Knowledge Graph Construction for Secondary Equipment Fault Diagnosis Based on Graph Attention

As one of the infrastructure foundations of social and economic development, the power system has a large scale and rich relationships among its complex and diverse elements. The construction of a knowledge graph for the power system aims to model the system through graph models, mine knowledge relationships within the system, and provide more comprehensive and in-depth information support for the operation, management, and optimization of the power system. However, most existing relationship extraction models only focus on the relationships between adjacent words, without considering the relationships between non-adjacent words. To address this issue, a security risk relationship extraction model based on graph attention is proposed, which propagates information between non-adjacent words through graph mechanisms. Meanwhile, to reduce the noise problem caused by the propagation of graph information, an objective function is constructed using the information bottleneck method, which maximizes the retention of useful information and compresses useless information. On a real-world dataset in the power domain, the proposed model outperforms several advanced models currently available.