基于内波统计特性的声速剖面预测方法  

SSP prediction method based on the statistical characteristics of internal waves

作　　者：苏林[1,2] 孙炳文 胡涛[1,2] 任群言 王文博[1,2] 郭圣明 马力 SU Lin;SUN Bingwen;HU Tao;REN Qunyan;WANG Wenbo;GUO Shengming;MA Li(Key Laboratory of Underwater Acoustic Environment, Institute of Acoustics, Chinese Academy of Sciences, Beijing 100190, China;University of Chinese Academy of Sciences, Beijing 100049, China)

机构地区：[1]中国科学院声学研究所中国科学院水声环境特性重点实验室,北京100190 [2]中国科学院大学,北京100049

出　　处：《哈尔滨工程大学学报》2021年第6期859-865,共7页Journal of Harbin Engineering University

基　　金：国家自然科学基金项目(11704396).

摘　　要：针对浅海内波环境下的声速剖面预测问题,本文提出了一种基于内波环境下时序声速剖面统计特征的声速剖面智能预报方法。通过分析有无内波情形下时序声速剖面的经验正交分解系数分布特点,给出内波存在时的特有分布特征,以此为基础,将声速剖面的预报问题简化为单一参数的时间序列预测问题。结合长短时记忆循环神经网络处理时间序列数据中长距离依赖信息的能力,实现内波环境下声速剖面的实时预测,对试验数据预测结果与测量值符合较好,最大均方根误差1.3 m/s。To solve the sound speed profile(SSP)prediction problem in shallow-water internal wave environments,an intelligent prediction method is proposed based on the statistical characteristics of internal waves.The specific distribution characteristics when internal waves exist are obtained by analyzing the distribution characteristics of the empirical orthogonal function coefficients of time-series SSP with and without internal wave existence,respectively.Accordingly,the SSP prediction problem is simplified as a single-parameter time-series prediction problem.It is combined with the ability of long short-term memory recurrent neural networks to process long-distance dependency information in time-series data.The real-time SSP prediction is realized in an internal wave environment.The test results show that this method has good accuracy in SSP prediction,and the maximum root mean square error is 1.3 m/s.

关 键 词：内波 统计特性 经验正交函数 长短时记忆 声速剖面 时间序列预测 神经网络 循环神经网络 

分 类 号：P731.3[天文地球—海洋科学]