针对旋转电机异常检测问题,提出了一种基于卷积自编码器(Convolutional Autoencoder,CAE)和单分类支持向量机(One Class SVM,OC-SVM)的异常检测方法.首先使用旋转电机正常运行时的电流数据进行无监督的CAE训练,自适应地提取特征;然后使用该特征训练OC-SVM;最后结合二者对新的样本进行异常检测.实验结果表明,对比基于时域特征结合OC-SVM和基于CAE特征结合K-means聚类的异常检测算法,所提方法的误检率和漏检率均有降低,最少降低了 43％,最多降低了 74％,证明了所提方法的有效性.
Rotation Motor Anomaly Detection Based on CAE-OC-SVM
A novel anomaly detection method for rotation motor based on convolutional autoencoder(CAE)and one-class support vector machine(OC-SVM)is proposed in this paper.Firstly,unsupervised training of the CAE is performed using current data collected during normal operation of the rotation motor,enabling adaptive feature extraction.secondly,the ex-tracted features are utilized to train the OC-SVM.Finally,the combination of CAE and OC-SVM is employed for anomaly detection on new samples.Experimental results demonstrate that compared to the anomaly detection algorithms based on time-domain features combined with OC-SVM and CAE features combined with K-means clustering,the proposed method achieves reduced false positive and false negative rates.The reduction ranges from 43％to 74％,affirming the effectiveness of the proposed approach.
万方数据
维普
