基于代理模型的YASA AFPM电机多目标分层优化设计  

Multi-objective hierarchical optimization design of YASA AFPM motor based on surrogate models

作　　者：刘慧娟[1] 刘广冬 易元元 LIU Huijuan;LIU Guangdong;YI Yuanyuan(School of Electrical Engineering,Beijing Jiaotong University,Beijing 100044,China)

机构地区：[1]北京交通大学电气工程学院,北京100044

出　　处：《电机与控制学报》2025年第3期59-73,共15页Electric Machines and Control

基　　金：北京交通大学自然科学横向项目(E24L02100)。

摘　　要：为了提高YASA AFPM电机的多目标电磁优化效率,降低优化算法复杂性和三维有限元分析成本,提出一种基于代理模型的多目标分层优化策略。首先,根据灵敏度分析,结合目标偏好,将优化目标和优化变量分为两层。然后,对于第一层5维变量的优化问题,建立一种Kriging代理模型,对模型精度进行了检验,并与人工神经网络和支持向量机代理模型进行了对比分析;对于第二层3维变量的优化问题,建立多项式响应面代理模型,研究优化变量之间的交互影响。最后,结合非支配排序遗传算法II得到最优Pareto解集,在可行解中选取转矩脉动最小的方案为最优解,实现转矩脉动降低68.9%,效率提升0.624%,成本下降2.2%,同时满足输出功率要求,验证了优化方法的有效性。In order to improve efficiency of multi-objective electromagnetic optimization of YASA AFPM motor,reduce the complexity of optimization algorithm and the cost of three-dimensional finite element simulation,a multi-objective hierarchical optimization strategy based on surrogate model was proposed.First,according to the sensitivity analysis,combined with the target preference,the optimization objectives and optimization variables were divided into two layers.Then,for the first layer of 5-dimensional variable optimization problem,a Kriging model was established,accuracy of the model was tested,and the comparison with artificial neural network and support vector machine surrogate models was analyzed.A polynomial response surface surrogate model was established for the second layer optimization problem with 3-dimensional variable,and the interaction between the optimized variables was studied.Finally,the optimal Pareto solution set was obtained by combining non-dominated sorting genetic algorithm II(NSGAII),and the scheme with the smallest torque ripple was selected as the optimization result among the feasible solutions.The torque ripple was reduced by 68.9%,the efficiency was improved by 0.624%,and the cost was decreased by 2.2%,satisfying the output power requirements.Effectiveness of the optimization strategy was validated.

关 键 词：无磁轭分段电枢轴向磁通永磁电机 多目标分层优化设计 分段偏移永磁体 Kriging代理模型 多项式响应面模型 非支配排序遗传算法Ⅱ 

分 类 号：TM351[电气工程—电机]