CTRL: Collaborative Temporal Representation Learning for Wind Power Forecasting

Accurate wind power forecasting is crucial for grid stability and renewable energy integration, but existing methods struggle to capture complex temporal dependencies in wind data. This paper introduces Collaborative Temporal Representation Learning (CTRL), a novel deep learning model that leverages collaborative representation learning to enhance forecasting accuracy and robustness. CTRL integrates Reversible Instance Normalization (RevIN), RNN-based hidden state learning, and a specialized collaborative representation unit to capture multi-directional temporal dynamics across different time scales and variables. Experimental results on two real-world wind power datasets demonstrate that CTRL significantly outperforms 20 existing methods, including state-of-the-art deep learning approaches, achieving up to 9.67% and 10.42% improvement in forecasting accuracy, respectively. These findings highlight the potential of collaborative representation learning for advancing wind power forecasting and facilitating the effective integration of renewable energy resources.