Power distribution network fault line positioning method and device

Abstract

The invention discloses a power distribution network fault line positioning method and device, and relates to the field of power distribution network fault identification, and the method comprises the steps: constructing a power distribution network topology according to the historical power supply data of a to-be-positioned fault power distribution network; optimizing a preset model parameter of a graph neural network model according to the node characteristics of each node of the power distribution network topology, performing iterative propagation calculation on the power distribution network topology based on the graph neural network model, and optimizing the model parameter of the graph neural network model based on a gradient descent method during iteration, training is completed until the model parameters accord with a preset threshold value, and a power grid fault positioning model is obtained; inputting real-time power supply data of a fault power distribution network to be positioned into the power grid fault positioning model to obtain fault positioning data; and inputting the fault positioning data into a preset text labeling model to obtain a fault label of the fault positioning data, and labeling real-time power supply data based on the fault label to obtain a fault positioning result.

Description

Technical Field

[0001] The present application relates to the field of distribution network fault identification, and in particular to a method and device for locating a distribution network fault line. Background Art

[0002] As the terminal structure of the power system directly facing users, the distribution network has a variety of power supply types, frequent load switching, and increasingly complex topology. The difficulty of locating and handling distribution network faults is gradually increasing. The previous manual judgment and decision-making method based on experience is no longer applicable to complex distribution networks. Therefore, a more efficient and accurate method for locating distribution network fault lines is needed.

[0003] Some methods for locating fault lines in intelligent distribution networks have been proposed and applied in the prior art, such as Petri nets, artificial neural networks, genetic algorithms, rough set decision-making, expert systems, and data mining. However, these methods often ignore the graph structure characteristics that exist due to the electrical connection relationship in the power grid structure. Therefore, when locating the fault line of the distribution network through the connection relationship between nodes, there is a certain degree of distortion, and the accuracy of locating the fault line of the distribution network is low. Summary of the invention

[0004] The present application provides a method and device for locating a fault line in a distribution network, so as to solve the technical problem that the method for locating a fault line in a distribution network in the prior art has distortion and leads to low positioning accuracy.

[0005] According to a first aspect of an embodiment of the present application, a method for locating a fault line in a distribution network is provided, comprising:

[0006] Construct the distribution network topology based on the historical power supply data of the fault distribution network to be located;

[0007] Optimizing the model parameters of a preset graph neural network model according to the node characteristics of each node of the distribution network topology, and performing iterative propagation calculation on the distribution network topology based on the graph neural network model. During iteration, optimizing the model parameters of the graph neural network model based on the gradient descent method until the model parameters meet the preset thresholds, and completing the training to obtain a power grid fault location model;

[0008] Inputting the real-time power supply data of the fault distribution network to be located into the power grid fault location model to obtain fault location data;

[0009] The fault location data is input into a preset text annotation model to obtain a fault label of the fault location data, and the real-time power supply data of the fault distribution network to be located is annotated based on the fault label to obtain a fault location result.

[0010] The present application first constructs a distribution network topology based on the historical power supply data of the faulty distribution network to be located, and then optimizes the model parameters of a preset graph neural network model based on the node characteristics of each node in the distribution network topology, which can integrate the node characteristics and graph structure characteristics of the distribution network topology into the graph neural network model to reduce model distortion, and then improve the accuracy of the power grid fault location model when the distribution network topology is iterated based on the graph neural network model to obtain a power grid fault location model; then, fault location data is obtained based on real-time power supply data, and the fault location data is input into a preset text annotation model to obtain a fault label, and then the real-time power supply data is annotated according to the fault label to obtain a fault location result, which can intuitively represent the fault location result and prevent errors and distortions caused by traditional manual judgment and positioning, thereby improving positioning accuracy.

[0011] In certain embodiments of the present application, the model parameters of the preset graph neural network model are optimized according to the node characteristics of each node of the distribution network topology, and the distribution network topology is iteratively propagated and calculated based on the graph neural network model. During iteration, the model parameters of the graph neural network model are optimized based on the gradient descent method until the model parameters meet the preset threshold value. The training is completed to obtain the power grid fault location model, which specifically includes:

[0012] The graph neural network model includes a self-attention mechanism;

[0013] Selecting nodes of the distribution network topology as current nodes in sequence, and obtaining a first node feature of the current node based on the self-attention mechanism, until all nodes of the distribution network topology are selected, thereby obtaining a plurality of first node features of the distribution network topology;

[0014] Optimizing the model parameters of the graph neural network model according to the multiple first node features to obtain a first graph neural network model;

[0015] Iterative propagation calculation is performed on the distribution network topology according to the first graph neural network model. During the iteration, the model parameters of the graph neural network model are optimized based on the gradient descent method until the training is completed when the model parameters meet the preset threshold, so as to obtain a power grid fault location model.

