Methods for Optimising Short-term Production Scheduling in Open-pit Mines

Abstract

Open-pit mine production planning involves decision-making for material extraction and allocation, divided into strategic and tactical planning. Strategic planning focuses on long-term economic value, while tactical planning addresses productivity and ore quality over shorter periods. This thesis contributes to tactical planning from both deterministic and stochastic perspectives. The deterministic approach solves static optimisation models with a focus on ore quality, while the stochastic approach manages uncertainty to ensure robust production plans under varying conditions.

Tactical planning models often involve numerous binary variables and bilinear constraints, posing computational challenges. Equipment uncertainties further complicate these models, making exact methods impractical. Traditional stochastic programming methods require many scenarios, adding to computational complexity.

To address these challenges, this thesis enhances the computational efficiency of a non-linear mixed integer programming model, enabling rapid responses to mining disruptions. A deterministic planner, Predictive CE, was developed using a sliding-window strategy, cross-entropy method, and probabilistic heuristic. Numerical tests show Predictive CE solves short-term non-linear production planning problems in minutes, producing solutions comparable to Gurobi, which struggled with large problems even after 10 hours. Predictive CE outperformed two alternative methods in speed, with quality gaps of 1.22% and 2.92%.

For managing operational uncertainty, a stochastic decision-making framework is proposed, featuring a robust planner, a discrete-event simulation module, and a dynamic truck-dispatching heuristic. The robust planner, central to this framework, optimises productivity under uncertainty. Simulation results show it can enhance productivity by up to 31.7% compared to a nominal planner, which only considers deterministic conditions.