Permanent magnet recloser power supply circuit design optimization method and system based on deep learning

Abstract

The invention relates to the technical field of power supply circuits, and discloses a permanent magnet recloser power supply circuit design optimization method and system based on deep learning. The method comprises the following steps: a power distribution network intelligent sensor collects electrical parameters of each node of the permanent magnet recloser to form a power supply state data set; the permanent magnet recloser dedicated topology sensing network extracts power supply path features and generates a topology feature matrix; performing multi-target constraint reconstruction based on the feature matrix to obtain a circuit topology reconstruction scheme; optimizing the power supply parameters by a deep reinforcement learning algorithm to obtain power supply circuit optimal configuration; and performing closed-loop regulation processing to generate an intelligent power supply control signal. The technical problem that an existing permanent magnet recloser power supply circuit is lack of intelligent topology perception, multi-target coordination optimization and adaptive parameter adjustment capability is solved.

Description

Technical Field

[0001] This application relates to the field of power supply circuit technology, and in particular to a method and system for designing and optimizing the power supply circuit of a permanent magnet recloser based on deep learning. Background Technology

[0002] The design of permanent magnet recloser power supply circuits mainly relies on traditional power system analysis methods and rule-based control strategies. Automatic power restoration after distribution network faults is achieved through preset protection settings and fixed reclosing logic. Traditional methods employ static circuit parameter configuration and a single fault diagnosis criterion, combined with time-current characteristic curves for coordinating the operation of the permanent magnet recloser. Simultaneously, SCADA systems are used for basic operational status monitoring and manual dispatching decisions. This approach can meet basic power supply reliability requirements in relatively simple distribution network structures.

[0003] However, existing data acquisition systems often suffer from problems such as missing data, noise interference, and incomplete coverage of operating conditions, resulting in inaccurate understanding of the operating status of permanent magnet reclosers. Secondly, there are issues with model interpretability and reliability. Traditional rule-based control methods lack a deep understanding of complex power supply network topology and struggle to handle nonlinear coupling relationships between multiple nodes. Thirdly, there are issues with real-time performance and adaptability. Fixed parameter configurations cannot adapt to dynamic changes in distribution network load and equipment aging, and lack sufficient flexibility when facing new operating environments or abnormal conditions.

[0004] The problem with existing technologies lies in the lack of a comprehensive technical solution capable of intelligently sensing changes in distribution network topology, adaptively learning optimal control strategies for permanent magnet reclosers, and achieving multi-objective coordinated optimization. Specifically, key technical issues that urgently need to be addressed include: how to establish a dedicated topology sensing mechanism for permanent magnet reclosers to accurately identify the correlation between power supply paths; how to construct a circuit topology reconstruction method under multi-objective constraints to balance power supply reliability and power quality; and how to achieve deep learning-based adaptive parameter optimization to cope with complex and changing operating conditions. Summary of the Invention

[0005] This application provides a deep learning-based method and system for optimizing the design of permanent magnet recloser power supply circuits. This method addresses the technical problems in existing permanent magnet recloser power supply circuit designs, which lack intelligent topology awareness, multi-objective coordinated optimization, and adaptive parameter adjustment capabilities.

[0006] Firstly, this application provides a deep learning-based method for optimizing the design of a permanent magnet recloser power supply circuit. This method includes: acquiring and processing multi-dimensional electrical parameters of each power supply node of the permanent magnet recloser using intelligent sensors in the distribution network to obtain a power supply status dataset containing node voltage amplitude, branch current, and power factor; inputting the power supply status dataset into a dedicated topology sensing network for the permanent magnet recloser to extract power supply path features, resulting in a power supply path feature matrix containing local power supply correlation and global network topology; performing multi-objective constraint reconstruction processing on the power supply circuit topology of the permanent magnet recloser based on the power supply path feature matrix to obtain a circuit topology reconstruction scheme containing optimal power supply path combinations and load allocation strategies; inputting the circuit topology reconstruction scheme into a deep reinforcement learning algorithm for power supply parameter optimization processing, resulting in an optimized power supply circuit configuration containing optimal power supply path switching strategies and voltage regulation parameters; and performing closed-loop regulation processing on the real-time power supply parameters of the permanent magnet recloser based on the optimized power supply circuit configuration to obtain an intelligent power supply control signal containing fault prediction results and power supply switching commands.

[0007] Optionally, the process of acquiring and processing multi-dimensional electrical parameters of each power supply node of the permanent magnet recloser using intelligent sensors in the distribution network to obtain a power supply status dataset including node voltage amplitude, branch current, and power factor includes: Based on the action response window of the permanent magnet recloser, the original electrical signals of each power supply node in the distribution network are adaptively segmented to obtain segmented electrical data containing pre-reclosing warning features and post-reclosing steady-state features. The segmented electrical data is input into a power supply node correlation analysis algorithm for topology correlation filtering to obtain a set of core power supply node electrical parameters after removing redundant node information. The electrical parameter set of the core power supply node is marked with importance according to the power supply path weight allocation strategy of the permanent magnet recloser, so as to obtain hierarchical electrical feature data including the main power supply path and the backup power supply path. The hierarchical electrical feature data is grouped by operating condition based on the power supply status similarity clustering method to obtain a power supply status dataset for training deep learning models.

[0008] Optionally, the step of inputting the power supply status dataset into a dedicated topology sensing network for permanent magnet reclosers for power supply path feature extraction processing, to obtain a power supply path feature matrix containing local power supply correlation and global network topology, includes: The power supply status dataset is input into the power supply adjacency matrix construction layer of the dedicated topology sensing network for permanent magnet reclosers to perform node correlation calculation and obtain a dynamic weighted adjacency matrix based on electrical distance and fault propagation probability. Based on the dynamic weighted adjacency matrix, a small-range power supply topology convolution is performed on the local feature extraction layer of the dedicated topology sensing network for permanent magnet reclosers to obtain a local topology feature map containing power supply association features of adjacent nodes. The local topological feature map is input into the medium receptive field extension layer of the dedicated topological sensing network for permanent magnet reconciliation and then processed by convolution of the power supply network at a medium distance to obtain a medium topological feature map containing the power supply coordination relationship within the region. Based on the aforementioned medium topological feature map, a large-scale power supply network convolution process is performed on the global receptive field extension layer of the dedicated topological sensing network for permanent magnet reclosers to obtain a global topological feature map containing cross-regional power supply influence relationships. Based on the power supply path attention weight mechanism, multi-scale feature fusion processing is performed on the local topology feature map, the intermediate topology feature map, and the global topology feature map to obtain a power supply path feature matrix that includes local power supply correlation and global network topology.

[0009] Optionally, the step of inputting the power supply status dataset into the power supply adjacency matrix construction layer of the dedicated topology sensing network for permanent magnet reclosers for node correlation calculation processing to obtain a dynamically weighted adjacency matrix based on electrical distance and fault propagation probability includes: Based on the opening and closing state transition characteristics of the permanent magnet recloser, the electrical distance measurement of each power supply node in the power supply state dataset is calculated to obtain the inter-node electrical distance matrix reflecting the influence range of the permanent magnet recloser. Based on the fault response time series of the permanent magnet recloser, the electrical distance matrix is ​​subjected to fault propagation path analysis to obtain a fault propagation probability matrix containing the probability of fault propagation from the source node to the target node; The electrical distance matrix and the fault propagation probability matrix are input into a weighted fusion algorithm for weight coefficient allocation, resulting in a node importance weight vector that reflects the power supply priority of the permanent magnet recloser. The importance weight vector of the node is dynamically adjusted based on the power supply switching logic of the permanent magnet recloser to obtain an adaptive weight coefficient that changes with the operating state of the permanent magnet recloser. The power supply node adjacency relationship is weighted according to the adaptive weight coefficient to obtain a dynamic weighted adjacency matrix based on electrical distance and fault propagation probability.

