Real-Time Automatic Detection of Sodium Aluminate Solution Concentration Based on PSO-BP Neural Network

Timely and precise determination of sodium aluminate solution is a prerequisite for optimizing the operation and improving the quality of the Bayer process in alumina production. An innovative prediction method that integrates a particle swarm optimization algorithm (PSO) with a back-propagation (BP) neural network model was introduced to achieve online control of sodium aluminate solution components in this paper. A hybrid prediction mode for forecasting the concentration of alumina and caustic alkali employing the PSO-BP neural network was developed, with conductivity, refractive index, and temperature serving as input variables. The feasibility and effectiveness of the proposed approach were validated through experimental data. The results demonstrate that the root mean square error (RMSE) and mean absolute error (MAE) predicted by the normalized PSO-BP model for the caustic alkali concentration are 2.360 and 1.778, and those for the predicted alumina concentration are 3.641 and 2.916, respectively, with smaller and more accurate errors than those of the traditional BP neural network model.