基于改进深度学习混合网络与小波分析的电机故障诊断方法  

Motor Fault Diagnosis Method Based on Improved Deep Learning Hybrid Network and Wavelet Analysis

作　　者：李志军 陈伟根[1] 周湶[1] 宗起振 卢应强 LI Zhijun;CHEN Weigen;ZHOU Quan;ZONG Qizhen;LU Yingqiang(State Key Laboratory of Power Transmission Equipment&System Security&New Technology,School of Electrical Engineering,Chongqing Univ.,Chongqing 400044,China;Guodian Nanjing Automation Haiji Technology Co.,Ltd.,Nanjing 211153,China)

机构地区：[1]重庆大学输配电装备及系统安全与新技术国家重点实验室,重庆400044 [2]江苏国电南自海吉科技有限公司,南京211153

出　　处：《三峡大学学报（自然科学版）》2021年第6期94-99,共6页Journal of China Three Gorges University：Natural Sciences

基　　金：国家创新研究群体基金项目(51321063);江苏省科技计划项目(KJZH7101)。

摘　　要：随着深度学习算法研究的深入与电力生产规模的进一步扩大,采用性能更优越的深度学习算法不断优化电机故障诊断的方法具有重要意义.针对浅层学习算法与传统深度学习混合网络存在的问题,将改进的深度学习混合网络与小波分析用于电机的故障诊断,对传统的深度学习混合网络进行了改进:一方面,为了提高其鲁棒性与抗噪能力,用降噪自动编码器代替传统自动编码器;另一方面,为了提高其准确率,用高斯伯努利受限玻尔兹曼机,代替传统的受限玻尔兹曼机层.并基于实例研究,在保证诊断准确率的前提下,选择了较为合理的隐藏层参数与训练样本集.结果表明:相比其他浅层算法与传统的深度学习网络,该方法能有效提高电机故障诊的可靠性和准确率.With the research of deep learning algorithm deeply and the expansion of power production scale further,it is of great significance to continuously optimize the method of motor fault diagnosis by using the deep learning algorithm which has better performance.Aiming at the problems existing in the shallow learning algorithm and the traditional deep learning hybrid network,the traditional deep learning hybrid network is improved in this paper:on the one hand,the automatic encoder of noise reduction is adopted instead of the traditional automatic encoder in order to improve its robustness and anti-noise ability.On the other hand,Gauss Bernoulli restricted Boltzmann machine is utilized to replace the traditional restricted Boltzmann machine layer in order to improve its accuracy.Based on the case study,the more reasonable hidden layer parameters and training sample set are selected on the premise of ensuring the accuracy of diagnosis.Meanwhile,the most reasonable hidden layer parameters and the training sample set are obtained considering the training time and accuracy.The results show that the proposed method can effectively improve the reliability and accuracy of motor fault diagnosis compared with the other shallow algorithm and the traditional deep learning network.

关 键 词：电机 故障诊断 深度学习 混合网络 小波分析 

分 类 号：TM343[电气工程—电机]