Relationship Extraction of Bushing Failure from Chinese Corpus Based on BERT-FC Model

As a crucial component of transformers, bushings operate in harsh environments for prolonged periods, leading to frequent failures and significant economic losses. With the digital transformation of power system, deep learning algorithms have been applied to failure diagnosis of power equipment. However, the subjectivity and unstructured nature of transformer bushing failure analysis text hinder automatic machine detection, affecting the maintenance efficiency of the transformer bushings. To address these issues, this paper utilizes the BERT-FC natural language processing model to extract information from bushing failure text. Firstly, equipment and failure information are isolated and labeled from the literature and reports, creating a dataset of equipment information and failure descriptions. Then, based on the textual representation of the dataset, the key weight parameters in BERT-FC are iteratively adjusted to form a model that identifies relationships among bushing equipment structure, failure cause, and classification. Finally, traditional deep learning models are compared with the model proposed in this paper, and the effectiveness of the presented method is evaluated by precision, recall, and $F$1 scores, reaching 92.8%, 92.2%, and 92.6%, respectively.