A Bayesian hierarchical model for traffic flow and waiting time prediction in carpooling lines
schedule le lundi 27 janvier 2020 de 17h30 à 18h30
Organisé par : F. Bechtold, W. Da Silva , A. Fermanian, S. Has, Y. Yu
Intervenant : Panayotis Papoutsis (LPSM)
Lieu : Jussieu, Salle Paul Lévy (16-26-209)
Sujet : A Bayesian hierarchical model for traffic flow and waiting time prediction in carpooling lines
The start-up ecov (ecov.fr) provides a range of innovative public-private carpooling services which respond to the mobility requirements in periurban and rural areas, in an economically and ecologically sound framework. These carpooling services, unlike their competitors who tend to focus a virtual service for densely urbanised regions, are also equipped with physical meeting points. In conjunction with a mobile telephone application, drivers participating in these carpooling services are invited to activate the geolocalisation on their mobile telephone. These GPS traces form the primary data source for our analysis which respond to the industrial requirements to characterise the driver behaviour. The two most important requirements are to forecast the driver flow and predict waiting times when a request of carpooling is made by a passenger. We propose a Bayesian hierarchical model based on multi-level coefficients with an autoregressive component. The multi-level coefficients are based on day types namely, workday, weekend or public holiday; and the regression parameters are estimated using Not-U-Turn samples (NUTS). The outputs of this model are discretised into predefined time intervals of a typical day as is the usual practise for transport providers, and then we carry out a validation with observed driver flows and waiting times.