Using Genetic Algorithms for Underground Stope Design Optimization in Mining: A Stochastic Analysis

Mine design and mine planning are essential to a mining operation as they do not only dictate the outline of the mine, but essentially determine the financial robustness and success of the mining operation. However, as mining is a conservative business mine design is still mostly done by hand. To indicate the effect of uncertainty in mine design, Vallee (2000) reported that 60% of the reviewed mines had an average production rate that was 30% lower than their designed capacity. To provide in better mine design researchers and companies have developed and proposed number of guidance tools in recent years to help optimize mine planning. However their efforts have mainly focussed on open-pit mine design whilst largely ignoring underground mining, which is more versatile and therefore more difficult to assess. Furthermore, the optimization algorithms that have been developed are mostly based on average-type or interpolated resource models, which do not allow assessing uncertainty. Several stochastic approaches have been proposed by Dimitrakopoulos and others as a tool to involve stochastic techniques in mine design optimization as a way to assess uncertainty. However, few practical stochastic solutions have been developed for underground mine design. To contribute to the recent efforts made by the TU Delft in the ‘Horizon 2020 – Real Time Mining’ European Union Research and Innovation Programme, this research has focussed on combining underground mine design optimization with several stochastic analyses. A stope layout optimizer was developed in Matlab based on standard genetic algorithms, a sub-group of evolutionary algorithms. This model was tested on two validation data sets where it showed good optimization performance. A stochastic optimization module was added and tested, but as the validation data set was linear, it showed no added value. Using a resource model of a copper / zinc VMS deposit a number of stochastic approaches were tested. By performing a minimal downside risk / maximum upside potential analysis a stochastic optimization and elimination approach was tested under different economic scenarios by using 20 unconditional economic block model simulations. A stochastic risk analysis using the same 20 economic block models was done to study the effect of stochastic probability on economic value, ore tonnage, waste tonnage, average ore grade and average arsenic content. Lastly, a stochastic design optimization was done using 100 unconditional economic block models and its optimization performance was compared to that for a traditional non-stochastic optimization. Both the minimal downside risk / maximum upside potential analysis and the stochastic design analysis showed better optimization results than traditional optimization using average type orebody models. However, as in the min / max approach emphasis was on robustness instead of profitability, it was unable to show the same degree of improvement over traditional optimization.Civil Engineering and GeosciencesGeoscience & EngineeringResource Engineerin