A Physics-Informed Offset-Free MPC approach for PV--Battery Islanded Microgrids without explicit disturbance modeling

Reliable operation of standalone PV/battery microgrids is challenged by uncertainties in both load demand and solar irradiance. Although model predictive control (MPC) has shown promise for power management, its performance deteriorates when prediction errors exceed certain thresholds, leading to suboptimal maximum power point tracking (MPPT) and voltage regulation. This work presents an enhanced offset-free MPC strategy that incorporates real-time irradiance prediction updates through a physics-informed correction mechanism. The proposed method dynamically adjusts the forecasts to compensate for the inaccuracies of the prediction, maintaining the stability of the system under significant uncertainty. Comparative simulations demonstrate that conventional MPC fails to ensure proper MPPT operation during periods of high prediction error, resulting in PV voltage deviations and reduced efficiency. In contrast, the proposed approach maintains consistent tracking performance while preserving DC bus voltage stability and battery state-of-charge (SoC) limits. Key innovations include (1) an adaptive prediction correction framework that improves irradiance estimation accuracy, (2) robust performance across varying levels of uncertainty, and (3) effective power balance under realistic operating conditions. The method proves particularly valuable for remote microgrid applications where prediction reliability is critical but often compromised by environmental variability.