This paper addresses the challenges in assessing computer network security due to complex, chaotic, and nonlinear network environments, which traditional methods struggle to accurately analyze. It proposes a W-SVM framework that integrates wavelet transform and SVM to improve accuracy. The framework retains the original dynamic network traffic trajectory, decomposes the traffic into frequency sub-bands, and uses SVM for prediction. Applied to a university network, the model, with specific time delay and embedding dimensions, shows high consistency with actual situations, excellent prediction accuracy, and enhanced network security protection.