异步并行的分布式协同进化MDO算法研究  

Asynchronous parallel distributed coevolutionary multidisciplinary design optimization

作　　者：陈琪锋[1] 戴金海[1] 

机构地区：[1]国防科技大学宇航科学与工程系,长沙410073

出　　处：《宇航学报》2002年第4期57-61,共5页Journal of Astronautics

摘　　要：为满足分布式计算机网络环境下多学科并行设计优化协调的要求 ,研究了分布式协同进化 MDO算法的网络异步并行实现 ,给出了异步并行的分布式协同进化 MDO算法。对导弹设计的计算实例 ,与单机上顺序执行的分布式协同进化 MDO算法相比 ,异步并行的分布式协同进化 MDO算法可有效缩短运行时间 ,而二者的收敛性能相当。算法基于 CORBA/ C++实现 ,灵活性、容错性好 。Adopting merits of evolutionary algorithm,distributed coevolutionary multidisciplinary design optimization (DCMDO) algorithm can be used directly to problems with discrete and integer design variables or with disjointed design space.To satisfy the need of multidisciplinary concurrent design optimization and coordination in modern distributed heterogeneous computer network environment,in this paper,we investigate the parallelization of DCMDO,and give an asynchronous parallel version of DCMDO algorithm- asynchronous parallel distributed coevolutionary multidisciplinary design optimization (APDCMDO) algorithm.The APDCMDO was executed on local computer network in an asynchronous parallel way based on Client/Server model.Not only inter disciplinary design optimization was parallelized,but also inner disciplinary design optimization was executed in parallel.Each discipline can have multiple optimization nodes concurrently.Optimization nodes can be added or shut down dynamically without affecting others.This makes APDCMDO both flexible and having good fault tolerance.The APDCMDO algorithm was applied to a missile aerodynamic shape/engine/control multidisciplinary design optimization problem.The convergence property of APDCMDO algorithm is close to single computer sequentially executed DCMDO algorithm,but the asynchronous parallel version effectively shortens the design time.The code of APDCMDO is based on CORBA/C++,thus can be conveniently used in CORBA based flying vehicle integrated design environment.

关 键 词：多学科设计优化 导弹设计 协同进化算法 并行计算 异步计算 

分 类 号：V423[航空宇航科学与技术—飞行器设计]