Research On Cloud Platform Anomaly Detection Algorithm By Ensemble Learning

With the development of cloud computing,how to ensure the efficiency and reliability of cloud platforms is getting more and more attention from enterprises.Operation and maintenance is a very important task in the management of cloud computing business platform.The operation and maintenance is to ensure the efficiency and availability of the cloud platform.Traditional operation and maintenance has problems such as complex system,high maintenance difficulty,and high risk pressure.Intelligent AIOps(Artificial Intelligence for Operations)uses machine learning to assist traditional operation and maintenance,greatly improving the efficiency of traditional operation and maintenance.In this paper,the detection direction of the abnormal point of intelligent operation and maintenance is studied.The object of detection is the monitoring data under the cloud platform,and the format is(time,monitoring value).Traditional abnormal point detection methods include threshold-based detection methods and distance-based detection methods.However,traditional abnormal point detection methods have problems of high false negative rate and false positive rate.For these shortcomings,the paper has done the following work:(1)An anomaly detection system based on integrated learning is designed.The whole system is divided into offline training and online detection.Offline training builds an abnormal point detection model through data processing,data clustering,feature engineering,model training,etc.On-line detection uses the abnormal point detection model to detect abnormal points and feeds the wrong and missing samples back to offline training and retraining.The whole process forms a closed loop.The abnormal point detection system gradually stabilizes.(2)Data imbalance is solved to some extent from the data level and the model level.During the operation and maintenance process,the anomalies rarely occur,resulting in a serious imbalance between the abnormal samples and the normal samples.Direct use of the original data for offline training will result in the abnormal point detection model preferting normal,resulting in underreporting.The paper oversamples the data and adjusts the loss function weight of the abnormal point detection model,which solves the data imbalance problem to a certain extent,and greatly improves the detection result.(3)The fusion of heterogeneous abnormal point detection models improves the accuracy of abnormal point detection.The low-variance outlier detection model is combined with the low-deviation anomaly detection model to effectively improve the detection effect.The paper uses the monitoring data with normal and abnormal tags to detect the abnormal point detection system.The results show that:(1)In the data processing stage,oversampling has the greatest impact on the improvement of the detection results.(2)In the model construction stage,model fusion has the greatest impact on the improvement of test results.(3)The effect of the anomaly detection model based on integrated learning is better than the traditional anomaly detection model.