Research On Key Technologies Of Fusion Navigation Algorithm During GNSS Outages

In the actual complex environment of the city,the GNSS signal is extremely vulnerable to the obstruction of buildings such as tall buildings,viaducts,trees,tunnels and so on.Under the condition that the observable environment is very poor,and even there is no GNSS signal,the traditional integrated navigation system degenerates into pure inertial navigation estimation.The inertial navigation system based on MEMS IMU due to the limitations of the manufacturing process and the scale effect of the material,the output of the MEMS gyroscope has a lot of noise,causing the navigation error to accumulate and diverge rapidly over time.In view of the practical problems encountered in participating in the key research and development plan of adaptive navigation software and hardware technology,this study explores the improvement in the urban complex GNSS environment from the two aspects of random error modeling of MEMS gyroscopes and neural network assisted integrated navigation system accuracy method.Firstly,for the random error modeling of MEMS gyroscopes,the application of ARMA(Auto Regression and Moving Average)model and wavelet threshold method in random error modeling of MEMS gyroscopes is studied.The experimental results show that the method of ARM combined with Kalman filtering can reduce the random error of the gyro and improve the accuracy of the gyro.Taking the Z axis as an example,the variance of the gyro static output after ARMA Kalman filtering is reduced by 24%,and errors such as angular random walk and zero bias instability are reduced by 16% and 38%,respectively.In the experimental results of the wavelet threshold method,compared with the case where the wavelet threshold is not applied,when the GNSS is disconnected for 60 s,the north position error is reduced from the maximum 5m to 2.4m,and the dynamic position error is reduced from 2.4m to 1.8m;disconnected In the case of 120 s,the north position error is reduced from a maximum of 31.19 m to 25.46 m,and the east position error is reduced from 11.56 m to 10.8m;when disconnected for 180 s,the north position error is reduced from a maximum of 25.7m to 22.75 m,east The position error was reduced from 10.31 m to 7.76 m.In the experimental results,compared to the case without wavelet denoising,when the GNSS is disconnected for 60 s,the north position error is reduced from the maximum 5m to 2.4m,and the dynamic position error is reduced from 2.4m to 1.8m;disconnected In the case of 120 s,the north position error is reduced from a maximum of 31.19 m to 25.46 m,and the east position error is reduced from 11.56 m to 10.8m;when disconnected for 180 s,the north position error is reduced from a maximum of 25.7m to 22.75 m,east The position error was reduced from 10.31 m to 7.76 m.The second is aimed at using neural network to assist integrated navigation after GNSS interruption.In order to improve the navigation accuracy during GNSS interruption,an algorithm based on Long Short-Term Memory(LSTM)neural network assisted inertial navigation system(SINS)is proposed.A method of generating pseudo-GNSS position increments by LSTM neural network to replace interrupted GNSS signal is proposed to suppress the rapid divergence of inertial navigation error after GNSS signal is interrupted.Experimental results are descripted as follows.When GNSS interrupts for 60 s,the LSTM algorithm reduces the maximum position error caused by pure SINS estimation by approximately 80%;when interrupting for 120 s,the LSTM algorithm reduces the maximum position error caused by pure SINS estimation by approximately 68.5%;during the 180 s GNSS interruption,compared with the pure inertial navigation algorithm,the LSTM algorithm can improve the navigation accuracy by 95%,and the MLP algorithm can improve the navigation accuracy by 50%.