TCEKG: A Temporal and Causal Event Knowledge Graph for Power Distribution Network Fault Diagnosis

Power distribution network fault diagnosis is important to ensure the smooth operation of the power grid. Line-level fault diagnosis techniques for distribution networks are almost non-existent, and some techniques using knowledge graphs are limited to some static entities, ignoring both more specific lines and the dynamic logic of event evolution. Currently, the Distribution Management System (DMS) is widely used in the country, and a large number of fault tickets exist in it. In this paper, we propose a practical event knowledge representation for distribution network fault diagnosis. Specifically, we design a series of abstract models to organize fault tickets and construct a Temporal and Causal Event Knowledge Graph (TCEKG), which can record temporal causal information and can be well integrated with domestic DMS. In addition, we design two TCEKG-based Fault Diagnosis Models (FDMs). To make the FDM more focused on recent events, we design a temporal decay mechanism for filtering events. Extensive experiments and ablation studies on four real-world datasets show that our TCEKG-based FDMs can effectively perform the distribution network fault diagnosis task by using TCEKG and are efficient in a single inference.