【作者】 王静莲；

【导师】 刘弘；

【作者基本信息】 山东师范大学 ， 计算机软件与理论， 2006， 硕士

【摘要】 随着科学技术进入一个多学科相互交叉、相互渗透、相互影响的时代,源于模拟某一自然现象或过程并且具有自组织、自学习、自适应等特征的神经网络、细胞自动机和进化计算等研究方向近年来得到了前所未有的蓬勃发展。生物多样性和协同进化论作为生物学新兴的一个重要理论,业已成为生态学的基础理论,并引用到人类社会而成为持续发展概念的基础。应用生物多样性协同进化理论研究生物进化和环境保护,以及指导人类的社会行为,要比达尔文以生存斗争为中心的进化论进步而积极得多;这也正是协同进化算法得到越来越多研究人员青睐的主要原因。进化计算技术在一些设计优化问题中已显示出巨大潜力和深远影响,但距离实现一个能与人的行为相匹配的系统还很远。尤其是在一些创新设计应用(如建筑、艺术、音乐、设计)中,由于评价标准主要依赖于人的主观想法,很难对个体的适应度函数进行描述。系统如何能在执行最优化的过程中,为个体提供进化机会的同时,也最大限度地避免产生退化的可能;而且可以实现一种智能化的评价方法等问题,成为计算机辅助概念设计的研究热点。在客观剖析相关领域存在的实际问题的基础上,本文具体所做贡献如下:1.多个体协同式强化学习的动力学模型的提出。借鉴生物科学中神经元相互刺激与抑制的量化关系,论文采用数学模型描述了多个体协同式强化学习的动力学行为。该模型通过种群内个体适应度之间的相互关联使得表现型低的个体受到抑制,而表现型高的个体得到强化;进而可使种群在后天生存竞争中的整体表现型得到快速提升以自适应更加复杂的生态环境特征。2.一种基于协同式强化学习机制的新协同进化算法的研究。通过分析生物在环境生态系统中的成长模式,论文强调生物进化是一个在生态环境里结合先天遗传进化和后天竞争学习进化的复杂过程以突显在现有进化模拟框架中引入一种后天强化学习思想的必要性;继而在传统遗传算法的基础上,论文采用了上述的多个体协同式强化学习的动力学模型。新算法增强了智能搜索的能力,加快了种群的收敛速度。3.新协同进化算法在计算机辅助概念设计中的应用测试。论文以建筑为例,基于组合原理和实例推理的创新设计机制,将上述新的协同进化算法应用于计算机辅助概念设计中,具体实现了许多构思巧妙、造型新颖的建筑样例。与此同时,实验测试数据也比较说明了本算法的高效运行性能。更多还原

【Abstract】 The researches on artificial neural networks, cellular automata andevolutionary computation characterized by self-organizing and self-learning andadaptation, rooting from simulating some natural phenomenon or the process,havebeen developed greatly as the science and technology entering the time when multidisciplines intercross, seeps mutually and affects mutually.The novel and important theory of biodiversity harmonious and cooperativeevolution has been the foundation of the bionomics and the sustainable developmentconcept.And coevolutionary algorithms have raised more and more attention fromresearchers recently, because it is more positive than Darwin’s evolution theorywhich must take struggle for surviving as the central in the research on biologyevolution and the environmental protection, as well as instructing humanity’ssocial behavior.The evolutionary computation technique had been shown great potential andfar-reaching impact on design optimization, but it is far from realizing a systemof matching the human performance. In particular, in some innovation designapplications fields, such as construction, art, music, design, it is difficultto evaluate the fitness because the measure depends mainly on the human mind. Howthe system can provide the evolution opportunity for the individual at the sametime maximum avoid degenerating in the execution optimization process and mayrealize the intelligent fitness function have become the hot spot in the researchon computer-aided conceptual design.To solve the related questions, there has been specific contribution in thisarticle as follows:1. The proposed novel multi-individual cooperative reinforcement learningmodelReferring from the stimulus and suppression theory of neurons in the physiology,this paper has proposed a novel coevolutionary model-multi-individual cooperativereinforcement learning model.The model made individuals of low phenotypesuppressed and high phenotype reinforced by the relation of individual fitness.Sothe global phenotype accelerated upgrading to self-adapt the complexity of theecological environment.2. Research on the novel coevolutionary algorithm based on the cooperativereinforcement learning mechanismBy analyzing the biological development model in the ecosystem,the paper hasemphasized biological evolution combined congenital/hereditary feature andcompetitive learning.And based on the traditional genetic algorithm,the paper hasintroduced the cooperative reinforcement learning model.The new coevolutionaryalgorithm not only strengthened capacity of intelligent search but alsoaccelerated the convergency of the population.3. The application-test of the new coevolutionary algorithm in thecomputer-aided conceptual design(CACD)The paper has studied the application of the new coevolutionary algorithm inthe computer-aided conceptual design(CACD) based on combination Principle andcase-based reasoning mechanism. And the system has showed many constructionexamples of ingenious idea and innovative shape. Finally, the experiment resultsshowed the new algorithm has higher-performance characteristic than traditionalgenetic algorithm.A study is conducted on coevolutionary technique and its application incomputer-aided conceptual design(CACD) in this paper.It is hoped that it couldpromoted the development of the relate research fields.更多还原