Sound Velocity Profile Prediction Method Based on RBF Neural Network

Abstract

As a marine environmental parameter, sound velocity has an important impact on sound propagation in the ocean. In the same sea area, the sound velocity profile (SVP) changes dynamically due to the influence of marine environment, season change and other factors. To accurately obtain the SVP of seawater in time is of great significance to improve the positioning accuracy of underwater acoustic equipment for marine research and development. As the main data source of physical oceanography research, Argo data has abundant ocean hydrological observations, which provides scientific reference basis for studying ocean temperature, salt, pressure structure and spatio-temporal variation of hydrological elements. Aiming at the problem that the SVP can’t be accurately obtained in time, this paper proposes a method of SVP inversion and prediction based on radial basis function (RBF) neural network. The method is based on the nonlinear function approximation capability of neural network, by using the measured temperature, salinity of the sea area and Argo data to build the sound velocity profile prediction model. The proposed SVP prediction method was verified with the Argo data of the Atlantic Ocean from 2004 to 2018. The results show that the prediction profiles based on neural network is closer to the actual SVPs that those of the average sound velocity method. Compared with error back propagation (BP) neural network, RBF neural network has the same accuracy and higher efficiency. Therefore, the SVP prediction method based on RBF neural network is more suitable for real-time or near real-time prediction of marine SVP.