| 摘要: |
| 海洋声速显著影响水下声传播特性,快速获取声速剖面对水声定位、目标探测和水下通信意义重大。单变量经验正交函数法(single Empirical Orthogonal Function regression,sEOF-r)通过建立声速剖面的经验正交系数与海表环境信息之间的线性回归关系反演声速剖面。上述方法主要存在以下不足：一是声速与海表信息并非简单的线性关系,二是逐格点反演对于大区域三维声速场重构而言计算效率低下。为此,本文基于HYCOM再分析资料,采用三维经验正交函数法(three dimensional Empirical Orthogonal Function, 3dEOF)和随机森林(Random Forest, RF)算法,建立声速与海表面温度、海平面高度等海表信息之间的非线性映射,提出一种区域逐深度层的全海深声速剖面快速反演方法(3dEOF-RF)。声速剖面反演结果表明：3dEOF-RF的反演精度和反演效率均有效提高,且反演结果具有较好的空间连续性,声速剖面的平均反演精度比经典sEOF-r法提高1.5 m/s,反演效率提高70.58%。 |
| 关键词: 声速剖面 EOF分解 随机森林 快速反演 |
| DOI： |
| 分类号: |
| 基金项目: |
|
| Fast reconstruction of Sound Speed Profile based on machine learning and empirical orthogonal function method |
|
lihongchen1, liming1, chenxi1, maokefeng1, wangpenghao1, liuyuhang2
|
|
1.National University of Defense Technology;2.Jiangsu Ocean University
|
| Abstract: |
| The ocean sound speed significantly impacts the underwater acoustic propagation characteristics, and rapid acquisition of sound speed profiles is crucial for underwater acoustic positioning, target detection, and underwater communication. The single Empirical Orthogonal Function regression (sEOF-r) method constructs a linear regression relationship between empirical orthogonal coefficients of sound speed profiles and surface environmental information to inversely derive sound speed profiles. However, this approach presents two main limitations: first, the relationship between sound speed and surface information is not straightforwardly linear; second, point-by-point inversion for reconstructing large-scale three-dimensional sound speed fields is computationally inefficient. To address these issues, this paper utilizes HYCOM reanalysis data and adopts a three-dimensional Empirical Orthogonal Function (3dEOF) approach in conjunction with a Random Forest (RF) algorithm to establish a non-linear mapping between sound speed and sea surface temperature, sea surface height, and other surface information. This novel methodology proposes a regional, depth-layered, full-water-column sound speed profile rapid inversion method, termed 3dEOF-RF. The results from the sound speed profile inversions demonstrate that both the inversion accuracy and efficiency have been effectively improved by the 3dEOF-RF method. Moreover, the inverted sound speed profiles exhibit good spatial continuity. On average, the 3dEOF-RF method achieves a 1.5 m/s improvement in sound speed profile inversion accuracy compared to the classical sEOF-r method, while enhancing inversion efficiency by 70.58%. |
| Key words: Sound velocity profile EOF decomposition Random forest Fast inversion |