate parameter increment after weighted average Its encrypted form The following formula is obtained by operating within the cryptographic domain: Among them, aggregate weight The value is dynamically determined based on the historical data quality score of the corresponding charging terminal and the size of the current increment. and These represent the ciphertext addition and scalar multiplication operations defined under the selected homomorphic encryption scheme, respectively.

[0056] Step S1033: The edge computing node uses its private key to aggregate the ciphertext result. Decryption is performed to obtain the aggregate parameter increment of the plaintext. To improve robustness and further prevent potential privacy breaches, Add Gaussian noise that meets differential privacy requirements This generates the final, noisy aggregate increment that can be used for model updates. .

[0057] Step S1034: The edge computing node obtains the parameters of the currently deployed first artificial intelligence model (ST-GNN). Noise-enhanced polymerization increment The specific update formula applied to the model parameters is as follows: ,in, Update the learning rate for the model on the edge side.

[0058] Step S1035: Using a small portion of the reserved validation data, update the first artificial intelligence model. Perform forward inference to evaluate the predictive performance of the initial charging strategy. If the performance evaluation metric is better than or at least not worse than the model before the update, then... The system will be officially deployed to replace the existing model and will be used to generate initial charging strategies for all charging terminals.

[0059] Step S104: When it is detected that the fluctuation of the power grid parameters exceeds the predetermined safety threshold or the total load of the charging terminal cluster in the area approaches the upper limit of the transformer capacity, a distributed collaborative decision-making mechanism is initiated among multiple associated edge computing nodes to dynamically negotiate and redistribute the charging power of each charging terminal.

[0060] The inherent contradiction between the single point of failure and latency risks of centralized cloud coordination and the high reliability and ultra-fast coordination requirements of the power grid in emergency situations is a fundamental one. If, as with existing technologies, the cloud center recalculates and issues commands in emergency situations, the response will be slow, and the entire system will fail if the cloud or communication link fails, posing a very high risk. To ensure system stability under extreme conditions through critical redundancy design, this application employs a distributed collaborative decision-making mechanism among multiple associated edge computing nodes when grid parameter fluctuations exceed a predetermined safety threshold or the total load of the charging terminal cluster in the area approaches the transformer capacity limit. This mechanism dynamically negotiates and redistributes the charging power of each charging terminal.

[0061] Specifically, the distributed collaborative decision-making mechanism in the above embodiment can be as follows: each edge computing node iteratively exchanges information on the total load and adjustable margin of the charging terminals it manages, based on a consensus algorithm, until all nodes reach a consensus on the global power allocation scheme. The consensus scheme satisfies the total transformer capacity constraint and maximizes the overall satisfaction of all charging terminals. Maximizing the overall satisfaction of all charging terminals can be achieved through the following process: determining the satisfaction function of each charging terminal, wherein the satisfaction function adopts a logarithmic function form and assigns dynamic priority weights to each charging terminal. The satisfaction of each terminal increases with the increase of the allocated charging power, but the rate of increase decreases. During the iterative process of the distributed collaborative decision-making mechanism, the goal is to maximize the sum of the satisfaction of all charging terminals, and the optimal power allocation scheme is dynamically solved with the upper limit of the total transformer capacity and the upper and lower limits of the allowed power of each charging terminal as constraints. The dynamic priority weights can be determined through the following process: based on the user's charging service contract level, the urgency of the vehicle's battery health status (SOH), and the current grid incentive price signal, the weight of each user is dynamically calculated through a fuzzy logic reasoning system to achieve a balance between fairness and efficiency. In the iterative process of the distributed collaborative decision-making mechanism, with the goal of maximizing the sum of satisfaction levels of all charging terminals and with constraints such as the upper limit of the total transformer capacity and the upper and lower limits of the allowable power of each charging terminal, the implementation method of dynamically solving the optimal power allocation scheme is as follows: Steps S1 to S4:

