Research On Anomaly Detection Algorithm Of Time Series Data Based On Deep Learning

With the continuous improvement of modern information technology,the Internet,the Internet of Things,5G and other applications have been widely popularized.Mass data is an objective existence and they are playing an increasingly important role.Time series data is an important structured form of data in mass data.How to efficiently detect anomalies in various time series data is an important research topic,which aims to accurately identify abnormal data from mass data and provide operators with the health conditions of each component of the monitored equipment.With the powerful learning ability of deep learning,this paper makes an in-depth study on the anomaly detection task of time series data.In view of the challenges brought by the special properties of time series data,such as unbalanced category,correlation between multiple features and time dependence,this paper proposes two kinds of time series data anomaly detection algorithms based on deep generative model.The main work of this paper is as follows:For the equipment monitored by multiple sensors,time series data are recorded in the form of multiple features,so we require the anomaly detection model to have strong feature learning ability and anomaly diagnosis ability.Aiming at this requirement,this paper proposes an anomaly detection model for time series data based on improved autoencoder network.The model is mainly composed of convolutional neural network,long short-term memory network and residual loss function.First,the model uses a convolutional neural network to learn the correlations between multiple features.Secondly,we use the long short-term memory network to capture the correlation between the sequences,and on this basis,we introduce the attention mechanism to avoid the deterioration of the model performance caused by the increase of the sequence length.The residual loss is then used to detect and diagnose anomalies.Finally,the model is evaluated on the actual datasets,and the experimental results show that the model performs well.Furthermore,an anomaly detection model for time series data based on optimized generation adversarial network is proposed.Based on the traditional generative adversarial network,this model introduces the time information as an additional condition into the traditional generative adversarial network,and solves the three possible misjudgment cases in the traditional generative adversarial network,thus improving the detection effect of the model.Compared with the traditional generative adversarial network,the experiment of anomaly detection on real datasets show that the generative adversarial network with added time information has better detection effect.