Abstract： In order to deal with the problem that empirical model of by-product gas consumption in iron and steel enterprises cannot accurately predict the by-product gas consumption, the analysis of by-product consumption user and its characteristics is carried out. Support vector machine is used to classify the by-product gas consumption. It combines Powell calculation, simulated annealing arithmetic calculation, and support vector regression machine (SVRM), a model of PSA (Powell simulated annealing)-SVRM to predict the by-product gas consumption has been established. By-product gas consumption data in some iron and steel enterprises are used in the model. The results show that the prediction accuracy for sintering, steelmaking and continuous casting processes is 94.8%, 94.9% and 100%, respectively, with the relative mean error of 2.5%, 2.8% and 2.1%, respectively, which indicates that this PSA-SVRM model is suitable to prediction of the by-product gas consumption. Wilcoxon sign rank test proves the effectiveness of PSA-SVRM model.