Power equipment fault diagnosis and control strategy recommendation method based on knowledge graph

Abstract

The invention discloses a power equipment fault diagnosis and control strategy recommendation method based on a knowledge graph, and relates to the technical field of power equipment diagnosis, and the method comprises the steps: collecting power equipment operation parameters, defining a node type and a relation type of the knowledge graph, and constructing a power equipment fault diagnosis knowledge graph; the method comprises the following steps of: extracting node features of nodes, and storing obtained node vectors into a vector database by taking optimization of a space distance between adjacent node vectors as a target to realize similarity query and semantic matching between the nodes; performing multi-hop reasoning in the knowledge graph by taking the alarm node as a starting point, screening a high-confidence path as a diagnosis conclusion, and generating a power equipment strategy recommendation list; evaluating the execution effect of the power equipment strategy recommendation list, adjusting the edge weight of the mapping knowledge domain corresponding relation, dynamically updating the structure and node vector of the power equipment fault diagnosis mapping knowledge domain, and achieving the self-adaptive optimization of the power equipment fault diagnosis mapping knowledge domain. According to the invention, the power equipment fault diagnosis capability is improved.

Description

Technical Field

[0001] This invention relates to the field of power equipment diagnostic technology, specifically to a method for power equipment fault diagnosis and control strategy recommendation based on knowledge graphs. Background Technology

[0002] In existing technologies, fault diagnosis and strategy recommendation for power equipment often rely on expert experience bases or static systems based on fixed rules. They typically process alarm information and historical data in isolation, lacking effective integration and utilization of deep semantic relationships between multi-source, heterogeneous data, such as operating parameters, meteorological conditions, and maintenance records, resulting in a one-sided diagnostic perspective. Once the system knowledge base is established, it is difficult to update and cannot learn autonomously from subsequent execution feedback, causing the recommended strategies to be lagging or ineffective, resulting in poor adaptability. Traditional methods have rigid reasoning capabilities, making it difficult to perform flexible and in-depth correlation reasoning when faced with complex new faults or multiple faults occurring concurrently, affecting the accuracy of diagnosis and the precision of strategy recommendation. Summary of the Invention

[0003] To address the aforementioned technical problems, this technical solution provides a knowledge graph-based method for recommending fault diagnosis and control strategies for power equipment, thus resolving the aforementioned issues.

[0004] To achieve the above objectives, the technical solution adopted by the present invention is as follows: Knowledge graph-based methods for recommending fault diagnosis and control strategies for power equipment include: S1. Based on the multi-dimensional system of power equipment, collect the operating parameters, alarm logs, maintenance records and meteorological data of the target area of ​​power equipment, define the node types and relationship types of the knowledge graph, and construct a knowledge graph for power equipment fault diagnosis. S2. Based on the knowledge graph of power equipment fault diagnosis, extract the node features of the knowledge graph of power equipment fault diagnosis, map the nodes into low-dimensional vectors, optimize the vector space distance between adjacent nodes as the objective, store the obtained node vectors in the vector database, and realize the similarity query and semantic matching between nodes. S3. If the monitoring system detects an abnormality in the power equipment, an alarm signal is immediately triggered. Starting from the alarm node, multi-hop reasoning is performed in the power equipment fault diagnosis knowledge graph to select high-confidence paths as diagnostic conclusions and generate a priority-ranked list of recommended power equipment strategies. S4. Based on the priority-ranked power equipment strategy recommendation list, evaluate the execution effect of the power equipment strategy recommendation list, adjust the edge weights of the corresponding relationships in the power equipment fault diagnosis knowledge graph, dynamically update the structure and node vectors of the power equipment fault diagnosis knowledge graph, and realize adaptive optimization of the power equipment fault diagnosis knowledge graph.

[0005] Preferably, step S1 specifically includes: Based on the power equipment monitoring system, the operating parameters of the power equipment are acquired in real time, including current, voltage, temperature, oil chromatography data and partial discharge signals. Based on the power equipment management system, obtain power equipment ledgers, models, technical parameters and connection topologies; Based on the power equipment maintenance record system, historical fault records, maintenance reports, test data and replacement parts records of power equipment are obtained; Based on the weather forecast APP for the target area, obtain meteorological data for the target area, including temperature, humidity, wind speed, rainfall and lightning data; Data from power equipment monitoring systems, equipment management systems, maintenance record systems, and weather forecast apps are integrated and preprocessed. Using power equipment, components, symptoms, faults, strategies, and meteorological data as entities, and the directed relationship types between entity nodes as relations, a Nebula Graph database architecture is constructed. The types of directed relationships between the entity nodes include: Component - belongs to - equipment, equipment - has - symptoms, symptoms - lead to - failure, meteorological data - induces - failure, failure - repair - strategy, symptoms - occur simultaneously - symptoms.