[0010] Optionally, the step of performing multi-objective constraint reconstruction processing on the permanent magnet recloser power supply circuit topology based on the power supply path feature matrix to obtain a circuit topology reconstruction scheme including the optimal power supply path combination and load distribution strategy includes: Based on the power supply reliability constraints of the permanent magnet recloser, the power supply path feature matrix is ​​subjected to power supply path feasibility screening to obtain a set of candidate power supply paths that meet the minimum number of operations required by the permanent magnet recloser. The candidate power supply path set is input into a multi-objective optimization algorithm for power quality constraint analysis to obtain voltage deviation control constraints and harmonic content control constraints. Based on the load carrying capacity limit of the permanent magnet recloser, the load allocation strategy is calculated and processed according to the voltage deviation control constraint and harmonic content control constraint to obtain the optimal power allocation ratio of each power supply branch. Based on the power supply path switching time constraint of the permanent magnet recloser, the optimal power allocation ratio is optimized by power supply timing coordination to obtain the main and backup power supply path switching sequence and switching time interval. The switching sequence of the primary and backup power supply paths is combined with the optimal power allocation ratio to perform topology reconstruction and integration processing, resulting in a circuit topology reconstruction scheme.

[0011] Optionally, the step of inputting the circuit topology reconstruction scheme into a deep reinforcement learning algorithm for power supply parameter optimization to obtain an optimized power supply circuit configuration including the optimal power supply path switching strategy and voltage regulation parameters includes: Based on the power supply state space definition of permanent magnet recloser, the circuit topology reconstruction scheme is processed by state vector encoding to obtain a state vector reflecting the current power supply network topology and load distribution. Based on the spatial constraints of the permanent magnet recloser power supply action, the state vector is subjected to executable action filtering processing to obtain a discrete action set of power supply path switching action and voltage regulation action. The state vector and discrete action set are input into a deep Q network for state-action value evaluation to obtain the long-term revenue prediction value corresponding to each power supply action. Based on the power supply reward mechanism of permanent magnet recloser, the long-term revenue prediction value is optimized by strategy gradient to obtain the optimal power supply path switching strategy and voltage regulation parameters. Based on the optimal power supply path switching strategy and voltage regulation parameters, the power supply configuration parameters are integrated to obtain an optimized power supply circuit configuration.

[0012] Optionally, the step of performing closed-loop adjustment processing on the real-time power supply parameters of the permanent magnet recloser according to the optimized configuration of the power supply circuit to obtain an intelligent power supply control signal containing fault prediction results and power supply switching instructions includes: Based on the permanent magnet recloser operation status monitoring, the power supply circuit optimization configuration is processed in real time to detect parameter deviations and obtain the deviation between the actual value and the target value of the power supply parameters. The deviation is input into the fault symptom identification algorithm for abnormal pattern analysis to obtain the potential fault types and fault occurrence probabilities of the permanent magnet recloser. Based on the potential fault types, a preventive adjustment strategy calculation is performed on the power supply switching decision of the permanent magnet recloser to obtain a preventive power supply path switching scheme before the fault occurs. The preventive power supply path switching scheme is processed in real time to generate execution instructions based on the closed-loop control feedback mechanism of permanent magnet recloser, so as to obtain fault prediction results and power supply switching instructions. The fault prediction results and power supply switching instructions are encapsulated and processed into control signals to obtain intelligent power supply control signals.

[0013] Secondly, this application provides a deep learning-based permanent magnet recloser power supply circuit design optimization system, the deep learning-based permanent magnet recloser power supply circuit design optimization system comprising: The data acquisition module is used to collect and process multi-dimensional electrical parameters of each power supply node of the permanent magnet recloser through intelligent sensors in the distribution network, and obtain a power supply status dataset including node voltage amplitude, branch current, and power factor. The extraction module is used to input the power supply status dataset into a dedicated topology sensing network for permanent magnet reclosers to perform power supply path feature extraction processing, and obtain a power supply path feature matrix that includes local power supply correlation and global network topology. The reconstructing module is used to perform multi-objective constraint reconstruction processing on the power supply circuit topology of the permanent magnet recloser based on the power supply path feature matrix, so as to obtain a circuit topology reconstruction scheme that includes the optimal power supply path combination and load distribution strategy. The optimization module is used to input the circuit topology reconstruction scheme into a deep reinforcement learning algorithm for power supply parameter optimization processing, so as to obtain an optimized power supply circuit configuration that includes the optimal power supply path switching strategy and voltage regulation parameters. The adjustment module is used to perform closed-loop adjustment processing on the real-time power supply parameters of the permanent magnet recloser according to the optimized configuration of the power supply circuit, so as to obtain an intelligent power supply control signal containing fault prediction results and power supply switching instructions.

[0014] Thirdly, a deep learning-based permanent magnet recloser power supply circuit design optimization device is provided, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor invokes the instructions in the memory to cause the deep learning-based permanent magnet recloser power supply circuit design optimization device to execute the aforementioned deep learning-based permanent magnet recloser power supply circuit design optimization method.

[0015] Fourthly, a computer-readable storage medium is provided, wherein instructions are stored therein, which, when executed on a computer, cause the computer to execute the aforementioned deep learning-based permanent magnet recloser power supply circuit design optimization method.

[0016] In the technical solution provided in this application, the deep learning-based permanent magnet recloser power supply circuit design optimization method acquires and processes multi-dimensional electrical parameters through distribution network intelligent sensors, realizing comprehensive perception of key parameters such as voltage amplitude, branch current, and power factor of each power supply node of the permanent magnet recloser. Compared with the traditional single-parameter monitoring method, this technical feature can construct a more complete power supply status dataset, laying a data foundation for subsequent intelligent analysis. At the same time, the power supply path feature extraction and processing of the dedicated topology sensing network for permanent magnet reclosers overcomes the problem of insufficient understanding of complex distribution network topology relationships in traditional methods. Through deep learning algorithms, it automatically identifies local power supply correlations and global network topology features. The resulting power supply path feature matrix can accurately reflect the electrical coupling relationship and fault propagation path between each power supply node, solving the technical problem of insufficient topology sensing capability in the prior art. In addition, the multi-objective constraint reconstruction processing technology realizes the coordinated optimization of power supply reliability and power quality. By comprehensively considering the optimal power supply path combination and load distribution strategy through the circuit topology reconstruction scheme, it avoids the system performance imbalance problem caused by single-objective optimization in traditional methods.

[0017] The core innovation of this application lies in the technical feature of using deep reinforcement learning algorithms for power supply parameter optimization. This algorithm adaptively optimizes the power supply path switching strategy and voltage regulation parameters of permanent magnet reclosers through a state-action value learning mechanism. Compared with traditional control methods based on preset rules, deep reinforcement learning algorithms can autonomously learn the optimal control strategy according to the dynamic changes in the operating state of the distribution network, significantly improving the adaptability and decision-making accuracy of permanent magnet reclosers under complex operating conditions. At the same time, the closed-loop regulation processing technology enables real-time optimization of power supply parameters and fault prevention. By continuously monitoring and feedback adjustment, intelligent power supply control signals are generated. This technology can not only predict potential faults and take preventive measures in advance, but also dynamically adjust the control strategy according to the actual operating effect, fundamentally solving the problems of parameter solidification and response lag in traditional methods. In the specific application field of permanent magnet recloser power supply circuit design optimization, the adaptive learning capability and multi-objective coordinated optimization characteristics of deep learning algorithms enable the entire power supply system to optimize power quality while ensuring power supply reliability, achieving a level of intelligent and refined control that is difficult to achieve with traditional technologies. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1This is a schematic diagram of an embodiment of the deep learning-based permanent magnet recloser power supply circuit design optimization method in this application. Figure 2 This is a schematic diagram of an embodiment of the deep learning-based permanent magnet recloser power supply circuit design optimization system in this application. Figure 3 This is a schematic block diagram of the structure of the permanent magnet recloser power supply circuit design optimization device based on deep learning in an embodiment of the present invention. Detailed Implementation

