An integrated machine learning approach for predicting the quantity and quality of mine waste rock drainage

Prediction of mine waste rock drainage is essential to waste rock management. In this research, multiple machine learning algorithms were integrated to predict the quantity and quality of waste rock drainage with regional weather data as the inputs: a tree-based algorithm to predict the occurrence of spring freshet, and two recurrent neural network algorithms to predict drainage flow rates and contaminant concentrations. The tree-based algorithms evaluated were decision tree, random forest, and AdaBoost; and the recurrent neural networks used were long short-term memory and gated recurrent unit. The machine learning approach developed was applied to a case study mine located in North America. The chemical constituents of interest (COIs) studied in this research were nitrate and selenium. Data, including 17-21 years of drainage flow rates data and 10-19 years of contaminant concentrations data, from three monitoring stations at the mine were used for the model training and validation. The data in the last monitoring year of the available dataset were assigned to the test set and the remaining data were assigned to the training set. According to the performance of all candidate algorithms, the random forest was selected for the prediction of spring freshet and the long short-term memory and the gated recurrent unit were used for the prediction of drainage flow rates and contaminant concentrations. The random forest algorithm was able to predict the occurrence of spring freshet with an accuracy of 91-95%. The long short-term memory and gated recurrent unit were capable of making robust predictions of the drainage flow rates and the contaminants concentrations. Assuming no significant changes in mining activities, this general approach may be capable of predicting the future loading of contaminants given the relevant weather data.Applied Science, Faculty ofMaterials Engineering, Department ofGraduat