On-line Prediction Method of Ferrous Concentration in Iron Removal Process by Goethite

To deal with the problems of detecting the concentration of ferrous ions in iron removal process by goethite for its complexity, long lag and nonlinearity, we propose an LSTM encoder-decoder model with attention to predict the ferrous ions concentration. To increase the robustness of the system and predict the changing trend of the ferrous ions, an error compensation model is designed, and we combine these two models into an integrated model in a parallel way. The industrial experiments in the iron removal process by goethite of zinc hydrometallurgy plant in China demonstrated the good performance of the proposed integration model. The attention values in the attention mechanism also show which input variable should be cared more in the real industrial field, which provides a solid foundation for the optimal control of iron removal process by goethite.