Online Car-Hailing Demand Prediction with Multi-Factor Spatio-Temporal Convolutional Network

For the problem of online car-hailing demand forecasting in the future period, this paper proposes a multi-factor spatio-temporal convolution prediction model (CNN-LSTMAtt-GCNs, CLAG). The model captures spatio-temporal dependence using convolution modules and extracts the spatial function similarity features of point of interest (POI) through the GCNs module, considering the influence of various external factors, such as weather and working day. The experimental results show that the CLAG model can effectively extract the spatial and temporal characteristics and capture the influence of external factors. Compared with the LSTM model, the MAE, RMSE and MAPE of the CLAG model are reduced by 36.89%, 43.83% and 52.67% respectively, and the <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$R^{2}$</tex> is improved by 24.46%. Furthermore, ablation experiments proved that the introduction of external factors and POI can effectively improve the prediction accuracy of the model.