基于集成学习的云平台异常点检测

Cloud platform anomaly detection based on ensemble learning

为提高云平台异常点检测的精度,解决单一检测系统误报率与漏报率高的问题,提出基于集成学习的异常点检测系统。为解决异常检测对象多样性的问题,构造监测序列的特征矩阵,采用自组织映射神经网络对监测序列进行聚类;对监测序列进行过采样,解决异常发生频率很低的问题;对异构的异常点检测器进行基于委员会的学习,集成各检测器的检测优点,提高检测的精度。通过带有标注的监测序列对异常点检测系统进行验证,结果表明,该系统效果优于单一检测系统,验证了设计的有效性。

To improve the accuracy of cloud platform anomaly detection and solve the problem of high false alarm rate and false negative rate of single detection system,an anomaly detection system based on ensemble learning was proposed.The monitoring sequence similarity matrix was constructed,and the self-organizing map neural network was used to cluster the monitoring sequences,which solved the problem of the diversity of the anomaly detection objects.The data were oversampled to solve the problem of low frequency of anomaly.The heterogeneous detection model integrated learning was used to improve the accuracy of anomaly detection.The anomaly detection system was verified by using the labeled monitoring sequence.The results show that the proposed system is better than the single detection system,which verifies the effectiveness of the design.