机械系统动力优化设计的并行遗传算法研究.pdfVIP

第八届全国振动理论及应用学术会议论文集，上海，2003 年 11 月 机械系统动力优化设计的并行遗传算法 1,2 1 韦凌云 ，赵玫 （1.上海交通大学振动冲击噪声国家重点实验室，上海 200030；2. 广西大学计算机信息与工程学院，南宁 530004） 摘 要： 针对机械系统动力优化设计问题非线性程度高、优化计算涉及动力学分析等特点，提出了一个 通用、易用和高效的并行遗传算法。该算法提供了一个新的全局寻优的并行遗传算法框架，采用自适应交叉 和变异概率进行交叉和变异操作，并根据个体间的相似性进行分类，在种群内形成的若干个小生境。再以同 步并行方式，在各个小生境内采用单纯形法进行快速地局部搜索。之后，从整个种群中选择优秀个体构成单 纯形，采用异步并行方式进行单纯形全局搜索。该算法有效地保持了局部搜索和全局探测的平衡，缓解了遗 传算法固有的过早收敛现象和局部搜索能力弱的问题。通过主从式模型实现并行计算，获得满意的并行效率， 提高了优化效率和优化质量。一个五自由度汽车悬挂系统的仿真算例验证了算法的效率、可行性和可靠性。 关键词： 遗传算法；并行计算；动力优化设计；机械系统 A parallel genetic algorithm for dynamic design optimization of mechanical systems WEI Ling-yun1,2, ZHAO Mei1 (1. State Key Lab of Vibration, Shock and Noise, Jiaotong University, Shanghai 20030; 2. School of Computer and Information Engineering, Guangxi University, Nanning 530004) Abstract: A parallel genetic algorithm is proposed in this paper to solve the dynamic design optimization problems, which is highly nonlinear and involved in complex dynamic analysis in very iterative step, in mechanical design. It provides a new architecture of parallel genetic algorithms, which organically merges the niche techniques, adaptation strategy of control parameters and Nelder-Mead simplex search into genetic algorithms to develop an available and robust global optimization. The proposed parallel algorithm divides the population into several niches according to individual similarity and then performs the simplex search with synchronous parallel mode to quickly exploit those niches, which have more than two individuals. After that, another simplex search is performed with asynchronous parallel mode to quickly locate the global 基