Online Detection of Component Concentration in Synthetic Sodium Aluminate Solution Using Orthogonal Regression and BP Neural Network

The online measurement of component concentrations in sodium aluminate solution is crucial for the Bayer alumina production process. In this paper, the orthogonal regression and back propagation (BP) neural network model were introduced to predict the caustic ratio and caustic alkali concentration of sodium aluminate solution, respectively. The function relationship among conductivity, refractive index, temperature and caustic ratio of sodium aluminate solution using the orthogonal regression model was established. The BP prediction mode performing conductivity, refractive index, temperature and caustic ratio (calculated by orthogonal regression function) as input variables was created. The experimental data were employed to verify the feasibility and effectiveness of the proposed approach, which demonstrated that the predicted errors generated by the orthogonal regression model are within ± 5%. The results indicated that the BP neural network is more accurate than the traditional radial basis function neural network and general regression neural network for the prediction of sodium aluminate solution concentration.