Mitigating Temporal Lag in Ultra-Short-Term Wind Speed Forecasting Using a VMD–S2S–Attn Model

Accurate ultra-short-term wind speed forecasting is critical for the reliable operation and scheduling of wind power systems, yet it remains challenging due to the strong nonlinearity, non-stationarity, and temporal dependence of wind speed time series. Data-driven models have been successfully used for predicting wind speed, often suffering from prediction lag and reduced accuracy. To address these challenges, an ultra-short-term wind speed forecasting strategy integrating Variational Mode Decomposition (VMD) with an Attention-enhanced Sequence-to-Sequence (S2S-Attn) architecture is proposed. Firstly, VMD decomposes the original wind speed time series into a set of band-limited intrinsic mode functions each with a relatively stable center frequency. Then, a sequence-to-sequence method aided with the attention mechanism conducts dynamic weighted modeling on information from historical time steps during the multi-step prediction process, which sufficiently captures nonlinear temporal dependencies in the wind speed time series and effectively improves multi-step prediction accuracy. Comprehensive experiments are conducted by inputting wind speed data measured at multiple heights, with window sizes ranging from 30 to 240 min and forecasting horizons from 10 to 60 min. The experimental results indicate that the proposed framework effectively mitigates temporal prediction lag across all heights and forecasting horizons. The model exhibits consistently higher predictive performance, with root mean square error values between 0.10 and 0.18, and R2 values remaining in the range of 0.98–0.99. The reliability and accuracy of the proposed framework are further validated by other metrics, such as Mean Absolute Error and Mean Absolute Percentage Error. Overall, the proposed model outperforms in both temporal alignment and predictive accuracy, confirming its effectiveness for ultra-short-term wind speed forecasting in multi-step predictions.