Anomaly Detection For Hyperdimensional Time Serie With Auto Encoder

Advances in technology have made it easier to access data,but the importance of data is growing.As the most common data of life,time series data is also increasing rapidly with the passage of time.In time series analysis,anomaly detection is one of the most impor-tant directions.This paper takes hyperdimensional time series data as the research object,combines feature extraction,neural network,deep learning and other methods,and uses Auto Encoder to carry out anomaly detection.The main work is as follows:Time series data in industry are large in volume and variable in number,and the vari-ables are often correlated with each other,which results in that many classical models can-not make good use of these data.To solve this problem,this paper proposes a feature ex-traction method based on PCA and sliding window.This method first uses PCA to reduce the dimensionality of the hyperdimensional data.The dimensionality of the reduced data is independent of each other and contains most of the information of the original data.Then the sliding window is used to segment the data and generate the statistical characteristics of each window.This method can enrich the feature space of time series and make use of the time dimension information of time series.Next,this paper uses autoencoder for anomaly detection.The Auto Encoder uses a small feature space to capture the representative features in the original sequence,while outliers are often non-representative,so the outliers can be found by comparing the recon-structed sequence with the original sequence.In order to make better use of time series data,this paper adopts the long and short time memory network as the encoding and decoding model of the Auto Encoder.Finally,an experiment is carried out on a real high-dimensional time series data set.The experiment shows that the proposed method is effective in anomaly detection.