[0016] The present application first calculates the first node characteristics of each node in the distribution network topology, and then optimizes the model parameters of the graph neural network model according to the first node characteristics to obtain the first graph neural network model, which can integrate the node characteristics and graph structure characteristics of the distribution network topology into the graph neural network model to reduce model distortion, and then iterates the distribution network topology based on the first graph neural network model, optimizes the model parameters based on the gradient descent method to obtain a power grid fault location model, and can quickly converge the model parameters to a preset threshold, meet the actual task requirements, improve the accuracy of the power grid fault location model, and thus improve the accuracy of the power grid fault line location.

[0017] In certain embodiments of the present application, obtaining the first node feature of the current node based on the self-attention mechanism specifically includes:

[0018] Based on the self-attention mechanism, calculate the attention coefficients of all neighboring nodes of the current node to the current node;

[0019] According to the attention coefficient, node features of all adjacent nodes of the current node are aggregated to obtain a first node feature of the current node.

[0020] This application first calculates the attention coefficient of all adjacent nodes of the current node for the current node based on the self-attention mechanism, and then aggregates the node features of all adjacent nodes of the current node according to the attention coefficient to obtain the first node feature of the current node. This enables the current node to include the node features of all adjacent nodes, strengthens the connection relationship between nodes, and is conducive to the subsequent optimization of the graph neural network model. Integrate the node features and graph structure features of the distribution network topology, thereby improving the matching degree between the model and the actual distribution network topology and improving the model accuracy.

[0021] In certain embodiments of the present application, the distribution network topology is iteratively propagated and calculated according to the first graph neural network model, and the model parameters of the graph neural network model are optimized based on the gradient descent method during iteration until the model parameters meet the preset threshold value. The training is completed to obtain a power grid fault location model, which specifically includes:

[0022] Iterative propagation calculation is performed on the distribution network topology according to the first graph neural network model. During the iteration, the model parameters of the graph neural network model are optimized based on the gradient descent method according to the gradient information of the first graph neural network model. The training is completed until the model parameters meet the preset threshold, so as to obtain a power grid fault location model.

[0023] This application first iterates the distribution network topology according to the first graph neural network model. During the iteration, the model parameters are optimized based on the gradient descent method according to the gradient information to obtain the power grid fault location model. The model parameters can be quickly converged to the preset threshold, which meets the actual task requirements and is conducive to improving the accuracy of the power grid fault location model.

[0024] In certain embodiments of the present application, the preset text annotation model specifically includes:

[0025] Based on preset division parameters, the preset standard fault location data and the corresponding standard fault label data are divided into fault training data, fault verification data and fault test data;

[0026] Iteratively training a preset initial text annotation model according to the fault training data and the fault verification data, training and updating the learning parameters of the initial text annotation model according to the fault training data in each round of iteration, and verifying and updating the hyperparameters of the initial text annotation model according to the fault verification data, until the initial text annotation model satisfies a preset training end condition, and the training is completed to obtain a preset text annotation model;

[0027] The initial text annotation model meets the preset training end conditions, specifically:

[0028] According to the fault test data, the accuracy and loss value of the initial text annotation model are calculated; wherein, during the calculation, the learning parameters and hyperparameters obtained in the current round are used as model parameters of the initial text annotation model;

[0029] If the accuracy is greater than the preset accuracy threshold and the loss value is less than the preset loss threshold, it is determined that the initial text annotation model of the current round meets the preset training end condition.

[0030] The present application first obtains fault training data, fault verification data and fault test data based on preset division parameters, and trains, verifies and updates learning parameters and hyperparameters based on the fault training data and fault verification data, and judges the training end conditions based on the fault test data, which can enable the learning parameters and hyperparameters of the model to quickly converge to preset thresholds, meet actual task requirements, and improve the accuracy of the preset text annotation model.

