海区声速剖面的现场反演方法

On-line retrieval methodology for sound speed profile of sea area

针对海区垂直声速剖面的预报问题,利用后向传播(BP)神经网络建立了反演模型,实现基于现场测量数据和历史数据的声速剖面实时预报。首先根据经验正交分析,由海区的历史平均数据提取出表征其主要变化特征的经验正交函数;构建三层BP网络,用前几阶经验正交函数及海区声速剖面历史样本对网络进行训练;再将海区表面温度现场遥感数据结合历史数据转化为声速样本,输入到BP网络进行海区垂直声速剖面的反演。经过实验数据分析,相比用平均声速剖面表示现场结果而言,反演剖面更加接近实际测量值,适用于海区垂直声速剖面的实时预报。

This paper constructed a retrieval model for ocean SSP （ Sound Speed Profile） with Back Propagation （ BP） neural network, which was used to on-line forecast SSP based on survey data and history data of the sea. Firstly, the main characteristic of SSP in the sea area was extracted from the mean value of history data set by the empirical orthogonal analyses. Then, a three-layer BP neural network was built and trained with empirical orthogonal functions and history SSP samples. The finished BP neural network was used to retrieve the vertical SSP with the satellite surface temperature and history data set. The simulation examples prove that the retrieval data is better than the mean value for showing the SSP of the sea. So the retrieval method can be used to real-time forecast the SSP.