Multi-stage stochastic linear programming for shared autonomous vehicle system operation and design with on-demand and pre-booked requests

TL;DR

This paper develops a multi-stage stochastic linear programming framework to optimize the design and operation of shared autonomous vehicle systems, considering both pre-booked and on-demand requests, using advanced dynamic programming techniques.

Contribution

It introduces a novel joint optimization model for SAV system design and operations, incorporating stochasticity and multiple planning stages, solved efficiently with SDDP.

Findings

Pre-booking options influence infrastructure and fleet strategies.

Dedicated pre-booked vehicles encourage advance reservations.

Solution approach achieves near-optimal results in complex scenarios.

Abstract

This study presents optimization problems to jointly determine long-term network design, mid-term fleet sizing strategy, and short-term routing and ridesharing matching in shared autonomous vehicle (SAV) systems with pre-booked and on-demand trip requests. Based on the dynamic traffic assignment framework, multi-stage stochastic linear programming is formulated for joint optimization of SAV system design and operations. Leveraging the linearity of the proposed problem, we can tackle the computational complexity due to multiple objectives and dynamic stochasticity through the weighted sum method and stochastic dual dynamic programming (SDDP). Our numerical examples verify that the solution to the proposed problem obtained through SDDP is close enough to the optimal solution. We also demonstrate the effect of introducing pre-booking options on optimized infrastructure planning and fleet sizing strategies. Furthermore, dedicated vehicles to pick-up and drop-off only pre-booked travelers can lead to incentives to reserve in advance instead of on-demand requests with little reduction in system performance.