Wind Power Modeling based on Data Augmentation and Stacking Integrated Learning

In order to improve the modeling accuracy, this paper investigates the problem of the wind power modeling method based on the data augmentation technology and stacking integrated learning algorithm. First, since the raw wind power data are scarce and unevenly distributed, various flexible data augmentation techniques including jittering, scaling, magnitude warping, and time warping are used to transform the original data to expand the dataset. Then, in order to build a stacking integrated model of wind power, the Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), and Convolutional Neural Network (CNN) based on time domain are selected as primary learners, and a stable Support Vector Regression (SVR) algorithm is used as a secondary learner. Finally, the ablation and comparative experiments are conducted according to the measured data of a wind farm in China. The experimental results show that the data augmentation technique significantly improves the robustness of each learner, and the proposed stacking integrated model is more effective compared with the single learner model, whose modeling accuracy has been effectively improved.