Research On Anomaly Detection Algorithm Of Time Series Data In Cloud Environment

In the cloud platform,the system architecture is complex and there are many types of services.Once the system crashes,it will cause incalculable losses.In order to ensure the security operation and maintenance of cloud environment,it is necessary to detect the KPI data effectively.However,the large number of KPI time series,the small number of tags and the characteristics of KPI time series make it difficult to detect exceptions.Based on the research background of cloud environment automatic operation and maintenance monitoring,this paper takes the anomaly detection algorithm which is suitable for the characteristics of KPI time series data in cloud environment as the research content,and finally realizes the automatic anomaly detection of KPI time series data.The main work is as follows:(1)In this paper,an anomaly detection algorithm based on tsfresh is proposed.The algorithm takes "abnormal time sequence sub segment" as the detection target,and uses tsfresh to automatically extract and filter the digital features after segmenting the KPI time sequence data.Then,XGBoost algorithm is used to model the anomaly recognition based on the extracted features,and the confidence degree of the anomaly sequence in KPI time series data is predicted accordingly,with the accuracy of 95.06%.(2)Based on the i Forest algorithm,the anomaly detection of time series data is carried out.The algorithm can detect two types of anomaly: "general outliers" and "fault outliers".First,the paper uses the waveform smoothing technology to process the single variable data and generate new samples,then delete a large number of noise information with high similarity to the abnormal points.Then,four machine learning algorithms KNN,LOF,One Class SVM and i Forest are used to detect the abnormal time series data.In order to reduce the difference between positive and negative samples and improve the detection performance of the algorithm,a suitable threshold can be set to filter the samples.After the optimization of the model,an anomaly detection model based on i Forest is built,with the accuracy of 83.91%.This study effectively improves the accuracy of anomaly detection of time series data in cloud environment,and provides an optional method and idea for automatic anomaly detection in cloud environment,which has certain research value.