A look-ahead policy for dynamic bike rebalancing with neighborhood interactions

Bike-sharing systems (BSSs) offer a flexible and eco-friendly transportation option in urban areas, but often encounter challenges due to imbalances in spatial and temporal demand. These imbalances can make it difficult for users to rent or return bikes. To restore efficiency within the system, rebalancing operations are crucial. Most bike rebalancing models assume that demand is dynamic and independent at station level, neglecting the interactions between neighborhoods. This phenomenon, known as roaming or spillover effects, occurs when users find a station unavailable and redirect to nearby alternatives. Ignoring these interactions distorts the representation of demand and results in suboptimal rebalancing strategies. This paper presents the Dynamic Bike Rebalancing Problem with Neighborhood Interactions (DBRP-NI), which explicitly models these spillover effects. To address the DBRP-NI, we have developed a heuristic framework based on a rolling horizon approach that incorporates look-ahead decision-making. We evaluated our solution method within a discrete-event simulation framework using real-world data from BSSs in Norway and the United States. The computational results demonstrate that our proposed method consistently outperforms benchmark policies, especially in large and imbalanced systems. It reduces long-distance roaming and enhances vehicle utilization while maintaining computational efficiency. Additional findings highlight the operational benefits of integrating neighborhood interactions and look-ahead strategies into rebalancing decisions for modern BSSs.