A Novel Method for Pattern Recognition of GIS Partial Discharge via Multi-Information Ensemble Learning

Area of Science:

• Electrical Engineering
• Materials Science
• Artificial Intelligence

Background:

• Partial discharge (PD) is a key indicator of internal insulation defects in gas-insulated switchgear (GIS).
• Accurate PD pattern recognition is crucial for diagnosing insulation faults and ensuring GIS operational reliability.
• Traditional single-feature analysis methods exhibit low recognition accuracy and varying diagnostic effectiveness for different insulation defects.

Purpose of the Study:

• To develop a novel multi-information ensemble learning approach for enhanced PD pattern recognition.
• To leverage rich insulation state information from PD for improved fault diagnosis in GIS.
• To overcome the limitations of single-feature analysis in PD diagnosis.

Main Methods:

• Acquisition of ultra-high frequency (UHF) and ultrasonic PD data from four typical GIS defects via experimentation.
• Utilizing a deep residual convolutional neural network (DRCNN) for automated extraction of discriminative PD features.
• Employing multi-information ensemble learning at the decision level for PD type classification.

Main Results:

• The proposed method achieved a diagnostic accuracy of 97.500% in experiments.
• The multi-information ensemble learning approach demonstrated higher accuracy and reliability compared to independent feature recognition.
• Significant improvement in the diagnosis accuracy for various GIS insulation defects was observed.

Conclusions:

• The novel multi-information ensemble learning method effectively enhances PD pattern recognition for GIS.
• This approach offers a more accurate and reliable solution for diagnosing internal insulation defects in GIS.
• The findings contribute to improved operational safety and maintenance strategies for gas-insulated switchgear.