Chinese Named Entity Recognition of Transformer Bushing Faults Based on ALBERT-BiLSTM-CRF

Lijun Jin, Yufang Zhang, Zhikang Yuan, Shuojie Gao, Miaosong Gu, Xianhui Liu, Yingyao Zhang

Abstract: With the digital transformation and upgrading of new power system, the demand for intelligent functionalities in traditional power equipment has increased. As one of the key equipment of transformers, bushings still encounter issues such as unpredictable accidents. There is an urgent need to employ digital methods to enhance the operation and maintenance efficiency of transformer bushings. Currently, operators have accumulated a substantial amount of experience on transformer bushing operation in the form of text. However, fault description in this text is highly specialized and varies from person to person, posing challenges for automated fault recognition by machines. This paper proposes a Chinese transformer bushing fault recognition method based on ALBERT-BiLSTM-CRF. First, a dataset of transformer bushing faults is constructed based on published papers, fault reports, analyses, records and standards. Next, the text is converted into a character sequence, and the semantic information is encoded through the ALBERT model to obtain the vector of the input sequence. Then, the BiLSTM network is used to capture contextual information. Finally, CRF is employed to obtain the dependencies between adjacent labels and predict the label of each character. The results demonstrate that ALBERT-BiLSTM-CRF effectively recognizes entities in Chinese bushing fault corpus, achieving an impressive F1 score of 96.60%, surpassing commonly used models.

External IDs:doi:10.1109/tia.2024.3523881