A novel spatiotemporal fusion wind speed prediction method under wind farm control center: BiTCN–transformer–cross-attention

This paper introduces a hybrid model for multivariate multi-wind farm wind speed prediction to reduce operational costs at wind farm control centers and enhance prediction accuracy. Initially, a parallel prediction model called BiTCN–Transformer–Cross-attention (BTTCA) is developed, which integrates spatiotemporal features using cross-attention. The BTTCA model is pre-trained using historical data from four typical wind farms, with input consisting of historical wind speed and related meteorological information. Subsequently, the pre-trained models are deployed via transfer learning to predict wind speed at various other wind farms managed by the control center. By improving prediction accuracy, the model minimizes manual interventions, optimizes resource allocation, and enhances the operational efficiency of wind farms, thereby effectively reducing operational costs. Additionally, the MOPOA is applied to refine the predictions of these 4 pre-trained models, maximizing their offline potential and improving prediction accuracy. Analysis of experimental results and comparisons with other algorithmic models suggests that this hybrid wind speed prediction model performs effectively.