Power communication network fault positioning method, device, equipment and product based on graph neural network

Abstract

The invention discloses an electric power communication network fault positioning method, device, equipment and product based on a graph neural network, and relates to the technical field of fault detection, the method comprises the following steps: collecting a data set of a target electric power communication network, each sample data in the data set comprising unstructured fault text data and structured alarm data; constructing a power communication network knowledge graph according to each piece of sample data; training a graph neural network according to each power communication network knowledge graph to obtain a fault positioning model; the graph neural network is used for representing a relationship between a fault entity and an alarm entity in the power communication network knowledge graph; obtaining to-be-predicted alarm data of the target power communication network, and converting the to-be-predicted alarm data into an alarm vector; and inputting an alarm vector into the fault positioning model to obtain a fault positioning result. According to the invention, the fault positioning accuracy can be improved.

Description

Technical Field

[0001] The present application relates to the field of fault detection technology, and in particular to a method, device, equipment and product for locating faults in a power communication network based on a graph neural network. Background Art

[0002] The power communication network is the foundation for supporting and ensuring the safe and stable operation of the power system. It undertakes important tasks such as production command and dispatch, and provides automated information transmission, administrative management, and other related services for the power grid. With the continuous advancement of communication technology and the continuous development of network services, the scale of the power communication network has continued to expand, gradually becoming diversified, heterogeneous, and complex. Faults in the power communication network are frequent and difficult to locate, posing a significant challenge to the stability and security of the power system. Therefore, to effectively address the risks posed by network failures, it is necessary to enhance the fault location capabilities of the power communication network so that appropriate measures can be taken immediately and the impact of faults on the operation of the power communication network can be reduced. Traditional power communication network fault diagnosis technologies are mostly rule-based, often lacking sufficient flexibility and accuracy, and cannot meet the high real-time and accuracy requirements of modern power communication systems. Summary of the Invention

[0003] The purpose of this application is to provide a method, device, equipment and product for fault location in a power communication network based on a graph neural network, which can improve the accuracy of fault location.

[0004] To achieve the above objectives, this application provides the following solutions:

[0005] In a first aspect, the present application provides a method for locating faults in a power communication network based on a graph neural network, comprising:

[0006] Collecting a data set of a target power communication network, wherein each sample data in the data set includes unstructured fault text data and structured alarm data;

[0007] Constructing a power communication network knowledge graph based on each sample data; the nodes in the power communication network knowledge graph include equipment in the target power communication network, fault entities extracted from the fault text data, and alarm entities extracted from the alarm data;

[0008] A graph neural network is trained based on the knowledge graphs of each power communication network to obtain a fault location model; the graph neural network is used to represent the relationship between the fault entity and the alarm entity in the power communication network knowledge graph;

[0009] Acquire the target power communication network's alarm data to be predicted, and convert the alarm data to be predicted into an alarm vector;

[0010] The alarm vector is input into the fault location model to obtain a fault location result.

[0011] Optionally, a knowledge graph of the power communication network is constructed based on the unstructured fault text data and the structured alarm data, specifically including:

[0012] Extracting a fault entity sequence from the fault text data using a power communication fault entity recognition model;

[0013] Filtering the alarm information corresponding to the fault entity from the alarm data, and mapping the alarm information to the alarm entity to obtain an alarm entity set;

[0014] Obtaining an entity relationship triple according to an entity set consisting of fault entities in the fault entity sequence and alarm entities in the alarm entity set and a preset rule;

[0015] The knowledge graph of the power communication network is constructed according to each triple and the communication topology graph of the power communication network.

[0016] Optionally, the power communication fault entity recognition model is a BERT-BiLSTM-CRF model.

[0017] Optionally, a graph neural network is trained based on the knowledge graphs of each power communication network to obtain a fault location model, specifically including:

[0018] For each electric power communication network knowledge graph, the alarm entities in the electric power communication network knowledge graph are converted into alarm vector samples, the alarm vector samples are used as input, and the fault entities in the electric power communication network knowledge graph are used as output to train the graph neural network.

[0019] Optionally, converting the alarm entity in the power communication network knowledge graph into an alarm vector sample specifically includes:

[0020] One-hot encoding is used to convert each alarm entity into an M-dimensional vector according to the index value of each alarm entity; M is the number of categories of the alarm entity;

[0021] The M-dimensional vector is used as the alarm vector sample of the corresponding alarm entity.

