Multi-step forecasting of ocean wave height using gate recurrent unit networks with multivariate time series

Abstract

Ocean wave height is an essential parameter for ocean engineering construction, planning decisions, and coastal hazards assessment. Long-term, accurate, and reliable ocean wave height forecasts are critical for the purposes mentioned above and have attracted more attention in recent years. This work proposed a novel method that achieves robust short-term and long-term ocean wave forecasting via the gate recurrent unit (GRU) network. The GRU-based wave forecasting model is established to learn long-term dependency among multivariate sequential data. The future wave height is predicted based on learned features via the proposed method. Case studies of 6 different stations along the coast of China are investigated. The results show that for 1-hour forecasts, the GRU network is superior to comparison methods in terms of all the error metrics. For 3-hour forecasts, the GRU network shows more robustness compared to the LSTM algorithm. The results also validate that the presented scheme is an efficient and reliable short-term and long-term wave forecasting approach. Applying the forecasting method in reality is essential for ocean safety, ocean exploitation, and many other fields.