[0006] Preferably, step S1 further includes: Based on a multi-dimensional power equipment system, structured data of power equipment is acquired in real time; Based on the power equipment management system and maintenance record system, the system acquires real-time text data of power equipment ledgers, models, technical parameters, connection topology, historical fault records, maintenance reports, test data and replacement parts records. Using the ApacheTika tool, the system extracts the power equipment text data, performs word segmentation and part-of-speech tagging on the power equipment text data, and identifies and extracts its entities and relationships. Based on the weather forecast APP for the target area, the meteorological data API interface is called regularly, and XML parsing technology is used to extract the meteorological data fields of the target area and convert them into a structured format; Based on structured data, text data, and meteorological data of power equipment, data entities and their corresponding relationships are extracted, entity-relationship-entity triples are designed, and the Nebula Graph graph database architecture is imported to construct a knowledge graph for power equipment fault diagnosis.

[0007] Preferably, step S2 specifically includes: Based on the knowledge graph of power equipment fault diagnosis, all node information is extracted, including node type, attributes, identifier and status, and all edge information is extracted, including edge type, source node of connection, edge weight and relation attributes. Unified node type; The nodes are entities, and the edges are relationships. Calculate the degree of each node in the knowledge graph for power equipment fault diagnosis, establish a node adjacency matrix, generate shortest path information between nodes, and obtain the structured features of the knowledge graph for power equipment fault diagnosis.

[0008] Preferably, step S2 further includes: The knowledge graph for fault diagnosis of power equipment is divided into training set, test set and validation set; Based on the knowledge graph of power equipment fault diagnosis in the training set, a GNN graph neural network model is trained to learn the feature representation of each node, aggregate the features of each node and its neighboring nodes, map each node to a low-dimensional dense vector representation, and capture the structural and semantic information of each layer of nodes in the knowledge graph of power equipment fault diagnosis. By using the stochastic gradient descent algorithm to optimize the distance between adjacent node vector spaces, the learning rate and number of training rounds of the GNN graph neural network model are continuously optimized. The training process is monitored using a knowledge graph of power equipment fault diagnosis in the validation set, and the learning rate and number of network layers of the GNN graph neural network model are adjusted to prevent overfitting. The trained GNN graph neural network model is evaluated using the power equipment fault diagnosis knowledge graph in the test set, and the similarity between nodes and the recall rate are calculated for quantitative evaluation. Based on the trained GNN graph neural network model, the knowledge graph of power equipment fault diagnosis is associated and paired to generate a stable vector that reflects the semantic association of all nodes. Based on generating stable vectors that reflect the semantic relationships of all nodes, vector standardization is performed to establish a node vector database. Based on all node vectors in the node vector database, store them in the Milvus vector database; Using any node vector as input, a k-nearest neighbor query is performed on the Milvus vector database to obtain the k most similar nodes in the vector space and their similarity scores, thus realizing node similarity query and semantic matching.

[0009] Preferably, step S3 specifically includes: Based on the monitoring system, the operating parameters of power equipment are monitored in real time. Combined with the known historical database of power equipment, a stable range of operating parameters of power equipment is selected as the benchmark range for the health of equipment operation. If the real-time operating parameters of the power equipment deviate from the equipment's operating health benchmark range, the power equipment is determined to be in an abnormal state at that moment, and an alarm signal is immediately triggered. By using the API interface, the alarm signal is parsed, the physical entity and abnormal symptoms of the power equipment at that moment are extracted, and the power equipment fault diagnosis knowledge graph is combined to locate the corresponding equipment node and symptom node of the power equipment at that moment. The starting node of the power equipment at that moment is the symptom node of the power equipment fault diagnosis knowledge graph. Using the symptom node of the power equipment fault diagnosis knowledge graph as the starting node of the power equipment at that moment, we traverse all the dependency edges of the node at that moment in the power equipment fault diagnosis knowledge graph to perform multi-hop reasoning, and gradually search for all fault nodes associated with the alarm anomaly to obtain the initial reasoning path set of the power equipment.

