Abstract: The selection of kernel function for support vector machine affects the forecast accuracy of dam monitoring model. Based on support vector machine structure risk minimization theory and wavelet frame theory, a new method using wavelet kernel function instead of RBF is proposed. The parameters of support vector machine are optimized by particle swarm optimization algorithm. Through monitoring data of a practical project, the results of new model are compared with those of support vector machine of RBF kernel function model and statistical regression model, which shows that the support vector machine with wavelet kernel function has better accuracy and generalization ability.