Robust design of microgrids under uncertainties: Comparative impact of modeling choices on sizing decisions

In this article, we assess how modeling choices affect the robust co-design of a microgrid under uncertainty. We investigate a five-household grid-connected system with solar, battery, and hydrogen storage, jointly optimized for Net Present Value and Renewable Energy Share (evaluated under a Conditional Value at Risk formulation) using NSGA-II. Four modeling axes are independently varied: time horizon, scenario multiplicity, component models, and parameterization. The resulting Pareto fronts are compared to a reference configuration. Our original methodology combines a Key Performance Indicator (KPI)-centered analysis using the newly introduced Differential Pareto Hypervolume metric with a decision-centered comparison of design variable distributions. This enables a systematic evaluation of modeling simplifications. Results show that temporal representation is a dominant factor: reducing data from 20 to 1 year strongly distorts KPIs and sizing, whereas 5-year horizons substantially mitigates errors by better capturing long-term degradation. Component modeling and scenario multiplicity emerge as a second-order factors: they have limited impact on global KPIs, mainly affecting storage and grid reliance. Parameterization is third-order factor, influencing KPIs more than the design itself—except for storage-related decisions. These findings question the common use of a single “representative year” and highlight that robust microgrid design must address both short-term variability and epistemic uncertainty. Neglecting long-term dynamics— such as capacity degradation—can outweigh short-term stochastic effects underscoring the need to prioritize modeling detail carefully to ensure robust design outcomes.