摘要
针对传统深度学习模型特征提取不全面、静态词向量语义表示质量低等问题,提出基于ChineseBERT-BiSRU-MCNN-AT的电网故障文本分类模型。ChineseBERT模型训练过程融入了字形和拼音信息,结合词的上下文信息进行动态学习,解决一词多义问题。多尺度语义协同模块BiSRU-MCNN提取故障文本局部语义和全局序列特征,确保提取特征的全面性,软注意力层赋予模型识别关键词的能力。通过对电网故障文本数据集进行实验,其结果表明ChineseBERT-BiSRU-MCNN-AT取得了最高的F1值。
To address the problems of incomplete feature extraction of traditional model and low semantic representation of static word vector,a power grid fault text classification model based on ChineseBERT-BiSRU-MCNN-AT is proposed.The Chinese-BERT integrates font and pinyin information,combines word context information for dynamic learning,and solves polysemy.The BiSRU-MCNN extracts the local semantic and global sequence features of text to ensure the comprehensiveness of the extracted features.The soft attention gives the model the ability to recognize keywords.Experiments on power fault texts show that the ChineseBERT-BiSRU-MCNN-AT achieves the highest F1,value.
作者
崔艳林
林旭
郭俊宏
周煜捷
CUI Yanlin;LIN Xu;GUO Junhong;ZHOU Yujie(Electric Power Dispatching and Control Center of Guangdong Power Grid Co.,Ltd.,Guangzhou 510600,China)
出处
《微型电脑应用》
2023年第8期64-67,共4页
Microcomputer Applications
基金
中国南方电网有限责任公司科技项目(036000KK52190045,GDKJXM20198558)。
