DYNAMIC MODELLING OF A SAG MILL-PEBBLE CRUSHER CIRCUIT BY DATA-DRIVEN METHODS

In a semi-autogenous grinding (SAG) mill and pebble crusher circuit, the behaviour of the comminution process is non-linear and time-varying due to wear and variations in the feed material. To describe such a complex system, data-driven models were introduced, along with a case study of a SAG mill circuit in Copper Mountain, British Columbia, Canada. This paper presented a mill power draw model using several regression algorithms like Artificial Neural Networks (ANN), K-Nearest Neighbours (KNN), Random Forest (RF), and Gradient Boosting method (GBM). The results were then combined by weighted mean squared errors to perform a more accurate ensemble model. The second method considered the comminution process as a time series problem, so the original dataset was re-structured, and a forecasting SAG power model was proposed by using Long Short-Term Memory algorithm (LSTM). This LSTM forecasting model applied 20 minutes of historical data to predict 2-minute SAG power draw in advance. It was shown that both the presented methods gave promising results for SAG mill power prediction