A Novel Point-Interval-Valued Wind Speed Prediction System from the Perspective of Mixed-Frequency Data

Accurate wind speed prediction is crucial for enhancing wind power generation efficiency and ensuring grid stability. While previous research has predominantly focused on point-valued or interval-valued predictions using common-frequency data, these approaches often fail to fully capture the multi-scale variability and uncertainty inherent in wind speed sequences. To address this limitation, this paper introduces a novel ensemble prediction system that incorporates mixed-frequency data with point-interval-valued modeling. The proposed framework systematically incorporates mixed-frequency characteristics by applying Symplectic Geometric Mode Decomposition to jointly denoise high-frequency and low-frequency point-interval-valued components; combining Mixed Data Sampling with artificial intelligence models to generate sub-model predictions and mixed-frequency point-interval-valued results and implementing an improved multi-objective ensemble mechanism using the Multi-Objective Rime optimization algorithm to optimally combine the sub-model outputs. Experimental evaluation using real-world wind speed data from two locations in Nanning, China, demonstrates the system's superior performance, achieving Point-Interval-Valued Mean Absolute Percentage Error values of 3.7984% and 4.7028%, respectively, and outperforming twenty benchmark models. The results highlight the effectiveness of mixed-frequency data in enhancing point-interval-valued prediction accuracy and provide a robust solution for wind energy applications.