430 / 2019-02-27 16:53:43

An empirical study on travel demand spatiotemporal patterns of dynamic internet based ride-hailing

ride-hailing, spatiotemporal patterns, travel demand prediction, deep learning

全体主题 > Spatial data collection, Analysis, and Visualization

The rapid growth of internet based ride-hailing brings great changes to residents’ travel and traffic, while there are still limited studies exam-ine the travel demand spatiotemporal patterns of such internet based ride-hailing trips by using empirical data. In this paper, we attempt to analyze the spatiotemporal patterns of internet based ride-hailing trav-el demand distribution from a month traffic order data in Chengdu, Sichuan Province, China, that provided by DiDi company. The statis-tical characteristics of data, OD spatiotemporal distributions, as well as location based travel distance are analyzed. We further present a deep learning approach for the demand prediction of ride-hailing service. To combine the complex non-linear spatial and temporal relations, a spatiotemporal model is proposed which consists of two views: mod-eling spatial correlations via Convolutional Neural Network (CNN), and modeling correlations between future demand values with histori-cal time points via Long Short Term Memory networks (LSTM). Re-sults depicted that our approach have effective prediction accuracy over traditional methods.