Abstract: In view of the disadvantages of SVM (support vector machine) such as a large number of support vectors and strict demand for kernel function, we introduce RVM (relevance vector machine) to establish dam safety model which has better performance. Kernel function and its parameters have important effects on the performance of the RVM model. Mixed kernel function in association with local and global kernels can improve the fitting accuracy and generalization ability of the model. The optimized parameters of the kernel function can be effectively found by using PSO (particle swarm optimization) algorithm. However, the defect of local optimal point easily occurs in normal PSO algorithm. In light of this, we apply an algorithm of improved particle swarm optimization (IPSO). On the basis of combined algorithms above, we establish a model for dam safety, and the results indicate that the performance of RVM model with hybrid kernel is superior to that of conventional model.

Key words: dam safety modeling, relevance vector machine, hybrid kernel function, adaptive particle swarm optimization, fitting accuracy, generalization ability