A review of machine learning applications for underground mine planning and scheduling

Production planning and scheduling optimisation for underground mining operations has continued to attract significant attention over the last decades. This has been necessitated by the growing need for operations to meet their shareholder's expectations sustainably under increasingly challenging operational dynamics. Several studies have been undertaken to utilise mathematical programming models such as mixed-integer programming, heuristics and simulation algorithms including combinations of these techniques for production scheduling optimisation with some notable achievements noted in extant literature. However, the limited reach of standalone mathematical optimisation models under increasing volumes of input data spurred by the booming information technology (IT) platforms has become more apparent and pertinent for increased scholarly attention. The growing emergence of big data, driven by the industrial digitisation and automation has seen an increased appetite for data-driven optimisation planning and scheduling largely in manufacturing and operations management. However, the scarcity of discussion in this novel and fast-evolving area in the underground mining space presents a glaring blind spot that appeals for thoughtful conversations to narrow that gap. This paper seeks to discuss opportunities for application of data analytics and machine learning to improve production planning and scheduling efficacy in underground mining. Specific focus will then be narrowed to opportunities for incorporating predictive analytics and machine learning to improve the accuracy of mathematical optimisation models. The overarching intent is to support the attainment of mineral production targets through enabling schedule dynamic response to variability in key determinant variables such as ore grade and tonnages.