Research On Integrated Navigation System Based On Neural Network Optimization

The combination of Strapdown Inertial Navigation System and Global Positioning System can complement each other's advantages,enhance strengths and avoid weaknesses.The system can greatly improve continuous long-term navigation accuracy of the system.However,the following problems still exist in the SINS/GPS integrated navigation based on the position and speed:There are a large number of gross errors in GPS observations,and they will reduce the estimation accuracy of Kalman Filtering;Linear KF must accurately determine the mathematical model and noise statistical characteristics of SINS/GPS in advance,but in reality SINS model parameters might change.There are also lots of non-linear factors in the SINS/GPS system;GPS signals of the SINS/GPS loose integrated navigation system may be severely out of lock.Because of the lack of measurement information,Kalman filter will be invalid.SINS which works alone will inevitably lead to more navigation errors.In response to the above problems,this article is discussed from the following aspects:Firstly,the basic working principle of each subsystem,combined mode and mathematical modeling in the SINS/GPS integrated navigation system are introduced in detail,which provides a theoretical premise for the application of the following information fusion algorithms and neural networks.Secondly,in order to solve the problem of RBFNN parameter optimization,the RBFNN based on K-Means clustering analysis is firstly introduced.Then,a model of RBF neural network based on the improved Beetle Antennae Search(BAS)and AP clustering analysis is proposed,which improves the structure of the RBF network and increases the convergence speed and identification ability of network.Thirdly,for the gross interference of GPS signals,the non-linear Robust Unscented Kalman Filtering is further studied based on the linear Robust KF.The gross error test is designed for RUKF Based on Hypothesis Testing(RUKFBHT).Based on the above algorithm,in order to improve the adaptive ability of RUKFBHT to the uncertain variance of the system model noise variance matrix in integrated navigation,this paper proposes to uses an improved RBFNN to assist RUKFBHT to adjust the estimated state in real time.The simulation results show that two improved UKF algorithms ensure the optimality of the filtering state estimation and improve the filtering accuracy.Finally,in view of the situation where the accuracy of SINS working alone navigation decreases when the GPS signal is seriously losing lock in the SINS/GPS loose integrated navigation system,a navigation information update strategy based on theO?_{SIN S}-?PV_Gmodel and improved RBFNN prediction is proposed.The input variables of the neural network are the angular velocity,specific force and speed of the current moment and one-step increments of the above variables.The expected output is the GPS position increment and speed increment.When the GPS signal is normal,improved RBFNN trains and learns the input sample information.When the GPS signal is out of lock,the neural network can predict the real-time position increment and speed increase of GPS,and provide measurement information for the improved UKF.Thus,the program can achieve continuous and reliable navigation under GPS loss of lock.