基于学习技术的多目标进化算法重组算子研究

项目名称： 基于学习技术的多目标进化算法重组算子研究

项目编号： No.61273313

项目类型： 面上项目

立项/批准年度： 2013

项目学科： 自动化技术、计算机技术

项目作者： 周爱民

作者单位： 华东师范大学

项目金额： 80万元

中文摘要： 多目标优化问题(MOP)是在科研与应用领域广泛存在的一类挑战性问题。由于MOP最优解往往是一个集合(Pareto最优解集)，进化算法等启发式方法已成为求解MOP的主流方法。目前多目标进化算法(MOEA)主要关注如何维护搜索种群的多样性和收敛性等算法框架(包含选择算子)的研究，忽略了如何从父体产生新个体(即重组算子)的研究。有效利用问题相关信息是启发式算法成功的关键之一。本项目拟着重研究MOEA重组算子的设计：深入分析并获取MOP最优解集和MOEA算法框架的先验信息，使用学习技术挖掘搜索群体中隐含最优解集的后验信息，在重组算子中采用合适的模式描述这些信息并用于指导新个体的产生。在此基础上设计高效的MOEA算法，并用图像处理等实际问题验证新算法的有效性。通过深入研究MOP Pareto最优解集特性和基于学习技术的MOEA重构算子设计原理，本项目的实施将为MOEA的设计和应用提供新思路。

中文关键词： 多目标优化；演化算法；学习；重组算子；图像处理

英文摘要： Multiobjective optimization problems (MOP) are a type of challenge problems in both scientific research and real-world application. Since the optimum of an MOP contains a set of solutions, named as Pareto optimal set, heuristic methods such as evolutionary algorithms have attacked much attention in recent years to approximate the Pareto optimal set. Currently, most of multiobjective evolutionary algorithms (MOEA) focus on how to maintain a set of solutions which is as diverse as possible and as close to the Pareto optimal set as possible, i.e., the framework(including selection procedure); but overlook how to generate new solutions from the parents, i.e., the recombination procedure. For this reason, this project addresses the recombination operator in MOEAs. The prior information about the MOPs and the MOEA framework will be extracted analytically, and the poster information from the population will be learned by machine learning techniques. The information will then be used to guide the generation of new trial solutions. New MOEAs based on the recombination operators will be validated by some real-world applications including image processing problems.By this project, (1) the properties of the MOPs and the principals to design recombination operators will be studied; and (2) new ideas and methods will be sugg

英文关键词： Multiobjective Optimization；Evolutionary Algorithm；Learning；Reproduction Operator；Image Processing