项目名称: 基于代理模型的实用多目标演化算法研究
项目编号: No.61303028
项目类型: 青年科学基金项目
立项/批准年度: 2014
项目学科: 自动化技术、计算机技术
项目作者: 陈彧
作者单位: 武汉理工大学
项目金额: 25万元
中文摘要: 近年来,多目标演化算法在科学和工程领域得到了广泛的应用。然而,实际应用问题的多目标优化模型的目标函数数学性质复杂,其不规则的高维可行域给高性能多目标演化算法的设计和应用带来了巨大的挑战。本项目拟从个体生成和评价策略等方面入手,采用理论分析与数值试验相结合的方法对多目标演化算法的设计和应用进行深入的研究。其具体研究内容包括:首先,分析种群体中个体所服从的概率分布对算法收敛速度的影响,设计一种能够适应不同可行域的个体生成策略;然后,探讨代理模型近似评价误差对多目标演化算法个体评价效率的影响,建立高效的个体评价策略;最后,设计一种统一的实用多目标演化算法框架,并将其应用到基因调控网络的推断当中。本项目采用理论分析与算法设计紧密结合的研究方案,其顺利实施将为多目标演化算法的研究建立一种新的研究方法和思路,其研究成果对多目标演化算法的理论研究、算法设计和应用具有重要的理论和实践意义。
中文关键词: 多目标演化算法;时间复杂度;代理模型;个体生成策略;基因调控网络
英文摘要: In recent years, multi-objective evolutionary algorithms(MOEAs) have been widely applied in the fileds of science and engineering. However, multi-objective optimization(MO) models of application problems generally have mathematically complicated objective functions, and their irregular and high-dimentional feasible regions have brought great changes to the design and application of high-performance MOEAs. Focusing on the generation strategies and evaluation methods of individuals, this project tries to study in depth the design and application of MOEAs via theoretical analysis and numerical experiment. The detailed contents are listed as follows. Firstly, we analyze the influence of probability distributions of individuals on the convergence speed of MOEAs, and try to design a generation strategy that can accommodate various feasible regions. Then, the affect of surrogate models is investigated, and an efficient evalution strategy can be found. At last, we try to construct a universal framework of MOEA to inference the gene regulatory network. This project employs a research scheme coupling theoretical analysis with algorithm design tightly, and its successful execution can lead to a novel research method of MOEAs. The research resluts contain theoretical and practical values important to the theoretical study
英文关键词: Multi-objective evolutioary algorithms;Time complexity;surrogate model;Generation strategies of individuals;Gene regulatory netowrk