[0020] This application provides a method and system for optimizing the design of a permanent magnet recloser power supply circuit based on deep learning. The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data used can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0021] For ease of understanding, the specific process of the embodiments of this application is described below. Please refer to [link / reference]. Figure 1 One embodiment of the deep learning-based permanent magnet recloser power supply circuit design optimization method in this application includes: Step S101: Collect and process multi-dimensional electrical parameters of each power supply node of the permanent magnet recloser through the distribution network intelligent sensor to obtain a power supply status dataset including node voltage amplitude, branch current and power factor. Step S102: Input the power supply status dataset into the dedicated topology sensing network for permanent magnet recloser to perform power supply path feature extraction processing, and obtain a power supply path feature matrix that includes local power supply correlation and global network topology. Step S103: Perform multi-objective constraint reconstruction processing on the power supply circuit topology of the permanent magnet recloser based on the power supply path feature matrix to obtain a circuit topology reconstruction scheme that includes the optimal power supply path combination and load distribution strategy. Step S104: Input the circuit topology reconstruction scheme into a deep reinforcement learning algorithm to optimize the power supply parameters and obtain an optimized power supply circuit configuration that includes the optimal power supply path switching strategy and voltage regulation parameters. Step S105: Perform closed-loop adjustment processing on the real-time power supply parameters of the permanent magnet recloser according to the optimized configuration of the power supply circuit to obtain an intelligent power supply control signal that includes fault prediction results and power supply switching instructions.

[0022] It is understood that the executing entity of this application can be a deep learning-based permanent magnet recloser power supply circuit design optimization system, or it can be a terminal or a server; the specific implementation is not limited here. This application's embodiments use a server as an example for illustration.

[0023] Specifically, the smart sensors in the distribution network collect multi-dimensional electrical parameters from each power supply node. This includes adaptive time-series segmentation of the original electrical signals using permanent magnet recloser action response window technology. This technology dynamically adjusts the size of the acquisition window based on the opening and closing characteristics of the permanent magnet recloser, accurately segmenting the continuous electrical signals according to the pre-reclosing warning features and the post-reclosing steady-state features. Then, the power supply node correlation analysis algorithm identifies the electrical coupling relationship between each node, eliminating redundant node information that has little impact on the power supply performance of the permanent magnet recloser. Finally, the power supply state similarity clustering method is used to classify data with similar electrical characteristics, forming a standardized power supply state dataset that includes node voltage amplitude, branch current, and power factor.

[0024] A dedicated topology sensing network for permanent magnet reclosers is used to perform deep feature extraction on the power supply status dataset. The network first calculates the electrical distance and fault propagation probability between nodes through a power supply adjacency matrix construction layer. The electrical distance is calculated based on the opening and closing state transition features of the permanent magnet recloser, reflecting the degree of influence of the permanent magnet recloser's actions on different nodes. The fault propagation probability is obtained from the fault response time series analysis of the permanent magnet recloser. Subsequently, power supply topology convolution processing is performed through multi-level feature extraction layers. The local feature extraction layer captures the direct power supply association features between adjacent nodes, the medium receptive field extension layer identifies the power supply coordination relationship within the region, and the global receptive field extension layer analyzes the power supply influence relationship across regions. Finally, the feature maps of different scales are fused through a power supply path attention weight mechanism to generate a power supply path feature matrix.

[0025] Based on the power supply path feature matrix, a multi-objective constraint reconstruction process is performed. First, candidate power supply paths that meet the minimum number of actions requirement are screened based on the power supply reliability constraints of the permanent magnet recloser. Then, voltage deviation control constraints and harmonic content control constraints are analyzed using a multi-objective optimization algorithm. Next, the optimal power allocation ratio of each power supply branch is calculated based on the load carrying capacity limit of the permanent magnet recloser. Then, the switching sequence and time interval of the main and backup power supply paths are optimized based on the power supply path switching time constraint. Finally, the circuit topology reconstruction scheme is integrated. Step S104 inputs the scheme into a deep reinforcement learning algorithm. The current power supply network topology and load distribution information are converted into a digital format that the algorithm can process through state vector encoding. Then, executable power supply path switching actions and voltage regulation actions are screened. The long-term benefits of each action are evaluated using a deep Q-network. The optimal parameter configuration is obtained through policy gradient optimization. By monitoring the operating status of the permanent magnet recloser in real time to detect power supply parameter deviations, an abnormal mode is analyzed and potential fault types are predicted using a fault symptom identification algorithm. A preventive power supply path switching scheme and execution instructions are generated based on a closed-loop control feedback mechanism. Taking a power distribution network in an industrial park as an example, when the system detects a fluctuation in the voltage amplitude of node A, the dedicated topology sensing network for the permanent magnet recloser identifies the correlation between this fluctuation and the adjacent nodes B and C. It calculates that the electrical distance between nodes A and B is small while the electrical distance between nodes AC and C is large, indicating that node B has a more significant impact on node A. Subsequently, multi-objective constraint reconstruction processing determines the strategy of transferring part of the load from node A to node B. Deep reinforcement learning algorithms optimize the switching timing to avoid transient impacts. The closed-loop regulation system adjusts the subsequent control strategy based on the actual switching effect. The whole process demonstrates the core role of deep learning algorithms in the optimization of the power supply circuit of the permanent magnet recloser.

[0026] In one specific embodiment, the process of performing step S101 may specifically include the following steps: Based on the action response window of the permanent magnet recloser, the original electrical signals of each power supply node in the distribution network are adaptively segmented to obtain segmented electrical data containing pre-reclosing warning features and post-reclosing steady-state features. The segmented electrical data is input into the power supply node correlation analysis algorithm for topology correlation filtering, resulting in a set of core power supply node electrical parameters after removing redundant node information; The electrical parameter set of the core power supply node is marked with importance according to the power supply path weight allocation strategy of permanent magnet recloser, and hierarchical electrical feature data including the main power supply path and the backup power supply path is obtained. The hierarchical electrical feature data is grouped by operating condition based on the power supply condition similarity clustering method to obtain a power supply condition dataset for training deep learning models.

[0027] Specifically, the permanent magnet recloser action response window technology determines the boundary conditions for time-series segmentation by analyzing the inherent action characteristics of the permanent magnet recloser. This technology defines the response window based on the complete time sequence of the permanent magnet recloser from receiving the trip signal to completing the reclosing action. Specifically, it includes four sub-windows: fault detection stage, trip delay stage, reclosing preparation stage, and reclosing execution stage. The adaptive time-series segmentation processing identifies the pre-reclosing warning characteristics and the post-reclosing steady-state characteristics by monitoring the rate of change and amplitude fluctuation of electrical signals in each sub-window. The pre-reclosing warning characteristics are mainly manifested as a gradual decrease in voltage amplitude and an abnormal increase in current harmonic content. The steady-state characteristics are manifested as the voltage amplitude returning to the normal range and the current waveform tending to stabilize. The segmented electrical data is classified and stored according to time tags and feature types. The power supply node correlation analysis algorithm uses mutual information theory to calculate the electrical correlation between nodes. The algorithm first calculates the joint probability distribution of electrical parameters of any two power supply nodes, and then quantifies the strength of the dependency relationship between nodes through mutual information values. Node pairs with high mutual information values ​​indicate that their electrical state changes have a strong coupling relationship, while node pairs with mutual information values ​​close to zero indicate that they are independent of each other. The topology correlation screening process removes redundant nodes that are not directly related to the operation of the permanent magnet recloser according to the preset correlation threshold. The core power supply node electrical parameter set only includes node data that has a significant impact on the power supply performance of the permanent magnet recloser.

[0028] The permanent magnet recloser power supply path weight allocation strategy determines the importance level of each power supply path based on power flow analysis and fault impact assessment. This strategy calculates a comprehensive score of the load capacity, power supply distance, and fault propagation risk of each power supply path. The weight of the main power supply path is allocated to the power supply branch that carries critical loads and has a low fault risk, while the weight of the backup power supply path is allocated to the power supply branch with a relatively small load capacity but high switching flexibility. The importance labeling process hierarchically labels the electrical parameters of each power supply node according to the weight level of its power supply path. The hierarchical electrical feature data includes information in four dimensions: node identifier, electrical parameter value, path weight, and hierarchical label.