[0022] Optionally, in the process of training the graph neural network according to the knowledge graph of each power communication network, GeneralConv is used to update the node features in the graph neural network, and a Softmax classifier is used to obtain the failure probability of each node, and the failure probability of each edge is determined according to the characteristics of each node;

[0023] The failure probability of each edge is expressed as:

[0024] in, is the failure probability of edge (i, j), edge (i, j) is the edge between node i and node j, W e is the binary classification weight, H edge,ij Represents the characteristics of edge (i, j); is the feature representation of the last layer of node i in the graph neural network, is the feature representation of node j in the last layer of the graph neural network.

[0025] Optionally, the loss function when training the graph neural network for each power communication network knowledge graph includes a node prediction loss function and an edge prediction loss function;

[0026] The node prediction loss function is expressed as:

[0027] The edge prediction loss function is expressed as:

[0028] in, Predict the loss for the node, is the edge prediction loss, N is the total number of nodes; is the predicted failure probability distribution of node i; Y i is the true fault label; E is the total number of edges; The predicted failure probability for edge (i, j); Y edge,ij is the true fault label of edge (i, j), and CrossEntropy() represents the cross entropy loss.

[0029] In a second aspect, the present application provides a graph neural network-based power communication network fault location device, wherein the graph neural network-based power communication network fault location device applies any of the graph neural network-based power communication network fault location methods described above, and the graph neural network-based power communication network fault location device includes:

[0030] A data set acquisition module, configured to acquire a data set of a target power communication network, wherein each sample data in the data set includes unstructured fault text data and structured alarm data;

[0031] A power communication network knowledge graph construction module is used to construct a power communication network knowledge graph based on each sample data; the nodes in the power communication network knowledge graph include equipment in the target power communication network, fault entities extracted from the fault text data, and alarm entities extracted from the alarm data;

[0032] A fault location model determination module is used to train a graph neural network based on each power communication network knowledge graph to obtain a fault location model. The graph neural network is used to represent the relationship between fault entities and alarm entities in the power communication network knowledge graph;

[0033] A vector conversion module is used to obtain the alarm data to be predicted of the target power communication network and convert the alarm data to be predicted into an alarm vector;

[0034] The fault location module is used to input the alarm vector into the fault location model to obtain a fault location result.

[0035] In a third aspect, the present application provides a computer device comprising: a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of any one of the above-mentioned methods for locating faults in a power communication network based on a graph neural network.

[0036] In a fourth aspect, the present application provides a computer program product, comprising a computer program, which, when executed by a processor, implements the steps of any one of the above-mentioned methods for locating faults in a power communication network based on a graph neural network.

[0037] According to the specific embodiments provided in this application, this application discloses the following technical effects:

[0038] The present application provides a method, apparatus, equipment and product for locating faults in an electric power communication network based on a graph neural network. The method models entities and their relationships in the electric power communication network in the form of a graph, and uses a graph neural network to represent the relationship between fault entities and alarm entities in the knowledge graph of the electric power communication network. This effectively combines the result features and attribute features of each node, thereby improving the accuracy of fault location. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0040] Figure 1 A schematic diagram of a flow chart of a method for locating faults in a power communication network based on a graph neural network according to an embodiment of the present application;

[0041] Figure 2 A schematic diagram of the process of constructing a knowledge graph for a power communication network according to an embodiment of the present application;

[0042] Figure 3 A schematic diagram of the structure of a power communication fault entity recognition model based on BERT-BiLSTM-CRF provided in one embodiment of the present application;

[0043] Figure 4A schematic diagram of a fault location process based on a graph neural network provided in one embodiment of the present application;

[0044] Figure 5 A schematic diagram of the effect of loss value in scenario 1 provided in an embodiment of the present application;

[0045] Figure 6 A schematic diagram of the loss value effect in scenario 2 provided in an embodiment of the present application;

[0046] Figure 7 A schematic diagram of the functional modules of a graph neural network-based power communication network fault location device provided in one embodiment of the present application;

[0047] Figure 8 A schematic diagram of the structure of a computer device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0048] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0049] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0050] In an exemplary embodiment, the present application provides a method for locating faults in a power communication network based on a graph neural network, such as Figure 1 As shown, the power communication network fault detection method based on graph neural network includes steps 101 to 105.

[0051] Step 101: Collect a data set of a target power communication network, where each sample data in the data set includes unstructured fault text data and structured alarm data.

