Residential microgrids that couple photovoltaic generation with lithium‐ion storage must curtail electricity expenditure while preserving battery health. Economic model predictive control (EMPC) is widely used for this task, yet a fixed formulation forces an unsatisfactory compromise: high‐fidelity models shrink linearization error but breach real‐time limits, whereas coarse models solve quickly at the price of missed savings when tariffs or demand shift. This study proposes a hyperstructural adaptive EMPC that reconfigures the optimization problem in real time. An intermediate‐fidelity virtual evaluation model gauges whole‐plant cost for each candidate setup; a stochastic local search then selects the most economical blend of prediction‐horizon length, time‐grid resolution, and admissible state‐of‐charge band, while a handover‐consistency constraint forces seamless trajectory transfer between successive solutions. All adaptations retain a mixed‐integer linear programming (MILP) structure, allowing deployment on ordinary embedded hardware under a strict solve‐time budget. Five‐day closed‐loop simulations on a grid‐connected household show that the adaptive scheme lowers total operating cost by about 1% and reduces capacity fade rate by roughly a factor of 2.1 relative to a static EMPC of equal baseline fidelity. Redirecting computation toward the most profitable configuration thus delivers tangible economic and durability gains for low‐voltage prosumers.
Vedel et al. (2026) studied this question.