Application of Natural Gradient Boosting and Grid Search in Fault Detection and Diagnosis of Electrical Power Grids

In recent years, the demand of electrical power grids had increased tremendously since; girds are primary sources for meeting the electricity demand. However, due to occurrence of faults continuously the power grids are failing in providing constant supply. Several methods are employed for detecting the faults but they failed in effectively classifying types of faults. Therefore, this research proposes Natural Gradient Boosting and Grid Search (NGBoost-Grid Search) for detection and diagnosis of electric faults in electrical power grid. Initially, the data is taken from publicly available electric fault line detection dataset. Then, these current signals are pre-processed by using Interquartile Range (IQR), Hilbert Huang Transformation (HHT) and min-max normalization. After that, Elastic Net technique is employed for selecting the optimal features from pre-processed signals for eliminating irrelevant signals. Finally, selected optimal features are classified by using NGBoost for distinguishing the fault current signals and healthy current signals, Grid search is integrated to hyper tune for improving the performance of NGBoost and gave effective classification of fault and normal current signals. The proposed NGBoost-Grid Search achieved attained better results such as accuracy (96.47%), precision (97.63%), Recall (89.94%) and Fl-score (93.29%) respectively when compared with existing Support Vector Machine (SVM).