[0010] Preferably, step S3 further includes: Based on the initial inference path set of power equipment, the product of the weights of all edges of each inference path is calculated, and the frequency of path occurrence is introduced as a prior constraint for score correction to obtain the confidence score of each inference path. Sort the paths in descending order and select the inference path with the highest confidence score as the main diagnostic result to obtain a list of paths sorted by confidence score; For each path with a confidence score, obtain the control strategy nodes connected to the faulty node, combine the historical success rate of the control strategy with the current equipment status, calculate the feasibility score of each control strategy, and generate a priority-ranked list of recommended power equipment strategies.

[0011] Preferably, step S4 specifically includes: Based on the generated priority-ranked power equipment strategy recommendation list, the result data of the power equipment control strategy execution on site are collected to evaluate its execution effect; Based on evaluating the effectiveness of power equipment control strategies, the edge weights of the corresponding relationships in the power equipment fault diagnosis knowledge graph are automatically adjusted. Establish an effectiveness evaluation mechanism to quantify the success of each implemented control strategy. If the fault is successfully resolved, it is marked as positive feedback, increasing the weight of the corresponding edge in the power equipment fault diagnosis knowledge graph. If the fault is not resolved or new problems are caused, it is marked as negative feedback, decreasing the weight of the corresponding edge in the power equipment fault diagnosis knowledge graph.

[0012] Preferably, step S4 further includes: For newly emerging types of power equipment faults and the effectiveness of control strategies, these are added as new nodes and relationships to the power equipment fault diagnosis knowledge graph, and the structure and node vector representation of the power equipment fault diagnosis knowledge graph are dynamically updated. Based on the newly generated node vectors, the Milvus vector database is updated synchronously to achieve adaptive optimization of the knowledge graph for power equipment fault diagnosis.

[0013] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention proposes a knowledge graph-based scheme for power equipment fault diagnosis and control strategy recommendation. This scheme constructs a dynamically evolving power equipment fault diagnosis knowledge graph, fusing scattered multi-source data into a unified semantic network. Vectorization technology is used to achieve deep semantic matching and rapid retrieval of faults and strategies. When equipment malfunctions, the system can perform intelligent reasoning based on the graph, automatically generating a prioritized strategy list, improving diagnostic efficiency and recommendation accuracy. The system can automatically adjust the weights and structure of the knowledge graph based on feedback from the on-site execution of strategies, enabling continuous learning and optimization of knowledge. This effectively overcomes the shortcomings of traditional static systems' knowledge aging, forming a closed loop of increasingly accurate fault diagnosis and strategy recommendation, comprehensively improving the intelligence level and adaptive capabilities of power equipment operation and maintenance. Attached Figure Description

[0014] Figure 1 A flowchart for a knowledge graph-based method for recommending fault diagnosis and control strategies for power equipment. Detailed Implementation

[0015] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.

[0016] Reference Figure 1 As shown, the knowledge graph-based method for recommending fault diagnosis and control strategies for power equipment includes: S1. Based on a multi-dimensional power equipment system, collect power equipment operating parameters, alarm logs, maintenance records, and meteorological data of the target area; define the node types and relationship types of the knowledge graph; and construct a power equipment fault diagnosis knowledge graph. The multi-dimensional power equipment system includes: a monitoring system, an equipment management system, and a maintenance record system. Step S1 specifically includes: Based on the power equipment monitoring system, the operating parameters of the power equipment are acquired in real time, including current, voltage, temperature, oil chromatography data and partial discharge signals. Based on the power equipment management system, obtain power equipment ledgers, models, technical parameters and connection topologies; Based on the power equipment maintenance record system, historical fault records, maintenance reports, test data and replacement parts records of power equipment are obtained; Based on the weather forecast APP for the target area, obtain meteorological data for the target area, including temperature, humidity, wind speed, rainfall and lightning data; Data from power equipment monitoring systems, equipment management systems, maintenance record systems, and weather forecast apps are integrated and preprocessed. Using power equipment, components, symptoms, faults, strategies, and meteorological data as entities, and the directed relationship types between entity nodes as relations, a Nebula Graph database architecture is constructed. The types of directed relationships between the entity nodes include: Component - belongs to - equipment, equipment - has - symptoms, symptoms - lead to - failure, meteorological data - induces - failure, failure - repair - strategy, symptoms - occur simultaneously - symptoms.

