Multi-scene whole station fault self-recovery method and system

Abstract

The invention provides a multi-scene whole station fault self-recovery method and system, and belongs to the technical field of power systems. The method comprises the following steps: calculating the attention weight of a scene to determine a scene recognition result, calculating a multi-step propagation accumulation fault to determine a fault equipment set, and calculating a shortest path from a switching power supply point to a voltage loss target point through an improved Yen algorithm to generate a candidate scheme set; constructing a multi-objective function of each path in the candidate scheme set, solving to obtain a non-dominated optimal solution set, performing comprehensive scoring on the non-dominated optimal solution set by using an entropy weight method, and generating a sorting scheme sequence; and calculating the similarity between the graph structure features of the sorting scheme sequence and the graph structure features of the historical schemes to determine the multi-criterion confidence score of the schemes, and selecting the scheme with the confidence score higher than a set threshold value as a final fault self-healing scheme. The load recovery speed of the whole station of the main network fault distribution network is improved, the cooperative processing efficiency of the main network fault and the distribution network fault is improved, and the risk handling capacity is improved.

Description

Technical Field

[0001] This invention belongs to the field of power system technology, and more specifically, relates to a method and system for self-healing faults in a multi-scenario substation. Background Technology

[0002] Unlike current fault isolation and recovery methods for small-disturbance scenarios involving feeder faults in distribution networks, traditional methods are not applicable when encountering large-scale, multi-area power outages during power grid maintenance or extreme weather conditions. Relying solely on manual fault detection, location, and emergency response planning is time-consuming, lacks real-time performance, and is inaccurate due to manual upper-lower level coordination for power restoration. This results in large-scale, prolonged power outages in the distribution network, making it difficult to ensure rapid restoration of de-energized lines, reducing customer satisfaction, weakening risk resistance, and impacting the safe and stable operation of the distribution network.

[0003] In recent years, with the continuous increase in power grid load and the increasing richness of grid structure, the use of diversified load resources and the construction of new power systems have been put on the agenda. Multi-scenario fault self-healing technology has emerged to meet the needs of multiple application scenarios in distribution networks and to quickly and proactively respond to power outages and restore power. This technology adopts integrated operation monitoring of main and distribution systems, achieving mutual integration and rapid collaboration. The interconnection network is automatically generated, and equipment faults are graded for self-healing. The power grid can autonomously detect problems and quickly provide load transfer solutions, achieving graded fault self-healing and precise isolation. The entire "full shutdown and full transfer" process is an intelligent "power outage self-healing" process. During this time, only the dispatcher needs to confirm the automatically generated transfer plan, adjust the transfer method with one click, and flexibly transfer power without worrying about power outages. One-click execution effectively shortens the power outage time from hours to minutes, achieving rapid power restoration. During this period, residents hardly feel the power outage, comprehensively improving the reliability of power supply in the distribution network and enhancing residents' sense of power accessibility.

[0004] Existing technical document 1 (CN117096850A) discloses an automatic detection and rapid self-healing method and system for medium-voltage busbar undervoltage faults. Its shortcoming is that it only handles medium-voltage busbar undervoltage faults in real time and lacks a contingency plan and self-healing mechanism for faults in the entire substation.

[0005] Existing technical document 2 (CN117096850A) discloses a method and system for self-healing of medium-voltage busbar faults and batch power transfer of substation faults. Its shortcomings are that it only handles medium-voltage busbar undervoltage faults in real time and batch power transfer of substation faults, and lacks classification, processing and self-healing mechanisms for whole-station faults under different scenarios. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides a multi-scenario whole-station fault self-healing method and system, which can automatically detect, handle emergency situations, and quickly restore power for power outages in different application scenarios and large areas, greatly shortening power outage time and helping to ensure more stable and safe operation of the distribution network.

[0007] The present invention adopts the following technical solution.

[0008] The first aspect of the present invention provides a multi-scenario whole-site fault self-healing method, comprising: The attention weight of the scenario is calculated by extracting the temporal electrical features and spatial topology features of the distribution network to determine the scenario identification result. Based on the scenario identification result, the multi-step propagation cumulative fault is calculated to determine the set of faulty equipment. Based on the set of faulty equipment, the transfer power supply point and the undervoltage target point are determined. The shortest path from the transfer power supply point to the undervoltage target point is calculated by improving the Yen algorithm. Based on the shortest path, a set of candidate solutions is generated. Construct a multi-objective function for each path in the candidate solution set, perform multi-objective optimization search on the multi-objective function using the Pareto algorithm to obtain a non-dominated optimal solution set, use the entropy weight method to give a comprehensive score to the non-dominated optimal solution set, and sort the comprehensive scores from smallest to largest to generate a sorted solution sequence; The graph structure features of the sorting scheme sequence are extracted, and the similarity between the graph structure features of the sorting scheme sequence and the graph structure features of historical schemes is calculated. Based on the similarity, the multi-criteria confidence score of the scheme is determined, and finally the scheme with the confidence score higher than the set threshold is selected as the final fault self-healing scheme.

[0009] Preferably, generating a candidate solution set includes: Based on the temporal electrical characteristics and spatial topology characteristics of the distribution network, the attention weight of the scene is determined. Based on the attention weight of the scene, the multi-scene characteristics of the distribution network are fused. Based on the fused multi-scene characteristics of the distribution network, the scene conditional probability is calculated. Based on the scene conditional probability, the scene recognition result is generated. The initial conditions of the graph convolutional diffusion model are determined based on the scene recognition results. The multi-step propagation cumulative failure probability of each node is calculated iteratively through the graph convolutional diffusion model. The set of faulty devices is determined based on the multi-step propagation cumulative failure probability. Based on the set of faulty equipment, the power supply point and the target point of power failure are determined. The shortest path from the power supply point to the target point of power failure is calculated by the improved Yen algorithm to determine the final path set. A candidate solution set is generated based on the final path set.

