摘要
为了提升全球定位系统(Global Positioning System,GPS)信号中断场景下传统滤波方法对非线性误差抑制不足的缺陷,提出了一种多模态融合的注意力机制卷积长短期记忆网络(Convolutional Long Short-Term Memory Neural Network Based on Attention Mechanism under Multi-modal Fusion,MF-ACLSTM)导航补偿方法。该模型通过卷积神经网络(Convolutional Neural Network,CNN)进行空间特征提取,结合长短期记忆神经网络(Long Short-Term Memory Neural Network,LSTM)进行时序特征建模,以有效融合惯性导航与GPS数据,并引入注意力机制,动态调整不同传感器模态对导航估计的贡献权重,提高数据融合的鲁棒性。在GPS信号中断时,通过历史惯性导航数据与模型的自回归机制,实现短期精确预测和长期误差抑制,使导航结果更精准;利用获得的实验数据对所提方法的性能进行了验证。实验结果表明,在GPS信号中断的场景下,所提方法较传统滤波方法在定位误差和长期稳定性等方面均有显著提升。
In order to address the limitation of the traditional filtering method in suppressing nonlinear errors in the scenario of Global Positioning System(GPS)signal interruption,a Convolutional Long Short-Term Memory Neural Network Based on Attention Mechanism under Multi-modal Fusion(MF-ACLSTM)is proposed.In this model,spatial feature extraction is carried out by Convolutional Neural Network(CNN),and temporal feature modeling is carried out by combining Long Short-Term Memory Neural Network(LSTM)to effectively integrate inertial navigation and GPS data,and attention mechanism is introduced to dynamically adjust the contribution weights of different sensor modalities to navigation estimation,improving the robustness of data fusion.When the GPS signal is interrupted,the autoregressive mechanism of the model,in combination with the historical inertial navigation data,is used to achieve short-term accurate prediction and long-term error suppression,so as to make the navigation results more accurate.Finally,the experimental data obtained are used to verify the performance of the proposed method.Experimental results show that,in the scenario of GPS signal interruption,the proposed method significantly outperforms traditional filtering methods in terms of positioning error and long-term stability.
作者
汤翔
章飞
汪勋
胡莹
TANG Xiang;ZHANG Fei;WANG Xun;HU Ying(Ocean College,Jiangsu University of Science and Technology,Zhenjiang 212003,China)
出处
《无线电工程》
2025年第6期1274-1282,共9页
Radio Engineering
