The locomotive loads distribution is dramatically influenced by the shims added to those springs． Considering the non-negligible nonlinear characteristics among locomotive bodies and secondary springs，a hybrid model is proposed by utilizing multi-output support vector regression as a compensation operator for the mechanism model． To improve the accuracy of compensation and obtain a set of optimal parameters，the genetic algorithm is used to globally optimize the parameters of the compensation operator． At the same time，the uniform design method is adopted to arrange the sampling scheme to reduce the sampling time． The proposed hybrid model is evaluated with the testing data set of a HXD1D high-power locomotive． The proposed model outperforms the mechanism model and the BP network hybrid model with respect to the root mean square error ( ＲMSE) ． An experimental result shows that the ＲMSE is reduced by 33. 53% in comparison with the mechanism model and 23. 69% less than that of the BP network model，which can further improve the accuracy of locomotive loads adjustment model． At the same time，the computation time of the proposed model is much less than that of the BP network hybrid model．