[0010] Preferably, the attention weights of the scene are expressed by the following formula:

[0011] In the formula, Let represent the attention weight of the k-th scenario at time t, where k represents the distribution network fault scenario, k=1 represents the bus fault scenario, k=2 represents the substation power outage scenario, and k=3 represents the loop power dispatch scenario. Represents an exponential function. Represents the similarity function of power distribution network scenarios. This represents the query vector for the k-th distribution network scenario. Represents the current power grid state key vector. , The weight matrix represents the key vector. Indicates the timing electrical characteristics of the distribution network. This represents the spatial topology characteristics of the distribution network.

[0012] Preferably, the characteristics of multiple scenarios in the power distribution network are integrated, as expressed by the following formula:

[0013] In the formula, This indicates the multi-scenario characteristics of integrated distribution networks. Represents the temporal-spatial feature weights. This represents the temporal feature component of the k-th scene. This represents the spatial feature component of the k-th scene.

[0014] Preferably, the scenario conditional probability is expressed by the following formula:

[0015] In the formula, Indicates scene condition indicator, Indicating the multi-scenario characteristics of integrated distribution networks The conditional probability in scenario k, This represents the conditional discriminator function. , This represents the weight vector for the k-th scene. This represents the bias term for the k-th scenario.

[0016] Preferably, the cumulative failure probability of multi-step propagation is expressed by the following formula:

[0017] In the formula, This represents the cumulative failure probability of node i across multiple propagation steps. This represents the failure probability of node i after the t-th propagation step. Representing a scene The maximum number of propagation steps.

[0018] Preferably, the power supply point and the target point of power loss are determined by the following formula:

[0019] In the formula, Indicates a power supply transfer point. Representing the distribution network topology There is a power supply point in the middle. To the depressurization target point path , This indicates finding the intersection; This indicates the existence of a power supply point and a target point of power failure. Belongs to the set of planned operations .

[0020] Preferably, determining the final path set includes: Dijkstra's algorithm is used to calculate the first shortest path from the power transfer point to the power failure target point. The candidate path set and the final path set are initialized, and the first shortest path is added to both the candidate path set and the final path set. For each deviation point on each path found, create a root path, which is a sub-path from the starting point to the deviation point. Temporarily remove all edges in the graph that pass through the root path except for the deviation point, and remove edges related to the set of faulty devices. In the modified temporary graph, calculate the shortest path from the deviation point to the depressurization target point. Merge the root path and the off-path into a new path, and verify whether the new path meets the scenario constraints. If the conditions are met and the new path is not in the candidate path set or the final path set, then add it to the candidate path set; Select the path with the smallest weight from the candidate path set as the next shortest path, add it to the final path set, and remove it from the candidate path set. Repeat the above process until the final path set reaches the set total number of paths.

[0021] Preferably, the multi-criteria confidence score is expressed by the following formula:

[0022] In the formula, This indicates a multi-criteria confidence score. , and These are the weighting coefficients. Representation scheme The similarity between the graph structure features of the current solution and the graph structure features of historical solutions. This indicates the rank of solution q in the sorted sequence of solutions. Indicates the number of sorting schemes. Let q represent the overall score of the q-th non-dominated optimal solution.

[0023] The second aspect of this invention discloses a multi-scenario whole-site fault self-healing system, and a multi-scenario whole-site fault self-healing method according to the first aspect, comprising: The initialization scheme module is used to calculate the attention weight of the scenario by extracting the temporal electrical characteristics and spatial topology characteristics of the distribution network to determine the scenario identification result, calculate the multi-step propagation cumulative fault based on the scenario identification result to determine the set of faulty equipment, determine the transfer power supply point and the undervoltage target point based on the set of faulty equipment, calculate the shortest path from the transfer power supply point to the undervoltage target point through the improved Yen algorithm, and generate a candidate scheme set based on the shortest path. The sorting module is used to construct the multi-objective function for each path in the candidate solution set. The Pareto algorithm is used to perform multi-objective optimization search on the multi-objective function to obtain the non-dominated optimal solution set. The entropy weight method is used to give a comprehensive score to the non-dominated optimal solution set and sort the comprehensive scores from smallest to largest to generate a sorted solution sequence. The output module is used to extract the graph structure features of the sorting scheme sequence, calculate the similarity between the graph structure features of the sorting scheme sequence and the graph structure features of historical schemes, determine the multi-criteria confidence score of the scheme based on the similarity, and finally select the scheme with the confidence score higher than the set threshold as the final fault self-healing scheme.

[0024] A third aspect of the present invention provides a terminal, including a processor and a storage medium; the storage medium is used to store instructions; the processor is used to operate according to the instructions to execute the steps of the multi-scenario whole-site fault self-healing method according to the first aspect.

[0025] A fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, including steps for implementing the multi-scenario whole-site fault self-healing method described in the first aspect when the program is executed by a processor.

[0026] Compared with the prior art, the beneficial effects of the present invention include at least the following: This invention addresses the technical problems in the background technology of "lacking a classification and self-healing mechanism for whole-station faults under different scenarios" and "relying solely on manual fault discovery, fault location, and emergency response plan formulation, which is time-consuming, lacks real-time performance, and has low accuracy" by using graph convolutional diffusion fault simulation and improved Yen path search. This invention enhances the distribution network's ability to intelligently identify and accurately distinguish three typical scenarios: bus faults, whole-station voltage loss, and closed-loop power dispatch, as well as its ability to automatically, quickly, and accurately assess the impact range and propagation path of complex faults. At the same time, it reduces the risk of delays, decision-making errors, and large-scale, long-term power outages caused by misjudgment of scenarios, manual reliance on fault location, and lack of cross-scenario contingency plans in the distribution network. This invention addresses the technical problems in the background technology, namely, the low accuracy caused by manually formulating emergency response plans, and the difficulty of traditional methods in quickly selecting globally near-optimal solutions under multiple conflicting objectives such as "minimizing load shedding, minimizing operations, and achieving the most balanced load," by employing a combination of multi-objective optimization and entropy weighting for ranking. This invention transforms the dispatcher's empirical trade-offs into a quantifiable multi-objective mathematical optimization problem, improving the decision-making ability and efficiency of the distribution network in automating and intelligently optimizing and ranking massive candidate solutions under complex constraints. It also reduces the problems of suboptimal solutions, operational redundancy, uncontrollable recovery time, and increased operational risks caused by limitations in human experience, single-objective decision-making, or simple rule-based ranking in the distribution network. This invention addresses the problems of low accuracy of schemes and insufficient reliability and consistency caused by reliance on manual review in the prior art, as well as the lack of intelligent scheme verification and comparison mechanisms based on historical successful experience, through the technical means of historical scheme similarity measurement and multi-criteria confidence fusion decision-making. This invention achieves intelligent comparison with historical knowledge base, improving the intelligence level of distribution network in the reliability verification of automatically generated schemes, historical experience fusion, and final execution decision-making; it reduces the probability of scheme adoption errors, increased operational risks, and self-healing execution failures caused by oversights in manual review, insufficient comparative analysis, or subjective experience bias in distribution network, ensuring the credibility and reliability accuracy of the final execution scheme. Attached Figure Description

