Abstract—In this study we present a data driven prediction approach to early prediction of transformer’s fault. To make such prediction we have collected dissolve gas data of transformer. We have solved this problem bagging based ensembled algorithm. Further we have found that our data has imbalanced class examples. To overcome this, we have removed class bias by using Synthetic Minority Over Sampling Technology (SMOTE). SMOTE is best known for generating synthetic data for minority classes. It is also proven to be better than random sampling. SMOTE oversamples the minority classes data by fitting the linear lines among them. In that way we can generate as many data as we want. Thus, it helped us in avoiding overfitting problem. Our empirical results show that proposed framework outperforms the state-of-the-art methods such as BP neural network, and support vector machine. Our method achieves 90.67 % precision accuracy which is better than the base lines. 

Index Terms—transformer fault prediction, Dissolved Gas Analysis (DGA), bagging, support vector machine, ensembled learning

Cite: Lu Peiwen, Huang Yongjing, and Khushnood Abbas, "Transformers Fault Prediction: An Improved Ensembled Method," International Journal of Electronics and Electrical Engineering, Vol. 8, No. 4, pp. 82-87, December 2020. doi: 10.18178/ijeee.8.4.82-87