Machine Learning Based Optimal Design of On-Road Charging Lane for Smart Cities Applications

The rapid advancement of electric vehicle (EV) technology toward environmentally friendly transportation emphasizes the necessity of dynamic wireless charging. However, challenges, such as the initial charging infrastructure cost, power transfer efficiency, and output power pulsation, pose significant limitations to dynamic wireless charging. Overcoming these challenges requires optimizing the design of various functional elements in dynamic charging, including the magnetic coupler, spacing between couplers, high-frequency inverter, and compensators. Despite the nonlinear relationships among these elements, obtaining mathematical relations proves cumbersome. This article proposes an effective machine learning (ML) approach to achieve the optimal design of the charging track, considering the cross-coupling effect. The algorithm not only aids in estimating the infrastructure cost of the charging lane but also predicts optimal design parameters using trained data. The ML approach, which predicts optimal design parameters with a trained dataset, is more efficient with reduced duration than conventional finite element analysis (FEA) tools and stochastic methods. The learning algorithms consider variables such as core structure, cross-coupling effect, and coil flux pipe length. Simulation and experimental prototype validation for a 3.3 kW system demonstrated an impressive efficiency of 93.21%.