Wave height prediction based on CNN-LSTM

The short-term wave height forecast is of great significance to the development and utilization of energy. To improve the accuracy of short-term wave height prediction, we propose a prediction model based on convolutional neural network (CNN) and long short term memory (LSTM) network as they have excellent feature extraction ability and are very good at processing time series data. This model leverages CNN to perform convolution and pooling calculation on the maximum wave height (Hmax), the zero up crossing wave period (Tz), the peak energy wave period (Tp), direction (related to true north) from which the peak period waves are coming from (Dir_Tp TRUE), approximation of sea surface temperature (SST) data to extract the feature map of wave height related data. To describe the timing dependence of wave height sequences, spectral feature information is used as the input information of the LSTM network to calculate the wave height prediction results. We use the actual measured data from Australia to verify the accuracy of the model, and the experimental results show that it has better prediction performance than LSTM, Support Vector Machines (SVM), Random Forest (RF) and other machine learning models.