3.2 Results The results of this agent permit not only training also predicts mine plans without specific knowledge of mining theories or algorithms. It shows in Figure 6 the mining plan with waste and ore requirements by period. The stacked bar chart displays ore and waste tonnages per period, while the line plot shows the corresponding copper grade trajectory. The agent learned to prioritize high-grade material in early periods to maximize discounted cash flows, resulting in a front-loaded grade profile that decreases from 1.08% to 0.67% over the 38-period life of mine. Total material moved: 3,040 Mt with a stripping ratio of 3.03:1. Figura 6 -Optimized production schedule generated by the Q-Learning agent . Figure 7 shows that the agent selects the optimal cut-off in the initial periods to maximize project’s Net Present Value (NPV), resulting in improved cash flows at the beginning of the mining operation. This type of optimization strategy is common in different approach like Mixed-Integer Lineal Programming (MILP) or metaheuristic algorithm (GRASP, TABU SEARCH). However, when using Reinforcement Learning (RFL) approach, the agent has the power of select other goals or include new paradigms in project evaluations such real options analysis, enabling of better opportunities throughout the mine’s life cycle.
RkJQdWJsaXNoZXIy MTM0Mzk2