VMD-CAT: A hybrid model for short-term wind power prediction
Huan Zheng, Zhenda Hu, Xuguang Wang, Junhong Ni, Mengqi Cui
Abstract
Accurate wind power prediction is essential to optimize the wind power scheduling and maximize the profits. However, the inertia and time-varying property of the wind speed pose a challenge to the wind power prediction task. The existing prediction models fail to efficiently mitigate the negative influence of these properties on the prediction results. Therefore, their generalization abilities require a further improvement. In this paper, the historical wind power segment is decomposed into sub-signals, which are considered as the fluctuation patterns of the wind power series, the variable support then is employed to describe the inertia and time-varying properties for the fluctuation patterns. The component-attention mechanism is introduced to formulate the correlation-relationship between each fluctuation pattern and the historical wind power segment, this mechanism is used to replace the self-attention mechanism for the Transformer model. A hybrid model combined VMD and Transformer is utilized for predicting the future wind power. Experiments performed on an actual wind power series validate the efficiency of the proposed model.