Aimed at the specific characteristics of runoff forecast and related issues, this paper first established a diversity changeable gray prediction model. Based on the analysis of the advantages and shorts of application of the diversity changeable gray prediction model ,non-time-varying immune neural network model, least squares support vector machine model in runoff forecast ,it discussed the practical significance of predictive modeling of parallel combination, and integrated the three model into parallel combination model so as to improve the precision of the sample data and full played their respective advantages and complement each other. Last, taking the annual runoff forecast of Xinjiang Yili Yamadu hydrological station for example, the paper set up the parallel combination model for annual runoff of the station, and confirmed the rationality, universality and reliability of the combination forecast model by comparative analysis of the predicted results of three individual models.