Short-Term Solar-Wind Power Forecasting via Physics-Informed MISO MFAC with Dual-Source Adaptive Forgetting

Accurate short-term forecasting of solar and wind power is crucial for reducing renewable energy uncertainty in smart grids. However, conventional model-free adaptive control (MFAC) methods lack physical consistency and suffer from a trade-off between transient tracking and steady-state smoothness. To address these limitations, this paper proposes a \textbf{Physics-Informed Multi-Input Single-Output MFAC (PI-MISO-MFAC)} framework with a dual-source adaptive mechanism. First, distinct from traditional heuristics, a \textbf{dual-source adaptive forgetting factor} is designed, which dynamically adjusts based on both \textit{input excitation intensity} and \textit{instantaneous prediction error}. This ensures fast tracking during sudden weather ramps and internal system drifts while maintaining robustness against noise. Second, a \textbf{physics-constrained projection algorithm} is introduced to the pseudo-partial derivative (PPD) estimation. By embedding prior knowledge of renewable generation characteristics (e.g., non-negativity of irradiance sensitivity), the interpretability and estimation accuracy are significantly enhanced. Experiments on the German 2016 dataset demonstrate that PI-MISO-MFAC outperforms Informer and optimized LSTM, achieving an MAE of 22.8 kW (18.5\% improvement), while strictly adhering to physical generation laws.