<p>Ferrous oxide (FeO) is a key indicator of sinter quality, significantly affecting blast furnace efficiency and energy consumption. To overcome the limitations of traditional FeO measurement methods—such as poor real-time performance and lack of uncertainty quantification—this study proposes an interval prediction model based on a hybrid Transformer and Long Short-Term Memory (LSTM) architecture integrated with adaptive bandwidth kernel density estimation (ABKDE). Gray relational analysis is first applied to select 15 process parameters highly correlated with FeO content as model inputs. The Transformer captures global dependencies, while the LSTM extracts temporal features for accurate point prediction. ABKDE is then used to model prediction error distributions, enabling uncertainty quantification and the construction of prediction intervals. Experimental results on production data from Chengde Iron and Steel Group show that the proposed model achieves superior point prediction performance, with MAE of 0.0891, MAPE of 1.0086%, RMSE of 0.1088, and R² of 0.9490. Under confidence levels of 85%, 90%, and 95%, the model also delivers the highest prediction interval coverage probability (PICP) and the lowest prediction interval normalized average width (PINAW) compared to benchmark models. This demonstrates its effectiveness as a reliable tool for intelligent control and quality optimization in the sintering process.</p>