Temporal-Spatial Graph Neural Network for Wind Power Forecasting Considering the Blockage Effects

Wind power forecasting (WPF) plays a critical role in ensuring the security, stability, and economics of the power grid. However, the variable nature of wind energy poses significant challenges to WPF accuracy, especially when considering the complex mutual influence between wind turbines in wind farms. To address this issue, a novel neural network for WPF has been proposed in this study. The proposed model utilizes a gated dilated inception network and graph neural network to learn temporal and spatial features concurrently. Additionally, a novel mechanism has been developed to calculate the mutual influence between wind turbines and improve forecasting accuracy by incorporating the blockage effect on each turbine. The proposed model has been validated on a real-world dataset for wind power forecasting with a prediction horizon of 48 hours. The experimental results demonstrate that the proposed model outperforms state-of-the-art methods in terms of forecasting accuracy. The study highlights the importance of considering the spatial and temporal features of wind farms to improve WPF accuracy and demonstrates the efficacy of the proposed model in addressing this challenge.