A Data-Driven Prediction Model of Blast Furnace Gas Generation Based on Spectrum Decomposition

Abstract

Blast furnace gas (BFG) is an important secondary energy in the iron and steel industries, and its efficient and reasonable utilization is the key to improving the economic efficiency of enterprises and the level of energy conservation and emission reduction. Aiming at the problems of difficult accurate modeling on the generation process and difficult prediction of real-time flow, this paper proposes a generation prediction model based on spectrum decomposition. Firstly, the major chemical reactions, production process, and data characteristics of blast furnace are analyzed, and the input variables for the prediction model are reasonably selected based on the correlation analysis results. Then, according to the spectrum characteristics, the BFG data is decomposed into low-frequency and medium-frequency parts by two finite impulse response filters. Next, for the low- and middle-frequency components of data, a low-frequency component prediction model based on the support vector regression, and a middle-frequency component prediction model based on the Elman neural network (ENN) are designed respectively. Finally, we decompose the spectrum of the actual industrial production data and find that the spectrum of the decomposed data basically meets the expected target, which verifies the effectiveness of the finite impulse response filters. In addition, we compare the prediction effect of the designed combined model with other models, such as the support vector regression, the back-propagation neural network, and the ENN. The final experimental results show the correctness, effectiveness, and superiority of the combined model and the spectral decomposition method proposed in this paper.