A power distribution automation terminal control optimization method, device, equipment and storage medium

Abstract

This application discloses a method, apparatus, device, and storage medium for optimizing control of distribution automation terminals, relating to the fields of power automation and intelligent distribution control technology. The method includes: acquiring real-time operating data of distribution terminals; allocating dedicated network slice resources according to the data priority of the real-time operating data, and performing bandwidth isolation on the dedicated network slice resources to obtain isolated network slice resources; fusing real-time operating data of multiple distribution terminals based on the isolated network slice resources to obtain a fused feature set; generating an optimized control strategy through a graph neural network based on the fused feature set and the distribution network topology; generating control commands according to the optimized control strategy and sending them to the distribution terminals to enable the distribution terminals to perform fault isolation and recovery operations, thereby achieving high reliability and low latency in communication transmission and optimization of control strategies during distribution network fault isolation and recovery.

Description

Technical Field

[0001] This application relates to the field of power automation and intelligent distribution control technology, and in particular to a method, device, equipment and storage medium for optimizing terminal control of distribution automation. Background Technology

[0002] With the deepening of smart grid construction, distribution automation systems are playing an increasingly important role in power system operation. When short-circuit or ground faults occur in the distribution network, rapid fault isolation and restoration of power to non-faulty areas are crucial to ensure the safe and stable operation of the power grid. The ultra-reliable, low-latency characteristics of 5G mobile communication technology provide new support for power system communication. Its end-to-end latency meets the stringent requirements of power automation control, making the optimization of distribution automation terminal control based on 5G mobile communication a key requirement for smart grid development.

[0003] However, in actual operation, especially in the scenario of distribution network fault isolation and recovery, traditional systems still face many technical limitations: when a fault occurs suddenly, the flooding effect caused by a large number of terminals simultaneously reporting fault alarm traffic can easily lead to network congestion, loss of critical fault control commands, or severe queuing due to fixed communication resource allocation; traditional algorithms have difficulty effectively combining the complex physical topology of the distribution network with communication routing, resulting in a lag in control strategy optimization; in addition, insufficient computing power at the field terminals and coordination with edge nodes make it difficult to cope with the surge in latency caused by dynamic switching of the distribution network topology, thus limiting the ability to quickly disconnect circuit breakers, isolate faults, and achieve precise control, which may cause large-scale power outages. Summary of the Invention

[0004] The main objective of this application is to provide a method, apparatus, device, and storage medium for optimizing control of distribution automation terminals, aiming to solve the technical problem of overcoming network congestion and loss of key instructions during the process of distribution network fault isolation and recovery, and to optimize the control strategy.

[0005] To achieve the above objectives, this application provides a method for optimizing terminal control in power distribution automation, comprising the following steps: Obtain real-time operating data from the power distribution terminal; Dedicated network slice resources are allocated based on the data priority of the real-time running data, and the dedicated network slice resources are bandwidth isolated to obtain isolated network slice resources. Based on the isolated network slice resources, the real-time operation data of multiple power distribution terminals are fused to obtain a fused feature set; Based on the fused feature set and the power distribution network topology, an optimized control strategy is generated using a graph neural network. Control commands are generated according to the optimized control strategy and sent to the power distribution terminal so that the power distribution terminal can perform fault isolation and recovery operations.

[0006] In one embodiment, the step of acquiring real-time operating data of the power distribution terminal includes: Acquire voltage, current, and power data from the power distribution terminal; The voltage data, current data, and power data are sampled according to a preset sampling frequency to obtain sampled data. The sampled data is subjected to Kalman filtering to obtain filtered data; Abnormal data in the filtered data are marked according to a preset voltage deviation threshold to obtain marked data; Real-time running data is generated based on the marked data.

[0007] In one embodiment, the step of allocating dedicated network slice resources according to the data priority of the real-time running data, and performing bandwidth isolation on the dedicated network slice resources to obtain isolated network slice resources includes: Obtain the status data of the power distribution terminal and the base station; A multi-agent reinforcement learning environment is constructed based on the state data. The network load is predicted through the multi-agent reinforcement learning environment, and the transmission path is adjusted according to the predicted network load to obtain the path optimization result. The fault types, terminal states, and network resources of the power distribution system are mapped into a graph structure, and semantic reasoning is performed based on the relationships between entities in the graph structure to obtain a priority score. The slice allocation problem is modeled as a non-cooperative game, and the reward function is calculated based on the power ratio and bandwidth resources of each agent to obtain the game optimization result. Dedicated network slice resources are allocated based on the priority score, the game optimization result, and the path optimization result, and the dedicated network slice resources are bandwidth isolated to obtain isolated network slice resources.

[0008] In one embodiment, the step of allocating dedicated network slice resources based on the priority score, the game optimization result, and the path optimization result, and performing bandwidth isolation on the dedicated network slice resources to obtain isolated network slice resources, includes: End-to-end latency is determined based on propagation latency, processing latency, queuing latency, and collaboration latency. The system monitors data integrity, and when the data corruption rate exceeds a preset corruption threshold, the power distribution terminal and the base station are used as negotiating agents. The retransmission path is determined through negotiation by the negotiation agent, and retransmission is triggered. Adjust the transmission service quality identifier parameters according to the end-to-end delay and the retransmission path; The slice allocation weight is determined based on the priority score, the game optimization result, the bandwidth utilization, and the end-to-end latency. Dedicated network slice resources are allocated based on the slice allocation weights and the path optimization results, and the dedicated network slice resources are bandwidth isolated to obtain isolated network slice resources.

[0009] In one embodiment, the step of fusing the real-time operating data of multiple power distribution terminals based on the isolated network slice resources to obtain a fused feature set includes: The real-time operating data of multiple power distribution terminals are obtained through the isolated network slice resources; The normalization weights are determined based on the signal strength of each of the power distribution terminals. The multiple real-time running data are weighted and fused according to the normalized weights to obtain a fused feature vector; Calculate the power fluctuation rate based on the power data within a preset time window; A fusion feature set is generated based on the fusion feature vector and the power volatility.

[0010] In one embodiment, the step of generating an optimized control strategy using a graph neural network based on the fused feature set and the distribution network topology includes: A power distribution network diagram is constructed based on the terminal nodes and communication connections in the power distribution network, wherein the terminal nodes and communication connections constitute the power distribution network topology. The fused feature set is used as the node feature input to the graph neural network for each terminal node in the power distribution network graph; The graph neural network extracts the features of the neighboring nodes of the current terminal node, and the neighboring node features are aggregated, transformed and nonlinearly mapped by a weight matrix to obtain the hidden layer features. By minimizing the loss function, an optimal control strategy corresponding to the hidden layer features is determined.

[0011] In one embodiment, the step of extracting the neighbor node features of the current terminal node through the graph neural network, and performing aggregation transformation and nonlinear mapping on the neighbor node features through a weight matrix to obtain the hidden layer features includes: The neighbor node features are linearly transformed using a weight matrix to obtain the transformed features. The transformed features are normalized and aggregated based on the number of nodes in the neighbor node set to obtain aggregated features; The aggregated features are concatenated with the local features of the current terminal node to obtain the concatenated features; Hidden layer features are generated by performing a non-linear mapping on the concatenated features using an activation function.

[0012] Furthermore, to achieve the above objectives, this application also proposes a power distribution automation terminal control optimization device, the device comprising: The data acquisition module is used to acquire real-time operating data of the power distribution terminal; The transmission allocation module is used to allocate dedicated network slice resources according to the data priority of the real-time running data, and to perform bandwidth isolation on the dedicated network slice resources to obtain isolated network slice resources. The fusion extraction module is used to fuse the real-time operating data of multiple power distribution terminals based on the isolated network slice resources to obtain a fusion feature set; The strategy generation module is used to generate an optimized control strategy based on the fused feature set and the power distribution network topology using a graph neural network. The control module is used to generate control commands according to the optimized control strategy and send them to the power distribution terminal so that the power distribution terminal can perform fault isolation and recovery operations.

[0013] Furthermore, to achieve the above objectives, this application also proposes a power distribution automation terminal control optimization device, the device comprising: a memory, a processor, and a power distribution automation terminal control optimization program stored in the memory and executable on the processor, the power distribution automation terminal control optimization program being configured to implement the steps of the power distribution automation terminal control optimization method as described above.

[0014] In addition, to achieve the above objectives, this application also proposes a storage medium, which is a computer-readable storage medium, storing a power distribution automation terminal control optimization program, which, when executed by a processor, implements the steps of the power distribution automation terminal control optimization method described above.

[0015] In addition, to achieve the above objectives, the present invention also provides a computer program product, the computer program product including a power distribution automation terminal control optimization program, which, when executed by a processor, implements the steps of the power distribution automation terminal control optimization method as described above.

