Dynamic rebalancing for Bike-sharing systems under inventory interval and target predictions

Abstract

Bike-sharing systems have become a popular transportation alternative. Unfortunately, station networks are often unbalanced, with some stations being empty, while others being congested. Given the complexity of the underlying planning problems to rebalance station inventories via trucks, many mathematical optimizations models have been proposed, mostly focusing on minimizing the unmet demand. This work explores the benefits of two alternative objectives, which minimize the deviation from an inventory interval and a target inventory, respectively. While the concepts of inventory intervals and targets better fit the planning practices of many system operators, they also naturally introduce a buffer into the station inventory, therefore better responding to stochastic demand fluctuations. We report on extensive computational experiments, evaluating the entire pipeline required for an automatized and data-driven rebalancing process: the use of synthetic and real-world data that relies on varying weather conditions, the prediction of demand and the computation of inventory intervals and targets, different reoptimization modes throughout the planning horizon, and an evaluation within a fine-grained simulator. Results allow for unanimous conclusions, indicating that the proposed approaches reduce unmet demand by up to 34% over classical models.