Abstract: Aiming at the problem of the limitations of the generalization performance of the single kernel extreme learning machine, we proposed to combine the reproducing kernel function with the polynomial kernel function to establish a new combined kernel extreme learning machine model, which had the advantages of global and local kernels, and selected cuckoo search algorithm to optimize its parameters. The simulation results show that it is feasible to use the combined kernel function based on the reproducing kernel as the kernel function of extreme learning machine. Compared with traditional support vector machine and single kernel extreme learning machine, the model has better generalization performance in multi\|valued classification and regression of experimental datasets.

Key words: cuckoo search algorithm, kernel extreme learning machine, combined kernel function