[0016] This application effectively avoids network congestion caused by a large number of terminals simultaneously reporting data in distribution network fault scenarios by allocating dedicated network slice resources according to data priority and isolating bandwidth, thus ensuring the reliable transmission of critical fault control commands. Through multi-agent reinforcement learning to predict network load and adjust transmission paths, combined with knowledge graph semantic reasoning to calculate priority scores and non-cooperative game theory models to calculate reward functions, dynamic adaptive allocation of network slice resources is achieved, improving communication resource utilization efficiency and transmission real-time performance. By fusing real-time operating data from multiple distribution terminals based on isolated network slice resources, the load on the central node is reduced, improving the timeliness of data processing. By constructing a graph neural network based on the fused feature set and distribution network topology, and using the aggregation transformation of neighbor node features and loss function minimization to determine the optimal control strategy, the application effectively combines the complex physical topology of the distribution network with communication routing relationships, improving the accuracy of the control strategy and the system response speed. By determining retransmission paths and triggering retransmissions through multi-agent negotiation, and dynamically adjusting transmission service quality identifier parameters based on end-to-end delay, the integrity and reliability of data transmission are further enhanced. Therefore, the application as a whole achieves high reliability and low latency in communication transmission and optimization of control strategies during distribution network fault isolation and recovery. Attached Figure Description

[0017] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0018] 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, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating the first embodiment of the power distribution automation terminal control optimization method of this application; Figure 2 This is a flowchart illustrating the second embodiment of the power distribution automation terminal control optimization method of this application; Figure 3 This is a flowchart illustrating the third embodiment of the power distribution automation terminal control optimization method of this application; Figure 4 This is a structural block diagram of the first embodiment of the power distribution automation terminal control optimization device of this application; Figure 5 This is a schematic diagram of the structure of the power distribution automation terminal control optimization equipment of this application.

[0020] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0021] It should be understood that the specific embodiments described herein are only used to explain the technical solutions of this application and are not intended to limit this application.

[0022] It should be noted that the executing entity of the embodiments of this application can be a computing service device with data processing, network communication, and program execution functions, such as a personal computer or mobile phone, or an electronic device capable of realizing the above functions, such as the aforementioned power distribution automation terminal control optimization device. The following embodiments will be described using the power distribution automation terminal control optimization device as an example.

[0023] This application provides a method for optimizing the control of a power distribution automation terminal, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the power distribution automation terminal control optimization method of this application.

[0024] In this embodiment, the power distribution automation terminal control optimization method includes the following steps: Step S10: Obtain real-time operating data of the power distribution terminal; It should be noted that a distribution terminal is an intelligent device deployed on the distribution network site to collect operating parameters and execute control commands. It achieves data interaction and remote control through built-in sensors and communication modules.

[0025] In addition, real-time operating data is a collection of electrical parameters and timestamp information that reflects the current working status of the power distribution terminal. It is obtained by sampling and processing the original signals and is used for subsequent analysis and control decisions.

[0026] Understandably, step S10 involves acquiring real-time operating data from the power distribution terminal, which means collecting raw signals such as voltage, current, and power from the power distribution terminal through sensors and communication interfaces, and then processing them to form a standardized data set that can be used by subsequent modules.

[0027] It should be understood that this step is performed on the power distribution terminal side, where the raw signals are collected and preliminarily processed.

[0028] Sensors are used to collect real-time parameters of the power distribution network, including voltage. Current and power ,in This represents the timestamp. The data sampling frequency is set to 1kHz to capture transient changes.

[0029] The collected data is noise filtered, and the Kalman filter algorithm is used to calculate the filtered data.

[0030] in, It is the state estimation vector (containing ), It is Kalman gain. These are observed values. It is the observation matrix. Indicates the estimated state. This is the gain matrix (calculated using covariance). For the original measurement, It is a linear observation model.

[0031] Generate preprocessed dataset and mark the abnormal threshold (e.g.) , (Nominal voltage).

[0032] In one feasible implementation, step S10 includes: S101: Acquire voltage, current, and power data from the power distribution terminal; It should be noted that voltage data is an electrical parameter that characterizes the potential difference between two points in a power distribution line. It records the instantaneous and effective values ​​of the voltage of each phase of the power distribution terminal in a time sequence and is used to determine whether the power grid is operating within the normal range.

[0033] In addition, current data is an electrical parameter that characterizes the amount of charge passing through the cross-section of a conductor per unit time. It records the instantaneous and effective values ​​of the current in each phase of the distribution terminal in a time sequence and is used to assess the line load level and fault characteristics.

[0034] Furthermore, power data is an electrical parameter that characterizes the rate of electrical energy conversion or transmission per unit time. It includes active and reactive power components and records the energy consumption status and load characteristics of the distribution terminal in a time-series format.

[0035] Understandably, step S101 involves acquiring the voltage, current, and power data of the power distribution terminal, i.e., collecting the raw electrical parameters of three-phase voltage, three-phase current, and three-phase power through the voltage transformer, current transformer, and power calculation unit built into the power distribution terminal.

[0036] S102: Sample voltage data, current data, and power data according to a preset sampling frequency to obtain sampled data; It should be noted that the preset sampling frequency is a parameter that is pre-set to determine the number of times the analog signal is discretized and sampled per unit time. It determines the time resolution and frequency band coverage of data acquisition. In this embodiment, it is set to 1 kHz to capture transient changes.

[0037] In addition, the sampled data is a data sequence obtained by discretizing a continuous analog signal according to a preset sampling frequency, which serves as the input basis for subsequent digital signal processing and feature extraction.

[0038] Understandably, step S102 involves sampling voltage data, current data, and power data according to a preset sampling frequency to obtain sampled data. That is, the continuous analog signals of voltage, current, and power are converted from analog to digital at a sampling interval of one thousand times per second to form a discrete digital signal sequence.

[0039] S103: Perform Kalman filtering on the sampled data to obtain filtered data; It should be noted that Kalman filtering is an optimal estimation algorithm based on a state-space model. It suppresses random noise interference through two recursive processes of prediction and update, and uses observation data to make an optimal estimate of the actual state.

[0040] In addition, the filtered data is a data sequence that has been denoised by the Kalman filter algorithm, which has a higher signal-to-noise ratio and higher accuracy in estimating the actual state compared to the original sampled data.

[0041] Understandably, step S103 involves performing Kalman filtering on the sampled data to obtain filtered data. This involves suppressing random noise in the sampled data through two recursive steps: state prediction and Kalman gain correction, resulting in a smooth data sequence that closely approximates the true state.

[0042] S104: Mark abnormal data in the filtered data according to the preset voltage deviation threshold to obtain the marked data; It should be noted that the preset voltage deviation threshold is a pre-set allowable range limit for voltage deviation from the nominal value, which is used to determine whether the current voltage data is in an abnormal working state. In this embodiment, it is set to five percent.

[0043] In addition, abnormal data refers to data points that exceed the preset voltage deviation threshold range, which indicates that the power grid may have fault symptoms such as short circuit, grounding, or sudden load change.

[0044] Furthermore, the labeled data is a set of data in which abnormal points exceeding the preset voltage deviation threshold are marked in the filtered data. It distinguishes normal data from suspicious data by adding abnormal labels.

[0045] Understandably, step S104 involves marking abnormal data in the filtered data according to a preset voltage deviation threshold, obtaining the marked data, that is, calculating the deviation ratio between the filtered data and the nominal voltage, marking data points with deviations exceeding 5% as abnormal, and generating a tagged data sequence.

[0046] S105: Generate real-time running data based on the marked data.

[0047] Understandably, step S105 generates real-time running data based on the marked data, that is, the voltage data, current data and power data after sampling, filtering and anomaly marking are integrated into a unified time series data set, forming real-time running data that can be used for subsequent transmission and analysis.

[0048] In this embodiment, the data acquisition and preprocessing in step S10 effectively eliminates sensor noise interference, improves data quality, and identifies potential fault signs in advance through anomaly marking, providing a clean data foundation with status indicators for subsequent high-reliability transmission and precise control.

[0049] Step S20: Allocate dedicated network slice resources according to the data priority of real-time running data, and perform bandwidth isolation on the dedicated network slice resources to obtain isolated network slice resources; It should be noted that data priority is a transmission level identifier based on data type and the urgency of the power grid. It is used to distinguish fault alarm data from routine monitoring data and ensure that critical control commands receive priority transmission resources.

[0050] In addition, dedicated network slice resources are independent logical network resources isolated for specific services in fifth-generation mobile communication networks through virtualization technology. They include dedicated bandwidth, time slots and quality of service guarantee mechanisms to meet the high reliability and low latency requirements of power distribution automation.

[0051] Furthermore, bandwidth isolation is a technical means of physically or logically isolating the transmission bandwidth of dedicated network slice resources from other service traffic. It prevents congested traffic from intruding into critical service channels through resource reservation and access control.

[0052] In addition, isolated network slice resources are network transmission resources that have undergone bandwidth isolation processing and can be exclusively used by power distribution automation services. They have guaranteed bandwidth and low latency transmission characteristics.

[0053] Understandably, step S20 involves allocating dedicated network slice resources based on the data priority of real-time running data, and isolating the dedicated network slice resources by bandwidth to obtain isolated network slice resources. That is, the urgency level of fault data is determined based on data priority, and a dedicated logical channel with quality of service guarantee is allocated in the fifth-generation mobile communication network. The transmission bandwidth of this channel is isolated from other service traffic through a resource reservation mechanism to form an independent and reliable transmission resource.

[0054] In the actual implementation, this step will preprocess the data. Data is transmitted to edge nodes via the 5G network and network slices are dynamically allocated.

