To solve the problem of spectral peak deviation for auto-regressive moving average (ARMA) model parametric spectrum estimation, an ARMA parameter estimation method combined by objective function and gene algorithm (GA) is proposed. An initial estimate of the model parameters is acquired by minimizing Least Square Error. The constraint equation is derived according to modern spectrum theory and the necessary condition for the extreme value of continuous function. The combined objective function is constructed and the GA is used to optimize the initial parameter estimation. The method is used for modeling the vertical vibration acceleration data of back centre under turning condition and its spectrum estimation. Results demonstrate its efficiency.