Ultra-short-term Forecast of Wind Speed in Qingdao Port Based on Spatiotemporal Large Graph Network Model

This study presents a novel spatiotemporal graph modeling method for ultra-short-term wind speed forecasting in Qingdao Port. The proposed model, termed GWSTGM, is based on graph wavelet convolutional neural networks for spatiotemporal graph modeling and aims to improve the predictions of wind speed. The model effectively captures the spatiotemporal correlations of attribute features between nodes in a graph by incorporating a dynamic adjacency matrix for node embedding learning. Additionally, stacked dilated causal convolution is employed to capture the temporal correlations among graph nodes. The proposed model is evaluated using three distinct datasets, including traffic road network speed data and meteorological wind velocity data. The results demonstrate that the GWSTGM model outperforms other benchmark methods in short-term wind speed prediction tasks, highlighting its potential for uncovering spatiotemporal structures in input data. Future research will focus on the model’s performance in large graph structures and dynamic graph structures, as well as its applicability in various application fields.