Learning-enhanced stochastic model predictive control with trustworthy forecasting for resilient multi-energy microgrid operation

The integration of electricity, heating, cooling, and hydrogen systems in multi-energy microgrids (MEMGs) enhances operational flexibility but introduces significant uncertainty, strong system coupling, and degradation-related challenges. This paper proposes a Learning-Enhanced Stochastic Model Predictive Control (LE-SMPC) framework that integrates trustworthy machine learning forecasting with constraint-aware optimisation. The proposed approach combines (i) degradation-aware modelling of fuel cells and batteries, (ii) permutation-based feature attribution embedded within the learning process to ensure interpretability, and (iii) stochastic MPC with chance constraints for robust decision-making under uncertainty. In this work, resilience is referred to as the ability to maintain safe and feasible operation under stochastic variability and progressive component degradation through adaptive control. Simulation results demonstrate that the proposed method achieves approximately 40% reduction in forecasting error (based on RMSE comparisons with baseline models), while consistently maintaining constraint satisfaction under representative degradation conditions, including up to 18% fuel cell voltage drop and 30% battery capacity loss, as well as varying demand levels. These results highlight that the proposed LE-SMPC framework improves prediction reliability, enhances control robustness, and provides transparent, operator-oriented decision support for multi-energy system management.