Artificial neural networks for predicting silicon content in raw iron from blast furnaces

Abstract

Artificial neural networks often perform better than conventional statistical methods for correlations. Prediction of silicon content in pig iron from blast furnaces has been rather difficult and inaccurate, when possible, and mathematical modelling of the process is still qualitatively inadequate for the purpose. This paper illustrates the feasibility of using a feed-forward neural network for predicting silicon content from actual industrial data. The eight inputs to the network were silicon contents of two previous tappings, and some operational parameters, and the output was the predicted silicon content of the next tapping. Levenberg-Marquardt method was used for training the network by minimising the sum of squares of the residuals. The output of each node was calculated by the logistic activation (sigmoid) function on the weighted sum of inputs to that node, with different gain values. It is shown therein that this technique results in better predictions compared to conventional statistical correlation methods.