Real-time wind power forecasting using an evolving fuzzy system based on Multivariate Gaussian and Chebyshev mapping

The increasing penetration of wind energy, driven by global energy transition goals, has intensified the demand for highly accurate short-term wind power forecasting to ensure the stability and efficiency of power systems. However, the inherently chaotic, intermittent, and nonlinear nature of wind power, coupled with rapidly changing meteorological conditions, poses significant challenges to forecasting accuracy in real-world operations. Most existing approaches focus on offline training and validation, lacking the adaptability required for real-time prediction in dynamic environments. To address this gap, we propose a novel online forecasting framework, termed Multivariate Gaussian Chebyshev Mapping Evolving Fuzzy System (MGCM-EFS). Unlike prior evolving fuzzy systems (EFSs) that rely on fixed thresholds or univariate memberships, MGCM-EFS jointly integrates multivariate Gaussian memberships, dual-side Chebyshev adaptation (antecedent thresholds and consequent forgetting), and density-utility pruning to preserve a compact yet expressive rule base in real time. Specifically, Chebyshev mapping is incorporated into both the antecedent and consequent parts to enhance nonlinear modeling capability, while the density-based pruning mechanism dynamically removes low-contribution rules to prevent model bloat and overfitting. This design enables MGCM-EFS to evolve its fuzzy rules and parameters in real time, effectively adapting to non-stationary wind power data streams. Experimental results on multi-country wind-power datasets show that MGCM-EFS achieves up to 29.3% higher prediction accuracy and up to 86.9% faster computation than state-of-the-art models, while providing second-level response and multi-scale adaptability suitable for real-time dispatch in high-frequency power systems.