联合匹配场和神经网络的声速时间场构建方法
A method for constructing a sound velocity time field by combining a matched field and neural network
李林洋 1徐天河 2王君婷 2黄威 3高凡 2舒建旭2
作者信息
- 1. 长安大学地质工程与测绘学院,陕西西安 710054
- 2. 山东大学空间科学与物理学院,山东威海 264209
- 3. 中国海洋大学信息科学与工程学部,山东青岛 266100
- 折叠
摘要
局域声速场变化严重影响了海洋声学定位与导航的精度,构建高精度、高分辨率的声速时间场尤为重要.针对局域声速场参考样本不足导致声速场构建精度较低的问题,本文依据匹配场和神经网络理论,建立了联合匹配场和神经网络的声速时间场构建方法.利用通信数据和全水深声速剖面数据,采用匹配场处理进行声速剖面仿真,其次通过后向传播神经网络拟合经验正交函数分解重构系数构建声速剖面,并与实测抛弃式温盐深剖面仪数据进行对比.结果表明:与平均声速剖面的1.079 m/s相比,考虑匹配场仿真声速剖面的联合匹配场和神经网络的声速时间场构建方法的均方根误差为0.665 m/s,提高了 38.4％.因此,该算法可以提高复杂海洋环境中声速时间场的构造精度.
Abstract
The variation in a local sound velocity field severely affects ocean acoustic positioning and navigation ac-curacy,and the construction of a high-precision and high-resolution sound velocity time field is particularly impor-tant.Aiming at the problem of the low accuracy of sound velocity field construction due to insufficient reference samples of the local sound velocity field,this paper establishes a construction method for a sound velocity time field based on the theory of a matched field and neural network.Matched field processing is adopted to simulate the sound velocity profile,using communication data and full water depth sound velocity profile data.Then,the sound velocity profile is constructed by a BP neural network fitting the EOF reconstruction coefficients and compared with the measured disposable temperature and salt depth profiler data.The results show that the root mean square error of MFPBP for simulating sound velocity profiles considering matched fields is 0.665 m/s,38.4％higher than 1.079 m/s for average sound velocity profiles.Therefore,the algorithm can improve the accuracy of constructing a sound velocity time field in complex marine environments.
关键词
声速时间场构建/匹配场处理/后向传播神经网络/声速剖面反演/经验正交函数分解/启发式算法/射线声学理论/第一模态系数Key words
construction of sound velocity time field/matching field processing/BP neural network/inversion of sound velocity profile/empirical orthogonal function decomposition/heuristic algorithm/theory of ray acoustics/first modal coefficient引用本文复制引用
基金项目
崂山实验室科技创新项目(LSKJ202205104)
山东省自然科学基金(ZR2023QD163)
中国博士后科学基金(2023M732041)
地理信息工程国家重点实验室基金(SKLGIE2022-ZZ2-02)
出版年
2023
国家科技期刊平台
NSTL
维普
万方数据