异步并行的分布式协同进化MDO算法研究

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，研究了分布工协同进化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行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，便于在飞行器一体化设计环境中应用。

Asynchronous parallel distributed coevolutionary multidisciplinary design optimization

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.