A novel intelligent optimization and guidance method for blast furnace oriented to increasing yield of hot metal

In the traditional blast furnace (BF) ironmaking process in China, a notable deviation exists between the theoretical and actual yield of hot metal, leading to unexpected iron loss and restricting the improvement of production capacity, which cannot adapt to the increasingly intensified smelting rhythm. Focusing on a BF in a Chinese steel enterprise, a deep neural network algorithm was designed to model the impact of multiple parameters on actual yield of hot metal in a single BF smelting cycle, successfully accomplishing the theoretical computation and real-time prediction of yield of hot metal for subsequent, unknown BF smelting cycle. Test results show that the proposed algorithm demonstrates an impressive prediction accuracy of 86.7% within an error range of ± 10 t and can swiftly complete the training and convergence process in 32.5 s. By integrating prediction results with Nomogram, a regulatory mechanism was engineered to minimize the deviation between theoretical and actual yield of hot metal. This mechanism ensures the yield enhancement of hot metal through dynamic adjustments of BF operational parameters. Industrial-scale application experiments confirmed that the intelligent operation and optimization system, developed in the laboratory, can maintain the yield deviation of hot metal within a stable range of 30 t, achieving a maximum reduction in iron loss rate of 17.65% compared to that before system operation. The findings provide robust support for the yield increase and efficiency improvement of the experimental BF.