[0055] Calculate transmission latency using 5G NSA mode:

[0056] in, It is the total delay. For propagation delay (calculated based on a distance of 510m). To handle the delay, Queuing delay. Symbol explanation: This represents the end-to-end latency (in milliseconds), with each component estimated using QoS parameters.

[0057] Using network slicing technology, 5QI parameters are assigned based on data priority: high-priority fault data (such as...) Dedicated slices are used to ensure bandwidth isolation. Slice allocation formula:

[0058] in, It is assigned to the data packet slices, It is the slice bandwidth. It refers to utilization rate. For slice index, Available bandwidth (Gbit / s). The utilization rate is 0-1.

[0059] Monitor the data corruption rate; if it exceeds 0.07%, trigger a retransmission. Output the transmitted data. To the edge node.

[0060] The purpose and function of this step is to leverage the low latency characteristics of 5G to achieve reliable transmission, reduce data loss, and improve system stability under high load scenarios by optimizing resource allocation through slicing.

[0061] This step introduces Multi-Agent Reinforcement Learning (MARL), treating each distribution terminal and base station as an agent to build a delay prediction model that dynamically adjusts transmission paths and resource allocation. The delay optimization formula is as follows:

[0062] Weighting coefficient Real-time learning and adjustment through MARL. Introducing a MARL-based delay prediction and path selection strategy:

[0063] Dynamic allocation and priority scheduling of network slices in a knowledge graph: Fault types, terminal states, and network resources of the power distribution system are mapped to a knowledge graph structure, and slice allocation is dynamically optimized through semantic reasoning. The optimized allocation formula is as follows:

[0064] Introducing a knowledge graph-based dynamic priority adjustment mechanism:

[0065] Adaptive retransmission for data integrity and edge collaborative optimization: Retransmission overhead is reduced through edge computing, and the retransmission threshold is dynamically adjusted using the following formula:

[0066] Edge collaborative optimization calculation formula:

[0067] Multi-agent collaborative 5G NSA mode transmission latency optimization: Total latency The calculation formula is expanded to The collaborative objective function is defined as follows:

[0068] Dynamic Network Slice Allocation and 5QI Parameter Optimization Based on Game Theory: The network slice allocation problem is modeled as a non-cooperative game, and the slice allocation formula is improved as follows:

[0069] The return function is: The formula for dynamically adjusting the 5QI parameter is:

[0070] Retransmission mechanism based on data integrity and multi-agent negotiation: Data corruption rate The retransmission path selection is based on multi-agent negotiation, and the optimization objective is:

[0071] In this embodiment, the slice allocation and bandwidth isolation in step S20 effectively avoid network congestion and loss of key control commands caused by a large number of terminals reporting data simultaneously in the case of power distribution network faults, thus ensuring the high reliability and low latency of communication transmission.

[0072] Step S30: Based on the isolated network slice resources, perform fusion processing on the real-time operation data of multiple power distribution terminals to obtain a fusion feature set; It should be noted that the fusion feature set is a high-dimensional feature vector set formed by weighted fusion and feature extraction of real-time operating data from multiple distribution terminals. It comprehensively reflects the regional operating status and topological correlation characteristics of the distribution network.

[0073] Understandably, step S30 is based on the fusion processing of real-time operating data of multiple distribution terminals using isolated network slice resources to obtain a fusion feature set. That is, it uses the isolated high-reliability transmission channel to aggregate real-time operating data of multiple terminals, and performs weighted fusion and fluctuation feature extraction on the edge side to form a fusion feature set that comprehensively represents the state of the regional power grid.

[0074] In practical implementation, data transmission is fused on edge devices. Extract key features.

[0075] Integrating data from multiple terminals:

[0076] in, It is a fusion of feature vectors. It is the number of terminals. These are weights (based on signal strength). Symbol explanation: The fused vector For normalized weights ( ).

[0077] Extracting features, such as power volatility:

[0078] in, It is the standard deviation. It is a time window. This is the average power. Symbol explanation: This represents volatility (unit: kW). This is the mean.

[0079] Generate feature sets This is used for subsequent optimization.

[0080] In one feasible implementation, step S30 includes: S301: Obtain real-time operating data of multiple power distribution terminals through isolated network slice resources; Understandably, step S301 involves obtaining real-time operating data from multiple power distribution terminals through isolated network slice resources. That is, by utilizing the low-latency, high-reliability transmission channel provided by the isolated network slice resources, real-time operating data such as voltage, current, and power from multiple power distribution terminals are received.

[0081] S302: Determine the normalization weights based on the signal strength of each power distribution terminal; It should be noted that signal strength is a physical quantity that characterizes the signal power level at the receiving end in a wireless communication link. It reflects channel quality and transmission reliability and is usually measured by the received power indication value.

[0082] In addition, the normalized weight is a weighting coefficient obtained by normalizing the signal strength of each distribution terminal. The sum of these weights is one, which is used to distinguish the credibility and importance of data from different terminals during data fusion.

[0083] Understandably, step S302 determines the normalized weight based on the signal strength of each power distribution terminal. That is, it measures the received signal power of each power distribution terminal when transmitting data through isolated network slice resources, and converts the signal strength into a normalized weighting coefficient. The stronger the signal, the higher the weight.

[0084] S303: Weighted fusion of multiple real-time running data according to normalized weights to obtain a fused feature vector; It should be noted that weighted fusion is a fusion method that uses normalized weights to sum the real-time operating data of multiple distribution terminals. It improves data reliability by having the data from high-weight terminals dominate the fusion result.

[0085] In addition, the fusion feature vector is a low-dimensional feature representation formed by weighted fusion of real-time operating data from multiple distribution terminals. It integrates the observation information from multiple terminals and suppresses single-point measurement errors.

[0086] Understandably, step S303 involves weighting and fusing multiple real-time operating data according to normalized weights to obtain a fused feature vector. That is, the voltage, current, and power data of each terminal are weighted and summed according to the determined normalized weights to generate a fused feature vector that comprehensively reflects the operating status of multiple terminals.

[0087] S304: Calculate the power fluctuation rate based on the power data within a preset time window; It should be noted that the preset time window is a pre-defined time interval for statistical power data, which determines the time scale and statistical significance of power fluctuation characteristics.

[0088] In addition, power volatility is a statistical measure that characterizes the degree of dispersion of power data within a preset time window. It reflects the stability and sudden change characteristics of the power grid load. The larger the value, the more severe the power fluctuation.

[0089] Understandably, step S304 involves calculating the power volatility based on the power data within a preset time window, that is, extracting the power data sequence within a preset time length, calculating the dispersion index of the sequence relative to the average power, and obtaining the volatility value characterizing power stability.

[0090] S305: Generate a fused feature set based on the fused feature vector and power volatility.

[0091] Understandably, step S305 generates a fusion feature set based on the fusion feature vector and power volatility, that is, the fusion feature vector and power volatility are combined and spliced ​​to form a fusion feature set that simultaneously contains multi-terminal fusion information and power stability features.

[0092] In this embodiment, the multi-terminal data fusion and feature extraction in step S30 effectively utilize the high reliability transmission characteristics of isolated network slice resources, reduce the data processing load of the central node, and improve the real-time performance and accuracy of regional power grid operation status perception.

[0093] Step S40: Generate an optimized control strategy using a graph neural network based on the fused feature set and the distribution network topology; It should be noted that the distribution network topology is a structured description that characterizes the electrical connection relationships and communication routing relationships between the terminal nodes in the distribution network. It includes a set of nodes and a set of connecting edges, reflecting the physical layout and information flow of the power grid.

[0094] Additionally, graph neural networks are a type of deep learning model specifically designed for processing graph-structured data. They aggregate neighbor node information through message passing mechanisms and learn high-dimensional embedding representations of nodes, making them suitable for data analysis with complex topological relationships.

[0095] Furthermore, the optimized control strategy is a control decision scheme used to guide the distribution terminal to perform switching actions and load adjustments. It is generated by graph neural network reasoning based on the current power grid state and topology relationship, aiming to achieve rapid fault isolation and power supply restoration.

[0096] Understandably, step S40 involves generating an optimized control strategy based on the fused feature set and the distribution network topology using a graph neural network. This means that the fused feature set is used as the node features of each terminal node in the distribution network graph, which is then input into the graph neural network. Through neighbor node feature aggregation and topological relationship reasoning, the optimal control decision under the current power grid condition is learned, and an optimized control strategy for fault isolation and recovery is output.

[0097] In this embodiment, the graph neural network strategy generation in step S40 effectively combines the complex physical topology and communication routing relationship of the distribution network, overcomes the shortcomings of traditional algorithms in integrating topology information, and improves the accuracy and response speed of the control strategy.

[0098] In the specific implementation, a graph neural network is used to optimize the control strategy, with the input feature set as the input. Output optimization instructions.

[0099] Constructing a power distribution network diagram ,node Indicates the terminal, edge Indicates a connection.

[0100] GNN forward propagation:

[0101] in, It is a node In the layer The hidden representation, It is an activation function. It is a weight matrix. It is a neighborhood set. Represented as a vector, These are trainable weights.

[0102] Output optimization control strategy ,in It is a loss function. It generates an instruction set. For example, a switch command.

