An Optimization Regression Model for Predicting Average Temperature of Core Dead Stock Column

Hearth activity is one of the most important factors which affect the smooth progress of production and even the life of blast furnace. However, the calculation of hearth activity depends on the empirical model entirely, and the model parameter acquisition is difficult. To overcome this deficiency, this paper presents a novel method based on an improved multiple linear regression model to predict average temperature of core dead stock column for evaluating it. In the algorithm, the Pearson correlation analysis, metallurgical formulas and the Akaike Information Criterion based on least square method are used to establish a multiple linear regression model. The method makes the estimation of hearth activity out of the empirical formula. And it is easy for the evaluated model to obtain parameters. Meanwhile, experimental results show our proposed method can achieve 0.69% average relative error on the test data set and average relative error of 0.57% on the training data set. Moreover, the function of low average temperature of core dead stock column warning can be realized.