Research And Implementation Of Power Protection Fault Dianosis Based On Knowledge Graph

With the increasing demand for electricity in recent years,more attentions have been paid to power security.In order to reduce potential failures and avoid the impact of power equipment failures,the electric power department often adopts rule-based maintenance and quick maintenance.However,this method still has some limitations: 1)The large amount of maintenance tasks are time-consuming and lack of targeted maintenance for high-risk faults;2)The post-event maintenance is not efficient,and the diverse and complicate fault data could lead to the long-time analysis and decision-making;3)There is no way to know the scope of the possible failure,and for the equipment that may be affected,it is difficult to prevent in advance.To address the above problems,this thesis proposes a method of using the knowledge graph to handle the power fault diagnosis.The massive power data is refined into power knowledge graph,which is the brain of the diagnosis system,can provide more accurate and interpretable solutions,and reduce the dependence on experts to some extent.The main contributions of this thesis are as follows:(1)Constructing a knowledge graph for power maintenance and fault diagnosis.Combined with the features of the power industry data,the power protection ontology is manually built according to the actual demands,and then the multi-source data is integrated for knowledge extraction.The knowledge extraction of structured data adopts the rule-based extraction method;The knowledge extraction of semi-structured data is to analyze the structure of power protection documents and write extraction rules,and supplement relevant semi-structured data from encyclopedia websites combined with crawler framework.The knowledge graph covers the important information required for power protection,including equipment,stations,personnel,orders,etc.This thesis further designs a regular incremental knowledge update method to avoid the impact of knowledge update on normal service.On the real-world power protection data,the extracted knowledge graph contains 87,720 entities,1,004,573 value attributes and 98,946 triples.(2)Proposing a risk warning model based on knowledge graph.According to the node’s own risk factors and the risk impact brought by other nodes,the model comprehensively considers the possibility of failure of the current node.Namely,if the failure risk of a node equipment increases significantly,the equipment related to this node will also be affected by this node.The model can realize the propagation analysis of risk impact and calculate the risk value of each node according to the weight.Moreover,the model can evaluate the status of power protection equipment in real time,judge possible faults,and notify relevant personnel for maintenance in advance to avoid abnormal events.The experimental results on the real-world historical data of power protection show that the accuracy of the power protection risk warning model designed in this thesis is 96%.(3)Proposing a method for power maintenance and fault diagnosis based on knowledge graph.Based on the historical fault records and fault plan documents,the fault knowledge graph is constructed,and the key nodes such as equipment and personnel in the fault knowledge graph are associated with the power protection knowledge graph to support the rapid positioning and accurate recommend service in case of fault.In this thesis,a joint extraction model is used to extract fault knowledge from unstructured documents.The fault knowledge is saved according to the hierarchical structure.Through knowledge fusion,it is matched with the power protection knowledge graph to ensure the knowledge consistency of the two graphs.In order to recommend similar historical faults when the faults occurs,fault knowledge is encoded into low dimensional vector representation in a specific way,and the similarity of two faults is calculated through the similarity between vectors.The fault knowledge graph contains 176,435 entities and 31,573 triples,which belong to 16 equipment types and 185 fault types respectively.The recall index of fault knowledge recommend can reach 90%and NDCG index can reach 88%,which can realize more accurate push.Based on the above work,this thesis designs and implements a fault diagnosis system based on knowledge graph.Currently,the system has been successfully applied in several major power guarantee scenarios,with stable operation and good application benefits.