[0103] Step S50: Generate control commands based on the optimized control strategy and send them to the power distribution terminal so that the power distribution terminal can perform fault isolation and recovery operations.

[0104] It should be noted that the control command is a digital command generated according to the optimized control strategy for operating the power distribution terminal switching equipment. It contains the switching action sequence and execution timing information, and is used to remotely control the opening and closing of the circuit breaker.

[0105] In addition, fault isolation is the process of separating the fault point from the normal power grid by disconnecting the switching equipment on both sides of the fault section when a short circuit or ground fault occurs in the distribution network. Its purpose is to prevent the fault from spreading and to protect the non-faulty sections.

[0106] Furthermore, the restoration operation is the process of restoring power to the non-faulty power outage section by closing the tie switch or backup power switch after the fault isolation is completed. Its purpose is to reduce the scope of the power outage and improve the reliability of power supply.

[0107] Understandably, step S50 generates control commands based on the optimized control strategy and sends them to the distribution terminal so that the distribution terminal can perform fault isolation and recovery operations. That is, the optimized control strategy generated by the graph neural network is transformed into specific switching action commands, which are sent to the distribution terminal through the isolated network slice resources to drive the circuit breaker to perform fault section isolation and non-fault section power restoration.

[0108] In this embodiment, the closed-loop transformation from control strategy to execution action is realized through the generation and issuance of instructions in step S50. The real-time delivery of control instructions is ensured by utilizing the highly reliable and low-latency isolated network slice resources, thereby achieving high reliability and low latency in communication transmission and optimization of control strategy during the fault isolation and recovery process of the distribution network.

[0109] In the specific implementation, the instructions will be optimized. Feedback is sent to the terminal, and parameters are adjusted based on the execution results.

[0110] Calculate the feedback error:

[0111] in, It is Euclidean error. It is the actual instruction executed.

[0112] Adaptive weight update:

[0113] in, It's the learning rate. For time The weight, It is the gradient.

[0114] like This triggers a re-collection, generating adjusted data. . This solution also includes steps for system performance evaluation and iterative optimization; Calculate system response time:

[0115] in, It is the total response time. Steps The delay.

[0116] Stability assessment metrics:

[0117] in, It is a stability score. For example... It is a score between 0 and 1.

[0118] like The GNN parameters are iteratively updated to generate an optimized model.

[0119] This embodiment effectively eliminates sensor noise interference, improves data quality, and identifies potential fault signs in advance by acquiring real-time operating data from distribution terminals and performing filtering and anomaly marking. By allocating dedicated network slice resources based on data priority and implementing bandwidth isolation, network congestion caused by a large number of terminals simultaneously reporting data in distribution network fault scenarios is effectively avoided, ensuring reliable transmission of critical control commands. Through multi-agent reinforcement learning to predict network load and adjust transmission paths, combined with knowledge graph semantic reasoning to calculate priority scores and a non-cooperative game model to calculate reward functions, dynamic adaptive allocation of network slice resources is achieved, improving communication resource utilization efficiency and real-time transmission performance. By fusing real-time operating data from multiple distribution terminals based on isolated network slice resources, the load on the central node is reduced, improving the real-time performance and accuracy of regional power grid operation status perception. By constructing a graph neural network based on the fused feature set and distribution network topology, and utilizing the aggregation transformation of neighbor node features and loss function minimization to determine the optimal control strategy, the complex physical topology and communication routing relationships of the distribution network are effectively combined, improving the accuracy and response speed of the control strategy. By generating control commands based on the optimized control strategy and sending them to the distribution terminal to perform fault isolation and recovery operations, a closed-loop transformation from control strategy to execution action is realized, ensuring the real-time delivery of control commands. This achieves high reliability and low latency in communication transmission and optimization of control strategies during the fault isolation and recovery process of the distribution network.

[0120] Reference Figure 2 , Figure 2 This is a flowchart illustrating the second embodiment of the power distribution automation terminal control optimization method of this application.

[0121] In one feasible implementation, step S20 may include: Step S201: Obtain status data of the power distribution terminal and base station; It should be noted that status data is a set of parameters characterizing the current operating conditions of power distribution terminals and base stations. It includes information such as equipment online status, channel quality, and remaining resource quantity, which is used to support environmental perception for subsequent intelligent decision-making.

[0122] Additionally, base stations are fixed communication devices deployed at the edge of the power distribution network in fifth-generation mobile communication networks. They are responsible for establishing wireless connections with power distribution terminals and forwarding data to the core network or edge nodes.

[0123] Understandably, step S201 involves acquiring the status data of the power distribution terminal and the base station, namely, collecting the online identification of the power distribution terminal, the channel quality indication, and the resource occupancy rate and coverage information of the base station, to form raw status data for network status assessment.

[0124] Step S202: Construct a multi-agent reinforcement learning environment based on the state data, predict the network load through the multi-agent reinforcement learning environment, adjust the transmission path according to the predicted network load, and obtain the path optimization result; It should be noted that the multi-agent reinforcement learning environment is a decision training framework composed of multiple intelligent interactive agents, in which each agent learns the optimal strategy through interaction with the environment. In this embodiment, it is used to coordinate the transmission decisions of the power distribution terminal and the base station.

[0125] In addition, network load is an indicator that represents the ratio of currently occupied resources to remaining resources in a communication network, reflecting the degree of channel congestion and the available transmission capacity.

[0126] Furthermore, the path optimization result is the optimal data transmission routing scheme calculated by the multi-agent reinforcement learning environment based on the predicted network load state, which aims to reduce end-to-end transmission latency.

[0127] Understandably, step S202 involves constructing a multi-agent reinforcement learning environment based on state data, predicting network load through the multi-agent reinforcement learning environment, adjusting the transmission path according to the predicted network load, and obtaining path optimization results. That is, the multi-agent interaction environment is initialized using the state data of the power distribution terminal and the base station, each agent predicts the network resource occupancy at future times, and dynamically adjusts the data transmission route from the terminal to the edge node based on the prediction results, generating path optimization results with minimum latency characteristics.

[0128] Step S203: Map the fault types, terminal states, and network resources of the power distribution system into a graph structure, and perform semantic reasoning based on the relationships between entities in the graph structure to obtain a priority score; It should be noted that fault type is a category of electrical fault that occurs in the distribution network. It includes short-circuit faults and grounding faults, and is used to distinguish different fault severity and handling priorities.

[0129] Additionally, the terminal status is a set of parameters that characterize the current operating mode of the power distribution terminal. It includes status indicators such as normal, alarm, and fault, reflecting the health level of the terminal.

[0130] Furthermore, network resources are the collective term for physical and logical resources such as spectrum, time slots, and bandwidth available for allocation in fifth-generation mobile communication networks. Their total amount and distribution directly affect the feasibility of slice allocation.

[0131] Additionally, graph structures are a structured knowledge representation that uses nodes and edges to represent entities and their relationships. They organize the semantic relationships between fault types, terminal states, and network resources in the form of graphs.

[0132] Furthermore, semantic reasoning is a process of logical deduction based on the relationships between entities in a graph structure, which mines implicit knowledge by traversing the paths between nodes.

[0133] Additionally, the priority score is a numerical indicator used to quantify the urgency of data, calculated through semantic reasoning. The higher the score, the more critical the data is to fault handling.

[0134] Understandably, step S203 maps the fault types, terminal states, and network resources of the power distribution system into a graph structure, and performs semantic reasoning based on the relationships between entities in the graph structure to obtain a priority score. That is, the fault categories such as short-circuit faults and ground faults, the normal or alarm states of power distribution terminals, and the available spectrum and bandwidth resources are abstracted into entity nodes and associated edges in the graph. By traversing the semantic association paths in the graph, the urgency of each data stream is deduced, and a priority score is generated to guide the allocation of slice resources.

[0135] Step S204: Model the slice allocation problem as a non-cooperative game, and calculate the reward function based on the power ratio and bandwidth resources of each agent to obtain the game optimization result; It should be noted that non-cooperative game theory is a game theory model in which each participant makes independent decisions to maximize their own gains, and there is no binding cooperation agreement between the participants. It is applicable to competitive resource allocation scenarios.

[0136] In addition, an intelligent agent is a participating entity with independent decision-making ability in a non-cooperative game model, which in this embodiment corresponds to network entities such as power distribution terminals and base stations.

[0137] Furthermore, the power percentage is the proportion of the power data transmitted by each agent in the total power data, which reflects the relative intensity of the agent's demand for network resources.

[0138] Additionally, the reward function is a mathematical expression used to quantify the gains obtained by each agent in a non-cooperative game after adopting a specific strategy. It comprehensively considers the power ratio and bandwidth resource usage.

[0139] Furthermore, the game optimization result is the output of the non-cooperative game model when it reaches a stable state under the interaction of the strategies of all parties, which represents the optimal slice selection strategy of each agent.

[0140] Understandably, step S204 models the slice allocation problem as a non-cooperative game and calculates the payoff function based on the power ratio and bandwidth resources of each agent to obtain the game optimization result. That is, the process of allocating network slice resources is abstracted into a competitive decision problem among multiple agents. The payoff value is calculated based on the power ratio of the data transmitted by each agent and the bandwidth resources occupied. Through game interaction, the strategy converges to a stable combination of strategies that none of the parties can improve their payoffs unilaterally, thus forming the game optimization result.

