Reducing Power Failures in Off-Grid Renewable Microgrids using Risk-Aware Energy Management

Abstract

Off-grid renewable microgrids can lose power when demand exceeds weather-driven generation and stored energy. Simple energy management often serves flexible loads too early. It then leaves little reserve for later critical demand. This paper presents a risk-aware energy management system that protects essential loads against uncertain 24-hour net demand. The method uses quantile gradient boosting, conformal interval calibration, and prequential bias correction. Monte Carlo paths convert forecast uncertainty into a conditional value-at-risk storage reserve. Load is divided into critical, flexible, and noncritical classes. Flexible demand may move by six hours. The experiment uses 10,356 public hourly records from the Rye microgrid in Norway. Training contains 9,515 hours, while testing covers 841 unseen hours. The measured system includes photovoltaic, wind, battery, and hydrogen equipment. Risk-aware control is compared with greedy, fixed-reserve, deterministic, and perfect-forecast controllers. The probabilistic forecast reaches 22.18 kW MAE and 80.02 percent interval coverage. Risk-aware control cuts critical outage hours from 422 to 346 against greedy control. This change equals 18.01 percent. Critical energy not served falls by 16.16 percent. The method finishes within two outage hours of the oracle controller. Its total load service remains 58.70 percent because winter resources are inadequate. Stress tests show similar gains under battery derating. Results show that risk-aware scheduling protects critical demand during scarcity. However, energy management cannot replace sufficient generation and storage capacity.