河北省科学院学报

烧结矿性能对高炉冶炼至关重要，为确保其质量稳定，需提前预测关键指标如转鼓指数。本研究提出了一种TCN-Attention-GRU-KAN混合模型，利用时间卷积网络(TCN)提取长期依赖关系，门控循环单元(GRU)捕捉短期动态关系，结合注意力机制融合特征，引入KAN层对GRU进行改进，增强特征表示能力。与GRU、TCN等模型相比，该模型在均方根误差、平均绝对误差和决定系数等指标上表现更优，决定系数达0.936 5,显著提高了预测精度。该模型具有较强的实际应用价值，可为企业优化配矿方案提供参考。

Sintering performance is crucial for blast furnace smelting. To stabilize product quality, key indicators such as the drum index must be predicted accurately before the process. This study proposes a TCN-Attention-GRU-KAN hybrid prediction model. The model leverages Temporal Convolutional Networks(TCN) to extract long-term dependencies and Gated Recurrent Units(GRU) to capture short-term dynamics. Feature fusion is enhanced through an attention mechanism, while the GRU is improved by introducing a Kolmogorov-Arnold Network(KAN) layer to strengthen feature representation. Compared with GRU and TCN models, the proposed model outperforms in metrics including Root Mean Square Error(RMSE), Mean Absolute Error(MAE), and the Coefficient of Determination(R²), achieving an R² value of 0.936 5 with significant improvement in prediction accuracy. The model exhibits strong practical applicability and can inform industrial optimization of burden blending strategies.