Applying Bayesian spatio-temporal models to demand analysis of shared bicycle

Abstract

Shared bicycle provides a cheap and healthy mobility alternative to travelers especially for the “first–last mile” trips. Although the temporal and spatial correlation of regional shared bicycle needs has been confirmed in the literature in recent years, the interdependencies between them are not yet fully understood. In this paper, a spatio-temporal Bayesian modeling method is proposed to quantify regional shared bicycle demand and identify the impact of various factors on the cycling trips. By combining the Integrated Nested Laplace Approximation (INLA) and Stochastic Partial Differential Equation (SPDE), it guarantees the establishment of the feasibility of algorithms on large-scale spatiotemporal data structures. In particular, the massive rental records of Mobike in Shanghai in August 2016 are used as the study observation. We establish a series of Bayesian models with different temporal and spatial structures, and uses the Deviation Information Criteria to verify the relevance of the models in the temporal and spatial dimensions. Moreover, the Kolmogorov–Smirnov test is proposed to fit different distributions to obtain the optimal distribution family of travel demand data. Our research efforts have further been made to evaluate the impact of meteorological factors, population density and per capita GDP on travel demand. The result shows that the model of temporal and spatial correlation structure can better assess the regional distribution of future shared bicycle riding demands, and the influence of temperature and precipitation on cycling demand is more significant. The study’s findings will help guide the development of future shared bicycle regional scheduling work, and improve economic benefits on the basis of meeting traveler’ needs.