Aiming at the hysteresis of FeO content in sinter detected by chemical analysis,which is difficult to reflect the sintering process state in real time,and the existing prediction models only focus on strong correlation parameters under specific working conditions,and it is difficult to adapt to the dynamic changes of parameter correlation,resulting in insufficient generalization ability and prediction accuracy of the model,a gated recurrent unit (GRU) network prediction model fused with temporal attention mechanism (TPA) is proposed. Firstly,the sintering production data collected at the industrial site is preprocessed,including outlier detection,outlier and missing value processing,and data standardization to ensure data quality and model training stability. Secondly,the attention mechanism is used to allocate the weight of each input feature parameter to realize the automatic screening of key features. Finally,the screened eigenvalues are input into the GRU network for training,and a prediction model of FeO content in sinter is constructed. Taking the actual production data of a sintering plant in 2024 as the research object,the prediction effect of the model is verified through experiments and compared with other neural network models. The results show that the model evaluation index R² reaches 0. 923 1,the root mean square error (RMSE) is 0. 179 1,and the mean absolute error (MAE) is 0. 203 5,showing strong generalization ability and prediction accuracy. This study can provide a real-time decision-making basis for batching optimization and process regulation in sintering production with important engineering application value for stabilizing the quality of sinter