Filter Algorithms For GNSS/INS Integrated Navigation System Assisted By Random Forest Regression

With the continuous advancement of technologies such as communication networks,artificial intelligence and satellite navigation,autonomous driving and intelligent transportation have become research hotspots in the field of navigation today,and they also put forward higher requirements for the accuracy and reliability of navigation positioning technology.The GNSS / INS integrated navigation system composed of global navigation satellites and inertial navigation systems can make up for the deficiencies of a single navigation system and improve the overall navigation accuracy and stability of the system.It is one of the important research directions in the field of navigation.This paper aims at the problem of rapid decline in system accuracy caused by the failure of GNSS signals in the environment of urban canyons,tunnels and underground parking lots.A GNSS/INS integrated navigation filtering algorithm assisted by random forest regression is studied,and the performance of the algorithm is analyzed and verified through vehicle experiments.The main research contents are as follows:(1)The basic principles and methods of GNSS / INS are summarized,including GNSS and INS data processing methods,integrated navigation mode classification and Kalman filter recursive equations and their discretization methods.The on-board experiment of integrated navigation was studied.The analysis and discussion of the processing results of the measured data verified the positioning accuracy and effectiveness of the integrated navigation system.(2)In view of the fact that the statistical characteristics of the integrated navigation system model noise are unknown,an improved Sage-Husa filter is proposed.A test factor is constructed by Mahalanobis distance of the prediction residual vector to detect the filter and judge whether adaptive filtering is needed,thus reducing the amount of filtering computation and facilitating the real-time estimation of filtering,abandon the negative terms in the calculation and measurement noise variance matrix,and obtain a high degree of stability without sacrificing a small filtering accuracy.The experimental results show that,when the design of the measurement noise variance matrix of the system is inaccurate,the improved Sage-Husa filtering latitude and longitude position errors are 0.17 m and 0.14 m,while the conventional Kalman filtering is 0.25 m and 0.21 m.It shows that the improved Sage-Husa filter can adjust the measurement noise variance matrix online in real time,making it conform to the noise characteristics of the current measurement value,thereby improving the convergence accuracy of the filter.(3)Aiming at the problem that the accuracy of the integrated navigation system drops rapidly when the GNSS signal is lost,this paper combines random forest regression and adaptive filtering algorithm to propose a random forest regression assisted integrated navigation filtering algorithm.The algorithm can construct the simulated GNSS displacement and speed increment through the trained random forest regression model when the GNSS signal is lost,and then use the improved Sage-Husa filter for information fusion.Experimental results show that the algorithm can assist the integrated navigation system when the GNSS signal is lost for a short time,suppress the divergence of positioning results caused by the combined navigation error due to the self-gliding of a single inertial navigation system,and effectively improve the positioning accuracy of the integrated navigation.