【作者】 时振涛；

【导师】 洪毅； 曾建潮；

【作者基本信息】 兰州理工大学 ， 控制理论与控制工程， 2013， 博士

【摘要】 复杂系统与复杂性科学被誉为21世纪的科学,是一种新兴的边缘、交叉学科,已成为现代系统科学的一个主要研究方向。作为复杂系统的研究工具,复杂系统的建模和仿真方法目前还在探索阶段,没有形成一个完整的体系,还需要继续深入研究。本文依据复杂适应系统理论的基本思想,将演化算法和人工社会方法相结合,构造出一种人工社会系统的建模方法——基于演化Agent的人工社会系统建模方法,并以企业组织这个复杂系统作为应用对象,结合微观经济学原理和复杂网络理论,应用本文建立的基于演化Agent的人工社会系统建模方法,建立了基于演化Agent的企业组织演化模型和企业人际关系网络的演化模型。本文主要完成以下工作：1.给出了基于演化Agent的人工社会系统建模方法。以复杂适应系统理论为基础,利用人工社会思想,运用多Agent建模方法的关键技术,根据人工社会系统中行为个体的局部细节,将演化算法应用于Agent的反应规则和各种局部行为中,实现Agent的类生物进化和行为优化,建立了具有高智能性和高适应性的微观Agent结构和模型。然后根据所建立的演化Agent的结构和模型,给出了基于演化Agent的人工社会系统建模方法和具体的建模流程,并分析了该方法的难点和模型的特点。2.建立了基于演化Agent的企业组织演化模型。针对企业组织系统的生长演化过程,运用基于演化Agent的人工社会系统建模方法,从微观主体的行为入手,依据主体的行为特点,结合演化算法,分别建立了企业组织演化模型的基本结构、适应性主体的基本行为模型和宏观整体演化的Echo模型,从而构造了基于演化Agent的企业组织演化模型,并借助Swarm仿真平台对其进行仿真实验,观察和分析模型中企业组织系统的演化规律及影响因素,探讨企业组织系统中的复杂适应性行为,验证了基于演化Agent的人工社会系统建模方法的有效性。3.应用改进微粒群算法对成员主体行为参数进行优化,提高基于演化Agent的企业组织模型的智能性。首先,针对企业组织系统演化过程中微观主体的自适应性行为,应用基于约束保持法的微粒群算法(CPPSO)优化成员主体的行为参数,根据成员对资源的需求偏好和对资源的工作能力,在满足成员最大化效用的基础上,设计了基于改进微粒群算法的成员主体行为参数的优化模型。然后,将该模型应用到基于演化Agent的企业组织演化模型中,并对其进行仿真实验,结果证明,在实际需求及约束条件下,CPPSO方法采用简洁的位置和速度更新实现参数寻优,不仅可较快获得行为参数的全局最优解,提高成员的学习能力,而且能有效地反映出成员理想效益曲线与实际效益曲线间的总体差异。4.应用遗传算法实现资源主体的进化。首先,以资源主体的属性特征为基础,对个体进行实数编码,根据资源主体与资源含量和成员对资源的需求之间的关系,建立适应度函数。然后,根据遗传算法“适者生存,优胜劣汰”的进化原则,采用遗传操作实现资源主体的进化,并在计算机上进行仿真,实现了基于演化Agent的企业组织演化模型中的资源主体的演化机制,同时,分析了资源主体的进化对企业组织演化过程的影响。5.构造了基于演化Agent的企业人际关系网络模型。根据Agent的行为特点,制定演化Agent节点的行为策略,构造智能Agent节点,使Agent节点能够随着企业组织的演化逐渐建立关联,形成企业人际关系网络模型,并通过仿真实验观察企业人际关系网络模型的生成和演化过程,同时,运用网络的统计特征论分析人际关系网络模型在演化过程中的变化规律,结果表明企业人际关系网络是随着Agent节点寻找资源的行为生成和演化的,网络结构也随着企业组织系统的演化不断变化,最终形成了具有“小世界”特性和“无标度”特性的复杂网络。更多还原

