Fault diagnosis method for oil-immersed transformers integrated digital twin model

Area of Science:

• Electrical Engineering
• Artificial Intelligence
• Materials Science

Background:

• Low accuracy in transformer fault diagnosis hinders reliable operation.
• Poor state perception and limited real-time collaboration impede effective maintenance.
• Existing methods struggle with complex fault identification in oil-immersed transformers.

Purpose of the Study:

• To develop an accurate and stable fault diagnosis method for oil-immersed transformers.
• To integrate digital twin technology for real-time monitoring and diagnosis.
• To enhance the intelligent operation and maintenance of transformers.

Main Methods:

• Fault sample balancing using Iterative Nearest Neighbor Oversampling (INNOS).
• Extraction of nine-dimensional ratio features and correlation analysis with dissolved gases.
• Feature fusion and dimensionality reduction via sparse principal component analysis (SPCA).
• Optimization of Kernel Extreme Learning Machine (KELM) parameters using Aquila Optimizer (AO) for the AO-KELM model.

Main Results:

• The proposed AO-KELM model achieved a high fault diagnosis accuracy of 98.1013%.
• Integration with digital twin models enabled real-time interaction between physical and virtual transformer spaces.
• The method demonstrated high diagnostic accuracy and strong stability in experimental results.

Conclusions:

• The digital twin-integrated fault diagnosis method significantly improves accuracy and stability.
• Real-time online diagnosis is enabled, enhancing transformer operational intelligence.
• This approach provides a valuable reference for intelligent operation and maintenance of transformers.