[0027] Figure 1 This is a schematic diagram of a multi-scenario whole-site fault self-healing method provided in accordance with an embodiment of the present invention. Detailed Implementation

[0028] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of this invention. The described embodiments are merely some embodiments of this invention, and not all embodiments. Based on the spirit of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the protection scope of this invention.

[0029] like Figure 1 As shown, Embodiment 1 of the present invention provides a multi-scenario whole-site fault self-healing method and system, including: Step 1: Calculate the scenario probability by extracting the temporal electrical characteristics and spatial topology characteristics of the distribution network, determine the set of faulty equipment based on the graph diffusion model, select the power supply point and target point for transfer, and use the improved Yen algorithm to search for k shortest paths to generate preliminary candidate transfer schemes.

[0030] Step 1.1: Determine the attention weight of the scene based on the temporal electrical characteristics and spatial topology characteristics of the distribution network; calculate the scene conditional probability based on the scene attention weight and fuse the multi-scene features of the distribution network based on the scene attention weight; generate the scene recognition result based on the scene conditional probability.

[0031] Step 1.1.1: Extract the time-series electrical features and spatial topology features of the distribution network from the real-time monitoring data of the distribution network.

[0032] More preferably, step 1.1.1 includes: Extracting time-series electrical features of the distribution network using LSTM It can be expressed by the following formula:

[0033] In the formula, LSTM represents a Long Short-Term Memory Neural Network, used to capture the temporal characteristics of transient processes during power distribution network faults. This represents the measured three-phase voltage value of the 10kV busbar, used for detecting busbar undervoltage and station-wide undervoltage scenarios. This represents the vector of three-phase current measurements on the low-voltage side of the main transformer, used to verify the no-current condition. It represents the protection signal sequence vector, including key criteria such as low-voltage backup protection signals and outgoing line protection action signals; This indicates the fault detection time window for the distribution network. Different values ​​are set according to different fault scenarios in the distribution network. For example, but not limited to bus fault scenarios, the detection window is 15 seconds. Timing starts when the voltage of any 10kV bus drops below 1kV, continuously monitoring the voltage, current, and protection signal sequence for 15 seconds to confirm the bus undervoltage state. In a substation undervoltage scenario, the detection window is 20 seconds. When all 10kV buses of the substation are detected to be undervoltage and there is no current in the low-voltage side switch of the main transformer, a 20-second continuous monitoring window is initiated to verify whether the substation's no-voltage and no-current conditions meet the automatic power transfer start-up requirements. In a closed-loop power dispatch scenario, the detection window is set to 5 seconds. Based on the characteristics of manual triggering, only a short confirmation of the steady-state operating status is needed. Operational safety is ensured through verification of steady-state data within 5 seconds, and the system is directly triggered by manual interface operation commands.

[0034] Extracting spatial topology features of the distribution network using GCN It can be expressed by the following formula:

[0035] In the formula, This represents the adjacency matrix of equipment connections within a substation, describing the electrical connection relationships between the outgoing lines of the busbar transformer switch. This represents the equipment feature matrix, which includes electrical parameter information such as equipment type, capacity, and status. The graph convolutional neural network is optimized for the radial topology of power distribution networks and used to extract spatial topological features.

[0036] It is worth noting that the present invention uses LSTM-based temporal feature extraction and GCN-based spatial topology feature extraction techniques to solve the problem of low efficiency in manually and serially extracting key fault features from massive real-time monitoring data. The present invention utilizes LSTM networks to quickly and automatically capture the temporal evolution patterns of voltage, current, and protection signals, and utilizes GCN optimized for the radial structure of distribution networks to efficiently and in parallel aggregate equipment connection and status information, thereby improving the automation efficiency of multimodal fault feature extraction and the completeness and accuracy of feature representation.

[0037] Step 1.1.2: Determine the attention weight of the scenario based on the time-series electrical characteristics and spatial topology characteristics of the distribution network obtained in Step 1.1.1, and calculate the multi-scenario characteristics of the distribution network based on the attention weight of the scenario.

[0038] More preferably, step 1.1.2 includes: The attention weights for a scene are expressed by the following formula:

[0039] In the formula, This represents the attention weight of the k-th scenario at time t, which directly reflects the matching probability between the current power distribution network state and each scenario. This represents the query vector for the k-th distribution network scenario. It includes: for bus fault scenarios, encoding single bus voltage anomalies and associated protection actions, focusing on capturing the timing correlation between a 10kV bus voltage drop to less than 0.1 pu and low-voltage backup protection signals; for substation-wide voltage loss scenarios, encoding substation-wide voltage and current loss and multi-criteria verification, focusing on the spatial distribution characteristics of simultaneous voltage loss of bus sections I / II, current loss of main transformers, and current loss of outgoing lines; and for loop-based power dispatch scenarios, encoding manual triggering and fault-free characteristics, identifying the combination characteristics of interface operation signals and normal electrical quantity changes. Represents an exponential function. Represents the current power grid state key vector. , The weight matrix represents the key vector; The similarity function for power distribution network scenarios is expressed by the following formula:

[0040] In the formula, This represents the Euclidean norm.

