Large-Scale Geomagnetic Navigation for High-Speed Aircraft Based on the Gradient Extraction Neural Network

Geomagnetic navigation technology is considered a potential alternative to global navigation satellite systems (GNSSs) and has been extensively studied in recent years. At the cruising altitude of high-speed aircraft (HA), it is challenging to utilize the main geomagnetic field for large-scale geomagnetic navigation due to the smooth variation of the main geomagnetic field, which can lead to inaccuracy in geomagnetic matching algorithms. To overcome this challenge, this article proposes a gradient extraction neural network model. First, the matching reference map is incorporated as part of the model’s input to enable the model to achieve position prediction over large areas. Then, by using the gradient extraction layer to construct a new feature vector and introducing the magnetic deviation vector, the model can quickly converge to the target position even with large initial position errors. Finally, the feasibility and accuracy of the gradient extraction layer and the GENN model are demonstrated through ablation studies and flight simulations. The flight simulation results indicate that under low noise conditions, the mean absolute error (MAE) of the integrated navigation system along the horizontal direction can be reduced to 53 m.