Modeling hot metal silicon content in blast furnace based on locally weighted SVR and mutual information

The operation mechanism of blast furnace ironmaking process is characteristic of nonlinearity, time lag, high dimension, big noise and distribution parameter etc. Accurate prediction of silicon content in hot metal is an essential part of blast furnace operation. In this paper, mutual information (MI) is used as a preprocessor of model to select the principal features of original data, and then an improved model of support vector regression (SVR) is presented to solve the silicon content prediction problem. The proposed model modifies the risk function of the SVR algorithm with the use of locally weighted regression (LWR). Additionally, based on Mahalanobis distance, the weighted distance algorithm for optimization the bandwidth of weighting function is proposed to improve the accuracy of the algorithm. The proposed model exhibits superior performance compared to that of the SVR and other common models. The hit rate reaches 87% in successive 100 heats in test set. It seems promising and determinant in providing the experts with the right tools for the prediction in this difficult problem, and it can satisfy the requirements of on-line prediction of silicon content in hot metal.