DGA-Based Early Transformer Fault Detection using Rough Set Theory Classifier

In the electricity transmission and distribution network, power transformer is one of the most important and expensive equipment. The transformers determine the availability and reliability of the power grid, and will have a direct impact on the electricity distribution to consumers and have an impact on the financial sector. One of the efforts to maintain the availability of the transformers is to identify potential faults earlier. Dissolved gas-in-oil analysis (DGA) is a popular technique for this. However, the DGA data is not linear, so need to find an analysis method that is capable to handle this characteristic. The purpose of this experiment is to develop an artificial intelligence based method that can accurately and reliably classify the type of transformer fault based on DGA data with three performance metrics, namely accuracy, precision, and recall. The method proposed is to apply Rough Set Theory for fault classifier. It was found that the proposed method has better performance compared to conventional methods. The accuracy of proposed method is 0.92, while for conventional methods, the highest is 0.72. In terms of precision and recall, in general the value is higher with a minimum of 0.75 compared to 0.04 of the conventional methods. The high values of precision and recall are the contribution of Synthetic Minority Over-sampling Technique which is applied in the pre-processing data. © 2021 IEEE.