Prediction of dielectric parameters of an aged mv cable: A comparison of curve fitting, decision tree and artificial neural network methods

Abstract

In most dielectric diagnosis studies on medium voltage cables, aging methods, which requires quite long measurement durations, are preferred and dielectric performance of cable is generally measured at the end of the test period. In addition, changes that occur during the aging cycle should be investigated. Predicting the future performance of a cable by using dielectric parameters measured during the aging cycle is quite important in terms of estimating possible failures. In this regard, the effectiveness of interpolation and extrapolation methods commonly used in the literature should be investigated in order to shorten aging durations and to predict future insulation performance. In this study, 12/20.8 kV rated voltage and XLPE insulated medium voltage cable was aged with 60 kV (5•U0) overvoltage for 80 cycles of 15 min. After each aging cycle, dielectric parameters (dissipation factor (tan δ), dielectric losses (Pk) and capacitance (C) were measured at rated voltage and mains frequency. Following the measurements, interpolation and extrapolation analyses were performed using artificial neural network (ANN), decision tree (DT) and curve fitting (CF) methods. As a result, interpolation and extrapolation performances of methods are comparatively discussed and introduced. It has been determined that ANN algorithm is the most successful method.