Wind Speed Prediction Method Based on CEEMDAN-UAR-BiLSTM-Attention

Given the nonlinearity, instability and uncertainty of wind speeds, effectively harnessing wind energy to ensure stable operation of power systems is critical. Based on uncertainty theory, this paper proposes a hybrid CEEMDAN-UAR-BiLSTM-Attention wind speed prediction model. The CEEMDAN algorithm decomposes the original wind speed time series into multiscale IMF (Intrinsic Mode Function) components, thereby reducing data complexity and mitigating noise interference. Subsequently, each component is modeled separately, the UAR model captures the linear trend of low-frequency components, while high-frequency components are fed into a bidirectional LSTM equipped with an attention mechanism. Leveraging the bidirectional structure of LSTM, the model reveals long-term dependencies while focusing on critical predictive information. During the prediction process, the combined model achieves reduced RMSE and MAE, attaining a coefficient of determination R 2 of 0.89. This provides robust support for wind power planning and operation, showcasing significant practical value and broad application prospects.