Energies, Vol. 19, Pages 349: A Short-Term Wind Power Forecasting Method Based on Multi-Decoder and Multi-Task Learning

Energies, Vol. 19, Pages 349: A Short-Term Wind Power Forecasting Method Based on Multi-Decoder and Multi-Task Learning

Energies doi: 10.3390/en19020349

Authors:
Qiang Li
Yongzhi Liu
Xinyue Yan
Haipeng Zhang
Siyu Wang
Ran Li

In short-term power forecasting for wind farms, factors such as weather conditions and geographic location lead to certain correlations in the power output of different wind farms, resulting in complex coupling relationships between them. Traditional wind power forecasting methods often predict each wind farm independently, without considering these coupling relationships. To address this issue, this paper proposes a multi-task Transformer model based on multiple decoders, which accounts for the intrinsic connections between different wind farms, enabling joint power forecasting across multiple sites. The proposed model adopts a single encoder-multiple decoder structure, where a unified encoder processes all input data, and multiple decoders perform prediction tasks for each wind farm separately. Testing on actual wind farm data from the Inner Mongolia region of China shows that, compared to other forecasting models, the proposed model significantly improves the accuracy of power predictions for different wind farms.