[0062] Step S1: Each edge computing node defines a satisfaction function for each charging terminal i it manages. The function satisfies: with the allocated charging power It increases strictly and incrementally, but its first derivative decreases (i.e., ... and At the same time, initialize the power allocation scheme for all charging terminals. To ensure that it meets the lower power limit of each terminal. and upper limit Local constraints;

[0063] Step S2: In each consensus iteration round k, each edge computing node performs the following operations: based on the current power allocation scheme Calculate the sum of local satisfaction levels for all charging terminals it manages. ,in, This represents the set of charging terminals managed by this node. The gradient of the sum of local satisfaction levels with respect to the power of each terminal is calculated. Generate an intended power adjustment along the gradient ascent direction. This makes the power after the intention adjustment It can improve local satisfaction;

[0064] Step S3: Each edge computing node adjusts its intended total load. and the margin for adjustment As part of the state vector, iterative exchanges and negotiations are conducted with neighboring nodes according to the aforementioned consensus protocol, with the goal of reaching a global total load distribution among all nodes. The consensus, and satisfy The total transformer capacity constraint is based on the total load quota determined through negotiation. Each node will submit its intended power adjustment plan. Projecting to a new feasible solution The solution must satisfy the following conditions. and Furthermore, among all solutions that satisfy these two conditions, the local satisfaction level corresponding to this solution is... Highest;

[0065] Step S4: Calculate the overall satisfaction of all charging terminals after this iteration. To determine whether the convergence condition is met: the increment of overall satisfaction. Less than the preset threshold And all edge computing nodes are allocated their total load quotas. If consensus is reached (i.e., the state vector converges), and if not, let k = k + 1, return to step S2 for the next iteration; if consensus is reached, output the current power allocation scheme. This is the optimal power allocation scheme that satisfies global constraints and maximizes overall satisfaction.

[0066] The consensus algorithm described above enables edge nodes to quickly achieve global consensus and dynamically allocate power by exchanging limited information with their neighbors, without relying on the center.

[0067] As for the triggering conditions for activating the distributed collaborative decision-making mechanism, they can be determined according to the following steps: continuously sample the power grid frequency signal and calculate its standard deviation within a sliding time window. ;like If the first threshold is exceeded, a primary warning is activated, and the data sampling frequency is increased; if, under the primary warning state, the first derivative of the frequency... Continuously exceeding the second threshold achieve If the fluctuation exceeds a predetermined safety threshold within a few seconds, the distributed collaborative decision-making mechanism is immediately triggered.

