Multistep time-series prediction model for blast furnace permeability index based on deep learning

The permeability index of a blast furnace is a key parameter that reflects the gas–solid flow balance, and its accurate prediction is crucial for ensuring stable furnace operation. To address this inherent strong time-series dependency, this paper proposes a deep-learning framework based on a Bayesian-optimized temporal fusion transformer (TFT) model for multistep permeability index prediction. First, a high-quality dataset suitable for multistep prediction was constructed using preprocessing techniques integrated with feature selection methods, including Spearman and maximal information coefficient, thereby laying a robust foundation for model training. Second, a multistep prediction model based on the Bayesian-TFT was developed, wherein Bayesian optimization was employed to refine the TFT hyperparameters, enhancing the efficiency of training convergence. Subsequently, a rolling window strategy combined with 10-fold cross-validation was adopted to improve the generalization capability of the model under complex operating conditions, achieving the prediction accuracies of 97.204% and 98.454% within acceptable error ranges. Finally, interpretability analysis was conducted to identify the core control parameters, providing a basis for the predictive regulation of permeability. This study enabled a one-hour-ahead multistep prediction of the permeability index, offering a novel approach for intelligent blast furnace ironmaking and demonstrating significant engineering value.