Bayesian neural network based inductance calculations of wireless power transfer systems

Abstract

This letter proposes a new method for obtaining self and mutual inductances in wireless power transfer (WPT) systems using a Bayesian neural network (BNN). Generally, inductance calculations using a field solver take a huge amount of time. Moreover, due to the complexity of WPT systems, there is no approximate equation for calculating inductances including ferrite shields. In this letter, nine structural parameters of a WPT system are experimentally used as inputs. The experimental results demonstrate that inductances obtained by the proposed method are within 5.1% in the maximum errors and within 1.1% in the mean absolute errors. The proposed method is about 748k times faster than the field solver in the CPU time required to obtain the inductances of one structure.