TSA-Net: Multivariate Time Series Anomaly Detection Based on Two-Stage Temporal Attention.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Industrial Monitoring

Background:

• Multivariate time series anomaly detection is crucial for industrial intelligent monitoring.
• Existing methods face challenges with high training costs and slow convergence, hindering adaptation to dynamic industrial conditions.

Purpose of the Study:

• To propose an efficient two-stage spatio-temporal attention detection framework (TSA-Net) for industrial anomaly detection.
• To address the limitations of existing methods regarding training costs and deployment efficiency.

Main Methods:

• TSA-Net employs a two-branch architecture with a structurally reparameterized temporal convolutional network (RepVGG-TCN) and a graph attention network (GAT).
• A cascading feedback mechanism refines predictions iteratively, while an adaptive gating mechanism fuses spatio-temporal features.

Main Results:

• TSA-Net achieved significant optimization, improving the F1 score by approximately 7% compared to state-of-the-art algorithms.
• Training time was reduced by up to 99% compared to complex Transformer-based models, demonstrating enhanced efficiency.

Conclusions:

• TSA-Net provides an efficient and rapid-deployment solution for high-dimensional anomaly detection in industrial settings.
• The framework's design optimizes structural complexity and improves model adaptability for dynamic environments.