Research On Application Of Gyro Error Modeling Based On The Wavelet And Diagonal Neural Network

In the engineering field,no matter what form of objects,information perception is an important issue. Information perception can be perceive themselves, perceive each other or perceive the environment. Navigation is a kind of information perception. No matter what carrier it is, we need to have the navigation information to assist motion control to reach the destination, so the accuracy of navigation is very important. In order to enhance the accuracy and reliability of the navigation system, the integrated navigation system has been proposed.By combining different types of sub-navigation systems, the integrated navigation system can overcome the limitations of the subsystem and enhance its accuracy.In the integrated navigation system, inertial navigation system is in the leading position,while the INS precision is decided by the inertial element. At the same time, the data fusion of integrated navigation needs each subsystem to build accurate mathematical model, or else there will be no advantage in integrated navigation. Therefore, the accurate mathematical model of inertial element error has the important meaning for error compensation and making up the integrated navigation. For this, this dissertation combine wavelet threshold denoising and diagonal neural network to model the inertial component random error from the combination modeling point of view. This model is used in a integrated navigation which is based on the pseudo range and the pseudo range rate to prove its validity.First of all, some models about the INS/GNSS integrated navigation are introduced and deduced, including inertial navigation error equations, principles about GNSS and integrated navigation system etc. And in the model derivation, the processing parameters are tried not to simplify to retain more complete information.Secondly, the Allan variance and Power Spectral Density are used to analyze the measured data of fiber optic gyro. The result show that the traditional hypothesis model with random number, correlated noise and white noise could not be used to accurately describe the random error. In addition, using the ARMA model is complex in operation and little expression. According to the analysis, the wavelet threshold denoising is used to gyro. When the wavelet is used, the dissertation provides a method to get the reference value of decomposition level and the various threshold rule effects were compared. Simulation results show that after denoising, gyro-frequency noise is effectively filtered out.Thirdly, after denoising, the characteristics of residual gyro random errors are low frequency and correlation. According to that, diagonal recurrent neural network (DRNN) is used for time series modeling. In order to speed the convergence of network, LM algorithm is used to improve the learning algorithm of neural network; the advantages of DRNN are talked about, and DRNN is demonstrated more convenient in time series modeling. After model checking, DRNN model residuals are zero mean white noise, which will also bring the convenience for integrated navigation filter. Simulation results show that DRNN model have very good effect.Finally, according to the method and conclusion in this dissertation, an integrated navigation system which is assisted by wavelet threshold denoising and DRNN modeling is presented. After that, sport car measured data is used to test and verify with the model.Simulation results show that the integrated navigation system which is based on pseudo range and pseudo range rate have better performance with this model's assist.