Jang Won Bae, Agent-Based Efficient Modeling and Simulation with Formalism, PhD Dissertation, Department of Industrial and Systems Engineering, KAIST, 2015 

Abstract

Agent-based modeling and simulation (ABMS) has been applied to complex systems for replicating, understanding, and analyzing problems from the complex systems. Complex systems consist of multiple components and their intricate relationships, and this trait is similar to the structure of agent-based model (ABM), which is one reason that ABM is considered as a representative method for modeling complex systems. However, as systems to be modeled are getting more complex, ABMS should be performed more efficient manner. Opposed to the prosperity of ABMs, the current practices of ABMS have inefficiently conducted: from the perspective of ABM development, most of ABM has been developed without reusing existing components of others’. Considering the cost for ABM development is considerable, model reuse would be a key for reducing the development cost. From the perspective of ABM simulation, the prolonged execution time of simulating ABM is a critical factor. As ABM describes complex systems, ABM generally holds lots of components and their intricate relationships. This complex model structure eventually results in the prolonged simulation execution. It causes heavy cost for ABM simulation because the simulation execution would be performed repeatedly to obtain statistical insights from the simulation results. This dissertation suggests a formalism-based approach to resolve those inefficiencies. Formalism is a mathematic-based representation for model specification and the model specifications should involve unambiguity and completeness. The model specifications that are mathematically represented can provide model reusability in ABM development. Such a trait is well represented in formalisms from system engineering field. Also, in the view of ABM simulation, model specifications in formalism are used to simplify ABM structure with reduction technique, and this simplification leads to the reduction of the simulation execution time. To realize the suggested approach, this dissertation proposes a formalism for ABM development, called Large-scale, Dynamic, Extensible, and Flexible (LDEF) formalism. LDEF adopts the structured modeling technique from DEVS, so it supports to build a larger model by the composition of other models. Also, LDEF inscribes ABM contexts, such as agents, environments, and their relationships, and dynamic interaction changes into its model specification, so specifications of ABMs are explicitly represented by LDEF. Also, this dissertation proposes a method for accelerating the simulation execution of ABMs that has hierarchical and dynamic structure. To do end, a meta-model structure for the event exchanges, called Coupling Relation Graph (CRG) is suggested. CRG is a directed graph that represents coupling relations among components of ABM. The proposed method abstracts model structures represented in CRG with techniques in graph theory so that it aids to reduce the procedure for event exchanges and eventually reduces the simulation execution time. To evaluate and prove the efficiency of the proposed methods, theoretical and empirical methods are applied to the example models that are developed by LDEF and simulated by the proposed method. The evaluation results prove that the proposed methods provide significant efficiency on ABMS, in particular, when a model holds a lot of components and frequent interaction changes. In addition, as a case study for ABMS, analyzing evacuation situation during bombardment is introduced.

@phdthesis{Bae:2015,

author = {Jang Won Bae},

advisor ={Il-Chul Moon},

title = {Agent-Based Efficient Modeling and Simulation with Formalism},

school = {KAIST},

year = {2015}

}