A fast recovery method and system based on regional power distribution network fault reconstruction

Abstract

This invention discloses a rapid recovery method and system based on regional distribution network fault reconstruction, belonging to the field of power system digital simulation and distribution automation technology. The method includes real-time acquisition of system operation data of the distribution network after a fault, constructing a distribution network topology map and extracting isolated subgraphs with link switches as source-target nodes, then constructing a subgraph feature information matrix, inputting it into a pre-trained subgraph neural network model to obtain connectivity prediction results for each potential connection link, generating and executing dynamic reconstruction control commands accordingly to achieve rapid distribution network recovery. The system includes a sensor acquisition module, a communication module, an edge computing module, and a control execution module. By integrating physical constraints and local graph features, this invention significantly reduces computational complexity and recovery time while ensuring the differences between subgraphs, achieving rapid and reliable recovery of the distribution network after a fault.

Description

Technical Field

[0001] This invention relates to the field of power system digital simulation technology, specifically to a rapid recovery method and system based on regional distribution network fault reconstruction. Background Technology

[0002] In recent years, frequent extreme weather events such as hurricanes, torrential rains, and droughts have significantly increased the risk of large-scale cascading failures in power grids. These large-scale blackouts, triggered by interdependent chains of failures, not only cause severe economic losses but also pose a significant threat to public safety. Notably, with the increasing penetration of renewable energy and enhanced grid interconnection, increased system uncertainty further amplifies the risk of cascading failures. Faced with this dual challenge, dynamic reconfiguration technology for distribution networks, capable of optimizing topology through real-time adjustment of switch states, has become a key means of improving grid resilience.

[0003] Distribution network reconfiguration is a complex high-dimensional combinatorial optimization problem. Its core lies in optimizing the network topology by adjusting the switch states to ensure the efficiency and reliability of power transmission. Traditional solutions mainly rely on heuristic algorithms and optimization algorithms, but these model-based solutions have two limitations: (1) With the increasing complexity of network structure and the diversification of controllable resources, the reconfiguration problem grows exponentially, making it difficult to meet the fast solution requirements of practical applications; (2) Existing studies mostly use assumptions and model simplifications, making it difficult to maximize the connectivity between power sources and loads under multi-source coordination, and different simplification strategies and mathematical processing methods have a significant impact on the solution results and efficiency. In contrast, neural network models can improve the solution speed by 20-100 times by explicitly modeling the complex physical constraints in the power system to solve the timing decision of network reconfiguration. In recent years, deep neural networks, convolutional neural networks, and graph neural networks have been widely used in power system operation and control links such as fault detection, optimal power flow, and state estimation. In particular, graph neural networks can effectively utilize the inherent network topology characteristics of the power system to ensure the integrity of the topology and the accuracy of calculation. This has achieved certain research results in distribution network fault reconfiguration.

[0004] However, graph neural networks tend to flatten the topology representation of power networks, often only capturing overall patterns and struggling to accurately identify and distinguish key structural differences. This is especially true in large-scale distribution networks with thousands of nodes, where relying solely on single features of nodes or edges for full-graph reasoning fails to accurately resolve differentiated topological information. Furthermore, when a system is isolated due to a fault, the same node may play different functional roles in different islands. The homogeneous node embeddings of traditional graph neural networks cannot characterize this dynamically changing semantic information, easily leading to connection prediction errors. In addition, in practical engineering applications, existing rapid recovery systems still largely employ centralized decision-making architectures, relying on global optimization and remote control of the master station system, resulting in inflexible algorithm deployment. Moreover, traditional systems are difficult to integrate effectively with neural network models, particularly lacking suitable hardware and software support for handling heterogeneous representations and real-time inference in graph neural networks.

