Timely detection and resolution of power equipment failures are vital for ensuring the stability and reliability of electrical grids. This paper presents a knowledge graph completion model (KGCM) that integrates multi-task learning with pre-trained models to enhance fault-handling knowledge graph representations. Entities and relations are encoded with a pre-trained model and refined through Translating Embeddings (TransE)-based structural constraints, while a Siamese network is introduced to improve feature extraction and robustness by predicting missing triplet components. The proposed framework was validated on a dataset from the State Grid Tianjin as well as standard benchmarks. Experimental results show that KGCM consistently outperforms the baseline Language Model Knowledge Embedding with TransE (LMKE-TransE) model in terms of Mean Reciprocal Rank (MRR) and Hits at N (Hits@N) metrics. In practical fault-action prediction, KGCM demonstrates superior accuracy, enabling more reliable diagnosis and handling of power grid faults. These findings confirm the effectiveness of combining pre-trained models, structural fine-tuning, and multi-task feature extraction for knowledge graph completion, and highlight the model’s potential to enhance automation and dependability in power grid fault management.
Huang et al. (Tue,) studied this question.