Identification of transformer fault based on dissolved gas analysis using hybrid support vector machine-modified...

Area of Science:

• Electrical Engineering
• Artificial Intelligence
• Power Systems Analysis

Background:

• Early detection of power transformer faults is vital for reducing maintenance costs and ensuring uninterrupted electricity supply.
• Dissolved Gas Analysis (DGA) is a common technique for identifying faults in oil-filled power transformers.
• Artificial intelligence methods, particularly with optimization techniques, offer promising results for DGA-based fault diagnosis.

Purpose of the Study:

• To propose a hybrid Support Vector Machine (SVM) model integrated with a modified Evolutionary Particle Swarm Optimization (EPSO) algorithm for accurate power transformer fault type determination.
• To evaluate the effectiveness of the proposed hybrid SVM-MEPSO-TVAC technique against unoptimized SVM and existing methods.
• To enhance the efficiency of the SVM training process through data reduction using stepwise regression.

Main Methods:

• Development of a hybrid Support Vector Machine (SVM) model.
• Integration of a modified Evolutionary Particle Swarm Optimization (EPSO) algorithm, specifically the MEPSO-TVAC variant, to optimize SVM parameters.
• Application of stepwise regression for data reduction prior to SVM training.

Main Results:

• The proposed hybrid SVM-MEPSO-TVAC technique achieved the highest correct fault identification percentage compared to other Particle Swarm Optimization (PSO) algorithms.
• The optimized SVM model demonstrated superior performance in classifying transformer fault types based on DGA data.
• Data reduction using stepwise regression effectively reduced SVM training time without compromising accuracy.

Conclusions:

• The developed hybrid SVM-MEPSO-TVAC technique presents a highly accurate and efficient solution for identifying power transformer fault types using DGA data.
• This optimized artificial intelligence approach offers a potential on-site diagnostic tool for power transformer maintenance.
• The study highlights the benefits of combining advanced optimization algorithms with machine learning for critical power system diagnostics.