Vehicle scheduling approach and its practice to optimise public bicycle redistribution in Hangzhou

• Published in: IET intelligent transport systems Vol. 12; no. 8; pp. 976 - 985
• Main Authors: Liu, Dongxu, Dong, Hongzhao, Li, Tiebei, Corcoran, Jonathan, Ji, Shiming
• Format: Journal Article
• Language: English
• Published: The Institution of Engineering and Technology 01.10.2018
• Subjects: backpropagation; back‐propagation neural network short‐term prediction model; bicycles; Case Study; disaggregate trip data; geographic information systems; GIS‐based prototype system; k‐means clustering; large‐scale PBSP redistribution; neural nets; optimisation; O‐D relationship; public bicycle redistribution optimization; public bicycle sharing programmes; public transit; public transport; region partition method; rolling horizon scheduling algorithm; self‐service stations; traffic information systems; vehicle scheduling approach; large-scale PBSP redistribution; public bicycle redistribution optimization; vehicle scheduling approach; geographic information systems; disaggregate trip data; public bicycle sharing programmes; O-D relationship; optimisation; k-means clustering; traffic information systems; public transit; rolling horizon scheduling algorithm; public transport; backpropagation; back-propagation neural network short-term prediction model; bicycles; self-service stations; region partition method; neural nets; GIS-based prototype system
• ISSN: 1751-956X; 1751-9578; 1751-9578
• DOI: 10.1049/iet-its.2017.0274

Public bicycle sharing programmes (PBSPs) have become increasingly popular across many urban areas in China. Hangzhou PBSP is the world's largest and forms this case study. The management of this large inventory of bicycles is a particularly challenging issue with the goal to ensure the demand for bicycles is met at all times across the network. To this end, an efficient scheduling approach is needed with the capacity to guide the redistribution of bicycles across the self-service stations. Drawing on 7 years of disaggregate trip data, this study first captures the usage dynamics across both space and time to extract the candidate stations and redistribution periods for vehicle scheduling. A region partition method with K-means clustering is proposed to satisfy the real-time requirement of large-scale PBSPs' redistribution. Moreover, drawing on the variations in demand a back-propagation neural network short-term prediction model is computed to inform the necessary prospective redistribution of bicycles to ensure demand is always met. Finally, a vehicle scheduling model employing a rolling horizon scheduling algorithm is established and implemented in a GIS-based prototype system. The prototype is evaluated through its effectiveness and found benefit for following 18 months of practical operation in Hangzhou PBSP.