Learning and Simulation of Spatiotemporally Multi-scale Pedestrian Flow Model

Congestion in urban spaces can lead to serious accidents, and mathematical modeling of pedestrian flows is important for prediction and prevention. Pedestrian flow behavior can be spatio-temporally divided into macro and micro scales, and most conventional models focus on a single block of urban space, the micro scale. In this study, we propose a model that simultaneously considers the macroscale. Information at each scale is encoded and decoded by LSTM to make predictions with interaction. The proposed model is trained from a dummy dataset reproducing the area in front of Shibuya station and measured data for basic investigation. Simulations will be conducted using the learned model to verify whether it is possible to reproduce the pedestrian flow in front of Shibuya Station.