Low-rank Regularized Long-term Forecasting Model for Wind Power Forecasting

This paper proposes a novel low-rank regularized long-term forecasting model for wind power forecasting, addressing the challenges of high-dimensional data complexity and model robustness. Wind power generation is characterized by stochasticity, uncertainty, and intermittency, which pose significant challenges to traditional forecasting methods when dealing with complex input variables. To tackle these issues, we employ Principal Component Analysis (PCA) for low-rank approximation to reduce data dimensionality, extract key features, and subsequently combine it with Light Gradient Boosting Machine (LGB) for prediction. The study begins with comprehensive data preprocessing, including missing value imputation and PCA for dimensionality reduction. The low-rank approximated data is then fed into the LightGBM model for training. Experimental results demonstrate consistent forecasting performance across multiple wind turbines, with some turbines exhibiting higher prediction errors due to their more complex wind power patterns. The comparison between predicted and actual values confirms the effectiveness of PCA in dimensionality reduction and LightGBM in capturing non-linear relationships. The proposed method not only enhances the accuracy of wind power forecasts but also improves the stability and interpretability of the model, making it a valuable tool for renewable energy management and grid integration,