The traditional parameter selection method using supportvectormach in e (SVM) is timeconsuming and difficult to obtain the optimal parameters becaus e it is based on artificial tests. So, this paper presents an automatic optimizat ionselection method based on genetic algorithm (GA). The global search capabil i ty of GA was adopted to obtain the optimum values automatically in the range of the SVM parameters. An SVM regression model was built to diagnose a smallscale water supply network. The results show that the forecast accuracy of the model o ptimized by GA is higher than that of the BP network. In addition, the overall d iagnosis performance of the SVM regression model using GA is superior to that of the traditional BP network.