A Blast Furnace Prediction Model Combining Neural Network with Partial Least Square Regression

The prediction of the important running variables of blast furnaces (BF) has been a major study subject as one of the most important means for monitoring the BF state in ferrous metallurgical industry. In this paper, a prediction model for BF by integrating a neural network (NN) with partial least square (PLS) regression is presented. The selection of influencing operational parameters of BF on variables to be predicted is developed according to the minimization of residuals based on the theory of path analysis. The selected influencing parameter data series are processed as the inputs of the prediction model. In order to validate this prediction model, the silicon content in hot metal of BF is taken as the variable to be predicted. The model is trained and evaluated with industrial data, and the results show that it works well. Further modification of this prediction model is also discussed to improve its industrial application.