A Closed-Loop MPC Framework with Hybrid Forecasting for Cost-, Carbon-, and Degradation-Optimal Residential Microgrids

The increasing penetration of renewable energy sources in residential power systems introduces significant operational challenges due to their inherent intermittency and uncertainty. In grid-connected microgrids, improper energy management can lead to elevated operational costs, increased carbon emissions, and rapidly increasing rates of battery degradation. This paper presents an advanced Energy Management System (EMS) based on a closed-loop Model Predictive Control (MPC) structure that employs highly accurate hybrid models to forecast residential load demand and photovoltaic generation. Forecasting of solar generation uses a GRU-XGBoost hybrid model, while load demand forecasting is accomplished using an LSTM model optimized with the Grey Wolf Optimizer. The EMS employs a multi-objective formulation that utilizes real-time carbon intensity signals and Time-of-Use tariffs to jointly optimize economic cost, environmental impact, and battery degradation. The proposed methodology is evaluated over one week on a representative grid-connected residential microgrid and compared with Particle Swarm Optimization and Rule-Based Control (RBC). Simulation results demonstrate that the proposed MPC-based EMS achieves up to a 7.1% reduction in operational costs and up to a 12% decrease in CO₂ emissions compared with RBC strategies. In addition, the predictive and degradation-aware formulation ensures smoother battery operation and enhanced system resilience under adverse weather conditions. These results confirm the effectiveness of the proposed approach for achieving cost-efficient and environmentally sustainable microgrid operation.