一种求解一般性多学科优化问题的分布式分解方法  

A Distributed Decomposition Method for General Problems in Multidisciplinary Design Optimization

作　　者：张代雨 张贝 王志东[1] 陈旭 吕海宁[2,3] ZHANG Dai-yu;ZHANG Bei;WANG Zhi-dong;CHEN Xu;LU Hai-ning(School of Naval Architecture and Engineering,Jiangsu University of Science and Technology,Zhenjiang 212003,China;State Key Laboratory of Ocean Engineering,Shanghai Jiao Tong University,Shanghai 200240,China;Yazhou Bay Institute of Deepsea SCI-TECH,Shanghai Jiao Tong University,Sanya 572000,China;China Ship Scientific Research Center,Wuxi 214082,China)

机构地区：[1]江苏科技大学船舶工程学院,江苏镇江212003 [2]上海交通大学海洋工程国家重点实验室,上海200240 [3]上海交通大学三亚崖州湾深海科技研究院,海南三亚572000 [4]中国船舶科学研究中心,江苏无锡214082

出　　处：《船舶力学》2023年第6期932-946,共15页Journal of Ship Mechanics

基　　金：国家自然科学基金资助项目(51909110);江苏省高等学校自然科学研究项目资助(19KJB570001)。

摘　　要：为了求解大规模多学科优化设计问题,目前已有多种分布式多学科优化方法被提出,但这些优化方法数值收敛困难,或局限于拟可分多学科设计优化问题,不适用于具有全局约束和全局目标的一般性多学科优化问题。为此本文提出一种分布式分解优化方法,其主要思想是通过两层分解策略,将一般性多学科优化问题分解为一个系统级问题和多个并行子优化问题,并证明该方法的最优解收敛于一般性多学科优化问题的Karush-Kuhn-Tucker点。最后,通过两个典型多学科优化算例对所提方法进行测试验证。结果表明该方法是一种具有良好收敛性能、竞争力强的分布式多学科优化方法。Many distributed multidisciplinary design optimization(MDO)methods have been present⁃ed for the optimal design of large-scale multidisciplinary systems,some of them are known to have nu⁃merical convergence difficulties while some of them are restricted to the quasi-separable MDO prob⁃lem.For solving the general MDO problem with global constraint and objective,a distributed decompo⁃sition method is presented to decompose the general MDO problem into one system-level problem and multiple concurrent sub optimization problems through a two-level decomposition process,whose solu⁃tion is proven to converge to the Karush-Kuhn-Tucker point of the general MDO problem.Two typi⁃cal cases are introduced to investigate the performance of the presented method.The results show that the presented method is a competitive distributed MDO method with a good convergence performance.

关 键 词：多学科设计优化 收敛证明 分解策略 最优灵敏度分析 

分 类 号：U662.2[交通运输工程—船舶及航道工程]