CEEMDAN-SE-HDBSCAN-VMD-TCN-BiGRU: A two-stage decomposition-based parallel model for multi-altitude ultra-short-term wind speed forecasting

Accurate wind speed forecasting is critical for enhancing wind farm efficiency and facilitating the integration of renewable energy into power systems. Existing wind speed forecasting methods typically rely on single-altitude measurements, neglecting the vertical variability inherent in wind dynamics. This study proposes CEEMDAN-SE-HDBSCAN-VMD-TCN-BiGRU, a two-stage decomposition-based parallel deep learning framework for multi-altitude ultra-short-term wind speed forecasting. The model first applies complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) to extract intrinsic mode functions, addressing nonlinearity and non-stationarity in the data. Sample entropy (SE) is then used to quantify the complexity of each intrinsic mode function, which are clustered into groups based on shared characteristics using hierarchical density-based spatial clustering of applications with noise (HDBSCAN). High-frequency and complex signals are further refined via variational mode decomposition (VMD), capturing intricate temporal patterns. The resulting sub-signals, along with clustered components, are input into a parallel forecasting model integrating temporal convolutional network (TCN) and bidirectional gated recurrent unit (BiGRU) to capture both long-range and bidirectional temporal dependencies. Evaluation on multi-altitude wind speed data from two wind farms demonstrates that the proposed framework outperforms both benchmark and state-of-the-art models. It achieves root mean square error values between 0.17736 to 0.44771 and coefficient of determination values ranging from 0.97394 to 0.99286. Compared with benchmark models, it delivers up to a 71.34% reduction in root mean square error and a 26.47% improvement in coefficient of determination. Other metrics, including mean absolute error and mean absolute percentage error, further confirm the reliability and accuracy of the proposed framework.