Combined with air material consumption features， this paper proposes a method combining the rough set（RS） and the support vector machine（SVM） for air material consumption prediction. Through attribute reduction of the incomplete information system of RS， redundant quantitative factors in the air material consumption information system are eliminated. On the basis of the importance of attributes， seven factors are reduced to three， and the core knowledge of the system is retained. After the introduction of SVM improved by the particle swarm optimization （PSO） model， this approach obtains the optimized parameter combination and establises the forecast model of air material consumption. The case study demonstrates that the prediction accuracy of the RS-PSO-SVM model is better than those of the PSO-SVM model or the RS- back propagation（ BP） model.