In order to solve the problems of Kalman filter dispersion and inability to accurately track the error changes in the vehicle combination navigation system， an improved adaptive filtering algorithm is designed. By analyzing the judging condition of filter divergence in principle， combining Sage-Husa filtering with multiple fading filtering algorithm， the fading factor is introduced in the filtering process. The filtering state and updating strategy are switched by the divergence condition， so that the state of filter returns to normal in time when divergence occurs. The improved algorithm is subjected to Jetlink inertial guidance/odometer combination navigation simulation tests and track of vehicle experiment. Simulation and experimental results are presented that the divergence of filter can be suppressed， and in the meanwhile， the adaptability and stability of the filter are enhanced to a certain extent.