论文部分内容阅读
针对工程电磁优化设计问题提出了一种基于统计近似模型和序贯采样技术的序贯推断方法.其中模型部分包括径向基和紧支径向基模型及其推断.序贯采样和优化过程分为粗优化和精优化两个循环过程.这种序贯优化技术相对于传统的直接优化方法有非常明显的优势.它能以较小的有限元采样点构造统计近似模型,并逐次缩减优化区域,进而极大地降低了有限元的计算量.同时序贯方法还可以为优化算法提供先验信息.最后为了说明新方法的有效性,分别以数值模拟和IEEE TEAM Workshop的基准优化问题来进行分析.结果表明:新方法的有限元计算量不足直接优化方法的1/10,且在优化目标的精度上也满足设计的要求.
A sequential inference method based on statistical approximation model and sequential sampling technique is proposed for the engineering electromagnetic optimization design problem, in which the model part includes the radial basis and the compact support radial basis model and its inference, and the sequential sampling and optimization process For the two processes of rough optimization and fine optimization.This sequential optimization technology has obvious advantages over the traditional direct optimization method.It can construct statistical approximation model with smaller finite element sampling points and reduce the optimized area , And then greatly reduce the computational load of finite element method.At the same time sequential method can also provide a priori information for the optimization algorithm.Finally, in order to illustrate the effectiveness of the new method, the numerical simulation and IEEE TEAM Workshop benchmark optimization problems are analyzed The results show that the new method has less than one tenth of the direct calculation method and meets the design requirements in the optimization of the target precision.