Spatio-temporal and multi-mode prediction for blast furnace gas flow

Abstract

The reasonable and stable distribution of blast furnace (BF) gas flow is the basis for maintaining the smooth operation of BF. Therefore, the accurate detection of the gas flow distribution is essential in the BF ironmaking process due to the direct impact on productivity, stability, and efficiency. However, there is a significant challenge to capture the complex interactions and dynamic changes of the ironmaking process by single predictive mode and two-dimensional (2D) distribution, leading to a lack of flexibility and interpretability in dealing with different abnormalities. To address this issue, a novel spatio-temporal multi-mode approach for three-dimensional (3D) BF gas flow prediction is proposed in this article. First, Pearson correlation analysis is employed to evaluate correlated variables in the spatial dimension. The precise temporal correlations among the multiple variables are matched with mutual information (MI) to extract spatio-temporal variables. Next, the spatio-temporal variables are decomposed utilizing variation mode decomposition (VMD), and the noise is removed with integrated correlation analysis and Fourier transform (FT) to identify and retain the relevant information. Finally, the MI-VMD-Informer is innovatively proposed to establish three different prediction modes based on spatio-temporal features, thus obtaining 2D and 3D gas flow distributions. The superiority of the proposed method is verified by actual BF production data.