PV forecasting-driven decision-support for solar microgrids with battery storage

We propose an integrated framework for solar microgrids that addresses PV intermittency by coupling forecasting with real-time, auditable decision support in a single pipeline. The approach combines a Random Forest forecaster—selected via head-to-head benchmarking for accuracy, interpretability, and edge feasibility—with a rule-based energy management system (EMS). On a held-out test set, the model attains R2 = 0.9310 and MAE = 8180.61 kWh for daily PV yield prediction. A Counterfactual Module converts forecasts into operator-readable “what-if” recommendations, identifying feasible adjustments to environmental and operational factors. A Decision-Support Module then executes transparent, rule-based decisions that coordinate energy flows among the PV array, the battery energy storage system (BESS), and the main grid subject to explicit constraints (e.g., state-of-charge, inverter ratings, and exchange limits). By closing the loop from sensors to forecast to counterfactuals to dispatch, the framework moves beyond accuracy-only studies and provides an explainability-to-action, edge-deployable solution for storage-aware scheduling and dispatch. Case studies illustrate how the pipeline transforms predictions into real-time, operator-auditable actions that support stable and efficient microgrid operation.