【作者】 陈冬；

【导师】 李智勇；

【作者基本信息】 湖南大学 ， 计算机科学与技术， 2010， 硕士

【摘要】 多目标进化算法(MOEA)是模拟生物进化理论而产生的高性能、自组织、高鲁棒性的多目标优化问题求解方法。MOEA对Pareto前沿的形状不敏感,算法一次运行产生多个Pareto非支配解。正因为这些优点,MOEA已经成为解决多目标优化问题的主流方法。当前MOEA主要依靠非支配排序推动种群朝Pareto前沿进化,在算法的后期非支配排序的推力不足,使得算法的全局寻优能力较弱。本文旨在深入探索和研究多目标进化算法蕴含的优化原理和进化机制,并探索将贝叶斯博弈模型引入到多目标进化算法中,以提高多目标遗传算法的全局寻优能力。本文的主要研究工作包括以下几个方面：(1)讨论了群智能算法的优化原理和进化算法在解决多目标优化问题中的求解模型,介绍了博弈论的基础知识,并对多目标进化算法的收敛性和多样性进行了研究；(2)提出了一种基于贝叶斯博弈模型的多目标遗传算法。将每个待优化目标视为一个博弈参与者,通过多个参与者的博弈共同拉动种群朝Pareto前沿搜索,使得算法表现出更好的收敛性和多样性。本文从理论上分析了基于贝叶斯博弈模型的多目标遗传算法的收敛性,并对算法的性能进行了对比验证,仿真实验结果表明算法在收敛性和解得分布性上优于NSGA-Ⅱ算法。(3)针对网格调度的特点,本文提出了基于SBG-MOGA的网格调度求解模型。本文对网格调度问题做了详细描述,建立了多目标优化问题模型,设计了符合网格调度问题特点的编码方式、精英集合以及遗传操作。最后通过仿真实验,验证了算法的收敛性和可行性。更多还原

【Abstract】 Multi-objective evolutionary algorithms (MOEAs) are proposed from simulating the biological evolutionary theory. These algorithms are.high-performance,self-organization and robustness.MOEAs are not sensitive to the figure of Pareto front and can get lots of results implemented once.Beacause of their advantage, MOEAs have been the mainstream on solving multi-objectives optimization problems.Conventional MOGAs use non-dominated sorting methods to push the population to move toward the real Pareto front.This approach has a good performance at earlier stages of the evolution, however it becomes hypodynamic at the later stages.The objective of this study is to explore and research the optimization principle and the evolutionary mechanism of MOEAs.In this paper we add the static Bayesian game strategy into MOGA and propose a novel multi-objective genetic algorithm(SBG-MOGA),which improves the optimition ability at the later stages of algorithm.The study focuses on the following aspects:(1)This article discusses the principles of swarm intelligence algorithm optimization and the solution model of multi-objective evolutionary algorithm,and then describes the game theory. And also multi-objective evolutionary algorithm convergence and diversity has been studied.(2)Multi-Objective Genetic Algorithm Based on Static Bayesian Game Strategy (SBG-MOGA) has been introduced.In SBG-MOGA, a objective that needs to optimize is regarded as a participant of a game.The game strategy in SBG-MOGA generates a tensile force over the population and this will obtain a better multi-objective optimization performance.The convergence of SBG-MOGA has been proofed in the paper. Moreover, the algorithm is verified by several simulation experiments and its performance is tested by different benchmark functions.(3)According to the characteristic of grid scheduling,this article proposed a grid scheduling solution model based on SBG-MOGA. In the paper, we gave a detailed description of scheduling problems on the grid, established the multi-objective optimization model of grid scheduling problem,and then improved the SBG-MOGA in order solving Grid scheduling problem.Finally, simulation experiments were given to verify the convergence and feasibility.更多还原