基于改进遗传算法的多目标决策方法研究及应用

【作者】 谭丹丹；

【导师】 曹斌； 王子牛；

【作者基本信息】 贵州大学 ， 计算机应用， 2007， 硕士

【摘要】 遗传算法作为一种非确定性的拟生态随机优化算法在过去20年中得到了广泛的应用。由于其具有不依赖于问题模型的特性、全局最优性、隐含并行性等特点，正越来越激起人们研究与应用的兴趣。多目标决策问题是从实际应用中产生的，它不论在经济、军事还是高科技领域都有着重要的研究价值。尽管多年来多目标决策问题已有许多求解方法，然而最近十几年来遗传算法已逐渐发展成为解决多目标决策问题的理想方法。但在实际应用中，人们常常发现，遗传算法会收敛到局部最优而不再进化，或者是群体中不能再产生性能超过父代的后代，群体中的各个个体之间非常相似即出现未成熟收敛现象。本文介绍了遗传算法和多目标决策的原理和方法，并围绕遗传算法中的早熟问题对遗传算法进行改进。基于配对策略，为了维持群体的多样性，该文设计的防乱伦遗传算法是有目的地选择配对个体，去除相似个体。然后通过三次试验对该改进算法进行验证，试验结果表明INGA能很好的摆脱早熟，并能搜索到最优值，且搜索的效率也很高。另外，本文以贵州省某药厂ERP系统的生产计划为例，提出多目标生产计划模型，并将改进的遗传算法应用于求解此多目标决策模型，从而使药厂摆脱了以往手工制定生产计划的过程，根据决策者的偏好，智能决策出客观、合理的生产计划。更多还原

【Abstract】 As an uncertain stochastic optimal algorithm, GA is applied in kinds of fields in the past 20 years. And because of its independence, global optimization and implicit parallelism, GA is developed and applied by more and more people.Arising from practical problems, Multi-objective Decision-making Problems (MDPS) plays a significant role in economy, military and other high-tech research fields. Although the approaches for solving MDPS have been available for many years, genetic algorithm have been developed to be ideal techniques for solving MDPS. But in practical application, people discover frequently that the genetic algorithm can converge on the local optimization and no longer evolve, or cannot have the descendant that it is capability cannot surpass it is father is in colony. In colony, each individual is extremely similar and appears the premature convergence phenomenon.In this paper, the elementary theory and methods of genetic algorithm and multi-objective decision-making is introduced. Some improvement of GA on the premature convergence is presented. The improved GA which was designed by this article purposefully chooses mating individual and eliminates similar individual based on mating strategies and maintaining population diversity. Then we test the INGA through two typical functions. The test result indicated that the INGA can get rid of premature convergence well and search the optimal value, the searching efficiency is very well.In addition, we take a ERP productive plan of pharmaceutical factory in the Guizhou Province as an example and propose a multi-objective productive plan model, then apply the improved genetic algorithm to solving it which make the pharmaceutical factory getting rid of the former productive plan process by hand-made. In this way, the impersonal and reasonable productive plan can be decided intelligently by favorite of decision-making person.更多还原