[0031] According to a second aspect of the implementation of the present application, there is provided a distribution network fault line location device, including a topology construction module, a model training module, a fault location module and a fault marking module;

[0032] The topology building module is used to build a distribution network topology based on historical power supply data of the fault distribution network to be located;

[0033] The model training module is used to optimize the model parameters of a preset graph neural network model according to the node characteristics of each node of the distribution network topology, and perform iterative propagation calculations on the distribution network topology based on the graph neural network model. During iteration, the model parameters of the graph neural network model are optimized based on the gradient descent method until the training is completed when the model parameters meet the preset threshold value, thereby obtaining a power grid fault location model;

[0034] The fault location module is used to input the real-time power supply data of the fault distribution network to be located into the power grid fault location model to obtain fault location data;

[0035] The fault labeling module is used to input the fault location data into a preset text labeling model to obtain a fault label of the fault location data, and to label the real-time power supply data of the fault distribution network to be located based on the fault label to obtain a fault location result.

[0036] In certain embodiments of the present application, the graph neural network model includes a self-attention mechanism, and the model training module includes a node feature acquisition submodule, a model parameter optimization submodule, and a model iterative training submodule;

[0037] The node feature acquisition submodule is used to sequentially select nodes of the distribution network topology as current nodes, and obtain a first node feature of the current node based on the self-attention mechanism, until all nodes of the distribution network topology are selected, thereby obtaining multiple first node features of the distribution network topology;

[0038] The model parameter optimization submodule is used to optimize the model parameters of the graph neural network model according to the multiple first node features to obtain a first graph neural network model;

[0039] The model iteration training submodule is used to perform iterative propagation calculations on the distribution network topology according to the first graph neural network model. During iteration, the model parameters of the graph neural network model are optimized based on the gradient descent method until the training is completed when the model parameters meet the preset threshold, thereby obtaining a power grid fault location model.

[0040] In certain embodiments of the present application, the node feature acquisition submodule includes an attention coefficient calculation unit and a node feature aggregation unit;

[0041] The attention coefficient calculation unit is used to calculate the attention coefficients of all neighboring nodes of the current node for the current node based on the self-attention mechanism;

[0042] The node feature aggregation unit is used to aggregate the node features of all adjacent nodes of the current node according to the attention coefficient to obtain the first node feature of the current node.

[0043] In certain embodiments of the present application, the model iteration training submodule includes a model iteration training unit;

[0044] The model iteration training unit is used to perform iterative propagation calculations on the distribution network topology according to the first graph neural network model. During iteration, the model parameters of the graph neural network model are optimized based on the gradient descent method according to the gradient information of the first graph neural network model. The training is completed when the model parameters meet the preset threshold, thereby obtaining a power grid fault location model.

[0045] In certain embodiments of the present application, the preset text annotation model specifically includes:

[0046] Based on preset division parameters, the preset standard fault location data and the corresponding standard fault label data are divided into fault training data, fault verification data and fault test data;

[0047] Iteratively training a preset initial text annotation model according to the fault training data and the fault verification data, training and updating the learning parameters of the initial text annotation model according to the fault training data in each round of iteration, and verifying and updating the hyperparameters of the initial text annotation model according to the fault verification data, until the initial text annotation model satisfies a preset training end condition, and the training is completed to obtain a preset text annotation model;

[0048] The initial text annotation model meets the preset training end conditions, specifically:

[0049] According to the fault test data, the accuracy and loss value of the initial text annotation model are calculated; wherein, during the calculation, the learning parameters and hyperparameters obtained in the current round are used as model parameters of the initial text annotation model;

[0050] If the accuracy is greater than the preset accuracy threshold and the loss value is less than the preset loss threshold, it is determined that the initial text annotation model of the current round meets the preset training end condition.

[0051] The present application first constructs a distribution network topology based on the historical power supply data of the faulty distribution network to be located, and then optimizes the model parameters of a preset graph neural network model based on the node characteristics of each node in the distribution network topology, which can integrate the node characteristics and graph structure characteristics of the distribution network topology into the graph neural network model to reduce model distortion, and then improve the accuracy of the power grid fault location model when the distribution network topology is iterated based on the graph neural network model to obtain a power grid fault location model; then, fault location data is obtained based on real-time power supply data, and the fault location data is input into a preset text annotation model to obtain a fault label, and then the real-time power supply data is annotated according to the fault label to obtain a fault location result, which can intuitively represent the fault location result and prevent errors and distortions caused by traditional manual judgment and positioning, thereby improving positioning accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 : A flow chart of a method for locating a fault line in a distribution network shown in certain embodiments of the present application;

[0053] Figure 2 : A module structure diagram of a distribution network fault line locating device shown in certain embodiments of the present application. DETAILED DESCRIPTION

[0054] The embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals represent the same or similar elements or elements having the same or similar functions from beginning to end. The embodiments described below in conjunction with the accompanying drawings are exemplary and are only used to explain some embodiments of the present application, and should not be construed as limitations on the embodiments of the present application. Based on the embodiments shown in the present application, all other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of the present application.

