【作者】 姚灿中；

【导师】 杨建梅；

【作者基本信息】 华南理工大学 ， 管理决策与系统理论， 2010， 博士

【摘要】 本文的产业复杂网络既包括产业竞争关系复杂网络,还包括新出现的大众生产的合作关系复杂网络。企业本来就是存在于由大量竞争对手形成的大规模网络之中的。产业竞争关系复杂网络模型及分析方法的提出,为经典的产业组织分析补充了企业之间竞争关系的网络视角,从而发展了产业组织的理论与方法。但是已有的产业复杂网络研究大都仅针对竞争关系复杂网络的,本文的研究对象则既包括产业竞争关系复杂网络,还包括了大众生产合作关系复杂网络。而且目前的产业竞争关系的复杂网络分析,大多是针对具体的单个产业进行的,缺少针对所有产业的普适的分析框架与方法,本文首先通过对多个具体的产业竞争关系复杂网络以及大众生产合作关系网络的分析,归纳出具有普适性的产业复杂网络分析的“静态拓扑特征—子图特征—动态生成机制—动力学性质”的分析框架。其次,深入探讨并提出了各种产业复杂网络的建模、仿真与分析的具体方法。在静态拓扑分析中,采用极大似然方法分析了产业复杂网络的度分布;在子图特征分析中,采用了k-clique子图、模体、层级结构、集团结构、分形与多重分形等分析方法;在动态生成机制分析中,通过建构模体异向匹配模型、拟Yule增长模型、双向适应度模型来揭示各种产业复杂网络的生成机理;在网络动力学性质分析中,构建了产业竞争关系复杂网络上的竞争扩散模型、大众生产平台之间的竞争模型、大众生产合作关系网络上的信息传播模型与多智能体模型、以及Ising模型与多智能体自适应模型等模型来进行研究。这些模型上不仅有方法方面来自于“综合”的创新性,也更加符合实际,比如多智能体模型将复杂网络中的“死”的节点视为“活”的智能体,将复杂自适应系统分析方法引进产业复杂网络的分析之中,使得产业复杂网络的分析方法考虑了企业实际上所具有的自主性。再次,在模型分析过程中还对相关的算法做了改进。比如在产业复杂网络的静态结构研究中,采用Powell着色法改进了复杂网络分形的盒维数算法;在产业复杂网络上的动力学性质研究中改进了粒子群算法等等。最后,需要指出的是以上框架、模型与方法不仅具有产业复杂网络分析方法方面的普适意义,而且具有新的洞察力,因此对产业复杂网络以及产业分析有一些新的发现。比如用极大似然方法估计出的产业复杂网络度分布的r_mian值,揭示了产业复杂网络满足幂律分布的企业最小竞争对手的数目,度的幂律分布刻画了产业内企业竞争对手数目的异质性以及大众生产者合作者数目的异质性。产业复杂网络的子图、模体、集团、分形与多重分形分析的发现,分别揭示了产业竞争存在着稳定与重叠的竞争结构,具有层级性,集团特征以及自相似等等特性。生成机制与网络上的动力机制分析也得出网络生成与竞争扩散等等方面的有关发现。更多还原

【Abstract】 The industry complex network in the paper include the industry competition complex net-work and the emerging collaboration network based on Peer Production. In fact, enterprisesexist in large networks formed by many competitors. Industry competition complex networkmodel and the analytical method complement the classical Organization Analysis theory fromnetwork perspective among the enterprises, which develops the industrial organization theoryand method. However,most industry complex network analysis focuses on competition net-work, the paper will both study the industry competition complex network and Peer Productionnetwork.In addition, the research on industry complex network mostly focus on single industryand lack general analytical framework and methods to all industries. First of all, through theanalysis of many specific real competition network and Peer Production network, the paperconstruct a universal framework of industry competition complex network for”static topologycharacteristics—subgraph characteristics—generating mechanism—dynamic characteristics”.Second, we make deep study on the static characteristics of the network topology, mainlyincluding the application of maximum likelihood method to estimate the degree distribution ofindustry complex networks, investigating the static structure especially the subgraph charac-teristic of industry complex network by the analysis of k-clique subgraph, motif, hierarchicalstructure, community structure, fractal and multifractal property. We propose the motif dis-assortative model, the imitated Yule process model and the bipartite fitness model to analyzethe generating mechanism of complex network. Again, we make research on the dynamicproperty of the industry competition network and Peer Production collaboration network. Weconstruct the competition spread model over the industry competition network and find thatThe enterprise location in the network topology who launch the competition have a significanteffect on the spread result. We also find that competitive effects spreading over the network haslocal characteristic. we construct and analyze the platform competition model of peer produc-tion,which reveals the evolution of peer production platform competition when the competitionintensity based on the network topology follows the power law and Gaussian distribution. Wefurther construct the information spread model based on Peer Production collaboration networkand analyze topological property of the individual who initiate the information as well as thenumber impact significantly on the spread result. Then we construct the multi-agent adaptivesystem model to study the information spread process and reveal the macro emergence of theadaptive system during different process. We also analyze the stability of Peer Production by constructing Ising model combining herding effect and study the the critical point of the sys-tem impacted by different network structures and external conditions, and finally we use MonteCarlo simulation to verify the critical analytical solution derived by mean field theory. We con-struct multi-agent self-adaptive system and analyze the stability of peer production system inshort term and long term. On the research of network dynamics, we construct the competitionspread model, Peer Production platform competition model, information spread based on PeerProduction network and multi-agent model, Ising model and multi-agent model, which are notonly”comprehensive”innovative but also more realistic, for example multi-agent model con-sider the”dead”node as a”live”agent, it bring the complex adaptive system into the analysisof industry complex network which make the industry complex network method concern theenterprises’practical autonomy. Further, During the analysis process of the static and dynamiccharacteristics of the network, we improve the corresponding algorithm. For example in theanalysis of static network topology, we improve the traditional box covering algorithm for com-plex network fractal property analysis and verify the efficiency of the the improved algorithmand in the process of dynamics analysis modeling, we improve PSO from the perspective ofstatic network topology and dynamic network topology, propose PSO based on SFL and DSF-PSO two improved models.Finally, be noted that the above framework, model and approach not only has the indus-try’s complex network analysis universal significance, but also has the new insights into theindustry complex network and industry analysis. For example we adopt the maximum likeli-hood method to estimate the r_min values of the industry complex network’s degree distributionwhich reveal the smallest competitors numbers of power-law distribution, and the degree fol-low the power-law distribution depict the heterogeneity within competitors of industry complexnetwork and heterogeneity within Collaborators of Peer Production network. The discoveryof subgraph, motif, community structure, fractal and multifractal property in industry complexnetwork respectively reveals the existence of a stable industrial competition and overlappingcompetition structure, with hierarchical, community characteristics, and self-similarity and soon.更多还原