An Efficient Inverse Modeling Method Using Translator-Inspired Neural Network and Dual-Annealing for a Compact WPT System

Abstract

This article proposes a novel prediction method for efficient inverse modeling of a compact wireless power transfer (WPT) system. Impedance matching circuit analysis with admittance (J-) inverter theory is used for coupling analysis. Then, a simple translator-inspired neural network is proposed to predict scattering (S-) parameters from 100 MHz to 1 GHz for each coupling case. The training dataset comprises 2900 randomly generated dimensions and the corresponding S-parameters via full-wave simulation. Consequently, the power transfer efficiency (PTE) can be figured out using the S-parameter, which varies from different coupled modes. Then, a dual-annealing algorithm is leveraged to find the best dimensions. The predicted dimensions, including the transmitter (Tx) and receiver (Rx) layouts, values of the etched capacitors, and WPT distances, meet the design purposes. Compared to the state-of-the-art approaches, the proposed method can efficiently optimize the WPT system with desired frequencies. The proposed inverse model reveals new insights and understanding of the prediction and optimization of inductively coupled WPT.