PREDICTION OF OCEAN WAVE HEIGHTS BASED ON A CONVOLUTIONAL-LSTM

Recent climate change has led to rising sea levels, increasing the importance of disaster prevention and prediction, particularly in coastal areas, for events such as wave overtopping, storm surges, and beach erosion. In the field of time series data prediction, recent advancements in artificial intelligence technologies have primarily focused on models like Recurrent Neural Networks (RNN) and Long Short-Term Memory (LSTM). However, these models have limitations in capturing spatial characteristics. In light of this, our study aims to establish a grid-based wave prediction model using a combination of Convolutional Neural Network (CNN) and LSTM, known as Convolutional LSTM (ConvLSTM), to account for spatial features in time series data prediction. Furthermore, our focus is on enhancing wave prediction performance with a simpler algorithm compared to previous research efforts. Additionally, we explored the potential for improving time series data prediction through the decomposition of temporal components within the input data. Overall, the application of time series decomposition elements in model training led to an improvement in accuracy of approximately 23percent across the entire study area.