Time Series Wave Forecasting with GRU with Attention Approach, Case Study In Jakarta Bay, Indonesia

Prediction of sea waves is crucial since various sectors rely heavily on this information, such as the business sector, daily fishing activities, transportation, and energy generation. Many studies have made predictions about ocean waves using statistical methods, conventional machine learning, and advanced neural networks. However, because there are certain limitations, such as statistical methods, which have limitations in their ability to capture non-stationary and non-linear data; an advanced model is needed to model accurately the nonlinear behaviour of wave data. This study uses a deep learning model, i.e. to make predictions of significant wave height using the GRU with Attention mechanism. The Attention mechanism is added, to increase prediction accuracy, since it can remember which data points significantly influence predicting future data. As a case study, we chose an area in the Java Sea, Indonesia, i.e., Jakarta Bay. We use hourly wave data collected from ERA5, in the period from January 2018 to December 2020. Besides the GRU with the Attention, we also compare it with the original GRU and LSTM models. The results obtained in this study indicate that the addition of the Attention mechanism in the GRU gives the best performance compared to the original GRU and LSTM. The GRU with Attention produces RMSE and MAPE of 0.013 and 0.023, respectively, whereas the original GRU gives 0.013 and 0.023, and the original LSTM gives 0.018 and 0.041, respectively.