[0052] Step 102: constructing a power communication network knowledge graph based on each sample data; the nodes in the power communication network knowledge graph include equipment in the target power communication network, fault entities extracted from the fault text data, and alarm entities extracted from the alarm data.

[0053] Step 103: Train a graph neural network based on each power communication network knowledge graph to obtain a fault location model; the graph neural network is used to represent the relationship between the fault entity and the alarm entity in the power communication network knowledge graph.

[0054] Step 104: Acquire the alarm data to be predicted of the target electric power communication network, and convert the alarm data to be predicted into an alarm vector.

[0055] Step 105: Input the alarm vector into the fault location model to obtain a fault location result.

[0056] A knowledge graph is a structured semantic knowledge base that describes the relationships between entities in the objective world in the form of a graph. This application models the entities and their relationships in the power communication network in the form of a graph. Furthermore, graph neural networks can effectively combine the structural and attribute characteristics of knowledge graphs, thereby effectively improving the accuracy and efficiency of fault location.

[0057] In an exemplary embodiment, step 102 constructs a power communication network knowledge graph based on unstructured fault text data and structured alarm data, specifically including:

[0058] Step 201: extracting a fault entity sequence from the fault text data using a power communication fault entity recognition model.

[0059] Step 202: Filter out alarm information corresponding to the fault entity from the alarm data, and map the alarm information to the alarm entity to obtain an alarm entity set.

[0060] Step 203: Obtain entity relationship triples according to the entity set consisting of the fault entities in the fault entity sequence and the alarm entities in the alarm entity set and preset rules.

[0061] Step 204: construct a knowledge graph of the power communication network based on each triple and the communication topology graph of the power communication network.

[0062] This application first uses a knowledge graph to represent the alarm data and topology of the power communication network in a graph-structured manner. It then uses GeneralConv to embed learning into this graph data, capturing complex fault propagation patterns. Experimental results demonstrate that this method can accurately identify fault locations and achieves excellent performance in metrics such as accuracy, recall, and F1 value. Future research could further optimize the structure of graph neural networks, such as by introducing attention mechanisms, to improve fault location accuracy. GeneralConv is a convolution function.

[0063] In an exemplary embodiment, the power communication fault entity recognition model is a Bidirectional Encoder Representation from Transformers-Bidirectional Long Short-Term Memory-Conditional Random Field (BERT-BiLSTM-CRF) model based on a transformer.

[0064] The construction of the electric power communication knowledge graph is mainly divided into three parts: entity extraction, relationship extraction, and knowledge storage. First, for unstructured fault text data, this application discloses an electric power communication fault entity recognition model based on BERT-BiLSTM-CRF to obtain key entity information of the fault text; for structured alarm data, the extracted fault data is used, combined with the topological structure, to filter out the alarm information corresponding to the fault, and directly map the relevant fields in the alarm information to the corresponding entity category; based on the acquired entity information and data characteristics, a method for extracting electric power communication relationships based on preset rules is proposed; finally, the acquired entity relationship information is constructed into triples, and the Neo4j graph database is used to realize the storage and visualization of the electric power communication knowledge graph, thereby completing the construction of the electric power communication knowledge graph. The specific process is as follows: Figure 2 As shown in the figure, the preset rules are known rules summarized based on historical experience.

[0065] Among them, the BERT-BiLSTM-CRF model includes the BERT pre-trained language model, BiLSTM network and CRF layer.

[0066] For the BERT-BiLSTM-CRF-based power communication fault entity recognition model, first, the fault text data is annotated to construct a data set suitable for BERT-BiLSTM-CRF model training; the training data is converted into word vectors through the BERT pre-trained language model, and then the BiLSTM network is used to encode the fault text features. Finally, the features output by BiLSTM are passed to the CRF layer for label constraint to ensure that the global optimal entity sequence is obtained, that is, the fault entity sequence in step 201. The BERT-BiLSTM-CRF model structure is as follows: Figure 3 shown.

[0067] The core component of the BERT pre-trained language model is the bidirectional Transformer encoder, which consists of multiple encoding units. Each encoding unit contains a self-attention mechanism. The output of the attention mechanism is obtained by calculating the similarity between the query (Q) and the key (K) and the normalized weighted value (V).

[0068]

[0069] Where: Q, K, V are the matrices of query, key, and value respectively, d k is the dimension of the key.

