[en] Support vector machine (SVM) is a learning machine based on statistical learning theory, which can get a model having good generalization. It can solve 'learning more' when dealing with small size. It can also avoid 'dimensional disaster' when solving nonlinear problems. This paper works on the parameters optimization for support vector regression machine (SVRM) and its applications. Solution path algorithm can save much CPU time when it is employed to optimize the regularization parameter of SVRM. Simulated annealing algorithm has good ability of finding global optimal solution. An improved solution path algorithm and simulated annealing algorithm are combined to optimize parameters of SVRM in the regression analysis of the acoustic and vibration data for complex practical problems. The numerical results show the model has good predictive capability. (authors)