Modeling Demand for Bikesharing Systems - Neighboring Stations as Source for Demand and Reason for Structural Breaks

Bicycle sharing systems (BSS) are becoming ever more popular all over the world. One of the remaining problems is that the distribution of rides between stations is not uniformly distributed and certain stations fill up or empty over time. These empty and full stations lead to demand for bikes and return boxes that cannot be fulfilled leading to unsatisfied and possibly even lost customers. To avoid this situation, bikes in the systems are redistributed by the provider. While redistribution of bikes in such systems is well studied, the underlying demand is not yet modeled to serve as an input to improve the redistribution. This gap is closed in this paper. We model demand for bikes and return boxes using data from the BSS Citybike Wien in Vienna, Austria. In particular, the influence of weather and full/empty neighboring stations on demand is studied using different count models. Furthermore, we show that forecasts from our model improve the forecast using historic demands. Lastly, the influence of new stations on the model parameters of a station and resulting structural breaks in the model are discussed.