Multi-task prediction of wind speed and time-varying wind shear coefficient using dynamic graph interactive neural network
This research presents a multi-task learning approach based on graph neural networks for simultaneous prediction of wind speed and wind shear coefficient. The method addresses the dynamic nature of wind shear coefficients that vary with meteorological conditions, overcoming challenges in coupling wind shear phenomena with wind speed prediction.
Jul 5, 2025
Incremental transfer learning based on temporal-frequency convolution interaction for multi-task prediction of wind speed and wind power
This paper introduces an incremental transfer learning approach based on temporal-frequency convolution interaction for multi-task prediction of wind speed and power in newly-built wind farms with insufficient historical data. The method integrates a temporal-frequency convolutional interactive neural network into a parallel framework with circular convolution and gated recurrent units, achieving significant reduction in prediction errors compared to classical LSTM algorithms.
Jul 5, 2025
A selective memory attention mechanism for chaotic wind speed time series prediction with auxiliary variable
This paper proposes a novel selective memory attention mechanism to enhance wind speed prediction accuracy by leveraging auxiliary variables. The method introduces an adaptive frequency-domain selection attention weight operator to parse effective information from different frequency intervals, significantly reducing prediction errors compared to classical LSTM algorithms.
Jul 5, 2025