Prediction of production rate of surface miner in coal mine: an application of single and ensemble machine learning methods

초록

Surface miner is an eco-friendly excavation machine which is gradually replacing the traditional drilling and blasting method of excavation. Hence, accurate determination of the production rate of surface miner is very imperative. This study therefore proposes machine learning based models for the prediction of production rate (PR) of surface miner in a coal mine. The artificial neural network (ANN), multivariate adaptive regression spline (MARS) and support vector regression (SVR) machine learning methods were developed using the machine parameters, rock mass and intact rock properties. The performance of the models were evaluated using the basic statistical indicators such as coefficient of determination (R2), root mean square error (RMSE) and mean absolute error (MAE). Three ensemble models combining the merits of the single models were then developed to enhance the performances of the proposed standalone models. Thereafter, a rigorous statistical selection regime was developed to select the most suitable models for the PR prediction. The outcome of the study revealed that ANN model has the highest overall R2 value of 0.975 with lowest RMSE of 8.95 and MAE of 7.16 out of the standalone models follow by the MARS model having R2 value of 0.95, RMSE of 12.98 and MAE of 10.28, then SVR model with R2 value of 0.93, RMSE of 15.6 and MAE of 12.3. All the three ensemble models have R2 values greater than those of MARS and SVR models while only non-linear neural ensemble (NNE) model one of the three ensemble models has the highest overall R2 value of 0.978 with RMSE of 8.65 and MAE of 6.84. The non-linear neural ensemble model was selected by the rigorous statistics as the best. A MATLAB implementation code for the practical implementation of the proposed models was presented. The sensitivity analysis was conducted and the coal thickness was found to be the most influencing model parameter. © The Author(s), under exclusive licence to Springer-Verlag GmbH Germany, part of Springer Nature 2024.

키워드