Bicycle sharing systems have proven to be very successful in several major cities and are now spreading all across the world. There exist more than 700 such systems that operate on five continents. The benefits for cities are multiple: from a greener image due to more eco-friendly means of transportation to the reduction of traffic congestion, noise and air pollution, they provide an alternative to private motorised vehicles, especially for short-distance trips. From the user’s perspective, they offer an affordable and efficient transport alternative with several benefits over the use of a personal bicycle with respect to maintenance, theft or storage issues.

A bicycle-sharing system is composed of a number of stations where a limited number of bikes can be parked. A user arrives at a station to pick up a bike. The ride ends when she returns the bike to any station. User experience and provider revenue can be hampered when the origin station is empty, which forces the user to either resort to another means of transport or try to find an available bike in another station. Similarly, if the destination station is full, the user must either wait until one bike is picked up, or return the bike to another station with at least one parking spot available. The problems caused by empty or full stations need to be solved if the full benefits of bicycle-sharing systems are to be obtained.

Researchers on the QUANTICOL project at INRIA, Edinburgh and IMT Lucca have been studying the problem of forecasting the future availability of bicycles in stations of a bike-sharing system. This is relevant in order to make recommendations guaranteeing that the probability that a user will be able to make a journey is sufficiently high. Probabilistic predictions of successful journeys are obtained from a time-inhomogeneous queueing theory model. The model has been parametrised and successfully validated across an entire one-year historical dataset from the Vélib’ system of the city of Paris.

The possibility of making probabilistic forecasts has significant added value, since it broadens the scope of the applicability of predictive models. It directly provides user-centric quantities of interest, useful for journey planning, such as the probability of finding a bike at the origin station (and dually, of finding an empty slot at the destination station).