[0029] The power supply condition similarity clustering method uses the K-means clustering algorithm to group the hierarchical electrical feature data into operating condition groups. This method first calculates the Euclidean distance between each electrical feature vector as a similarity measure, and then iteratively optimizes the data points with similar electrical features to be classified into the same cluster center. The operating condition grouping process determines the number of clusters based on the electrical response mode of the permanent magnet recloser under different operating conditions. Each cluster represents a typical power supply condition, including similar load distribution patterns, voltage levels and harmonic features. The training dataset includes the central feature vector of each cluster, the distribution variance of the data within the cluster, and the separation index between clusters. Taking a manufacturing enterprise's power distribution network as an example, when the permanent magnet recloser detects a voltage drop on the output side of the main transformer, the action response window technology identifies the warning characteristics of a step-like decrease in voltage amplitude before the drop and the steady-state characteristics of rapid voltage recovery after reclosing. The correlation analysis algorithm finds that the voltage drop is strongly correlated with the load changes of the three downstream workshops. The weight allocation strategy classifies the workshop carrying precision machining equipment as the main power supply path and assigns it the highest weight, while classifying the auxiliary equipment workshop as the backup power supply path. The clustering method clusters the three operating conditions of normal operation, light load operation, and heavy load operation according to the similarity of electrical features, forming a training dataset containing typical electrical feature patterns.

[0030] In one specific embodiment, the process of performing step S102 may specifically include the following steps: The power supply status dataset is input into the power supply adjacency matrix construction layer of the dedicated topology sensing network for permanent magnet reclosers to perform node correlation calculation and obtain a dynamic weighted adjacency matrix based on electrical distance and fault propagation probability. Based on the dynamic weighted adjacency matrix, a small-range power supply topology convolution is performed on the local feature extraction layer of the dedicated topology sensing network for permanent magnet reclosers to obtain a local topology feature map containing power supply association features of adjacent nodes. The local topological feature map is input into the medium receptive field extension layer of the dedicated topological sensing network for permanent magnet reclosers and then processed by convolution of the power supply network at medium distance to obtain a medium topological feature map containing the power supply coordination relationship within the region. Based on the medium topological feature map, the global receptive field extension layer of the dedicated topological sensing network for permanent magnet reclosers is subjected to large-scale power supply network convolution processing to obtain a global topological feature map containing cross-regional power supply influence relationships. Based on the power supply path attention weight mechanism, multi-scale feature fusion processing is performed on local topology feature maps, medium-sized topology feature maps, and global topology feature maps to obtain a power supply path feature matrix that includes local power supply correlation and global network topology.

[0031] Specifically, the power supply adjacency matrix construction layer establishes the topological connection relationship between each power supply node of the permanent magnet recloser through node correlation degree calculation. This layer first calculates the electrical distance between each node based on the opening and closing state transition characteristics of the permanent magnet recloser. The electrical distance reflects the propagation attenuation degree of the impact of the permanent magnet recloser's operation on the electrical state of different nodes. The calculation process determines the distance weight by analyzing the change amplitude of voltage and current of each node during the permanent magnet recloser's opening to closing process. The smaller the distance, the stronger the electrical coupling between nodes. Then, the propagation path and propagation speed of the fault from the source node to the target node are analyzed based on the fault response time series of the permanent magnet recloser. The fault propagation probability is calculated by statistically analyzing the frequency and time delay of fault propagation between nodes in historical fault data. The dynamic weighted adjacency matrix weights and fuses the electrical distance and the fault propagation probability. Each element in the matrix represents the comprehensive correlation strength between the corresponding node pairs. After receiving the dynamically weighted adjacency matrix, the local feature extraction layer performs small-range power supply topology convolution processing. This layer uses a sliding window technique to perform convolution operations on the adjacency matrix. The convolution kernel size is set to cover the range of directly adjacent nodes of the permanent magnet reconcile. The convolution processing extracts the power supply association patterns between adjacent nodes by scanning the local regions of the adjacency matrix one by one. Each convolution operation converts the local adjacency relationship into a feature vector. The local topology feature map contains information on the power supply association strength and association type of each node and its directly adjacent nodes. Each position in the feature map corresponds to a power supply node, and its feature value reflects the connection density and influence of the node in the local power supply network.

[0032] The medium receptive field extension layer takes the local topology feature map as input for mid-range power supply network convolution processing. This layer covers a larger network range by increasing the convolution kernel size. Mid-range convolution processing can capture the regional power supply coordination relationship within the influence range of the permanent magnet recloser. When the convolution operation slides on the local feature map, it considers the feature combination of multiple adjacent nodes at the same time. The medium topology feature map records the coordination and cooperation pattern and load distribution relationship between nodes in each power supply area. The feature values ​​in the feature map represent the degree of coordination of power supply resources and the potential for configuration optimization in the area.

[0033] The global receptive field extension layer performs large-scale power supply network convolution processing based on a medium-sized topology feature map. This layer uses the largest-sized convolution kernel to cover the entire permanent magnet recloser power supply network. Global convolution processing identifies cross-regional power supply influence relationships and electrical coupling effects between remote nodes. The global topology feature map reflects the macroscopic topology structure and long-distance power transmission characteristics of the entire power supply network. The feature map contains power supply support capabilities and fault isolation boundary information between each region.

[0034] The power supply path attention weighting mechanism performs multi-scale feature fusion processing on the topology feature maps at three scales. The mechanism first calculates the importance weight of each feature position in each feature map. The weight calculation is based on the load carrying capacity and switching frequency of the permanent magnet recloser on different power supply paths. The attention mechanism fuses the local, intermediate and global feature maps into a unified feature representation through weighted averaging. During the fusion process, the local feature map is responsible for providing fine node connection information, the intermediate feature map is responsible for providing regional coordination information, and the global feature map is responsible for providing overall topology information. The power supply path feature matrix finally contains the comprehensive feature vectors of each power supply node at different scales and the multi-level association relationships between nodes. Taking the power distribution network of an industrial park as an example, when the permanent magnet recloser connects the main transformer and the power distribution rooms of three workshops, the adjacency matrix construction layer calculates that the electrical distance between the main transformer and workshop A is the smallest and the electrical distance between the main transformer and workshop C is the largest. The local feature extraction layer identifies that there is a direct power supply relationship between workshop A and workshop B. The medium receptive field extension layer finds that there is a coordinated relationship of mutual support among the three workshops during peak load periods. The global receptive field extension layer identifies the power supply dependence relationship between the entire park and the external power grid. The attention weight mechanism assigns different fusion weights according to the power consumption importance of each workshop, and finally forms a power supply path feature matrix containing multi-level topology information.

[0035] In one specific embodiment, the process of inputting the power supply status dataset into the power supply adjacency matrix construction layer of the dedicated topology sensing network for permanent magnet reclosers for node correlation calculation can specifically include the following steps: Based on the opening and closing state transition characteristics of permanent magnet reclosers, electrical distance measurement is calculated for each power supply node in the power supply state dataset to obtain the inter-node electrical distance matrix reflecting the influence range of permanent magnet reclosers. Based on the fault response time series of the permanent magnet recloser, the electrical distance matrix is ​​analyzed to obtain a fault propagation probability matrix containing the probability of the fault spreading from the source node to the target node. The electrical distance matrix and the fault propagation probability matrix are input into a weighted fusion algorithm for weight coefficient allocation, resulting in a node importance weight vector that reflects the power supply priority of the permanent magnet recloser. The node importance weight vector is dynamically adjusted based on the power supply switching logic of the permanent magnet recloser to obtain adaptive weight coefficients that change with the operating state of the permanent magnet recloser. The power supply node adjacency relationship is weighted by adaptive weighting coefficients to obtain a dynamic weighted adjacency matrix based on electrical distance and fault propagation probability.