[0055] In the description of this application, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include at least one of the features. In the description of this application, unless otherwise clearly and specifically defined, "multiple" and "several" mean two or more.

[0056] Existing methods for locating faulty lines in distribution networks include methods such as Petri nets, artificial neural networks, genetic algorithms, rough set decision-making, expert systems, and data mining. However, these methods ignore the graph structure characteristics that exist due to the electrical connection relationship in the power grid structure. Therefore, when locating faulty lines in distribution networks through the connection relationship between nodes, there is a certain degree of distortion, and thus the accuracy of locating faulty lines in distribution networks is low.

[0057] Based on the above technical background, please refer to Figure 1 The embodiment of the present application provides a method for locating a fault line in a distribution network, including steps S101 to S104, each of which is specifically as follows:

[0058] Step S101: construct a distribution network topology according to historical power supply data of a faulty distribution network to be located.

[0059] Specifically, the historical power supply data includes the voltage level, maximum allowable load, power generation, current, power supply frequency, historical load and installation status of power supply protection devices of each node of the fault distribution network to be located.

[0060] Step S102: Optimize the model parameters of a preset graph neural network model according to the node characteristics of each node in the distribution network topology, and perform iterative propagation calculations on the distribution network topology based on the graph neural network model. During iteration, optimize the model parameters of the graph neural network model based on the gradient descent method until the training is completed when the model parameters meet the preset threshold, thereby obtaining a power grid fault location model.

[0061] In certain embodiments of the present application, the output of the power grid fault location model is the status of each node of the distribution network topology.

[0062] In certain embodiments of the present application, the model parameters of the preset graph neural network model are optimized according to the node characteristics of each node of the distribution network topology, and the distribution network topology is iteratively propagated and calculated based on the graph neural network model. During iteration, the model parameters of the graph neural network model are optimized based on the gradient descent method until the model parameters meet the preset threshold value. The training is completed to obtain the power grid fault location model, which specifically includes:

[0063] The graph neural network model includes a self-attention mechanism;

[0064] Selecting nodes of the distribution network topology as current nodes in sequence, and obtaining a first node feature of the current node based on the self-attention mechanism, until all nodes of the distribution network topology are selected, thereby obtaining a plurality of first node features of the distribution network topology;

[0065] Optimizing the model parameters of the graph neural network model according to the multiple first node features to obtain a first graph neural network model;

[0066] Iterative propagation calculation is performed on the distribution network topology according to the first graph neural network model. During the iteration, the model parameters of the graph neural network model are optimized based on the gradient descent method until the training is completed when the model parameters meet the preset threshold, so as to obtain a power grid fault location model.

[0067] The present application first calculates the first node characteristics of each node in the distribution network topology, and then optimizes the model parameters of the graph neural network model according to the first node characteristics to obtain the first graph neural network model, which can integrate the node characteristics and graph structure characteristics of the distribution network topology into the graph neural network model to reduce model distortion, and then iterates the distribution network topology based on the first graph neural network model, optimizes the model parameters based on the gradient descent method to obtain a power grid fault location model, and can quickly converge the model parameters to a preset threshold, meet the actual task requirements, improve the accuracy of the power grid fault location model, and thus improve the accuracy of the power grid fault line location.

[0068] In certain embodiments of the present application, obtaining the first node feature of the current node based on the self-attention mechanism specifically includes:

[0069] Based on the self-attention mechanism, calculate the attention coefficients of all neighboring nodes of the current node to the current node;

[0070] According to the attention coefficient, node features of all adjacent nodes of the current node are aggregated to obtain a first node feature of the current node.

[0071] In some embodiments of the present application, the node features of all neighboring nodes of the current node are aggregated according to the attention coefficient to obtain the first node feature of the current node, specifically:

[0072] The attention coefficient is used as a weight to perform weighted summation of node features of all adjacent nodes of the current node to obtain a first node feature of the current node.

[0073] This application first calculates the attention coefficient of all adjacent nodes of the current node for the current node based on the self-attention mechanism, and then aggregates the node features of all adjacent nodes of the current node according to the attention coefficient to obtain the first node feature of the current node. This enables the current node to include the node features of all adjacent nodes, strengthens the connection relationship between nodes, and is conducive to the subsequent optimization of the graph neural network model. Integrate the node features and graph structure features of the distribution network topology, thereby improving the matching degree between the model and the actual distribution network topology and improving the model accuracy.

