摘要
针对牵引电机故障诊断研究中所采用的神经网络方法,提出在模型训练阶段引入K折交叉验证。该方法在划分训练集与测试集期间,使验证集能够遍历所有数据集,从多方向开始学习,从而在一定程度上避免了局部极小的问题。训练完成后,以神经网络作为分类器进行故障识别。神经网络学习算法采用随机梯度下降的方法,每次投入一组数据集进行训练,大大提高了训练速度。Eclipse+Anaconda仿真结果证明:与传统神经网络电机故障诊断方法相比,该方法可以在一定程度上避免过拟合现象,同时避免局部极小。此外,在Matlab环境下,单独比较支持向量机采用交叉验证前后的故障分类效果。对比结果表明:交叉验证方法从多方向开始学习,对于提升故障诊断的准确率有较好作用。
Aiming at the neural network method used for researching traction motor fault diagnosis,the Kfolds cross validation is proposed for the training phase of model.During dividing the training set and test set,this method enables the validation set to go through all the data sets,and learn from multiple directions to avoid the local minimum to a certain extent.After training,the neural network is used as classifier to identify the faults.The stochastic gradient descent is used in neural network learning algorithm,to train one data set at a time,which greatly improves the training speed.The results of Eclipse+Anaconda simulation prove that this method is better compared with the traditional neural network motor fault diagnosis method.It can avoid the overfitting phenomenon and the local minimum.In addition,the fault classification effects of support vector machine before and after using the cross validation are compared under the environment of Matlab,the result proves that the cross validation can improve the accuracy of fault diagnosis,because it starts learning from multiple directions.
作者
王惠中
乔林翰
贺珂珂
段洁
WANG Huizhong;QIAO Linhan;HE Keke;DUAN Jie(School of Electrical and Information Engineering,Lanzhou University of Technology,Lanzhou 730050,China)
出处
《自动化仪表》
CAS
2018年第4期22-25,共4页
Process Automation Instrumentation
基金
甘肃省教育厅科研基金资助项目(0903-07)
关键词
电机故障诊断
K折交叉验证
随机梯度下降
神经网络
拟合
支持向量机
Motor fault diagnosis
K-fold cross validation
Stochastic gradient descent
Neural network
Fitting
Support vector machine(SVM)