[0068] While the consensus algorithm of the aforementioned distributed collaborative decision-making mechanism enables edge nodes to quickly achieve global consensus and dynamically allocate power without relying on a central authority, simply by exchanging limited information with their neighbors, the mechanism faces two core technical challenges in its implementation: efficiency and convergence, and real-time satisfaction of physical constraints. Efficiency and convergence refer to designing an efficient and reliable iterative mechanism during distributed negotiation among multiple edge nodes to ensure that all nodes can quickly reach consensus, rather than getting stuck in infinite loops or oscillations. Real-time satisfaction of physical constraints refers to ensuring that intermediate results meet local physical limitations at each step of the iterative negotiation, avoiding the generation of theoretically feasible but practically unenforceable power allocation schemes, thereby guaranteeing the security and feasibility of the decision-making process. To address the two core technical challenges mentioned above in the implementation of distributed collaborative decision-making mechanisms, one existing approach is centralized optimization followed by broadcasting. This involves performing centralized optimization calculations at a master edge node or in the cloud, deriving a global allocation scheme, and then broadcasting it directly to other nodes for execution. However, this approach completely contradicts the original intent of distributed and collaborative systems, reintroducing single-point failure risks and communication latency bottlenecks. If the master node or its communication link fails, the entire collaborative mechanism will collapse, failing to meet high reliability requirements. Another existing approach is a simple average consensus protocol. This uses a standard average consensus algorithm where all nodes repeatedly exchange information and average their state values ​​until convergence. While this algorithm can achieve consensus, the consensus value may be an "average," failing to guarantee that intermediate results after each iteration satisfy the local physical constraints of each node. For example, during iteration, a node's calculated power allocation value might momentarily exceed its actual capacity, which is dangerous and unacceptable in a real power grid. It only cares about the final result, not the safety of the process. To provide a safe, efficient, and reliable mathematical tool for solving the core challenges in distributed negotiation, this application proposes that each edge computing node, based on a consensus algorithm, iteratively exchange information on the total load and adjustable margin of the charging terminals it manages until all nodes reach a consensus on the global power allocation scheme. This can be achieved by each edge computing node sharing its state vector... Broadcast to its communication neighbor nodes, among which, For total load, To allow for adjustable power margin, each node receives the state vectors of all its neighbors and updates its own state based on the following consensus protocol: ,in, It is the consensus weight. It is a distributed control law designed based on local constraints to ensure that the result after each iteration satisfies the local physical limitations. This iterative negotiation scheme based on a consensus algorithm, on the one hand, explicitly defines the adjustable power margin. Incorporating state vectors and introducing distributed control laws designed based on local constraints into the consensus protocol. This allows the edge computing nodes to proactively adjust their states to within their physical limitations after each iteration, ensuring that every intermediate state during the negotiation process is safe and executable, fundamentally avoiding dangerous or invalid power allocation instructions. On the other hand, the consensus protocol... It provides a structured, mathematically analyzable, and efficient iterative framework, where each edge computing node only needs to exchange simplified state information with its neighbors (i.e., Instead of all the original data, this consensus protocol uses a carefully designed control law to guide the system state to converge quickly and stably to a consensus solution that satisfies both global constraints (transformer capacity) and all local constraints, avoiding oscillations or non-convergence. In summary, the iterative negotiation scheme based on the consensus algorithm does not rely on any central node for computation and coordination. Each edge computing node independently updates its own state based solely on local information and limited information obtained from its neighbors, using the same protocol rules. This ultimately leads to globally consistent optimization decisions, greatly enhancing the system's robustness and reliability. The failure of some nodes will not cause the entire system to collapse, thus achieving true decentralized autonomous collaboration.

[0069] As mentioned above, The distributed control law is designed based on local constraints to ensure that the result after each iteration satisfies the local physical constraints. In the above embodiments, although the distributed control law ensures that local physical constraints are met during iteration, in distributed negotiation, each edge computing node only possesses local information. How to design a control law so that the autonomous decisions of each node not only satisfy local constraints but also collaboratively optimize the global objective (i.e., overall satisfaction), avoiding the solution from falling into a local optimum or suboptimal state, is a pressing technical problem. Furthermore, how to ensure convergence to a high-quality solution (e.g., a Pareto optimal solution, meaning that no other solution can improve the satisfaction of any node without harming the satisfaction of other nodes), thereby guaranteeing the fairness and efficiency of resource allocation, is another pressing technical problem in distributed negotiation. To address this, this application solves by designing the distributed control law in the above-mentioned distributed negotiation in the following manner. Define a local cost function This function penalizes the deviation between its total load and the transformer capacity allocation value, as well as the cost of using its adjustable margin; in each consensus iteration, a gradient descent-based optimization method is used to calculate the distributed control law. This ensures that the global consensus solution converges to a Pareto optimal solution that satisfies system constraints while also achieving a high overall satisfaction level; specifically, a gradient descent-based optimization method is used to compute the distributed control law. Specifically, the solution that allows the global consensus solution to converge to the Pareto optimal solution with the best overall satisfaction while satisfying system constraints can be: based on the current state vector Calculate the local cost function gradient , where the local cost function Designed to penalize deviations between total load and transformer capacity allocation, as well as the cost of using adjustable margins, thus quantifying the cost of the current state; based on gradients... The gradient descent algorithm is used to calculate the distributed control law. The update amount, so that the state vector To reduce the cost function The direction is adjusted; specifically, the distributed control law is updated to... (where η is the learning rate), or incorporate a more complex optimizer to accelerate convergence; in updating the distributed control law Then, the application is ensured through constraint handling mechanisms (e.g., projected gradient method or Lagrange multiplier method). The resulting new state vector Satisfy local physical constraints to ensure that the intermediate state after each iteration is a feasible solution; repeat the above steps until the state vectors of all edge computing nodes are calculated. No longer significantly changing (i.e., reaching consensus), and the cost function Minimize, at which point the global consensus solution satisfies the system constraints and converges to the Pareto optimal solution with overall satisfaction.

