Optimizing Shared E-Scooter Operations Under Demand Uncertainty: A Framework Integrating Machine Learning and Optimization Techniques

The emergence of dockless shared e-scooters as a new form of shared micromobility offers a viable solution to specific urban transportation problems, including the first-mile–last-mile issue, parking constraints, and environmental emissions. However, this sharing service faces several challenges in daily operation, particularly related to demand volatility, battery recharging, maintenance, and regulations, owing to their trip and physical characteristics. Therefore, this study proposed a new data-driven rebalancing framework for dockless shared e-scooters that incorporates demand and variance prediction, and Monte Carlo sampling to simulate the expected demand. Thus, demand uncertainty and the collection of low-battery and broken e-scooters were included in the rebalancing formulation to minimize user dissatisfaction and operating costs. Rebalancing optimization is an NP-hard problem; in this study, the small-size problem was solved using the integer linear programming (ILP) solver GNU Linear Programming Kit, and the large-size problem was solved using the proposed hybrid ant colony optimization–ILP algorithm (ACO–ILP). This framework was evaluated on a real-world dataset from Minneapolis, Minnesota, which demonstrated that the demand and variance prediction efficiently allocated the uncertainty while reducing the overall uncertainty, leading to shorter driving distances and lower rebalancing costs relative to baseline cases.