[0036] Specifically, the permanent magnet recloser opening and closing state transition feature calculates the electrical distance by recording the change pattern of electrical parameters of each power supply node during the transition from the open state to the closed state of the permanent magnet recloser. This feature includes the voltage drop amplitude, current interruption time, voltage recovery speed, and current establishment process of each node at the moment of opening of the permanent magnet recloser. The electrical distance measurement calculation process determines the distance value between nodes by comparing the similarity of the impact of the permanent magnet recloser state transition on the electrical parameters of different nodes. The smaller the distance value, the more similar the electrical response of the two nodes is when the permanent magnet recloser operates. The electrical distance matrix between nodes records the electrical coupling strength between any two nodes. A diagonal element of zero in the matrix indicates that the distance between the node and itself is zero, and a larger value of the off-diagonal element indicates that the electrical independence of the corresponding node pair is stronger. The permanent magnet recloser fault response time series includes the time delay from the permanent magnet recloser detecting a fault signal to executing a tripping action, the time interval from tripping to reclosing action, and the time nodes of successful or failed reclosing. The fault propagation path analysis and processing is based on the statistical analysis of the propagation direction and propagation delay of the fault between different nodes based on historical fault records. The analysis process determines the propagation path by identifying the time sequence of the fault's initial node and the subsequently affected nodes. The propagation probability is obtained by calculating the ratio of the frequency of fault propagation between specific node pairs to the total number of faults. The fault propagation probability matrix records the probability value of any source node propagating the fault to the target node. The higher the value of the matrix element, the greater the possibility of fault propagation between the corresponding node pairs.

[0037] The weighted fusion algorithm assigns weight coefficients to the electrical distance matrix and the fault propagation probability matrix. The algorithm first normalizes the two matrices to eliminate dimensional differences, and then determines the fusion weights based on the power supply priority of the permanent magnet recloser. Nodes with higher power supply priority correspond to larger electrical distance weights and smaller fault propagation weights. The weight coefficient allocation process uses a linear combination to sum the normalized electrical distance and fault propagation probability. The node importance weight vector contains the comprehensive importance score of each power supply node in the permanent magnet recloser power supply network. Nodes with higher scores have higher protection priority in power supply circuit optimization.

[0038] The power supply switching logic of the permanent magnet recloser is based on a dynamic adjustment strategy that determines the node weights according to load type, power supply path redundancy, and switching cost. This logic includes weight allocation rules in normal operation mode and weight redistribution rules in fault emergency mode. The dynamic adjustment process selects the corresponding adjustment strategy according to the current operating status of the permanent magnet recloser. During normal operation, the weights of important load nodes are higher and the weights of auxiliary load nodes are lower. During fault emergency, the weights of critical power supply path nodes increase significantly and the weights of nodes in the fault-affected area decrease accordingly. The adaptive weight coefficient is updated in real time with the operating status of the permanent magnet recloser, reflecting the changes in the relative importance of each node under different operating conditions.

[0039] The weighted assignment process assigns numerical values ​​to the adjacency relationships of power supply nodes based on adaptive weight coefficients. This process uses the weight coefficients as correction factors for the elements of the adjacency matrix. The strength of the adjacency relationship is equal to the original correlation degree multiplied by the weight coefficient of the corresponding node pair. Each element in the dynamic weighted adjacency matrix reflects both the physical connection relationship between nodes and the influence of the permanent magnet recloser's operating state on the importance of the connection. The matrix element values ​​are dynamically updated according to the permanent magnet recloser's operating mode. Taking a chemical plant's power distribution network as an example, when the permanent magnet recloser connects the reactor workshop, packaging workshop, and office building, the opening and closing state transition characteristics show that the voltage drop in the reactor workshop is the most severe when the permanent magnet recloser is opened, while the voltage drop in the office building is the slowest. Based on this, it is calculated that the electrical distance between the reactor workshop and the packaging workshop is less than the electrical distance between the reactor workshop and the office building. Fault response time series analysis reveals that the probability of a fault in the reactor workshop propagating to the packaging workshop is higher than the probability of it propagating to the office building. The weighted fusion algorithm assigns the highest weight coefficient based on the production importance of the reactor workshop. When the power supply switching logic detects a sudden increase in the load of the reactor workshop, it dynamically increases the weight coefficient of the reactor workshop and correspondingly decreases the weight coefficient of the office building. The resulting dynamic weighted adjacency matrix prominently reflects the strong correlation between the reactor workshop and other nodes.

[0040] In one specific embodiment, the process of executing step S103 may specifically include the following steps: Based on the power supply reliability constraints of the permanent magnet recloser, the power supply path feature matrix is ​​processed to screen the feasibility of the power supply path, and a set of candidate power supply paths that meet the minimum number of operations required by the permanent magnet recloser is obtained. The candidate power supply path set is input into a multi-objective optimization algorithm for power quality constraint analysis, resulting in voltage deviation control constraints and harmonic content control constraints. Based on the load carrying capacity limit of the permanent magnet recloser, the load allocation strategy is calculated and processed according to the voltage deviation control constraint and harmonic content control constraint to obtain the optimal power allocation ratio of each power supply branch. Based on the power supply path switching time constraint of the permanent magnet recloser, the optimal power allocation ratio is optimized by power supply timing coordination to obtain the switching sequence and switching time interval of the main and backup power supply paths. The circuit topology reconstruction scheme is obtained by integrating the switching sequence of the main and backup power supply paths with the optimal power allocation ratio through topology reconstruction.

[0041] Specifically, the reliability constraints of the permanent magnet recloser power supply are determined based on the mechanical life and electrical performance limitations of the permanent magnet recloser to determine the feasibility boundary of the power supply path. These constraints include parameters such as the maximum allowable number of operations of the permanent magnet recloser, the shortest operation interval time, and the continuous operation capability. The power supply path feasibility screening process checks whether the number of operations required by the permanent magnet recloser for each candidate power supply path within a preset time window exceeds the minimum number of operations required. The screening process compares the historical operation frequency of each path in the power supply path feature matrix with the reliability constraint threshold. The candidate power supply path set only includes power supply paths that meet the requirements for long-term stable operation of the permanent magnet recloser.

[0042] After receiving the set of candidate power supply paths, the multi-objective optimization algorithm performs power quality constraint analysis. The algorithm uses the Pareto front search method to simultaneously optimize multiple power quality indicators. The voltage deviation control constraint determines the voltage quality control boundary by calculating the deviation between the voltage of each power supply node and the rated voltage. The harmonic content control constraint determines the harmonic pollution control boundary by analyzing the harmonic distortion rate of each power supply path. The power quality constraint analysis uses the voltage deviation and harmonic content as constraints of the optimization objective function to form a feasible solution region in a multi-dimensional constraint space.

[0043] The load carrying capacity limit of the permanent magnet recloser is determined based on the rated capacity and overload capacity of the permanent magnet recloser to determine the upper limit of power distribution for each power supply branch. The load distribution strategy calculation process takes voltage deviation control constraints and harmonic content control constraints as input constraints, and solves the optimal power distribution ratio for each power supply branch through a linear programming algorithm. During the calculation process, the total capacity constraint of the permanent magnet recloser, the power balance constraint of each branch, and the power quality constraint must be met. The optimal power distribution ratio ensures that the power distribution of each power supply branch meets the load demand without exceeding the carrying capacity limit of the permanent magnet recloser.

[0044] The power supply path switching time constraint of the permanent magnet recloser is determined based on the operating characteristics of the permanent magnet recloser and the grid stability requirements. The timing limit for path switching is determined. This constraint includes parameters such as the shortest switching interval time, the maximum switching duration, and switching sequence constraints. The power supply timing coordination optimization process determines the power supply branches that need to be switched according to the optimal power allocation ratio. Then, the switching sequence of the primary and backup power supply paths is calculated based on the switching time constraint. The switching sequence arrangement ensures that critical loads are given priority in power supply and secondary loads are appropriately delayed in switching. The switching time interval is determined according to the operation recovery time of the permanent magnet recloser and the transient stability requirements of the grid.

