基于计算试验设计与代理模型的飞行器近似优化策略探讨

Discuss on Approximate Optimization Strategies Using Design of Computer Experiments and Metamodels for Flight Vehicle Design

现代飞行器设计优化中广泛应用高精度分析模型以提高设计可信度与综合性能,但是也带来了计算复杂性问题。为了有效缓解计算耗时的问题,基于计算试验设计与代理模型的飞行器近似优化策略成为研究热点。近似优化策略通过构造合理的近似模型引导优化过程快速收敛到最优解,从而达到降低计算成本,缩短设计周期的目的。根据广泛的文献调研,对飞行器近似优化策略的发展现状进行详细探讨。给出近似优化策略的定义、求解流程、特点以及关键技术。对计算试验设计方法、代理模型方法、精度校验与代理模型选择方法等技术进行综述。围绕静态与自适应两类近似优化策略,重点讨论典型的代理模型管理与更新策略与收敛准则。针对飞行器多学科设计优化问题,探讨近似优化策略与分解策略在求解效率与收敛性方面的技术特点。通过数值算例对各项关键技术的特点进行较详尽的对比分析与总结,并依托飞行器设计优化工程实例对近似优化策略的综合性能进行探讨,指出不同近似优化策略的适用范围。研究结果表明,飞行器近似优化策略在优化效率、全局收敛性以及鲁棒性等方面体现出较显著的优势,具有良好的工程应用前景。指出飞行器近似优化策略的未来研究方向。

Although the wide use of high fidelity analysis models in modern flight vehicle design is beneficial to improving the design credibility and overall performance of flight vehicle systems, it also causes high computational cost and complexity. In order to alleviate the computational difficulty, approximate optimization strategies using design of computer experiments（Do CE） and metamodels for flight vehicles have become more and more popular, which construct reasonable approximation models to enable efficient convergence to the optimal solution with much less computational burden and shorter design cycles. An extensive literature survey of the state-of-the-art of approximate optimization strategies in the context of flight vehicle design is provided. The definition, solution process, features and key technologies of approximate optimization strategies are presented, and then the development of Do CE, metamodeling, accuracy assessment and metamodel selection, as well as corresponding typical methodologies, are reviewed. Moreover, metamodel management and updating schemes and termination criteria used in both static and adaptive approximate optimization strategies are specifically discussed. The efficiency and convergence behaviors of approximate optimization strategies for solving multidisciplinary design optimization（MDO） problems are analyzed via comparison with decomposition-based strategies. A number of well-known numerical benchmark problems are employed to discuss the characteristics of the aforementioned key technologies. Furthermore, the overall performance and applicability of different approximate optimization strategies are discussed through flight vehicle design applications. Comparative studies demonstrate that approximate optimization strategies show obvious advantages in optimization efficiency, convergence and robustness, which are important for engineering applications. Future research directions of approximate optimization strategies are given.