The increasing integration of renewable energy introduces uncertainty in microgrid operation, making effective energy management more challenging. Rolling-horizon optimization is used to address this challenge by enabling periodic decision updates; however, most existing approaches rely on fixed optimization horizons and predetermined update frequencies. When prediction accuracy decay (PAD) is considered in adaptive rolling-horizon approaches, it is represented using a fixed decay value, not an online indicator that compares forecasted and actual renewable generation during operation. This leads to suboptimal re-optimization timing, unnecessary computational effort, excessive battery switching, or delayed corrective actions. To address these limitations, this paper proposes a PAD-driven adaptive rolling horizon (ARH) approach, in which re-optimization is triggered using an online PAD indicator computed from the percentage deviation between forecasted and realized renewable generation data. Re-optimization is activated when the PAD indicator exceeds a predefined threshold, enabling adaptive scheduling updates according to real-time forecasting degradation. The problem is formulated as a robust mixed-integer linear programming (MILP) model of a high renewable penetration microgrid, including battery degradation and switching penalties. The energy self-sufficiency ratio (SSR) is used as a key sustainability performance indicator to assess the extent to which local renewable generation and storage satisfy microgrid demand. The proposed approach is first compared with a fixed rolling-horizon approach using a fixed re-optimization interval of 1 h, where the results show a profit improvement of 10.5%. A sensitivity analysis tested the proposed approach under bounded renewable forecast uncertainty levels up to ±15 and different battery capacities. The results indicate that performance is strongly influenced by forecast accuracy and battery capacity, with higher economic gains under low uncertainty and more conservative operation under high uncertainty.
Elgazzar et al. (Mon,) studied this question.