[0070] A single attention head is prone to being limited in capturing only a certain type of semantic information. To address this issue, the multi-head attention mechanism uses multiple independent attention heads to perform self-attention calculations in parallel. Each head learns a different representation subspace and finally merges the information from different subspaces to more comprehensively capture the semantic relationships of the input data.

[0071] A multi (Q,K,V)=C(head1,head2,head3,L,head h )W O (2)

[0072]

[0073] Where: A multi (Q, K, V) is a multi-head attention mechanism; C() is vector concatenation; W O is the weight matrix; head i is the output vector of the i-th head; is the weight matrix learned independently by the i-th head.

[0074] The BiLSTM grid captures the dependencies between the context in the sequence through a bidirectional structure, so that the output of each time step is affected by both the current input and its context information.

[0075] The BiLSTM grid cannot directly consider the label relationship and global constraints between sequences. Introducing the CRF layer on the basis of the BiLSTM grid can model the label transition probability, optimize the global constraints, and improve the accuracy of entity recognition.

[0076] By analyzing the structural characteristics of power communication data and combining it with the extracted entity information, we use preset rules to define a series of relationship sets to extract power communication relationships. This application defines three relationship types, which aim to associate entities (including fault entities and alarm entities) into triples, as shown in Table 1.

[0077] Table 1 Definition of power communication relationship rules

[0078]

[0079] In Table 1, A and B represent nodes, GEN represents generation relationship, CAU represents cause relationship, and DER represents derive relationship. A_GEN_B means entity A generates entity B, and the relationship type is generation relationship. A_CAU_B means entity A causes entity B, and the relationship type is cause relationship. A_DER_B means entity A derives entity B, and the relationship type is derive relationship.

[0080] This application chooses Neo4j to store and visualize the obtained triples. Neo4j is a graph-based database that provides an intuitive graphical interface to help quickly understand the fault propagation path.

[0081] In an exemplary embodiment, a graph neural network is trained based on each power communication network knowledge graph to obtain a fault location model, specifically including: for each power communication network knowledge graph, converting the alarm entity in the power communication network knowledge graph into an alarm vector sample, taking the alarm vector sample as input, and taking the fault entity in the power communication network knowledge graph as output to train the graph neural network.

[0082] After completing the construction of the power communication knowledge graph, how to effectively apply it to fault location becomes the key. Although the knowledge graph has structured and integrated the knowledge in the power communication field, to achieve accurate fault location, it is still necessary to use a graph neural network model that can process graph structures and deeply mine the information contained in the nodes and edges in the knowledge graph. The specific process is as follows: Figure 4 As shown, Figure 4 XX transformer represents a device under a substation. XX line is the connection line between two devices, such as an optical cable. AU_ALS and R_LOS are both alarm information. AU_ALS represents the auxiliary alarm indication signal, and R_LOS represents the remote signal loss.

[0083] In an exemplary embodiment, converting the alarm entity in the power communication network knowledge graph into an alarm vector sample specifically includes:

[0084] One-hot encoding is used to convert each alarm entity into an M-dimensional vector according to the index value of each alarm entity; M is the number of categories of the alarm entity.

[0085] The M-dimensional vector is used as the alarm vector sample of the corresponding alarm entity.

[0086] In an exemplary embodiment, obtaining the alarm data to be predicted of the target power communication network and converting the alarm data to be predicted into an alarm vector specifically includes: node feature representation and initialization.

[0087] The system obtains the alarm information to be predicted due to the fault. After preprocessing such as de-redundancy, it uses an encoder to convert the alarm data into an alarm vector representation. These alarm vectors are used as the initial input features of the nodes in the current power communication network knowledge graph. In other words, the device that generated the alarm information is used as the initial input feature of the device node that generated the alarm information. The alarm vector is obtained using one-hot encoding.

[0088] The electric power communication network knowledge graph contains M different types of alarms. Each alarm category is represented by an index m (ranging from 0 to M-1). One-hot encoding maps the category index m to an N-dimensional vector, where the mth dimension is 1 and the other dimensions are 0.

[0089] One-Hot encoding is used to convert the alarm information to be predicted into an M-dimensional vector according to the index value of the alarm entity.

[0090]

[0091] Where: e m [k] represents the value of the one-hot encoding in the kth dimension corresponding to the alarm type index m; δ mk is the Kronecker delta function, which indicates whether k is equal to m.

[0092] Map the one-hot encoding of the alarm to be predicted to the feature vector representation of the corresponding node, that is, the alarm vector.

