Connecting planning horizons in mining complexes with reinforcement learning and stochastic programming

Abstract

Listen

Translate article

Connecting short- and long-term production schedules in mining complexes is essential to ensure that the long-term production schedule is achievable at shorter timescales. Previous research that addresses optimizing mining complexes under uncertainty focus on simultaneously optimizing different components in the mining complex to capitalize on advantageous synergies. Typically, short- and long-term production schedules are optimized separately in a number of stages. This poses risk of schedule misalignment, which can adversely affect the economic outcome of a mining complex and the ability to meet long-term production forecasts at shorter timescales. A framework is proposed to jointly optimize short- and long-term production schedules by connecting planning horizons with stochastic mathematical programming and reinforcement learning. The solution approach is tested in a large operating copper mining complex and demonstrates significant improvements in the resulting production and financial forecasts.