【Abstract】 As the21st-centurys science, the complexity science is a new interdisciplinary discipline, and has become a main research direction of modern system science. As a tool for the complex system research, the modeling and simulation method of complex system is still at the stage of exploring, has not formed an integrated system, so it is necessary to be further studied. Firstly, based on complex adaptive system theory, evolutionary algorithm and artificial social method, a modeling method for artificial social system, called evolutionary agent-based artificial social system modeling method, is constructed in the dissertation. Secondly, taking enterprise system as application background, combining with the principles of microeconomics and complex network theory, utilizing the established evolutionary agent-based artificial social system modeling method, we establish the enterprise evolution model and the enterprise relationship network evolution model based on the evolutionary agent. The main achievements are obtained as follows:1. The evolutionary agent-based artificial social system modeling method is proposed. According to complex adaptive system theory and artificial social idea, by using the multi-agent modeling, depending on the local details of individual in artificial social system, the right evolutionary algorithms is applied in the agent’s reaction rules and various local behaviors, so the agent’s similar biological evolution and behavior optimization are implemented, a high intelligent and high adaptability micro agent structure and model is constructed. Moreover, the evolutionary agent-based artificial social system modeling method is formed by applying the established evolutionary agent model. Finally, the difficulties and the characteristics of the model are analyzed.2. The evolutionary agent-based enterprise evolutionary model is constructed. Towards the evolution process of enterprise system, by applying the evolutionary agent-based artificial social system modeling method, analyzing the characteristics of the behavioral agents, the appropriate evolutionary algorithms is selected to establish the basic structure of the enterprise system, the micro adaptive agents model, the overall macro Echo model, respectively, and the evolutionary agent-based enterprise organization evolutionary model. In order to show the validity of the method, the simulation is performed in the SWARM platform, the evolutionary relations between micro-agents behaviros and macro-complex phenomena are observed, some complex adaptive behaviors of enterprise organization are discussed. 3. The member behavior parameters optimization model based on improved particle swarm algorithm is constructed. According to the micro-agents adaptive behaviors in the enterprise organization system, the particle swarm optimization algorithm with constraint-preserving mechanism(CPPSO) is applied to solve the member behavior parameters optimization problems. Depending on the demand preference and the productivity of member, a member behavior parameters optimization model based on the improved particle swarm algorithm in order to maximize the utility of member. Furthemore, the optimization model is applied in the evolutionary agent-based enterprise evolution model, the simulation results shows that under the actual demand and constraint conditions, the use of CPPSO algorithm can obtain the global optimal solution fairly quickly, greatly improve the learning ability of member, and reflect the difference between the ideal benefit and the practical benefit of member effectively.4. The evolution of resource agents based on genetic algorithm is implemented. Firstly, based on the properties of resource agents, agent is encoded with real-number. Secondly, by analyzing the relationship between the resource agents and the resource’s value, the member’s demand, the fitness function is designed. Then, according to the evolutionary laws of survival of the fittest of genetic algorithm, the evolution of resource agents using genetic operations is presented. Through the simulation, the evolutionary mechanism of resource agents is realized in the evolutionary agent-based enterprise evolution model. In addition, the effect of the resource evolution on the evolution process of enterprise organization is analyzed in detail.5. The enterprise relationship network model based on evolutionary agent is constructed. According to the characteristics of agent behaviors, the evolutionary agents’behavior strategies are designed, the intelligent agent nodes are presented. Accompanying the enterprise organization evolutions, the connection between agent nodes are gradually established, and the enterprise relationship network model is gradually formed. Through the simulation, the generation and evolution of enterprise relationship network model are presented, moreover, the evolution laws of enterprise relationship network are analyzed using the statistic characteristics of complex networks. The simulation results show that enterprise relationship network is gradually generated and evolved during the process of the agent nodes searching for resources, the network structure is constantly evolving with the enterprise evolutions, finally, the complex networks with the small-world and scale-free characteristic is formed.更多还原