Due to the correlation error of high order and degree term of gravity field model spherical harmonic coefficient, the ground mass migration inversion results is presented by stripe error by using GRACE(Gravity Recovery And Climate Experiment)time-variable gravity field data. The fitting polynomial filtering with sliding window is one of the effective methods to deal with correlation errors. In order to improve the de-correlation effect of this algorithm, the optimal parameters of this algorithm are analyzed and selected based on the de-correlation results of GRACE time-variable model data. The experimental results show that the de-correlatin starting order, sliding window size and fitting polynomial order respectively should be set for 15, 5 7, 4.