[0074] In certain embodiments of the present application, the distribution network topology is iteratively propagated and calculated according to the first graph neural network model, and the model parameters of the graph neural network model are optimized based on the gradient descent method during iteration until the model parameters meet the preset threshold value. The training is completed to obtain a power grid fault location model, which specifically includes:

[0075] Iterative propagation calculation is performed on the distribution network topology according to the first graph neural network model. During the iteration, the model parameters of the first graph neural network model are optimized based on the gradient descent method according to the gradient information of the first graph neural network model. The training is completed until the model parameters meet the preset threshold value, so as to obtain a power grid fault location model.

[0076] In certain embodiments of the present application, the distribution network topology is iteratively propagated and calculated according to the first graph neural network model. During iteration, the model parameters of the graph neural network model are optimized based on the gradient descent method according to the gradient information of the first graph neural network model until the training is completed when the model parameters meet the preset threshold, and the power grid fault location model is obtained, which is specifically:

[0077] Iterative propagation calculation is performed on the distribution network topology according to the first graph neural network model, and the preset loss of the first graph neural network model is calculated during iteration. According to the gradient information of the first graph neural network model, the model parameters of the graph neural network model are optimized based on the gradient descent method until the model loss of the graph neural network model meets the preset threshold, and the training is completed to obtain a power grid fault location model; wherein the preset loss is determined by the current task.

[0078] In some implementations of the present application, the preset loss is determined by the current task, specifically:

[0079] If the current task is fault node category identification, the preset loss is the cross entropy loss;

[0080] If the current task is fault degree prediction, the preset loss is the mean square error loss.

[0081] This application first iterates the distribution network topology according to the first graph neural network model. During the iteration, the model parameters are optimized based on the gradient descent method according to the gradient information to obtain the power grid fault location model. The model parameters can be quickly converged to the preset threshold, which meets the actual task requirements and is conducive to improving the accuracy of the power grid fault location model.

[0082] Step S103: inputting the real-time power supply data of the fault distribution network to be located into the power grid fault location model to obtain fault location data.

[0083] In certain implementations of the present application, the fault location data is specifically:

[0084] If the current task is fault node category identification, the fault location data is the state of each node in the distribution network topology and the corresponding state probability;

[0085] If the current task is fault degree prediction, the fault location data is the state of each node in the distribution network topology and the corresponding state value.

[0086] Specifically, the state value is used to characterize the degree of the state of each node in the distribution network topology.

[0087] Step S104: inputting the fault location data into a preset text annotation model to obtain a fault label of the fault location data, and annotating the real-time power supply data of the fault distribution network to be located based on the fault label to obtain a fault location result.

[0088] In certain embodiments of the present application, the preset text annotation model specifically includes:

[0089] Based on preset division parameters, the preset standard fault location data and the corresponding standard fault label data are divided into fault training data, fault verification data and fault test data;

[0090] Iteratively training a preset initial text annotation model according to the fault training data and the fault verification data, training and updating the learning parameters of the initial text annotation model according to the fault training data in each round of iteration, and verifying and updating the hyperparameters of the initial text annotation model according to the fault verification data, until the initial text annotation model satisfies a preset training end condition, and the training is completed to obtain a preset text annotation model;

[0091] The initial text annotation model meets the preset training end conditions, specifically:

[0092] According to the fault test data, the accuracy and loss value of the initial text annotation model are calculated; wherein, during the calculation, the learning parameters and hyperparameters obtained in the current round are used as model parameters of the initial text annotation model;

[0093] If the accuracy is greater than the preset accuracy threshold and the loss value is less than the preset loss threshold, it is determined that the initial text annotation model of the current round meets the preset training end condition.

[0094] In certain embodiments of the present application, a preferred implementation of the initial text annotation model and the preset text annotation model is a Transformer model.

[0095] The present application first obtains fault training data, fault verification data and fault test data based on preset division parameters, and trains, verifies and updates learning parameters and hyperparameters based on the fault training data and fault verification data, and judges the training end conditions based on the fault test data, which can enable the learning parameters and hyperparameters of the model to quickly converge to preset thresholds, meet actual task requirements, and improve the accuracy of the preset text annotation model.

