The full use of blast furnace gas (BFG) can effectively reduce the consumption of primary energy, but the working condition of the blast furnace site are constantly changing, and the relationship between gas supply and demand is always in an unbalanced state, resulting in the phenomenon of gas escaping. In order to improve the utilization rate of gas, a BFG prediction method based on Temporal Attention (T-Attention) network is proposed. Firstly, the key factors affecting BFG content are selected by combining the blast furnace smelting mechanism and the maximum mutual information coefficient (MIC). Then, in view of the random perturbation in the collected data, a wavelet analysis is used to remove the noise in the data. In the modeling process, the gated recurrent unit (GRU) is used to capture the periodic fluctuation law in multivariate data, and the attention mechanism is integrated to calculate the coupling relationship between each variable and the predicted value in each sample in real time and distribute the weights, so as to improve the dynamic adaptability and interpretability of the model. Finally, the field data of blast furnace of a steel plant is used for verification. The results show that the prediction effect of the T-Attention network model is better than that of the traditional method, and it can accurately predict the indicators in BFG, and provide timely and accurate decision-making reference for BFG scheduling and energy conservation and pollution reduction in the later stage.