A Degradation Warning Method for Ultra-High Voltage Energy Devices Based on Time-Frequency Feature Prediction

Area of Science:

• Electrical Engineering
• Materials Science
• Signal Processing

Background:

• Resistance plate deterioration in ultra-high voltage devices poses a significant risk.
• Accurate monitoring of leakage currents is crucial for predicting device failure.
• Existing methods struggle with noise and accurate long-term forecasting.

Purpose of the Study:

• To propose an improved algorithm for analyzing leakage current characteristics.
• To enhance the prediction accuracy of future trends in DC characteristics.
• To enable early short-term warnings for lightning arrester conditions.

Main Methods:

• Developed an improved symplectic geometric mode decomposition-wavelet packet (ISGMD-WP) algorithm.
• Enhanced the I-Informer prediction network with improved embedding and distillation layers.
• Integrated predictions from adjacent columns to mitigate power grid fluctuations.

Main Results:

• The ISGMD-WP algorithm achieved a high decomposition ability evaluation index (EIDC) of 1.95 under intense noise.
• The I-Informer network demonstrated low mean absolute error (MAE) of 0.02538 and root mean square error (RMSE) of 0.03175 in long-term predictions.
• The integrated approach enables accurate fault prediction and timely warnings for energy devices.

Conclusions:

• The proposed ISGMD-WP algorithm effectively isolates prominent harmonics of leakage current.
• The enhanced I-Informer network accurately forecasts future changes in DC characteristics.
• The study successfully provides a method for early fault warnings in ultra-high voltage energy devices.