[0070] Step S105: Upload key operation logs and decision data during the charging process to the cloud blockchain node for secure storage via an encrypted channel.

[0071] If critical operation logs and decision data during the charging process are stored locally or in a centralized database, this data is easily tampered with locally or lost due to centralized server failure, resulting in insufficient credibility. However, leveraging the immutability and traceability of blockchain, a reliable chain of evidence can be provided for all critical operations. Therefore, this application can upload critical operation logs and decision data during the charging process to a cloud blockchain node for secure evidence storage via an encrypted channel. Specifically, as an embodiment of this application, uploading critical operation logs and decision data during the charging process to a cloud blockchain node for secure evidence storage via an encrypted channel can be as follows: at an edge computing node, a quantum-resistant cryptographic algorithm is used to digitally sign the hash value of the critical operation logs and decision data; the signed hash value, timestamp, and device identifier are combined to form a data packet for evidence storage; and the data packet is broadcast to the cloud blockchain network via an encrypted channel. On the cloud-based blockchain network side, the blockchain's smart contract automatically verifies the validity of the digital signature in the received evidence storage transaction data packet. After successful verification, the smart contract writes the hash value in the evidence storage transaction data packet as the transaction content into a new block. The new blockchain is then connected to the main blockchain, and a transaction receipt containing the hash value of that block is generated and returned to the edge computing node, completing the verifiable and secure evidence storage.

[0072] From the above appendix Figure 1 As illustrated by the example of electric vehicle charging management, on the one hand, deploying a spatiotemporal graph convolutional network model at edge computing nodes close to the data source for millisecond-level policy generation effectively reduces the latency of remote data transmission to the cloud. This allows the system to react instantly to rapid changes in the power grid state, greatly improving control timeliness and ensuring the real-time safety and stability of the power grid. On the other hand, the spatiotemporal graph convolutional network model at the edge nodes focuses on the initial global optimization of regional power grid safety and efficiency, while the local decision-making model at the charging terminal further integrates personalized factors such as vehicle battery health status and real-time electricity prices for in-depth optimization, ensuring that the final executed strategy is no longer... It is not a compromise of a single objective, but a truly multi-objective collaborative optimal solution, thereby simultaneously ensuring grid security, user economy, and vehicle battery life. Thirdly, by constructing a distributed collaborative decision-making mechanism among edge computing nodes, when grid parameters exceed limits or load approaches capacity, multiple edge nodes can autonomously and quickly exchange and negotiate information without relying on the cloud center, ultimately reaching a consistent power allocation scheme that satisfies global constraints (e.g., total transformer capacity) and maximizes overall satisfaction. This decentralized collaborative approach not only avoids the risk of single-point failures but also forms a kind of "collective intelligence," jointly maintaining the stable operation of the regional power grid. In summary, the technical solution of this application achieves millisecond-level response for electric vehicle charging, multi-objective collaborative optimization, and rapid collaborative handling of grid faults.

