This paper presents a novel framework for customised modular bus systems that leverages travel demand prediction and modular autonomous vehicles to optimise services proactively. The proposed framework addresses two prediction scenarios with different forward-looking operations: optimistic operation and pessimistic operation. A mixed integer programming model in a space-time-state network is developed for the optimistic operation to determine module routes, schedules, formations and passenger-to-module assignments. For the pessimistic case, a two-stage optimisation procedure is introduced. The first stage involves two formulations (i.e., deterministic and robust) to generate cost-saving plans, and the second stage adapts plans with control strategies periodically. A Lagrangian heuristic approach is proposed to solve formulations efficiently. The performance of the proposed framework is evaluated using smartcard data from Beijing and two state-of-the-art machine learning algorithms. Results indicate that the proposed framework outperforms the real-time approach in operating costs and highlights the role of module capacity and time dependency.

The integration of autonomous vehicles and on-demand customized bus systems is expected to be beneficial for responding to real-time demands. This paper investigates the autonomous customized bus (ACB) system that leverages passenger demand prediction to enhance service quality and vehicle utilization. A novel ACB service design optimization model that determines vehicle movements and passenger-to-vehicle assignments is developed for the real-time routing and scheduling problem. Then, a rolling horizon approach, incorporating travel demand prediction, proactive dispatching and reactive adjustment, is proposed to address the studied problem. The performance of the introduced ACB system is evaluated using smartcard data from Beijing and the state-of-the-art machine learning algorithm. Results show that the proposed ACB system can effectively improve system performance and service level in terms of operating cost and passenger waiting time compared to reactive operations.