Taking GPS/SINS integrated navigation system as an application background in light of the problem that the conventional Kalman filter can easily diverge because of lack of prior knowledge and outliers, an improved adaptive Kalman filtering is proposed.The algorithm is based on the combination of Sage_Huse adaptive filter and fading memory filter which can suppress the filter divergence caused by lack of prior knowledge, and then a compression function which can effectively identify and deal with outliers is introduced,so the divergence problem caused by outliers can be solved.Simulation results indicate that the improved adaptive filtering algorithm can suppress the divergence caused by the inaccurate models and outliers,and simultaneously the filter accuracy of the horizontal positions is improved 6 times and 5.7 times, and the filter accuracy of the height position is improved 2.39 times compared to the traditional algorithms,at the same time it is better in adaptability and stability.