[0141] Step S205: Allocate dedicated network slice resources according to priority scores, game optimization results and path optimization results, and perform bandwidth isolation on the dedicated network slice resources to obtain isolated network slice resources.

[0142] Understandably, step S205 involves allocating dedicated network slice resources based on priority scores, game optimization results, and path optimization results, and then performing bandwidth isolation on these dedicated network slice resources to obtain isolated network slice resources. This involves allocating dedicated logical channels with quality of service guarantees for distribution automation services in the fifth-generation mobile communication network based on priority scores obtained from knowledge graph reasoning, game optimization results generated from non-cooperative game theory, and path optimization results output from multi-agent reinforcement learning. This results in isolated network slice resources that can be used for high-reliability, low-latency transmission.

[0143] In one feasible implementation, step S205 includes: S2051: Determine the end-to-end delay based on propagation delay, processing delay, queuing delay, and coordination delay; It should be noted that propagation delay is the time required for a signal to travel from the transmitter to the receiver in the form of electromagnetic waves in a wireless channel, and it is determined by the transmission distance and propagation speed.

[0144] Additionally, processing latency is the time required for base station or core network equipment to encode, decode, modulate, demodulate, and process received data according to protocols, reflecting the equipment's computing power.

[0145] Furthermore, queuing delay is the time that data packets spend waiting in the device's buffer queue for scheduled transmission, and it increases with network load.

[0146] In addition, collaboration latency is the time required for signaling and consensus among agents during multi-agent negotiation, reflecting the efficiency of distributed collaboration.

[0147] Furthermore, end-to-end delay is the sum of propagation delay, processing delay, queuing delay, and coordination delay, which represents the total time delay from when data is sent from the distribution terminal to when it is received by the edge node.

[0148] Understandably, step S2051 determines the end-to-end delay based on the propagation delay, processing delay, queuing delay, and cooperation delay. That is, the propagation time of the signal in the wireless channel, the processing time of the base station, the waiting time of the data in the queue, and the interaction time of multi-agent negotiation are calculated respectively, and the above four components are added together to obtain the total transmission delay.

[0149] S2052: Monitor data integrity. When the data corruption rate exceeds the preset corruption threshold, the power distribution terminal and base station will be used as negotiating agents. It should be noted that data integrity is an attribute that indicates that data has not been lost, tampered with, or damaged during transmission. It ensures the consistency between received and sent data through verification mechanisms.

[0150] Additionally, the data corruption rate is the proportion of erroneous data detected by the receiver out of the total received data, reflecting the impact of wireless channel interference and network congestion on transmission quality.

[0151] Furthermore, the preset corruption threshold is a pre-set upper limit for the allowed data corruption rate, which serves as the determination boundary for triggering the retransmission mechanism. In this embodiment, it is set to 0.07 percent.

[0152] Additionally, the negotiation agent is an intelligent entity with negotiation and decision-making capabilities that is activated when the data corruption rate exceeds a preset corruption threshold. It is jointly undertaken by the power distribution terminal and the base station.

[0153] Understandably, step S2052 involves monitoring data integrity. When the data corruption rate exceeds a preset corruption threshold, the power distribution terminal and the base station are used as negotiation agents. That is, the correctness ratio of the received data is detected in real time. When the proportion of erroneous data exceeds 0.07%, the power distribution terminal and the base station are activated as negotiation participants to prepare for the retransmission negotiation process.

[0154] S2053: The retransmission path is determined through negotiation by the negotiation agent, and retransmission is triggered; It should be noted that the retransmission path is an alternative transmission route determined by the negotiating agent after the data corruption rate exceeds the preset corruption threshold. It is different from the original transmission path in order to avoid congested or high-interference channels.

[0155] Understandably, step S2053 involves negotiating and determining a retransmission path through a negotiation agent and triggering retransmission. That is, the power distribution terminal and the base station interact through signaling to reselect a transmission route with lower latency and higher reliability, and initiate the retransmission of data based on the route.

[0156] S2054: Adjust the transmission quality of service identifier parameters based on end-to-end delay and retransmission path; It should be noted that the Transmission Service Quality Identifier parameter is an identifier used in fifth-generation mobile communication networks to distinguish service priority and resource guarantee level, and it determines the priority order of data packet scheduling and resource allocation strategy.

[0157] Understandably, step S2054 adjusts the transmission service quality identifier parameters based on end-to-end delay and retransmission path. That is, based on the calculated total transmission delay and the retransmission route characteristics determined by renegotiation, the service level identifier of the data stream is dynamically modified so that critical fault data can obtain higher scheduling priority and resource guarantee.

[0158] S2055: Determine slice allocation weights based on priority scores, game optimization results, bandwidth utilization, and end-to-end latency; It should be noted that bandwidth utilization is the ratio of occupied bandwidth to total available bandwidth in a dedicated network slice, reflecting the current scarcity of slice resources.

[0159] In addition, the slice allocation weight is a weighted coefficient calculated by comprehensively considering priority scores, game optimization results, bandwidth utilization and end-to-end latency. It is used to quantify the relative importance of different factors in slice allocation.

[0160] Understandably, step S2055 determines the slice allocation weight based on the priority score, game optimization result, bandwidth utilization, and end-to-end latency. Specifically, it involves weighting the priority score obtained from knowledge graph reasoning, the game optimization result output from non-cooperative game, the bandwidth occupancy ratio of the current slice, and the calculated total latency to obtain a comprehensive weight value used to guide resource allocation.

[0161] S2056: Allocate dedicated network slice resources based on slice allocation weights and path optimization results, and perform bandwidth isolation on the dedicated network slice resources to obtain isolated network slice resources.

[0162] Understandably, step S2056 involves allocating dedicated network slice resources based on slice allocation weights and path optimization results, and performing bandwidth isolation on the dedicated network slice resources to obtain isolated network slice resources. That is, based on the determined slice allocation weights and path optimization results obtained from multi-agent reinforcement learning, a dedicated logical channel is allocated in the fifth-generation mobile communication network, and bandwidth isolation is implemented on the channel to obtain isolated network slice resources with high reliability and low latency characteristics.

[0163] In practical implementation, under 5G Non-Standalone (NSA) network mode, data transmission latency consists of propagation latency, processing latency, and queuing latency. Traditional methods lack adaptability to dynamic network environments. When a large-scale fault occurs in the distribution network, hundreds or thousands of distribution terminals simultaneously send alarm traffic to the base station, causing instantaneous network congestion (i.e., "burst traffic flooding"). This embodiment treats each distribution automation terminal (FTU / DTU) and 5G access network base station as independent agents in reinforcement learning. This embodiment introduces Multi-Agent Reinforcement Learning (MARL), treating each distribution terminal and base station as an agent, and dynamically optimizing latency parameters through distributed learning to predict and reduce end-to-end latency.

[0164] Constructing a latency prediction model: Based on a distributed strategy of MARL, the agent learns the variation pattern of latency components by interacting with the environment and dynamically adjusts the transmission path and resource allocation.

[0165] Combine real-time QoS parameters: Utilize the 5QI (5G QoS Identifier) ​​metric of the 5G network to dynamically update the weights of processing latency and queuing latency.

[0166] The delay optimization formula is as follows:

[0167] piece, Total end-to-end delay (ms); Propagation delay (e.g., at a distance of 510m) ms; Processing delays; Queue delay; Prediction delay bias (ms, a dynamic adjustment term obtained through MARL prediction); where, prediction delay bias The physical essence of the power grid is that the intelligent agent learns the traffic surge pattern during historical faults through the MARL algorithm and predicts in advance the base station channel queuing delay jitter caused by the sudden traffic surge due to the fault. The weighting coefficients are adjusted in real time through MARL learning to satisfy... .

[0168] To further optimize latency, a latency prediction and path selection strategy based on MARL is introduced:

[0169] in, Optimal latency after optimization; Path selection strategy; Path strategy space; path The reward function; path End-to-end delay; Number of paths. Within the critical milliseconds of fault recovery, the terminal uses the reward function. Real-time evaluation and selection of the 5G wireless transmission path with the lowest packet loss rate and lowest latency. Avoid congested base station logical channels.

[0170] This embodiment constructs a dedicated knowledge graph (KG) for power distribution automation faults. Entities in the graph strictly correspond to: substation outgoing switches, sectionalizing circuit breakers, tie switches, Class I users (e.g., hospitals / government), Class II users, short-circuit faults, and grounding faults. Relationships in the graph strictly correspond to: upstream and downstream topological relationships, power supply priority, and delay constraints.

[0171] Construct a knowledge graph for power distribution automation: The graph includes entities (terminal equipment, fault types, base stations), relationships (priority, bandwidth requirements, delay constraints), and attributes (real-time status, historical data).

[0172] Dynamic slice allocation: Based on knowledge graph reasoning, combined with the priority of fault data Based on network conditions, dynamically adjust slice resources. The optimized allocation formula is as follows:

[0173] in, Assigned to data packets Network slice index; Network slice collection; slice Available bandwidth; slice Utilization rate; Data packets based on knowledge graph computing Priority score; slice For data packets The contribution of delay; The weighting coefficients sum to 1.

