基于LSTM与扩展卡尔曼滤波的无人机机载电子磁干扰补偿研究

Research on UAV Onboard Electronic Magnetic Interference Compensation Based on LSTM and Extended Kalman Filter

摘要: 针对无人机磁探测任务中因机载电子系统、姿态机动等因素诱发的高维度非线性磁干扰问题，开展了基于深度学习与状态估计融合的改进磁补偿模型研究。提出了一种基于LSTM与扩展卡尔曼滤波(EKF)的耦合补偿模型。该模型发挥了LSTM网络对电机转速、相电流等异构特征的时序非线性演化规律的捕获能力，并结合EKF对非线性观测残差的实时后验修正，克服了单一神经网络在实测工况下易产生预测漂移与高频波动的问题。仿真结果表明，本文方法在应对复杂耦合干扰时，补偿精度较传统T-L模型及单一深度学习模型显著提升，改善比达到17.25。在无人机实机飞行试验中，补偿后的均方根误差降至1.7403 nT，改善比较次优模型提升了约39%。研究表明，该耦合算法能够有效抑制机载电子设备产生的动态磁畸变，显著提升磁测数据质量，为高精度磁探测提供了技术支撑。

Abstract: To address the high-dimensional non-linear magnetic interference in Unmanned Aerial Vehicle (UAV) magnetic detection tasks caused by onboard electronic systems and attitude maneuvers, this study investigates an improved magnetic compensation model based on the fusion of deep learning and state estimation. A coupled compensation model based on Long Short-Term Memory (LSTM) and Extended Kalman Filter (EKF) is proposed. This model leverages the capability of LSTM networks to capture the temporal non-linear evolution of heterogeneous features, such as motor speed and phase current, while integrating EKF for real-time posterior correction of non-linear observation residuals. This approach overcomes the inherent limitations of single neural networks, such as prediction drift and high-frequency fluctuations under practical operating conditions. Simulation results demonstrate that the proposed method significantly enhances compensation accuracy compared to the traditional Tolles-Bergen (T-L) model and single deep learning models when dealing with complex coupled interference, achieving an improvement ratio (IR) of 17.25. In actual UAV flight tests, the compensated Root Mean Square Error (RMSE) was reduced to 1.7403 nT, representing an improvement of approximately 39% over the sub-optimal model. The research indicates that the coupled algorithm effectively suppresses dynamic magnetic distortions produced by onboard electronic equipment (OBE) and significantly improves the quality of magnetic measurement data, providing robust technical support for high-precision magnetic detection.