Study On Adaptive Kalman Filter Algorithm And Application In Precise Point Positioning

Precise point positioning(PPP)refers to the acquisition of phase-based observations by a single receiver,the use of satellite precise ephemeris and precision clock differential products,and the correction of related errors by models or parameters to enable them to reach the positioning level of dynamic decimeter level and static centimeter level.The application of precise point positioning is not limited to conventional positioning,but also can be used for precise orbit determination,earthquake monitoring and inversion of troposphere and ionosphere.The application prospect is very broad and has become a research hotspot in the field of navigation and positioning at home and abroad.However,through research,it is found that when Kalman filtering is used to estimate precise point positioning parameters,it is often affected by observation anomalies and dynamic model anomalies.When there is an anomaly(gross error)in an epoch observation value of PPP,the accuracy of PPP parameter estimation will be greatly reduced,and even it will not converge for a long time.The anomaly of the dynamic model is mainly reflected in the slow convergence speed caused by the inaccurate dynamic noise of the system.The reason is that theoretically the state parameter covariance will reach a stable value approaching zero as recursion progresses.If the state covariance is inaccurate,it will take a long time to stabilize,which increases the convergence time of PPP.In view of this,based on the theoretical research of PPP and Kalman filtering,this paper proposes a new method to solve the above problems,mainly including the following two parts:(1)Aiming at the problems of large deviation of parameter estimation and filter divergence caused by abnormal PPP observation,the author has conducted in-depth research on observation residual and innovation residual,and found that both residuals can well detect and diagnose the time when PPP observation is abnormal.Therefore,the above two kinds of residuals are used to establish the robust factor,and the original observation noise covariance matrix is replaced by the robust factor to achieve the purpose of robust.(2)When Kalman filter is used for PPP parameter estimation,the dynamic model abnormality of Kalman filter is mainly reflected in the inaccuracy of dynamic noise.If the dynamic noise is inaccurate,the formula is deduced according to Kalman filter,thus the calculated state covariance prediction value is also inaccurate,resulting in slight deviation between the filter gain matrix and the actual value.Since the gain matrix is involved in the estimation of the state and its covariance,the unreliability of the gain matrix will not only increase the convergence time of PPP,but also affect the accuracy of parameter estimation.Considering that the innovation residuals contain state information,so the innovation residuals can also reflect the abnormal state.The method in this paper is to construct the adaptive factor of adaptive Kalman filter according to the innovation residuals,and then modify the prediction value of state covariance through the classification factor matrix to make it relatively more reliable.Therefore,the reliability of the filter gain matrix will be correspondingly improved,thus achieving the purpose of improving the parameters and their covariance.Finally,the performance of the two robust Kalman filters under static/dynamic conditions and the reliability of the adaptive Kalman filter under static conditions are verified by experiments on multiple IGS tracking stations and groups of observation data.The experimental results show that both robust Kalman filters can suppress gross errors and improve the accuracy of parameter estimation to varying degrees.For the adaptive Kalman filter under static PPP,the predicted value of the state covariance and the time when the estimated value tends to be stable are lower than those of the standard Kalman filter,and finally the convergence time of PPP can be well reduced,and the accuracy of parameter estimation is also improved to some extent.Figure[33]Table[3]References[81]...