Real-time moisture control in sintering process using offline-online NARX neural networks

Abstract Sinter is the main raw material for blast furnace iron making. To provide high quality sinter, the moisture content of the mixture in the sintering need to be in the best range. However, most of the sintering is still artificial water adding which leads to a great variation in the moisture content of the mixture. The present work proposes a sintering parameter identification model using a nonlinear autoregressive model with exogenous (NARX). By exploiting the real-time and historical performing data, we set up a mixture adding water model involved the water and the major mixtures among sintering. Then, a combination of offline deep supervisor learning and online self-learning NARX algorithm is proposed. Finally, in the experimental stage, the results suggest the proposed method can effectively predict the moisture with an acceptable degree of accuracy.