Estimation of Support Vector Machine Parameters Based on Breakdown Voltage Characteristics

In this work, breakdown voltage characteristics of vegetable oil as insulation medium is studied. Breakdown voltage is measured under different electrode gap and ramp rate of applied voltage. Weibull distribution is used to analyse the measured results. The time to failure and the corresponding voltage magnitude depends on electrode gap and ramp rate. The breakdown results are further processed to train Support Vector Machine (SVM), a machine learning algorithm. The electric field features corresponding to breakdown voltage are extracted from electric field distribution and are used to train Support Vector Machine. The parameters that configure SVM during training process for prediction or classification of new data are estimated. The breakdown mechanism influences values of these parameters.