元演变自适应进化计算研究

Modern evolutionary algorithms, which incorporate mechanisms of local search and population diversity preservation, are important methods for solving NP-hard problems. Given a problem, how to adaptively employ the methods of the algorithm's core components (i.e., local search, population diversity preservation, recombination operation) as well as their parameter values during the evolution, thus efficiently and effectively delivering the solution, is currently a very hot research topic. Along with the continuously emerging NP-hard problems, addressing such a topic has become an urgent task. In this proposal, we first construct a meta-evolution based adaptive evolution framework, which organically combines the adaptive usage as well as dynamic evolution of the methods of the algorithm's core components. Based on the framework, we then propose a set of effective theories, methods and techniques to systematically study the above topic and give the overall solution. Specifically, the proposal will study, step by step, how to achieve effective adaptation of the methods of the above three core components, and give several theoretical achievements as well as efficient, effective and robust evolutionary computation, which will be further utilized to solve the clustering problem. Such a proposal will greatly enrich and expand the existing research both from the theoretical and technical aspects. The outcomes of the proposal possess high theoretical values and the developed framework can be applied to adaptively, efficiently and effectively solve various optimization and search problems.

融合局部搜索及群体多样性保持机制的现代进化算法是解决NP-困难问题的重要方法。给定一个问题，如何在算法运行过程中自适应使用其核心部件（局部搜索、群体多样性保持、算法基本操作）的方法及其参数值来快速、有效给出问题解是目前一个热点课题。随着NP-困难问题不断涌现，研究解决这一课题已成为迫切需求。本项目拟建立元演变自适应进化框架，将部件方法的自适应使用及其动态演变有机结合为一体，并基于该框架提出一套有效的原理、方法和技术来系统研究该课题，给出整体解决方案。具体的，项目将针对如何实现上述三大核心部件方法的有效自适应使用展开层层递进研究，给出若干理论成果以及快速、有效且具普适性的进化计算，并实际解决聚类这一数据挖掘领域极具挑战性的问题。本项目的研究将极大丰富、拓展现有成果，力图从理论与技术层面完善自适应进化算法理论体系，成果具较高理论价值，开发的算法框架可应用于各种优化问题的自适应、快速、有效求解。

本项目根据计划研究建立元演变自适应进化框架，并基于该框架提出一套有效的原理、方法和技术，对现代进化算法的三大核心部件的自适应使用展开系统研究，探索其与算法性能以及普适性的关系，给出快速、有效且具普适性和鲁棒性的进化计算方法，并解决相关优化应用问题。具体工作包括研究：1、建立元演变自适应进化框架；2、元演变自适应多局部搜索；3、元演变自适应多小生境方法；4、元演变自适应算法基本操作控制；5、应用于数据分析、模式识别、控制、安全、信息隐藏等领域。期间邀请了国内外专家交流访问5人次，培养硕士生7名，在IEEE Transactions等期刊和会议上发表带标注论文30篇，其中包括5篇领域内的顶级、中科院SCI 1区期刊论文，申请发明专利4项。