Underground mine planning and scheduling optimization: Opportunities for embracing machine learning augmented capabilities

Mine Planning and scheduling in underground mining remains a key focal point for optimization studies to drive operational excellence and sustainable attainment of shareholders’ expectations. Manual and mathematical optimization methods have dominated the optimization criterion over the years with the latter gaining momentum due to its superiority in obtaining optimal solutions in both long-term and short-term planning horizons. However, the solitary utilization of mathematical models for the optimization of mine plans and production schedules has exposed fatal flaws related to the accuracy of planning inputs and overall prediction accuracy that require resolution through combinatorial models or a paradigm shift. Specifically, manual schedules and standalone mathematical models struggle to handle parameter perturbations and variability of key planning inputs over time. As such, optimal solutions from these models fail to intimately relate to the mining processes a short period barely after the production plan is published for execution. Fortunately, the technological footprint of Internet of Things (IoT) and the increasing data capturing and computational capabilities of modern-day computing hardware and software are evolving the production planning landscape enormously. Large volumes of historical performance data that have accumulated over the years for operations, and have been lying idle, are now readily available for in-depth analysis through data analytics and machine learning techniques to derive secondary useful insights that can improve the visibility of the operations beyond the immediate horizon. In particular, the enhanced prediction capabilities of machine learning models are perceived to be a game changer in the optimization of underground mining plans and production schedules through improved awareness of the progression of production activities over time. The prediction capabilities of machine learning models provide opportunities for improving the accuracy and validity of input parameters over longer time horizons and changing operating environments. Such an occurrence is undoubtedly expected to facilitate planning and mitigatory controls to be embedded in the mine planning and scheduling process. Further, this will potentially improve the accuracy of production forecasts, compliance with schedules, a true representation of schedules to operations, and above all, improve the accuracy and optimization of production schedules and plans.