Data-Driven Anomaly Detection in High-Voltage Transformer Bushings with LSTM Auto-Encoder.

Area of Science:

• Electrical Engineering
• Power Systems
• Machine Learning

Background:

• Bushing failures in high-voltage (HV) power transformers can lead to significant financial losses due to power outages.
• Identifying insulation deterioration is crucial for preventing catastrophic bushing failures.
• Continuous monitoring of bushing measurements can indicate equipment condition anomalies.

Purpose of the Study:

• To develop a real-time anomaly detection method for HV transformer bushings.
• To utilize machine learning for monitoring current magnitude and phase angle from bushing taps.
• To ensure the reliability and health of critical power infrastructure.

Main Methods:

• A Long Short-Term Memory Auto-Encoder (LSTMAE) network was employed for anomaly detection.
• The LSTMAE model learns normal operational patterns of bushing current and phase angle.
• Anomaly detection is based on evaluating changes in measurements using the Mean Absolute Error (MAE) metric.

Main Results:

• The proposed machine learning method successfully detected anomalous events in real-world data.
• The system demonstrated real-time anomaly detection capabilities for HV transformer bushings.
• The approach proved to be fast, self-supervised, and flexible in identifying deviations.

Conclusions:

• The LSTMAE-based method provides an effective solution for real-time anomaly detection in HV transformer bushings.
• This approach enhances the reliability of power supply by preemptively identifying potential equipment failures.
• The self-supervised and flexible nature of the method makes it suitable for practical industrial applications.