Intelligent Prediction for the Wind Speed Based on Hybrid Strategy

Interval prediction of wind speed is essential for ensuring wind power generation stability and efficiency. Addressing limitations in capturing nonlinear features using traditional interval methods, this paper proposes a novel approach integrating an improved wavelet threshold (WT), a Deep Learning model (BiTCN-BiGRU), and the Nutcracker Optimization Algorithm (NOA). Initially, NOA separately optimizes the WT denoising parameters and the BiTCN-BiGRU model hyperparameters. Subsequently, the NOA-optimized WT (NOA-WT) smooths the wind speed data. To identify chaotic characteristics and capture temporal dependencies, Phase Space Reconstruction (PSR) analyzes the smoothed data. Finally, the NOA-optimized QR-BiTCN-BiGRU model (NOA-QR-BiTCN-BiGRU) generates the wind speed prediction intervals. Empirical comparisons against other deep learning methods, under consistent hyperparameters and network architecture, demonstrate that the combined NOA-WT-QR-BiTCN-BiGRU model achieves superior performance. This methodology offers valuable insights for interval forecasting of other complex nonlinear time series in practical applications.