Monte Carlo sensitivity analysis results from 1,000 stochastic simulations using the QLearning basic_parallel model. (Top) Summary statistics showing the expected NPV of $967.85M with a standard deviation of $98.4M (CV=10.2%), indicating robust model performance. The 84.7% probability of achieving NPV >$900M demonstrates high confidence in project viability. (Middle-left) NPV distribution histogram approximates a normal distribution (slightly right-skewed), with P10-P90 range of $252.7M representing the uncertainty band. (Middle-right) Cumulative probability S-curve enables target-based decision making. (Bottom-left) Tornado diagram reveals copper price as the dominant sensitivity factor (±$166.7M impact), followed by discount rate (±$89.3M) and operating costs (±$54.2M). (Bottom-right) Value at Risk analysis shows 95% confidence interval of [$805.2M, $1,130.5M]. The balanced strategy (P50) is recommended for optimal risk-adjusted returns with coefficient of variation of 10.2% and positive skewness favoring upside potential. 4. CONCLUSIONS This research demonstrates that integrating Lane’s Theory with Reinforcement Learning (RL) provides a superior framework for strategic mine planning compared to traditional deterministic models. The following key findings highlight the impact of this approach: • Dynamic Adaptability: Unlike static MILP models, the RL agent successfully internalizes market volatility and operational stochasticity, providing real-time cut-off grade policies that maintain an optimal NPV path without the need for full re-optimization. • Computational Efficiency: The proposed architecture achieves convergence to optimal solutions significantly faster than iterative metaheuristic methods. This reduction in the "decision gap" allows planners to evaluate multiple "what-if" scenarios in hours rather than days. • Economic Superiority: Empirical results confirm that the autonomous agent consistently identifies higher-value extraction sequences by effectively balancing the trade-offs between processing capacities and ore quality over the Life of Mine (LOM). • Scalability for Mining 4.0: The modular nature of the MDP environment developed in this study serves as a scalable foundation for fully autonomous mine management systems, bridging the gap between theoretical economic optimization and real-time operational execution. In conclusion, this AI-driven approach not only maximizes the economic potential of nonrenewable resources but also fosters a "smarter" and more resilient mining industry, directly addressing the global challenge of efficient and responsible mineral delivery. REFERENCES Lane, K. F. (1988). The Economic Definition of Ore. Cleveland-Queensland, Australia: COMET Strategy Pty Ltd.
RkJQdWJsaXNoZXIy MTM0Mzk2