Collaborative optimization of train operation using safe reinforcement learning and ray parallelism solution for nested modeling problem

doi:10.1016/j.swevo.2026.102310

Collaborative optimization of train operation using safe reinforcement learning and ray parallelism solution for nested modeling problem

2026-02-10

0

PRE

AI

DOI：10.1016/j.swevo.2026.102310

Original

Original request for help

Save

Abstract

En 中文

In urban rail transit, to reduce operational costs and enhance passenger service quality, this paper proposes a nested integrated optimization model consisting of inner and outer layers. The inner-layer optimization model formulates an inter-station running curve optimization framework based on a Ray-parallel computing architecture. The outer-layer optimization model addresses the cooperative utilization of regenerative braking energy, the allocation of running time, and the calculation of average passenger waiting time. To solve this nested model and enhance the safety of the training process, the paper employs a safe reinforcement learning framework. Finally, to validate the effectiveness of the proposed approach, case experiments are conducted based on an actual line. The simulation results demonstrate that the proposed method can significantly reduce both the total net energy consumption of the trains and the average passenger waiting time during an operational cycle.

Keywords:

train operation optimization

safe reinforcement learning

ray parallelism

nested optimization model

regenerative braking energy

Collaborative optimization of train operation using safe reinforcement learning and ray parallelism solution for nested modeling problem

Abstract

Journal

Organization