Research On Anomaly Detection Method Of Multivariate Time Series Based On Deep Learning

With the development of High-speed railway scale,the anomaly detection method is not able to meet the safety needs of traction system.In recent years,deep learning has been gradually applied to anomaly detection tasks,not only to prevent dynamically,but also without human involvement,and to present a more obvious advantage over traditional machine learning in processing large amounts of data.Therefore,this paper will study how to use deep learning technology to improve the performance of traction system anomaly detection.In addition,due to insufficient abnormal data,existing detection methods are unable to provide exception category information,resulting in the inability to locate the source of the anomaly quickly.The main work of this article is as follows:First of all,this paper proposes a multivariate time series anomaly detection method based on deep autoencoder,which uses a multi-scale feature association matrix to describe the state of the current system,and designs an anomaly discrimination rule based on the matrix,and improves the deep automatic coding structure,which uses the reel neural network to extract spatial information of the multi-scale feature association matrix,and introduces an improved jump connection layer to optimize the autoencoder structure,aiming at adapting the low-level feature information in the encoder network.Helps better rebuild multiscale feature association matrices.Then,in view of the problem that the anomaly detection can't provide the abnormal category and exception classification,a kind of classification method based on semisupervision is proposed,which uses semi-supervised clustering to classify the exception,and optimizes the effect of time series clustering by using DTW distance and DBA average sequence in the clustering method.Finally,the model evaluation is carried out on the High-speed railway traction data set and the network public data set.Experiments show that the proposed anomaly detection method and abnormal classification method can improve the performance of anomaly detection and classification of traction system to some extent.