[0174] Predict the impact range of faults using graph inference and optimize slice scheduling:

[0175] in, Related entity set; Entity impact factor (0–1); Urgency level (0–1); Connectivity in the graph.

[0176] When data packet When a trip control command for a certain switch is included, the knowledge graph uses semantic reasoning to discover that a failure to disconnect the switch will trigger a trip of the upper-level main transformer protection (i.e., an impact factor). and urgency (Extremely high), thus assigning the data packet an extremely high priority score, forcing the 5G network to allocate a dedicated isolation slice for it. This ensures that control commands are never lost.

[0177] When high-voltage line failure or flashover causes partial data corruption (data corruption rate) When the signal strength increases, do not blindly retransmit, otherwise it will exacerbate network congestion.

[0178] Data integrity detection: Real-time monitoring of data corruption rate at edge nodes The retransmission threshold is dynamically adjusted based on data priority. Adaptive retransmission strategy: If... This triggers a retransmission; the dynamic calculation formula for the threshold is as follows:

[0179] in, Dynamic retransmission threshold; Base threshold (set to 0.07%); Data priority; Current network latency; Historical average delay; Adjustment coefficient.

[0180] If the current data is an extremely important fault trip output control command ( (Very large), dynamic retransmission threshold The threshold will be automatically lowered, which means that the system has a very low fault tolerance rate for critical instructions, and any damage will immediately trigger a retransmission; conversely, if it is ordinary electrical energy acquisition data, the threshold will be relaxed to avoid occupying valuable fault repair communication bandwidth.

[0181] Introducing an edge node collaborative verification mechanism to reduce retransmission overhead:

[0182] in, Retransmission coordination costs; Set of edge nodes; distance; Load rate; Calculate the cost; Weighting coefficients.

[0183] During retransmission, the terminal does not need to request data from the cloud master station thousands of miles away, but instead receives it directly from the nearest power distribution edge gateway node. Data retransmission and collaborative verification significantly reduced the round-trip time for emergency repair orders.

[0184] This embodiment acquires status data from distribution terminals and base stations, providing an accurate operational basis for subsequent network load prediction and transmission path adjustment. By constructing a multi-agent reinforcement learning environment and predicting network load, dynamic optimization of transmission paths is achieved, effectively reducing data propagation and processing latency. By mapping fault types, terminal states, and network resources into a graph structure and performing semantic reasoning, a scientific priority score is obtained, ensuring that critical fault data receives priority transmission resources. By modeling the slice allocation problem as a non-cooperative game and calculating the reward function based on power proportion and bandwidth resources, competitive optimization of network slice resources is achieved, improving resource utilization efficiency. By calculating propagation latency, processing latency, queuing latency, and cooperation latency to obtain end-to-end latency, monitoring data integrity, and initiating multi-agent negotiation and retransmission when the data corruption rate exceeds a preset corruption threshold, the transmission service quality identification parameters are dynamically adjusted to determine reasonable slice allocation weights. Ultimately, precise allocation and bandwidth isolation of dedicated network slice resources are achieved, effectively avoiding network congestion and loss of critical control commands, ensuring the integrity and real-time performance of data transmission, and thus achieving high reliability and low latency in communication transmission during distribution network fault isolation and recovery.

[0185] This embodiment refers to Figure 3 , Figure 3 This is a flowchart illustrating the third embodiment of the power distribution automation terminal control optimization method of this application.

[0186] In one feasible implementation, step S40 includes: Step S401: Construct a power distribution network diagram based on the terminal nodes and communication connections in the power distribution network. The terminal nodes and communication connections constitute the power distribution network topology. It should be noted that a terminal node is a physical or logical unit in a power distribution network that has independent data acquisition and communication capabilities. Its corresponding power distribution terminal is represented in a graph structure and is used to carry node characteristics and topological association information.

[0187] In addition, communication connection relationships are a description of the electrical communication or information exchange paths between various terminal nodes in a power distribution network, which includes the establishment method and transmission attributes of wired or wireless connections.

[0188] Furthermore, a power distribution network graph is a graph data structure consisting of terminal nodes as vertices and communication connections as edges. It is used to formally describe the topological layout and information flow of a power distribution network.

[0189] Understandably, step S401 involves constructing a power distribution network graph based on the terminal nodes and communication connections in the power distribution network. This means identifying all terminal devices with communication capabilities in the power distribution network and their interconnection paths, abstracting the terminal devices as vertices and the connection paths as edges, and constructing a power distribution network graph for graph neural network computation.

[0190] Step S402: Input the fused feature set as the node feature input of each terminal node in the power distribution network diagram into the graph neural network; It should be noted that node features are attribute vectors assigned to each terminal node in the power distribution network diagram. In this embodiment, they are provided by a fused feature set and are used to characterize the operating status of the power distribution terminal corresponding to the node.

[0191] Understandably, step S402 involves using the fused feature set as the node feature input of each terminal node in the power distribution network diagram into the graph neural network. That is, the feature vector obtained after fusion processing is assigned to the corresponding terminal node in the power distribution network diagram as the initial input for message passing and feature aggregation of the graph neural network.

[0192] Step S403: Extract the features of the neighboring nodes of the current terminal node through a graph neural network, and perform aggregation transformation and nonlinear mapping on the neighboring node features through a weight matrix to obtain the hidden layer features; Understandably, step S403 involves extracting the features of neighboring nodes of the current terminal node through a graph neural network, and then performing aggregation transformation and nonlinear mapping on the neighboring node features through a weight matrix to obtain hidden layer features. That is, the feature information of the neighboring nodes directly connected to the current terminal node is collected using the message passing mechanism of the graph neural network, the collected neighboring features are weighted, aggregated and transformed through a weight matrix, and then nonlinearly mapped to generate high-dimensional hidden layer features for subsequent control strategy reasoning.

[0193] In one feasible implementation, step S403 includes: S4031: The features of neighboring nodes are linearly transformed by the weight matrix to obtain the transformed features; It should be noted that the transformed features are intermediate representations obtained by linearly weighting the features of neighboring nodes using the weight matrix. Their dimension is consistent with the column space of the weight matrix and is used for subsequent normalization and aggregation.

[0194] Understandably, step S4031 involves performing a linear transformation on the features of neighboring nodes using a weight matrix to obtain the transformed features. In other words, the learnable weight matrix is ​​used to perform matrix multiplication on the features of neighboring nodes, mapping the original features to a new feature space and generating the transformed features.

[0195] S4032: Normalize and aggregate the transformed features based on the number of nodes in the neighbor node set to obtain aggregated features; Understandably, step S4032 involves normalizing and aggregating the transformed features based on the number of nodes in the neighbor node set to obtain the aggregated features. This involves counting the total number of neighbor nodes of the current terminal node, summing the transformed features of each neighbor node, and dividing by the number of neighbor nodes to eliminate the influence of differences in the number of neighbors on the feature scale, thus obtaining the normalized aggregated features.

[0196] S4033: Concatenate the aggregated features with the local features of the current terminal node to obtain the concatenated features; Understandably, step S4033 involves concatenating the aggregated features with the local features of the current terminal node to obtain the concatenated features. In other words, the normalized aggregated neighbor features are concatenated with the node features of the current terminal node itself in the feature dimension to form concatenated features that simultaneously cover neighborhood information and its own state.

[0197] S4034: The concatenated features are non-linearly mapped using an activation function to generate hidden layer features.

[0198] Understandably, step S4034 generates hidden layer features by performing nonlinear mapping on the concatenated features through an activation function. That is, the activation function is used to perform element-wise nonlinear transformation on the concatenated features, introducing nonlinear expressive power to generate the hidden layer features of the current layer of the graph neural network.

[0199] Step S404: Determine the optimization control strategy corresponding to the hidden layer features by minimizing the loss function.

[0200] Understandably, step S404 determines the optimal control strategy corresponding to the hidden layer features by minimizing the loss function. That is, it defines a loss function that measures the deviation between the predicted value of the control strategy and the expected target, and reduces the deviation to a minimum through an optimization algorithm, thereby deriving the optimal control strategy that minimizes the loss function from the hidden layer features.

[0201] In the specific implementation, the concept of multi-agent system and game theory collaborative optimization is introduced to model the network slice allocation and delay optimization problem as a distributed multi-agent game problem.

[0202] In distribution network automation relay transmission or collaborative protection (such as distributed FA), multiple adjacent switch terminals need to communicate with each other to confirm the fault range.

[0203] Delay Calculation and Optimization Model: Total Delay The calculation formula is:

[0204] in, This indicates the additional coordination delay introduced by multi-agent collaboration. The coordination weighting coefficients are used to balance the collaboration costs and the benefits of delay optimization. Among them, the coordination delay... The essence of the power technology is the time consumed by adjacent circuit breaker terminals (intelligent agents) to conduct peer-to-peer (P2P) negotiations on the 5G side in order to perform "relay protection blocking signaling interaction" or "network-wide distributed fault collaborative location".

