The FeO content of sinter acts as a crucial indicator for sinter quality as well as for maintaining stability during blast furnace smelting. To improve the accuracy of FeO prediction, this study proposes a novel method based on a VAE–CNN–BiLSTM hybrid model. First, a variational autoencoder (VAE) is employed for anomaly detection and missing value imputation to improve the completeness and reliability of raw production data. Second, a convolutional neural network (CNN) is used to automatically extract key features from the sintering process. To optimize the model input, a 1 × 1 convolution weight analysis is applied to identify the factors most strongly correlated with FeO variation. Finally, a bidirectional long short‐term memory (BiLSTM) network models the refined data to accurately predict FeO content. Experimental results demonstrate that the proposed model achieves an R 2 of 0.9897, a mean absolute error of 0.01764, and an root mean square error of 0.00146 on the test set. Overall, the VAE–CNN–BiLSTM‐based prediction method significantly improves the accuracy and robustness of FeO content forecasting, providing reliable data support for sintering process optimization and the stable blast furnace operation.
Li et al. (Tue,) studied this question.