一种基于上下文残差学习的异常检测方法及系统
By employing a contextual residual learning method, the relative difference pattern between the query window and the normal reference window is calculated. Combined with curvature perception and mask perception low-rank feature trajectories, the problem of time series anomaly detection in heterogeneous domains is solved, achieving robust anomaly detection performance across domains.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SHANDONG UNIV
- Filing Date
- 2026-01-08
- Publication Date
- 2026-07-17
AI Technical Summary
Existing time series anomaly detection methods are difficult to transfer between heterogeneous domains and cannot effectively detect anomalies in domains with different channel dimensions, sampling rates, noise mechanisms, and temporal dynamics, resulting in poor detection performance in cross-domain scenarios.
A context-based residual learning approach is adopted. By acquiring detection data from the target domain, a pre-trained time-series anomaly detection model is used to calculate the relative difference pattern between the query window and the normal reference window. Combined with curvature-aware basic residuals and mask-aware low-rank feature trajectories, a domain-invariant representation is constructed to achieve cross-domain anomaly detection.
Stable anomaly detection is achieved in cross-domain scenarios. It can suppress domain-dependent scale, bias and trend effects, maintain noise resistance and robustness, and generalize to different sensors and environments to capture the intrinsic characteristics of anomalous behavior.
Smart Images
Figure CN121479701B_ABST