Grade prediction modeling for froth flotation based on NARX-NN and ensemble learning

In mineral flotation, concentrate and tailing grade are indices directly relating to plant economics. Realizing online detection of grade is costly and difficult. In China, offline analysis is employed in most flotation plants, resulting in delayed detection and control. A grade prediction model based on nonlinear autoregressive networks with exogenous inputs (NARX) and ensemble learning is herein proposed. Concentrate and tailing grades are influenced by many factors in the complex flotation process. The prediction model structure is determined by mechanism analysis in that process. Based on the correlation analysis between process conditions and the grade, process condition input and feedback delays are determined. To improve the model accuracy and generalization ability, a NARX neural network was trained as the base learner and support vector regression was the second learner. A grade prediction model was established and shown to effectively track grade dynamic fluctuations with high prediction accuracy and stability.