A novel approach to generate occupancy schedules based on human mobility and lstm-gan model

Occupant behavior significantly impacts urban energy consumption, yet urban building energy modeling (UBEM) often relies on simplistic occupancy schedules, leading to performance gaps. This study use GPS mobility data to generate data-driven occupancy schedules by Variational Autoencoders (VAEs) and Generative Adversarial Networks (GANs) for 700+ buildings in Phoenix, AZ, USA. VAEs produce smoother distributions with stable performance, while GANs capture peak occupancy but with higher volatility.

These generated schedules were integrated into UBEM for this research area, replacing standard schedules. Simulation results showed energy consumption variations of ±5% – ±20%, highlighting the importance of data-driven schedules in improving urban building energy simulation.