Real-time scheduling optimization for autonomous public transport vehicles to meet booking demands

• A new ‘predict-then-optimize’ modelling framework. • Real-time booking promotes autonomous public transport, with online demand tracked by a rolling horizon approach. • NP-hardness was proven prudently. • A tailored two-phase algorithm with valid cuts, Monte Carlo simulation, neighborhood and local search. • Results show up to 35% less waiting time and 13% smaller fleet size for two implementation scenarios. The booking service, a key feature of autonomous public transport vehicle (APTV) systems, has been designed to introduce a new, real-time, on-demand, and reliable element to service improvement, similar to ride-hailing. However, the current APTV system has yet to fully realize the potential of a smart public transport service in optimizing the balance between supply and demand. This study proposes a real-time, multi-objective programming model that aims to minimize three key factors: passenger waiting times, timetable deviations, and fleet size. Recognized as an NP-hard problem, the model is linearized to reduce computational complexity, with real-time demands tracked through a rolling horizon method. A predict-then-optimize approach is introduced to enable timely responses to new bookings. A customized two-phase algorithm incorporating three enhancements − valid cuts, Monte Carlo simulation, and neighborhood and local search − significantly improves solution efficiency. A case study in Auckland, New Zealand, evaluates the proposed approach. The findings reveal significant improvements in booking service performance, with two scenarios achieving a 35 % and 27 % reduction in passenger waiting time and a 13 % and 12 % decrease in fleet size compared to the current conventional bus line. These results were attained with minimal deviations from the original schedule, validating the effectiveness of the developed methodology.