[0005] To address the aforementioned issues, there is an urgent need for a rapid recovery method and system based on regional power distribution network fault reconstruction to resolve the problems associated with traditional methods. Summary of the Invention

[0006] The purpose of this invention is to provide a rapid recovery method and system based on regional power distribution network fault reconfiguration. By integrating physical constraints and key local features, it significantly reduces computational load and time costs while ensuring the differences in subgraphs corresponding to different links.

[0007] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0008] A rapid recovery method based on regional power distribution network fault reconfiguration includes:

[0009] Step 1: Collect real-time system operation data of the distribution network after a fault;

[0010] Step 2: Based on system operation data, construct a distribution network topology map, and extract the distribution network topology structure with link switches as the source-target nodes to obtain an island sub-map;

[0011] Step 3: Perform feature processing on the isolated subgraphs and construct a subgraph feature information matrix;

[0012] Step 4: Input the subgraph feature information matrix into the pre-trained subgraph neural network model to obtain the connectivity prediction results of each potential connection link;

[0013] Step 5: Generate dynamic reconfiguration control commands based on connectivity prediction results, and execute the control commands to restore the distribution network.

[0014] Furthermore, in step 2, the topology of the distribution network is extracted using link switches as source-target nodes to obtain an island subgraph, specifically:

[0015] The two endpoints of the link switch are used as the source node and the target node;

[0016] Determine the range of k-hop neighbors for the source and target nodes, obtain all neighbor nodes within this range, and collectively form an isolated subgraph.

[0017] Furthermore, the method for determining the number of hops k is as follows:

[0018] (1)

[0019] In the formula, As a DG regulation factor, R is the attenuation coefficient, and R is the mean line impedance. ρ For the topological density parameters of the island, Let be the set of nodes in the k-jump subgraph, and Ɛ be the fluctuation sensitivity parameter. This represents the weighting of the load proportion of isolated areas. This represents the node voltage threshold within the island. For load fluctuation rate, This represents the number of busbars within the isolated island.

[0020] Furthermore, in step 3, feature processing is performed on the isolated subgraphs to construct a subgraph feature information matrix, specifically as follows:

[0021] Calculate the feature distances of each neighbor node within the k-hop range relative to the source node and the target node;

[0022] Discrete labels are assigned to each neighbor node within a k-hop range based on feature distance calculation.

[0023] Based on the hierarchy determined by the discrete labels, the feature vectors of neighboring nodes are weighted and aggregated, and the aggregation result is concatenated with the features of the source or target node itself and the label embedding vector to form a subgraph feature information matrix.

[0024] Further, the feature distances of each neighbor node within the k-hop range relative to the source node and the target node are calculated, specifically as follows:

[0025] (2)

[0026] In the formula, For the feature distance, V, ρ These represent the node voltage and distributed power supply penetration rate, respectively. These are the weighting coefficients. The subscript u... , These are the source node, the target node, and any neighbor node in the target link, respectively. The reference voltage, This represents the maximum penetration rate of distributed power sources in isolated islands.

[0027] Furthermore, discrete labels are assigned to each neighbor node within a k-hop range based on feature distance, specifically as follows:

[0028] (3)

[0029] In the formula, Assign discrete labels to nodes. These are the system's rated parameters, used to fix the label layering granularity. This represents the baseline penetration rate of distributed power sources in isolated islands.

[0030] Furthermore, based on the hierarchy determined by the discrete labels, the feature vectors of neighboring nodes are weighted and aggregated. The aggregated result is then concatenated with the features of the source or target node itself and the label embedding vector to form a subgraph feature information matrix, specifically:

[0031] The feature vectors of the remaining neighboring nodes within the isolated subgraph are accumulated and added to the target node. The feature vectors of the neighboring nodes are then weighted and aggregated according to the hierarchy determined by the discrete labels, resulting in:

[0032] (4)

[0033] In the formula, These are the feature vectors aggregated from the target node u. Let ω be the feature vector of the neighboring node. , Neighboring nodes Active power, reactive power, and voltage. The maximum number of tag levels. For the weight function, , The attenuation coefficient is set to 0.5. Let k be the set of k-hop neighbors of the source node u;

[0034] The final subgraph feature information matrix is ​​as follows:

[0035] (5)

[0036] (6)

[0037] In the formula, For the label embedding vector, d e The embedding dimension has a value of 16 or 32. It is a one-hot encoding; The embedding matrix is ​​used to map discrete labels to trainable parameter matrices of low-dimensional continuous variables. Let be the feature vector of node u.

[0038] Further, in step 4, the subgraph feature information matrix is ​​input into the pre-trained subgraph neural network model to obtain the connectivity prediction results of each potential connection link, specifically:

[0039] Power encoding and node feature encoding are performed on the feature information matrix of the subgraph, and LayerNorm is used to suppress the noise effect caused by feature scale differences.

[0040] The feature information of the encoded subgraph feature information matrix is ​​updated based on the graph convolutional neural network model. The update process is performed with residual connections, and the residual connection structure is batch normalized.

[0041] Connectivity prediction is calculated on the updated results to obtain the connectivity prediction results.

[0042] This invention also provides a rapid recovery system based on regional power distribution network fault reconfiguration, used to implement the aforementioned rapid recovery method based on regional power distribution network fault reconfiguration, comprising:

[0043] The sensor acquisition module is used to collect system operation data of the distribution network in real time after a fault.

[0044] The communication module is used to transmit the system's operating data and control commands;

[0045] An edge computing module includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the program to implement the steps described in claims 1-8;

[0046] The control execution module is used to perform switching operations based on connectivity prediction results.

[0047] Furthermore, the sensing acquisition module includes at least one of a synchronous phasor measurement unit, a feeder terminal unit, a distribution terminal unit, and a smart meter deployed at the distribution network node;

[0048] The edge computing module is specifically an edge computing gateway deployed in a power distribution substation or key node, or a station-side server deployed in a power distribution automation master station or substation; the edge computing gateway or station-side server has a built-in artificial intelligence acceleration chip.

[0049] The control execution module includes a remotely controllable automatic sectionalizing switch, a tie switch, and a distributed power grid-connected switch.

[0050] In summary, the present invention has at least one of the following beneficial technical effects:

[0051] 1. Significantly improves reconstruction accuracy and reliability: By introducing an island subgraph extraction mechanism, this invention can accurately capture the heterogeneous topological features between islands in the distribution network after a fault, avoiding connection prediction errors caused by homogeneous node embedding in traditional graph neural networks. Combined with a feature aggregation method that incorporates physical rules, it significantly improves the accuracy of connectivity judgment and dynamic reconstruction.

[0052] 2. Significantly improves reconstruction decision speed: The complex global network reconstruction problem is decomposed into a local link prediction problem centered on link switches, greatly reducing the dimensionality of the solution. Combined with efficient feature engineering and a lightweight subgraph neural network model, it achieves rapid decision-making at the millisecond to second level, meeting the stringent real-time requirements of fault recovery.

[0053] 3. Effectively reduces computational resource consumption: Compared with traditional graph neural network methods that process the entire distribution network, the local subgraph extraction and computation strategy adopted in this invention significantly reduces the computational load and memory consumption during model training and inference, making the algorithm easy to deploy on edge computing devices or resource-constrained industrial controllers.

[0054] 4. Enhanced Model Generalization and Adaptability: The method proposed in this invention does not rely on specific network topologies or fixed fault scenarios. By dynamically determining the k-hop range and adaptive feature aggregation, the model can adaptively cope with distribution networks of different sizes and structures, and maintain stable performance under various complex operating conditions such as fault evolution, branch switching, and distributed power output fluctuations.

[0055] 5. Deep integration of physical constraints to ensure the feasibility of the scheme: In the process of feature extraction and aggregation, Kirchhoff's laws, voltage constraints and other core physical rules of power systems are innovatively incorporated to ensure that the model's learning and decision-making process conforms to the basic laws of power grid operation, thereby ensuring the physical feasibility and security of the generated reconfiguration scheme. Attached Figure Description

[0056] Figure 1 This is a schematic diagram of the algorithm flow of the present invention;

[0057] Figure 2 This is a schematic diagram of the IEEE 33-node system.

[0058] Figure 3 This is a schematic diagram of the system architecture of the present invention. Detailed Implementation

[0059] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0060] like Figure 1 and Figure 2 As shown, this invention provides a fast recovery method based on regional distribution network fault reconfiguration (SDGCN), comprising:

[0061] Step 1: Collect real-time system operation data of the distribution network after a fault;

[0062] Step 2: Based on system operation data, simplify the actual line distribution wiring diagram into a distribution network topology diagram, and extract the distribution network topology with link switches as the source-target nodes to obtain the island sub-diagram;

[0063] Step 3: Perform feature processing on the isolated subgraphs and construct a subgraph feature information matrix;

[0064] Step 4: Input the subgraph feature information matrix into the pre-trained subgraph neural network model to obtain the connectivity prediction results of each potential connection link;

[0065] Step 5: Generate dynamic reconfiguration control commands based on connectivity prediction results, and execute the control commands to restore the distribution network.

[0066] In step 2, the topology of the distribution network is extracted using link switches as the source-target nodes to obtain an island subgraph, specifically:

[0067] Obtain the undirected topology graph G=(V,E) of the distribution network in the area after the fault, where V represents the set of bus nodes and E is the set of power lines;

[0068] For a fault-affected distribution network, extract all neighboring nodes within a k-hop range (source and target nodes of the line link switch) as an isolated subgraph node. The method for determining k is as follows:

[0069] (1)

[0070] In the formula, As a DG regulation factor, R is the attenuation coefficient, and R is the mean line impedance. ρ For the topological density parameters of the island, Let be the set of nodes in the k-jump subgraph, and Ɛ be the fluctuation sensitivity parameter. This represents the weighting of the load proportion of isolated areas. This represents the node voltage threshold within the island. For load fluctuation rate, This represents the number of busbars within the isolated island.

[0071] In step 3, feature processing is performed on the isolated subgraphs to construct a subgraph feature information matrix, specifically as follows:

[0072] For the two endpoints (u, v) of the link switch, calculate the characteristic distances of each neighboring node within the k-hop range relative to the source node and the target node, specifically as follows:

[0073] (2)

[0074] In the formula, For the feature distance, V, ρ These represent the node voltage and distributed power supply penetration rate, respectively. These are the weighting coefficients. The subscript u... , These are the source node, the target node, and any neighbor node in the target link, respectively. The reference voltage, This represents the maximum penetration rate of distributed power sources in isolated islands.

[0075] Discrete labels are assigned to each neighbor node within a k-hop range based on feature distance calculation, specifically:

[0076] (3)

[0077] In the formula, Assign discrete labels to nodes. These are the system's rated parameters, used to fix the label layering granularity. This represents the baseline penetration rate of distributed power sources in isolated islands.

[0078] Based on the hierarchy determined by the discrete labels, the feature vectors of neighboring nodes are weighted and aggregated. The aggregation result is then concatenated with the features of the source or target node itself and the label embedding vector to form a subgraph feature information matrix, specifically:

[0079] The feature vectors of the remaining neighboring nodes within the isolated subgraph are accumulated and added to the target node. The feature vectors of the neighboring nodes are then weighted and aggregated according to the hierarchy determined by the discrete labels, resulting in:

[0080] (4)

[0081] In the formula, These are the feature vectors aggregated from the target node u. Let ω be the feature vector of the neighboring node. , Neighboring nodes Active power, reactive power, and voltage. The maximum number of tag levels. For the weight function, , The attenuation coefficient is set to 0.5. Let k be the set of k-hop neighbors of the source node u;

[0082] The final subgraph feature information matrix is ​​as follows:

[0083] (5)

[0084] (6)

[0085] In the formula, For the label embedding vector, d e The embedding dimension has a value of 16 or 32. It is a one-hot encoding; The embedding matrix is ​​used to map discrete labels to trainable parameter matrices of low-dimensional continuous variables. Let be the feature vector of node u.

[0086] In step 4, the subgraph feature information matrix is ​​input into the pre-trained subgraph neural network model to obtain the connectivity prediction results for each potential connection link, specifically:

[0087] The feature information matrix of the subgraph, i.e., the feature information matrix of the target node u, is subjected to power encoding and node feature encoding to enhance feature separability. LayerNorm is used to suppress the noise effect caused by feature scale differences. The calculation method is as follows:

[0088] (7)

[0089] (8)

[0090] (9)

[0091] In the formula, These are the active power and reactive power feature values ​​of the target node and its neighboring nodes, respectively; LayerNorm is used to normalize the power features, stabilize the training process, and prevent gradient explosion / vanishing caused by excessively large or small power. Let be the original attribute vector of the target node u, W be the weight matrix, and b be the bias term.

[0092] The graph convolutional neural network model is used to update the feature information of subgraphs using the update process of the graph convolutional network. The update mechanism of the graph convolutional network is as follows:

[0093] (10)

[0094] (11)

[0095] in, represents the node feature matrices of the l-th and l+1-th layers respectively, N is the number of link switch endpoints after subgraph extraction, and d is the dimension of the aggregated target node feature information; Let be the adjacency matrix of the graph. It is the identity matrix that ensures nodes can be aggregated with their own characteristics; To add self-loops, the adjacency matrix D is... The degree matrix, diagonal elements ; The learnable weight matrix of the l-th layer; For activation functions;

[0096] The update process involves residual joins, specifically as follows:

[0097] (12)

[0098] in, These are the initial input features; For the first l The residual projection matrix of the layer is used to linearly map the initial features to the current spatial layer;

[0099] The residual connection structure is subjected to batch normalization, specifically as follows:

[0100] (13)

[0101] (14)

[0102] (15)

[0103] In the formula, , where m is the number of samples in the mini-batch. Let be the feature vector of the i-th sample in the l-th layer; The mean, Indicates variance; It is a very small constant to prevent the denominator from being zero; These are scaling factor and offset factor, respectively. They are learnable parameters, typically initialized to 1 and 0, and are used to restore the network's expressive power.

[0104] The connectivity prediction is calculated based on the training results, and the predicted probability is:

[0105] (16)

[0106] (17)

[0107] In the formula, These are the weight matrices for the first and second layers, respectively. These are the first and second layer bias vectors, respectively. For the Sigmoid function, This is a real label.

[0108] Combination Figure 2 The performance of the algorithm of this invention is verified using the IEEE 33-node system shown. Figure 2 In the diagram, nodes 1 through 33 are node numbers. Node 1 connects to the external power grid, while nodes 7, 11, 12, 21, 25, 27, and 33 connect to distributed power sources. Dashed lines represent line tie switches, and diagonal lines indicate line faults. The baseline voltage for the entire experiment was 12.66 kV, and the total network load was 6.431 + j2.821 MVA. By introducing random faults into the network, 21 different fault topologies were generated. Each topology contained 300 sets of scenario data, with all samples divided into three groups: 10% for validation, 80% for training, and 10% for evaluation. During training, the number of cycles was set to 300, and the batch size was 512.

[0109] In the comparative case analysis, the proposed algorithm (SDGCN) was compared with the performance of a conventional graph convolutional network (GCN) without subgraph extraction, and the widely used "subgraph-embedding-prediction" algorithm (SEAL). The performance of each model in fault network processing and computation tasks was analyzed using multiple evaluation metrics, including F1 score, training loss, fault network information capture capability (ICC), and precision. Specific results are shown in Table 1.

[0110] Table 1 Performance evaluation of the three algorithms

[0111]

[0112] As can be seen from this table, the SDGCN algorithm proposed in this invention is superior to the other two conventional algorithms in terms of accuracy, ability to capture network information, and training loss.

[0113] To enable the method described in this invention to be applied in actual power distribution networks, such as... Figure 3 The invention also provides a rapid recovery system based on regional power distribution network fault reconstruction, which is used to implement the above-mentioned rapid recovery method based on regional power distribution network fault reconstruction. This architecture maps the algorithm steps to specific hardware modules, realizing full-process automation from data acquisition to control execution.

[0114] The hardware system comprises the following four layers:

[0115] 1. Sensing and Acquisition Layer

[0116] Hardware composition: This layer mainly includes intelligent sensing devices deployed at key nodes of the distribution network (such as busbars, switching stations, and distributed power grid connection points), such as synchronous phasor measurement units (PMU), feeder terminal units (FTU), distribution terminal units (DTU), smart meters, and remote terminal units (RTU) of SCADA systems.

[0117] Functional Correspondence (Step 1): These devices are responsible for collecting real-time system operation data after a fault, including node voltage V, current I, active power P, reactive power Q, real-time on / off status of switches, and real-time output of distributed generation (DG), used to calculate parameters such as permeability ρ. The collected analog signals are converted into standardized digital data through local analog-to-digital conversion.

[0118] 2. Communication Network Layer

[0119] Hardware composition: This layer consists of wired and wireless communication equipment, including industrial Ethernet switches, fiber optic ring network equipment, power line carrier (PLC) communication modules, and 5G / 4G wireless private network communication terminals.

[0120] Function: Provides a highly reliable, low-latency bidirectional data channel. Uplink, it is responsible for aggregating and transmitting massive amounts of operational data collected by the sensing layer to the computing layer; downlink, it is responsible for accurately and quickly distributing control commands generated by the computing layer to various execution devices. This layer meets power system communication standards, ensuring that the transmission latency of critical data is in the millisecond range.

[0121] 3. Edge / Station-level Computing Layer

[0122] Hardware components: This layer includes edge computing gateways deployed in substations or important power distribution nodes, and station-side monitoring servers deployed in power distribution automation master stations or substations.

[0123] Edge computing gateway: It adopts a high-performance embedded hardware platform, equipped with a multi-core processor, and has sufficient memory and solid-state storage (eMMC / SSD).

[0124] Functions correspond to algorithm steps:

[0125] Topology Construction and Subgraph Extraction (corresponding to step 2): The computing unit (edge ​​gateway or station server) receives global data transmitted through the communication network and dynamically constructs and maintains the undirected topology graph G=(V,E) of the distribution network in memory. According to formula (1), the subgraph extraction range k of each potential connection link to be evaluated (i.e., the two ends of the link switch) is dynamically calculated, and the k-hop neighbor search algorithm centered on the source node and the target node is run to extract the corresponding island subgraph data structure from the global graph.

[0126] Feature processing and matrix construction (corresponding to step 3): For each extracted isolated subgraph, the computing unit calls its mathematical computing library, calculates the feature distance of each neighboring node according to formula (2), assigns discrete labels to it according to formula (3), and finally completes feature weighting aggregation and splicing according to formulas (4)-(6) to generate a standard subgraph feature information matrix M. u

[0127] Neural network model inference (corresponding to step 4): The pre-trained Subgraph Neural Network (SDGCN) parameter file (.pt) is stored in the storage medium of the computing unit. When making fault recovery decisions, the model is loaded into memory. The GPU performs the forward inference computation of the model, which encapsulates a series of operations described in formulas (7) to (17), such as feature encoding, graph convolution, residual connection, normalization, and classifier computation, and finally outputs the connectivity prediction probability of each potential connection link. This layer of hardware supports online model updates.

