Hierarchical multiagent reinforcement learning for sustainable truck dispatching in open-pit mining

作者：Kaiqi Zhao, Zhengke Liu, Wenzhuo Du, Huabo Lu, Bin Zhou, Guizhen Yu, Xiaolei Ma · 发表于：Transportmetrica A Transport Science · 年份：2025 · DOI：10.1080/23249935.2025.2522360 · 被引用次数：4 · 研究领域：Mining Techniques and Economics、Belt Conveyor Systems Engineering、Advanced Manufacturing and Logistics Optimization

Efficient mining truck dispatching is vital for green mining, as trucks are major greenhouse gas (GHG) emission emitters in open-pit mining. Autonomous truck technology, already adopted in several mines, reduces human-related inefficiencies and enables centralised dispatching. However, real-time challenges – like intersection bottlenecks and operational disruptions – necessitate an advanced dispatch system. This study introduces a hierarchical multiagent reinforcement learning framework to optimise truck dispatching and routing under these constraints. To address congestion and uncertainties often overlooked by current methods, we incorporate first-in-first-out (FIFO) rules at intersections. A case study in a large-scale open-pit mine in Inner Mongolia, China, demonstrated the framework's effectiveness. Compared with benchmarks, the method achieved up to an 8.2% increase in production and a 4.2% reduction in GHG emissions. FIFO integration led to production gains of up to 15.1% and emission reductions of up to 9.5%. These findings underscore reinforcement learning’s potential for sustainable mining in real-world applications.