Hanging represents an extremely hazardous anomalous condition during blast furnace ironmaking, directly threatening the stability and safety of furnace operations. Timely and accurate prediction is a critical technological requirement for ensuring smooth blast furnace operation. Existing diagnosis methods struggle to fully uncover the data evolution patterns and feature correlations before and after hanging events due to insufficient utilization of multi-sensor information. To address these challenges, this paper proposes a method that integrates attention-optimized temporal convolutional networks (TCN) and bidirectional gated recurrent units (BiGRU), offering a novel and effective solution for predicting the occurrence of upper hanging. By integrating mechanistic knowledge to precisely select core sensitive data—such as blast furnace body static pressure and stock rod height—as modeling data, and using a combined oversampling and undersampling strategy to balance normal and abnormal samples. The model design incorporates the TCN to capture long-range temporal dependencies and the BiGRU to incorporate contextual information. Additionally, the proposed method introduces an attention mechanism that identifies critical time steps, aiming to enhance both the accuracy and interpretability of upper hanging prediction. Test results reveal an overall accuracy of 90.83% for the TCN-BiGRU-Attention, effectively capturing dependencies within time-series data to achieve outstanding prediction accuracy and stability in upper hanging. In comparison to other algorithms, the proposed hybrid model excels in upper hanging feature extraction and recognition, which significantly enhances the objectivity and timeliness of detection, providing reliable guidance for stable blast furnace operation.
Meng et al. (Thu,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: