DCFF-MTAD: A Multivariate Time-Series Anomaly Detection Model Based on Dual-Channel Feature Fusion

Area of Science:

• Data Science
• Machine Learning
• Signal Processing

Background:

• Automated monitoring of complex systems relies on detecting anomalies in multivariate time-series data.
• Increasing data volume and dimensionality pose challenges for traditional anomaly detection methods.

Purpose of the Study:

• To develop a robust and effective multivariate time-series anomaly detection model.
• To improve anomaly detection performance by focusing on spatial and temporal features.

Main Methods:

• A dual-channel feature extraction module combining spatial short-time Fourier transform (STFT) and a graph attention network.
• Fusion of spatial and temporal features for enhanced detection.
• Incorporation of the Huber loss function for improved model robustness.

Main Results:

• The proposed model demonstrated superior anomaly detection performance compared to state-of-the-art methods on three public datasets.
• Validation of the model's effectiveness and practicality in real-world shield tunneling applications.

Conclusions:

• The dual-channel feature extraction model offers a significant advancement in multivariate time-series anomaly detection.
• The model's robustness and practical applicability are confirmed through comparative studies and real-world testing.