【摘要】 针对传统值解析法、有限元等作为分析模型进行电机多目标优化设计，存在建模难度大、时间成本高的问题，提出一种基于代理模型的电机优化设计框架，该框架由遗传算法优化的极限学习机(GA-ELM)以及多目标粒子群优化(MOPSO)算法组成，并用于一台永磁直线同步电机(PMLSM)的结构优化。基于单变量扫描、主效应分析以及试验设计(DOE)的方法建立了模型训练样本库，在保证样本质量的同时降低了样本容量、节约了建模时间；采用GA-ELM搭建了电机的代理模型，进一步提高了原模型精度；基于MOPSO优化算法引入扰乱子完成对模型的多目标寻优，获得了三维Pareto最优前沿解集。最后依据优化结果加工样机，实验验证了该优化设计框架所得优化结果的正确性，且结果表明优化后的电机平均推力提高了11.19%,推力波动降低了21.95%。更多还原

【Abstract】 Aiming at the problems of difficult modeling and high time cost in traditional multi-objective optimization design of motors with analysis method and finite element as analysis models, a motor optimization design framework based on surrogate model was proposed. The framework consists of a genetic algorithm extreme learning machine(GA-ELM) and a multiple objective particle swarm optimization(MOPSO) and is used for structural optimization of a permanent magnet linear synchronous motor(PMLSM). Based on univariate scanning, main effect analysis and design of experiment(DOE), a model training sample library was established, which reduces the sample size and saves the modeling time while ensuring the sample quality; The surrogate model built using GA-ELM further improves the accuracy of the original model. Using MOPSO as an optimization algorithm, a multi-objective optimization of the model was conducted, and a three-dimensional Pareto optimal frontier solution set was obtained. Finally, the prototype was processed according to the optimization results. The experiment verifies the correctness of the optimization results obtained by the optimization design framework. The results show that the average thrust of the optimized motor is increased by 11.19%, and the thrust ripple is reduced by 21.95%.更多还原