Accurate prediction of blast furnace gas (BFG) generation is crucial for achieving rational utilization of gas and intelligent control of gas systems. Due to measurement instrument errors and production fluctuations, measured values of BFG generation are prone to anomalies. The current approach focuses on the self-identification and self-correction of BFG generation data, without leveraging the correlations among process parameters to conduct data governance on BFG generation. To address the above issues, a blast furnace was selected as the research object to construct a data governance model, which identified and corrected abnormal data from BFG generation measurements by utilizing the correlation between BFG generation and blast volume. Kalman filtering was applied to eliminate noise, and backpropagation neural network (BPNN), long short-term memory network (LSTM) and extreme gradient boosting (XGBoost) prediction models were built for result analysis. The results showed that the prediction accuracy of all models was significantly improved after data governance. The BPNN model demonstrated the optimal prediction performance with a mean absolute error (MAE) of 90.21 m³/min and a mean absolute percentage error (MAPE) of 2.48%, while LSTM and XGBoost models achieved MAE values of 133.92 m³/min and 125.19 m³/min, with corresponding MAPE values of 3.71% and 3.45%, respectively.