Simulation and visualisation for a wind power prediction model based on structural attention LSTM and environmental correction

With the increasing share of renewable energy, its volatility poses challenges to grid dispatching, making wind power prediction crucial. Existing methods mainly include point forecasting and probabilistic forecasting, but the former struggles to capture fluctuations, while the latter lacks reasonable scenario generation for grid integration. Additionally, current approaches fail to fully utilise wind farm spatial structures and environmental factors, limiting prediction accuracy and generalisation. To address this, this paper proposes a scenario generation model (SLEP) based on structural attention LSTM and environmental correction. SLEP integrates temporal wind power characteristics, turbine spatial structures, and environmental factors, built upon TimeGAN. SA-LSTM combines a graph convolutional network (GCN) with LSTM to capture spatiotemporal wind power features, while the environmental correction module (ERM) employs cross-attention to embed environmental variables, improving sample adaptability. Experiments show that SLEP outperforms existing methods in accuracy, scenario diversity, and environmental adaptability, providing reliable support for grid dispatching.