火电厂故障诊断文本的实体抽取模型构建  

Construction of entity extraction model for fault diagnosis text in thermal power plants

作　　者：陈宏[1,2] 王云博 穆思澎 陈阳 CHEN Hong;WANG Yunbo;MU Sipeng;CHEN Yang(School of Mechanical and Power Engineering,Zhengzhou University,Zhengzhou 450001,China;Department of Mechanical and Electrical Engineering,Hami Vocational and Technical College,Hami 839099,China;Ulster College,Shaanxi University of Science and Technology,Xi’an 710016,China)

机构地区：[1]郑州大学机械与动力工程学院,郑州450001 [2]哈密职业技术学院机电系,新疆哈密839099 [3]陕西科技大学阿尔斯特学院,西安710016

出　　处：《重庆理工大学学报（自然科学）》2024年第11期206-212,共7页Journal of Chongqing University of Technology：Natural Science

基　　金：国家自然科学基金项目(52275138)。

摘　　要：针对火电厂故障诊断领域文本存在实体边界模糊、文本特征不够充分、模型识别效果不明显等问题,提出一种改进BERT-BiLSTM-CRF故障诊断领域文本实体识别模型。为了提高BERT模型在中文语境下的性能,对模型参数进行改进,使用对抗训练方法提高模型精度,使模型F1值提高0.0206。对已构建的数据集进行实体命名识别实验,实验结果表明:改进BERT-BiLSTM-CRF实体识别模型在数据集上的F1值达到0.9016,相较于其他模型F1值有所提升,验证了该模型的有效性。To address such issues as blurred entity boundaries,insufficient text features,and unremarkable model recognition effects in the field of fault diagnosis for thermal power plants,we propose a text entity recognition model based on improved BERT-BiLSTM-CRF for fault diagnosis.Entity naming recognition experiments are conducted on a newly built dataset.Our results indicate the entity recognition model based on the improved BERT-BiLSTM-CRF achieves an F 1 score of 0.9016,which is superior to those of other models,validating the effectiveness of our model.To enhance the performance of the BERT model in a Chinese context,model parameters are optimized,and adversarial training methods are employed to improve model accuracy,which is up by 0.0206 in F 1 score.

关 键 词：实体命名识别 预训练语言模型 火电厂 故障诊断 对抗训练 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术] TK268[自动化与计算机技术—计算机科学与技术]