[0093] h v =e m (5)

[0094] Where: h v represents the feature vector of node v; e m It is the one-hot code of the alarm type corresponding to the node.

[0095] In an exemplary embodiment, in the process of training the graph neural network based on the knowledge graph of each power communication network, GeneralConv is used to update the node features in the graph neural network, the Softmax classifier is used to obtain the failure probability of each node, and the failure probability of each edge is determined based on the characteristics of each node.

[0096] The graph neural network used in this application has richer aggregation capabilities and stronger expression capabilities than the graph convolutional network (GCN). It can effectively capture the fault propagation pattern affected by neighbors and show superior performance.

[0097]

[0098] Where: is the feature of node v in the k+1th layer; is the neighbor set of node v; α vu is the adjacency weight coefficient; W is the trainable weight matrix; b is the bias term; σ(·) is the activation function.

[0099] In the fault location task, node fault determination uses a fully connected layer W c The final features are mapped to a binary classification (normal / faulty), that is, predicting whether a device has failed. Finally, Softmax classification can be used to obtain the class probability distribution of the node.

[0100]

[0101] Where: is the feature representation of node i in the last layer; W c is the binary classification weight; is the predicted failure probability of node i; Softmax() outputs the probability of each node belonging to different categories after normalization.

[0102] Similarly, predicting whether a certain optical cable is faulty is to perform edge fault determination.

[0103]

[0104] in, is the failure probability of edge (i, j), edge (i, j) is the edge between node i and node j, W e is the binary classification weight, H edge,ij Represents the characteristics of edge (i, j); is the feature representation of the last layer of node i in the graph neural network, is the feature representation of node j in the last layer of the graph neural network.

[0105] In an exemplary embodiment, the loss function used when training the graph neural network for each electric power communication network knowledge graph includes a node prediction loss function and an edge prediction loss function, both of which use cross entropy loss (CrossEntropy).

[0106] The node prediction loss function is expressed as:

[0107]

[0108] The edge prediction loss function is expressed as:

[0109]

[0110] in, Predict the loss for the node, is the edge prediction loss, N is the total number of nodes; is the predicted failure probability distribution of node i; Y i is the true fault label; E is the total number of edges; The predicted failure probability for edge (i, j); Y edge,ij is the true fault label of edge (i, j), and CrossEntropy() represents the cross entropy loss. An edge fault refers to a failure in the connection line between two nodes, such as a fiber optic cable failure.

[0111] The final optimization goal is to minimize the node and edge losses simultaneously.

[0112]

[0113] in, For the total loss.

[0114] In an exemplary embodiment, a graph neural network-based power communication network fault location method of the present application is simulated and verified.

[0115] The graph neural network consists of two GeneralConv graph convolution layers, two hidden layers, and two output layers.

[0116] The experimental data for simulation verification mainly comes from the telecommunications company. The experimental data specifically includes: structured alarm information, each alarm information includes fields such as alarm source, alarm name, alarm start time and alarm clearing time; unstructured fault text data, which records the faults occurring in the power communication network and their descriptions. The unstructured fault text data includes information such as the device where the fault occurred, the fault type, the occurrence time and the completion time; network topology data, which describes the connection between devices, communication paths and other information.

[0117] Simulation Setup: PyCharm was used to support deep learning and graph data processing, while the knowledge graph was stored and inferred using the Neo4j graph database. The model used two GeneralConv graph convolutional layers with a hidden layer dimension of 16 and a learning rate of 0.005. The model was trained for 300 epochs to ensure convergence. The Adam optimizer was used for optimization, and a weighted cross-entropy loss function was employed to address class imbalance.

[0118] Simulation results: In order to verify the applicability of general graph convolution in the task of fault location in power communication networks, two different experimental scenarios are designed: In scenario 1, there are no derivative alarms generated by alarm propagation, such as Figure 5 As shown; Scene 2 has derivative alarm phenomena, such as Figure 6The precision (P), accuracy (ACC), recall (R) and F1 score under different experimental scenarios are compared, as shown in Table 2.

