GNSS/INS Integrated Navigation System Based On Time Series Neural Network And Attention Mechanism

In recent years,the integrated navigation system combined by inertial navigation system(INS)and global satellite navigation system(GNSS)is widely used to enhance the position,speed and attitude information of driverless vehicles.However,GNSS signal is easily affected by the complex external environment,and the signal will be interrupted in urban valleys,tunnels and other places with serious shielding,so it is unable to complete the positioning service.Therefore,it is of great significance to study how GNSS / INS integrated navigation system can provide reliable navigation service when GNSS signal is lost in special environment.The methods to solve the problem of GNSS signal loss in GNSS / INS integrated navigation system are mainly divided into two categories.One is to establish the relationship between INS navigation information and INS navigation information error by using machine learning or deep learning method,and correct INS navigation information through the navigation information error output by machine learning or deep learning method when GNSS signal is lost.Another method is to establish the relationship between INS navigation information and GNSS signal by using machine learning or deep learning method.When GNSS signal is lost,the pseudo GNSS information output by machine learning or deep learning method is taken as GNSS signal and input into Kalman filter algorithm together with INS information to correct INS navigation information.At present,these two methods have some defects when using machine learning deep learning method to solve the problem of GNSS signal loss.They do not make full use of historical navigation information,do not consider the impact of different navigation information components on navigation results,and do not consider the noise generated by inertial sensors in INS.Aiming at the above problems,this paper makes a systematic research on GNSS /INS integrated navigation.This paper presents a GNSS / INS integrated navigation system assisted by neural network and signal denoising.Firstly,in order to provide position information for the navigation system during GNSS signal interruption,the timing neural network(TCN)uses the specific force,angular velocity,velocity and attitude angle of the current vehicle to predict the pseudo GNSS position.Secondly,because the different information components of the vehicle make different contributions to the accuracy of predicting the pseudo GNSS position by the sequential neural network,the attention mechanism is used to assign different weights to each information component of the vehicle,so as to improve the accuracy of predicting the pseudo GNSS position by the sequential neural network.Finally,because the original data measured by inertial sensor has large noise,the improved empirical mode threshold filter is used to reduce the noise of the original data measured by inertial sensor.Using the idea of module differentiation,this paper first verifies the effectiveness of the denoising algorithm proposed in this paper,and then uses the IMU signal denoised by the denoising algorithm to verify the performance of the neural network assisted GNSS / INS integrated navigation system proposed in this paper.In order to evaluate the effectiveness of this method,this paper uses the field test data of teacher Yan Gongmin to compare this method,the fusion algorithm based on LSTM and the fusion algorithm based on MLP.The results show that this method can significantly improve the navigation accuracy when GNSS signal is interrupted.