【作者】 杨城；

【导师】 孙世新；

【作者基本信息】 电子科技大学 ， 计算机应用技术， 2009， 博士

【摘要】 目前，复杂性科学已经成为现代系统科学的一个主要研究方向，被誉为“21世纪的科学”。由于复杂系统自身的非线性、层级性等复杂性特征，基于还原论思想的传统理论和研究方法在复杂性研究中往往显得力不从心，而基于复杂适应系统理论的建模与仿真技术，采用演化的观点，将系统中的单元看作“活的”适应性主体，通过在计算机中建立现实对象的多主体模型，来研究具体的复杂系统。随着计算机科学技术的迅猛发展，这种基于CAS理论的计算机仿真技术已经成为研究复杂系统的必要手段和主要途径，并广泛应用于生物、生态、社会、经济等不同领域的具体研究。有鉴于此，本文作者结合自己对一些具体复杂系统的研究体会，从理论、方法、应用等几个方面，对复杂适应系统的仿真技术进行了全面深入的研究。首先，从理论上归纳并总结了复杂性、复杂性科学、复杂系统等复杂性研究的基本概念和特征，分析了CAS理论的基本思想——适应造就复杂性，阐述了CAS理论的几个重要概念：聚集、非线性、流、多样性、标志、内部模型和积木块，明确指出CAS系统的核心就是系统的适应性主体，并在此基础上讨论了适应性主体的特征及其主体性描述。其次，从演化算法上分析了复杂适应系统的演化理论，在对CAS系统的演化过程进行数学描述的基础上，提出了层级CAS演化的数学模型，并对遗传算法、分类器系统和协同进化分别进行了讨论，阐述了各自的特点、算法以及在复杂适应系统研究中的应用，尤其对分类器系统在主体适应和学习中的应用、协同进化对应CAS系统中组织层级结构的重要意义进行了重点分析。再次，从仿真技术上归纳并总结了计算机仿真的基本理论，包括基本的概念、关系和流程等，然后重点研究了多主体系统的建模理论，阐述基于多主体的建模思想、方法和步骤，接着对比分析了几种常见的CAS仿真平台，建立了CAS系统的基本模型CAS-CM和层级模型CAS-HM，并对仿真模型的稳定性和计算复杂性进行了讨论。最后，在仿真应用上，研究了两个具体的复杂系统模型：少数者博弈模型和多人的重复囚徒困境模型，分别从不同的角度对前面的理论和方法进行了实践和创新。模型一对适应性主体的内部模型进行了分析和创新，提出了基于非完备策略和缺席的等级制度的个体策略机制，从而推动个体协作程度和系统适应度的整体上升；模型二提出了分级编码的策略方案和基于协议竞争模式的双层演化模型，并对演化过程中的阶段性均衡进行了定性分析和定量计算，为多人囚徒问题的研究开辟了新的思路。纵观全文，作者从基础理论、演化算法、仿真技术及实际应用等多个方面对复杂适应系统的建模与仿真进行了深入分析和研究，在层级演化的数学模型、微观主体的策略机制和CAS系统的仿真建模等诸多方面作出了有益的探索和创新，取得了不错的研究成果和贡献。更多还原

【Abstract】 Regarded as the "science of the 21st century", complexity science has become a major academic research area in the domain of modern system science. The classical academic theories and research methodologies based on reductionism are not capable of explaining and examining the complexity of a complex system due to its characteristics of non-linearity, hierarchy, etc. However, the modeling and simulation approach based on the Complex Adaptive Systems (CAS) provides an alternative method for complex system research. This approach treats individual elements in a system as "living" adaptive agents from an evolutionary perspective and establishes multi-agent models in a computer for real objects. With the rapid development of computer science and technology, the CAS-based computer simulation technology has become an essential tool for complex system research and has been widely used in various study fields such as biology, ecology, economics and sociology. With author’s own experience on some specific complex systems, this dissertation intends to discuss the CAS-based simulation technology from theoretical, methodological and applicable levels.To start with, the section reviews the basic concepts and properties of various complexity studies and analyzes the principle of the CAS theory: adaption builds complexity, with description of a number of key concepts in the CAS theory including aggregation, non-linearity, flow, diversity, tagging, internal model and building blocks. It concludes that the essence of CAS is the adaptive agents, with further description of properties and subjectivity of these adaptive agents.The second part of the dissertation analyzes the CAS evolutionary theory from the aspect of evolutionary algorithm and establishes the mathematical model for hierarchical CAS based on the mathematical description of CAS evolutionary process. It then discusses Genetic Algorithm (GA), Classifier Systems and Co-evolution, respectively, regarding their own characteristics, algorithms, and applications in CAS. Particular emphasis is given to the application of classifier systems in the study of adaptive agents and the relationship between co-evolution and CAS.Thirdly, the dissertation summarizes the basic theories of computer simulation, including its general concepts, relationships and procedures. Then it puts emphasis on the principles and methods for the modeling and simulation of Multi-Agent Systems (MAS). With comparatively introducing the major CAS simulation platforms, the dissertation proposes software frameworks for CAS Common Model (CAS-CM) and CAS Hierarchical Model (CAS-HM). It also discusses the stability and the computational complexity of simulation models.The rest of the dissertation studies two specific simulation applications on CAS: the model of Minority Game (MG) and the model of N-player Iterated Prisoner’s Dilemma (NIPD), which respectively practice and innovate the previous theories and methodologies from different aspects. In the MG model, through analyzing internal model of adaptive agent, a new mechanism of individual strategic with incomplete strategies and default hierarchy has been proposed, which can greatly improve the overall performance and the cooperation degree among individuals. The NIPD model proposes the strategy based on the classified coding and explores the CAS hierarchy modeling illustrated by a Double-Layer evolutionary model. It also make a qualitative and quantitative analysis on the stage equilibrium of the game process, which providing new ways to research for NIPD.Overall, this dissertation comprehensively studies the modeling and simulation of complex adaptive systems in the aspects of fundamental theories, evolutionary algorithms, simulation techniques, and practical applications. With useful exploration and innovation, it makes contributions to the mathematical model of hierarchical CAS, the mechanism of individual strategic and the CAS evolutionary modeling.更多还原