[0073] Please see the appendix Figure 2 This application provides an electric vehicle charging management device, which may include a receiving module 201, a first strategy generation module 202, a second strategy generation module 203, a coordination module 204, and an uploading module 205, as detailed below:

[0074] The receiving module 201 is used to receive real-time operating data from multiple charging terminals at the edge computing node;

[0075] The first strategy generation module 202 is used to generate an initial charging strategy for each charging terminal within a millisecond-level response time by using a first artificial intelligence model deployed on an edge computing node and based on real-time running data. The first artificial intelligence model is a prediction model based on the spatiotemporal graph convolutional network ST-GNN.

[0076] The second strategy generation module 203 is used to optimize the initial charging strategy locally using a second artificial intelligence model deployed on the charging terminal, generate the final charging strategy and execute it. The second artificial intelligence model is a decision model built with a deep reinforcement learning (DRL) algorithm.

[0077] The collaboration module 204 is used to initiate a distributed collaborative decision-making mechanism among multiple associated edge computing nodes when the power grid parameter fluctuation exceeds a predetermined safety threshold or the total load of the charging terminal cluster in the area approaches the upper limit of the transformer capacity, so as to dynamically negotiate and redistribute the charging power of each charging terminal.

[0078] Upload module 205 is used to upload key operation logs and decision data during the charging process to the cloud blockchain node for secure storage via an encrypted channel.

[0079] From the above appendix Figure 2As illustrated by the example of an electric vehicle charging management device, on the one hand, deploying a spatiotemporal graph convolutional network model based on millisecond-level policy generation at edge computing nodes close to the data source effectively reduces the latency of remote data transmission to the cloud. This allows the system to react instantly to rapid changes in the power grid state, greatly improving control timeliness and ensuring the real-time safety and stability of the power grid. On the other hand, the spatiotemporal graph convolutional network model at the edge nodes focuses on the initial global optimization of regional power grid safety and efficiency, while the local decision-making model at the charging terminal further integrates personalized factors such as vehicle battery health status and real-time electricity prices for in-depth optimization, ensuring that the final executed strategy is no longer... It is not a compromise of a single objective, but a truly multi-objective collaborative optimal solution, thereby simultaneously ensuring grid security, user economy, and vehicle battery life. Thirdly, by constructing a distributed collaborative decision-making mechanism among edge computing nodes, when grid parameters exceed limits or load approaches capacity, multiple edge nodes can autonomously and quickly exchange and negotiate information without relying on the cloud center, ultimately reaching a consistent power allocation scheme that satisfies global constraints (e.g., total transformer capacity) and maximizes overall satisfaction. This decentralized collaborative approach not only avoids the risk of single-point failures but also forms a kind of "collective intelligence," jointly maintaining the stable operation of the regional power grid. In summary, the technical solution of this application achieves millisecond-level response for electric vehicle charging, multi-objective collaborative optimization, and rapid collaborative handling of grid faults.

[0080] Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. For example... Figure 3 As shown, the electronic device 3 in this embodiment mainly includes: a processor 30, a memory 31, and a computer program 32 stored in the memory 31 and executable on the processor 30, such as a program for an electric vehicle charging management method. When the processor 30 executes the computer program 32, it implements the steps described in the above-described embodiment of the electric vehicle charging management method, for example... Figure 1 The steps S101 to S105 are shown. Alternatively, when the processor 30 executes the computer program 32, it implements the functions of each module / unit in the above-described device embodiments, for example... Figure 2 The functions of the receiving module 201, the first strategy generation module 202, the second strategy generation module 203, the collaboration module 204, and the uploading module 205 are shown.

