A multi-factor clustering integration paradigm for wind speed point-interval prediction based on feature selection and optimized inverted transformer

Accurate wind speed prediction is essential for enhancing wind power integration and ensuring grid stability. These limitations include insufficient consideration of external factors, oversimplified handling of temporal correlations. This paper proposes a multi-factor clustering integration model for wind speed point-interval prediction, incorporating advanced feature selection and an optimized inverted Transformer. The approach begins with a characteristic contribution assessment method to identify critical factors impacting prediction accuracy, ensuring the model leverages the most influential features. Subsequently, through an intrinsic feature extraction method, the wind speed series are further segmented into multiple clusters, capturing multi-scale dependencies and complex temporal patterns that may be overlooked by traditional models. This segmented approach enables a dual-phase forecasting framework, where the optimized inverted Transformer is applied to each cluster, increasing both predictive stability and precision by aligning forecasts with specific data patterns. Additionally, a point to interval prediction mechanism generates probabilistic intervals that effectively capture the uncertainty inherent in wind speed data. Experiments conducted on two datasets confirm the model's superiority, achieving the lowest mean squared error among comparison models. This integrated methodology enhances the accuracy, robustness, and interpretability of short-term wind speed forecasts, providing a comprehensive solution to the inherent challenges of wind speed prediction.