A Reproducibility Analysis of Synthetic Population Generation

For the development of agent based traffic simulation model, population synthesis is critical to the accuracy of simulation outcomes. This paper attempts to develop the synthetic population generation based on the Simulated Annealing (SA) algorithm for the activity-based travel demand model. This algorithm leads to estimate the activity schedules according to the multi-dimensional characteristics of the synthetic populations. However, appropriate rules have not been established for the estimation of parameters in simulated annealing, and it requires a significant amount of time to find optimal solution. In order to apply SA into the synthetic population, hill climbing and cooling schedule should be considered. In this study, total absolute error was calculated to prevent hill climbing and used Metropolis- Hasting algorithm to determine whether to select or dismiss follow-up distribution. In addition, stability of the algorithm was determined through scenario analysis of the optimal combination of iteration and temperature “T” on the cooling schedule. Based on this result, the current condition of household travel diary survey and census data were used to compare the IPF(Iterative Proportional Fitting) of a previous methodology with the result of establishing suggested algorithm, performing procedures of creating synthetic population, and suggesting the validity of algorithm created with the synthetic population based on SA through statistical verification.