Service Level Requirements for Real Life–Sized Bicycle Sharing Systems.

This article focuses on a two-step optimization approach to managing service level requirements (SLRs) in large-scale bicycle sharing systems (BSSs). The first step, called the target-level problem (TLP), uses a two-stage stochastic program solved via Benders decomposition to allocate bicycles across stations to maximize the expected number of satisfied trips, incorporating real-world data from nine major cities. The second step, the bicycle rebalancing problem (BRP), designs vehicle routes to adjust bicycle quantities to these target levels, employing adaptive large neighborhood search heuristics under two cost policies to ensure route feasibility and efficiency. Computational experiments demonstrate that significant reductions in bicycle inventory—sometimes up to 50%—can maintain high service levels, offering practical insights for BSS operators on optimizing fleet size and station capacities. The study highlights model M2, which integrates station and scenario bounds, as the most effective formulation for solving the TLP at real-life scales.