[0205] Multi-agent cooperation mechanism: Based on distributed cooperation, the terminal agent adjusts its data transmission strategy through reinforcement learning to minimize [data loss / damage]. The collaborative objective function is defined as follows:

[0206] in, Indicates terminal The strategy (including sending frequency, priority, etc.) This represents the set of policies for other terminals. For terminal set, For the terminal End-to-end delay.

[0207] Each terminal agent adjusts its own transmission strategy. (Such as retransmission frequency and fixed value reporting timing) to minimize the total latency of all terminals participating in fault isolation across the entire network and prevent control commands from colliding with each other.

[0208] Imagine 5G sliced ​​bandwidth as a road. In the event of a sudden failure, the "fault control flow" (high priority) and the "normal telemetry sampling flow" (low priority) will simultaneously compete for lanes. This method models this as a non-cooperative game.

[0209] Slice Allocation Game Model: The network slice allocation problem is modeled as a non-cooperative game. The slice allocation formula is improved as follows:

[0210] in, For the terminal Priority weights, The game payoff function is defined as follows:

[0211] in, Prioritize terminal data. As a priority threshold, For end-user demand, To be assigned to a slice The terminal set, To reduce slice switching costs, To switch the penalty coefficient.

[0212] If a certain business slice The inside was crammed with ordinary electricity meter data collection flow ( (Very large), its rate of return It will decrease; and the critical fault trip control flow ( (Extremely large) It possesses absolute power in the competition, and it will exert influence through penalties. This forces low-priority traffic to automatically "give way" (switch to other ordinary slices), thereby freeing up 5G core slice resources for fault circuit interrupters.

[0213] 5. Dynamic adjustment of QI parameters: During the game, the adjustment formula is as follows:

[0214] in, As the baseline value, To adjust the coefficient, This represents the upper limit of utilization.

[0215] In the dynamic equilibrium of the game, based on the current channel utilization... Based on the urgency of power data, the 5QI parameters of this control channel are dynamically adjusted stepwise towards higher service quality.

[0216] Data corruption rate monitoring and retransmission trigger: Data corruption rate .like This triggers a retransmission. When the data corruption rate... When a retransmission is triggered, the terminal agents negotiate in a distributed manner using the Contract Net Protocol. The retransmission path selection is based on multi-agent negotiation, and the optimization objective is:

[0217] in, This is the total path delay. For path energy consumption, This is the energy consumption weighting coefficient.

[0218] Multi-agent negotiation mechanism: Terminal agents select retransmission paths and slice resources through a distributed negotiation protocol (such as the Contract Network protocol) to ensure that the impact on other high-priority data transmission is minimized, and output optimized retransmission data. To the edge node.

[0219] The malfunctioning FTU acts as the "bidder," issuing "bids" to nearby DTUs or edge gateways that are unaffected by the fault and have idle channels; surrounding terminals respond according to their own current channel latency. and remaining computing energy consumption A bidding process will be conducted. Ultimately, a green retransmission path that will have the least impact on the power supply to other core loads will be selected. Optimized retransmitted data It delivers data quickly to edge nodes, completely avoiding the technical risk of the entire distribution network automation master station system being paralyzed due to blind retransmission.

[0220] This embodiment constructs a distribution network graph based on the terminal nodes and communication connections in the distribution network, formalizing the complex physical topology and communication routing relationships of the distribution network into a graph structure, providing a structural foundation for subsequent topology sensing. By inputting the fused feature set as the node features of each terminal node in the distribution network graph into a graph neural network, deep fusion of multi-source operational data and network topology is achieved, enabling each node to possess a computable state representation. By extracting features from neighboring nodes and performing linear transformation of the weight matrix, normalization aggregation, local feature concatenation, and nonlinear mapping of the activation function, neighborhood topology information is effectively aggregated while preserving the node's own state, generating a high-dimensional hidden layer feature that contains local correlations and global evolutionary laws. By minimizing the loss function to determine the optimal control strategy corresponding to the hidden layer features, the control decision can approach the optimal target, improving the accuracy of the strategy output. This overcomes the shortcomings of traditional algorithms in combining the complex physical topology and communication routing of the distribution network, achieving predictive optimization and improving the accuracy and response speed of the control strategy.

[0221] Reference Figure 4 , Figure 4 This is a structural block diagram of the first embodiment of the power distribution automation terminal control optimization device of this application.

[0222] like Figure 4 As shown, the power distribution automation terminal control optimization device proposed in this application includes: The data acquisition module 10 is used to acquire real-time operating data of the power distribution terminal; The transmission allocation module 20 is used to allocate dedicated network slice resources according to the data priority of real-time running data, and to isolate the dedicated network slice resources by bandwidth to obtain isolated network slice resources. The fusion extraction module 30 is used to fuse real-time operating data of multiple power distribution terminals based on isolated network slice resources to obtain a fusion feature set; The strategy generation module 40 is used to generate an optimized control strategy based on the fused feature set and the power distribution network topology through a graph neural network. The control module 50 is used to generate control commands based on the optimized control strategy and send them to the power distribution terminal so that the power distribution terminal can perform fault isolation and recovery operations.

[0223] Based on the first embodiment of the power distribution automation terminal control optimization device described in this application, a second embodiment of the power distribution automation terminal control optimization device of this application is proposed.

[0224] In this embodiment, the data acquisition module 10 is also used to acquire voltage data, current data and power data of the power distribution terminal; Voltage data, current data, and power data are sampled according to a preset sampling frequency to obtain sampled data; Perform Kalman filtering on the sampled data to obtain the filtered data; The abnormal data in the filtered data are marked according to the preset voltage deviation threshold to obtain the marked data; Real-time running data is generated based on the tagged data.

[0225] In this embodiment, the transmission distribution module 20 is also used to acquire status data of the power distribution terminal and the base station; A multi-agent reinforcement learning environment is constructed based on state data. The network load is predicted through the multi-agent reinforcement learning environment, and the transmission path is adjusted according to the predicted network load to obtain the path optimization result. The fault types, terminal states, and network resources of the power distribution system are mapped into a graph structure, and semantic reasoning is performed based on the relationships between entities in the graph structure to obtain priority scores. The slice allocation problem is modeled as a non-cooperative game, and the reward function is calculated based on the power ratio and bandwidth resources of each agent to obtain the game optimization result. Dedicated network slice resources are allocated based on priority scoring, game optimization results, and path optimization results, and bandwidth isolation is performed on the dedicated network slice resources to obtain isolated network slice resources.

[0226] In this embodiment, the transmission allocation module 20 is further configured to determine the end-to-end delay based on the propagation delay, processing delay, queuing delay, and cooperation delay; Monitor data integrity; when the data corruption rate exceeds the preset corruption threshold, the power distribution terminal and base station will be used as negotiating agents. The retransmission path is determined through negotiation by the intelligent agent, and retransmission is triggered. Adjust the transmission quality of service (QoS) identifier parameters based on end-to-end delay and retransmission paths; The slice allocation weights are determined based on priority scores, game optimization results, bandwidth utilization, and end-to-end latency. Dedicated network slice resources are allocated based on slice allocation weights and path optimization results, and bandwidth isolation is performed on the dedicated network slice resources to obtain isolated network slice resources.

[0227] In this embodiment, the fusion extraction module 30 is also used to obtain real-time operating data of multiple power distribution terminals through isolated network slice resources; The normalization weights are determined based on the signal strength of each power distribution terminal; Multiple real-time running data are weighted and fused according to normalized weights to obtain a fused feature vector; Calculate the power fluctuation rate based on the power data within a preset time window; A fusion feature set is generated based on the fusion feature vector and power volatility.

[0228] In this embodiment, the strategy generation module 40 is also used to construct a power distribution network diagram based on the terminal nodes and communication connection relationships in the power distribution network, wherein the terminal nodes and communication connection relationships constitute the power distribution network topology. The fused feature set is used as the node feature input of each terminal node in the power distribution network graph into the graph neural network; The features of the neighboring nodes of the current terminal node are extracted by a graph neural network, and the features of the neighboring nodes are aggregated, transformed and nonlinearly mapped by a weight matrix to obtain the hidden layer features. By minimizing the loss function, an optimal control strategy corresponding to the hidden layer features is determined.

[0229] In this embodiment, the strategy generation module 40 is also used to perform a linear transformation on the features of neighboring nodes through a weight matrix to obtain the transformed features; The transformed features are normalized and aggregated based on the number of nodes in the neighbor node set to obtain the aggregated features. The aggregated features are concatenated with the local features of the current terminal node to obtain the concatenated features. Hidden layer features are generated by performing a non-linear mapping on the concatenated features using an activation function.

[0230] Other embodiments or specific implementations of the power distribution automation terminal control optimization device of this application can be found in the above-described method embodiments, and will not be repeated here.

[0231] This application provides a power distribution automation terminal control optimization device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the power distribution automation terminal control optimization method in the above embodiment 1.