[0041] Based on the attention weight of the distribution network scenario and the fusion of multiple scenario features of the distribution network, it is expressed by the following formula:

[0042] In the formula, This indicates the multi-scenario characteristics of integrated distribution networks. Represents the temporal-spatial feature weights. This represents the temporal feature component of the k-th scene. Let k represent the spatial feature component of the k-th scenario, where k represents the distribution network fault scenario, k=1 represents the bus fault scenario, k=2 represents the substation power outage scenario, and k=3 represents the loop power dispatch scenario.

[0043] It is worth noting that this invention solves the problem of diluted discriminative information caused by simple splicing or average fusion of features from different fault scenarios in traditional methods by using scene weight calculation and feature fusion based on attention mechanism. This invention quickly determines the contribution of each scenario by calculating the attention weight of the current power grid state and each preset scenario, and then performs weighted fusion of temporal and spatial features accordingly, thereby improving the computational efficiency of cross-scenario feature fusion and the accuracy of scene-related information extraction, and generating a more discriminative comprehensive feature vector.

[0044] Step 1.1.3: Calculate the scenario conditional probability based on the multi-scenario characteristics of the integrated distribution network, and generate the scenario recognition result based on the scenario conditional probability.

[0045] More preferably, step 1.1.3 includes: The conditional probability of a scenario is expressed by the following formula:

[0046] In the formula, Indicates scene condition indicator, This indicates the presence of a low-voltage backup protection signal; otherwise, it indicates the absence of a low-voltage backup protection signal. This is the core criterion for determining the presence of a low-voltage backup protection signal in bus fault scenarios. This indicates that the conditions for a complete site-wide load and flow reduction are met; otherwise, the conditions are not met. This is the core criterion for determining whether a site-wide load reduction scenario is met. This indicates a manual trigger flag; otherwise, it will not be triggered. This is the core criterion for closed-loop power regulation scenarios. Indicating the multi-scenario characteristics of integrated distribution networks The conditional probability in scenario k, This represents the conditional discriminator function. , This represents the weight vector for the k-th scene. This represents the bias term for the k-th scenario.

[0047] Scene recognition results are generated based on scene conditional probability. It can be expressed by the following formula:

[0048] In the formula, Indicates the scene detection timestamp. Indicates the effective duration of the scene. This indicates the probability that only scenario k is true. Exceeding the preset threshold At that time, Set as scene recognition result .

[0049] It is worth noting that this invention solves the problems of poor flexibility and low fault tolerance caused by scene determination relying on fixed logical rules by using a technical means of scene probability calculation and threshold decision based on a condition discriminator. This invention inputs the fused features and the scene core criterion indicator into the condition discriminator to quickly calculate the probability of belonging to each scene, and achieves hard decision by comparing it with a preset threshold, which significantly improves the decision efficiency and recognition accuracy of multi-scene classification, and realizes reliable and rapid differentiation of bus fault, station undervoltage, and loop power dispatch scenarios.

[0050] Step 1.2: Determine the initial conditions of the graph convolutional diffusion model based on the scene recognition results, iteratively calculate the multi-step propagation cumulative fault probability of each node through the graph convolutional diffusion model, and determine the set of faulty devices based on the multi-step propagation cumulative fault probability.

[0051] Step 1.2.1: Initialize the initial conditions of the graph convolutional diffusion model according to the fault scenario type and scenario conditional probability. The initial conditions include the initial fault state of the distribution network node, the adjacency matrix of the distribution network scenario, and the equipment feature matrix.

[0052] The initial fault states of distribution network nodes are constructed using the following formula:

[0053] In the formula, This represents the initial fault state of node i at level 0, where 1 indicates an initial fault and 0 indicates normal. Indicates scenario-based The initial set of fault nodes, in the case of a bus fault scenario. In the event of a complete site-wide power failure, In the scenario of closed-loop power regulation, {Switch to be operated, busbar to be adjusted} The threshold value represents the conditional probability of the scenario; in this invention, it is set to 0.7, and i represents the node number. , This indicates the total number of nodes in the distribution network topology.

[0054] Constructing an adjacency matrix for a power distribution network scenario It can be expressed by the following formula:

[0055] In the formula, This represents the physical connection adjacency matrix based on the actual topology of the distribution network. Represents the Hadamard product, implementing scene connectivity constraints. Representing a scene The influence mask matrix, in the case of bus fault scenarios. It means if and only if the line ,List The corresponding node's influence mask matrix within the same voltage level, with a value of 10kV for the same voltage level, in the case of a total station power outage scenario. This indicates all rows on the entire site. ,List The corresponding node influence mask matrix, in the closed-loop power regulation scenario, Indicates the upward movement of the planned operation path. ,List The influence mask matrix of the nodes.

[0056] Constructing the device feature matrix It can be expressed by the following formula:

[0057] In the formula, This indicates the equipment type code, including busbars, switches, and transformers, etc. This indicates the current status, including the device's open state, closed state, energized state, and de-energized state. It indicates electrical parameters, including rated capacity, current, voltage, etc.

[0058] It is worth noting that this invention solves the problem of a single and unspecific initial state setting when constructing a fault propagation model by using a scenario-adaptive graph model initialization technique. Based on the identified scenario types and their probabilities, this invention determines the initial fault nodes, constructs an adjacency matrix with scenario masks and a comprehensive equipment feature matrix, thereby improving the preparation efficiency of graph diffusion model initialization and its modeling accuracy for specific fault scenarios, and laying an accurate graph structure foundation for accurately simulating fault propagation.

[0059] Step 1.2.2: Construct a graph convolutional diffusion model using the initial conditions of the graph convolutional diffusion model. Determine the fault state of the final layer nodes by performing a set number of iterations on the graph convolutional diffusion model. Determine the multi-step propagation cumulative fault probability of each node based on the fault state of the final layer nodes.

