Development and Evaluation of a Short-Term Current Forecasting Model with a Deep Learning Error Correction Algorithm

In this study, based on the COAWST (Coupled Ocean-Atmosphere-Wave-Sediment Transport) model and the GRU (Gated Recurrent Unit) algorithm, a short-term forecasting model named as COAWST-GRU for current speed profiles was developed. The model used the GRU algorithm and optimization strategy to make an error correction to the forecasted current speed from the COAWST's outputs, in order to obtain more accurate in-situ current speed forecasts. According to the comparison of the model's prediction results in Cangnan offshore wind farm with the submerged buoy observation data, the COAWST-GRU model significantly improved the accuracy of the current prediction, and the mean error was reduced by about 67% when compared to the original COAWST model. In the vertical direction, the model realized an overall correction of the error in all depth layers, which had a high potential for application in marine engineering construction.