浙江电力2025，Vol.44Issue(7)：24-32,9.DOI:10.19585/j.zjdl.202507003
基于BERT-MRC的电网现场作业文本关键实体识别方法
A BERT-MRC-based method for key entity recognition in power grid field operation texts
摘要
Abstract
Ensuring production safety in power grids requires effective control and inspection of field operations,where accurate recognition of key equipment entities in operation texts serves as the foundation for intelligent control and inspection.However,existing power entity recognition methods rely heavily on large volumes of manually anno-tated text data to train models,making them difficult to apply to field operation texts,which are generated rapidly,exist in large quantities,and often involve nested entities and other complex relationships.Based on an analysis of the characteristics of power grid field operation texts,this paper proposes a key entity recognition method tailored for risk control and inspection of power grid operations.The method enhances recognition performance while signifi-cantly reducing the model's dependence on labeled data.First,bidirectional encoder representations from trans-formers(BERT)are employed to obtain text data vectors that incorporate contextual features.Then,leveraging BERT-machine reading comprehension(MRC),the entity recognition task is reformulated as an MRC task to build the model.Finally,a few-short learning(FSL)method based on the Noisy Student is applied to iteratively train the model,greatly reducing the reliance on labeled data.Experiments conducted on real-world power grid field opera-tion texts demonstrate the effectiveness of the proposed method.关键词
实体识别/机器阅读理解/电网现场作业/风险管控稽查/BERT/小样本学习Key words
entity recognition/MRC/grid field operation/risk control and inspection/BERT/FSL引用本文复制引用
费正明,袁可为,黄弘扬,张亦翔,尹凡,周辉,罗华峰..基于BERT-MRC的电网现场作业文本关键实体识别方法[J].浙江电力,2025,44(7):24-32,9.基金项目
国家电网华东分公司科技项目(520800230008) （520800230008）
