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

With the development of information technology,the application of information technology in industry has gradually developed,and one of the prominent applications is the Industrial Internet.The production of the Industrial Internet has greatly improved the production efficiency with the help of information technology.The production efficiency not only includes the manufacturing efficiency,but also the various sensor data in the industrial process with the application of artificial intelligence technology.The anomaly detection capability of parameter indicators helps the industry reduce operating costs and improve detection efficiency.However,the sensor data in the Industrial Internet has the characteristics of dynamic,large-scale and time correlation,so there are challenges such as the difficulty of modeling and capturing the statistical characteristics of the data distribution of these time series data,and the difficulty of quantitative analysis of outliers.The research of time series data and the problem of anomaly detection emerge as the times require.Based on the above challenges,this paper mainly carries out the following three aspects of anomaly detection research based on time series data.(1)This paper proposes a data completion algorithm based on Long Short-Term Memory(LSTM)on the missing value scene of onedimensional time series data,the LSTM-Completion data completion model,through Modeling and forecasting of time series data solves the problem of missing values in time series data scenarios.And through experiments on the Power Demand dataset,it is verified that the model in this paper has superior data in the mean squared error(Root Mean Squared Error,RMSE)index compared with the data completion baseline model(Random forest,RF-model).Complete performance.(2)In order to solve the problems of sample imbalance and time correlation in the actual time series data anomaly detection scenario,this paper adopts an unsupervised learning-based time series anomaly detection algorithm,that is,based on LSTM prediction model and Gaussian function anomaly scoring Anomaly detection algorithm for time series data of the controller.This paper conducts experiments and analysis on the anomaly detection algorithm of time series data based on the traditional LSTM prediction model,and finds that the anomaly scoring function of this model does not fit the actual distribution of the sample,and proposes a new regression-based Gaussian function(Gaussian function)LSTM-GAUSS anomaly detection model for anomaly scorer.And through experiments on the datasets Power Demand Dataset and KPI Dataset,it is verified that the LSTM-GAUSS model in this paper has better anomalies than other baseline models in terms of precision,recall and F1-score.Check performance.(3)This paper studies and analyzes the effects of the parameters in the LSTM-GAUSS anomaly detection model,such as the size of Lookback,on the anomaly detection performance of the model,and the Gaussian function anomaly scorer compared to the Gaussian distribution anomaly scorer in the sample.The advantage of fit.