A real-time monitoring system of composite impact loads was constructed by a Fiber Bragg Grating(FBG) sensor network, and the wavelet packet feature extraction and a Support Vector Regression(SVR) were used to identify the impact location. For the impact response signals at the same position measured by different FBG sensors, the wavelet packet energy spectrum analysis shows that some specifically frequency bands of sensor signals are sensitive to the impact. The relation between impact location and wavelet energy was studied and the sixth decomposition level wavelet packet energy was chosen as the characteristic vector of the impact location. The SVR whose tuning parameters have been optimized was used to established the sample regression model and predict the impact location. The result shows that network testing error of the SVR is 4.81％. The research provides a practical reference for the impact performance evaluation of the structures from carbon fiber reinforced plastics.