Wave Forecast using Bidirectional GRU and GRU Method Case Study in Pangandaran, Indonesia

Abstract

Wave forecasting is needed to support marine activities, especially planning for ship navigation, maintenance, and offshore engineering construction. Especially in an open sea, forecasting waves can be challenging due to the stochastic nature of the waves. In this research, we use deep learning models to forecast significant wave height, i.e., the Gated Recurrent Unit (GRU) both unidirectional and bidirectional variants. As a case study, we use six-year time series wave data in Pangandaran, West Java, Indonesia. We use the historical wave data to forecast the next one, three, and seven days’ wave conditions. Results of prediction show that the GRU gives better results than the experiment’s Bidirectional GRU (BiGRU). To predict one day ahead, the GRU gives an RMSE value of 0.2184 and R2 value of 0.9863, whereas the BiGRU gives a prediction with an RMSE value of 0.2206 and R2 value of 0.9869.