A Study on Error Compensation and Optimization of BDS/INS Integrated Navigation Based on Deep Learning

This study addresses the challenges posed by sensor manufacturing limitations and inherent technical constraints in inertial navigation systems (INS), particularly the effect of nonlinear random noise that leads to rapid error divergence. To improve the performance of BeiDou Navigation Satellite System/INS (BDS/INS) integrated navigation, we propose an adaptive error compensation method based on deep learning. This approach leverages the strengths of deep learning to model errors without strict assumptions on noise distributions, combining conventional filtering techniques with deep learning to avoid the pitfalls of end-to-end black-box models. A carefully designed deep neural network enhances generalization while reducing parametric complexity and computational cost. Experiments conducted using an in-vehicle data acquisition platform under simulated BDS signal outages show that the proposed method significantly outperforms traditional Extended Kalman Filter-based approaches, improving recognition accuracy by 77.70 % across various road segments. These results demonstrate the method's robustness and practical potential for high-precision navigation applications.