摘要
针对微机电惯性测量单元(MEMS IMU)因非线性、时变误差引起的导航误差快速发散问题,提出一种基于深度学习校准MEMS IMU的航迹推算方法。通过构建深度可分离时间卷积-注意力机制神经网络(DSTCN-Attention),对MEMS IMU的时变非线性误差进行特征提取,并利用注意力机制筛选有效数据生成误差补偿量。将校准后的IMU数据输入不变扩展卡尔曼滤波系统进行航迹推算,公开数据集验证表明:所提方法较基于卷积神经网络的轨迹推算方法,水平轨迹绝对误差降低66%。
To address the issue of rapid navigation error divergence caused by nonlinear and time-varying error in micro-electromechanical inertial measurement units(MEMS IMUs),a deep learning-based calibration method for MEMS IMU dead reckoning is proposed.By constructing a depthwise separable temporal convolutional network with attention mechanism(DSTCN-Attention),the time-varying nonlinear error of the MEMS IMU is extracted,and the attention mechanism is utilized to filter effective data for generating error compensation values.The calibrated IMU data is then fed into an invariant extended Kalman filter system for dead reckoning.Validation on public datasets shows that the proposed method reduces the absolute error of the horizontal trajectory by 66%compared to trajectory estimation methods based on convolutional neural networks.
作者
乔美英
赵开东
韩昊天
邱运强
QIAO Meiying;ZHAO Kaidong;HAN Haotian;QIU Yunqiang(School of Electrical Engineering and Automation,Henan Polytechnic University,Jiaozuo 454003,China;Henan International Joint Laboratory of Direct Drive and Control of Intelligent Equipment,Jiaozuo 454003,China)
出处
《中国惯性技术学报》
北大核心
2025年第8期770-777,共8页
Journal of Chinese Inertial Technology
基金
国家自然科学基金(U1404510)
河南省自然科学基金资助项目(No.232300421152)
河南省科技攻关项目(222102220076)。
关键词
深度学习
惯性导航
校准
轨迹预测
deep learning
inertial navigation
calibration
trajectory prediction
