Sparse and Efficient Rebalancing Operations: Is More Always Better?

【主讲】Mabel Chou，新加坡国立大学商学院副教授

【主题】稀疏性和效率的再均衡：多一定好吗？

【时间】2019年5月14日（周二）下午14:00

【地点】清华经管学院伟伦楼453

【语言】英语

【主办】管理科学与工程系

【简历】Mabel Chou的简历

【Speaker】Mabel Chou, Associate Professor, Department of Analytics & Operations, Business School, National University of Singapore

【Topic】Sparse and Efficient Rebalancing Operations: Is More Always Better?

【Time】Tuesday, May 14, 2019, 14:00pm

【Venue】Room 453, Weilun Building, Tsinghua SEM

【Language】English

【Organizer】Department of Management Science and Engineering

【Abstract】 Motivated by the Bike Angels Program in New York’s Citi Bike system, we study the use of volunteers as an alternative to rebalance bikes in a bicycle sharing system before the morning and evening peak hours. We develop a distributionally robust method to design a static sparse network to support the re-distribution activities and use a state-dependent approach to determine the rebalancing activities of the volunteers in a bike sharing system. While this approach allows the empty bikes to be re-positioned to satisfy the demands and maximize availability of bikes to the users, it unfortunately may lead to excessive redundant moves. We show that sparse network structure such as those used in the Bike Angels Program can reduce this issue and improve the performance of the simple state-dependent heuristic. Depending on the operating conditions, our method often produces a hybrid structure that outperforms both the long chains and the fixed pick-up/drop-off structure used in the Bike Angels Program. We use a data set from the Hubway system in Boston (with 60 stations) to demonstrate that using a small set of arcs (15%) to support the rebalancing activities of the volunteers in the system only causes a small loss (around 4%) of the efficiency, and helps to improve the performance of state-dependent-based re-balancing activities in the system - reducing the redundant rebalance moves by over 50%, while managing to achieve better or similar bike availability level in the system when compared to the state-dependent approach in a fully flexible network. In other words, a little flexibility in the rebalancing operations not only can perform as well as a fully flexible system in terms of system efficiency (in the off-line case), but also can improve the performance of online state-dependent heuristic by concentrating the rebalancing activities on those suitably selected arcs.