Wide angle range wind direction ultra-short-term interval prediction based on an improved loss function

Wind direction prediction is essential for the intelligent management of wind farms. This study proposes an ultra-short-term wind direction single-step interval prediction method based on an improved loss function (ILF). The ILF considers the continuity of 0° and 360°. Based on the ILF, the network structure is adjusted to standardize the output range of prediction values. In addition, based on the point prediction method, an interval prediction model for wind direction is further constructed. This interval prediction process uses a distance-based classification of predicted values. Then, actual wind direction data with a large angle range is used to verify the effectiveness of the method. Point prediction results show that the ILF-based model is significantly superior to the mean absolute error (MAE) and mean squared error (MSE) based models. In addition, the same pattern is observed in the other networks. A relatively small fluctuation and other time scale data are also used to verify the performance of the model. The results indicate that the prediction method based on the ILF also has advantages in the two experiments. Finally, the interval prediction results indicate that the proposed method is effective, with an average width of 13.76 % at a 90 % confidence level.