エージェントベースの人口推計モデルにおける属性決定手法の評価

抄録

In agent-based simulation using actual social statistics, attributes of all the agents have to been decided to meet the statistics. However, they can’t be decided uniquely because statistics gives only limited number of aggregated values. To this problem, Hara et al. have proposed a method of formulating the problem as a constrained optimization problem, and solving it using simulated annealing. They applied it to estimation of future population. However, they didn’t carry out evaluation of computational performance of their model. In this paper, the authors improved implementation of their model to enhance computational performance, discuss convergence of the optimization process and evaluate computational load of their method. Estimation results are compared with an estimation using conventional cohort-component method.