Multivariable Prediction of Blast Furnace Status Driven by Data

The generalized regression neural network (GRNN) combined with particle swarm optimization (PSO) model for multivariable prediction of the blast furnace (BF) is proposed in this article. The smelting processes are introduced to confirm the inputs and target variables in the first step, and then correlation analysis (CA) is conducted to obtain the exact time lags between the inputs and target variables. The weighting factors of different inputs are calculated based on CA, which puts the inputs in different weights to improve the specificity of data. The GRNN multivariable prediction model is adopted, in which the PSO algorithm is used to optimize the smoothing factor of the GRNN; thus, the GRNN-PSO model with the best performance is constructed. Finally, the gas utilization rate (GUR) and top temperature of the BF status are chosen as the target variables, and the performance of the GRNN-PSO multivariable prediction model is evaluated under different time scales. The accuracies of the first prediction target GUR with 12, 24, and 100 h of the samples were 1.0, 1.0, and 0.998, respectively, and they were 1.0, 1.0, and 0.852 for the second prediction target of top temperature. The accuracy of the proposed weighted input scheme was increased by 3.1%. Experimental results verified the effectiveness of the proposed GRNN-PSO model.