Long short-term memory based on random forest-recursive feature eliminated for hot metal silcion content prediction of blast furnace

Abstract

The variation of hot metal silicon content simultaneously reflects the product quality and the thermal state of blast furnace (BF), and it is a vital economic index to monitor in the iron-making process. This paper proposes a model to predict the change trend of silicon content, which integrates random forest-recursive feature eliminated (RF- RFE) with long short-term memory (LSTM). In the process of feature selection, the ensemble learning method of RF can obtain subsets with high classification accuracy, and RFE further removes some irrelevant features to determine the best input variables. Then, LSTM is presented as the predictor, which conquers the shortcoming of traditional RNN vanishing gradient, and the gating mechanism processes the correlation information before and after the time series adequately. The effectiveness of the proposed solution for silicon content prediction of a nonlinear process is proved by a case study of real production data from BF. The results also show that the proposed method has better prediction accuracy than other data-driven models.