Methods for Short-term Wind Power Forecasting under Small Sample Conditions of Extreme Weather such as Cold Waves

Extreme weather events such as cold waves pose significant challenges to the safe operation of wind power and new types of power systems dominated by wind power. Providing accurate wind power forecasting in response to these challenges is an effective countermeasure. To this end, this paper proposes a short-term combined wind power forecasting method under small sample conditions based on TimeGAN-EMD-LSTM (Temporal Adversarial Generative Network, Empirical Mode Decomposition, and Long Short-Term Memory Neural Network), which effectively improves forecasting accuracy. The method first uses the TimeGAN algorithm for small sample expansion, then employs ensemble empirical mode decomposition to decompose the expanded wind power into high and low frequency sequences. Separate LSTM neural network models are established for each decomposed and reconstructed sub-sequence and meteorological factors. Finally, the results of each sub-sequence forecasting model are superimposed to obtain the wind power forecast value. Compared with conventional LSTM forecasting algorithms, the accuracy of the combined forecasting algorithm under cold wave weather is significantly improved, with the RMSE reduced by 46.82% and the MAE reduced by 68.22%.