3:20 PM - 3:40 PM
[E4-02] Towards Pseudo People Flow: Developing a Deep Generative Model based on Person Trip Survey Data to Reproduce Large-Scale Daily People Activity Profiles.
Keywords:Generative model, Pseudo people flow, Person trip survey data, Mobility data
The lack of cost-effective access to the human mobility dataset constrains deeper scientific investigations and restricts policymakers’ ability to evaluate policy effectiveness. Recent achievements in generative pre-training models have made it possible to generate high-quality synthetic data as a practical substitute. In this study, we explore how existing generative pre-training models can be harnessed to reconstruct human daily behavior by modeling it as a sequence of activity choices. We evaluate the performance of our proposed approach using the Person Trip survey data collected from 20 metropolitan areas in Japan, encompassing approximately 6 million individuals’ typical daily activities. Our experimental results show that the proposed approach is proficient at generating representative, realistic, and diverse synthetic human daily activity pattern data. This research significantly contributes to the generation of synthetic human mobility datasets, such as the Pseudo People Flow. Consequently, it aids in enhancing our understanding and accuracy of human mobility, demonstrating a promising avenue for future studies in this domain.