[0081] For example, the computer program 32 of the electric vehicle charging management method mainly includes: receiving real-time operating data from multiple charging terminals at an edge computing node; using a first artificial intelligence model deployed on the edge computing node, generating an initial charging strategy for each charging terminal within a millisecond-level response time based on the real-time operating data received by the edge computing node, wherein the first artificial intelligence model is a prediction model based on a spatiotemporal graph convolutional network ST-GNN; using a second artificial intelligence model deployed locally on the charging terminal, performing local optimization on the initial charging strategy generated by the first artificial intelligence model, generating a final charging strategy and executing it, wherein the second artificial intelligence model is a decision model constructed using a deep reinforcement learning DRL algorithm; when it is detected that the fluctuation of the power grid parameters exceeds a predetermined safety threshold or the total load of the charging terminal cluster in the area approaches the upper limit of the transformer capacity, initiating a distributed collaborative decision-making mechanism among multiple associated edge computing nodes, dynamically negotiating and redistributing the charging power of each charging terminal; and uploading key operation logs and decision data during the charging process to a cloud blockchain node for secure storage through an encrypted channel. The computer program 32 can be divided into one or more modules / units, one or more modules / units are stored in memory 31 and executed by processor 30 to complete this application. One or more modules / units can be a series of computer program instruction segments capable of performing specific functions, which describe the execution process of computer program 32 in electronic device 3. For example, computer program 32 can be divided into the functions of receiving module 201, first strategy generation module 202, second strategy generation module 203, coordination module 204, and uploading module 205 (a module in the virtual device). The specific functions of each module are as follows: receiving module 201 is used to receive real-time running data from multiple charging terminals at the edge computing node; first strategy generation module 202 is used to generate an initial charging strategy for each charging terminal within a millisecond response time based on real-time running data using a first artificial intelligence model deployed on the edge computing node, wherein the first artificial intelligence model is a prediction model based on spatiotemporal graph convolutional network ST-GNN; second strategy generation module 203 is used to generate an initial charging strategy for each charging terminal within a millisecond response time using a first artificial intelligence model deployed on the edge computing node, based on real-time running data; the second strategy generation module 204 is used to generate an initial charging strategy for each charging terminal using a first artificial intelligence model deployed on the edge computing node, based on real-time running data; the first artificial intelligence model is a prediction model based on spatiotemporal graph convolutional network ST-GNN; the second strategy generation module 205 is used to generate an initial charging strategy for each charging terminal within a millisecond response time. Block 203 is used to optimize the initial charging strategy locally using a second artificial intelligence model deployed locally on the charging terminal, generate the final charging strategy, and execute it. The second artificial intelligence model is a decision model built with a deep reinforcement learning (DRL) algorithm. The collaboration module 204 is used to initiate a distributed collaborative decision-making mechanism among multiple associated edge computing nodes when the grid parameter fluctuation exceeds a predetermined safety threshold or the total load of the charging terminal cluster in the area approaches the transformer capacity limit. This mechanism dynamically negotiates and redistributes the charging power of each charging terminal. The upload module 205 is used to upload key operation logs and decision data during the charging process to the cloud blockchain node for secure storage through an encrypted channel.

[0082] Electronic device 3 may include, but is not limited to, processor 30 and memory 31. Those skilled in the art will understand that... Figure 3 This is merely an example of electronic device 3 and does not constitute a limitation on electronic device 3. It may include more or fewer components than shown, or combine certain components, or different components. For example, electronic device may also include input / output devices, network access devices, buses, etc.

[0083] The processor 30 may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor.

[0084] The memory 31 can be an internal storage unit of the electronic device 3, such as a hard disk or RAM. The memory 31 can also be an external storage device of the electronic device 3, such as a plug-in hard disk, Smart Media Card (SMC), Secure Digital (SD) card, or Flash Card. Furthermore, the memory 31 can include both internal and external storage units of the electronic device 3. The memory 31 is used to store computer programs and other programs and data required by the electronic device. The memory 31 can also be used to temporarily store data that has been output or will be output.

