SIMULATING CONTINUOUS-TIME HUMAN MOBILITY TRAJECTORIES
Abstract
Recent pandemic events have greatly emphasized the need to understand how humans navigate in modern day cities for effective public health policy implementation. In this paper, we propose a two-stage generative model, DeltaGAN, to simulate realistic human mobility trajectories. Compared with existing work where time was discretized, DeltaGAN generates continuous visitation time to better capture temporal irregularity in human mobility behaviors. Conditioned on the generated time, DeltaGAN synthesizes realistic trajectories by limiting the range of accessible location candidates. Experimental results demonstrate that our model achieves consistently better performance than baselines when comparing distribution similarities with real-world GPS trajectories via 6 individual trajectory and geographical metrics. We further validate the utility of DeltaGAN on COVID-19 spread simulation and observe the diffusion process under generated trajectories is consistent with that under real data.
Sirisha Rambhatla
University of Waterloo
