An effective algorithm for offshore air temperature prediction with LSTM neural network and wavelet decomposition and reconstruction

Abstract

Offshore air temperature is an important parameter in marine scientific research. The change of offshore air as an indicator of the marine ecological environment is not only related to the growth of offshore organisms but also affects the development of the marine economy. Effective prediction of offshore air temperature is significant. The prediction model of offshore air temperature data is established by using wavelet decomposition and reconstruction algorithm combined with long short-term memory neural network (LSTM). We use the wavelet decomposition to decompose the offshore air temperature data of the ocean station into the overview signal and the detail signal, and the decomposed signal is reconstructed by a single branch to obtain the reconstructed signal. Then, input the reconstructed signals into LSTM model to predict the future offshore temperature. Finally, through the experiments, the proposed model is verified that has more advantages than the prediction effect of the LSTM offshore air temperature prediction model based on seasonal-trend decomposition procedure based on seasonal trend loss (STL) decomposition and the traditional LSTM prediction model. The proposed model has better prediction accuracy for offshore air temperature, and the prediction model can achieve effective prediction of the offshore air temperature.