Research On Prediction And Anomaly Detection Algorithm Of Time Series In Cloud Environment

The cloud environment is a large-scale,distributed and complex system.Due to the interdependence and invocation of the various functional layers in the cloud environment,the efficient operation and maintenance of the cloud environment becomes a major problem.The main form of KPI data monitored daily in the cloud environment is time series.The prediction and anomaly detection of time series in the cloud environment has been the two hotspots at home and abroad.Algorithms with high prediction accuracy and high anomaly detection accuracy can help us discover potential problems in the cloud environment and stop losses in time to avoid large losses,which is of great significance for improving the high availability of the cloud environment.Based on the previous research results,this paper takes time series in cloud environment as the research object,and improves the prediction accuracy of time series and the accuracy rate of anomaly detection.This paper proposes efficient and accurate prediction algorithm and anomaly detection algorithm that is suitable for the characteristics of time series in cloud environment.The main work of this paper is as follows:1.The domestic and international research algorithms related to prediction and anomaly detection arc sorted and classified,and the research trends,latest achievements and shortcomings in this field are found.2.The characteristics,classification,anomaly types,analysis methods,prediction algorithms and anomaly detection algorithms of time series in cloud environment are summarized.3.Exploring the commonality between the prediction algorithm and the anomaly detection algorithm,and combining the two,so that the model and the result of the prediction algorithm can be simultaneously applied in the anomaly detection algorithm to improve the execution efficiency of the algorithm.4.The EWT-IF-varRNN prediction model based on empirical wavelet transform,Isolation Forest and common variants of recurrent neural network is proposed to improve the predictive ability of the model through complementary advantages.5.The OCPD algorithm based on prediction and improved multi-dimensional SAX vector representation is proposed to detect outliers and change points in time series.6.An improved multi-dimensional SAX vector representation method is proposed for the discrimination and screening of change points in time series.7.The effectiveness and accuracy of the EWT-IF-varRNN prediction model and OCPD algorithm proposed in this paper are verified by comparing experiments with the monitoring data set in the real business cloud environment.The research and contributions made in this paper improve the prediction accuracy and anomaly detection accuracy of time series in the cloud environnent.What's more,we provide our own insights and prospects for future research.