Iron temperature regulation method based on process optimisation and fuzzy control

This paper investigates the accurate prediction of molten iron temperature in blast furnace smelting. A prediction model based on Kolmogorov–Arnold Network-Long Short-Term Memory-Attention(KAN-LSTM-Attention) is proposed, combined with fuzzy control to achieve closed-loop temperature regulation. The model integrates the nonlinear mapping capability of KAN, the temporal feature extraction capability of Long Short-Term Memory (LSTM), and the dynamic weighting capability of the Attention mechanism. Experimental results show that the model exhibits low prediction error and good fitting performance on both training and test sets, achieving an 84% qualification rate within the industrial tolerance of ±15 °C. Furthermore, by adopting a fuzzy control strategy—through steps including fuzzification, rule reasoning, inference decision-making, and defuzzification—stable control of the molten iron temperature is realized, maintaining it within the range of 1495 °C to 1505 °C.