A dual-dimensionality reduction attention mechanism with fusion of high-dimensional features for wind power prediction

As wind farm monitoring technologies improve, the complexity of wind power data increases. This added complexity, particularly feature redundancy in high-dimensional data, can negatively affect the performance of prediction models. To address this, we propose a dual-dimensional reduction attention mechanism that reduces redundancy and highlights the most important features. The mechanism includes an active reduction layer that evaluates features based on their relevance to power predictions and a passive reduction layer that assigns appropriate weights to these features. We also introduce an adaptive variational mode decomposition (VMD) algorithm, which combines the Crisscross Search Optimization (CSO) with VMD to analyze wind power data. A novel entropy-based fitness function is used within this framework. For prediction, a Bidirectional Gated Recurrent Unit (BiGRU) model is used to capture the relationships between input features. The method is tested on real offshore and onshore wind farm data, and the results show that it outperforms existing methods in accuracy for high-dimensional data.