[0045] The topology reconfiguration integration process integrates the switching sequence of the primary and backup power supply paths with the optimal power allocation ratio. This process first determines the activation time window of each power supply path based on the switching sequence, and then maps the power allocation ratio to the corresponding time window to form a dynamic power allocation scheme. During the integration process, the capacity constraints of the permanent magnet recloser at different switching stages and the power coordination relationship between each power supply branch need to be considered. The circuit topology reconfiguration scheme includes the connection status of each power supply path, power allocation parameters, and switching timing arrangement. Taking a data center power distribution network as an example, when a permanent magnet recloser connects a server room, a cooling room, and an office area, the power supply reliability constraint requires that the number of times the permanent magnet recloser operates per hour does not exceed a specified limit. After screening, three candidate power supply paths that meet the operation frequency requirement are obtained. Multi-objective optimization algorithm analysis finds that the server room is most sensitive to voltage deviation, while the cooling room is most sensitive to harmonic content. The load distribution strategy calculation process determines the maximum power allocation ratio for the server room based on the power quality sensitivity of each area and the capacity limit of the permanent magnet recloser. The power supply timing coordination optimization process arranges the server room as the main power supply path and gives it the highest switching priority. The cooling room and office area are used as backup power supply paths and switched in order of importance. The topology reconstruction integration process forms a circuit topology reconstruction scheme that includes dynamic power allocation and hierarchical switching timing.

[0046] In one specific embodiment, the process of executing step S104 may specifically include the following steps: Based on the state space definition of permanent magnet recloser power supply, the circuit topology reconstruction scheme is processed by state vector encoding to obtain a state vector reflecting the current power supply network topology and load distribution. Based on the spatial constraints of the power supply action of the permanent magnet recloser, the executable action selection process is performed on the state vector to obtain the discrete action set of power supply path switching action and voltage regulation action. The state vector and discrete action set are input into a deep Q network for state-action value evaluation to obtain the long-term revenue prediction value corresponding to each power supply action. Based on the power supply reward mechanism of permanent magnet recloser, the long-term revenue prediction value is optimized by strategy gradient processing to obtain the optimal power supply path switching strategy and voltage regulation parameters. The power supply configuration parameters are integrated and processed based on the optimal power supply path switching strategy and voltage regulation parameters to obtain the optimized power supply circuit configuration.

[0047] Specifically, the power supply state space definition of the permanent magnet recloser includes dimensional information such as the power supply network topology, the load status of each node, the opening and closing status of the permanent magnet recloser, and electrical parameters. The state vector encoding process converts the discrete topology information in the circuit topology reconstruction scheme into a continuous numerical vector. The encoding process first expands the adjacency matrix of the power supply network into a one-dimensional vector to represent the topology connection relationship. Then, the load power, voltage amplitude, current magnitude, and other parameters of each node are arranged in the order of the nodes to form a load distribution vector. Next, the opening and closing status, number of operations, and remaining capacity of the permanent magnet recloser are encoded into an operation state vector. Finally, the topology vector, load vector, and operation vector are connected to form a state vector that reflects the current power supply network topology and load distribution. The spatial constraint of permanent magnet recloser power supply actions determines the permissible range of operations based on the physical limitations of the permanent magnet recloser and the power grid operation rules. This constraint includes restrictions such as the number of power supply paths that the permanent magnet recloser can control, the range of voltage regulation, and the time interval between action executions. The executable action screening process determines which power supply actions are physically feasible based on the capacity margin of the permanent magnet recloser and the load demand in the current state vector. The screening process checks whether each candidate action violates the capacity constraint, voltage constraint, and timing constraint of the permanent magnet recloser. The discrete action set contains the power supply path switching actions and voltage regulation actions that have passed the screening. Each action is represented by a discrete code to indicate its type, parameters, and execution conditions.

[0048] The deep Q-network receives the state vector and discrete action set and performs state-action value evaluation. This network uses a multi-layer neural network structure to learn the long-term value function of state-action pairs. The value evaluation process calculates the Q-value of each discrete action through forward propagation, taking the state vector as input and outputting the Q-value. The Q-value represents the expected cumulative reward for performing a specific action in the current state. During network training, the optimal state-action value mapping relationship is learned through experience replay and a target network update mechanism. The long-term reward prediction reflects the expected contribution of each power supply action to the long-term operating performance of the permanent magnet recloser. The permanent magnet recloser power supply reward mechanism designs a reward function based on indicators such as power supply reliability, power quality, and operating efficiency. This mechanism provides positive rewards for actions that improve power supply reliability, negative rewards for actions that reduce power quality, and penalties for actions that exceed the capabilities of the permanent magnet recloser. The policy gradient optimization process adjusts the policy parameters of the deep Q-network based on the reward mechanism. The optimization process updates the network weights by calculating the policy gradient to maximize the expected reward. The optimal power supply path switching strategy determines the best power supply path to be selected in different states, and the voltage regulation parameters determine the optimal voltage regulation amplitude under various operating conditions.

[0049] The power supply configuration parameter integration process combines the optimal power supply path switching strategy and voltage regulation parameters to form a power supply control scheme. The integration process converts the strategy parameters into specific instructions that the permanent magnet recloser controller can execute, including configuration information such as activation conditions, switching sequence, voltage setpoints and protection parameters for each power supply path. The power supply circuit optimization configuration includes the optimal control parameters and action strategies of the permanent magnet recloser under various operating conditions. The configuration parameters are dynamically adjusted according to the actual operating status to adapt to load changes and fault conditions.

[0050] In one specific embodiment, the process of executing step S105 may specifically include the following steps: Based on the monitoring of the permanent magnet recloser's operating status, real-time parameter deviation detection and processing are performed on the optimized configuration of the power supply circuit to obtain the deviation between the actual value and the target value of the power supply parameters. The deviation is input into the fault symptom identification algorithm for abnormal mode analysis to obtain the potential fault types and fault occurrence probabilities of the permanent magnet recloser. Based on the potential fault type, a preventive adjustment strategy calculation is performed on the power supply switching decision of the permanent magnet recloser to obtain a preventive power supply path switching scheme before the fault occurs. Based on the closed-loop control feedback mechanism of permanent magnet recloser, the preventive power supply path switching scheme is processed in real time to generate execution instructions, and the fault prediction results and power supply switching instructions are obtained. The fault prediction results and power supply switching commands are encapsulated and processed into control signals to obtain intelligent power supply control signals.

[0051] Specifically, the permanent magnet recloser operation status monitoring performs real-time parameter deviation detection and processing by collecting the electrical parameters, mechanical status, and control signals of the permanent magnet recloser in real time. This monitoring includes key parameters such as the coil current, contact resistance, operating time, and insulation status of the permanent magnet recloser. The deviation detection and processing compares the actual values ​​of the monitored power supply parameters with the target values ​​set in the optimized configuration of the power supply circuit item by item, and calculates the absolute and relative deviations of each parameter. The deviation includes values ​​in multiple dimensions such as voltage deviation, current deviation, power deviation, and frequency deviation. The magnitude and trend of the deviation reflect the degree of difference between the current operating status of the permanent magnet recloser and the ideal state.

[0052] After receiving deviation data, the fault symptom identification algorithm performs abnormal pattern analysis. The algorithm uses pattern matching technology to compare the characteristic patterns of the current deviation with a pre-stored fault symptom pattern library. The abnormal pattern analysis identifies the abnormal pattern type by calculating the statistical characteristics of the time-series changes of the deviation, including different patterns such as gradual anomalies, sudden anomalies, and periodic anomalies. The algorithm determines the potential fault type based on the amplitude distribution, rate of change, and duration of the deviation. The potential fault types include typical permanent magnet recloser faults such as poor contact, insulation aging, mechanical wear, and control failure. The probability of fault occurrence is calculated using Bayesian inference based on the current symptom intensity and historical fault statistics.

