Fault Prediction Method for SINS/GNSS Integrated Navigation System Based on SD-RBF Neural Network

In response to the problem of difficult prediction of slowly changing faults in Strapdown Intertial Navigation System / Global Navigation Satellite System(SINS/GNSS) integrated navigation system, this paper proposes a sliding window based double-layer Radial Basis Function(SD-RBF) fault prediction model when the system model and data distribution characteristics are difficult to obtain. This algorithm combines the explanatory model with the RBF based time series model. It uses system residuals to establish a time series prediction model and extracts fault trend information through residual analysis, achieving the prediction of slowly changing faults in SINS/GNSS integrated navigation. The simulation results show that the algorithm can predict slowly changing faults in 3 seconds, and the accuracy can reach 88%.