Explainable Multi‐Target Forecasting and Resilience Assessment for Microgrid Energy Management

The growing use of decentralised microgrids has resulted in the creation of a need for forecasting frameworks that not only support operational planning but also provide interpretable insights into prediction behaviour under variable renewable generation as well as varying load demand. In this research, it is proposed to use an explainable forecasting and resilience‐analysis framework to predict solar generation, wind generation, hydrogeneration, and load demand simultaneously, using a multi‐target Random Forest model. The forecasting layer is interlinked with SHapley Additive exPlanations (SHAP) and local interpretable model‐agnostic explanations (LIME) to understand the significance of global features and the local drivers of decision‐making. Moreover, a pair of indicators with a resilience orientation, such as net energy balance tracking and cumulative energy debt, is applied to assess how the system can withstand energy adequacy over time under current conditions. The system comprises 2.5 MWp solar, 1.8 MW wind, and 1.2 MW hydro power. Experimental analysis of a multi‐source renewable energy dataset shows that the proposed framework achieves a good predictive performance in solar ( R 2 = 0.999, root mean square error (RMSE) = 1.89 kWh, mean absolute percentage error (MAPE) = 1.85) and wind generation ( R 2 = 0.998, RMSE = 1.09 kWh, MAPE = 0.89%), whereas load demand remains more challenging because of its higher stochasticity ( R 2 = 0.28, RMSE = 110.91 kWh, MAPE = 35.2%). The explainability analysis indicates that the temporal variables and factors related to temperature conditions contribute to the formation of demand and generation forecasts, and the error hotspot analysis has identified the key periods of operational stress (especially during evening transition periods). The resilience analysis also reveals that the combined forecasting–interpretation model can be used to support the identification of deficit‐prone periods and help to characterise the relationship between predictive uncertainty and microgrid balancing conditions, with the estimate of the net energy surplus of about 2500 kWh. In general, the research indicates that explainable artificial intelligence combined with multi‐target forecasting can be used to offer a helpful and practically relevant decision‐support layer to support transparent and resilience‐conscious microgrid operation.