Attention-GRU神经网络辅助的SINS/DVL组合导航算法  

SINS/DVL integrated navigation algorithm assisted by Attention-GRU neural network

作　　者：王立辉 刘恩东 吴璠 胡桥[2] 郝程鹏[3] 吴敏 WANG Lihui;LIU Endong;WU Fan;HU Qiao;HAO Chengpeng;WU Min(Key Laboratory of Micro-inertial Instrument and Advanced Navigation Technology,Ministry of Education,School of Instrument Science and Engineering,Southeast University,Nanjing 210096,China;School of Mechanical Engineering,Xi'an Jiaotong University,Xi'an 710049,China;Institute of Acoustics,Chinese Academy of Sciences,Beijing 100190,China)

机构地区：[1]东南大学仪器科学与工程学院微惯性仪表与先进导航技术教育部重点实验室,南京210096 [2]西安交通大学机械工程学院,西安710049 [3]中国科学院声学研究所,北京100190

出　　处：《中国惯性技术学报》2024年第6期565-571,共7页Journal of Chinese Inertial Technology

基　　金：国家自然科学基金(61773113,52371337,62371446);江苏省重点研发计划(BE2022389)。

摘　　要：针对多普勒测速仪(DVL)短时失效导致定位精度下降的问题,提出了一种注意力机制门控循环单元(Attention-GRU)辅助的捷联惯导(SINS)/DVL组合导航算法。当DVL测速有效时,使用SINS/DVL组合导航信息对Attention-GRU神经网络进行训练。当DVL故障时,使用训练完毕的Attention-GRU神经网络预测DVL速度辅助校正SINS。仿真结果表明:当DVL故障时,Attention-GRU相对纯惯性导航系统和GRU在匀速运动状态的平均速度误差分别减小了71.35%和3.48%,平均位置误差分别减小了34.76%和1.74%;在运动状态变化时平均速度误差分别减小了58.45%和14.67%,平均位置误差分别减小了9.82%和2.27%。An algorithm of strapdown inertial navigation system/Doppler velocity log(SINS/DVL)integrated navigation assisted by attention-gated recurrent unit(Attention-GRU)is proposed to address the problem of degraded positioning accuracy caused by temporary failures of DVL in special terrains.During effective DVL measurements,the Attention-GRU neural network is trained by using SINS/DVL integrated navigation information.In the event of DVL failure,the trained Attention-GRU neural network predicts the DVL velocity to assist in correcting the SINS results.Simulation results demonstrate that when DVL is faulty,the Attention-GRU method reduces the average velocity error by 71.35%and 3.48%,and the average position error by 34.76%and 1.74%,respectively,compared with pure inertial navigation and GRU in constant velocity motion.During motion state changes,the Attention-GRU method reduces the average velocity error by 58.45%and 14.67%,and the average position error by 9.82%and 2.27%,respectively,compared with pure inertial navigation and GRU.

关 键 词：组合导航 自适应卡尔曼滤波 DVL故障 门控循环单元 注意力机制 

分 类 号：U666.1[交通运输工程—船舶及航道工程]