[0119] Table 2 Comparison of experimental results in different scenarios

[0120] Scenario Precision (P) Accuracy (ACC) Recall (R) <![CDATA[F1]]> Scene 1 99.23％ 96.99％ 97.73％ 98.47％ Scene 2 96.88％ 95.52％ 98.41％ 97.64％

[0121] The comparison results lead to the following conclusions: Generalized graph convolution is an effective method for fault location in power communication networks. In the absence of derivative alarms, it can accurately identify the fault point with an accuracy of 96.99%; in the presence of derivative alarms, the accuracy reaches 95.52%, demonstrating that generalized graph convolution maintains strong adaptability when handling complex alarm propagation patterns. The loss value effect graph shows a steady decline with no significant overfitting, demonstrating that the proposed fault location model has good generalization capabilities and is adaptable to power communication network fault location tasks in diverse scenarios.

[0122] This application discloses a knowledge graph model applicable to power communication networks by analyzing the topological structure and alarm fault types of power communication networks. This model constructs a graph structure containing device nodes, alarm nodes, fault nodes, and the connection relationships between each device. The relationship between device nodes and alarm nodes is a generation relationship, between device nodes and fault nodes is a generation relationship, and between fault nodes and alarm nodes is a cause relationship. Based on this, GeneralConv is used as the core operator of the graph neural network to effectively integrate node neighbor information, train and infer graph data, and accurately infer fault sources during the testing phase. Experimental results show that this method performs well in different power communication fault location scenarios and can accurately infer fault sources.

[0123] Based on the same inventive concept, the embodiments of the present application also provide a graph neural network-based power communication network fault location device for implementing the graph neural network-based power communication network fault location method involved above. The implementation solution provided by this device is similar to the implementation solution described in the above method. Therefore, the specific limitations of one or more embodiments of the graph neural network-based power communication network fault location device provided below can be found in the above-mentioned limitations of the graph neural network-based power communication network fault location method, and will not be repeated here.

[0124] In an exemplary embodiment, Figure 7 As shown, a power communication network fault location device based on a graph neural network is provided, including:

[0125] The data set acquisition module is used to acquire the data set of the target power communication network, wherein each sample data in the data set includes unstructured fault text data and structured alarm data.

[0126] The power communication network knowledge graph construction module is used to construct a power communication network knowledge graph based on each sample data; the nodes in the power communication network knowledge graph include equipment in the target power communication network, fault entities extracted from the fault text data, and alarm entities extracted from the alarm data.

[0127] The fault location model determination module is used to train a graph neural network based on each power communication network knowledge graph to obtain a fault location model. The graph neural network is used to represent the relationship between the fault entity and the alarm entity in the power communication network knowledge graph.

[0128] The vector conversion module is used to obtain the alarm data to be predicted of the target power communication network and convert the alarm data to be predicted into an alarm vector.

[0129] The fault location module is used to input the alarm vector into the fault location model to obtain a fault location result.

[0130] In an exemplary embodiment, a computer device is provided. The computer device may be a server or a terminal. The internal structure diagram thereof may be as follows: Figure 8 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O) and a communication interface. The processor, memory and input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The database of the computer device is used to store power communication network fault location data based on graph neural network. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a power communication network fault location method based on graph neural network is implemented.

[0131] Those skilled in the art will understand that Figure 8The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present application and does not constitute a limitation on the computer device to which the solution of the present application is applied. A specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement. In an exemplary embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor implements the steps of the above-mentioned method embodiments when executing the computer program.

[0132] In an exemplary embodiment, a computer program product is provided, including a computer program. When the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0133] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0134] Those skilled in the art will understand that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, database or other media used in the embodiments provided in this application may include at least one of non-volatile and volatile memory. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory may include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0135] The databases involved in the various embodiments provided herein may include at least one of a relational database and a non-relational database. Non-relational databases may include, but are not limited to, distributed databases based on blockchains. The processors involved in the various embodiments provided herein may be, but are not limited to, general-purpose processors, central processing units, graphics processing units, digital signal processors, data processing logic of programmable logic devices, and the like.

[0136] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0137] This document uses specific examples to illustrate the principles and implementation methods of this application. The description of the above examples is only intended to help understand the method and core concept of this application. At the same time, for those skilled in the art, based on the concept of this application, there may be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A method for locating faults in a power communication network based on graph neural networks, characterized in that: The power communication network fault location method based on graph neural network includes: Collecting a data set of a target power communication network, wherein each sample data in the data set includes unstructured fault text data and structured alarm data; Constructing a power communication network knowledge graph based on each sample data; the nodes in the power communication network knowledge graph include equipment in the target power communication network, fault entities extracted from the fault text data, and alarm entities extracted from the alarm data; A graph neural network is trained based on the knowledge graphs of each power communication network to obtain a fault location model; the graph neural network is used to represent the relationship between the fault entity and the alarm entity in the power communication network knowledge graph; Acquire the target power communication network's alarm data to be predicted, and convert the alarm data to be predicted into an alarm vector; The alarm vector is input into the fault location model to obtain a fault location result.

