Aiming at the time-consuming problem of large-scale structure optimization, a method (SVM-GA) which combines Support Vector Machine (SVM) and Genetic Algorithm (GA) is proposed. In this method, approximation model is constructed based on SVM to substitute the time-consuming FEA. GA is applied to optimization. The multi-objective structural dynamics optimization of a satellite is adopted as an example to illustrate the effectiveness of the SVM-GA method. Approximation model based on SVM is constructed to simulate the nonlinear mapping relation between structure design parameters of honeycomb plate and carbon fiber frame and frequency response of acceleration and stress. Second order response surface is constructed as a contrast. Results show that the SVM-GA method can reduce the time consumption of optimization and have a good precision.