REINFORCEMENT LEARNING FOR MINING PRODUCTION PLANNING

Machine learning (ML) and artificial intelligence (AI) significantly impact society on the last years, including, in solving complex Mixed-Integer Linear Programming (MILP) problems. This paper investigates the use of Proximal Policy Optimization (PPO), a reinforcement learning algorithm, for assisting the solution of MILP in mining production planning. Efficient planning is crucial for optimizing resources and ensuring economic feasibility in mining, where traditional methods struggle with large datasets and dynamic conditions. We compare PPO's performance with open-source and commercial solvers using a real-world MILP case study. Our results demonstrate that PPO improves the process of finding feasible and optimal solutions in complex problem instances, highlighting its potential for more efficient and scalable solutions.