The hydraulic vibration system of two freedom hydraulic vibration roller was the research object in this paper. Three horizontal subsets were selected in every experimental factor which was ob- tained from the three一dimensional orthogonal experiment to obtain the samples of the training neural network. The mathematic model was built by artificial neural network theory and the optimum tom- bined parameters of horizontal subsets were established by means of Mtalab simulation. The results show that target value of the optimum combined is 0. 552 3 and the stiffness of the system is 3. 3 N/ mm through the artificial neural network mathematics model. This meets the project requirement be- cause of the relative error of 10.46% between experimental results and experimental target value.