Based on the structure of the time response formula in the multi-variable model and the accumulated sequence with quasi-exponential law, this paper researchs the problem of background value optimization of the model. After analyzing background value error of traditional MGM(1,m??) model, the paper uses the functions with non-homogeneous exponential law to fit the accumulated sequences for every variable and gets the optimal formula of background value of MGM(1,m??) model through a series of mathematical derivation. And the optimization effect is verified by examples. The result shows that the proposed method can significantly improve the simulation and prediction accuracy of the traditional MGM(1,m??) model.