[0060] More preferably, step 1.2.2 includes: Construct a graph convolutional diffusion model, expressed by the following formula:

[0061] In the formula, Indicates the first Fault status of layer nodes, Represents the augmented adjacency matrix. , Represents the identity matrix. Degree matrix, , Indicates the degree of node i. express The (i, j)th element, Represents the learnable weights. This represents a learnable bias. This represents the activation function.

[0062] The graph convolutional diffusion model is iterated through a set number of L rounds to obtain the fault state of the node in the final Lth layer. , Represents a node i ,node j The final feature vector, where N is the number of nodes in the Lth layer.

[0063] The conditional probability of fault propagation is determined based on the fault state of the Lth node in the final layer, expressed by the following formula:

[0064] In the formula, Indicates a fault from the node i propagation to nodes j The conditional probability, Graph neural networks representing distribution network optimization Let i represent the set of neighbors of node i. Let i be a neighboring node of node i.

[0065] The cumulative failure probability in multi-step propagation is determined based on the conditional probability of failure propagation, expressed by the following formula:

[0066] In the formula, This represents the cumulative failure probability of node i across multiple propagation steps. This represents the failure probability of node i after the t-th propagation step. Representing a scene Maximum propagation steps, maximum propagation steps in bus fault scenarios Maximum propagation steps in a site-wide pressure failure scenario Maximum propagation steps in a closed-loop power regulation scenario .

[0067] It is worth noting that this invention solves the problem of inaccuracy and incompleteness caused by relying on empirical estimation or simple topology search for fault impact range by using multi-layer convolution diffusion and scene-aware fault propagation probability calculation techniques. This invention simulates the spread of fault state in the network through multi-layer graph convolution iteration and uses graph neural networks to calculate the conditional probability of fault propagation between nodes, thereby accumulating the comprehensive fault probability of nodes, improving the computational efficiency of fault range simulation and the dynamic and refined level of fault probability assessment.

[0068] Step 1.2.3: Determine the set of faulty devices based on the cumulative fault probability of multi-step propagation.

[0069] Boundary device set It can be expressed by the following formula:

[0070] In the formula, V represents the set of all nodes, V = {1, 2, ..., N}, which includes all equipment nodes in the distribution network. Representing a scene Under the following boundary threshold, in the bus fault scenario, To avoid excessive isolation in the event of a site-wide failure, To ensure full coverage in closed-loop power regulation scenarios, Only specify the equipment. Representing a scene The set of influence ranges under the following conditions: the set of influence ranges under the bus fault scenario represents the union of the electrical neighbor sets of all nodes within the fault bus set; the set of influence ranges under the whole station under the power outage scenario represents all 10kV equipment; and the set of influence ranges under the loop power dispatch scenario represents the nodes on the planned operation path.

[0071] The set of faulty devices is generated using the following formula:

[0072] In the formula, Represents the set of faulty devices. This indicates the key equipment being installed, including the main transformer and important outgoing lines. This indicates the set critical equipment failure threshold. Indicates the key equipment number.

[0073] It is worth noting that this invention solves the problem of over-isolation or under-isolation caused by rigid fault determination criteria, such as, but not limited to, relying solely on electrical quantities, through a fault boundary and device set generation technology based on probability thresholds. This invention sets differentiated probability thresholds based on scenarios, filters out nodes whose cumulative fault probability exceeds the threshold to form boundaries, and includes them in the inspection of key equipment, thereby improving the decision-making efficiency and judgment accuracy of the final fault device set generation, and achieving accurate and reliable definition of the fault range.

[0074] Step 1.3: Based on the set of faulty equipment, determine the power supply point and the target point of power failure. Calculate the shortest path from the power supply point to the target point of power failure using the improved Yen algorithm to obtain the final path set. Generate a candidate solution set based on the final path set.

[0075] More preferably, step 1.3 includes: Based on the set of faulty equipment, determine the power supply point and the target point of power loss, expressed by the following formula:

[0076] In the formula, Indicates a power supply transfer point. This is a busbar fault scenario. express This is another normal 10kV busbar within this station. Representing the distribution network topology There is a power supply point in the middle. To the depressurization target point path , From the set of faulty devices The 10kV faulty busbar extracted from it This indicates finding the intersection; This is a scenario where the entire site experiences a power outage. For example, but not limited to, 10kV tie line switches and power supply lines of adjacent substations. From the set of faulty devices The 10kV busbar of the entire station was extracted; This is a closed-loop power regulation scenario. This indicates the planned power transfer point. This indicates the existence of a power supply point and a target point of power failure. Belongs to the set of planned operations ; This indicates the target point of depressurization.

[0077] By improving the Yen algorithm to calculate the shortest path from the power transfer point to the target point of power failure, the final path set is obtained, including: Use Dijkstra's algorithm to calculate the first shortest path p1 from source to target, initialize the candidate path set and the final path set, and add p1 to the two sets. For each deviation point on each found path, create a root path, which is a sub-path from the starting point to the deviation point. Temporarily remove all edges in the graph that pass through the root path except for the deviation point, and remove edges related to the set of faulty devices. In the modified temporary graph, calculate the shortest path from the deviation point to the target. Merge the root path and the off-path into a new path, and verify whether the new path meets the scenario constraints. If the conditions are met and the new path is not in the candidate path set or the final path set, then add it to the candidate path set; Select the path with the smallest weight from the candidate path set as the next shortest path, add it to the final path set, and remove it from the candidate path set. Repeat the above process until the final path set reaches the set total number of paths, expressed by the following formula:

[0078] In the formula, This represents an improvement on Yen's k-shortest path algorithm. Indicates the number of search paths. Represents the set of k final paths. .

[0079] Based on the pre-defined operations for each path, a set of candidate solutions is generated from the final set of paths.

[0080] It is worth noting that this invention solves the problems of search space explosion and low efficiency when finding feasible transfer paths for multiple source-target points in complex power grids by using the K-shortest path search technique based on the improved Yen algorithm. This invention introduces fault equipment avoidance and scenario constraint verification into the classic Yen algorithm, and dynamically prunes the graph structure each time it deviates from the path search, thereby improving the enumeration efficiency and generation speed of feasible transfer paths that meet safety constraints, and ensuring the diversity and feasibility of the candidate solution set.

