Bike-sharing demand prediction based on a multi-level model integrating macro and micro information

Accurate bike-sharing demand prediction helps relevant authorities make informed decisions on bike placement and advance scheduling. Although a great deal of work has been carried out in this field, we found that most studies have focused on modeling at a single granularity, ignoring the interactions between multiple granularities. To address this issue, we propose a Multi-level Model for Bike-sharing Demand Prediction (MMBDP). This model starts with an embedding module that determines the macro- and micro-granularity representations of bike-sharing demand from the initial study area. Then, macro- and micro-modules are built to perform spatiotemporal modeling to extract the spatiotemporal correlation information at different granularities. Finally, the fusion module is designed to integrate the information from each granularity and perform the final prediction. Experiments on real datasets show that the proposed model outperforms the baselines. Ablation experiments further reveal the superiority of our model. The model helps to address the bike supply-demand imbalance problem and promotes the sustainable development of urban transportation systems.