[0232] The following reference Figure 5This document illustrates a structural schematic diagram of a power distribution automation terminal control optimization device suitable for implementing embodiments of this application. The power distribution automation terminal control optimization device in the embodiments of this application may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Description), PMPs (Portable Media Players), vehicle terminals (e.g., vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 5 The power distribution automation terminal control optimization device shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0233] like Figure 5 As shown, the power distribution automation terminal control optimization device may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory 1002 or a program loaded from a storage device 1003 into a random access memory 1004. The random access memory 1004 also stores various programs and data required for the operation of the power distribution automation terminal control optimization device. The processing unit 1001, the read-only memory 1002, and the random access memory 1004 are interconnected via a bus 1005. An input / output interface 1006 is also connected to the bus. Typically, the following systems can be connected to the input / output interface 1006: input devices 1007 including, for example, a touch screen, touchpad, keyboard, mouse, image sensor, microphone, accelerometer, gyroscope, etc.; output devices 1008 including, for example, a liquid crystal display (LCD), speaker, vibrator, etc.; storage devices 1003 including, for example, magnetic tape, hard disk, etc.; and communication devices 1009. Communication device 1009 allows the distribution automation terminal control optimization equipment to communicate wirelessly or wiredly with other devices to exchange data. Although the figures show distribution automation terminal control optimization equipment with various systems, it should be understood that implementation or possession of all the systems shown is not required. More or fewer systems may be implemented alternatively.

[0234] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from read-only memory 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0235] The distribution automation terminal control optimization device provided in this application, employing the distribution automation terminal control optimization method in the above embodiments, can solve the technical problem of overcoming network congestion and loss of key commands during distribution network fault isolation and recovery, and achieving control strategy optimization. Compared with the prior art, the beneficial effects of the distribution automation terminal control optimization device provided in this application are the same as those of the distribution automation terminal control optimization method provided in the above embodiments, and other technical features in this distribution automation terminal control optimization device are the same as those disclosed in the previous embodiment method, and will not be repeated here.

[0236] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0237] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0238] This application provides a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, which are used to execute the power distribution automation terminal control optimization method in the above embodiments.

[0239] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems or devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0240] The aforementioned computer-readable storage medium may be included in the power distribution automation terminal control optimization equipment; or it may exist independently and not be assembled into the power distribution automation terminal control optimization equipment.

[0241] The aforementioned computer-readable storage medium carries one or more programs that, when executed by the distribution automation terminal control and optimization device, enable the distribution automation terminal control and optimization device to write computer program code for performing the operations of this application in one or more programming languages ​​or a combination thereof. These programming languages ​​include object-oriented programming languages ​​such as Java, Smalltalk, and C++; and also conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network, such as a local area network (LAN) or a wide area network (WAN), or connected to an external computer (e.g., via the Internet using an Internet service provider).

[0242] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0243] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0244] The readable storage medium provided in this application is a computer-readable storage medium that stores computer-readable program instructions (i.e., a computer program) for executing the above-described distribution automation terminal control optimization method. This solves the technical problem of overcoming network congestion and loss of key instructions during distribution network fault isolation and recovery, and optimizing control strategies. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in this application are the same as those of the distribution automation terminal control optimization method provided in the above embodiments, and will not be repeated here.

[0245] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the power distribution automation terminal control optimization method described above.

[0246] The computer program product provided in this application can solve the technical problem of overcoming network congestion and loss of key instructions in the process of power distribution network fault isolation and recovery, and optimizing the control strategy. Compared with the prior art, the beneficial effects of the computer program product provided in this application are the same as those of the power distribution automation terminal control optimization method provided in the above embodiments, and will not be repeated here.

[0247] The above description is only a part of the embodiments of this application and does not limit the scope of protection of this application. All equivalent structural transformations made under the technical concept of this application and using the content of this application specification and drawings, or direct / indirect applications in other related technical fields, are included in the scope of protection of this application.

Claims

1. A method for optimizing terminal control in power distribution automation, characterized in that, The method includes: Obtain real-time operating data from the power distribution terminal; Dedicated network slice resources are allocated based on the data priority of the real-time running data, and the dedicated network slice resources are bandwidth isolated to obtain isolated network slice resources. Based on the isolated network slice resources, the real-time operation data of multiple power distribution terminals are fused to obtain a fused feature set; Based on the fused feature set and the power distribution network topology, an optimized control strategy is generated using a graph neural network. Control commands are generated according to the optimized control strategy and sent to the power distribution terminal so that the power distribution terminal can perform fault isolation and recovery operations.

2. The method as described in claim 1, characterized in that, The step of acquiring real-time operating data of the power distribution terminal includes: Acquire voltage, current, and power data from the power distribution terminal; The voltage data, current data, and power data are sampled according to a preset sampling frequency to obtain sampled data. The sampled data is subjected to Kalman filtering to obtain filtered data; Abnormal data in the filtered data are marked according to a preset voltage deviation threshold to obtain marked data; Real-time running data is generated based on the marked data.

3. The method as described in claim 1, characterized in that, The step of allocating dedicated network slice resources according to the data priority of the real-time running data, and performing bandwidth isolation on the dedicated network slice resources to obtain isolated network slice resources includes: Obtain the status data of the power distribution terminal and the base station; A multi-agent reinforcement learning environment is constructed based on the state data. The network load is predicted through the multi-agent reinforcement learning environment, and the transmission path is adjusted according to the predicted network load to obtain the path optimization result. The fault types, terminal states, and network resources of the power distribution system are mapped into a graph structure, and semantic reasoning is performed based on the relationships between entities in the graph structure to obtain a priority score. The slice allocation problem is modeled as a non-cooperative game, and the reward function is calculated based on the power ratio and bandwidth resources of each agent to obtain the game optimization result. Dedicated network slice resources are allocated based on the priority score, the game optimization result, and the path optimization result, and the dedicated network slice resources are bandwidth isolated to obtain isolated network slice resources.

4. The method as described in claim 3, characterized in that, The step of allocating dedicated network slice resources based on the priority score, the game optimization result, and the path optimization result, and performing bandwidth isolation on the dedicated network slice resources to obtain isolated network slice resources, includes: End-to-end latency is determined based on propagation latency, processing latency, queuing latency, and collaboration latency. The system monitors data integrity, and when the data corruption rate exceeds a preset corruption threshold, the power distribution terminal and the base station are used as negotiating agents. The retransmission path is determined through negotiation by the negotiation agent, and retransmission is triggered. Adjust the transmission service quality identifier parameters according to the end-to-end delay and the retransmission path; The slice allocation weight is determined based on the priority score, the game optimization result, the bandwidth utilization, and the end-to-end latency. Dedicated network slice resources are allocated based on the slice allocation weights and the path optimization results, and the dedicated network slice resources are bandwidth isolated to obtain isolated network slice resources.

5. The method as described in claim 1, characterized in that, The step of fusing the real-time operating data of multiple power distribution terminals based on the isolated network slice resources to obtain a fused feature set includes: The real-time operating data of multiple power distribution terminals are obtained through the isolated network slice resources; The normalization weights are determined based on the signal strength of each of the power distribution terminals. The multiple real-time running data are weighted and fused according to the normalized weights to obtain a fused feature vector; Calculate the power fluctuation rate based on the power data within a preset time window; A fusion feature set is generated based on the fusion feature vector and the power volatility.

6. The method as described in claim 1, characterized in that, The step of generating an optimized control strategy using a graph neural network based on the fused feature set and the distribution network topology includes: A power distribution network diagram is constructed based on the terminal nodes and communication connections in the power distribution network, wherein the terminal nodes and communication connections constitute the power distribution network topology. The fused feature set is used as the node feature input to the graph neural network for each terminal node in the power distribution network graph; The graph neural network extracts the features of the neighboring nodes of the current terminal node, and the neighboring node features are aggregated, transformed and nonlinearly mapped by a weight matrix to obtain the hidden layer features. By minimizing the loss function, an optimal control strategy corresponding to the hidden layer features is determined.

7. The method as described in claim 6, characterized in that, The step of extracting the features of the neighboring nodes of the current terminal node through the graph neural network, and performing aggregation transformation and nonlinear mapping on the neighboring node features through a weight matrix to obtain the hidden layer features includes: The neighbor node features are linearly transformed using a weight matrix to obtain the transformed features. The transformed features are normalized and aggregated based on the number of nodes in the neighbor node set to obtain aggregated features; The aggregated features are concatenated with the local features of the current terminal node to obtain the concatenated features; Hidden layer features are generated by performing a non-linear mapping on the concatenated features using an activation function.

8. A power distribution automation terminal control optimization device, characterized in that, The device includes: The data acquisition module is used to acquire real-time operating data of the power distribution terminal; The transmission allocation module is used to allocate dedicated network slice resources according to the data priority of the real-time running data, and to perform bandwidth isolation on the dedicated network slice resources to obtain isolated network slice resources. The fusion extraction module is used to fuse the real-time operating data of multiple power distribution terminals based on the isolated network slice resources to obtain a fusion feature set; The strategy generation module is used to generate an optimized control strategy based on the fused feature set and the power distribution network topology using a graph neural network. The control module is used to generate control commands according to the optimized control strategy and send them to the power distribution terminal so that the power distribution terminal can perform fault isolation and recovery operations.

9. A power distribution automation terminal control optimization device, characterized in that, The device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the power distribution automation terminal control optimization method as described in any one of claims 1 to 7.

10. A storage medium, characterized in that, The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, it implements the steps of the power distribution automation terminal control optimization method as described in any one of claims 1 to 7.

Images