[0017] Step S1 also includes: Based on a multi-dimensional power equipment system, structured data of power equipment is acquired in real time; Based on the power equipment management system and maintenance record system, the system acquires real-time text data of power equipment ledgers, models, technical parameters, connection topology, historical fault records, maintenance reports, test data and replacement parts records. Using the ApacheTika tool, the system extracts the power equipment text data, performs word segmentation and part-of-speech tagging on the power equipment text data, and identifies and extracts its entities and relationships. Based on the weather forecast APP for the target area, the meteorological data API interface is called regularly, and XML parsing technology is used to extract the meteorological data fields of the target area and convert them into a structured format; Based on structured data, text data, and meteorological data of power equipment, data entities and their corresponding relationships are extracted, entity-relationship-entity triples are designed, and the Nebula Graph graph database architecture is imported to construct a knowledge graph for power equipment fault diagnosis.

[0018] When using it, please refer to the steps outlined above: In existing power equipment fault diagnosis technologies, background technologies typically rely on a single system data source, lacking effective integration of multi-dimensional information such as equipment operating parameters, historical maintenance records, and meteorological environment. This results in an incomplete fault diagnosis knowledge system with weak correlations, and traditional relational databases struggle to efficiently handle complex entity relationships, thus hindering the accuracy of fault tracing and early warning. This step integrates multi-dimensional system data from monitoring, management, maintenance, and meteorology to construct a knowledge graph based on Nebula Graph. This enables deep correlations between equipment, symptoms, faults, strategies, and meteorological entities, improving the knowledge coverage and reasoning efficiency of fault diagnosis. It also supports multi-dimensional fault cause analysis and enhances the dynamic diagnosis capabilities for complex faults and meteorological-induced faults.

[0019] S2. Based on the knowledge graph of power equipment fault diagnosis, extract the node features of the knowledge graph of power equipment fault diagnosis, map the nodes into low-dimensional vectors, optimize the vector space distance between adjacent nodes as the objective, store the obtained node vectors in the vector database, and realize the similarity query and semantic matching between nodes. Step S2 specifically includes: Based on the knowledge graph of power equipment fault diagnosis, all node information is extracted, including node type, attributes, identifier and status, and all edge information is extracted, including edge type, source node of connection, edge weight and relation attributes. Unified node type; The nodes are entities, and the edges are relationships. Calculate the degree of each node in the knowledge graph for power equipment fault diagnosis, establish a node adjacency matrix, generate shortest path information between nodes, and obtain the structured features of the knowledge graph for power equipment fault diagnosis.

[0020] Step S2 also includes: The knowledge graph for fault diagnosis of power equipment is divided into training set, test set and validation set; Based on the knowledge graph of power equipment fault diagnosis in the training set, a GNN graph neural network model is trained to learn the feature representation of each node, aggregate the features of each node and its neighboring nodes, map each node to a low-dimensional dense vector representation, and capture the structural and semantic information of each layer of nodes in the knowledge graph of power equipment fault diagnosis. By using the stochastic gradient descent algorithm to optimize the distance between adjacent node vector spaces, the learning rate and number of training rounds of the GNN graph neural network model are continuously optimized. The training process is monitored using a knowledge graph of power equipment fault diagnosis in the validation set, and the learning rate and number of network layers of the GNN graph neural network model are adjusted to prevent overfitting. The trained GNN graph neural network model is evaluated using the power equipment fault diagnosis knowledge graph in the test set, and the similarity between nodes and the recall rate are calculated for quantitative evaluation. Based on the trained GNN graph neural network model, the knowledge graph of power equipment fault diagnosis is associated and paired to generate a stable vector that reflects the semantic association of all nodes. Based on generating stable vectors that reflect the semantic relationships of all nodes, vector standardization is performed to establish a node vector database. Based on all node vectors in the node vector database, store them in the Milvus vector database; Using any node vector as input, a k-nearest neighbor query is performed on the Milvus vector database to obtain the k most similar nodes in the vector space and their similarity scores, thus realizing node similarity query and semantic matching.

[0021] When using it, please refer to the steps outlined above: Traditional power equipment fault diagnosis methods often rely on expert experience or rule-based systems, which suffer from limited knowledge coverage, high dependence on manual intervention, and insufficient semantic association mining. These methods struggle to effectively handle the complex structured and unstructured information in knowledge graphs, leading to insufficient node feature extraction, low semantic matching accuracy, and a lack of vectorized representation capabilities for deep inter-node relationships. This step utilizes graph neural networks to perform low-dimensional vectorized representation of knowledge graph nodes, aiming to optimize the vector space distance between adjacent nodes. It effectively integrates node structure and semantic information to generate stable vectors with semantic associations. By leveraging a vector database, it achieves efficient similarity querying and semantic matching, improving the accuracy, scalability, and automation level of fault diagnosis.

