Short Term Demand Forecasting of Bike Sharing Stations Based on LSTM Model

With the increasingly prominent role of shared bicycles in urban traffic, station demand forecasting has become the key to scheduling optimization. In view of the research gap in the targeted application of LSTM model in this field, this study collected the Toronto shared single vehicle operation data, built the LSTM prediction model, used to predict the demand of each station in the next three hours, and evaluated the prediction performance of the model from a multidimensional perspective. In this study, the overall prediction results of 140 time points in the future of 15 stations and the prediction of 40 time points in the future of specific stations are visualized. The experimental results demonstrate that the LSTM model attains high levels of fitting accuracy for the demand of shared bicycle stations, which is significantly better than ARIMA model. It can provide a scientific reference for the intelligent scheduling of shared vehicles, and expanding the input characteristic dimension can further improve the prediction accuracy.