Ensemble Learning Framework to Detect Partial Discharges and Predict Power Line Faults

Partial discharges (PD) are small discharges which, in the medium and long term, can severely damage electric transmission lines’ equipment. If not timely solved, partial discharges can lead to a permanent damage and stop the equipment from functioning properly. A solution to this problem could be implemented on current standard electrical grids as part of their transition towards smart grids. In this work, we propose an ensemble learning approach to predict partial discharges. The ensemble model is formed of four submodels based on deep Recurrent Neural Networks (Long Short-Term Memory) and linear regression. Multiple tests, various feature extraction techniques and hyperparameter optimizations were conducted. The obtained results are promising and show the high performance obtained by the proposed model with an accuracy of 97.3%, outperforming many previous works in the field.