In order to consider the effect of historical information on future navigation results,and use deeper integrated navigation information for fusion,an INS/GNSS deep integrated navigation method based on graph optimization is proposed.By taking measurement information and state propagation as constraint information,the optimal cost function is constructed in time domain,and the optimal state estimation is obtained by using the Levenberg-Marquardt method.The method is evaluated and analyzed by INS/GPS deep integrated navigation system simulation experiment.The simulation results show that compared with the conventional Kalman filtering method,the proposed method reduces the mean position errors of the three axes by 38.5%,21.0%,and 30.9% respectively; and the mean velocity errors are reduced by 31.4%,52.8% and 57.3%.The proposed algorithm can effectively improve the positioning accuracy.