This is the authors' abstract. We don't add key points for this paper.
Escalating renewable penetration necessitates precise, near-real-time wind projections for reliable grid dispatch and systemic equilibrium. To address key challenges – including non-stationary wind dynamics, multi-scale temporal fluctuations, redundant input features, and stringent low-latency deployment constraints – this study proposes a novel end-to-end forecasting framework that is modular, computationally efficient, and inherently interpretable. The framework introduces dual correlation-based feature selection to enhance input relevance; employs a Multi-scale Exponential Moving Average (MEMA) operator for robust dual-stream decomposition of trend and seasonal components; applies Selective Kolmogorov–Arnold Network (SKAN) for residual correction; employs a Gated Adaptive Fusion (GAF) mechanism to dynamically regulate the contributions of trend-oriented and fluctuation-oriented representations under different operating conditions; and utilizes an Improved Grey Wolf Optimizer (IGWO) for systematic hyperparameter tuning. Evidence obtained from comprehensive trials involving real-world wind energy records confirms the superiority of this framework over several state-of-the-art forecasting models. Compared with the strongest model among the conventional forecasting baselines evaluated in this study (DLinear), the proposed framework reduces the mean absolute error (MAE) and mean squared error (MSE) by 27.25% and 38.06%, respectively, while achieving an 87.85% reduction in inference latency. These findings underscore the framework’s practical viability for real-time decision support in wind-integrated power grid operations.
Answers are grounded in this paper and cite their sources.
No takes yet. Share an insight, caveat, or question.
Yang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: