Power distribution network topology detection method and system based on double-graph structure

Abstract

The invention provides a power distribution network topology detection method and system based on a double-graph structure, and the method comprises the steps: obtaining the measurement data of each node of a power distribution network every first preset time, and extracting a first feature and a second feature from the measurement data; calculating a comprehensive electrical distance index according to the first feature and the second feature, constructing a comprehensive distance matrix according to the electrical distance, and constructing a node attribute similarity graph according to the comprehensive distance matrix; inputting the node attribute similar graph into a pre-trained double-graph structure graph neural network model, generating a node embedding vector, and performing supervision training on the node embedding vector through a structure perception loss function by using an actual topological graph; and calculating connection scores of all the potential edges, judging whether the edges are connected or not according to the connection scores, and outputting a current topological structure of the power distribution network according to a judgment result. According to the method, the power distribution network topology inference can be realized without depending on the prior topology structure information.

Description

Technical Field

[0001] This invention relates to the field of topology detection technology, and in particular to a distribution network topology detection method and system based on a dual-graph structure. Background Technology

[0002] Accurate and real-time distribution network topology is fundamental for the efficient operation of advanced applications such as system state estimation, fault location, reactive power optimization, and protection coordination. In transmission networks, topology can usually be directly obtained through data acquisition and monitoring control systems. However, in distribution networks, due to their large scale, numerous switches, incomplete communication coverage, frequent network reconfigurations, and the integration of distributed energy resources, it is difficult for operators to grasp the actual interconnections of the grid in real time and accurately. The lack or lag in switch status information severely impacts the distribution network's state awareness, control capabilities, and operational efficiency.

[0003] To address this challenge, the industry has proposed various distribution network topology detection methods, which can be mainly divided into state estimation-based methods and data-driven methods. State estimation-based methods require pre-establishing a topology library containing all possible topologies and using real-time measurement data to perform state estimation for each topology in the library. Finally, the true topology is identified based on criteria such as minimizing residuals. The accuracy of these methods heavily depends on the precision of distribution network parameters, and their computational complexity increases dramatically with the size of the topology library, making real-time application difficult in large-scale distribution networks. Data-driven methods, especially deep learning-based methods, predict system topologies by training models using historical data, avoiding the need to build a complete topology library. However, traditional deep learning models (such as fully connected neural networks) can only utilize node attribute information (such as voltage and phase angle) and cannot effectively utilize the inherent graph structure information of the distribution network, resulting in limited generalization ability and detection accuracy.

[0004] Graph neural networks (GNNs), as a novel deep learning model capable of simultaneously processing node attributes and graph structure information, offer potential for solving this problem. However, existing GNN-based topology detection methods heavily rely on a known baseline graph structure (e.g., partially known line connections) as input during both the training and inference phases. This "pre-known topology" premise is often unsatisfactory in real-world distribution networks, as the objective is precisely to detect this unknown topology. Therefore, existing GNN models are difficult to apply to scenarios without any prior structural information, limiting their practical application in distribution network topology detection. Summary of the Invention

[0005] The purpose of this invention is to provide a distribution network topology detection method and system based on a dual-graph structure, aiming to solve the problem that traditional distribution network topology detection technology is limited in use due to its reliance on prior topology information.

[0006] In a first aspect, the present invention provides a distribution network topology detection method based on a dual-graph structure, the method comprising:

[0007] Measurement data of each node in the power distribution network is acquired at a first preset time interval, and a first feature and a second feature are extracted from the measurement data.

[0008] A comprehensive electrical distance index is calculated based on the first feature and the second feature, a comprehensive distance matrix is ​​constructed based on the electrical distance, and a node attribute similarity graph is constructed based on the comprehensive distance matrix;

[0009] The node attribute similarity graph is input into a pre-trained dual-graph structure graph neural network model to generate node embedding vectors. The node embedding vectors are then trained under supervision using a structure-aware loss function on an actual topological graph to align the node embeddings with the real topology and update the graph neural network parameters.

[0010] Calculate the connection score of all potential edges, determine whether each edge is connected based on the connection score, and output the current topology of the distribution network based on the determination result.

[0011] Secondly, the present invention provides a distribution network topology detection system based on a dual-graph structure, the system comprising:

[0012] The measurement data acquisition module is used to acquire measurement data of each node of the power distribution network at a first preset time interval, and extract a first feature and a second feature from the measurement data;

[0013] The node graph construction module is used to calculate the comprehensive electrical distance index based on the first feature and the second feature, construct a comprehensive distance matrix based on the electrical distance, and construct a node attribute similarity graph based on the comprehensive distance matrix;

[0014] The parameter update module is used to input the node attribute similarity graph into a pre-trained dual-graph structure graph neural network model, generate node embedding vectors, and use the actual topology graph to supervise the training of the node embedding vectors through a structure-aware loss function, so that the node embeddings are aligned with the real topology, thereby updating the graph neural network parameters.

[0015] The connectivity detection module is used to calculate the connectivity score of all potential edges, determine whether each edge is connected based on the connectivity score, and output the current topology of the distribution network based on the determination result.

[0016] Thirdly, the present invention provides a storage medium that stores one or more programs, which, when executed by a processor, implement the above-described distribution network topology detection method based on a dual-graph structure.

[0017] Fourthly, the present invention provides an electronic device, the electronic device comprising a memory and a processor, wherein:

[0018] The memory is used to store computer programs;

[0019] When the processor executes the computer program stored in the memory, it implements the above-described distribution network topology detection method based on a dual-graph structure.

[0020] Compared with the prior art, the present invention has the following advantages:

[0021] This invention extracts the Pearson coefficient and normalized Euclidean distance of node voltages through multi-feature fusion, constructs a comprehensive electrical distance index, and dynamically generates a node attribute similarity graph, effectively balancing the impact of long-term correlation and short-term fluctuations on topology judgment. Combined with a pre-trained dual-graph neural network model, it dynamically aggregates neighbor features using an attention mechanism, and through multi-task supervised training (including binary cross-entropy, ranking loss, and physical constraint loss), it ensures that node embeddings are highly aligned with the real topology, significantly improving the model's generalization ability in complex scenarios. Finally, it achieves topology inference by calculating connectivity scores in real time, without relying on prior topology information. Attached Figure Description

[0022] Figure 1 This is a flowchart of a power distribution network topology detection method based on a dual-graph structure proposed in an embodiment of the present invention;

[0023] Figure 2 This is a schematic diagram of a power distribution network topology detection system based on a dual-graph structure proposed in an embodiment of the present invention.

[0024] The following detailed description, in conjunction with the accompanying drawings, will further illustrate the present invention. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Unless otherwise defined, the technical or scientific terms used herein should have the ordinary meaning understood by those skilled in the art. The terms "comprising" and similar expressions used herein mean that the element or object preceding the word covers the element or object listed after the word and its equivalents, but does not exclude other elements or objects.

[0026] like Figure 1As shown, an embodiment of the present invention proposes a distribution network topology detection method based on a dual-graph structure. The method includes steps S101 to S104, wherein:

[0027] Step S101: Acquire measurement data of each node of the power distribution network at first preset time intervals, and extract the first feature and the second feature from the measurement data;

[0028] It should be noted that the measurement data includes voltage amplitude and voltage phase angle, and the measurement data of the distribution network node is defined as a time series matrix. Where T is the time window length, N is the number of nodes, and F is the feature dimension.

[0029] In some embodiments, the first feature refers to the Pearson coefficient, and the second feature refers to the normalized Euclidean distance. Specifically, the distance between the two nodes v is calculated according to the following formula. i and v j The Pearson coefficient within the time window T is used to characterize the trend similarity of voltage fluctuations between nodes:

[0030]

[0031] Calculate the normalized Euclidean distance between nodes using the following formula:

[0032]

[0033] Where, ρ ij For node v i and v j The Pearson coefficient between x i，t，f x j，t，f Let f be the measured value of the f-th feature of node i at time t, and let j be the measured value of the f-th feature of node j at time t. Let i be the average value of the data for node i within the current time window. Let j be the average value of the data within the current time window. For node v i and v j The normalized Euclidean distance between them, E ij For node v i and v j The original Euclidean distance between them, E max This is the maximum value in the original Euclidean distance.

[0034] Step S102: Calculate the comprehensive electrical distance index based on the first feature and the second feature, construct a comprehensive distance matrix based on the electrical distance, and construct a node attribute similarity graph based on the comprehensive distance matrix;

[0035] First, the comprehensive electrical distance is calculated using the following formula:

[0036]

[0037] Among them, D ij For node v i and v j The overall electrical distance between them, where α is an adjustment factor;

[0038] Then, a fully connected comprehensive distance matrix is ​​constructed based on the comprehensive electrical distance of all node pairs.

[0039] Specifically, the integrated distance matrix is ​​an N×N symmetric matrix, and D ij This corresponds to the element in the i-th row and j-th column of the matrix.

[0040] Furthermore, in this step, an improved degree-constrained Kruskal algorithm needs to be applied to process the composite distance matrix to generate a sparse graph that conforms to the physical characteristics of the distribution network. Specifically, this includes:

[0041] All potential edges in the comprehensive distance matrix are weighted by D. ij Sort by size from smallest to largest;

[0042] Initialize a graph containing N nodes and set the maximum allowed degree threshold for each node, for example, the value can be set to the upper limit of the branches of the distribution network feeder;

[0043] Iterative selection of sorted edges e ij An edge is added to the graph only if the following dual constraints are met:

[0044] Condition 1 (No Cyclic Constraints): Node v i and v j The circuit is not connected in the current graph (no loops are generated);

[0045] Condition 2 (Degree Constraint): After adding this edge, node v i and v j The degrees all do not exceed the maximum permissible degree threshold;

[0046] The iteration stops when the number of edges added to the graph reaches N-1 or no more edges can be added that meet the conditions. Finally, the graph structure generated by the above algorithm is used as a node attribute similarity graph, and its adjacency matrix A is output. NASG This graph integrates the amplitude and fluctuation trend characteristics of node voltages, which is used for message passing in subsequent graph neural networks.

[0047] Step S103: Input the node attribute similarity graph into the pre-trained dual-graph structure graph neural network model to generate node embedding vectors, and use the actual topology graph to supervise the training of the node embedding vectors through the structure-aware loss function so that the node embeddings are aligned with the real topology in order to update the graph neural network parameters;

[0048] It should be noted that in this step, the constructed Node Attribute Similarity Graph (NASG) is input into the model, and its edge weight information is used to improve the information transmission mechanism of the GraphSAGE layer.

[0049] Specifically, for any node v in the k-th layer i Instead of using the traditional, simple mean aggregation, it calculates the attention weights of neighboring nodes according to the following formula:

[0050]

[0051] Where, α ij For node v i and v j Attention weights between them, λ is a learnable distance decay coefficient, N NASG (v i ) represents the neighbor set, D iu For node v i and v u Comprehensive electrical distance between them;

[0052] Then, aggregated features for the current layer are generated based on the attention weights. And perform node embedding:

[0053]

[0054] in, Let W be the node embedding vector of neighbor node j in the previous layer (layer k) and the node embedding vector of node j in the current layer (layer k+1), respectively. ReLU is the activation function (Rectified Linear Unit). (k+1) Here, is the trainable weight matrix for layer k+1, and CONCAT is the concatenation operation, which combines the "features of its own" and the "features of its neighbors".

[0055] To capture the complex nonlinear interactions between nodes, a trainable bilinear interaction matrix M is introduced to replace the simple dot product operation when calculating the connectivity score. For any pair of potential connected nodes a and b, their final embedding vector is z. a and z b Connection probability score P(∈P) red The calculation formula is:

[0056]

[0057] Furthermore, in some embodiments, positive and negative sample pairs need to be constructed: during the pre-training phase, training samples are constructed using the Actual Topology Graph (ATG). This involves partitioning the positive sample set... and negative sample set In addition, triplet samples (e) also need to be constructed. + e - That is, selecting an existing edge e from the set of positive samples. + Select a non-existent edge e from the negative sample set. - This is used for subsequent training of sorting constraints.

[0058] Furthermore, in some embodiments, a composite loss function comprising classification loss, ranking regularization term, and physical consistency constraint term needs to be constructed:

[0059] L total =L BCE +γ1·L Rank +γ2·L Phy

[0060]

[0061] Among them, L total For the composite loss function, γ1 and γ2 are both balance coefficients, L BCE For binary cross-entropy classification loss, L Rank For the interval-based ranking loss, L is used to ensure that the predicted score of a positive edge is at least one preset interval higher than the predicted score of a negative edge. Phy The voltage phase angle physical constraint loss is intended to penalize edges that are predicted to be high-probability connections but have excessively different physical states, P(e ij (e) represents the edge e between node i and node j. ij The predicted connection probability, θ i θ j They are nodes v i and v j voltage phase angle, θ thres y is the maximum permissible phase angle difference threshold for normal operation of the distribution network. e Let ξ be the true label of the edge, 0 or 1, P(e) be the predicted connection probability of any edge e, and ξ be the interval threshold. + P(e) represents the predicted score of the positive sample edge. - ) represents the predicted score of the negative sample edge.

[0062] Finally, by minimizing L... total Update network parameters.

[0063] Step S104: Calculate the connection score of all potential edges, determine whether each edge is connected based on the connection score, and output the current topology of the distribution network based on the determination result.

[0064] It should be noted that in this step, the set of all physically possible power lines in the distribution network is first defined as the potential edge set. For each edge e in this set ab Extract the final layer node embedding z from its two endpoints a and b. a and z b The connection score s is calculated by dot product operation and sigmoid activation function. ab This score represents the probability that the edge is either closed or connected.

[0065]

[0066] Where T is the transpose.

[0067] Next, depending on the operating mode or testing requirements of the distribution network, choose one of the following two methods to determine connectivity:

[0068] Method 1: When the goal is to detect the complete topology of the distribution network and the network is preset to be radial, execute the Top-k selection strategy. Specifically, this includes: selecting the set... The connection score of all potential edges in s ab Sort the edges in descending order from highest to lowest score; select the edges corresponding to the top k scores after sorting as connected edges, and the rest as disconnected edges; where k takes the value of N-1 (N is the total number of network nodes) to ensure that the generated topology satisfies the tree structure constraint of the radial network.

[0069] Method 2: When the goal is to detect specific switch states, missing edge states, or meshed topologies containing loops, a threshold determination strategy is implemented. This includes: setting a connectivity probability threshold δ (e.g., 0.5); and setting each potential edge e... ab Connection score s ab Compare each value with the threshold δ; if s ab If s > δ, then the edge is determined to be connected / closed; if s ab If the value is less than δ, then the edge is determined to be in an open / disconnected state; this method allows for the existence of loops or non-N-1 edges in the generated topology.

[0070] Furthermore, in some embodiments, after acquiring real-time measurement data from multiple nodes in the distribution network, the method further includes denoising preprocessing of the measurement data. Specifically, this involves employing Gaussian kernel smoothing technology, using a Gaussian kernel function to smooth the voltage amplitude and voltage phase angle data of the time series, in order to eliminate the influence of measurement noise on topology detection. The Gaussian kernel function is expressed as follows:

[0071]

[0072] Where, x * For the current evaluation point in time, x i Here, b represents the historical sampling data points, b is the bandwidth parameter controlling the width of the smoothing window, and Γ is the normalization constant used to ensure that the sum of the kernel function integrals is 1.

[0073] In addition, in some embodiments, it is also necessary to extract independent measurement data of the A-phase, B-phase and C-phase nodes in the distribution network separately;

[0074] Based on the independent measurement data of each phase, a phase-specific node attribute similarity graph is constructed for each phase.

[0075] Three independent dual-graph structured graph neural network models are used to perform parallel training or inference on phase-specific node attribute similarity graphs and phase-specific actual topology graphs corresponding to phases A, B, and C, respectively.

[0076] The outputs of the three phase-specific models are combined, and the overall three-phase topology of the distribution network is output by merging the edge connectivity determination results of each phase.

[0077] In summary, this invention extracts the Pearson coefficient and normalized Euclidean distance of node voltages through multi-feature fusion, constructs a comprehensive electrical distance index, and dynamically generates a node attribute similarity graph, effectively balancing the impact of long-term correlation and short-term fluctuations on topology judgment. Combined with a pre-trained dual-graph neural network model, it dynamically aggregates neighbor features using an attention mechanism, and through multi-task supervised training (including binary cross-entropy, ranking loss, and physical constraint loss), it ensures that node embeddings are highly aligned with the real topology, significantly improving the model's generalization ability in complex scenarios. Finally, it achieves topology inference by calculating connectivity scores in real time, without relying on prior topology information.

[0078] like Figure 2 As shown, one embodiment of the present invention proposes a distribution network topology detection system based on a dual-graph structure, the system comprising:

[0079] The measurement data acquisition module 10 is used to acquire measurement data of each node of the power distribution network at a first preset time interval, and extract a first feature and a second feature from the measurement data;

[0080] The node graph construction module 20 is used to calculate the comprehensive electrical distance index based on the first feature and the second feature, construct a comprehensive distance matrix based on the electrical distance, and construct a node attribute similarity graph based on the comprehensive distance matrix;

[0081] The parameter update module 30 is used to input the node attribute similarity graph into a pre-trained dual-graph structure graph neural network model, generate node embedding vectors, and use the actual topology graph to supervise the training of the node embedding vectors through a structure-aware loss function, so that the node embeddings are aligned with the real topology, thereby updating the graph neural network parameters.

[0082] The connectivity detection module 40 is used to calculate the connectivity score of all potential edges, determine whether each edge is connected based on the connectivity score, and output the current topology of the distribution network based on the determination result.

[0083] In another aspect, the present invention also proposes a storage medium on which one or more programs are stored, which, when executed by a processor, implement the above-described distribution network topology detection method based on a dual-graph structure.

[0084] In another aspect, the present invention also proposes an electronic device, including a memory and a processor, wherein the memory is used to store a computer program, and the processor is used to execute the computer program stored in the memory to realize the above-mentioned distribution network topology detection method based on a dual-graph structure.

[0085] Those skilled in the art will understand that the logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can mean any means that can contain stored, communicated, propagated, or transmitted programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0086] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0087] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0088] While embodiments of the present invention have been described in detail above, it will be apparent to those skilled in the art that various modifications and variations can be made to these embodiments. However, it should be understood that such modifications and variations fall within the scope and spirit of the invention as set forth in the claims. Furthermore, the invention described herein may have other embodiments and can be implemented or carried out in various ways.

Claims

1. A distribution network topology detection method based on a dual-graph structure, characterized in that, The method includes: Measurement data of each node in the power distribution network is acquired at a first preset time interval, and a first feature and a second feature are extracted from the measurement data. A comprehensive electrical distance index is calculated based on the first feature and the second feature, a comprehensive distance matrix is ​​constructed based on the electrical distance, and a node attribute similarity graph is constructed based on the comprehensive distance matrix; The node attribute similarity graph is input into a pre-trained dual-graph structure graph neural network model to generate node embedding vectors. The node embedding vectors are then trained under supervision using a structure-aware loss function on an actual topological graph to align the node embeddings with the real topology and update the graph neural network parameters. Calculate the connection score of all potential edges, determine whether each edge is connected based on the connection score, and output the current topology of the distribution network based on the determination result.

2. The distribution network topology detection method based on a dual-graph structure according to claim 1, characterized in that, The step of acquiring measurement data of each node in the distribution network at first preset time intervals and extracting the first feature and the second feature from the measurement data includes: The measurement data includes voltage amplitude and voltage phase angle, and the measurement data of the distribution network nodes is defined as a time series matrix. ,in, Where N is the length of the time window, N is the number of nodes, and F is the feature dimension; Calculate the two nodes using the following formula. and In the time window Pearson coefficient: ； Calculate the normalized Euclidean distance between nodes using the following formula: ； in, For nodes and The Pearson coefficient between them , Let f be the measured value of the f-th feature of node i at time t, and let j be the measured value of the f-th feature of node j at time t. Let i be the average value of the data for node i within the current time window. Let j be the average value of the data within the current time window. For nodes and The normalized Euclidean distance between them For nodes and The original Euclidean distance between them This is the maximum value in the original Euclidean distance.

3. The distribution network topology detection method based on a dual-graph structure according to claim 2, characterized in that, The steps of calculating the comprehensive electrical distance index based on the first feature and the second feature, and constructing the comprehensive distance matrix based on the electrical distance, include: The comprehensive electrical distance is calculated using the following formula: ； in, For nodes and The combined electrical distance between them As a regulating factor; Construct a fully connected integrated distance matrix based on the integrated electrical distance of all node pairs.

4. The distribution network topology detection method based on a dual-graph structure according to claim 3, characterized in that, The step of constructing a node attribute similarity graph based on the comprehensive distance matrix includes: Sort all potential edges in the composite distance matrix in ascending order of composite electrical distance, and initialize a graph containing N nodes, setting the maximum allowable degree threshold for each node; Iteratively select the sorted edges, when the node and The graph is currently disconnected, and after adding this edge, the node... and If the degree of any edge does not exceed the maximum allowed degree threshold, then the edge is added to the graph. When the number of edges added to the graph reaches When an edge cannot be added to meet the condition, stop iterating and output a similarity graph of node attributes.

5. The distribution network topology detection method based on a dual-graph structure according to claim 4, characterized in that, The step of inputting the node attribute similarity graph into a pre-trained dual-graph structure graph neural network model to generate node embedding vectors includes: The attention weights of neighboring nodes are calculated using the following formula: ； in, For nodes and Attention weights between them A learnable distance attenuation coefficient. Gather for the neighbors, For nodes and Comprehensive electrical distance between them; Generate aggregated features for the current layer based on the attention weights. And embed the nodes: ； ； in, , Let be the node embedding vector of neighbor node j in the previous layer and the node embedding vector of node j in the (k+1)th layer, respectively. This is the activation function (Rectified Linear Unit). Let be the trainable weight matrix of the (k+1)th layer. This is for splicing operations.

6. The distribution network topology detection method based on a dual-graph structure according to claim 5, characterized in that, The step of using the actual topology graph to supervise the training of the node embedding vectors through a structure-aware loss function, so as to align the node embeddings with the real topology and update the parameters of the graph neural network, includes: Construct the loss function according to the following formula: ； in, For composite loss function, , All are balance coefficients. For binary cross-entropy classification loss, For the interval-based ranking loss, The voltage phase angle physical constraint loss, The edge between node i and node j The predicted connection probability value, , They are nodes and voltage phase angle, The maximum permissible phase angle difference threshold for normal operation of the distribution network. For the true labels of the edges, Let e ​​be the predicted connection probability of any edge e. This is the interval threshold. The predicted score for the positive sample edge. The predicted score for the negative sample edge.

7. The distribution network topology detection method based on a dual-graph structure according to claim 6, characterized in that, The steps of calculating the connection scores of all potential edges, determining whether each edge is connected based on the connection scores, and outputting the current topology of the distribution network based on the determination results include: Set a connectivity probability threshold , each potential edge Connection fraction With threshold Compare them one by one; like If so, then the edge is determined to be connected / closed; if If so, the edge is determined to be in an open / disconnected state.

8. A distribution network topology detection system based on a dual-graph structure, characterized in that, The system includes: The measurement data acquisition module is used to acquire measurement data of each node of the power distribution network at a first preset time interval, and extract a first feature and a second feature from the measurement data; The node graph construction module is used to calculate the comprehensive electrical distance index based on the first feature and the second feature, construct a comprehensive distance matrix based on the electrical distance, and construct a node attribute similarity graph based on the comprehensive distance matrix; The parameter update module is used to input the node attribute similarity graph into a pre-trained dual-graph structure graph neural network model, generate node embedding vectors, and use the actual topology graph to supervise the training of the node embedding vectors through a structure-aware loss function, so that the node embeddings are aligned with the real topology, thereby updating the graph neural network parameters. The connectivity detection module is used to calculate the connectivity score of all potential edges, determine whether each edge is connected based on the connectivity score, and output the current topology of the distribution network based on the determination result.

9. A storage medium, characterized in that, The storage medium stores one or more programs that, when executed by a processor, implement the distribution network topology detection method based on a dual-graph structure as described in any one of claims 1-7.

10. An electronic device comprising a memory and a processor, wherein: The memory is used to store computer programs; When the processor executes the computer program stored in the memory, it implements the power distribution network topology detection method based on a dual-graph structure as described in any one of claims 1-7.

Images