Short-Term Surface Wind Field Forecasting at Multiple Stations Based on MSTFRDGC-BOBiLSTM

Wind field forecasting is crucial in meteorology, impacting aviation safety, wind power generation, and weather warnings. Wind field variations are complex, influenced by various meteorological factors, and historical data are often incomplete, increasing the difficulty of prediction. In response, we propose a novel model integrating a Multivariable Spatiotemporal Graph Convolutional Network and a Bayesian Optimization Bidirectional Long-Short-Term Memory Network, termed MSTFRDGC-BO-BiLSTM, for short-term wind field forecasting. This model employs an interpolation attention mechanism and adaptive graph convolutional networks to extract spatial features and handle data missingness, while utilizing a Bayesian-optimized BiLSTM to precisely model temporal sequence changes. Experiments demonstrate that the model is significantly effective in wind field forecasting and has practical application value.