A representation-enhanced graph temporal convolutional network under complex missing patterns for equipment anomaly...

Area of Science:

• Industrial IoT
• Machine Learning
• Time Series Analysis

Background:

• Multivariate time series anomaly detection in equipment is critical for operational reliability.
• Existing methods struggle with missing data, compromising anomaly detection accuracy.
• Complex missing data patterns in industrial equipment pose significant challenges.

Purpose of the Study:

• To propose a novel method for anomaly detection in equipment under complex missing data patterns.
• To enhance the representation of system health status by integrating reconstruction and prediction.
• To improve the reliability and accuracy of anomaly detection in the presence of missing data.

Main Methods:

• Developed a representation-enhanced graph temporal convolutional network (REGTCN).
• Integrated reconstruction-based and prediction-based paradigms for joint optimization.
• Utilized a missing-tolerant masked graph attention (MGAT) network for reconstruction.
• Employed an adaptive multi-scale temporal convolutional interaction network (AMTCIN) for prediction.

Main Results:

• The proposed REGTCN method effectively handles complex missing data patterns.
• Experimental results show superior performance compared to baseline models across various missing-data scenarios.
• The integrated framework enhances the representation of system health status.

Conclusions:

• The REGTCN method offers a robust solution for multivariate time series anomaly detection with missing data.
• This approach significantly improves the reliability of anomaly detection in industrial equipment.
• The study highlights the importance of addressing missing data challenges in time series analysis.