Hybrid optimization of metro train scheduling an‎d composition using deep reinforcement learning an‎d proximal policy optimization algorithm

10/19/2026 12:00:00 AM

Optimization of operation in urban rail transportation systems, especially in metro lines, is one of the fundamental challenges in managing dynamic deman‎d an‎d energy consumption. In this research, an intelligent model for adaptive train scheduling an‎d wagon composition is presented, which is developed based on a deep reinforcement learning algorithm with asymptotic policy optimization (PPO) method. In this model, the train scheduling an‎d composition problem is formulated as a Markov decision process (MDP), in which the learning agent, considering the current state of the system — including passenger deman‎d level, train capacity an‎d travel distances — makes an optimal choice of two key variables: Headway an‎d number of wagons (Train Composition). The model reward function is designed based on a combination of passenger waiting costs, energy consumption an‎d train composition changes to establish a balance between service quality an‎d operating costs. Simulation results show that the proposed model has been able to reduce waiting time, increase service rate, an‎d improve energy efficiency compared to traditional methods. This research has taken an effective step towards making urban transportation systems smarter by providing a new framework for using reinforcement learning algorithms in optimizing subway operations.

يادگيري تقويتي عميق، بهينه‌سازي سياست مجانبي (PPO)، زمان‌بندي تطبيقي مترو، تركيب قطار، بهره‌برداري هوشمند، تقاضاي پويا

Deep reinforcement learning, asymptotic policy optimization (PPO), adaptive subway scheduling, train composition, smart operation, dynamic demand

mohammad heydary

masoud yaghini