Knowledge Graph Completion for Power Grid Main Equipment Using Pretrained Language Models.

The safe and stable operation of power systems relies on the timely diagnosis of defects in power grid equipment. To achieve this, knowledge graph (KG) can be used to model power grid equipment defect knowledge, and knowledge graph embedding (KGE) can be utilized to embed KG into low dimensional vector spaces for deep learning models. However, pre-trained language model-based KGE methods may not perform as well as structure-based methods due to their limitations in explicitly representing domain-specific knowledge and supplementary information about entities. In this study, a hybrid KGE model called PLMSM was proposed to address this issue. PLMSM combines pre-trained language models with structure-based models to input entities and their supplementary information into a pre-trained language model to obtain their embeddings, which are then combined with the embeddings generated by a structure-based model for entity completion tasks. The model was optimized through efficient negative sampling and addressed the issue of inaccurate predictions caused by long-tail entities in the power grid defects KG. The experimental results showed that PLMSM achieved good performance in Entity completion tasks on the power grid equipment defects KG. This proposed model has potential applications in power grid equipment defect diagnosis and maintenance.