Estimating sea surface wind speed from underwater noise using improved CNN-LSTM model

A method for estimating surface wind speed by combining the wind-induced noise characteristics with an improved convolutional neural network (CNN)–long short-term memory (LSTM) model is proposed. Firstly, the energy spectrum level of noise is calculated by data preprocessing to reflect the variations in actual noise intensity. Secondly, the energy correlation matrix is derived from the energy spectrum level to identify wind-induced noise features as input feature vectors. On this basis, a multi-feature-based estimation model is established to estimate wind speed by combining the feature extraction characteristics of convolutional neural network with the sequential information learning capabilities of long short-term memory networks. The experimental results in the South China Sea show that the root mean square error of the wind speed estimation of the proposed model is less than 0.3, the correlation coefficient with actual wind speed series is higher than 0.97, which is in good agreement with the real value, and the evaluation indexes are significantly better than the long short-term memory model.