Multi-objective optimization of high-speed permanent magnet machine based on fast hierarchical surrogate model integrating PDP-LightGBM

Purpose This paper aims to tackle the challenges in optimizing high-speed permanent magnet (HSPM) machines, which exhibit intricate coupling and high-dimensional nonliearity. Traditional design optimization faces challenges like high computational cost, long duration and inability to accurately depict their complex nonlinearity. To address this, this paper introduces a swift hierarchical computational surrogate model based on Partial Dependence Plots (PDP)-LightGBM, integrated with Grey Wolf Optimization (GWO) for multi-objective optimization of HSPM machines. Design/methodology/approach Aiming at the problems of difficult optimization, long optimization time and high cost of HSPM machines, a method combining the fast hierarchical computational surrogate model based on PDP-LightGBM with the GWO algorithm is proposed to achieve high average torque, low torque ripple, high power density and high efficiency. Findings While saving computing resources, the optimal HSPM machine was found, which features high average torque, low torque ripple, high power density and high efficiency. Finally, the optimization results of the algorithm are verified through the finite element analysis method. Originality/value This paper proposes a PDP-LightGBM hierarchical surrogate model for HSPM machines. Targeting high-dimensional nonconvex electromagnetic optimization, it compares Particle Swarm Optimization (PSO) and GWO in multi-objective optimization of HSPM machines, showing GWO outperforms PSO in solution quality. The method effectively addresses surrogate model construction challenges, reducing computational cost and resource consumption significantly.