[0085] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed. That is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above-described device can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0086] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0087] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0088] In the embodiments provided in this application, it should be understood that the disclosed apparatus / device and method can be implemented in other ways. For example, the apparatus / device embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0089] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0090] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0091] If integrated modules / units are implemented as software functional units and sold or used as independent products, they can be stored in a storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program for the electric vehicle charging management method can be stored in a storage medium. When executed by a processor, this computer program can implement the steps of the various method embodiments described above, namely, receiving real-time operating data from multiple charging terminals at the edge computing node; using a first artificial intelligence model deployed on the edge computing node, generating an initial charging strategy for each charging terminal within a millisecond-level response time based on the real-time operating data received by the edge computing node; wherein the first artificial intelligence model is based on a spatiotemporal graph convolutional network ST-... The system employs a GNN prediction model; a second AI model deployed locally on the charging terminal optimizes the initial charging strategy generated by the first AI model, generating and executing the final charging strategy. The second AI model is a decision-making model built using a deep reinforcement learning (DRL) algorithm. When grid parameter fluctuations exceed a predetermined safety threshold or the total load of the charging terminal cluster in the area approaches the transformer capacity limit, a distributed collaborative decision-making mechanism is initiated among multiple associated edge computing nodes to dynamically negotiate and redistribute the charging power of each charging terminal. Key operation logs and decision data during the charging process are uploaded to a cloud blockchain node via an encrypted channel for secure storage. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. Storage media can include any entity or device capable of carrying computer program code, recording media, USB flash drives, external hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media. It should be noted that the contents of the storage medium may be appropriately added to or subtracted from the contents according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the storage medium may not include electrical carrier signals and telecommunication signals.

[0092] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit it. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. These modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application. The specific embodiments described above further illustrate the purpose, technical solutions, and beneficial effects of this application. It should be understood that the above descriptions are merely specific embodiments of this application and are not intended to limit the protection scope of this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this invention.

Claims

1. A method for managing electric vehicle charging, characterized in that, The method includes: The edge computing node receives real-time operational data from multiple charging terminals. Using a first artificial intelligence model deployed on the edge computing node, an initial charging strategy for each charging terminal is generated within a millisecond response time based on the real-time running data. The first artificial intelligence model is a prediction model based on the spatiotemporal graph convolutional network ST-GNN. The initial charging strategy is optimized locally using a second artificial intelligence model deployed on the charging terminal to generate and execute the final charging strategy. The second artificial intelligence model is a decision model built with a deep reinforcement learning (DRL) algorithm. When the power grid parameter fluctuations exceed the predetermined safety threshold or the total load of the charging terminal cluster in the area approaches the upper limit of the transformer capacity, a distributed collaborative decision-making mechanism is initiated among multiple associated edge computing nodes to dynamically negotiate and redistribute the charging power of each charging terminal. Key operation logs and decision data during the charging process are uploaded to a cloud blockchain node via an encrypted channel for secure storage.

2. The electric vehicle charging management method according to claim 1, characterized in that, The first artificial intelligence model treats charging terminals as nodes in a graph, dynamically constructing a graph structure with grid parameters and environmental conditions as edge features, to simultaneously predict the optimal charging power probability distribution of all charging terminals within a short-term time window in the future; the dynamic construction of the graph structure with grid parameters and environmental conditions as edge features includes: Based on the physical distance between the charging terminals and the topology of the power grid connection, a static spatial adjacency matrix is ​​constructed; The line power flow values ​​from the real-time collected power grid operation parameters and the temperature values ​​from the environmental status information are used as dynamic edge weights to weight and update the static spatial adjacency matrix, forming a dynamic spatiotemporal graph structure.

3. The electric vehicle charging management method according to claim 2, characterized in that, The method utilizes a first artificial intelligence model deployed on the edge computing node to generate an initial charging strategy for each charging terminal within a millisecond-level response time based on the real-time operational data, including: The dynamic spatiotemporal graph structure is input into the spatial graph convolution module, which integrates a multi-head attention mechanism to extract and output the spatial feature sequence formed by the nonlinear energy flow interaction between charging terminals. The spatial feature sequence is input into the temporal convolutional network (TCN) module to extract and output its long-term dependency features in the temporal dimension from the spatial feature sequence. The long-term dependency features are input into the output layer of the hybrid density network (MDN), and the MDN output layer calculates and outputs Gaussian mixture model parameters that characterize the probability distribution of future charging power for each charging terminal.

