Ultra-Short-Term Wind Power Forecasting Based on the Improved Dlinear Model

Wind energy plays a more significant role in power grids due to its eco-friendly and renewable characteristics. This study proposes an improved prediction approach using the DLinear model, which incorporates the interactive learning module along with the channel and token attention mechanisms. The initial step involves decomposing the input time series into trend and seasonal components with a moving average kernel, followed by applying interactive learning to improve the modeling of these components. The combined framework demonstrates superior performance on multiple private datasets, significantly improving prediction accuracy and reliability compared to existing baseline models. The experimental findings show that the proposed method achieves a lower mean square error (MSE) and a lower mean absolute error (MAE) than several current models on two private datasets, thereby showcasing the combined effectiveness of these two modules.