[0022] S3. If the monitoring system detects an abnormality in the power equipment, an alarm signal is immediately triggered. Starting from the alarm node, multi-hop reasoning is performed in the power equipment fault diagnosis knowledge graph to select high-confidence paths as diagnostic conclusions and generate a priority-ranked list of recommended power equipment strategies. Step S3 specifically includes: Based on the monitoring system, the operating parameters of power equipment are monitored in real time. Combined with the known historical database of power equipment, a stable range of operating parameters of power equipment is selected as the benchmark range for the health of equipment operation. If the real-time operating parameters of the power equipment deviate from the equipment's operating health benchmark range, the power equipment is determined to be in an abnormal state at that moment, and an alarm signal is immediately triggered. By using the API interface, the alarm signal is parsed, the physical entity and abnormal symptoms of the power equipment at that moment are extracted, and the power equipment fault diagnosis knowledge graph is combined to locate the corresponding equipment node and symptom node of the power equipment at that moment. The starting node of the power equipment at that moment is the symptom node of the power equipment fault diagnosis knowledge graph. Using the symptom node of the power equipment fault diagnosis knowledge graph as the starting node of the power equipment at that moment, we traverse all the dependency edges of the node at that moment in the power equipment fault diagnosis knowledge graph to perform multi-hop reasoning, and gradually search for all fault nodes associated with the alarm anomaly to obtain the initial reasoning path set of the power equipment.

[0023] Step S3 also includes: Based on the initial inference path set of power equipment, the product of the weights of all edges of each inference path is calculated, and the frequency of path occurrence is introduced as a prior constraint for score correction to obtain the confidence score of each inference path. Sort the paths in descending order and select the inference path with the highest confidence score as the main diagnostic result to obtain a list of paths sorted by confidence score; For each path with a confidence score, obtain the control strategy nodes connected to the faulty node, combine the historical success rate of the control strategy with the current equipment status, calculate the feasibility score of each control strategy, and generate a priority-ranked list of recommended power equipment strategies.

[0024] When using it, please refer to the steps outlined above: In existing technologies, fault diagnosis of power equipment often relies on manual experience or single-rule judgment, lacking the ability to reason deeply about complex and related faults. Strategy recommendations often ignore real-time equipment status and historical execution effects, resulting in low diagnostic accuracy and delayed response. This step introduces multi-hop reasoning and confidence assessment based on knowledge graphs, which can automatically uncover complex relationships between abnormal symptoms and potential faults. It integrates path frequency and historical strategy data to generate a priority recommendation list, improving the accuracy of diagnosis and the feasibility of strategies, and realizing a leap from passive alarm to intelligent decision-making.

[0025] S4. Based on the priority-ranked power equipment strategy recommendation list, evaluate the execution effect of the power equipment strategy recommendation list, adjust the edge weights of the corresponding relationships in the power equipment fault diagnosis knowledge graph, dynamically update the structure and node vectors of the power equipment fault diagnosis knowledge graph, and realize adaptive optimization of the power equipment fault diagnosis knowledge graph. Step S4 specifically includes: Based on the generated priority-ranked power equipment strategy recommendation list, the result data of the power equipment control strategy execution on site are collected to evaluate its execution effect; Based on evaluating the effectiveness of power equipment control strategies, the edge weights of the corresponding relationships in the power equipment fault diagnosis knowledge graph are automatically adjusted. Establish an effectiveness evaluation mechanism to quantify the success of each implemented control strategy. If the fault is successfully resolved, it is marked as positive feedback, increasing the weight of the corresponding edge in the power equipment fault diagnosis knowledge graph. If the fault is not resolved or new problems are caused, it is marked as negative feedback, decreasing the weight of the corresponding edge in the power equipment fault diagnosis knowledge graph.

[0026] Step S4 also includes: For newly emerging types of power equipment faults and the effectiveness of control strategies, these are added as new nodes and relationships to the power equipment fault diagnosis knowledge graph, and the structure and node vector representation of the power equipment fault diagnosis knowledge graph are dynamically updated. Based on the newly generated node vectors, the Milvus vector database is updated synchronously to achieve adaptive optimization of the knowledge graph for power equipment fault diagnosis.

