Kukic, M., and Bierlaire, M. (2023)

Hybrid Simulator for Capturing Dynamics of Synthetic Population

23rd Swiss Transport Research Conference, Ascona, Switzerland

This paper presents a novel hybrid framework that combines model- based and data-driven approaches for generating and maintaining syn- thetic population. The existing generators produce a snapshot of synthetic data that becomes outdated over time, requiring a complete regeneration using the newest datasets for updates. However, periodic census data can be leveraged to simplify the generation process and provide up-to-date synthetic populations at any moment in time. We propose a method that generates a baseline dataset of synthetic individuals using the Markov Chain Monte Carlo simulation and projects it over time by simulating events that impact generated attributes of synthetic individuals. The projected sample should replicate the distribution of the newest version of the dataset, but may not provide a perfect fit due to data collection bi- ases. To account for these differences, we use information from the newest data to re-sample and correct the projected synthetic sample. To validate our approach, we use Swiss census data from 2010, 2015, and 2021. We generate the baseline sample using 2010 Swiss census data and project it to 2021. We compare existing state-of-the-art techniques for dynamic projection with our hybrid approach against Swiss census data from 2021. The results show that the synthetic sample generated in the past can be improved by integrating information from the newest data without using complete regeneration.