4. The electric vehicle charging management method according to claim 1, characterized in that, The reward function R of the second artificial intelligence model is designed as follows: The and These represent the battery health status of the vehicle at time t and time t-1, respectively. The initial policy power issued to the edge computing node. This is the actual adjusted power. For real-time electricity price costs, , and The weighting coefficients are used to determine the localization optimization, which at least considers the vehicle battery health status characteristics.

5. The electric vehicle charging management method according to claim 1, characterized in that, The triggering conditions for initiating the distributed collaborative decision-making mechanism are determined according to the following steps: The power grid frequency signal is continuously sampled, and its standard deviation within a preset sliding time window is calculated. ; like If the first threshold is exceeded, a primary warning will be activated, and the data sampling frequency will be increased. If, under the initial warning state, the first derivative of the frequency... Continuously exceeding the second threshold achieve If the fluctuation exceeds a predetermined safety threshold within a few seconds, the distributed collaborative decision-making mechanism is immediately triggered.

6. The electric vehicle charging management method according to claim 1, characterized in that, The distributed collaborative decision-making mechanism includes: each edge computing node iteratively exchanges information on the total load and adjustable margin of the charging terminals it manages based on a consensus algorithm, until all nodes reach a consensus on the global power allocation scheme. The consensus scheme satisfies the total capacity constraint of the transformer and maximizes the overall satisfaction of all charging terminals. Each edge computing node, based on a consensus algorithm, iteratively exchanges information on the total load and adjustable margin of the charging terminals it manages, until all nodes reach a consensus on the global power allocation scheme, including: Each of the aforementioned edge computing nodes sets its state vector Broadcast to its communication neighbor nodes, the For the total load, the This allows for adjustable power margin; Each node receives the state vectors of all its neighbors and updates its own state based on the following consensus protocol: The It is the consensus weight, the aforementioned It is a distributed control law designed based on local constraints, used to ensure that the result after each iteration satisfies the local physical constraints.

7. The electric vehicle charging management method according to claim 1, characterized in that, The process of uploading key operation logs and decision data during the charging process to a cloud blockchain node for secure storage via an encrypted channel includes: At edge computing nodes, quantum-resistant cryptographic algorithms are used to digitally sign the hash values ​​of critical operation logs and decision data; The signed hash value, timestamp, and device identifier are combined to form a single evidence storage transaction data packet. The data packet containing the evidence storage transaction is broadcast to the cloud blockchain network through the encrypted channel.

8. An electric vehicle charging management device, characterized in that, The device includes: The receiving module is used to receive real-time operating data from multiple charging terminals at the edge computing node; The first strategy generation module is used to generate an initial charging strategy for each charging terminal within a millisecond-level response time by using a first artificial intelligence model deployed on the edge computing node and based on the real-time running data. The first artificial intelligence model is a prediction model based on the spatiotemporal graph convolutional network ST-GNN. The second strategy generation module is used to optimize the initial charging strategy locally using a second artificial intelligence model deployed on the charging terminal, generate the final charging strategy and execute it. The second artificial intelligence model is a decision model built with a deep reinforcement learning (DRL) algorithm. The collaboration module is used to initiate a distributed collaborative decision-making mechanism among multiple associated edge computing nodes when the power grid parameter fluctuation exceeds the predetermined safety threshold or the total load of the charging terminal cluster in the area approaches the upper limit of the transformer capacity, so as to dynamically negotiate and redistribute the charging power of each charging terminal. The upload module is used to upload key operation logs and decision data during the charging process to the cloud blockchain node for secure storage via an encrypted channel.

9. An electronic device, the device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method as described in any one of claims 1 to 7.

10. A storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 7.

Images