Time Series Generative Adversarial Network‐Empirical Mode Decomposition‐Stacking Ensemble Wind Power Prediction Model for Extreme Weather Conditions

Aiming at the problems of low prediction accuracy, large prediction deviation, and scarce samples that make modelling difficult under extreme weather conditions such as cold waves, this paper proposes a wind power prediction model based on time series generative adversarial network (TimeGAN)—empirical mode decomposition (EMD)—stacking ensemble. First, the TimeGAN algorithm is used to generate samples of meteorological and power data to build a database of extreme weather meteorological and power data. Next, based on EMD, the expanded wind power data are decomposed into high and low‐frequency sequences. For each subsequence obtained from decomposition and reconstruction and the corresponding meteorological factors, an improved particle swarm optimisation (IPSO) algorithm is utilised to optimise the parameters of several commonly used prediction methods and to compare their prediction accuracy. Then, combining the performance of the algorithms and the principle of maximum difference, the optimal algorithm combination is determined by using prediction error evaluation indicators and Pearson correlation coefficients to construct a stacking ensemble learning‐based multi‐algorithm combination model for extreme weather power prediction. Finally, the effectiveness of the proposed method in improving prediction performance is verified through a real cold wave case, demonstrating its good practicality.