An Agent-based Framework And Generation Of An Artificial Society

Nowadays, individual-based modeling and simulation provides an effective method to study complex systems, and the successful application of artificial society in reconstruction and reproduction of complex systems attracts increasing amounts of researchers in multiple disciplines. It is a synthetic integration to reconstruct an artificial society based on multi-field knowledge. Two basic problems should be considered: how to design a united framework for an artificial society and how to generate all the construction models in it. To our best knowledge, the reconstruction and generation of a synthetic artificial society is still at beginning, and lots of efforts are required.In the background of the application and demand of an artificial society’s reconstruction in emergency managment of public health, the dissertation aims to generate accurate and reasonable computational models of the artificial society. We do much work on the framework design, construction methods study, logical relations establishment between models. The main contributions are listed as followings:(1) The framework and formalization methods of a large-scale artificial society are designed. In this framework, structures and contents of the basic construction models and their correlations are explicitly specified, and individual status, migration and interaction are characterized and integrated using multi-source data. Two mapping relations are established to describe correlations between individuals and environments; geospatial-related behavior schedule is designed to model the diversity and regularity of individual daily behavior; social-relations-driven algorithms are proposed to model the interaction tendency between individuals.(2) The method of generating large-scale synthetic population is proposed. With limited and statistical demographic data, each individual is endowed with proper social attributes under the proposed generation rules and algorithms. The reconstruction of a family structure is a core step, and constraint rules between family members are established, according to which individual social attributes are generated. The generated population is not only statistically equivalent to a real population, but also promotes the heterogeneity and resolution of the generated population.(3) The method of reconstructing geographic environment and algorithms of mapping population to environment entities are proposed. With geographic information data, the environment entity is designed as a hierarchical structure according to administrative divisions, which is endowed with abilities to access individuals’ information who locate in it. Under this mechanism, an individual is able to perceive other individual’s information in the same local environment. Further, population distribution data is utilized to estimate the statistics of high-resolution environment entities, and the environment entity database is designed and generated to provide locations for individual daily behavior. For each individual, an environment list is also generated to specify the detailed locations where the individual performs his activities. During the process of establishing mapping relations between individuals and environment entities, we discretize the continuous geographic space to reduce the computing complexity of geospatial searching.(4) The method of generating multilayer social networks and algorithms of social-relation-driven interaction are proposed. According to the spatial distribution of generated population and environment, a framework of multilayer social networks is proposed, and all types of individual social relations are integrated in one model. We design a modular and hierarchical structure to generate multilayer social networks, and the networks grow from local clustering networks to large range but sparse ones. Individual social interaction is quantified as a non-stochastic process by utilizing social-relation-driven interaction algorithms, and it promotes the interacting resolution and tendency in the population.(5) The transmission of influenza H1N1 and control policies in an artificial society are studied. Based on the proposed framework and methods, we design and generate three artificial societies on different scales: classroom, campus, and city of Beijing. Transmission experiments of influenza H1N1 are carried on, and individual spatial migrations, interactions, and topological characteristics of social networks(both static and dynamic) are analyzed. The emergency managements are also evaluated by reproducing the transmission process. From the experiments, it can be seen that artificial computational environment can reproduce complex events, and proves that the proposed framework and generation method are efficient and reasonable.The research of this dissertation belongs to the fundamental theory and methodology in modeling and simulation domain. It is a meaningful effort on developing a methodology of individual-based artificial society, and the proposed generation methods of synthetic population, geographic environment, spatial behavior and multilayer social networks promote the reconstruction and reproduction of complex systems and also the evaluation of emergency management.