A novel temporal-spatial graph neural network for wind power forecasting considering blockage effects

Wind Power Forecasting is crucial for the operational security, stability, and economic efficiency of the power grid, yet it faces significant accuracy challenges due to the variable nature of wind energy and complex interactions within wind farms. This study introduces a novel neural network model specifically designed for Wind Power Forecasting, incorporating both a gated dilated inception network and a graph neural network. This innovative approach enables the concurrent analysis of temporal and spatial features of wind energy, significantly enhancing the forecasting accuracy. A pivotal feature of this model is its unique mechanism to compute the mutual influence between wind turbines, with a particular focus on the blockage effect, a key factor in turbine interactions. The model's efficacy is validated using a real-world dataset, targeting a 48-hour prediction horizon. The experimental outcomes demonstrate that this model achieves superior performance compared to state-of-the-art methods, with a notable improvement of 6.87% in Root Mean Square Error and 8.77% in Mean Absolute Error. This study not only highlights the model's enhanced forecasting capabilities but also emphasizes the importance of integrating spatial and temporal dynamics in wind farms for improving Wind Power Forecasting accuracy.