[0081] Step 2: Construct a multi-objective function for each path in the candidate solution set, perform multi-objective optimization search on the multi-objective function using the Pareto algorithm to obtain a non-dominated optimal solution set, use the entropy weight method to give a comprehensive score to the non-dominated optimal solution set, and sort the comprehensive scores from smallest to largest to generate a sorted solution sequence.

[0082] Step 2.1: Construct the multi-objective function for each path in the candidate solution set, expressed by the following formula:

[0083] In the formula, Represents the a-th candidate solution Load shedding ratio, Represents the a-th candidate solution Total load shedding Indicates the total load before the fault. Represents the a-th candidate solution The ratio of switching operands, Represents the a-th candidate solution The number of switches that need to be operated. Indicates the maximum number of allowed switching operations. Represents the a-th candidate solution Load balancing Represents the a-th candidate solution Maximum load rate of each line after transfer Represents the a-th candidate solution Minimum load rate of each line after transfer, Represents the a-th candidate solution Average load rate of each line after transfer.

[0084] It is worth noting that this invention solves the problem of inconsistent dimensions and difficulty in direct comparison when evaluating the multi-dimensional performance of different candidate schemes by constructing a normalized multi-objective evaluation function. This invention transforms the load shedding amount, switching operands, and load balancing degree into unified relative proportions or exponents, which quickly realizes the standardization and quantification of each scheme in different dimensions, improves the computational efficiency and comparability of indicators in the multi-dimensional performance evaluation of the schemes, and provides standard input for subsequent multi-objective optimization.

[0085] Step 2.2: Use the Pareto algorithm to perform multi-objective optimization search on the multi-objective function to obtain the non-dominated optimal solution set.

[0086] Step 2.3: Prioritize the non-dominated optimal solution set according to the entropy weight method.

[0087] Step 2.3.1: Calculate the entropy weights for the multi-objective objectives based on the non-dominated optimal solution set, expressed by the following formula:

[0088] In the formula, Let represent the entropy weight of target d, where d∈{1,2,3} corresponds to the load shedding ratio, the switching operand ratio, and the load balancing degree, respectively. The information entropy of target d is expressed by the following formula:

[0089] In the formula, This represents the number of non-dominated optimal solutions in the set of non-dominated optimal solutions. The weight of the q-th non-dominated optimal solution in objective d is expressed by the following formula:

[0090] In the formula, Let d represent the objective of the q-th non-dominated optimal solution.

[0091] Step 2.3.2: Determine the final comprehensive weight of the multi-objectives based on their entropy weights, expressed by the following formula:

[0092] In the formula, This represents the final overall weight of objective d. This represents the scene-adaptive weights for target d. This represents the weighted fusion coefficient.

[0093] Step 2.3.3: Calculate the comprehensive score of the non-dominated optimal solution based on the final comprehensive weight of the multi-objectives, sort the scores from smallest to largest, and obtain the sorted sequence of solutions, expressed by the following formula:

[0094] In the formula, This represents the overall score of the q-th non-dominated optimal solution; the smaller the value, the better the solution.

[0095] The comprehensive scores of all non-dominated optimal solutions are sorted from smallest to largest to obtain a sequence of sorting schemes.

[0096] It is worth noting that this invention solves the problem of subjective weighting bias when selecting the final execution scheme from the Pareto optimal solution set by using a comprehensive scoring and ranking technique based on entropy weighting. This invention objectively calculates the weights based on the degree of dispersion of each target value in the solution set, i.e., information entropy, and combines it with prior knowledge of the scenario to calculate the comprehensive score, thereby improving the objectivity of the comprehensive score, the computational efficiency, and the scientific nature of the final ranking result.

[0097] Step 3: Extract the graph structure features of the sorting scheme sequence, calculate the similarity between the graph structure features of the sorting scheme sequence and the graph structure features of historical schemes, determine the multi-criteria confidence score of the scheme based on the similarity, and finally select the scheme with the confidence score higher than the set threshold as the final fault self-healing scheme.

[0098] Step 3.1: Input the sorted scheme sequence into a graph convolutional network to extract the graph structure features of the sorting scheme; Historical schemes are input into a graph convolutional network to extract the graph structure features of the historical schemes.

[0099] Step 3.2: Compare the similarity between the graph structure features of the sorting scheme and the graph structure features of the historical schemes, expressed by the following formula:

[0100] In the formula, Represent the sorting scheme The similarity between the graph structure features of the current solution and the graph structure features of historical solutions. This represents the number of historical solutions in the historical solution database that are most similar to the sorted solution. This represents the set of historical solutions most similar to the sorting scheme. A label indicating that a historical plan was successfully executed. Represent the sorting scheme Graph eigenvectors, Representing historical schemes The graph eigenvectors.

[0101] It is worth noting that this invention solves the problems of low efficiency and poor accuracy caused by relying on text descriptions or simple rule matching when finding historical cases for reference in the current solution by using K-nearest neighbor retrieval technology based on feature vector similarity and historical success labels. This invention quickly retrieves the K most similar cases from the historical database by calculating the similarity of graph feature vectors, and improves the efficiency of historical experience retrieval and the matching accuracy and confidence value of the obtained reference cases by weighting the similarity with historical execution success labels.

[0102] Step 3.3: Based on the comprehensive score of all non-dominated optimal solutions in Step 2.3.3 and the similarity between the graph structure features of the ranking scheme and the graph structure features of historical schemes, determine the multi-criteria confidence score of the scheme, expressed by the following formula:

[0103] In the formula, The multi-criteria confidence score for scheme q is represented. , and These are the weighting coefficients. This indicates the rank of solution q in the sorted sequence of solutions. Indicates the number of sorting schemes. Representation scheme The similarity between the graph structure features of the current scheme and the graph structure features of historical schemes.

[0104] The scheme with a confidence score greater than the set threshold is selected as the final fault self-healing scheme.

