Identification and Classification of Incipient Discharges in GIS Adopting Machine Learning Techniques

In the present study, the PD due to particle movement, surface discharge, floating particle and corona defects in GIS are detected by adopting UHF sensor. The FFT analysis of UHF signal for all four defects shows the dominant frequency around 1GHz. To identify and classify the various PD source in GIS, multi classifier machine learning techniques are adopted. The spectral data of the obtained UHF PD signals for each of the four types of discharge signals were obtained after performing the Fast Fourier transform of the UHF signals frequency data. Random Forest and K Nearest Neighbour classifiers are adopted in the present study for PD classification. In order to improve the performance of RF classifier, the inbuilt hyperparameters are optimized for a better performance and accuracy, using a technique called Random Search CV. The modified Random Forest classifier showed significant improvement in the classification accuracy.