A short-term wind power forecasting method based on evolution-framed fuzzy GANs

Abstract

Wind power forecasting is an essential technology for a reliable power system. Although many methods demonstrated good performance in wind power prediction, there is limited research on forecasting under extreme conditions, which pose challenges for wind power generation and management. To address this problem, this paper proposes an evolutionary short-term wind power forecasting method that uses a three-module evolutionary framework to generate and optimize samples in extreme conditions for better model performance. In this framework: first, a supervised deep convolutional generative adversarial network (S-DCGAN) is proposed so that reasonable wind power samples can be generated. Second, a special fuzzy inference system (FIS) is designed according to similarities in wind power correlations, which makes the selected samples more diversified. Third, a statistical method is adopted to identify the required samples. Finally, with this evolutionary framework, the required training samples can be properly generated and the forecasting performance can be greatly improved. Four groups of experiments are conducted to evaluate the proposed method, and models trained using the generated samples by the proposed method improve averaged MAE and RMSE by more than 5% in wind power forecasting. The result of the experiment shows that the proposed method is effective.