Pre-processing Techniques for Leakage Current Data: Enabling Machine Learning in High Voltage Insulator Monitoring

Leakage current flows over polluted insulators having pollutants covering the insulators, thus, making leakage current a useful indicator of pollution levels. In this study, to obtain raw leakage current data, insulator samples with different contamination levels representing different pollution classes were prepared and they were subjected to clean fog test under high voltages. High resolution leakage current, temperature, and humidity were recorded through advance Data Acquisition (DAQ) System. The time to insulator flashover varied for each insulator sample thus varying the leakage current data sample sizes. Moreover, Electromagnetic interference was observed in the temperature and humidity data, which was removed through filtering. Portions of the leakage current data were selected from each pollution class to create uniform size datasets. Each dataset is further split into 300 sub-datasets which will be further used for feature extraction and machine learning training. This paper outlines the experimental setup, data collection, and preprocessing techniques applied on insulator Leakage Current to make it ready for training machine learning models.