A data-driven clustering method for redistribution timing of a public bicycle sharing program

Public bicycle sharing programs (PBSPs) have become increasingly popular across many urban areas worldwide. The major challenge faced by PBSP operators is to find a suitable scheduling plan for bicycle redistribution to ensure that bicycle demand at all stations can be met at all times. Redistribution timing, which is to find the best time to dispatch vehicles to PBSP stations for bicycle distribution, is one of the key factors affecting the bicycle redistribution efficiency, PBSP operation cost and PBSP service quality. However, there are only a few studies on bicycle redistribution timing. A data-driven clustering method for redistribution timing based on mass of PBSP trip data is proposed to determine an optimal redistribution time. The method includes a data-driven model to describe the bicycle mobility, a dynamic redistribution timing algorithm to obtain the candidate time in need for bicycle redistribution, and a macro redistribution time clustering algorithm to acquire eventual reasonable redistribution times of PBSP considering the balance of operation cost and service quality. Finally, taking Hangzhou PBSP as an example, a data-based experiment is conducted to analyze the bicycle movement characteristics of typical stations in different PBSP regions. The macro bicycle redistribution times of these stations on both working days and holidays are captured resorting to the proposed method. The experimental results show that the method could promote both of the operational efficiency and service quality of the Hangzhou PBSP.