Optimal Ferrite Structure Design of DD-Type Magnetic Coupler Based on Improved Deep Reinforcement Learning for WPT System

Optimization of the core structure in the magnetic coupler of a wireless power transfer system is critical for enhancing the power density and system efficiency. This study reveals for the first time that the mutual inductance of DD-type coils initially increases but then decreases as the number of cores in the DD-type receiver increases, contradicting the inherent assumption that mutual inductance is maximized when the core is fully distributed. Furthermore, a core optimization method based on improved deep reinforcement learning is introduced. By integrating a residual structure into the deep neural network, the feature extraction ability of the network is improved. The survival of the fittest strategy is used to identify core locations with high contributions to mutual inductance to improve the optimization speed. The core region is divided into inner and outer layers, and a hierarchical selection strategy is proposed for selecting the two regions separately to reduce the optimization space. A 30-kW prototype and experimental platform were developed to validate the proposed core structure. Compared with the conventional fully filled core structure, the optimized receiver achieved a 28% reduction in core volume and a 2.3% increase in mutual inductance, while maintaining misalignment tolerance, shielding effectiveness, and overall efficiency.