Due to the frequent fluctuations of generation and consumption of blast furnace gas and difficulties to be predicted effectively in iron and steel works, the dynamic Lssvm prediction model with a timing update and self-correction function is established. The model is based on decomposed volatility and trend data of the generation and consumption after excluding the data "noise" by wavelet analysis, which combined with the actual blast furnace operating conditions. The amount of blast furnace gas generation of 3 200 m³ blast furnace and the consumption of the hot blast stove are taken as sample data to predict the future data in eight hours. The results show that the mean absolute error of the Lssvm prediction model with wavelet analysis has declined to 2.77%, the Update_Lssvm model is established to predict the accuracy of blast furnace gas generation date is 1. 55%, and blast furnace gas consumption of hot stove is 4.23%. The predict randomness problem of generation and consumption of blast furnace gas under the variable condition has been settled. Compared with other forecasting models, the prediction accuracy of the Update Lssvm model has been enhanced. The model not only has the generalization ability, but also provides a theoretical basis for optimal operation of blast furnace gas.