Research And Implementation Of Web Application Performance Anomaly Detection Based On Time Series Analysis

With the continuous development of Internet technology,various Web application systems are springing up in the Internet.Meanwhile,with the widespread use of smartphones,tablets and browsers,as well as the spread of cloud computing and virtualization technologies,the architecture of Internet applications is becoming increasingly complex.The factors that affect the performance of these Internet application systems become more and more complex.In the face of increasingly complex applications,it is necessary to monitor the performance of these Web applications in order to ensure their smooth and reliable operation.The performance data generated during the operation of Web application system is mainly expressed in the form of time series data.High precision of prediction and anomaly detection algorithm that can help us to quickly find the potential performance issue in the process of Web application system is running,to predict the possibility of abnormal conditions in advance,to an early warning system management personnel,make the system and network management personnel in a timely manner to find the problems existing in the system operation process,early intervention,reduce the losses caused by the fault,application system and improve the availability and reliability of the application system.The performance time series data generated during the operation of Web application system is non-linear and non-stationary,which makes the prediction of the application performance time series data highly challenging.This article selects monitored application system as the research object,the response time of the data to the corresponding application system response time data processing,forecasting,and through the anomaly detection algorithm to test the performance status of the application system,and finally realize the characteristics for response time series data of efficient performance prediction model and the anomaly detection model.The main research work is as follows:(1)This paper obtains performance data based on the application performance management system platform,adopts Kafka message middleware to receive response time performance data,and implements the cleaning of performance data through data preprocessing,which improves the quality of data and provides a good data foundation for prediction model and anomaly detection model.(2)In this paper,a long-term and short-term memory neural network prediction model IPSO-LSTM based on improved particle swarm optimization algorithm is proposed,which reduces the optimization time of the prediction model and improves the prediction accuracy of the prediction model.Based on the improved particle swarm optimization algorithm to determine the LSTM both short-term and long-term memory neural network prediction model of training vector,and the number of each network layer neurons,cut short-and long-term memory neural network prediction model for the optimal parameters of the time,improve the accuracy of the prediction model to predict,for behind the performance data of anomaly detection laid the good foundation.(3)On the basis of analyzing the anomaly detection method of performance data,this paper proposes an anomaly detection method based on dynamic threshold.The relative deviation between the predicted value and the measured value is used as the judgment basis for the occurrence of anomalies.Because the performance data of Web application system is constantly changing according to time,this paper adopts the detection method of sliding interval and relative deviation to detect the abnormal performance time series data generated during the operation of Web application system.This method realizes the on-line detection of performance data,and can conveniently and quickly detect the realtime performance data generated during the operation of the Web application system.