Data-driven prioritization strategies for inventory rebalancing in bike-sharing systems

The popularity of bike-sharing systems has constantly increased throughout the recent years. Most of such success can be attributed to their multiple benefits, such as user convenience, low usage costs, health benefits and their contribution to environmental relief. However, satisfying all user demands remains a challenge, given that the inventories of bike-sharing stations tend to be unbalanced over time. Bike-sharing system operators must therefore intervene to rebalance station inventories to provide both available bikes and empty docks to the commuters. Due to limited rebalancing resources, the number of stations to be rebalanced often exceeds the system’s rebalancing capacity, especially close to peak hours. As a consequence, operators are forced to manually select a subset of stations that should be prioritized for rebalancing. While most of the literature has concentrated either on predicting optimal station inventories or on the rebalancing itself, the identification of critical stations that should be prioritized for rebalancing has received little attention. Given the importance of this step in current operating practices, we propose three strategies to select the stations that should be prioritized for rebalancing, using features such as the predicted trip demand and the inventory levels at the stations themselves. Two sets of computational experiments aim at evaluating the performance of the proposed prioritization strategies on real-world data from Montreal’s bike-sharing system operator. The first set of experiments focuses on both the 2019 and 2020 seasons, each of which exhibits distinct travel patterns given the restrictive measures implemented in 2020 to prevent the spread of COVID-19. One of these strategies significantly improves by reducing the estimated lost demand by up to 65%, while another strategy reduces the estimated number of required rebalancing operations by up to 33% when compared to the prioritization scheme currently in use at the considered bike-sharing system. The second set of experiments evaluates the performance of the proposed strategies when rebalancing decisions are optimized in a rolling horizon planning. The results highlight various benefits of the proposed strategies, which are efficiently solved as transportation problems and improve lost demand over two intuitive baselines.

Mots clés

bike-sharing; demand prediction; rebalancing; inventory management