OD Demand Forecasting for the Large-Scale Dockless Sharing Bike System: A Deep Learning Approach

摘要

Bike sharing, as a rapidly emerging transportation mode, can provide reliable data support with ample travel information. Bike sharing's ordering data has highly application value for obtaining bike share origin-destination (OD) volumes. This paper, considers the ordering data generated by Beijing's Mobike users as the research object, and proposes a deep-learning-based method that considers both temporal and spatial correlations. Two deep learning structures, including recurrent neural network (RNN) and long short-term memory (LSTM) network, are proposed to forecast OD volumes. Experiment results show that the LSTM network has a relatively small prediction error, with 8.72% mean absolute percentage error (MAPE) for 15-minute time intervals, and thus, able to accurately estimate bike sharing's OD volume.