.png){kind=icon}
.png){kind=icon}
Return
Pressure difference prediction model based on blast furnace big data and GWO-FE-CatBoost
.png){kind=icon}
PRE
AI
J
M
Q
M
Z
Q
DOI：10.1007/s42243-026-01850-z.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Maintaining stable pressure difference within an appropriate range was a critical approach to ensuring high-quality production, high yield, and low consumption in blast furnaces. A pressure difference prediction model was established based on blast furnace big data, enabling operational staff to proactively manage blast furnaces. Data preprocessing techniques were employed to enhance blast furnace industrial data governance, thereby improving data quality. Feature engineering (FE) was utilized to construct a feature set for pressure difference characterization, incorporating 58 blast furnace parameters selected through recursive feature elimination and 32 derived features generated via variational mode decomposition. These features exhibited strong correlations with pressure difference. Gradient boosting decision tree, light gradient boosting machine, and categorical boosting (CatBoost) were implemented to establish the pressure difference prediction model. The application of FE reduced the mean squared error of prediction models by an average of 14.32%. The hit rate within the ± 2.5 kPa error margin increased by 9.26%. After hyperparameter optimization was performed using the grey wolf optimizer (GWO), the GWO-FE-CatBoost framework achieved optimal predictive performance, yielding a mean absolute percentage error of 0.78%, a mean squared error of 3.07, and a hit rate of 88.62%. Additionally, a periodic optimization mechanism was proposed, which enhanced the hit rate by 7.88% during the testing period. This approach effectively maintained stable high-precision predictive performance. The robust pressure difference prediction results gained unanimous recognition from field engineers. This reliable predictive trend proved essential for facilitating proactive blast furnace control.
Keywords:
Blast furnace
Pressure difference prediction
Feature engineering
CatBoost
Optimization mechanism
Journal
IF:
3.6
Papers:
3.6K
Citations:
6.1K
Organization
Cited Papers