Abstract

Dissertation submitted in partial fulfilment of the requirements for the Degree of Master of Science in Geospatial Technologies Forecasting bike availability is of great importance when turning the shared bike into a reliable, pleasant and uncomplicated mode of transport. Several approaches have been developed to forecast bike availability in station-based bike sharing systems. However, dockless bike sharing systems remain fairly unexplored in that sense, despite their rapid expansion over the world in recent years. To fill this gap, this thesis aims to develop a generally applicable methodology for bike availability forecasting in dockless bike sharing systems, that produces automated, fast and accurate forecasts. To balance speed and accuracy, an approach is taken in which the system area of a dockless bike sharing system is divided into spatially contiguous clusters that represent locations with the same temporal patterns in the historical data. Each cluster gets assigned a model point, for which an ARIMA(p,d,q) forecasting model is fitted to the deseasonalized data. Each individual forecast will inherit the structure and parameters of one of those pre-build models, rather than building a new model on its own. The proposed system was tested through a case study in San Francisco, California. The results showed that the proposed system outperforms simple baseline methods. However, they also highlighted the limited forecastability of dockless bike sharing data.