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

<abstract> <p>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 critical challenges operating high-quality bike-sharing is rebalancing bike stations from being full or empty. However, complex characteristics spatiotemporal dependency on usage demand may lead difficulties for traditional statistical models dealing with this relationship. To address issue, we propose a graph-based neural network model learn representation spatial-temporal graph. The has ability use graph-structured data takes both spatial- temporal aspects into consideration. A case study about Nanjing, large city China, conducted based proposed method. results show that algorithm can predict short-term relatively high accuracy low computing time. predicted errors hourly station level prediction often within 20 bikes. provide helpful tools other similar shared mobility systems.</p> </abstract>