Availability simulation and transfer prediction for bike sharing systems based on discrete event simulation

Abstract

To improve the operational efficiency of Bike Sharing Systems (BSSs), predicting users’ real demand plays an essential role. Previous demand predictions have heavily relied on the historical rides (explicit demand) recorded in the BSS management information system. Since there are no bikes available for rent, users' unfulfilled ride expectations (implicit demand) cannot be captured in the system. This paper presents an approach for simulating station-level availability and predicting transfers for BSS. Firstly, the (s,S) inventory strategy combined with an iterative optimization-seeking algorithm to improve the existing simulation model. Then, the K-means method is used to cluster the similarity among stations. The transfer prediction uses source data to aid the learning of target data quickly. Finally, a real case study is conducted to validate the practicability. The simulation experiment indicates that it is possible to estimate the maximum percentage of underestimation of real demand at a station. Through regression analysis, this paper investigates the complex nonlinear relationship between the percentage underestimation of real demand and the easily accessible implicit demand influencing factors. In practical applications, managers can use this method to optimize the number of bikes placed for BSSs.