[0128] 4. Control Execution Layer

[0129] Hardware composition: This layer consists of various remotely operable switching devices, mainly including automatic sectionalizing switches, tie switches, and grid-connected switches for distributed power sources. These switches have built-in electric operating mechanisms and intelligent controllers, and have "remote signaling, remote measurement, and remote control" functions.

[0130] Functional Correspondence (Step 5): Receive the dynamic reconfiguration control command set issued from the computing layer. The commands are transmitted to the controllers of each target switch via the communication network in the form of standard power protocols. After parsing the commands, the controller drives the operating mechanism to perform "closing" or "opening" operations, thereby changing the network topology and realizing the restoration of the distribution network. To increase safety, the switch controller can integrate simple local logic verification (e.g., voltage difference, phase angle checks) before performing critical operations.

[0131] The beneficial effects of hardware systems working together:

[0132] By closely integrating the aforementioned layered hardware architecture with the algorithm of this invention, the following engineering advantages are further achieved:

[0133] Distributed computing reduces central pressure: Deploying highly complex graph neural network inference tasks at the edge, close to the data source, significantly reduces the amount of data that needs to be uploaded to the main station, lowers the demand for computing resources and communication bandwidth on the main station, and improves the overall throughput and robustness of the system.

[0134] Hardware and software decoupling and efficient collaboration: The algorithm design takes into account the characteristics of embedded platforms (such as fixed-dimensional matrix operations), and is easy to optimize in parallel using processor vector instruction sets, giving full play to hardware computing power and achieving millisecond-level inference.

[0135] Easy integration and smooth evolution: This architecture is highly compatible with existing power distribution automation systems. It can be implemented by adding edge computing gateways to existing substations or ring main units and upgrading the main station software in a modular fashion, effectively protecting users' existing investments.

[0136] High reliability and autonomous capability: Edge computing nodes have certain local storage and computing capabilities. Even when communication with the master station is interrupted due to extreme situations, they can still make rapid autonomous recovery decisions on critical links by using locally cached models and recent data, which greatly enhances the survivability of the distribution network.

[0137] Embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0138] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0139] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0140] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0141] Contents not described in detail in this specification are prior art known to those skilled in the art. It is hereby indicated that the above description is intended to help those skilled in the art understand this invention, but does not limit the scope of protection of this invention. Any equivalent substitutions, modifications, improvements, or simplifications of the above descriptions that do not depart from the essential content of this invention fall within the scope of protection of this invention.

Claims

1. A rapid recovery method based on regional power distribution network fault reconfiguration, characterized in that, include: Step 1: Collect real-time system operation data of the distribution network after a fault; Step 2: Based on system operation data, construct a distribution network topology diagram, and extract the distribution network topology structure with link switches as source-target nodes to obtain an island subgraph; use the two endpoints of the link switches as source nodes and target nodes; Determine the k-hop range of the source node and the target node, obtain all neighboring nodes within this range, and together they form an isolated subgraph; Step 3: Perform feature processing on the isolated subgraphs and construct a subgraph feature information matrix; Specifically: Calculate the feature distances of each neighbor node within the k-hop range relative to the source node and the target node; Discrete labels are assigned to each neighbor node within a k-hop range based on feature distance calculation. Based on the hierarchy determined by the discrete labels, the feature vectors of neighboring nodes are weighted and aggregated, and the aggregation result is concatenated with the features of the source or target node itself and the label embedding vector to form a subgraph feature information matrix. Step 4: Input the subgraph feature information matrix into the pre-trained subgraph neural network model to obtain the connectivity prediction results for each potential connection link; specifically: Power encoding and node feature encoding are performed on the feature information matrix of the subgraph, and LayerNorm is used to suppress the noise effect caused by feature scale differences. The feature information of the encoded subgraph feature information matrix is ​​updated based on the graph convolutional neural network model. The update process is performed with residual connections, and the residual connection structure is batch normalized. The updated results are used to perform connectivity prediction calculations to obtain the connectivity prediction results. Step 5: Generate dynamic reconfiguration control commands based on connectivity prediction results, and execute the control commands to restore the distribution network.

