Research On SINS/GPS Integrated Initial Alignment Technology Based On Nonlinear Filtering

Initial alignment is one of the key technologies of strapdown inertial navigation system(SINS),the alignment accuracy directly affects navigation positioning accuracy,and the time determines the startup speed of navigation system.In recent years,with the improvement of navigation accuracy,speed and environmental adaptability for vehicle,integrated initial alignment has been one of the important directions of initial alignment technology.Among them,SINS/GPS integrated initial alignment is the most widely used.In practical applications,limited by the measurement errors of inertial device,the system modeling errors,the viciousness of environment and so on,SINS/GPS integrated initial alignment system has a large azimuth misalignment angle after coarse alignment on inertial frame and has the characteristics of strong nonlinear,time variation and uncertainty.It doesn't apply to the condition of kalman filter.As a result,it needs to build nonlinear error models with a large azimuth misalignment angle and adopts nonlinear filters in the progress of precision alignment.The observability of system state variables can reflect the speed and accuracy of filtering estimation,,then affects the integrated performance.So it is necessary to analyze the observability before the design of filters.Based on error models of SINS/GPS integrated initial alignment system with a large azimuth misalignment angle,a singular value decomposition method based on piece-wise constant system is applied to quantitatively or qualitatively analyze the nonlinear system observability in different conditions of vehicle motion and observation,thereby improving the error model.Considering that traditional nonlinear filters require the covariance matrix to maintain positive definiteness and symmetry,this article mainly studys square-root transformation of nonlinear filters.By the performance comparison of SINS/GPS integrated initial alignment methods respectively based on square unscented kalman filter(SUKF)and square cubature kalman filter(SCKF),it is obvious that the method based on SCKF can get a higher accuracy of azimuth alignment,especially for higher dimensional systems.Therefore the key point is to improve and optimize SCKF.Nonlinear filters are strict with alignment system,which can lead to poor alignment accuracy and even divergence in some cases of unknown system noise statistics and model with errors.In allusion to the condition,improved nonlinear noise statistics estimator and improved strong tracking filter(STF)are proposed.The former is based on sage-husa adaptive filter,which can adjust noise statistical property matrixes in real time and improve the adaptive ability of the system;Based on STF,the latter introduces the multiple fading factors to adjust the filtering gain during the square-root updating of state prediction covariance,and improves system robustness in the case of model mismatch.Above all,three filters are designed,including adaptive square cubature kalman filter(ASCKF),strong tracking square cubature kalman filter(ST-SCKF)and improved strong tracking and adaptive square cubature kalman filter(ST-ASCKF).By analyzing and comparing the performances of SINS/GPS integrated initial alignment methods respectively based on these filters,it can get the simulation result,which shows the improved SCKF algorithm can obviously increase the accuracy of azimuth alignment with a large azimuth misalignment angle,under the condition of unknown system noise statistics and model with errors.