| 摘要: |
| 舱内航磁干扰补偿是飞行器地磁导航面临的关键问题，由于舱内环境的干扰磁场更强且干扰源更加复杂，无法满足经典线性补偿模型的应用限定条件，导致模型补偿精度下降。近年来，由数据驱动的神经网络模型因其强大的学习能力成为一种潜在的补偿方法，但其在舱内场景的补偿精度和泛化能力受限。对于高空磁测序列数据而言，磁特征变化较为平缓，前后时刻特征之间存在一定的约束关系。因此，提出了一种结合注意力机制的神经网络补偿模型。该方法利用注意力机制挖掘不同时刻的输入特征对当前补偿结果的重要性，以提高舱内场景下模型的补偿精度。为了验证该方法的有效性，分别采用固定翼有人机实测数据和旋翼无人机实测数据开展补偿实验。实验结果表明，该方法的补偿精度高于经典磁补偿模型，在两种飞行平台上的补偿均方误差分别达到了0.40 nT和0.22 nT，与经典模型相比，补偿误差分别降低了97.9%和96.7%；与无注意力机制的神经网络模型相比，补偿误差分别降低了63.9%和81.1%。 |
| 关键词: 航磁干扰补偿 神经网络 注意力机制 地磁导航 舱内磁干扰 |
| DOI： |
|
| 基金项目:中国科学院青年创新促进会基金（E13314010D） |
|
| A neural network aeromagnetic compensation method combining attention mechanism |
| LIU Yuxin,LI Wen,WEI Dongyan,SHEN Ge |
| (Aerospace Information Research Institute, Chinese Academy of Sciences, Beijing 100094, China; School of Electronic, Electrical and Communication Engineering, University of Chinese Academy of Sciences, Beijing 100094, China) |
| Abstract: |
| Compensation for in-cabin aeromagnetic interference is one of the key issues in aircraft geomagnetic navigation. Due to the stronger magnetic interference and more complex interference sources, the in-cabin environment cannot meet the specific application conditions of classical linear compensation models, resulting in a decrease in compensation accuracy. In recent years, data-driven neural network models have emerged as a potential compensation method due to their powerful learning ability. However, the compensation accuracy and generalization ability of this method are limited in cabin scenes. For high-altitude magnetic survey sequence data, the changes in magnetic characteristics are relatively gentle, and there is a certain constraint relationship between the characteristics of the previous and subsequent moments. Therefore, a neural network compensation model that combines the attention mechanism is proposed. In this method, the attention mechanism is used to mine the importance of the input features at different moments to the current compensation result, in order to improve the compensation accuracy of the model inside the cabin. To verify the effectiveness of this method, compensation experiments are conducted using measured data from fixed-wing manned aerial vehicles and rotary-wing unmanned aerial vehicles. The experimental results show that the compensation accuracy of this method is higher than that of the classical magnetic compensation model, and the compensation mean square errors on two flight platforms reach 0.40 nT and 0.22 nT, respectively. Compared to the classical model, the compensation errors are reduced by 97.9% and 96.7% respectively. Compared to the neural network model without attention mechanism, the compensation errors are reduced by 63.9% and 81.1% respectively. |
| Key words: Aaeromagnetic interference compensation Neural network Attention mechanism Geomagnetic navigation In-cabin magnetic interference |
