基于CNN-BiLSTM特征融合的异常检测算法

ANOMALY DETECTION ALGORITHM BASED ON CNN-BILSTM FEATURE FUSION

目前运维数据的异常检测大都针对单一特征,随着系统复杂度的增加,系统异常可能与内在相关联的监控数据特征相关,将这些特征汇聚到一起进行分析和处理,可以提高异常检测算法的可靠性。CNN可以提取数据的空间维度特征,LSTM可以获得时序数据的时间维度特征,Attention可以有效地对空间和时间特征进行加权,提出一种基于CNN-BiLSTM特征融合的异常检测算法,并引入Attention机制。在多个数据中心运维数据上的实验结果表明,该算法可以有效提高异常检测的效果。

At present, anomaly detection of operation data is mostly for single feature. As the complexity of the system increases, system anomalies may be related to multi-dimensional inherently related features of monitoring data. Converging together for analyzing and processing can improve the reliability of anomaly detection. CNN can extract the spatial dimension features of data, LSTM can obtain the time dimension features of time series data, and Attention can effectively weight spatial and temporal features. An anomaly detection algorithm based on CNN-BiLSTM feature fusion is proposed, and adds attention mechanism. Experimental results on multiple operation data in data center show that this algorithm can effectively improve the effect of anomaly detection.