A Two-Fold Simulated Annealing to Reconstruct Household Composition from Statistics

In this paper, we propose a two-fold simulated annealing (a 2-fold SA) to reconstruct household compositions from real statistics. A social simulation has received attention in recent years, which is one of research topics to see the real world in terms of social sciences. We need individual data such as age, gender, work etc. To construct a realistic simulation model. However, it is difficult to use an individual data because the government restricts the secondary use of their individual data. Therefore, we should reconstruct artificial data from the real statistics using a reconstructing method. We have improved Ikeda´s method to minimize the error between reconstructed households and the real statistics using an evolutionary computation (EC). However, the method takes a longer computation time than Ikeda´s method due to time consuming processes of EC. We should try to reduce the computation time in view of future applications. In this paper, we propose a 2-fold SA in order to reduce the computation time. Our simulation results indicate that the proposed 2-fold SA is better in the accuracy and the computation time.