Demand forecasting of online car‐hailing by exhaustively capturing the temporal dependency with TCN and Attention approaches

Abstract

Abstract With the development of the car‐hailing industry, it has become an indispensable way of travel in our lives. Accurate prediction of online car‐hailing demand can provide the basis for real‐time vehicle dispatch and dynamic pricing for online car‐hailing companies. Most previous studies on online car‐hailing demand forecasting have only considered the temporal and spatial factors separately, while the effects of time series and spatial series on online car‐hailing demand have not been considered. In this paper, the temporal, spatial, and weather features of online car‐hailing are analyzed and they are used as input features of the model. In addition, an attention mechanism is added to the model in order to select a small amount of key feature data from a large amount of feature data and give more weight to the key data, and an attention mechanism‐based TCN (Temporal Convolutional Network) prediction model (TCN+Attention) was developed to better highlight the key features that affect the prediction of online car demand and improve the prediction accuracy of online car demand. Finally, taking the data of Ningbo City as an example, the data is divided into 10 min, 15 min, and 30 min time intervals for prediction, and it is combined with other models and with other prediction models (SVR, LightGBM, Random Forest, Stacking Integrated Learning, LSTM, LSTM+ Attention, and TCN) results in comparative analysis. Experiments show that the TCN+Attention model of online car‐hailing demand prediction has higher accuracy compared with other models.

Keywords

• data mining
• demand forecasting
• feature extraction
• feedforward neural nets
• road traffic