A Short-Term Wind Speed Forecasting Model

Accurate ultra-short-term wind speed forecasting is crucial for wind energy development. However, wind speed sequences exhibit pronounced volatility, and existing research struggles to fully extract their nonlinear patterns. Aiming to improve the prediction accuracy, this paper proposes a hybrid model that integrates Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN), Discrete Wavelet Transform (DWT), component reconstruction, and a Nested Long Short-Term Memory (Nested LSTM) architecture for wind speed prediction. Firstly, the CEEMDAN method is applied to decompose the original sequence, reducing its instability and nonlinearity, resulting in components of varying complexity. Next, use the DWT method to reduce the instability of the high-complexity modal components, while reconstructing the low-complexity components by summation. Finally, put the processed component data into the Nested LSTM model for training. This method uses the NREL Flatirons Campus (M2) wind farm data and reproduces several models for comparative experiments. The results demonstrate that, compared to advanced second-order decomposition prediction models, the proposed model achieves improvements of 59.64%, 59.72%, and 38.25% in MAE, RMSE, and MAPE, respectively.