Optimization Model Design of Permanent Magnet Switch Based on GAN and Finite Element Analysis

Abstract

With the advancement of intelligent technologies, the optimization design of permanent magnet switches (PMS) faces increasingly complex electromagnetic performance requirements. Traditional methods, relying on experience and physical experiments, are inefficient and costly. This paper proposes a hybrid optimization model (FEGAN-DQN) based on Generative Adversarial Networks (GAN), Finite Element Analysis (FEA), and Deep Q-Networks (DQN), aiming to achieve efficient and accurate PMS optimization design. Experimental results demonstrate that FEGAN-DQN excels on the PMSM and MagLev datasets, achieving a Mean Squared Error (MSE) of 1.23 on the PMSM dataset, significantly outperforming traditional models such as k-NN (26.10) and RF (16.42). This model effectively optimizes electromagnetic performance, providing an intelligent solution for industrial design and advancing development in the power and automation fields.