Smart Mining Fleet Dispatching System to Reduce Greenhouse Gas Emissions Using Deep Reinforcement Learning
Date:
Artificial intelligence is expected to make future mining operations smarter and greener. To more efficiently dispatch mining haul trucks and reduce fossil fuel GHG emissions, a smart truck dispatching solution is developed in this study using deep reinforcement learning. The proposed algorithm trains each truck in the fleet to make real-time decisions based on loaded material, road traffic, estimated queueing, and maintenance needs. Compared to conventional approaches, the deep reinforcement learning solution improves productivity while reducing over 30% of GHG emissions, and overcomes the difficulty of handling operational randomness such as unexpected changes in fleet size and ore grade. The cost-effectiveness and scaling-up potential of the proposed solution are also evaluated. Comparison with other decarbonization technologies suggests that upgrading existing fleet management systems with AI costs about 200% and 1200% less than fleet electrification and carbon capture and storage, respectively, and can be rapidly adopted in the near-term. These advantages make smart fleet dispatching systems a feasible way to improve productivity and reduce GHG emissions for mining operations before other low-carbon technologies become cost-effective.