基于SVM的航位推算误差补偿  

Error Compensation for Dead Reckoning Based on SVM

作　　者：李鑫[1] 王晓鸣[1,2,3] 武建国 赵基伟 忻加成 陈凯 张彬 LI Xin;WANG Xiaoming;WU Jianguo;ZHAO Jiwei;XIN Jiacheng;CHEN Kai;ZHANG Bin(School of Mechanical Engineering,Tianjin University of Science and Technology,Tianjin 300202,China;Tianjin HanhaiLanfan Marine Technology Co.,Ltd,Tianjin 300300,China;Tianjin Key Laboratory of Deep-sea Intelligent Mobile SurveyEquipment,Tianjin 300300,China)

机构地区：[1]天津科技大学机械工程学院,天津300202 [2]天津瀚海蓝帆海洋科技有限公司,天津300300 [3]天津市深远海智能移动勘测装备研发重点实验室,天津300300

出　　处：《水下无人系统学报》2024年第6期1009-1017,共9页Journal of Unmanned Undersea Systems

基　　金：国家重点研发计划项目(2020YFC1521704)。

摘　　要：在使用机器学习方法对自主水下航行器(AUV)航位推算进行误差补偿时,通常采用神经网络算法。但神经网络需要大量的训练样本才能达到稳定的训练结果。为了解决此问题,文中对支持向量机(SVM)在航位推算的误差补偿问题进行研究。利用SVM训练出误差补偿模型,对航位推算进行误差补偿,提高了导航精度。误差补偿模型选取AUV的俯仰角、横滚角和航向角,多普勒计程仪(DVL)对地的前向、右向和天向速度以及航位推算时间等7个参数作为输入参数,以全球卫星定位系统(GPS)和惯导+DVL组合提供的经纬度与航位推算的经纬度差作为模型的输出,训练出误差补偿模型。对比神经网络算法,在数据量较少的前提下,SVM训练模型和神经网络训练模型的相对误差分别为0.28%和0.93%。进而通过湖上试验得出,SVM训练模型能够将航位推算的相对误差控制在0.5%内。In the use of machine learning methods for error compensation in dead reckoning of an autonomous undersea vehicle(AUV),the neural network algorithm is commonly used.However,neural networks require a large number of training samples to achieve stable training results.To solve this problem,research was conducted on the application of support vector machine(SVM)for error compensation in dead reckoning.By utilizing SVM,an error compensation model was trained to correct the errors in dead reckoning,thereby improving navigational accuracy.The error compensation model takes seven parameters as input:pitch angle,roll angle,course angle,forward,right,and upward velocity of the Doppler velocity log(DVL)relative to the ground,and dead reckoning time of the AUV.The difference in latitude and longitude provided by the global positioning system(GPS)and inertial navigation system(INS)+DVL combination compared with latitude and longitude obtained from dead reckoning serves as the output of the model.The SVM trained model and the neural network trained model show a relative error of 0.28%and 0.93%,respectively,when the amount of data is limited.Through lake tests,it is concluded that the model trained by SVM can control the relative error of dead reckoning within 0.5%.

关 键 词：自主水下航行器 航位推算 支持向量机 误差补偿 

分 类 号：TP630.33[自动化与计算机技术—控制理论与控制工程] U666.1[自动化与计算机技术—控制科学与工程]