基于BERT-BiLSTM-CRF模型的开关柜设备运维实体抽取方法研究

Research on Extraction Method of Switch Cabinet Equipment Operation and Maintenance Entity Based on BERT-BiLSTM-CRF Model

开关柜设备的检修过程繁琐,需要人工手持设备对其进行检测,效率低下。为了提高开关柜设备的运维工作效率,本文将实体抽取技术运用于开关柜设备运维。首先将知网作为语料库爬取相关文献,然后对爬取的文献进行标注和分类,选择了三种实体抽取模型:使用训练集和验证集对模型进行训练和验证、使用测试集对模型的抽取能力进行测试并计算相关评价指标,最后对三种不同的实体抽取模型进行了比较,结果表明,采用BERT预训练模型的关注上下文信息的BERT-BiLSTM-CRF模型F1分数最高,可达到97.24,与BiLSTM-CRF模型和HMM模型相比分别提高了1.13和3.15。

The switch cabinet plays an important role in the power system.However,the maintenance process of the switch cabinet is often tedious,and requires manual handheld devices for detection,is inefficient.The troubleshooting work also puts forward a higher ability requirement for the operation and maintenance personnel.In order to improve the efficiency of operation and maintenance of switch cabinet,this paper studies the application of entity extraction technology to the operation and maintenance of switch cabinet.This paper first uses CNKI as the corpus to crawl relevant documents,then labels and classifies the crawled documents,selects three entity extraction models,uses the training set verification set to train and verify the model,uses the test set to test the model's extraction ability,calculates relevant evaluation indicators,and compares three different entity extraction models,The BERT-BiLSTM-CRF model,which adopts the BERT pre training model and can focus on context information,has the highest F1 score,reaching 97.24,which is 1.13 and 3.15 higher than BiLSTM-CRF model and HMM model respectively.