Short-term wind power forecasting method for extreme cold wave conditions based on small sample segmentation
Lin Lin, Jie Xu, Jianfei Liu, Hao Zhang, Pengchen Gao
Abstract
With the intensification of global warming, the frequency of extreme weather events has increased, drawing significant attention from countries worldwide to the deteriorating environmental conditions. In this context, nations have accelerated the transition of their energy structures to reduce dependence on fossil fuels and lower carbon emissions. Renewable energy, including wind power, has become a key focus of energy policies due to its clean and sustainable nature. Cold waves, as one of the most common extreme weather events, cause significant fluctuations in wind power over short periods, greatly increasing the difficulty of wind power forecasting. To address this issue, this paper proposes a segmented wind power forecasting method based on the generation of small sample cold wave data. First, the characteristics of wind power fluctuations during cold wave conditions are analyzed. Given the scarcity, extreme values, and high volatility of the sample data, a Sequence Variational Autoencoder (SeqVAE) algorithm is employed to generate numerical weather prediction data and corresponding power samples. Then, a cold wave power loss extraction method based on Graph Convolutional Networks (GCN) and Bidirectional Gated Recurrent Units (BiGRU) is introduced for the entire time period. On this basis, the Light Gradient Boosting Machine (LightGBM) method is used to predict power during normal weather periods, while a LightGBM-Transformer method is proposed for predicting power losses during such periods. Finally, the method is validated using datasets, and results show that the proposed approach effectively improves forecasting accuracy.