基于小样本迁移代理模型的高速永磁电机应力优化
Stress Optimization of High-speed Permanent Magnet Motor Based on Few-shot Transfer Learning Surrogate Model
谢冰川 1张岳 2刘光伟 1张凤阁 1李永胜3
作者信息
- 1. 沈阳工业大学电气工程学院,辽宁省 沈阳市 110870
- 2. 山东大学电气工程学院,山东省 济南市 250100
- 3. 山东天瑞重工有限公司,山东省 潍坊市 261000
- 折叠
摘要
针对目前电机代理模型样本空间计算效率低的问题,该文提出迁移径向基函数神经网络代理模型.该模型的样本信息源于足量可快速获取或先验积累的低可信度数据和少量高可信度的数据.基于高速永磁电机应力优化问题,研究该模型对小样本的学习能力、对电机相似拓扑结构的学习能力以及基函数对迁移模型精度的影响,证明该模型在工程优化中的效率优势.最后展望迁移学习技术在电机优化中的发展趋势.
Abstract
This paper proposes a transfer learning radial basis function neural network surrogate model to improve the computational efficiency of the sample space of the traditional motor surrogate model.The sample data of this model originate from enough low-fidelity data that can be quickly obtained or prior cumulative and few high-fidelity data.Based on the stress optimization problem of the high-speed permanent magnet motor,the learning ability of the model with few-shot samples,the learning ability to similar topological structures,and the influence of the radial basis function of the model are studied,and this model is proved efficient in the engineering optimization field.Finally,the development trend of transfer learning technology in motor optimization is prospected.
关键词
迁移学习/代理模型/径向基函数神经网络/高速永磁电机/电机优化Key words
transfer learning/surrogate model/radial basis function neural network/high-speed permanent magnet motor/motor optimization引用本文复制引用
基金项目
国家重点研发计划项目(2021YFE0108600)
国家自然科学基金项目(52077121)
国家自然科学基金项目(51920105011)
出版年
2024
国家科技期刊平台
NSTL
万方数据