A Robust and Accurate Multivariate Time Series Anomaly Detection in Fluctuating Cloud-Edge Computing Systems

Cloud-edge computing uses edge infrastructure to extend cloud computing further away from the data source, compensating for some of the limitations of conventional cloud computing. Multivariate time series anomaly detection can reflect the operating status of edge cloud computing systems. Many anomaly detection methods have been developed but still face challenges, especially irregular fluctuations in data leading to false positive detection. In addition, good detection robustness is necessary to capture complex data patterns in cloud-edge computing systems, but few existing methods focus on it. Furthermore, monitoring data usually lack labels, while manual data labeling is ineffective. To address these issues, we propose an unsupervised multivariate time series anomaly detection method: Correlative-GNN with Multi-head Self-attention and Auto-Regression Ensemble Method (CGNN-MSAR). Our approach provides parallel Graph Neural Networks(GNNs) to learn inter-dependencies in multivariate time series from feature and time dimensions to achieve fewer false positives without any prior knowledge. Moreover, Our method integrates a multi-head self-attention mechanism and AR models to improve the robustness of the model. Extensive experiments have been conducted on five publicly available datasets. Specifically, CGNN-MSAR outperforms its competitors with an F1-Score of 0.86 on average and is 18.2% better than state-of-the-art baseline methods.