Modélisation statistique cyclique des locations de Vélo'v à Lyon
Abstract
Vélo'v is Lyon's community bicycle program. It is studied here at a global level, to assess the evolution with time of the number of hired bikes, and find the relevant factors to explain the evolution. An empirical analysis first reveals the daily and weekly patterns in a cyclostationary manner, jointly with the non-stationary evolutions over time-scales of the day and larger. Combining this model with linear statistical regression, a model is proposed for the prediction of the number of bikes hired per hour. We show that the season (its weather and the existence of holidays), the popularity of the program (as given by the number of subscribed users) and the time during the week are the most relevant factors to predict the number of locations at a given hour. This is confronted to the real database of Vélo'v trips.
Origin : Publisher files allowed on an open archive
Loading...