The Construction Methods Of Knowledge Graph For Power Communication Equipment Fault Information

With the rapid development of smart grid and energy Internet,massive heterogeneous power communication devices are constantly connected to the power infrastructure,which greatly promotes the digitalization of power grid.However,the failure of power communication devices may cause grid disturbance and threaten grid security.The huge number of devices makes the fault diagnosis data grow exponentially,and due to the different functions of the devices,the faults vary widely and have different impacts,leading to poor supervision effectiveness and difficulty in locating fault information during the operation and maintenance process.The existing power fault database has redundant structure,complex association and difficult to analyze,and the fault information retrieval efficiency is low and accuracy is poor,which greatly hinders the intelligent and digital development of the power grid.In view of the above problems,this article investigates the knowledge extraction method of power communication equipment faults,constructs the fault knowledge graph of power communication equipment,and realizes the visual management of power communication equipment fault information.And based on this,we study the fast retrieval method of fault information,realize the accurate positioning of power communication equipment faults,and provide technical support for the intelligent operation and maintenance of power grid.For the problem of unclear relationship and incomplete content of fault entities in power communication equipment,this article proposes a word embedding transformation of fault data based on ALBERT-Bi LSTM pre-training and shared coding model to solve the problem of text information loss.And the deep mining of corpus features through parameter layer sharing draws on the coupling between the quantum task of contextual relationship enhancement to realize the joint coding of entity relationship information.The proposed MEKJE method performs relationship extraction among multiple fault entities for equipment faults to achieve joint acquisition of fault entities and relationships.For the problem of lack in domain database and feature learning for fault retrieval and location of power communication equipment,this article reconstructs and designs the ontology layer of power knowledge graph with grid dispatching business as the core to form a scientific and reasonable fault business dispatching ontology layer.And the knowledge graph construction software is used to visualize the fault knowledge graph of power grid communication equipment.Based on this,PF2 RM is proposed to improve the query effect of fault retrieval and location of power communication equipment.Among them,the PFR method enhances the aggregation degree of fault entities by designing graph neighbor fault entity clusters to solve the cold start fault location problem.Meanwhile,the UPRR method achieves accurate prediction of conventional fault retrieval by forming user retrieval subgraphs of past states and current states.In this article,various classical knowledge extraction models such as MEKJE and Bi LSTM-CRF are evaluated in comparison on SDH dataset and Cluener dataset,respectively.The F1 score of this model reaches 78.6%,with a 5% improvement in accuracy and 8%improvement in recall rate.Meanwhile,this article evaluates PF2 RM against various retrieval algorithms such as Ripple Net based on the knowledge graph of power communication equipment faults.The model improves NDCG and AUC by 2-10%,and has 10%,6%,8%and 2% improvement in accuracy,diversity,recall and F1 score,which provides an effective solution for digitalization and intelligence of power grid.