变可信度近似模型及其在复杂装备优化设计中的应用研究进展

Survey of Multi-fidelity Surrogate Models and their Applications in the Design and Optimization of Engineering Equipment

变可信度近似模型通过融合不同精度分析模型的数据,可有效平衡近似模型预测性能和建模成本之间的矛盾,在复杂装备优化设计中受到广泛的关注。综述变可信度近似模型及其在复杂装备优化设计中的应用研究进展。概述三类常用变可信度近似建模方法的基本思想,并介绍变可信度近似建模方法研究的最新进展。回顾面向变可信度近似模型试验设计方法的发展现状,包括一次性试验设计方法和序贯试验设计方法。综述直接影响变可信度近似模型优化设计求解精度和效率的两类近似模型管理策略,探讨基于变可信度近似模型的智能优化和可靠性/稳健性优化这两个领域前沿问题。归纳总结变可信度近似模型应用于复杂装备优化设计的现状。针对变可信度近似建模及其优化设计给出了一些应用建议,并指出未来值得深入研究的方向。

Multi-fidelity(MF) surrogate models have attracted significant attention recently in engineering design optimization since they can make a trade-off between high prediction accuracy and low computational cost by augmenting the small number of expensive high-fidelity(HF) samples with a large number of cheap low-fidelity(LF) data. This work summarizes the state-of-the-art of MF surrogate modeling approaches and their applications in engineering design optimization. Firstly, the concept of three types of commonly used MF surrogate models is provided and the developments of extensions of them are reported. Secondly, the design of experiments for the MF surrogate models are summarized, including the one-shot design and sequential design approaches. Thirdly, two model management strategies, which directly determine the accuracy and efficiency of MF surrogate model-assisted design optimization approaches, are presented. Besides, the hot topics, MF surrogate model-assisted intelligent optimization algorithms and reliability/robust optimization are discussed. Fourthly, the applications of MF surrogate models in the practical engineering design domain are summarized. Finally, some suggestions about the usage of the MF surrogate models and their applications are provided, followed by the discussion of the deserved future work.