基于动态罚因子的多学科协同优化算法及其在船舶设计中的应用

针对标准协同优化算法求解复杂系统工程问题的缺陷,提出了一种改进的协同优化算法,并将其应用于油船总体概念设计阶段。改进协同优化算法将系统级一致性约束最优化问题通过罚函数方法转化为一个无约束优化问题。同时,给出了两种不同的基于差异信息的动态可调罚系数,以保证在优化初期,系统级设计变量与学科级共享变量相差较大时,惩罚力度也大,促使一致性差异在总目标函数中占主导地位,则一致性差异将迅速下降。随着优化的进行,罚系数变小,惩罚力度减轻,目标函数的收敛加快。通过对MDO测试函数算例与标准协同优化和其他典型的改进协同算法的比较,验证了该方法在优化结果的可靠性和稳定性等方面有优势。最后,应用改进的协同优化算法求解以油船造价为系统级目标协同浮性与稳性、快速性等4个子学科的多学科优化问题以体现其工程实用性。

Application of improved multi-discipline collaborative optimization in ship conceptual design based on dynamic penalty factors

Facing the shortcomings of traditional collaborative optimization, such as time-consuming, be-ing sensitive to the initial points and not converging. An improved collaborative optimization is presented and then is applied to ship conceptual design. The main feature of the proposed method is converting system-level consistency constrained optimization problem to an unconstrained optimization problem based on penal-ty function. To ensure the efficiency of the proposed method, two types of dynamically adjustable penalty factors which always reflect the different information are presented. The effects of dynamically adjustable penalty factors are reflected in the optimization process. In the early stage, the larger the difference between the system-level design variables and discipline shared variables is, the larger the punishment is. So that the consistency difference would dominate the overall objective function and prompt the consistency differ-ences decline rapidly. As the process of optimization, dynamically adjustable penalty factors decreased, the convergence of the objective function would be accelerated. Comparing with the traditional collaborative op-timization and other typical improved methods via a numerical example, the better convergence, stability and reliability of the presented collaborative optimization are demonstrated. Finally, this improved collaborative optimization is used to solve an oil tank conceptual design optimization problem that with a cost system and four performance sub-disciplines and a satisfied optimization result is also achieved.