In order to improve the prediction accuracy of river flow, the support vector machine(SVM)and the AR model are coupled to construct the SVM model for the river flow prediction with three-core hybrid kernel function. Taking the monthly runoff in the Weihe River Basin as an example, firstly through the time series analysis, the runoff data of Weihe River Basin was divided into trend data, seasonal data and random fluctuation data. A data set suitable for support vector machine algorithms was constructed with the AR model, and then was divided into the training set and the test set by 4∶1. Secondly, with the linear combination, this study constructed a three-core hybrid kernel function composed of polynomial kernel function, radial basis kernel function and Sigmoid kernel function. On the training set, the genetic algorithm was used to determine the relevant parameters, and predictions were conducted on the test set. It was found that when the genetic algorithm is used to determine the parameters, it will bring greater uncertainty with greater differences in results, thus it need more discussions on the parameter uncertainty brought by genetic algorithm. Through the function construction and statistical analysis, the general method and process of parameter selection of the three-core hybrid kernel function are given and verified. With this method, the uncertainty of the genetic algorithm can be reduced on the test set, and a more accurate flow prediction result can be achieved. The mean square error between the predicted flow rate and the actual flow rate is reduced from about 150 to about 130.