Exploring the Impact of Spatiotemporal Granularity on the Demand Prediction of Dynamic Ride-Hailing

Dynamic demand prediction is a key issue in ride-hailing dispatching. Many methods have been developed to improve the demand prediction accuracy of an increase in demand-responsive, ride-hailing transport services. However, the uncertainties in predicting ride-hailing demands due to multiscale spatiotemporal granularity, as well as the resulting statistical errors, are seldom explored. This paper attempts to fill this gap and to examine the spatiotemporal granularity effects on ride-hailing demand prediction accuracy by using empirical data for Chengdu, China. A convolutional, long short-term memory model combined with a hexagonal convolution operation (H-ConvLSTM) is proposed to explore the complex spatial and temporal relations. Experimental analysis results show that the proposed approach outperforms conventional methods in terms of prediction accuracy. A comparison of 36 spatiotemporal granularities with both departure demands and arrival demands shows that the combination of a hexagonal spatial partition with an 800 m side length and a 30 min time interval achieves the best comprehensive prediction accuracy. However, the departure demands and arrival demands reveal different variation trends in the prediction errors for various spatiotemporal granularities.

Automated summary

This paper investigates the issue of demand prediction in ride-hailing dispatching and proposes a method based on a convolutional long short-term memory model combined with a hexagonal convolution operation. Experimental analysis using empirical data for Chengdu, China shows that the proposed approach outperforms conventional methods in terms of prediction accuracy. The comparison of 36 spatiotemporal granularities reveals that a hexagonal spatial partition with an 800 m side length and a 30 min time interval achieves the best comprehensive prediction accuracy, although departure demands and arrival demands exhibit different variation trends in prediction errors for various spatiotemporal granularities.