Machine learning based multi-method interpretation to enhance dissolved gas analysis for power transformer fault...

Area of Science:

• Electrical Engineering
• Materials Science

Background:

• Accurate dissolved gas analysis (DGA) is crucial for power transformer reliability.
• Existing DGA interpretation methods (DRM, RRM, IRM, DTM, DPM) can yield conflicting results and encounter limitations.
• Inconsistent DGA interpretation poses risks to transformer management and operational safety.

Purpose of the Study:

• To develop a novel, more accurate, and consistent DGA interpretation technique.
• To integrate established DGA methods into a unified framework.
• To enhance the reliability of power transformer condition assessment.

Main Methods:

• A multi-method approach integrating existing DGA interpretation techniques.
• Implementation of a scoring index and random forest machine learning principles.
• Validation using DGA data from transformers under diverse operational conditions.

Main Results:

• The proposed multi-method DGA technique demonstrates superior accuracy and consistency compared to individual conventional methods.
• The random forest-based multi-method achieved higher accuracy than using the scoring index alone.
• The new approach effectively addresses limitations of ratio-based DGA interpretation.

Conclusions:

• The integrated multi-method DGA approach offers significant improvements in transformer health assessment.
• Machine learning, specifically random forest, enhances the predictive power of DGA interpretation.
• This novel technique contributes to more reliable power system operation through better transformer management.