海区声速剖面反演预测方法
投稿时间:2018-03-26  修订日期:2018-04-16  点此下载全文
引用本文:
摘要点击次数: 114
全文下载次数: 0
作者单位邮编
胡军* 湘潭大学 411105
中文摘要:针对海区声速剖面实时预测问题,采用遗传算法(GA)优化的径向基(RBF)函数神经网络建立反演预测模型(GA-RBF),实现基于海区表面实测温度数据以及历史数据的声速剖面反演预测。以我国南海某区域为对象,用2006—2015年的ARGO数据为样本对模型进行训练,用2016和2017年数据对模型进行检验。以均方根作误差(RMSE)为检验指标,用平均声速剖面表示实际声速剖面两年6月和12月的均方根误差平均值是2.3932m?s-1、2.1763 m?s-1;本文模型反演预测的均方根误差平均值是0.8454 m?s-1、0.8148 m?s-1,表明反演预测模型的 更接近实测声速剖面,可用于海区垂直声速剖面的实时预测。
中文关键词:声速剖面  反演预测  遗传算法  径向基函数神经网络  ARGO数据  均方根误差
 
Inversion prediction method for sound speed profile of sea area
Abstract:This paper constructed a inversion prediction model for ocean Sound Speed Profile (SSP) with Radial-Basis Function (RBF) neural network optimized by Genetic Algorithm (GA) (GA-RBF), which was used to on-line SSP prediction based on survey data and historical data of the sea. Training the model with the ARGO data from 2006 to 2015 years of the South China Sea region, and test the model with the data from 2016 to 2017 years. The average root-mean-square error (RMSE) of the mean value in June and December of two years are 2.3932 m?s-1、2.1763 m?s-1 , and the prediction model are 0.8454 m?s-1、0.8148 m?s-1. It is learnt from the RMSE result analysis that the prediction model is better than the mean value for showing SSP of the sea and it can be used to real-time forecast the SSP.
keywords:Sound Speed Profile (SSP)  inversion prediction  Genetic Algorithm (GA)  Radial Basis Function (RBF) neural network  ARGO data  Root-Mean-Square Error (RMSE)
查看全文   查看/发表评论   下载pdf阅读器