Accuracy enhancement of fault diagnosis for powertransformers with a hybrid approach integrating robustand tree-based algorithms

Power transformers (PTs) are a significant component of power grids that transmit and distribute electricity generated by renewable energy sources. Nevertheless, PTs are susceptible to faults that can cause costly outages and disruptions. Over the past decades, the technique of dissolved gas analysis (DGA) has been extensively employed in oil-immersed transformer fault diagnosis. There are various methods to identify faults using DGA. Due to its superior accuracy compared to other techniques, the dual pentagon method (DPM) is utilized for fault diagnosis of PTs in this research. On the other hand, implementing DPM on large amounts of DGA data can be challenging. To address this problem, we proposed data-driven algorithms such as tree-based algorithms counting Decision Tree Classifier (DTC), Random Forest Classifier (RFC), eXtreme Gradient Boosting Classifier (XGBC), Light-GBM (LGBM) Classifier, Adaptive Boosting (AdaBoost) Classifier, and Categorical Boosting (CatBoost) Classifier. Furthermore, four data scaling techniques have been used for more effectiveness because the dataset contains outliers. The outcomes of the data analysis and Python simulation demonstrate that the suggested approach performs better than the previous methods. From the simulation analysis, the robust Light-GBM method has achieved an accuracy of 96.08%, and MCC of 95.41%, which is higher compared to the existing techniques.