Search papers, labs, and topics across Lattice.
This paper introduces MSCENet, a novel framework designed for multivariate time series anomaly detection that effectively captures both temporal dynamics and inter-series correlations. By employing a combination of dilated convolutions and Mixhop graph convolutions, MSCENet enhances detection accuracy across multiple temporal scales, addressing the limitations of traditional methods. Experimental results on real-world datasets demonstrate its superior performance in identifying subtle anomalies in complex data environments.
MSCENet achieves unprecedented accuracy in detecting anomalies in multivariate time series by integrating advanced spatio-temporal learning with graph-based correlation modeling.
In the field of multivariate time series anomaly detection, against the backdrop of increasing data complexity and complex dependencies across multiple temporal scales, traditional methods often struggle to simultaneously capture temporal dynamic features and intricate inter-series correlations. To address this, we propose an innovative framework, MSCENet, which leverages advanced spatio-temporal learning and multi-scale learning techniques to enhance detection accuracy. MSCENet includes a fine-grained temporal convolution module that captures complex temporal dependencies through dilated convolutions, enabling the detection of both short- and long-term patterns. Additionally, the framework models inter-series relationships as a graph structure, using Mixhop graph convolutions to adaptively capture spatial dependencies across varying time scales. To support robust anomaly detection, the multi-scale gated convolution module in MSCENet integrates spatial and temporal attributes through gated mechanisms, facilitating the detection of subtle variations across multiple scales. Experimental evaluations on real-world datasets: SMD, PSM, and SWaT. It provides an adaptable and high-performance solution for anomaly detection in complex time series data environments.
