Combined Random Forest and NSGA-II for Optimal Design of Permanent Magnet Arc Motor

This article presents an optimization design method for a double-stator hybrid excited permanent magnet arc motor (DS-HE-PMAM). The proposed optimization method combining the machine learning algorithm random forest (RF) and the nondominated sorting genetic algorithm-II (NSGA-II) contributes to achieving high average torque, low torque ripple, high back electromotive force (EMF), and low total harmonic distortion of the back EMF. First, the motor structure and working principle of the DS-HE-PMAM are illustrated. The selection of parameters to be optimized is determined based on an analytical model. Then, a variable importance measure-based new sensitivity analysis method is implemented to evaluate the influence of each structural parameter on the selected design objectives. The finite-element analysis (FEA)-based DS-HE-PMAM model is developed to obtain the sample data regarding input structural parameters and output design objectives. Based on the sample data, a powerful machine learning algorithm called RF is employed to fit the function relationship between output design objectives and input structural parameters. After that, an intelligent search algorithm named NSGA-II is introduced to search for the optimal solution to the structural parameters combination and obtain the optimal motor performances. Finally, the electromagnetic characteristics of the initial and optimized models of the DS-HE-PMAM are compared and analyzed, and both FEA and prototype experiments verify the feasibility and superiority of the proposed optimization method.