摘要
充分利用历史电力设备缺陷描述文本可对新出现的设备故障进行快速分类,提升运维人员的检修效率,为此针对缺陷描述文本具有复杂语义等特点,提出基于多头注意力循环卷积神经网络(multi-head attention recurrent convolutional neural networks,MAT-RCNN)的电力设备缺陷分类方法。首先对电力设备缺陷描述文本进行研究,并分析部分文本分类模型的局限;然后采用分布式表示方法将词语表示为向量形式,并将多头注意力机制与优化的RCNN结合,构建基于MAT-RCNN的电力设备缺陷描述文本分类模型;最后,通过算例比较分析,证明所提方法在语义学习能力、分类效果等方面优于RNN等常规方法。
In order to make full use of the record texts of historical power equipment defects to quickly classify the new power equipment faults and improve maintenance efficiency of the operation and maintenance personnels,this paper proposes a power equipment defect classification method for the complex semantics of defect record texts based on multi-head attention recurrent convolutional neural network(MAT-RCNN).It firstly studies the description texts of the power equipment defects and analyzes the limitations of some text classification models.Then,it uses the distributed representation method to express the words in vector form and combines the multi-head attention mechanism with the optimized RCNN to build a text classification model of power equipment defects based on MAT-RCNN.Finally,through comparative analysis of numerical examples,the paper proves that the proposed method is superior to RNN and other conventional methods in semantic learning ability and classification effect.
作者
陆世豪
祝云
周振茂
LU Shihao;ZHU Yun;ZHOU Zhenmao(Guangxi Key Laboratory of Power System Optimization and Energy Technology,Guangxi University,Nanning,Guangxi 530004,China;Laibin Power Supply Bureau of Guangxi Power Grid Co.,Ltd.,Laibin,Guangxi 546100,China)
出处
《广东电力》
2021年第6期30-38,共9页
Guangdong Electric Power
基金
广西创新驱动发展专项资金项目(桂科AA19254034)。
关键词
多头注意力
循环卷积神经网络
文本分类
电力设备缺陷文本
深度语义学习
multi-head attention
recurrent convolutional neural networks(RCNN)
text classification
power equipment defect text
deep semantic learning
