🤖 AI Summary
Addressing the challenge of unsupervised anomaly detection in multivariate time series—where both temporal dynamics and inter-variable dependencies must be jointly captured—this paper proposes a temporally aware unsupervised clustering framework. Methodologically, it integrates sliding-window segmentation, dynamic time warping (DTW)-driven time-series clustering, and autoencoder-based feature learning, jointly optimizing reconstruction error and cluster-based outlier scores to simultaneously discriminate anomalies in magnitude and shape. Its key innovation lies in embedding temporal structural priors directly into the clustering process, substantially enhancing modeling capacity for complex cross-variable dependencies and nonlinear temporal evolution. Extensive experiments on multiple benchmark datasets demonstrate that the method significantly outperforms state-of-the-art unsupervised and semi-supervised baselines, achieving superior detection accuracy, robustness to noise and distribution shifts, and inherent interpretability through interpretable cluster assignments and reconstruction residuals.