Leakage-Safe K-Means-Assisted Multi-Objective Energy Management for a Solar and Wind–Battery Micro-grid

This study evaluates a leakage-safe, K-Means-assisted multi-criteria energy management framework for a solar-wind-battery micro-grid. Random Forest provides one-hour-ahead load predictions, while design-only K-Means clustering identifies operating regimes for battery and generator dispatch. A chronological design-validation-blind-test protocol separates model development, controller feasibility screening, and final evaluation. Random Forest achieved a blind-test RMSE of 3.343 kW, MAE of 2.315 kW, and R² of 0.840, corresponding to 45.68% RMSE skill relative to persistence. On the blind test, the reported regime-adaptive candidate achieved zero unserved energy, reduced generator energy by 59.34% and renewable curtailment by 23.85% relative to prediction-only control, and reduced battery throughput by 37.49% relative to reactive control. These gains involved explicit trade-offs in generator use and battery utilization. However, the validation grid was non-discriminative: all 81 feasible candidates produced identical objective values and normalized scores. The study therefore reports the candidate as a feasibility-screened representative policy, not as a uniquely optimized solution. The results demonstrate the operational trade-offs obtainable within the stated simulator while identifying controller-parameter sensitivity as a necessary next methodological step.