基于MOEA和物理规划的多学科鲁棒协同优化建模与求解策略研究

项目名称： 基于MOEA和物理规划的多学科鲁棒协同优化建模与求解策略研究

项目编号： No.51305073

项目类型： 青年科学基金项目

立项/批准年度： 2014

项目学科： 机械、仪表工业

项目作者： 李海燕

作者单位： 东北大学

项目金额： 23万元

中文摘要： 复杂产品研制过程中通常存在由设计条件、认知能力、近似模型以及环境等引起的不确定性，导致其设计性能不稳定，可能造成巨大损失。对此，本项目开展面向复杂产品的多学科鲁棒协同优化理论研究：(1)采取多阶段优化策略，解决协同优化优化结果对初始点选取敏感和松弛因子选取较难的问题；(2)建立基于协同优化框架的耦合不确定性传播模型，以解决不确定性在各学科间通过耦合变量相互传播而导致的计算复杂性；(3)对于子学科不具有物理目标的问题，设计融合鲁棒协同优化框架与MOEA进化过程的途径；(4)针对物理规划和MOEA两种方法，探求具有物理目标子学科的一致性目标函数的特殊处理方式，在保证学科间一致性的前提下，提出提高子学科物理目标鲁棒优化性能的方案；(5)研发多学科鲁棒协同优化原型系统，验证研究成果的有效性。本项目研究对面向复杂产品的多学科设计优化理论做出有益探索,可在一定程度上提高多学科设计优化技术的工程实用性。

中文关键词： 多学科设计优化；不确定性；鲁棒协同优化；线性物理规划；NSGA-II算法

英文摘要： The design process of complex product is usually affected by the uncertainty factors caused by design condition, cognitive ability, approximate model, environment, etc., which may lead to the performance instability and huge losses. In this project, therefore, the following theories on multidisciplinary robust collaborative optimization are investigated to improve the optimization performance of complex product: (1) The multi-stage optimization strategy is employed to deal with the problems that the result of collaborative optimization is sensitive to the initial point and a rational relaxation factor for equality-constrained system level compatibility requirement is difficult to determine; (2) A coupling uncertainty propagation model based on the bi-level optimization framework of collaborative optimization is proposed to reduce the computational complexity of interdisciplinary uncertainty caused by the transmission of coupling variable; (3) The combination mode of robust collaborative optimization framework with the evolution process of multi-objective evolutionary algorithm is presented for the multidisciplinary design optimization problem without physical objectives in subsystems; (4) The special treatment method sloving the interdisciplinary compatibility function of the subsystem with physical objectives i

英文关键词： Multidisciplinary design optimization；Uncertainty；Robust collaborative optimization；Linear physical programming；NSGA-II algorithm