[0053] The power supply switching decision for permanent magnet reclosers is based on the calculation of preventive adjustment strategies according to the potential fault type and fault probability. This decision mechanism determines the urgency of preventive measures based on the threat level of different fault types to power supply reliability. The preventive adjustment strategy calculation includes various preventive measures such as load transfer strategy, backup path activation strategy, and protection setting adjustment strategy. The calculation process considers factors such as the remaining reliability of permanent magnet reclosers, load importance level, and backup capacity constraints. The pre-fault preventive power supply path switching scheme determines the power supply path adjustment operations to be performed before the fault occurs, including specific details such as switching timing, switching sequence, and switching parameters.

[0054] The closed-loop control feedback mechanism of the permanent magnet recloser generates and processes execution instructions in real time for the preventive power supply path switching scheme. This mechanism verifies the feasibility of the preventive scheme based on the current power grid operating status and load demand. The feedback mechanism checks whether the execution of the scheme will cause other equipment to overload or voltage exceed the limit. The real-time execution instruction generation process converts the feasibility-verified preventive scheme into specific instructions that the permanent magnet recloser controller can recognize. The fault prediction results include information such as fault type, predicted occurrence time, and impact range assessment. The power supply switching instructions include operation instructions such as target power supply path, switching time, and switching parameters.

[0055] The control signal encapsulation process encapsulates fault prediction results and power supply switching commands according to communication protocol requirements. This encapsulation includes data processing steps such as command priority setting, timestamp addition, and checksum calculation. The intelligent power supply control signal uses a standardized data format to ensure the permanent magnet recloser controller can correctly parse and execute control commands. The control signal consists of three basic components: a command header, a data payload, and a checksum tail. Taking a hospital power distribution network as an example, when the permanent magnet recloser supplies power to the operating room, ICU, and general wards, operational status monitoring detects a slow downward trend in the operating room's power supply voltage. Deviation detection processing calculates that the actual voltage value is lower than the target value by a preset threshold. The fault symptom identification algorithm identifies this gradual voltage drop pattern as a sign of increased contact resistance of the permanent magnet recloser contacts and calculates that the probability of this fault occurring within the next two hours is high. The preventative adjustment strategy calculation determines to pre-transfer the operating room load to the backup power supply path. The closed-loop control feedback mechanism verifies that the backup path has sufficient capacity and will not affect the power supply to other loads. Finally, an intelligent power supply control signal containing fault warning information and load transfer commands is generated.

[0056] The above describes the deep learning-based permanent magnet recloser power supply circuit design optimization method in the embodiments of this application. The following describes the deep learning-based permanent magnet recloser power supply circuit design optimization system in the embodiments of this application. Please refer to [link / reference]. Figure 2 One embodiment of the deep learning-based permanent magnet recloser power supply circuit design optimization system in this application includes: The data acquisition module is used to collect and process multi-dimensional electrical parameters of each power supply node of the permanent magnet recloser through intelligent sensors in the distribution network, and obtain a power supply status dataset including node voltage amplitude, branch current, and power factor. The extraction module is used to input the power supply status dataset into a dedicated topology sensing network for permanent magnet reclosers to perform power supply path feature extraction processing, and obtain a power supply path feature matrix that includes local power supply correlation and global network topology. The reconstructing module is used to perform multi-objective constraint reconstruction processing on the power supply circuit topology of the permanent magnet recloser based on the power supply path feature matrix, so as to obtain a circuit topology reconstruction scheme that includes the optimal power supply path combination and load distribution strategy. The optimization module is used to input the circuit topology reconstruction scheme into a deep reinforcement learning algorithm for power supply parameter optimization processing, so as to obtain an optimized power supply circuit configuration that includes the optimal power supply path switching strategy and voltage regulation parameters. The adjustment module is used to perform closed-loop adjustment processing on the real-time power supply parameters of the permanent magnet recloser according to the optimized configuration of the power supply circuit, so as to obtain an intelligent power supply control signal containing fault prediction results and power supply switching instructions.

[0057] above Figure 2The deep learning-based permanent magnet recloser power supply circuit design optimization system in this embodiment of the invention is described in detail from the perspective of modular functional entities. The deep learning-based permanent magnet recloser power supply circuit design optimization device in this embodiment of the invention is described in detail from the perspective of hardware processing.

[0058] Reference Figure 3 This invention also provides a deep learning-based permanent magnet recloser power supply circuit design optimization device. This deep learning-based permanent magnet recloser power supply circuit design optimization device can be a server, and its internal structure can be as follows: Figure 3 As shown. This deep learning-based permanent magnet recloser power supply circuit design optimization device includes a processor, memory, display screen, input device, network interface, and database connected via a system bus. The processor in this computer design provides computing and control capabilities. The memory of this deep learning-based permanent magnet recloser power supply circuit design optimization device includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database of this deep learning-based permanent magnet recloser power supply circuit design optimization device stores the data corresponding to this embodiment. The network interface of this deep learning-based permanent magnet recloser power supply circuit design optimization device is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements the above-described method.

[0059] Those skilled in the art will understand that Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present invention and does not constitute a limitation on the deep learning-based permanent magnet recloser power supply circuit design optimization device applied thereto.

[0060] The present invention also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein the computer-readable storage medium stores instructions that, when the instructions are executed on a computer, cause the computer to perform the steps of the deep learning-based permanent magnet recloser power supply circuit design optimization method.

[0061] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0062] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a deep learning-based permanent magnet recloser power supply circuit design optimization device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0063] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A deep learning-based method for optimizing the design of a permanent magnet recloser power supply circuit, characterized in that, The method includes: By using smart sensors in the distribution network to collect and process multi-dimensional electrical parameters of each power supply node of the permanent magnet recloser, a power supply status dataset including node voltage amplitude, branch current, and power factor is obtained. The power supply status dataset is input into a dedicated topology sensing network for permanent magnet reclosers to perform power supply path feature extraction processing, resulting in a power supply path feature matrix that includes local power supply correlation and global network topology. Based on the power supply path feature matrix, the power supply circuit topology of the permanent magnet recloser is reconstructed using a multi-objective constraint to obtain a circuit topology reconstruction scheme that includes the optimal power supply path combination and load distribution strategy. The circuit topology reconstruction scheme is input into a deep reinforcement learning algorithm for power supply parameter optimization, resulting in an optimized power supply circuit configuration that includes the optimal power supply path switching strategy and voltage regulation parameters. Based on the optimized configuration of the power supply circuit, the real-time power supply parameters of the permanent magnet recloser are adjusted in a closed loop to obtain an intelligent power supply control signal that includes fault prediction results and power supply switching instructions.

2. The deep learning-based permanent magnet recloser power supply circuit design optimization method according to claim 1, characterized in that, The process involves collecting and processing multi-dimensional electrical parameters of each power supply node of the permanent magnet recloser using intelligent sensors in the power distribution network. This yields a power supply status dataset including node voltage amplitude, branch current, and power factor. Based on the action response window of the permanent magnet recloser, the original electrical signals of each power supply node in the distribution network are adaptively segmented to obtain segmented electrical data containing pre-reclosing warning features and post-reclosing steady-state features. The segmented electrical data is input into a power supply node correlation analysis algorithm for topology correlation filtering to obtain a set of core power supply node electrical parameters after removing redundant node information. The electrical parameter set of the core power supply node is marked with importance according to the power supply path weight allocation strategy of the permanent magnet recloser, so as to obtain hierarchical electrical feature data including the main power supply path and the backup power supply path. The hierarchical electrical feature data is grouped by operating condition based on the power supply status similarity clustering method to obtain a power supply status dataset for training deep learning models.

