测向误差特征辅助两步式网络的水声纯方位定位方法  

A two-step network assisted by direction-finding error underwater acoustic bearings-only localization method

作　　者：孙大军 傅笑盈[1,2,3] 滕婷婷 SUN Dajun;FU Xiaoying;TENG Tingting(Acoustic Science and Technology Laboratory,Harbin Engineering University Harbin 150001;Key Laboratory of Marine Information Acquisition and Security(Harbin Engineering University),Ministry of Industry and Information TechnologyHarbin 150001;College of Underwater Acoustic Engineering,Harbin Engineering University Harbin 150001)

机构地区：[1]哈尔滨工程大学水声技术重点实验室,哈尔滨150001 [2]工业和信息化部海洋信息获取与安全工信部重点实验室(哈尔滨工程大学),哈尔滨150001 [3]哈尔滨工程大学水声工程学院,哈尔滨150001

出　　处：《声学学报》2023年第2期291-302,共12页Acta Acustica

基　　金：国家自然科学基金项目(61871145)资助。

摘　　要：围绕水声分布式纯方位定位问题,针对传统方法的远距离定位精度低、定位结果易受初值影响等缺点,提出了一种测向误差特征辅助两步式全连接层神经网络(DFE-TS-FCNN)的纯方位定位方法。使用神经网络进行定位,提高远距离定位精度并消除初值影响,输入特征是目标方位角测量值和测向误差标准差估计值。使用两步式网络结构抑制网络过拟合,分类网络确定目标区域后,再用对应的定位网络估计目标位置。蒙特卡洛仿真实验中,所提方法在近距离达到了与迭代加权最小二乘算法和迭代总体最小二乘算法相近的定位精度,在远距离定位精度大幅提高、约束均方根误差(RMSE)小于2.5 km的条件下,最远可定向距离相比传统方法从12.6 km提升至22.7 km。在实际数据中,该方法也获得了较好的定位结果。This paper is concerned with the underwater acoustic distributed bearings-only passive localization.To overcome the problems of low localization accuracy over long distances and the localization results being easily affected by the initial values,a two-step fully connected neural network assisted by direction-finding error(DFE-TS-FCNN)bearings-only localization method is presented.The neural network is used to improve localization accuracy over long distances as well as eliminate the influence of initial values.Target direction measurements and standard deviation estimates of direction-finding error are used as input features.A two-step network structure is used to prevent overfitting of the networks.The target region is determined by the classification network,and subsequently estimated by the localization network corresponding to that specific region.In the Monte Carlo simulation experiment,similar localization accuracy is achieved compared to the iterative weighted least-squares algorithm and the iterative total least-squares algorithm under close distances,while simultaneously localization accuracy is improved over long distances.When the root mean square error(RMSE)is less than 2.5 km,the furthest directional distance increases from 12.6 km to 22.7 km compared with traditional algorithms.Excellent localization results have also been demonstrated in real data.

关 键 词：纯方位定位 全连接层神经网络 最小二乘算法 最大似然算法 均方根误差 

分 类 号：TB56[交通运输工程—水声工程]