[0027] When using it, please refer to the steps outlined above: In existing power equipment fault diagnosis systems, background technologies typically rely on static knowledge graphs and fixed rules for strategy recommendation. This lack of real-time feedback and adaptive optimization mechanisms for strategy execution results in delayed knowledge updates, insufficient recommendation accuracy, and an inability to dynamically adjust the knowledge graph structure to adapt to new fault types. This step improves the accuracy and real-time performance of strategy recommendations by introducing dynamic weight adjustment based on execution results and an adaptive optimization mechanism for the knowledge graph. This enhances the system's adaptability to complex faults. Simultaneously, continuous learning optimizes the knowledge base, achieving an intelligent and precise upgrade in power equipment fault diagnosis.

[0028] Based on the above, the specific implementation method is as follows: The scheme was implemented using a 110kV main transformer in a substation as a specific example. The monitoring system of the transformer was used to collect oil temperature, winding temperature, oil chromatography and partial discharge data in real time. The equipment management system is used to obtain its model, technical parameters, and connection topology with circuit breakers and disconnect switches. Historical maintenance reports can be extracted from maintenance records, such as the seal replacement due to oil leakage in the casing in 2023. By accessing local meteorological data via API, including thunderstorm and high temperature warnings, the main transformer, bushing, acetylene exceedance, oil temperature overheating, internal discharge, seal replacement strategy, thunderstorm weather, etc. are defined as entities. And according to the relationship of component-belonging-equipment, symptom-cause-failure, meteorological data-inducing-failure, an interconnected knowledge graph is built in Nebula Graph. By using graph neural networks to learn the graph, all these entities are transformed into low-dimensional vectors with semantic information and stored in the Milvus vector database, making the vectors of the two nodes, namely, oil temperature too high and cooler failure, very close in space. One day, the monitoring system detected that the oil temperature and acetylene content of the main transformer exceeded the standard at the same time, and immediately triggered an alarm. After analyzing the alarm signal, the system located the two symptom nodes of excessive oil temperature and excessive acetylene in the knowledge graph. Starting from these, the system performed multi-hop reasoning. First, it found the fault nodes directly related to these two symptoms, such as internal discharge and insulation aging. Then it found that the thunderstorm weather node was also connected to internal discharge through the induction relationship. The system calculates the confidence level of each inference path, combines the path weight and historical occurrence frequency, and finally determines that internal discharge is the most likely fault at present with a confidence level of 92%. The strategy nodes in the spectrum connected to internal discharge, such as oil chromatography tracking analysis and power outage internal inspection, are activated. The system then combines the historical success rate of these strategies with the current equipment load to generate a priority recommendation list, suggesting that the operation and maintenance personnel immediately arrange oil chromatography tracking and prepare for power outage inspection. Based on the system's recommendation, the on-site maintenance personnel executed the power outage internal inspection strategy, successfully discovered and repaired the fault point of the tap changer contact erosion. After the system collected feedback on the successful execution of this strategy, it automatically strengthened the weight of the relationship edge in the knowledge graph from the internal discharge fault to the power outage internal inspection strategy. The tap changer contact erosion mentioned in this maintenance record was added as a new fault node, and its relationship with the acetylene exceedance symptom and contact replacement strategy was added as a new triple to the knowledge graph. This update of the graph structure immediately triggered the GNN graph neural network model to perform incremental training, regenerated the vector representation of all nodes, and simultaneously updated the Milvus vector database. After multiple such cycles, the system achieved self-advancing. When the main transformer of another substation showed excessive acetylene levels accompanied by abnormal partial discharge signals, the system could not only quickly diagnose the internal discharge fault, but also, based on updated knowledge, recommend with higher confidence the specific strategy of focusing on checking the tap changer. The newly accumulated contact erosion fault vectors and the historical arc burn fault vectors form an aggregation in the vector space, enabling the system to identify potential similar fault types through semantic matching via vector similarity queries even when faced with different combinations of symptoms in the future. This demonstrates a powerful generalized diagnostic and predictive warning capability.

[0029] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention. The scope of protection claimed by the appended claims and their equivalents is defined.

