Abstract: The performance of support vector machine depends on the selection of kernel functions and kernel parameters. The Mexican Hat wavelet kernel function is constructed based on the wavelet analysis theory which could satisfy the Mercer conditions. Further, the Mexican Hat wavelet kernel function and the B-spline kernel are combined with the LSSVM respectively and accordingly MW-LSSVM and BS-LSSVM are proposed. Subsequently, optimizations for penalty parameters and kernel parameters are conducted using particle swarm optimization (PSO) and thus the PSO-MW-LSSVM and PSO-BS-LSSVM algorithms are proposed for spatial wind pressure prediction. The numerical analysis shows that the proposed method not only significantly outperforms the conventional RBF-LSSVM and BS-LSSVM in forecasting accuracy and generalization ability, but also has great engineering application prospects due to its stability.

Key words: wavelet kernel function, spline kernel function, least square support vector machine, wind pressure prediction, particle swarm