[0105] It is worth noting that this invention solves the risk of relying on a single indicator or human experience when making a final decision by integrating similarity, ranking and scoring multi-criteria confidence assessment technology. This invention integrates historical similarity, solution quality ranking and multi-objective scoring into a unified confidence score through a weighted formula, and improves the comprehensive consideration efficiency and automation level of the final implementation solution decision by automatically filtering through thresholds, and reduces the risk of subjective misjudgment.

[0106] Embodiment 2 of this invention proposes a multi-scenario whole-site fault self-healing system, which, according to the multi-scenario whole-site fault self-healing method described in Embodiment 1, includes: The initialization scheme module is used to calculate the attention weight of the scenario by extracting the temporal electrical characteristics and spatial topology characteristics of the distribution network to determine the scenario identification result, calculate the multi-step propagation cumulative fault based on the scenario identification result to determine the set of faulty equipment, determine the transfer power supply point and the undervoltage target point based on the set of faulty equipment, calculate the shortest path from the transfer power supply point to the undervoltage target point through the improved Yen algorithm, and generate a candidate scheme set based on the shortest path. The sorting module is used to construct the multi-objective function for each path in the candidate solution set. The Pareto algorithm is used to perform multi-objective optimization search on the multi-objective function to obtain the non-dominated optimal solution set. The entropy weight method is used to give a comprehensive score to the non-dominated optimal solution set and sort the comprehensive scores from smallest to largest to generate a sorted solution sequence. The output module is used to extract the graph structure features of the sorting scheme sequence, calculate the similarity between the graph structure features of the sorting scheme sequence and the graph structure features of historical schemes, determine the multi-criteria confidence score of the scheme based on the similarity, and finally select the scheme with the confidence score higher than the set threshold as the final fault self-healing scheme.

[0107] Embodiment 3 of the present invention proposes a terminal, including a processor and a storage medium; the storage medium is used to store instructions; the processor is used to operate according to the instructions to execute the steps of the multi-scenario whole-site fault self-healing method according to Embodiment 1.

[0108] Embodiment 4 of the present invention proposes a computer-readable storage medium storing a computer program thereon, including steps for implementing the multi-scenario whole-site fault self-healing method described in Embodiment 1 when the program is executed by a processor.

[0109] Compared with the prior art, the beneficial effects of the present invention include at least the following: This invention addresses the technical problems in the background technology of "lacking a classification and self-healing mechanism for whole-station faults under different scenarios" and "relying solely on manual fault discovery, fault location, and emergency response plan formulation, which is time-consuming, lacks real-time performance, and has low accuracy" by using graph convolutional diffusion fault simulation and improved Yen path search. This invention enhances the distribution network's ability to intelligently identify and accurately distinguish three typical scenarios: bus faults, whole-station voltage loss, and closed-loop power dispatch, as well as its ability to automatically, quickly, and accurately assess the impact range and propagation path of complex faults. At the same time, it reduces the risk of delays, decision-making errors, and large-scale, long-term power outages caused by misjudgment of scenarios, manual reliance on fault location, and lack of cross-scenario contingency plans in the distribution network. This invention addresses the technical problems in the background technology, namely, the low accuracy caused by manually formulating emergency response plans, and the difficulty of traditional methods in quickly selecting globally near-optimal solutions under multiple conflicting objectives such as "minimizing load shedding, minimizing operations, and achieving the most balanced load," by employing a combination of multi-objective optimization and entropy weighting for ranking. This invention transforms the dispatcher's empirical trade-offs into a quantifiable multi-objective mathematical optimization problem, improving the decision-making ability and efficiency of the distribution network in automating and intelligently optimizing and ranking massive candidate solutions under complex constraints. It also reduces the problems of suboptimal solutions, operational redundancy, uncontrollable recovery time, and increased operational risks caused by limitations in human experience, single-objective decision-making, or simple rule-based ranking in the distribution network. This invention addresses the problems of low accuracy of schemes and insufficient reliability and consistency caused by reliance on manual review in the prior art, as well as the lack of intelligent scheme verification and comparison mechanisms based on historical successful experience, through the technical means of historical scheme similarity measurement and multi-criteria confidence fusion decision-making. This invention achieves intelligent comparison with historical knowledge base, improving the intelligence level of distribution network in the reliability verification of automatically generated schemes, historical experience fusion, and final execution decision-making; it reduces the probability of scheme adoption errors, increased operational risks, and self-healing execution failures caused by oversights in manual review, insufficient comparative analysis, or subjective experience bias in distribution network, ensuring the credibility and reliability accuracy of the final execution scheme.

[0110] This disclosure can be a system, method, and / or computer program product. A computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for causing a processor to implement various aspects of this disclosure.

[0111] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the claims of the present invention.

Claims

1. A multi-scenario whole-site fault self-healing method, characterized in that: The attention weight of the scenario is calculated by extracting the temporal electrical features and spatial topology features of the distribution network to determine the scenario identification result. Based on the scenario identification result, the multi-step propagation cumulative fault is calculated to determine the set of faulty equipment. Based on the set of faulty equipment, the transfer power supply point and the undervoltage target point are determined. The shortest path from the transfer power supply point to the undervoltage target point is calculated by improving the Yen algorithm. Based on the shortest path, a set of candidate solutions is generated. Construct a multi-objective function for each path in the candidate solution set, perform multi-objective optimization search on the multi-objective function using the Pareto algorithm to obtain a non-dominated optimal solution set, use the entropy weight method to give a comprehensive score to the non-dominated optimal solution set, and sort the comprehensive scores from smallest to largest to generate a sorted solution sequence; The graph structure features of the sorting scheme sequence are extracted, and the similarity between the graph structure features of the sorting scheme sequence and the graph structure features of historical schemes is calculated. Based on the similarity, the multi-criteria confidence score of the scheme is determined, and finally the scheme with the confidence score higher than the set threshold is selected as the final fault self-healing scheme.

