Support Vector Machine Approach for Calculating the AC Resistance of Air-Core Reactor

Abstract

In this paper, a rapid and accurate machine learning approach is developed to predict the winding ac resistance of air-core reactors. By applying the pairing comparison method to the finite-element simulations of real reactor models, reliable and simplified models are derived by eliminating the factors that have a negligible influence on the winding ac resistance. The support vector machine (SVM) approach is introduced into building a regressive function for calculating the ac resistance of layered windings. In the SVM-based learning algorithm, a 3-degree resistance factor kernel is proposed through factorial experiment and kernel construction. The numerical experiments show that the proposed kernel can achieve better generalization and computational performance.