Hot blast stoves are dominant by-product gas consumers. Gas consumption data of hot blast stove group is weak in regularity and fluctuates sharply, making it difficult to predict. To accurately predict the gas consumption of a hot blast stove group, a prediction method of gas consumption of a hot blast stove group based on BP neutral network (BPNN) was proposed. It decomposed the gas consumption data of the hot blast stove group into several sub-datasets of each hot blast stove in the group, and predicted the gas consumption of every single hot blast stove according to the periodic gas consumption characteristics, and finally obtained the predicted gas consumption of the hot blast stove group by data recomposition. The real data of a hot blast stove group of an iron and steel site was collected as a case study. The results showed that the proposed model had a mean absolute error (MAE) of 2 978.74 m³/min, a mean absolute percentage error (MAPE) of 6.59%, and a symmetric mean absolute percentage error (SMAPE) of 6.88%. Compared to the traditional artificial neural network model, the values of MAE, MAPE and SMAPE of the proposed model had been reduced by 61.86%, 70.88% and 66.60%, respectively.