2. The method for locating faults in a power communication network based on graph neural network according to claim 1, characterized in that: The knowledge graph of the power communication network is constructed based on unstructured fault text data and structured alarm data, including: Extracting a fault entity sequence from the fault text data using a power communication fault entity recognition model; Filtering the alarm information corresponding to the fault entity from the alarm data, and mapping the alarm information to the alarm entity to obtain an alarm entity set; Obtaining an entity relationship triple according to an entity set consisting of fault entities in the fault entity sequence and alarm entities in the alarm entity set and a preset rule; The knowledge graph of the power communication network is constructed according to each triple and the communication topology graph of the power communication network.

3. The method for locating faults in a power communication network based on graph neural network according to claim 2, characterized in that: The power communication fault entity recognition model is a BERT-BiLSTM-CRF model.

4. The method for locating faults in a power communication network based on graph neural network according to claim 2, characterized in that: The graph neural network is trained based on the knowledge graphs of each power communication network to obtain the fault location model, which specifically includes: For each electric power communication network knowledge graph, the alarm entities in the electric power communication network knowledge graph are converted into alarm vector samples, the alarm vector samples are used as input, and the fault entities in the electric power communication network knowledge graph are used as output to train the graph neural network.

5. The method for locating faults in a power communication network based on graph neural network according to claim 4, characterized in that: Converting the alarm entities in the power communication network knowledge graph into alarm vector samples specifically includes: One-hot encoding is used to convert each alarm entity into an M-dimensional vector according to the index value of each alarm entity; M is the number of categories of the alarm entity; The M-dimensional vector is used as the alarm vector sample of the corresponding alarm entity.

6. The method for locating faults in a power communication network based on graph neural network according to claim 1, characterized in that: In the process of training the graph neural network based on the knowledge graph of each power communication network, GeneralConv is used to update the node features in the graph neural network, and the Softmax classifier is used to obtain the failure probability of each node. The failure probability of each edge is determined based on the characteristics of each node; The failure probability of each edge is expressed as: in, is the failure probability of edge (i, j), edge (i, j) is the edge between node i and node j, W e is the binary classification weight, H edge,ij Represents the characteristics of edge (i, j); is the feature representation of the last layer of node i in the graph neural network, is the feature representation of node j in the last layer of the graph neural network.

7. The method for locating faults in a power communication network based on graph neural network according to claim 1, characterized in that: The loss functions for training graph neural networks in the knowledge graphs of each power communication network include node prediction loss function and edge prediction loss function; The node prediction loss function is expressed as: The edge prediction loss function is expressed as: in, Predict the loss for the node, is the edge prediction loss, N is the total number of nodes; is the predicted failure probability distribution of node i; Y i is the true fault label; E is the total number of edges; The predicted failure probability for edge (i, j); Y edge,ij is the true fault label of edge (i, j), and CrossEntropy() represents the cross entropy loss.

8. A power communication network fault location device based on graph neural network, characterized in that: The graph neural network-based power communication network fault location device applies the graph neural network-based power communication network fault location method according to any one of claims 1 to 7, and the graph neural network-based power communication network fault location device includes: A data set acquisition module, configured to acquire a data set of a target power communication network, wherein each sample data in the data set includes unstructured fault text data and structured alarm data; A power communication network knowledge graph construction module is used to construct a power communication network knowledge graph based on each sample data; the nodes in the power communication network knowledge graph include equipment in the target power communication network, fault entities extracted from the fault text data, and alarm entities extracted from the alarm data; A fault location model determination module is used to train a graph neural network based on each power communication network knowledge graph to obtain a fault location model. The graph neural network is used to represent the relationship between fault entities and alarm entities in the power communication network knowledge graph; A vector conversion module is used to obtain the alarm data to be predicted of the target power communication network and convert the alarm data to be predicted into an alarm vector; The fault location module is used to input the alarm vector into the fault location model to obtain a fault location result.

9. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the power communication network fault location method based on graph neural network according to any one of claims 1 to 7.

10. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, it implements the power communication network fault location method based on graph neural network according to any one of claims 1 to 7.

Images