A spatio-temporal deep learning model for short-term bike-sharing demand prediction

Bike-sharing systems are widely operated in many cities as green transportation means to solve the last mile problem and reduce traffic congestion. One of the critical challenges in operating high-quality bike-sharing systems is rebalancing bike stations from being full or empty. However, the complex characteristics of spatiotemporal dependency on usage demand may lead to difficulties for traditional statistical models in dealing with this complex relationship. To address this issue, we propose a graph-based neural network model to learn the representation of bike-sharing demand spatial-temporal graph. The model has the ability to use graph-structured data and takes both spatial- and temporal aspects into consideration. A case study about bike-sharing systems in Nanjing, a large city in China, is conducted based on the proposed method. The results show that the algorithm can predict short-term bike demand with relatively high accuracy and low computing time. The predicted errors for the hourly station level usage demand prediction are often within 20 bikes. The results provide helpful tools for short-term usage demand prediction of bike-sharing systems and other similar shared mobility systems.