CrossAD: Time Series Anomaly Detection with Cross-scale Associations and Cross-window Modeling

TL;DR

CrossAD tackles time series anomaly detection by explicitly modeling cross-scale associations and transcending fixed-window limitations. It introduces cross-scale reconstruction with a three-stage design (multi-scale generation and embedding, scale-independent encoding, next-scale generation) and a cross-window mechanism using a learnable sub-series query library plus a global multi-scale context to share information across windows. Empirical results across seven real-world datasets show state-of-the-art performance on multiple metrics, with ablations confirming the value of cross-scale reconstruction and the global context. The approach is practically impactful for robust anomaly detection across varying temporal granularities and streaming contexts.

Abstract

Time series anomaly detection plays a crucial role in a wide range of real-world applications. Given that time series data can exhibit different patterns at different sampling granularities, multi-scale modeling has proven beneficial for uncovering latent anomaly patterns that may not be apparent at a single scale. However, existing methods often model multi-scale information independently or rely on simple feature fusion strategies, neglecting the dynamic changes in cross-scale associations that occur during anomalies. Moreover, most approaches perform multi-scale modeling based on fixed sliding windows, which limits their ability to capture comprehensive contextual information. In this work, we propose CrossAD, a novel framework for time series Anomaly Detection that takes Cross-scale associations and Cross-window modeling into account. We propose a cross-scale reconstruction that reconstructs fine-grained series from coarser series, explicitly capturing cross-scale associations. Furthermore, we design a query library and incorporate global multi-scale context to overcome the limitations imposed by fixed window sizes. Extensive experiments conducted on multiple real-world datasets using nine evaluation metrics validate the effectiveness of CrossAD, demonstrating state-of-the-art performance in anomaly detection.

Figures (14)

• Figure 1: Example of cross-scale associations. Scales 1, 2, and 3 represent three series obtained by downsampling from the same original series, ranging from fine to coarse. The blue arrows indicate the associations between coarse-grained and fine-grained series under normal conditions, while the red arrows highlight the associations that are disrupted during anomaly occurrences.
• Figure 2: The architecture of CrossAD, which takes scale number $m$=3 as an example.
• Figure 3: Sub-series of the same color have similar contexts.
• Figure 4: Visualization of the traditional reconstruction criterion and cross-scale reconstruction criterion across five anomaly patterns. The first row shows the original series. The second and third rows display anomaly scores based on the traditional reconstruction criterion and cross-scale reconstruction criterion, respectively. Red areas indicate anomalies.
• Figure 5: Visualization of cross-scale reconstruction results. (a)-(e) represent the five scales ranging from coarse to fine, where the blue line indicates the input series, the orange lines depict the reconstructed series at each scale, and the green lines show the corresponding anomaly scores. (f) illustrates the final anomaly score obtained by fusing the anomaly scores from all five scales, with red regions marking the actual anomaly events.