Abstract: Combination forecast modeling is essentially a process that selects and makes use of the information gained through single forecast models.How to make judgment about the conditions that the single forecast model should meet is analysed.Moreover,the co-integration theory is applied to screen models,and a comprehensive method is proposed to screen the single forecast model for combination forecast modeling in three aspects so as to improve forecast precision and simplify computation.In the initial single forecast model,some representative single forecast methods,such as state space models and neural networks,are adopted except for the common forecast methods.The adoption of these fine forecast models improves the precision of combination forecast to a certain extent.