Researches On The Theories And Algorithms Of The Error Analysis And Compensation For Integrated Navigation System

The combination of the inertial navigation system and the satellite systems is one of the most important approaches that can improve the navigation precision and the reliability. However, the errors of inertia components are remarkable factors in influencing INS or GPS/INS. This dissertation mainly focuses on the foundation of the error models, the compensation of the errors and the controlling of the errors by using the methods of Robust Spectral Analysis, wavelet analysis, adaptive filtering and so on. Finally, the applications in FOG north determining, the initial alignment of SINS and the integrated navigation are introduced. The main works and contributions are summarized as follows:1,In the FOG signals, there exists the low-frequency colored noise. First it is fitted and predicted for the initial signal. The robust spectral analysis is applied to differentiate the useful signal, colored noise and periodic noise from the initial signal. In simulating example, it is testified that this method can eliminate the periodic noise and degrade the influence of colored noise.2,For the practical correlated noise in the gyro, periodic function fitting and wavelet transform are applied to degrade the periodic noise respectively. Then the higher-order AR models are introduced for the correlated noise fitting. Finally the two AR models are applied in GPS/INS navigation. And the result based on the wavelet transform and the higher-order AR model shows a major improvement in the precision of navigation.3,Two-Position north determining data of FOG north seeker is analyzed and processed. Robust estimation is applied to calculate the coefficient of the trend part of the signal to reduce its influence. Average, robust estimation and wavelet transform are used and compared for the noise and the disturbance of the signal.4,In Kalman filtering of SINS refined initial alignment, when the inaccurate model is constructed or the systematic covariance matrix is not consistent with the actual noise, it will degrade the filtering accuracy or even lead to radiation. In order to solve this problem, a new method based on Elman neural network and Kalman filtering is presented in this paper. First, the reliable state estimation of Kalman filtering for the known system is taken to train the Elman neural network. Then the trained neural network is applied to estimate the state parameters for the unknown system. By the simulating data, it is determined that this new algorithm can get over the shortcomings of Kalman filtering in SINS refined initial alignment.5,A new method is presented to determine the initial attitude based on the wavelet transform and three-parameter sequential robust adjustment. First the wavelet multiresolution analysis is applied to de-noise the noise components from the measurements of gyros and accelerometers. Then the original attitude angles are calculated by the three-parameter sequential robust adjustment. By the simulating stationary data, it is determined that this new method can ensure high alignment accuracy in short time.6,A new algorithm based on wavelet threshold de-noising for GPS/INS is presented to improve the precision of integrated navigation. First, frequency-spectral analysis for output signals of inertia elements is given to decide the scale of wavelet multiresolution analysis and the measures for their high frequency coefficients. The high frequency coefficients of the scale which mostly represents high frequent noise will be set zero and those of the scale which represents low frequent noise and useful signals will be dealt with by using the threshold value. By the calculation, it is shown that the new algorithm can effectively resist the influence of the errors of inertia elements and improve the precision of navigation.7,An adaptive Kalman filtering is applied in GPS/DR integrated navigation to control the influences of outlying movement disturbances. The multi adaptive factors of the predicted states are given by the predicted residuals. By an actual calculation, it is shown that both the adaptive factor and the multi adaptive factors can resist the influence of the state disturbances, and the precision of the navigation is higher than the standard Kalman filtering.