Prediction of a Blast Furnace Burden Distribution Variable

Abstract

A model for prediction of a variable that characterizes the burden distribution in the blast furnace is presented. The neural network model makes use of short-term temperature measurements from an above-burden probe as well as information about the charging program, and predicts a normalized temperature change, which can be considered to reflect the burden distribution. The basic structure of both an on-line and an off-line model is presented. By applying the models on process data from a Finnish blast furnace it is demonstrated that accurate predictions are obtained of the temperature changes caused by the dumps for the entire charging cycle.