Application of LSTM Neural Network in RISS/GNSS Integrated Vehicle Navigation System

The integrated navigation solution of GNSS and INS has been widely utilized in the field of land vehicle navigation. Compared with the SINS solution, the Reduced Inertial Sensor System (RISS) decreases the number of inertial sensors. Meanwhile, it cuts the cost of the system and simplifies the calculation of the mechanization solution. For this reason, the RISS/GNSS integrated navigation system has been proved to be an excellent solution for the low-cost vehicle navigation system. However, since the drifts of inertial sensors accumulate over time, the position results are unstable during GNSS outages, which leads to a sharp decline in the navigation accuracy of the MEMS-based RISS/GNSS vehicle system. A novel vehicle navigation solution assisted by the neural network is proposed to solve the above problems. In this paper, the LSTM neural network is adopted to complement the RISS/GNSS integrated navigation system. The navigation data can be predicted by establishing the relationship between the observations of inertial sensors, the odometer, and the attitudes calculated by the integrated solution during GNSS outages. The accuracy of the navigation system is improved with the aid of the LSTM neural network. The dataset collected in the urban area of Toronto, Canada, was tested in this paper. The results show that the RISS/GNSS integrated navigation method based on the LSTM neural network has significantly improved the position performances compared to the traditional integrated navigation algorithm. The algorithm can not only adapt to the complex road conditions but also significantly enhance the environmental adaptability of the vehicle navigation system. This method has a good application prospect in the field of urban road driving.

Automated summary

This paper presents the problems of the vehicle navigation method based on RISS/GNSS integrated navigation system and proposes a solution to improve navigation accuracy using neural networks. The results of the experiments show that this method significantly improves the position performances and enhances the environmental adaptability of the vehicle navigation system.