Optimal IPT Core Design for Wireless Electric Vehicles by Reinforcement Learning

Abstract

In this article, optimal inductive power transfer (IPT) core structures for wireless electric vehicle (WEV), which can be derived by optimal reinforcement learning (RL) algorithms, are newly proposed. Because the IPT cannot be theoretically analyzed to find a maximum value of mutual inductance for the optimal core structure design, intuitive and iterative process based on finite element method analysis are usually implemented. This conventional method, however, is not preferred due to numerous possible combinations and computation times. For this reason, RL algorithms are designed to optimize nonlinear system design, enabling the WEV IPT to be efficiently designed with high mutual inductance, even in the presence of severe misalignment conditions. Contrary to the conventional RL algorithm for the IPT core design, the proposed RL algorithm can follow higher mutual inductance by shorter episodes; hence, 50% of computation time reduction and 2% of maximum mutual inductance were achieved. A prototype of WEV IPT system designed by the proposed RL algorithm was fabricated, satisfying the standard J2954 of the society of automotive engineers for WPT3/Z3 case. As a result, it is found that the proposed WEV IPT can be manufactured, considering the desired number of cores for reasonable cost and weight of the vehicle assembly.