Improving reliability of transformers based on DGA analysis using machine learning techniques

Abstract

Classifiers, Artificial Neural Networks (ANN), Fuzzy Logic (FL) and Adaptive Neuro Fuzzy Inference System (ANFIS) have been used as methods to detect faults using data obtained from Dissolved Gas Analysis (DGA). DGA provides reasonably good results in detecting insipient faults but improvement on the method's accuracy has been done. Comparative analysis using the mentioned methods have been done on IEC 599 standard, Rogers Ratio Method and Doernenburg's method. Fault databases have been used to train the models to improve the diagnostic capability. ANFIS has shown superiority on Classifiers, ANN and FL which is evident from the obtained results. ANFIS being a union of all the said methods, it has a higher prediction accuracy and is user friendly thereby, providing a promising surrogate in reinstating the conventional methods.