A Closed-Loop Intelligent Control Framework for Automated Railway Shunting in Marshalling Yards

Abstract

Accurate and efficient railway shunting operations are crucial for the operation of intelligent marshalling yards. This article proposes a closed-loop scheduling method that integrates fuzzy analytic hierarchy process (F-AHP), graph neural network (GNN), and multi-objective optimization algorithm to achieve intelligent and automated shunting operations through the "perception decision execution" chain. The system integrates multi-source sensor data (train position, switch status, track occupancy, etc.), uses GNN for track conflict prediction, and determines multi-objective weights based on F-AHP. Combined with multi-objective optimization, it generates Pareto optimal scheduling scheme. The experiment was conducted at a large marshalling yard in the southwest region, and the results showed that compared with manual scheduling, the system reduced the average operating time by 35.7% ± 2.1%, single task energy consumption by 21.4% ± 1.5%, and scheduling conflict rate by 87.5% ± 3.2%, while improving judgment accuracy to 97.8% (evaluated over 20 runs with p < 0.01).The research has verified the comprehensive advantages of the proposed method in terms of efficiency, energy consumption, and safety, and has the ability to transplant and expand across stations. To support reproducibility, we also specify the optimization model, variable and constraint definitions, the GNN architecture (features, loss, hyper-parameters), and the rolling-horizon settings with quantified latency budgets.