[0096] The present application first constructs a distribution network topology based on the historical power supply data of the faulty distribution network to be located, and then optimizes the model parameters of a preset graph neural network model based on the node characteristics of each node in the distribution network topology, which can integrate the node characteristics and graph structure characteristics of the distribution network topology into the graph neural network model to reduce model distortion, and then improve the accuracy of the power grid fault location model when the distribution network topology is iterated based on the graph neural network model to obtain a power grid fault location model; then, fault location data is obtained based on real-time power supply data, and the fault location data is input into a preset text annotation model to obtain a fault label, and then the real-time power supply data is annotated according to the fault label to obtain a fault location result, which can intuitively represent the fault location result and prevent errors and distortions caused by traditional manual judgment and positioning, thereby improving positioning accuracy.

[0097] Corresponding to the previous method, see Figure 2 , the embodiment of the present application provides a distribution network fault line location device, including a topology construction module 210, a model training module 220, a fault location module 230 and a fault marking module 240;

[0098] The topology building module 210 is used to build a distribution network topology based on the historical power supply data of the fault distribution network to be located;

[0099] The model training module 220 is used to optimize the model parameters of the preset graph neural network model according to the node characteristics of each node of the distribution network topology, and perform iterative propagation calculation on the distribution network topology based on the graph neural network model. During iteration, the model parameters of the graph neural network model are optimized based on the gradient descent method until the model parameters meet the preset threshold value, and the training is completed to obtain the power grid fault location model;

[0100] The fault location module 230 is used to input the real-time power supply data of the fault distribution network to be located into the power grid fault location model to obtain fault location data;

[0101] The fault labeling module 240 is used to input the fault location data into a preset text labeling model to obtain a fault label of the fault location data, and to label the real-time power supply data of the fault distribution network to be located based on the fault label to obtain a fault location result.

[0102] In certain embodiments of the present application, the graph neural network model includes a self-attention mechanism, and the model training module 220 includes a node feature acquisition submodule, a model parameter optimization submodule, and a model iterative training submodule;

[0103] The node feature acquisition submodule is used to sequentially select nodes of the distribution network topology as current nodes, and obtain a first node feature of the current node based on the self-attention mechanism, until all nodes of the distribution network topology are selected, thereby obtaining multiple first node features of the distribution network topology;

[0104] The model parameter optimization submodule is used to optimize the model parameters of the graph neural network model according to the multiple first node features to obtain a first graph neural network model;

[0105] The model iteration training submodule is used to perform iterative propagation calculations on the distribution network topology according to the first graph neural network model. During iteration, the model parameters of the graph neural network model are optimized based on the gradient descent method until the training is completed when the model parameters meet the preset threshold, thereby obtaining a power grid fault location model.

[0106] In certain embodiments of the present application, the node feature acquisition submodule includes an attention coefficient calculation unit and a node feature aggregation unit;

[0107] The attention coefficient calculation unit is used to calculate the attention coefficients of all neighboring nodes of the current node for the current node based on the self-attention mechanism;

[0108] The node feature aggregation unit is used to aggregate the node features of all adjacent nodes of the current node according to the attention coefficient to obtain the first node feature of the current node.

[0109] In certain embodiments of the present application, the model iteration training submodule includes a model iteration training unit;

[0110] The model iteration training unit is used to perform iterative propagation calculations on the distribution network topology according to the first graph neural network model. During iteration, the model parameters of the first graph neural network model are optimized based on the gradient descent method according to the gradient information of the first graph neural network model. The training is completed when the model parameters meet the preset threshold, thereby obtaining a power grid fault location model.

[0111] In certain embodiments of the present application, the preset text annotation model specifically includes:

[0112] Based on preset division parameters, the preset standard fault location data and the corresponding standard fault label data are divided into fault training data, fault verification data and fault test data;

[0113] Iteratively training a preset initial text annotation model according to the fault training data and the fault verification data, training and updating the learning parameters of the initial text annotation model according to the fault training data in each round of iteration, and verifying and updating the hyperparameters of the initial text annotation model according to the fault verification data, until the initial text annotation model satisfies a preset training end condition, and the training is completed to obtain a preset text annotation model;

[0114] The initial text annotation model meets the preset training end conditions, specifically:

[0115] According to the fault test data, the accuracy and loss value of the initial text annotation model are calculated; wherein, during the calculation, the learning parameters and hyperparameters obtained in the current round are used as model parameters of the initial text annotation model;

[0116] If the accuracy is greater than the preset accuracy threshold and the loss value is less than the preset loss threshold, it is determined that the initial text annotation model of the current round meets the preset training end condition.

