Revision as of 10:02, 4 May 2026

Abstract

As a heavy-haul railway, China Energy Railway mainly transports coal and has the characteristics of stable supply and large freight volume. The order approval process follows a centralized and unified model. This approach suffers from prolonged approval cycles, extended intervals between order submission and actual transportation (with submissions required a year in advance), and vague planned delivery times, making it difficult to meet clients’ demands for precise delivery timelines and the flexibility required for high-value scattered cargo orders. To stabilize long-term revenue for transport enterprises and enhance client satisfaction, this study introduces a client value coefficient and a delivery-time satisfaction function to evaluate order value. A dynamic order approval model for China Energy Railway freight services is constructed and solved using a deep reinforcement learning algorithm. Given the large volume of orders and the complexity of order requirements in China Energy Railway, which involve multiple auxiliary decision variables, some discrete decision variables are adjusted to continuous variables to accelerate model training. This adjustment is combined with the HyAR algorithm to enhance the training efficiency of intelligent agents. Finally, the model’s performance is tested using freight data from China Energy Railway in March 2024. Under constrained capacity conditions, the dynamic order approval model achieves 3.1% improvement in comprehensive revenue compared to static approval methods.OPEN ACCESS Received: 23/09/2025 Accepted: 24/11/2025

Document