首页|CNN-LSTM车辆运动状态识别的AUKF组合导航方法

CNN-LSTM车辆运动状态识别的AUKF组合导航方法

AUKF integrated navigation method based on CNN-LSTM vehicle motion state recognition

扫码查看
原文链接
  • NETL
  • NSTL
  • 万方数据
针对固定的噪声协方差难以适应车辆不同运动行为下噪声统计特性差异大的问题,提出了一种基于卷积神经网络与长短期记忆网络(CNN-LSTM)的车辆运动状态识别自适应无迹卡尔曼滤波(AUKF)组合导航方法.首先,应用CNN-LSTM网络模型进行车辆运动状态识别,解决车辆自我运动不确定性的问题;其次,将特定运动状态约束下的噪声协方差应用于UKF的时间更新与量测更新;最后,将所提方法在采集的数据集上进行验证.实验结果表明,与经典的UKF算法相比,所提方法的位置均方根误差与速度均方根误差分别下降了 22.67%与 2.63%,验证了所提方法的有效性.
Aiming at the problem that the fixed noise covariance is difficult to adapt to the large difference of noise statistical characteristics under different movement behaviors of vehicles,an adaptive unscented Kalman filter(AUKF)integrated navigation method based on convolutional neural network and long short-term memory network(CNN-LSTM)for vehicle motion state recognition is proposed.Firstly,the CNN-LSTM network is employed to discern the vehicle motion state,addressing inherent uncertainties in self-motion.Secondly,noise covariance under specific motion state constraints is incorporated into both the time and measurement updates of the UKF.Finally,the proposed method is validated on the collected data set.Experimental results show that compared with the classical UKF algorithm,the root mean square error of position and velocity for the proposed method are reduced by 22.67%and 2.63%,respectively,which verifies its effectiveness.

integrated navigationvehicle motion state recognitioncombined neural networkadaptive unscented Kalman filternoise covariance

刘宁、谢越栋、胡彬、范军芳、苏中

展开 >

北京信息科技大学 高动态导航技术北京市重点实验室,北京 100192

北京信息科技大学 现代测控技术教育部重点实验室,北京 100192

组合导航 车辆运动状态识别 组合神经网络 自适应无迹卡尔曼滤波 噪声协方差

北京市自然科学基金面上项目国家重点研发计划课题北京市科技新星计划交叉学科合作课题现代测控技术教育部重点实验室开放课题资助

42440912020YFC1511702202111

2024

10.13695/j.cnki.12-1222/o3.2024.08.008

中国惯性技术学报
中国惯性技术学会

中国惯性技术学报

CSTPCD北大核心
影响因子：0.792
ISSN：1005-6734
年,卷(期)：2024.32(8)