Considering the complexity and inaccuracy of the dynamic model for satellite and the fact that parameters to be estimated in orbit determination are correlated temporally with each other, an optical multi-model merging method for integrative satellite precise orbit determination based on GEOSTAR system is proposed. The B-spine function with knots-free is adopted to de- scribe the motion state of a satellite and thus to realize the reduction the error of the dynamic model. Combining the bi-satellite distance observation model, the precise trajectory determination process can be transformed into a nonlinear optimization problem of the multi model merging with the parameters of orbit spine expression coefficients and observation system error coefficients. By introducing the model structure to determinate the selection rule of optimal merging weight and adopting the nonlinear optimization method under LSE rule to search the optimal spine knots dis- tribution, the optimal estimation for the parameters to-be-estimated can be obtained, and the more precise determination of the LEO can be completed. The theoretical analysis and simulation show that this method is more effective, not only in increasing the ultimate precision of satellite state, but also in making the structure of state estimation more robust.