Physics-Guided Attention Model for Complex Aeromagnetic Interference Compensation in Aircraft Cabins

Aeromagnetic compensation is crucial for suppressing carrier platform interference during magnetic measurements. Unlike traditional methods, in-cabin magnetic measurements are more susceptible to complex couplings between multisource nonlinear interferences such as irregular wiring and unknown interferences that are difficult to describe using physical models. To improve the accuracy of magnetic measurements in the cabin, a cabin magnetic noise suppression method based on a physics-guided attention network (PGAN) is proposed, which combines physics-based electromagnetic constraints with deep learning to achieve robust modeling and compensation of airborne magnetic interference. This method first uses the Tolles-Lawson (T-L) model to suppress the inherent magnetic interference of the platform. Subsequently, the Biot-Savart law was explicitly embedded into the neural network architecture, imposing physical constraints on the modeling of current-induced magnetic fields in airborne electrical equipment. At the same time, a fine-grained feature selection (FS) method in the frequency domain was proposed to replace traditional preprocessing-based FS methods (usually decoupled from neural networks) and achieve adaptive recognition and weighted discrimination of measurement signals related to magnetic interference. Finally, the cross-attention mechanism dynamically integrates magnetic field measurement data, aircraft attitude, and selected multisensor measurement signals to achieve adaptive magnetic compensation in the aircraft cabin. Experiments on real flight datasets demonstrated that the proposed method achieved a root mean square error (RMSE) of 10.75 nT, and the standard deviation (STD) of compensated measurement errors remains below 10 nT in multiple experiments. The corresponding expanded uncertainty is 7.18 nT, demonstrating high compensation accuracy and measurement reliability. Meanwhile, ablation studies have confirmed the necessity of each component.