| As the technology of cloud data centers continues to mature,more and more business systems are deployed on the cloud to achieve dynamic scalability and comprehensive operation and maintenance of resources.Therefore,how to collect multi-source data in the cloud network environment and perform anomaly detection through data modeling becomes an important issue.Firstly,the basic methods of multi-source data collection in the cloud environment are introduced and the current data collection schemes are comprehensively analyzed.At the same time,the advantages and disadvantages of various existing monitoring tools are also described.Secondly,aiming at the large amount of multi-source data in the cloud environment,the imbalance of positive and negative samples and the high data dimension,an anomaly detection model is constructed based on Deep Belief Network integrated learning.The model solves the problem of positive and negative sample imbalance of multi-source operation and maintenance data,and makes use of the feature extraction function of Deep Belief Network effectively to reduce the dimension of multi-source timing Key Performance Indicator data.Then,in conjunction with Logistic Regression and Restricted Boltzmann Machines,the anomaly detection is implemented,and the recognition rate of anomaly obtained by a single weak classifier is over 99%.In the integration process of multiple weak classifiers,this thesis proposes an self-adaptive threshold voting algorithm.By traversing all the available thresholds,the maximum result of F1-score is selected as the optimal threshold result.Through self-adaptive threshold voting algorithm,the multiple weak classifiers are integrated so that the generalization of the model is improved,the accuracy of the anomaly detection model is over 99.23%,and the recall of model is 99.38%. |
