李正良, 彭思思, 王涛. 基于混合加点准则的代理模型优化设计方法[J]. 工程力学, 2022, 39(1): 27-33. DOI: 10.6052/j.issn.1000-4750.2020.12.0925
引用本文: 李正良, 彭思思, 王涛. 基于混合加点准则的代理模型优化设计方法[J]. 工程力学, 2022, 39(1): 27-33. DOI: 10.6052/j.issn.1000-4750.2020.12.0925
LI Zheng-liang, PENG Si-si, WANG Tao. A SURROGATE-BASED OPTIMIZATION DESIGN METHOD BASED ON HYBRID INFILL SAMPLING CRITERION[J]. Engineering Mechanics, 2022, 39(1): 27-33. DOI: 10.6052/j.issn.1000-4750.2020.12.0925
Citation: LI Zheng-liang, PENG Si-si, WANG Tao. A SURROGATE-BASED OPTIMIZATION DESIGN METHOD BASED ON HYBRID INFILL SAMPLING CRITERION[J]. Engineering Mechanics, 2022, 39(1): 27-33. DOI: 10.6052/j.issn.1000-4750.2020.12.0925

基于混合加点准则的代理模型优化设计方法

A SURROGATE-BASED OPTIMIZATION DESIGN METHOD BASED ON HYBRID INFILL SAMPLING CRITERION

  • 摘要: 在工程优化设计中,采用数值仿真模拟计算结构响应需耗费大量的时间和计算成本,给计算密集型的优化设计带来了巨大挑战,因此基于代理模型的序列优化设计方法得到了深入研究和广泛应用。对代理模型的序列优化方法框架进行了简要的概述;针对现有方法中存在的不足,发展了一类模型无关的混合加点准则,使优化循环过程中产生的新样本点分布在当前最小值邻域以及设计空间中交叉验证误差较大的区域,局部开发与全局搜索并行,能够更加准确地找到全局最优解;将发展的混合加点准则引入到代理模型优化框架中,并结合粒子群优化算法,提出了一种高效的代理模型优化设计方法;通过数学算例和工程算例对建议方法进行了验证。算例结果表明,与基于传统加点准则的代理模型优化设计方法比较,建议方法能够兼顾计算精度与效率,具有更好的全局寻优特性。

     

    Abstract: In engineering optimization design, numerical simulation calculation of structural response is expensive and time-consuming, which brings great challenges to compute-intensive optimization design. Therefore, the surrogate-based sequential optimization method has been well studied and widely used. Firstly, the framework of the surrogate-based sequence optimization is summarized at first. Secondly, a model-independent hybrid infill sampling criterion is developed in view of the insufficiency in existing methods. The new sample points generated during optimization process are distributed in the neighborhood of the current minimum value and the region with largest cross-validation error in the design space. The local exploitation and global exploration can be carried out simultaneously to find accurate global optimal solution. Thirdly, hybrid infill sampling criterion is introduced into the surrogate-based optimization framework combined with particle swarm optimization algorithm, and an efficient surrogate-based optimization design method is proposed. Finally, the proposed method is verified by mathematical and engineering examples. Compared with the optimization method by the grounds of traditional criterions, the proposed method can keep the tradeoff of accuracy and efficiency, which has better global optimization characteristics.

     

/

返回文章
返回