3. The deep learning-based permanent magnet recloser power supply circuit design optimization method according to claim 1, characterized in that, The step of inputting the power supply status dataset into a dedicated topology sensing network for permanent magnet reclosers for power supply path feature extraction processing yields a power supply path feature matrix that includes local power supply correlation and global network topology, including: The power supply status dataset is input into the power supply adjacency matrix construction layer of the dedicated topology sensing network for permanent magnet reclosers to perform node correlation calculation and obtain a dynamic weighted adjacency matrix based on electrical distance and fault propagation probability. Based on the dynamic weighted adjacency matrix, a small-range power supply topology convolution is performed on the local feature extraction layer of the dedicated topology sensing network for permanent magnet reclosers to obtain a local topology feature map containing power supply association features of adjacent nodes. The local topological feature map is input into the medium receptive field extension layer of the dedicated topological sensing network for permanent magnet reconciliation and then processed by convolution of the power supply network at a medium distance to obtain a medium topological feature map containing the power supply coordination relationship within the region. Based on the aforementioned medium topological feature map, a large-scale power supply network convolution process is performed on the global receptive field extension layer of the dedicated topological sensing network for permanent magnet reclosers to obtain a global topological feature map containing cross-regional power supply influence relationships. Based on the power supply path attention weight mechanism, multi-scale feature fusion processing is performed on the local topology feature map, the intermediate topology feature map, and the global topology feature map to obtain a power supply path feature matrix that includes local power supply correlation and global network topology.

4. The deep learning-based permanent magnet recloser power supply circuit design optimization method according to claim 3, characterized in that, The step of inputting the power supply status dataset into the power supply adjacency matrix construction layer of the dedicated topology sensing network for permanent magnet reclosers for node correlation calculation processing yields a dynamically weighted adjacency matrix based on electrical distance and fault propagation probability, including: Based on the opening and closing state transition characteristics of the permanent magnet recloser, the electrical distance measurement of each power supply node in the power supply state dataset is calculated to obtain the inter-node electrical distance matrix reflecting the influence range of the permanent magnet recloser. Based on the fault response time series of the permanent magnet recloser, the electrical distance matrix is ​​subjected to fault propagation path analysis to obtain a fault propagation probability matrix containing the probability of fault propagation from the source node to the target node; The electrical distance matrix and the fault propagation probability matrix are input into a weighted fusion algorithm for weight coefficient allocation, resulting in a node importance weight vector that reflects the power supply priority of the permanent magnet recloser. The importance weight vector of the node is dynamically adjusted based on the power supply switching logic of the permanent magnet recloser to obtain an adaptive weight coefficient that changes with the operating state of the permanent magnet recloser. The power supply node adjacency relationship is weighted according to the adaptive weight coefficient to obtain a dynamic weighted adjacency matrix based on electrical distance and fault propagation probability.

5. The deep learning-based permanent magnet recloser power supply circuit design optimization method according to claim 1, characterized in that, The process of performing multi-objective constraint reconstruction of the permanent magnet recloser power supply circuit topology based on the power supply path feature matrix to obtain a circuit topology reconstruction scheme that includes the optimal power supply path combination and load distribution strategy includes: Based on the power supply reliability constraints of the permanent magnet recloser, the power supply path feature matrix is ​​subjected to power supply path feasibility screening to obtain a set of candidate power supply paths that meet the minimum number of operations required by the permanent magnet recloser. The candidate power supply path set is input into a multi-objective optimization algorithm for power quality constraint analysis to obtain voltage deviation control constraints and harmonic content control constraints. Based on the load carrying capacity limit of the permanent magnet recloser, the load allocation strategy is calculated and processed according to the voltage deviation control constraint and harmonic content control constraint to obtain the optimal power allocation ratio of each power supply branch. Based on the power supply path switching time constraint of the permanent magnet recloser, the optimal power allocation ratio is optimized by power supply timing coordination to obtain the main and backup power supply path switching sequence and switching time interval. The switching sequence of the primary and backup power supply paths is combined with the optimal power allocation ratio to perform topology reconstruction and integration processing, resulting in a circuit topology reconstruction scheme.

6. The deep learning-based permanent magnet recloser power supply circuit design optimization method according to claim 1, characterized in that, The step of inputting the circuit topology reconstruction scheme into a deep reinforcement learning algorithm for power supply parameter optimization processing, to obtain an optimized power supply circuit configuration including the optimal power supply path switching strategy and voltage regulation parameters, includes: Based on the power supply state space definition of permanent magnet recloser, the circuit topology reconstruction scheme is processed by state vector encoding to obtain a state vector reflecting the current power supply network topology and load distribution. Based on the spatial constraints of the permanent magnet recloser power supply action, the state vector is subjected to executable action filtering processing to obtain a discrete action set of power supply path switching action and voltage regulation action. The state vector and discrete action set are input into a deep Q network for state-action value evaluation to obtain the long-term revenue prediction value corresponding to each power supply action. Based on the power supply reward mechanism of permanent magnet recloser, the long-term revenue prediction value is optimized by strategy gradient to obtain the optimal power supply path switching strategy and voltage regulation parameters. Based on the optimal power supply path switching strategy and voltage regulation parameters, the power supply configuration parameters are integrated to obtain an optimized power supply circuit configuration.

7. The deep learning-based permanent magnet recloser power supply circuit design optimization method according to claim 1, characterized in that, The process of performing closed-loop adjustment of the real-time power supply parameters of the permanent magnet recloser based on the optimized configuration of the power supply circuit to obtain an intelligent power supply control signal containing fault prediction results and power supply switching commands includes: Based on the permanent magnet recloser operation status monitoring, the power supply circuit optimization configuration is processed in real time to detect parameter deviations and obtain the deviation between the actual value and the target value of the power supply parameters. The deviation is input into the fault symptom identification algorithm for abnormal pattern analysis to obtain the potential fault types and fault occurrence probabilities of the permanent magnet recloser. Based on the potential fault types, a preventive adjustment strategy calculation is performed on the power supply switching decision of the permanent magnet recloser to obtain a preventive power supply path switching scheme before the fault occurs. The preventive power supply path switching scheme is processed in real time to generate execution instructions based on the closed-loop control feedback mechanism of permanent magnet recloser, so as to obtain fault prediction results and power supply switching instructions. The fault prediction results and power supply switching instructions are encapsulated and processed into control signals to obtain intelligent power supply control signals.

8. A deep learning-based system for optimizing the design of a permanent magnet recloser power supply circuit, characterized in that, For implementing the deep learning-based permanent magnet recloser power supply circuit design optimization method as described in any one of claims 1-7, the deep learning-based permanent magnet recloser power supply circuit design optimization system comprises: The data acquisition module is used to collect and process multi-dimensional electrical parameters of each power supply node of the permanent magnet recloser through intelligent sensors in the distribution network, and obtain a power supply status dataset including node voltage amplitude, branch current, and power factor. The extraction module is used to input the power supply status dataset into a dedicated topology sensing network for permanent magnet reclosers to perform power supply path feature extraction processing, and obtain a power supply path feature matrix that includes local power supply correlation and global network topology. The reconstructing module is used to perform multi-objective constraint reconstruction processing on the power supply circuit topology of the permanent magnet recloser based on the power supply path feature matrix, so as to obtain a circuit topology reconstruction scheme that includes the optimal power supply path combination and load distribution strategy. The optimization module is used to input the circuit topology reconstruction scheme into a deep reinforcement learning algorithm for power supply parameter optimization processing, so as to obtain an optimized power supply circuit configuration that includes the optimal power supply path switching strategy and voltage regulation parameters. The adjustment module is used to perform closed-loop adjustment processing on the real-time power supply parameters of the permanent magnet recloser according to the optimized configuration of the power supply circuit, so as to obtain an intelligent power supply control signal containing fault prediction results and power supply switching instructions.

9. A device for designing and optimizing the power supply circuit of a permanent magnet recloser based on deep learning, characterized in that, The device includes a memory and a processor, wherein the memory stores a computer program that can run on the processor, and the processor executes the computer program to implement the deep learning-based permanent magnet recloser power supply circuit design optimization method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is run by the processor, it causes the processor to execute the deep learning-based permanent magnet recloser power supply circuit design optimization method as described in any one of claims 1 to 7.

Images