Integrated location and operation for emergency road service: A machine learning-driven robust approach

Abstract

In emergency road services, the location of service centers and the allocation of mechanics are crucial for ensuring a rapid response. Their importance becomes even more pronounced when faced with uncertain accident repair demands. In this paper, we explore an integrated service area design and repair service operation problem in a transportation system under demand uncertainty. We formulate the problem as a two-stage stochastic mixed-integer linear programming (MILP) model. In the first stage, it determines the locations for service centers and allocates mechanics accordingly. In the second stage, it manages the mechanics to serve demands. To address incomplete demand distribution information, we further extend the stochastic model to a distributional robust optimization (DRO) model, which assumes that the true distribution lies in an ambiguity set and concerns the performance under the worst-case distribution within the set. To enhance the computational efficiency, we develop alternating direction method of multipliers (ADMM) algorithm and further integrate it with machine learning (ML) techniques to propose an ML-ADMM algorithm. Our numerical results using real data show that our approach can obtain a solution with a limited optimality gap while reducing computational time by 41.23%. Out-of-sample results show that DRO model demonstrates superior risk-averse performance, surpassing both risk-neutral and robust optimization methods. We reveal that choosing either too few or too many centers is not ideal, which can result in higher costs and lower service levels. While it is necessary to ensure all demands are served timely from a safety perspective, it is not cost-effective.