Aimed at the shortcomings of optimal algorithm parameters of support vector machine (SVM) being difficult to determine and basic particle swarm optimization algorithm (PSO) being easy to fall into local extreme and other issues, ,the paper proposed immune particle swarm optimization algorithm (IAPSO) by using IAPSO algorithm to search SVM learning parameters, constructed IAPSO-SVM forecast model, and compared it with PSO-SVM, GA-SVM mode.Taking monthly runoff forecast of a hydrological statione in Yunnan province as an example ,it trained and forecasted the model by using data before 43 years and after 10 years the station was built. The results show that the absolute values of average relative errors of prediction by use of IAPSO-SVM model in 3 months dry season after runoff instance of 10 years are 3.32%, 6.52% and 6.55%.The prediction accuracy by IAPSO-SVM model is better than that by PSO-SVM model and GA-SVM model.The result showed that IAPSO-SVM model has higher prediction accuracy and generalization ability. IAPSO algorithm used density selection mechanism and immunization theory to improve the global searching ability and convergence speed of basic particle swarm optimization algorithm, and has stronger global optimization ability. The SVM learn parameters got by using the IAPSO algorithm to optimize can effectively improve the prediction accuracy and generalization ability.