2. The multi-scenario whole-site fault self-healing method according to claim 1, characterized in that: The generated candidate solution set includes: Based on the temporal electrical characteristics and spatial topology characteristics of the distribution network, the attention weight of the scene is determined. Based on the attention weight of the scene, the multi-scene characteristics of the distribution network are fused. Based on the fused multi-scene characteristics of the distribution network, the scene conditional probability is calculated. Based on the scene conditional probability, the scene recognition result is generated. The initial conditions of the graph convolutional diffusion model are determined based on the scene recognition results. The multi-step propagation cumulative failure probability of each node is calculated iteratively through the graph convolutional diffusion model. The set of faulty devices is determined based on the multi-step propagation cumulative failure probability. Based on the set of faulty equipment, the power supply point and the target point of power failure are determined. The shortest path from the power supply point to the target point of power failure is calculated by the improved Yen algorithm to determine the final path set. A candidate solution set is generated based on the final path set.

3. The multi-scenario whole-site fault self-healing method according to claim 2, characterized in that: The attention weights for a scene are expressed by the following formula: In the formula, Let represent the attention weight of the k-th scenario at time t, where k represents the distribution network fault scenario, k=1 represents the bus fault scenario, k=2 represents the substation power outage scenario, and k=3 represents the loop power dispatch scenario. Represents an exponential function. Represents the similarity function of power distribution network scenarios. This represents the query vector for the k-th distribution network scenario. Represents the current power grid state key vector. , The weight matrix represents the key vector. Indicates the timing electrical characteristics of the distribution network. This represents the spatial topology characteristics of the distribution network.

4. The multi-scenario whole-site fault self-healing method according to claim 3, characterized in that: The characteristics of multiple scenarios in the integrated power distribution network are expressed by the following formula: In the formula, This indicates the multi-scenario characteristics of integrated distribution networks. Represents the temporal-spatial feature weights. This represents the temporal feature component of the k-th scene. This represents the spatial feature component of the k-th scene.

5. The multi-scenario whole-site fault self-healing method according to claim 4, characterized in that: The conditional probability of a scenario is expressed by the following formula: In the formula, Indicates scene condition indicator, Indicating the multi-scenario characteristics of integrated distribution networks The conditional probability in scenario k, This represents the conditional discriminator function. , This represents the weight vector for the k-th scene. This represents the bias term for the k-th scenario.

6. The multi-scenario whole-site fault self-healing method according to claim 5, characterized in that: The cumulative failure probability after multi-step propagation is expressed by the following formula: In the formula, This represents the cumulative failure probability of node i across multiple propagation steps. This represents the failure probability of node i after the t-th propagation step. Representing a scene The maximum number of propagation steps.

7. The multi-scenario whole-site fault self-healing method according to claim 6, characterized in that: The power supply point and the target point of power failure are determined using the following formula: In the formula, Indicates a power supply transfer point. Representing the distribution network topology There is a power supply point in the middle. To the depressurization target point path , This indicates finding the intersection; This indicates the existence of a power supply point and a target point of power failure. Belongs to the set of planned operations .

8. The multi-scenario whole-site fault self-healing method according to claim 7, characterized in that: Determine the final set of paths, including: Dijkstra's algorithm is used to calculate the first shortest path from the power transfer point to the power failure target point. The candidate path set and the final path set are initialized, and the first shortest path is added to both the candidate path set and the final path set. For each deviation point on each path found, create a root path, which is a sub-path from the starting point to the deviation point. Temporarily remove all edges in the graph that pass through the root path except for the deviation point, and remove edges related to the set of faulty devices. In the modified temporary graph, calculate the shortest path from the deviation point to the depressurization target point. Merge the root path and the off-path into a new path, and verify whether the new path meets the scenario constraints. If the conditions are met and the new path is not in the candidate path set or the final path set, then add it to the candidate path set; Select the path with the smallest weight from the candidate path set as the next shortest path, add it to the final path set, and remove it from the candidate path set. Repeat the above process until the final path set reaches the set total number of paths.

9. The multi-scenario whole-site fault self-healing method according to claim 1, characterized in that: The multi-criteria confidence score is expressed by the following formula: In the formula, This indicates a multi-criteria confidence score. , and These are the weighting coefficients. Representation scheme The similarity between the graph structure features of the current solution and the graph structure features of historical solutions. This indicates the rank of solution q in the sorted sequence of solutions. Indicates the number of sorting schemes. Let q represent the overall score of the q-th non-dominated optimal solution.

10. A multi-scenario whole-site fault self-healing system, and a multi-scenario whole-site fault self-healing method according to any one of claims 1-9, characterized in that: The initialization scheme module is used to calculate the attention weight of the scenario by extracting the temporal electrical characteristics and spatial topology characteristics of the distribution network to determine the scenario identification result, calculate the multi-step propagation cumulative fault based on the scenario identification result to determine the set of faulty equipment, determine the transfer power supply point and the undervoltage target point based on the set of faulty equipment, calculate the shortest path from the transfer power supply point to the undervoltage target point through the improved Yen algorithm, and generate a candidate scheme set based on the shortest path. The sorting module is used to construct the multi-objective function for each path in the candidate solution set. The Pareto algorithm is used to perform multi-objective optimization search on the multi-objective function to obtain the non-dominated optimal solution set. The entropy weight method is used to give a comprehensive score to the non-dominated optimal solution set and sort the comprehensive scores from smallest to largest to generate a sorted solution sequence. The output module is used to extract the graph structure features of the sorting scheme sequence, calculate the similarity between the graph structure features of the sorting scheme sequence and the graph structure features of historical schemes, determine the multi-criteria confidence score of the scheme based on the similarity, and finally select the scheme with the confidence score higher than the set threshold as the final fault self-healing scheme.

11. A terminal, comprising a processor and a storage medium; characterized in that: The storage medium is used to store instructions; The processor is configured to operate according to the instructions to perform the steps of the multi-scenario whole-site fault self-healing method according to any one of claims 1-9.

12. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps of the multi-scenario whole-site fault self-healing method as described in any one of claims 1-9.

Images