摘要：pH value is an essential factor in the control of froth flotation process. However, it cannot be measured online because the online-pH detector is easily damaged due to the poor field conditions, and maintenance is always delayed. Therefore, considering of the characteristics that the pH value fluctuate around a prescribed value due to the variation of the operating conditions when the ore is stable, and the prescribed control range of pH value changes when the ore type changes, multiple RBF networks based on sample classification and adaptive retraining strategy are proposed corresponding to these two characteristics for the online estimation of the pH value. Simulation results using the industrial data collected in a flotation process of bauxite show that an improvement in predictive accuracy and fitting capability can be achieved by adaptive multiple neural networks (Adaptive MNN) (RMSE=0.0957, R2 =0.6503) in comparison with the MNN (RMSE=0.1591, R2 =0.2312) and the single RBF neural network model (RMSE=0.2023, R2 = 0.1930).
