Condition Diagnosis of Composite Insulator Based on Knowledge Graph

Composite insulators have been widely used in power transmission system for the lightweight, high strength, and great hydrophobicity. However, the aging of the polymers in composite insulators is a key issue affecting the reliable operation, posing challenges for the condition diagnosis of composite insulators. Knowledge graph is an emerging technique for structuring existing experience and knowledge, which shows promising prospects on maintenance efficiency enhancement for composite insulators. This article introduces the condition diagnosis approach for composite insulators based on knowledge graph. First, the annotated corpus of composite insulator was constructed based on the technical reports, handbooks, open published articles, and standards. This work defined nine types of entities and nine types of relations. Then, the a lite bidirectional encoder representations from transformer (ALBERT)–bidirectional long short-term memory (BiLSTM)–conditional random field (CRF) model and the ALBERT-bidirectional gated recurrent unit (BiGRU)–attention model are employed to realize the entities and relations recognition and the ${F}1$ scores of which reach 92.49% and 92.36%, respectively, exhibiting better performance than any other models. At last, the first knowledge graph on composite insulator was constructed. There are 820 entity-relation–entity triplets, covering the basic information, environmental conditions, test methods, fault mechanisms, and so on. This work will provide an artificial intelligence tool for condition diagnosis of composite insulators.