Subdomain Adaptation Capsule Network for Partial Discharge Diagnosis in Gas-Insulated Switchgear.

Area of Science:

• Electrical Engineering
• Artificial Intelligence
• Materials Science

Background:

• Deep learning, particularly Convolutional Neural Networks (CNNs), shows promise for partial discharge (PD) diagnosis in gas-insulated switchgear (GIS) within laboratory settings.
• Challenges remain in field applications due to CNNs overlooking feature relationships and requiring extensive sample data, hindering robust and high-precision diagnosis.

Purpose of the Study:

• To develop a novel method for accurate and robust PD diagnosis in GIS field environments.
• To address the limitations of traditional deep learning models in handling real-world field data complexities.

Main Methods:

• A subdomain adaptation capsule network (SACN) was employed for PD diagnosis in GIS.
• Capsule networks were utilized for enhanced feature extraction and representation.
• Subdomain adaptation transfer learning was applied to align local data distributions and mitigate subdomain confusion.

Main Results:

• The SACN achieved a diagnostic accuracy of 93.75% on field data for GIS.
• The proposed SACN demonstrated superior performance compared to conventional deep learning techniques.
• The method effectively improved feature representation and adapted to field data variations.

Conclusions:

• The subdomain adaptation capsule network (SACN) presents a viable and effective solution for high-precision PD diagnosis in GIS field applications.
• The SACN's ability to improve feature representation and adapt to subdomain variations highlights its potential for real-world power system monitoring.
• This approach offers significant potential for enhancing the reliability and safety of gas-insulated switchgear through advanced diagnostic capabilities.