Spectral Level Prediction Model of Ocean Ambient Noise Based on GA-LM-BP Neural Network

Abstract

Efficient and accurate prediction of ocean ambient noise spectrum level is very important to improve the detection capability of sonar equipment. The more factors taken into account, the more complex the establishment and application of the prediction model, which makes a low prediction efficiency. As a data-driven technology, neural networks have the ability to accurately predict the state of complex systems and can avoid complex physical modeling. In this study, a neural network model is built to predict ocean ambient noise spectrum level based on the data of sea water depth, temperature, salinity, sea surface wind speed and rainfall. The model is based on Genetic Algorithm (GA), Levenberg–Marquardt algorithm (LM) and Back Propagation (BP) neural network. The use of GA and LM makes the model combine the powerful mapping ability of neural network and the global search characteristic of GA. The model is used to predict the variation characteristics of spectral levels with frequency, depth, wind speed and rainfall rate, respectively. The predicted values are compared with the real values, for example, the RMSE values are all nearly below 2.04. The results show that the GA-LM-BP neural network prediction model is accurate and effective, and has flexible input factor scalability, which provides a paradigm framework for the establishment of multi-source and multi-factor spectral level prediction model of ocean ambient noise spectrum level based on deep learning.