Wind power prediction using stacking and transfer learning
Xu Cheng, Yu Cao, Zhiyuan Song, Chenguang Zhang
Abstract
As countries focus more on renewable energy, especially wind power, predicting wind power output accurately is crucial for managing power grids and saving costs. This paper presents a new method for ultra-short-term wind power prediction using a combination of Stacking and Transfer Learning. To improve accuracy, we first reduce the data dimensions using PCA. Then, we use several models like LSTM, BiLSTM, GRU, BiGRU, and LSTM-Attention as base learners. These models are combined using a Stacking ensemble model. We also use Transfer Learning to share trained models between tasks, which helps improve performance. Tests with real data from a wind farm show that our method is more accurate than single models.