Wave height forecasting correction model based on transformer improved UNet

Ocean waves are irregular combinations of waves with multiple wave heights, periods, and directions of travel, and are an important element of the marine environment. Accurate wave forecasting plays an important role in reducing wave accidents and guiding offshore production activities. In response to the spatial dependence of wave forecasting data, this paper establishes an intelligent correction model for wave forecasting based on Transformer improved UNet (T-UNet). Introducing the Transformer structure into the encoder of the traditional UNet model and using a dual sampling module instead of traditional upsampling in the decoder to improve feature extraction capability, the improved model can better understand the dependency relationships between different positions in the sequence and achieve effective correlation between local and global features. This article uses wave height and wind speed forecast data and significant wave height reanalysis data provided by the European Centre for Medium Range Weather Forecasts (ECMWF) to conduct comparative experiments with traditional UNet models. The experimental results show that the root mean square error and average absolute error of the T-UNet model for the correction of future 24-hour forecasts are smaller than those of the UNet model. Compared with the root mean square error before correction, it has decreased by more than 72.4%, effectively reducing the error of numerical forecast data and improving the forecasting ability of wave height.