Investigation into simulation based optimization of a continuous mining operation

Optimization in the mining industry is often only done on a discontinuous and long-term basis. The RTRO-Coal project aims to improve the mining efficiency by introducing a continuous optimization cycle. As a part of the RTRO-Coal project, an experimental simulation model, validated in the lab (TRL 4), for short-term mine optimization was developed. In the first part of this research, the experimental simulation model is extended and validated. Before validating the simulation model, historical production data from the Profen open-pit coalmine in Germany is analysed and processed for simulation model calibration and benchmarking during the validation. The simulation model input parameters, such as resource model, excavator digging rates, excavator scheduling and stochastic failure data are derived and calculated from this historical data. To prepare the simulation model for usage with an optimizer, the simulation model is embedded in a framework with an interface to the simulation controller, which sends the input parameters to the simulation model and processes the results. With the simulation parameters set up, a validation strategy is developed and three experimental setups are set up with varying parameters to validate the model and establish its accuracy under various circumstances. The three experiments show that while the simulation model generally provides a good match with the reality when its parameters are derived from historical data, it deviates from the reality when these parameters are replaced with stochastic variables. Especially the equipment failures are shown to vary greatly between the considered experiment scenarios, which makes it hard to model them without increasing the uncertainty. While this causes the model to deviate on a daily level, over a longer period the predicted production matches the reality relatively well. By validating and demonstrating the simulation model in the industrially relevant environment, TRL 6 is achieved. The second part of the research focusses on simulation based optimization using the verified simulation model. Several algorithms are considered that meet the required conditions for the computationally expensive simulation model and the large solution space. The Key Performance Indicators (KPIs) for optimization are decided and a penalty function which calculates the input parameter fitness based on the simulation model results is defined. For two optimization algorithms, a Simulated Annealing and a Genetic Algorithm, custom implementations are developed based on the desired properties of the optimizer considering optimization time and convergence speed. The performance of the two developed algorithms is examined by running several experimental optimizations with different configurations. Because these experiments show that the Simulated Annealing has better performance on a local scale and the Genetic Algorithm on a global scale, a hybrid Genetic and Simulated Annealing algorithm is developed which combines both methods to an algorithm that is able to find a good solution in a reasonable time given the complex optimization problem. It is thus concluded that simulation based optimization can increase the efficiency of a continuous mining operation.Resource EngineeringGeoscience & EngineeringCivil Engineering and Geoscience