Fleet management in free-floating bike sharing systems using predictive modelling and explorative tools

Abstract

For redistribution and operating bikes in a free-floating systems, two measures are of highest priority. First, the information about the expected number of rentals on a day is an important measure for service providers for management and service of their fleet. The estimation of the expected number of bookings is carried out with a simple model and a more complex model based on meterological information, as the number of loans depends strongly on the current and forecasted weather. Secondly, the knowledge of a service level violation in future on a fine spatial resolution is important for redistribution of bikes. With this information, the service provider can set reward zones where service level violations will occur in the near future. To forecast a service level violation on a fine geographical resolution the current distribution of bikes as well as the time and space information of past rentals has to be taken into account. A Markov Chain Model is formulated to integrate this information.

We develop a management tool that describes in an explorative way important information about past, present and predicted future counts on rentals in time and space. It integrates all estimation procedures. The management tool is running in the browser and continuously updates the information and predictions since the bike distribution over the observed area is in continous flow as well as new data are generated continuously.