Prediction of hot metal silicon content in blast furnace based on EMD and DNN

Abstract

In the blast furnace iron-making process, the prediction of the silicon content in hot metal is one of the most important but difficult. This paper proposes a novel combined algorithm based on Empirical Mode Decomposition(EMD) and Dynamic Neural Network (DNN) for predicting the silicon content of hot metal in blast furnace. To eliminate the mutual interference of different frequency components of the original historical data, the EMD algorithm decomposes the original historical data into a series of different frequency and stationary Intrinsic Mode Functions (IMFs) and a residue. And then each IMFs and the residue was approximated to their Nonlinear Autoregressive Model (NARM) and predicted by DNN, finally the prediction of silicon content will be obtained by summing the prediction of each IMFs and the residue. At last, with an experiment of some sample data of silicon content that collected from an ironworks in China to verify our algorithm, the results indicate that the combined algorithm we proposed has better perferrmanc than a single algorithm without EMD, which shows the validity of the proposed algorithm.