【作者】 易文明；

【导师】 肖人彬；

【作者基本信息】 华中科技大学 ， 管理科学与工程， 2011， 博士

【摘要】 在市场竞争日益激烈的情况下,企业的内外环境也日趋复杂,需求的多样化及不确定性给各类企业都带来挑战。为了敏捷地响应客户个性化的需求,生产企业必须优化传统运作模式来提高生产的柔性和敏捷性,同时企业之间需要协作组建虚拟企业应对市场机会；另外,虚拟软件企业这种特殊产品形式的虚拟企业的发展也引起了学界的关注。本文从企业协作的观点出发,选取典型企业形式进行研究,分别以生产制造型企业、虚拟企业及虚拟软件企业为对象,并采用多智能体仿真技术建立了各自的企业协作行为仿真模型,并根据仿真实验结果,提出了企业协作行为的管理策略。本文主要研究工作如下：(1)针对实体生产企业生产管理中的协作行为,分析了拉式生产系统中看板控制策略所存在的局限性,论述了看板控制策略与群集智能中的固定阈值蚁群劳动分工模型之间的相似性。提出将后者引入看板控制策略,以弥补看板控制策略缺乏全局优化能力的不足,实现平衡生产。据此对基本的固定阈值蚁群劳动分工模型进行了改进,重新设计蚂蚁的活动环境、属性特征以及状态转移规则等,建立了多状态蚁群劳动分工模型。通过一个多状态蚁群劳动分工模型执行过程的示例,给出了其算法实现步骤,进而选取一个典型的生产加工实例,采用该模型进行了计算仿真实验,结果表明基于多状态蚁群劳动分工模型的动态看板控制策略可以平衡各工序的生产效率,增强生产系统应对生产任务变化的能力,降低关键工序的限制,实现全局优化,使得生产平稳性得以提高,加工过程的工作负荷趋于平衡。(2)针对虚拟企业的协作行为,以虚拟企业中自我管理团队为例,分析了自我管理团队双层学习行为,以企业内部的协作行为为对象。将虚拟企业团队成员与任务定义为不同智能体,在自我能力估计误差率和彼此能力估计误差率等概念的基础上建立了多智能体仿真模型,采用Visual Basic开发了STLBMSS系统,并用一个实例对该仿真模型进行了确认。仿真实验结果揭示了双层学习行为对团队任务处理效率的影响机理,同时也表明STLBMSS用于分析自我管理团队学习行为的有效性。(3)进一步,针对软件企业此类特殊的虚拟企业,探讨软件企业与其外部的企业协作行为。以复杂适应系统(CAS)理论为基础,分析了虚拟软件企业的从组建到解散这一生命周期中的整体协作方式以及涌现现象,采用多智能体建模与仿真方法,将虚拟软件企业及其外部市场环境抽象为不同的智能体,建立了虚拟软件企业演化多智能体模型,并使用Anylogic仿真软件实现了模型。通过设计和运行仿真实验,得出了虚拟软件企业演化过程中盟主企业选择合作伙伴的机制和软件企业规模的发展趋势。虚拟软件企业能否成功很大程度上取决于其组建过程,而组建过程中最关键的环节是选择合适的合作伙伴。以虚拟软件企业合作伙伴选择问题作为研究对象,分析了其微观作用机制,采用多智能体建模与仿真方法,建立了合作伙伴选择多智能体模型,设计了相应的仿真实验,研究了不同决策目标下的合作伙伴选择机制。研究表明,成本最低法、工期最短法和质量最优法这三种单目标决策方法能够达到各种决策目标的最优化,但是稳定性较差；综合最佳法这种多目标决策方法能够实现软件项目总工期和总质量之间的平衡,并且稳定性较好。更多还原

【Abstract】 Under the circumstance of increasing competitive market, the internal and external environment is increasingly complex for enterprises, and the diversification of demand and uncertainty are a challenge to various types of enterprises. For quick response to personalized demand of customers, production companies must optimize the traditional operation mode to increase production flexibility and agility, and also the companies should collaborate to build the virtual enterprise to respond to market opportunities; in addition, virtual software enterprises, a particular form of virtual enterprise, also attracted scholars increasingly. In this paper, form the view of enterprise collaboration, it is selected the typical forms of enterprises, such as manufacturing enterprise, virtual enterprise and virtual software enterprise, and based on this enterprises, the related multi-agent simulation models are established, and according to the simulation results, the management recommendations on dynamic behaviors were proposed. The main study works of this paper are the following.Firstly, this paper analyzed the limitations of kanban control strategy on pull production systems, discussed the similarity between kanban control strategy and the fixed threshold model of ant colony labor division. In order to compensate for the lack of global optimization and achieve production balance, we proposed a new kanban control strategy. We improved the fixed threshold model of ant colony labor division, re-designed of the environment, property characteristics of state transition rules of ants, and established a multi-state model of ant colony division of labor. Through an implementation process example of the multi-state model of ant colony division of labor, the algorithm steps were given. Then, we select a typical example of the production process, used the model to simulated. The results showed that the dynamic kanban control strategy based on multi-state model of ant colony division of labor can balance the production efficiency of processes, increase the ability of production systems to respond to changeable production tasks, reduce restrictions on the key processes to achieve global optimization, increase the smoothness of production process, balance the workload.Secondly, this paper took the self-management teams of virtual enterprise as a example, analyzed the double-layer learning behaviors of self-management teams defined the team members and tasks as different agent. Based on the self-ability estimation error rate and mutual capacity estimation error rate, we established a multi-agent simulation model and achieved STLBMSS by Visual Basic, confirmed the model by an instance. The results revealed influence mechanism of the double double-layer learning behaviors on team task treatment efficiency, and also showed the effectiveness of STLBMS for analyzing the learning behavior of self- management team.Thirdly, based on the complex adaptive system (CAS) theory, this paper analyzed overall operation and emergence phenomenon of the virtual software enterprises from formation to dissolution, used the multi-agent modeling and simulation method, abstracted the virtual software enterprises as agents, established a virtual software enterprise evolution multi-agent model and used Anylogic to achieved the model. According to simulation results, we found the mechanism of dominant selecting partners and the development trends of enterprises on the evolution process of virtual software enterprises. The success of virtual software enterprises largely depend on the process of its formation. The most crucial aspect of formation is choosing the right partners. In this paper, we took the partner selection of virtual software enterprises as study object, analyzed micro-mechanism of the partner selection, and used multi-agent modeling and simulation method to establish a multi-agent model, designed simulation experiments to study partner selection mechanism under different goals. The results showed that lowest cost method, shortest duration method and best quality method achieved the optimal goal, but less stable. The comprehensive best method achieved the balance between the total duration and total quality.更多还原