基于卷积神经网络的电力设备缺陷文本分类模型研究  

A Classification Model of Power Equipment Defect Texts Based on Convolutional Neural Network

作　　者：刘梓权 王慧芳[1] 曹靖[1] 邱剑[1] 

机构地区：[1]浙江大学电气工程学院

出　　处：《电网技术》2018年第2期644-650,共7页Power System Technology

摘　　要：电网生产管理系统中存在大量闲置的设备缺陷记录文本。针对电力设备缺陷文本的特点,构建了基于卷积神经网络的缺陷文本分类模型。首先通过分析大量电力设备缺陷记录,归纳了电力设备缺陷文本的特点;然后参考中文文本分类的一般流程,并考虑缺陷文本的特点,建立了一种基于卷积神经网络的电力缺陷文本分类模型;最后通过算例对基于卷积神经网络的缺陷分类模型和多种传统机器学习分类模型进行全面比较。算例结果表明,所提出的缺陷文本分类模型能显著降低分类错误率,在分类效率上也比较可观。Currently a large amount of equipment defect record texts are left unused in power grid management system. Aiming at features of power equipment defect texts, a classification model of defect texts based on convolutional neural network is established. Firstly, the features of power equipment defect texts are extracted by analyzing a large number of defect records. Then, referencing general process of Chinese text classification, the paper establishes a classification model of defect texts based on convolutional neural network, considering the features of defect texts. Finally, the model is compared with multiple traditional machine learning classification models in an example. Results indicate that the proposed defect text classification model can significantly reduce error rate with considerable efficiency.

关 键 词：电力文本处理 缺陷分类 卷积神经网络 机器学习 

分 类 号：TM721[电气工程—电力系统及自动化]