A graph neural model for predicting wind speed behavior based on the effect of wind speed point coupling

In this paper, HI-MSTG (Hierarchical Linear Attention Blocks with Multiscale Interactive Spatiotemporal Graph Neural Networks) is proposed to address the key issues of 3D Spatiotemporal Wind Speed Prediction in Wind Farms with Complex Terrain. We propose a three-stage progressive framework to tackle the challenges of modeling nonlinear features, spatiotemporal dynamics, and the coupled multidimensional spatial relationships of wind speed points in complex terrains: To begin, the dynamic feature screening method is used to obtain wind speed-related operational parameters of wind turbines; then, a novel principal component analysis is used to group all wind turbines in the wind farm with spatial and temporal correlation; and finally, three spatial scale wind speed coupled prediction models are established by combining the multi-source data. The experimental results demonstrate that the model effectively identifies wind direction and yaw angle as key variables through dynamic feature extraction, offering a novel approach to unit classification in complex terrain. The model achieves an accuracy of 87.3 % and a stability of 90 % in the 4-h prediction. Compared to traditional neural network models and actual operational data, this method significantly enhances the representation of spatiotemporal features. It effectively reduces the complexity of multidimensional dependency modeling in complex terrain wind farm forecasting. The application cases of real wind farms demonstrate the model's engineering practical utility in 3D Spatiotemporal wind speed prediction.