面向超多目标优化的分解进化算法

项目名称： 面向超多目标优化的分解进化算法

项目编号： No.61473241

项目类型： 面上项目

立项/批准年度： 2015

项目学科： 自动化技术、计算机技术

项目作者： 张青富

作者单位： 香港城市大学深圳研究院

项目金额： 80万元

中文摘要： 在许多科学研究与工程应用领域，许多优化问题需考虑三个以上相互冲突的目标函数，忽视其中任何一个目标就有可能带来严重的负面影响。这些问题被称为超多目标优化问题。多目标进化算法是目前求解两个或三个目标优化问题的主流方法，但是随着目标个数的增加，这些算法的性能常会急剧下降。超多目标优化研究尚处于起步阶段，它是目前多目标进化研究领域最重要和最富有挑战性的课题之一。国内外的一些初步研究表明，分解进化算法是求解超多目标优化的最有潜力的途径之一。本项目将在超多目标优化测试题构造、算法评估指标设计、适合于超多目标分解进化算法的分解技术，超多目标分解进化算法中的动态计算资源分配方法以及权重调整策略等方面展开深入研究。本项目还将结合领域知识求解汽车侧面碰撞超多目标优化问题。本项目的顺利实施，一方面将解决超多目标优化研究中的一些基本问题；另一方面将提供求解超多目标优化问题的有效分解进化算法。

中文关键词： 进化计算；计算智能；多目标优化

英文摘要： The growing competition in industry is enforcing the need for faster, better and cheaper products and processes. Nowadays many real world engineering applications has to consider many intrinsically conflicting functional requirements (objectives), subject to constraints on resource and technical specifications. Failure to account for any objective may imply punitive cost. An Example is an eleven objective car side-impact problem. These applications have contributed to growing emergence of many-objective optimization problems. Although multiobjective evolutionary algorithms (MOEA) have been widely accepted as a mainstream multiobjective optimizer for dealing with two or three objectives, most current MOEAs fail to scale up with the number of objectives. Many-objective evolutionary optimization is still in its early infancy and many real-world many-objective problems can not well be solved. This proposed project is to address this challenge. To this end, it will design many objective test suites which can resemble real world problems, propose new performance indicators for evaluating performances of many-objective algorithms, investigate decomposition techniques, dynamic computational resource allocation methods and weight adjustment strategies for many-objective optimization, and develop new decomposition based algorithmic framework for many-objective optimization and apply them for solving a car side-impact problem.

英文关键词： Evolutionary Computation;Computational Intelligence;Multiobjective Optimization