Claims

1. A knowledge graph-based method for recommending fault diagnosis and control strategies for power equipment, characterized in that, include: S1. Based on the multi-dimensional system of power equipment, collect the operating parameters, alarm logs, maintenance records and meteorological data of the target area of ​​power equipment, define the node types and relationship types of the knowledge graph, and construct a knowledge graph for power equipment fault diagnosis. S2. Based on the knowledge graph of power equipment fault diagnosis, extract the node features of the knowledge graph of power equipment fault diagnosis, map the nodes into low-dimensional vectors, optimize the vector space distance between adjacent nodes as the objective, store the obtained node vectors in the vector database, and realize the similarity query and semantic matching between nodes. S3. If the monitoring system detects an abnormality in the power equipment, an alarm signal is immediately triggered. Starting from the alarm node, multi-hop reasoning is performed in the power equipment fault diagnosis knowledge graph to select high-confidence paths as diagnostic conclusions and generate a priority-ranked list of recommended power equipment strategies. S4. Based on the priority-ranked power equipment strategy recommendation list, evaluate the execution effect of the power equipment strategy recommendation list, adjust the edge weights of the corresponding relationships in the power equipment fault diagnosis knowledge graph, dynamically update the structure and node vectors of the power equipment fault diagnosis knowledge graph, and realize adaptive optimization of the power equipment fault diagnosis knowledge graph.

2. The method for recommending power equipment fault diagnosis and control strategies based on knowledge graphs according to claim 1, characterized in that, Step S1 specifically includes: Based on the power equipment monitoring system, the operating parameters of the power equipment are acquired in real time, including current, voltage, temperature, oil chromatography data and partial discharge signals. Based on the power equipment management system, obtain power equipment ledgers, models, technical parameters and connection topologies; Based on the power equipment maintenance record system, historical fault records, maintenance reports, test data and replacement parts records of power equipment are obtained; Based on the weather forecast APP for the target area, obtain meteorological data for the target area, including temperature, humidity, wind speed, rainfall and lightning data; Data from power equipment monitoring systems, equipment management systems, maintenance record systems, and weather forecast apps are integrated and preprocessed. Using power equipment, components, symptoms, faults, strategies, and meteorological data as entities, and the directed relationship types between entity nodes as relations, a Nebula Graph database architecture is constructed. The types of directed relationships between the entity nodes include: Component - belongs to - equipment, equipment - has - symptoms, symptoms - lead to - failure, meteorological data - induces - failure, failure - repair - strategy, symptoms - occur simultaneously - symptoms.

3. The method for recommending fault diagnosis and control strategies for power equipment based on knowledge graphs according to claim 2, characterized in that, Step S1 also includes: Based on a multi-dimensional power equipment system, structured data of power equipment is acquired in real time; Based on the power equipment management system and maintenance record system, the system acquires text data in real time, including power equipment ledgers, models, technical parameters, connection topology, historical fault records, maintenance reports, test data, and replacement parts records. Using the Apache Tika tool, the system extracts the power equipment text data, performs word segmentation and part-of-speech tagging on the power equipment text data, and identifies and extracts its entities and relationships. Based on the weather forecast APP for the target area, the meteorological data API interface is called regularly, and XML parsing technology is used to extract the meteorological data fields of the target area and convert them into a structured format; Based on structured data, text data, and meteorological data of power equipment, data entities and their corresponding relationships are extracted, entity-relationship-entity triples are designed, and the Nebula Graph graph database architecture is imported to construct a knowledge graph for power equipment fault diagnosis.

4. The method for recommending power equipment fault diagnosis and control strategies based on knowledge graphs according to claim 3, characterized in that, Step S2 specifically includes: Based on the knowledge graph of power equipment fault diagnosis, all node information is extracted, including node type, attributes, identifier and status, and all edge information is extracted, including edge type, source node of connection, edge weight and relation attributes. Unified node type; The nodes are entities, and the edges are relationships. Calculate the degree of each node in the knowledge graph for power equipment fault diagnosis, establish a node adjacency matrix, generate shortest path information between nodes, and obtain the structured features of the knowledge graph for power equipment fault diagnosis.

