Joint seminar with HEC Montréal and the Department of Logistics and Operations Management.

Bike sharing systems are essential for urban transportation, but geographical and temporal imbalances in bike demand require nightly reallocation to maintain service levels and minimize demand loss. This study addresses two key challenges in optimizing static bike rebalancing: accurately predicting station-level bike demand with demand substitution and optimizing the routing of multiple rebalancing vehicles. We propose a data-driven solution featuring predictors that account for time dependencies, weather conditions, and demand substitution by nearby stations. A sequential simulation-based demand loss estimator determines optimal rebalancing quantities. Additionally, a mixed integer linear programming model optimizes vehicle routing, supported by a data-driven decomposition algorithm that simplifies the multivehicle routing problem into parallel single-vehicle problems. Extensive experiments with New York City Citi Bike data show the accuracy of our demand predictors, the significance of demand substitution, and the efficiency of our optimization framework.