Hot Metal Silicon Prediction in a Blast Furnace by Using a Deep Neural Network

The blast furnace hot metal silicon (hot metal silicon in a sort [Si]) is an important parameter in steel making process. The vast variation in [Si] and considerable time delay exists in the offline analysis procedure; focusing on this, a real-time model was developed to achieve an online prediction and control [Si]. In this model, principal component analysis has been used for the input variable selection. A deep neural network combined with a genetic algorithm is used to predict [Si] content in advance, and fuzzy logic has been used to control [Si]. As [Si] content has a relationship with furnace parameters, the change of parameters in the furnace is reflected indirectly by [Si] content. Every input parameter has different inherent time lags in silicon transfer. The final 23 parameters with prominent influence on final [Si] have been included in the model. The model [Si] prediction accuracy is above 95% with an error between ± 0.1.