A novel deep learning-based convolutional neural network – long short-term memory model for predicting weekly significant wave heights
Abstract
Significant wave height is an important part of wave monitoring, and its accurate prediction is of great significance for reducing marine disasters and utilizing wave energy. In this study, a hybrid model combining convolutional neural network (CNN) and long short-term memory (LSTM) with transformer encoder is proposed to predict the weekly significant wave heights. To improve the prediction accuracy, wavelet threshold denoising (WTD) technique was used to denoise the data. The performance of the hybrid CNN-LSTM-Encoder model was compared with four other artificial intelligence models. The results show that the mean absolute percentage error (MAPE), root mean squared error (RMSE), mean absolute error (MAE) and coefficient of determination (R2) values of the CNN-LSTM-Encoder model are 1.3 %, 0.029 m, 0.024 m and 0.9833, respectively, for the training datasets and 1.6 %, 0.037 m, 0.029 m and 0.964, respectively, for the test datasets, indicating that the proposed CNN-LSTM-Encoder model has high prediction accuracy for weekly significant wave heights. In addition, the sensitivity, uncertainty and generalization ability of the hybrid CNN-LSTM-Encoder model were also analyzed in this study.