Wind Speed Forecasting Based on Trend Augmentation and Dual-Phase Inverted Transformer

Wind speed prediction is crucial for power system operation management and electricity trading. Accurate wind speed prediction relies on the extraction of temporal features and correlation characteristics between different variables. However, existing methods, especially recurrent neural network-based models and Transformer-based models, typically focus on extracting temporal features, often neglecting the extraction of inter-variable correlation characteristics. In this paper, a Dual-phase Inverted Transformer with trend augmentation is introduced. By deploying dual-phase encoders and inverted attention mechanism considering phase difference between sequences, model's capacity to capture covariate dependency is enhanced. Measured wind speed data collected from three wind stations is applied to test model's performance. The results demonstrate that by applying the proposed model, the performance of wind speed forecasting is improved, enabling more accurate wind power prediction. The proposed strategy enhances the predictability and controllability of wind power generation, promoting the wider application of wind energy.