2. The rapid recovery method based on regional power distribution network fault reconfiguration according to claim 1, characterized in that, The method for determining the number of jumps k is as follows: （1） In the formula, As a DG regulation factor, R is the attenuation coefficient, and R is the mean line impedance. ρ For the topological density parameters of the island, Let be the set of nodes in the k-jump subgraph, and Ɛ be the fluctuation sensitivity parameter. Weighting based on the proportion of isolated loads. This represents the node voltage threshold within the island. For load fluctuation rate, This represents the number of busbars within the isolated island.

3. The rapid recovery method based on regional power distribution network fault reconfiguration according to claim 2, characterized in that, Calculate the feature distances of each neighbor node within the k-hop range relative to the source node and the target node, specifically as follows: （2） In the formula, For the feature distance, V, ρ These represent the node voltage and distributed power supply penetration rate, respectively. For weighting coefficients, different subscripts u, , These are the source node, the target node, and any neighbor node in the target link, respectively. The reference voltage, This represents the maximum penetration rate of distributed power sources in isolated islands.

4. The rapid recovery method based on regional power distribution network fault reconfiguration according to claim 3, characterized in that, Discrete labels are assigned to each neighbor node within a k-hop range based on feature distance calculation, specifically: （3） In the formula, Assign discrete labels to nodes, u, , These are the source node, the target node, and any neighbor node in the target link, respectively. These are the system's rated parameters, used to fix the label layering granularity. This represents the baseline penetration rate of distributed power sources in isolated islands.

5. A rapid recovery method based on regional power distribution network fault reconfiguration according to claim 4, characterized in that, Based on the hierarchy determined by the discrete labels, the feature vectors of neighboring nodes are weighted and aggregated. The aggregation result is then concatenated with the features of the source or target node itself and the label embedding vector to form a subgraph feature information matrix, specifically: The feature vectors of the remaining neighboring nodes within the isolated subgraph are accumulated and added to the target node. The feature vectors of the neighboring nodes are then weighted and aggregated according to the hierarchy determined by the discrete labels, resulting in: （4） In the formula, These are the feature vectors aggregated from the source node u. Let ω be the feature vector of the neighboring node. , Neighboring nodes Active power, reactive power, voltage The maximum number of tag levels. For the weight function, , The attenuation coefficient is set to 0.

5. Let k be the set of k-hop neighbors of the source node u; The final subgraph feature information matrix is ​​as follows: （5） （6） In the formula, For the label embedding vector, d e The embedding dimension has a value of 16 or 32. It is a one-hot encoding; The embedding matrix is ​​used to map discrete labels to a trainable parameter matrix of low-dimensional continuous variables. Let be the feature vector of the source node u.

6. A rapid recovery system based on regional power distribution network fault reconfiguration, used to implement the rapid recovery method based on regional power distribution network fault reconfiguration as described in any one of claims 1-5, characterized in that, include: The sensor acquisition module is used to collect system operation data of the distribution network in real time after a fault. The communication module is used to transmit the system's operating data and control commands; An edge computing module includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the program to implement the method described in claims 1-5; The control execution module is used to perform switching operations based on connectivity prediction results.

7. A rapid recovery system based on regional power distribution network fault reconfiguration according to claim 6, characterized in that, The sensing and acquisition module includes at least one of a synchronous phasor measurement unit, a feeder terminal unit, a distribution terminal unit, and a smart meter deployed at a power distribution network node. The edge computing module is specifically an edge computing gateway deployed in a power distribution substation or key node, or a station-side server deployed in a power distribution automation master station or substation; the edge computing gateway or station-side server has a built-in artificial intelligence acceleration chip. The control execution module includes a remotely controllable automatic sectionalizing switch, a tie switch, and a distributed power grid-connected switch.

Images