A Grid Based Simulation Environment for Parallel Exploring Agent-Based Models with Vast Parameter Space
Agent-based simulation models with large experiments for a precise and robust result over a vast parameter space are becoming a common practice,where enormous runs intrinsically require highly intensive computational resources.This paper proposes a grid based simulation environment,named Social Macro Scope(SOMAS)to support parallel exploration on agent-based models with vast parameter space.We focus on three types of simulation methods for agent-based models with various objectives: 1)forward simulation to conduct experiments in a straightforward way by simply operating sets of parameter values to obtain sets of results; 2)inverse simulation to search for solutions that reduce the error between simulated results and actual data by means of solving inverse problem,which executes the simulation steps in a reverse order and employs optimization algorithms to fit the simulation results to the desired objectives; and 3)model selection to find optimal model structure with subset of parameters and procedures,which conducts two-layer optimization to obtain a simple and more accurate simulation result.We have confirmed the practical scalability and efficiency of SOMAS by a case study in history simulation domain.
Agent-Based Simulation Grid Computing Forward Simulation Inverse Simulation Model Selection
Chao Yang Isao Ono Setsuya Kurahashi Bin Jiang Takao Terano
Business School,Hunan University,China;Department of Computational Intelligence and Systems Science, Department of Computational Intelligence and Systems Science,Tokyo Institute of Technology,Japan Graduate School of Business Sciences,University of Tsukuba,Japan Department of Computational Intelligence and Systems Science,Tokyo Institute of Technology,Japan;Col
国际会议
南昌
英文
1-14
2013-09-26（万方平台首次上网日期，不代表论文的发表时间）