5. The method for recommending power equipment fault diagnosis and control strategies based on knowledge graphs according to claim 4, characterized in that, Step S2 also includes: The knowledge graph for fault diagnosis of power equipment is divided into training set, test set and validation set; Based on the knowledge graph of power equipment fault diagnosis in the training set, a GNN graph neural network model is trained to learn the feature representation of each node, aggregate the features of each node and its neighboring nodes, map each node to a low-dimensional dense vector representation, and capture the structural and semantic information of each layer of nodes in the knowledge graph of power equipment fault diagnosis. By using the stochastic gradient descent algorithm to optimize the distance between adjacent node vector spaces, the learning rate and number of training rounds of the GNN graph neural network model are continuously optimized. The training process is monitored using a knowledge graph of power equipment fault diagnosis in the validation set, and the learning rate and number of network layers of the GNN graph neural network model are adjusted to prevent overfitting. The trained GNN graph neural network model is evaluated using the power equipment fault diagnosis knowledge graph in the test set, and the similarity between nodes and the recall rate are calculated for quantitative evaluation. Based on the trained GNN graph neural network model, the knowledge graph of power equipment fault diagnosis is associated and paired to generate a stable vector that reflects the semantic association of all nodes. Based on generating stable vectors that reflect the semantic relationships of all nodes, vector standardization is performed to establish a node vector database. Based on all node vectors in the node vector database, store them in the Milvus vector database; Using any node vector as input, a k-nearest neighbor query is performed on the Milvus vector database to obtain the k most similar nodes in the vector space and their similarity scores, thus realizing node similarity query and semantic matching.

6. The method for recommending fault diagnosis and control strategies for power equipment based on knowledge graphs according to claim 1, characterized in that, Step S3 specifically includes: Based on the monitoring system, the operating parameters of power equipment are monitored in real time. Combined with the known historical database of power equipment, a stable range of operating parameters of power equipment is selected as the benchmark range for the health of equipment operation. If the real-time operating parameters of the power equipment deviate from the equipment's operating health benchmark range, the power equipment is determined to be in an abnormal state at that moment, and an alarm signal is immediately triggered. By using the API interface, the alarm signal is parsed, the physical entity and abnormal symptoms of the power equipment at that moment are extracted, and the power equipment fault diagnosis knowledge graph is combined to locate the corresponding equipment node and symptom node of the power equipment at that moment. The starting node of the power equipment at that moment is the symptom node of the power equipment fault diagnosis knowledge graph. Using the symptom node of the power equipment fault diagnosis knowledge graph as the starting node of the power equipment at that moment, we traverse all the dependency edges of the node at that moment in the power equipment fault diagnosis knowledge graph to perform multi-hop reasoning, and gradually search for all fault nodes associated with the alarm anomaly to obtain the initial reasoning path set of the power equipment.

7. The method for recommending power equipment fault diagnosis and control strategies based on knowledge graphs according to claim 6, characterized in that, Step S3 also includes: Based on the initial inference path set of power equipment, the product of the weights of all edges of each inference path is calculated, and the frequency of path occurrence is introduced as a prior constraint for score correction to obtain the confidence score of each inference path. Sort the paths in descending order and select the inference path with the highest confidence score as the main diagnostic result to obtain a list of paths sorted by confidence score; For each path with a confidence score, obtain the control strategy nodes connected to the faulty node, combine the historical success rate of the control strategy with the current equipment status, calculate the feasibility score of each control strategy, and generate a priority-ranked list of recommended power equipment strategies.

8. The method for recommending power equipment fault diagnosis and control strategies based on knowledge graphs according to claim 7, characterized in that, Step S4 specifically includes: Based on the generated priority-ranked power equipment strategy recommendation list, the result data of the power equipment control strategy execution on site are collected to evaluate its execution effect; Based on the evaluation of the effectiveness of power equipment control strategies, the weights of the corresponding edges in the knowledge graph for power equipment fault diagnosis are automatically adjusted.

9. The method for recommending power equipment fault diagnosis and control strategies based on knowledge graphs according to claim 8, characterized in that, Step S4 also includes: Establish an effectiveness evaluation mechanism to quantify the success of each implemented control strategy. If the fault is successfully resolved, it is marked as positive feedback, increasing the weight of the corresponding edge in the power equipment fault diagnosis knowledge graph. If the fault is not resolved or new problems are caused, it is marked as negative feedback, decreasing the weight of the corresponding edge in the power equipment fault diagnosis knowledge graph.

10. The method for recommending power equipment fault diagnosis and control strategies based on knowledge graphs according to claim 8, characterized in that, Step S4 also includes: For newly emerging types of power equipment faults and the effectiveness of control strategies, these are added as new nodes and relationships to the power equipment fault diagnosis knowledge graph, and the structure and node vector representation of the power equipment fault diagnosis knowledge graph are dynamically updated. Based on the newly generated node vectors, the Milvus vector database is updated synchronously to achieve adaptive optimization of the knowledge graph for power equipment fault diagnosis.

Images