基于多智能体的城市群政府合作建模与仿真——嵌入并反馈于一个异构性社会网络

摘要

以社会网络视阈下的城市群政府合作过程中的多期进化博弈为例,通过概念模型、数学模型和计算机模型的完整建模过程,在微观主体互动层面建立一个考虑个体成本收益差异的非对称进化博弈系统(并同时考虑全体博弈互动和部分博弈互动的交互),在宏观社会结构层面建立一个实现节点之间连接概率定制化的异构性社会网络模型(且同时考虑全局信息交互和局部信息交互的更替),并通过多智能体系统(特别是交互意愿变量的引入)实现了微观互动(博弈行为)和宏观涌现(网络结构)之间的双向反馈和动态交互.在构建模拟系统的基础上,设计并运行了一系列的实验方案,对仿真模型及模拟结果进行信度检验,并结合多次模拟实验样本数据的统计分析,考察全局交互比例,局部连接概率等网络结构变量、合作收益、惩罚措施等博弈参数变量、以及决策个性等主体属性变量对城市群政府博弈形势与合作水平的动态影响,为促进城市群政府合作和区域一体化进程提供决策依据和政策参考,也是政治学与公共行政学科计算实验研究的一次创新尝试.%To analyze the evolutionary game process of multi-cities' government collaboration in the context of social networks,through an integral modeling procedure of conceptual model,mathematical model,and computer model,a multi-agent simulation system,integrating evolutionary game theory and social networks model,is built.An asymmetrical evolutionary game system considering the individual differences on cost and benefit is constructed for the micro level of agent interaction (considering the interaction between entire game and partial game),and a heterogeneous social network evolving system,considering the linking probability differences between nodes,is built for the macro level of social structure (considering the interchange between global information exchange and local information exchange),to achieve the bidirectional feedback and dynamic interaction between micro game interaction and macro social network based on multi-agent system (particularly by introducing the interacting willingness).The simulation experiments investigate the dynamic effect of network structure variables such as global interacting proportion and local interlinking probability,game coefficient variables such as governance benefit and punishment measure,and agent property variable such as decision-making personality composition,on the evolving process and revenue outcome of multi-cities' continuous game.Policy suggestions are given for multi-cities' coordination and regional integration.Besides,through multiple simulation experiments of each of the experiment plan,samples of data are built for the reliability test of the simulation model and results,and for the statistical analysis of experiment results and findings.This research provides new thoughts in computational simulation experiment research in the field of politics and public administration.