IRNet-BFGS: A Compensation Method for Vector Magnetometer Systems of Autonomous Vehicles With Dynamic Interference Suppression
Abstract
Accurate geomagnetic measurements are essential for autonomous vehicle positioning but are often degraded by magnetic interference from ferromagnetic structures and onboard electronics. To address this issue, this article proposes a geomagnetic interference compensation framework that combines dynamic interference suppression and fixed-source calibration. First, a vector interference model incorporating odometry information is developed to characterize dynamic disturbances. Based on this model, a motion-aware interference compensation network (IRNet) integrating model-driven and data-driven mechanisms, is designed to suppress motor-induced magnetic interference. For fixed interference compensation, a robust parameter estimation method combining Huber loss and the robust Broyden–Fletcher–Goldfarb–Shanno (RBFGS) is employed to improve calibration accuracy. Experimental results show that, compared to variational mode decomposition (VMD), the proposed algorithm reduces the three-axis and scalar mean absolute error (MAE) by 25.58%, 40.15%, 59.95%, and 55.92%, respectively. Positioning experiments further demonstrate that the compensated geomagnetic data significantly improve localization accuracy.