[0117] The present application first constructs a distribution network topology based on the historical power supply data of the faulty distribution network to be located, and then optimizes the model parameters of a preset graph neural network model based on the node characteristics of each node in the distribution network topology, which can integrate the node characteristics and graph structure characteristics of the distribution network topology into the graph neural network model to reduce model distortion, and then improve the accuracy of the power grid fault location model when the distribution network topology is iterated based on the graph neural network model to obtain a power grid fault location model; then, fault location data is obtained based on real-time power supply data, and the fault location data is input into a preset text annotation model to obtain a fault label, and then the real-time power supply data is annotated according to the fault label to obtain a fault location result, which can intuitively represent the fault location result and prevent errors and distortions caused by traditional manual judgment and positioning, thereby improving positioning accuracy.

[0118] It should be understood that the device provided in the embodiments of the present application corresponds to the aforementioned method, and a distribution network fault line locating device provided in the embodiments of the present application can implement a distribution network fault line locating method provided in any embodiment of the present application.

[0119] Adaptively, the embodiments of the present application also provide a computer device and a computer-readable storage medium.

[0120] The computer device comprises: a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor;

[0121] Wherein, when the processor executes the computer program, a method for locating a fault line in a distribution network of the present application is implemented.

[0122] The computer-readable storage medium stores a plurality of instructions, and the instructions are suitable for being loaded by a processor to execute a method for locating a fault line in a distribution network of the present application.

[0123] The above are some implementation methods of the present application, and the purpose, technical scheme and beneficial effects of the present application are further described in detail. It should be clear that the above-mentioned partial implementation methods of the present application cannot be understood as limiting the present application. It is particularly pointed out that for those skilled in the art, any changes, modifications, equivalent substitutions and variations made within the spirit and principles of the present application should be included in the protection scope of the present application.

Claims

1. A method for locating a fault line in a distribution network, characterized in that: include: Construct the distribution network topology based on the historical power supply data of the fault distribution network to be located; Optimizing the model parameters of a preset graph neural network model according to the node characteristics of each node of the distribution network topology, and performing iterative propagation calculation on the distribution network topology based on the graph neural network model. During iteration, optimizing the model parameters of the graph neural network model based on the gradient descent method until the model parameters meet the preset thresholds, and completing the training to obtain a power grid fault location model; Inputting the real-time power supply data of the fault distribution network to be located into the power grid fault location model to obtain fault location data; The fault location data is input into a preset text annotation model to obtain a fault label of the fault location data, and the real-time power supply data of the fault distribution network to be located is annotated based on the fault label to obtain a fault location result.

2. A method for locating a fault line in a distribution network according to claim 1, characterized in that: The method optimizes the model parameters of the preset graph neural network model according to the node characteristics of each node of the distribution network topology, and performs iterative propagation calculation on the distribution network topology based on the graph neural network model. During iteration, the model parameters of the graph neural network model are optimized based on the gradient descent method until the model parameters meet the preset threshold value. The training is completed to obtain the power grid fault location model, which specifically includes: The graph neural network model includes a self-attention mechanism; Selecting nodes of the distribution network topology as current nodes in sequence, and obtaining a first node feature of the current node based on the self-attention mechanism, until all nodes of the distribution network topology are selected, thereby obtaining a plurality of first node features of the distribution network topology; Optimizing the model parameters of the graph neural network model according to the multiple first node features to obtain a first graph neural network model; Iterative propagation calculation is performed on the distribution network topology according to the first graph neural network model. During the iteration, the model parameters of the first graph neural network model are optimized based on the gradient descent method until the training is completed when the model parameters meet the preset threshold, so as to obtain a power grid fault location model.

3. A method for locating a fault line in a distribution network according to claim 2, characterized in that: The obtaining of the first node feature of the current node based on the self-attention mechanism specifically includes: Based on the self-attention mechanism, calculate the attention coefficients of all neighboring nodes of the current node to the current node; According to the attention coefficient, node features of all adjacent nodes of the current node are aggregated to obtain a first node feature of the current node.

4. A method for locating a fault line in a distribution network according to claim 2, characterized in that: The iterative propagation calculation of the distribution network topology is performed according to the first graph neural network model, and the model parameters of the graph neural network model are optimized based on the gradient descent method during iteration until the model parameters meet the preset threshold value. The training is completed to obtain the power grid fault location model, which specifically includes: Iterative propagation calculation is performed on the distribution network topology according to the first graph neural network model. During the iteration, the model parameters of the graph neural network model are optimized based on the gradient descent method according to the gradient information of the first graph neural network model. The training is completed until the model parameters meet the preset threshold, so as to obtain a power grid fault location model.

5. A method for locating a fault line in a distribution network according to claim 1, characterized in that: The preset text annotation model specifically includes: Based on preset division parameters, the preset standard fault location data and the corresponding standard fault label data are divided into fault training data, fault verification data and fault test data; Iteratively training a preset initial text annotation model according to the fault training data and the fault verification data, training and updating the learning parameters of the initial text annotation model according to the fault training data in each round of iteration, and verifying and updating the hyperparameters of the initial text annotation model according to the fault verification data, until the initial text annotation model satisfies a preset training end condition, and the training is completed to obtain a preset text annotation model; The initial text annotation model meets the preset training end conditions, specifically: According to the fault test data, the accuracy and loss value of the initial text annotation model are calculated; wherein, during the calculation, the learning parameters and hyperparameters obtained in the current round are used as model parameters of the initial text annotation model; If the accuracy is greater than the preset accuracy threshold and the loss value is less than the preset loss threshold, it is determined that the initial text annotation model of the current round meets the preset training end condition.

6. A distribution network fault line locating device, characterized in that: It includes topology construction module, model training module, fault location module and fault marking module; The topology building module is used to build a distribution network topology based on historical power supply data of the fault distribution network to be located; The model training module is used to optimize the model parameters of a preset graph neural network model according to the node characteristics of each node of the distribution network topology, and perform iterative propagation calculations on the distribution network topology based on the graph neural network model. During iteration, the model parameters of the graph neural network model are optimized based on the gradient descent method until the training is completed when the model parameters meet the preset threshold value, thereby obtaining a power grid fault location model; The fault location module is used to input the real-time power supply data of the fault distribution network to be located into the power grid fault location model to obtain fault location data; The fault labeling module is used to input the fault location data into a preset text labeling model to obtain a fault label of the fault location data, and to label the real-time power supply data of the fault distribution network to be located based on the fault label to obtain a fault location result.

7. A distribution network fault line locating device according to claim 6, characterized in that: The graph neural network model includes a self-attention mechanism, and the model training module includes a node feature acquisition submodule, a model parameter optimization submodule and a model iterative training submodule; The node feature acquisition submodule is used to sequentially select nodes of the distribution network topology as current nodes, and obtain a first node feature of the current node based on the self-attention mechanism, until all nodes of the distribution network topology are selected, thereby obtaining multiple first node features of the distribution network topology; The model parameter optimization submodule is used to optimize the model parameters of the graph neural network model according to the multiple first node features to obtain a first graph neural network model; The model iteration training submodule is used to perform iterative propagation calculations on the distribution network topology according to the first graph neural network model. During iteration, the model parameters of the graph neural network model are optimized based on the gradient descent method until the training is completed when the model parameters meet the preset threshold, thereby obtaining a power grid fault location model.

8. A distribution network fault line locating device according to claim 7, characterized in that: The node feature acquisition submodule includes an attention coefficient calculation unit and a node feature aggregation unit; The attention coefficient calculation unit is used to calculate the attention coefficients of all neighboring nodes of the current node for the current node based on the self-attention mechanism; The node feature aggregation unit is used to aggregate the node features of all adjacent nodes of the current node according to the attention coefficient to obtain the first node feature of the current node.

9. A distribution network fault line locating device according to claim 7, characterized in that: The model iteration training submodule includes a model iteration training unit; The model iteration training unit is used to perform iterative propagation calculations on the distribution network topology according to the first graph neural network model. During iteration, the model parameters of the first graph neural network model are optimized based on the gradient descent method according to the gradient information of the first graph neural network model. The training is completed when the model parameters meet the preset threshold, thereby obtaining a power grid fault location model.

10. A distribution network fault line locating device according to claim 6, characterized in that: The preset text annotation model specifically includes: Based on preset division parameters, the preset standard fault location data and the corresponding standard fault label data are divided into fault training data, fault verification data and fault test data; Iteratively training a preset initial text annotation model according to the fault training data and the fault verification data, training and updating the learning parameters of the initial text annotation model according to the fault training data in each round of iteration, and verifying and updating the hyperparameters of the initial text annotation model according to the fault verification data, until the initial text annotation model satisfies a preset training end condition, and the training is completed to obtain a preset text annotation model; The initial text annotation model meets the preset training end conditions, specifically: According to the fault test data, the accuracy and loss value of the initial text annotation model are calculated; wherein, during the calculation, the learning parameters and hyperparameters obtained in the current round are used as model parameters of the initial text annotation model; If the accuracy is greater than the preset accuracy threshold and the loss value is less than the preset loss threshold, it is determined that the initial text annotation model of the current round meets the preset training end condition.

Images