Ultra-Short-Term Wind Power Forecasting Based on DT-DSCTransformer Model

When using the Transformer model for wind power prediction, the accuracy of the model predictions tends to be reduced due to the shift in the wind power data distribution, channel mixing, and the inability of the model to establish strong correlations. To address these challenges, this paper proposes an ultra-short-term wind power prediction model based on the DT-DSCTransformer. First, the model applies DT's self-learning standardization and de-standardization parameters to standardize the input and de-standardize the output, mitigating the impact forecasting of data distribution shifts on prediction accuracy. Second, the proposed De-Stationary Channel Attention (DSCAttention) mechanism is introduced. By incorporating De-Stationary Attention (DSAttention) into the channel attention mechanism while maintaining channel independence, the model establishes stronger inter-channel correlations, addressing the performance degradation caused by channel mixing and weak correlations. Finally, experimental analysis demonstrates that the proposed model achieves the highest prediction accuracy compared to commonly used time series forecasting models.