Towards Fine-grained Flow Forecasting: A Graph Attention Approach for Bike Sharing Systems

Abstract

As a healthy, efficient and green alternative to motorized urban travel, bike sharing has been increasingly popular, leading to wide deployment and use of bikes instead of cars. Accurate bike-flow prediction at the individual station level is essential for bike sharing service. Due to the spatial and temporal complexities of traffic networks and the lack of data-driven design for bike stations, existing methods cannot predict the fine-grained bike flows to/from each station. To remedy this problem, we propose a novel data-driven spatio-temporal Graph attention convolutional neural network for Bikestation-level flow prediction (GBikes). We develop data-driven and spatio-temporal designs, and model bike stations (nodes) and inter-station bike rides (edges) as a graph. In particular, we design a novel graph attention convolutional neural network (GACNN) with attention mechanisms capturing and differentiating station-to-station correlations. Multi-level temporal closeness, spatial distances and other external factors (e.g., weather and points of interest) are jointly considered for comprehensive learning and accurate prediction of bike flows at each station. Extensive experiments upon a total of over 11 million trips collected from three large-scale bike-sharing systems in New York City, Chicago, and Los Angeles have corroborated GBikes's significant improvement of accuracy, robustness and effectiveness over prior work.