A conjugationg gradient algorithm based on Local learning strategy is presented to improve approximation quality and extensive capability of the WNN(Wavelet Neural Network), as well as to Improve the parameters optimization and process simulation in the machining of vibration drilling on variation parameters. And then the three input weight-|values are selected for the improved net model by taking advantage of the grey relevancy analysis. The selection principle for dilation and translation factors of wavelet function and network weight-|value is also given. The experiment shows that the proposal method improves the simulation effect and network performance.