Time Series Forecasting of Dockless Bike-Sharing OD with the Weather

With the mobile internet booms, the bike-sharing service has been greatly expanded. At present, dockless bike-sharing has the problems of low utilization rate and turnover rate, and limited distribution. Understanding the OD distribution in different regions can help the service operators solve these problems effectively. In this paper, we propose a temporal convolution network prediction model to predict the bike-sharing OD distribution. This study uses temporal convolution network (TCN) to extract the temporal features and utilizes fully connected network (FCNN) to model weather influence. This study proposes a forecast method of FC-TCN, which can effectively fuse TCN and FCNN, considering the multiple factors. The model accuracy is tested through a case study of Tianjin, China. The results show that this model can predict OD distribution more precisely than other deep learning methods such as LSTM and GRU.