OPERATIONAL RISK QUANTIFICATION FOR PREVENTIVE CONTROL IN LOW-VOLTAGE GRIDS WITH DISTRIBUTED ENERGY RESOURCES

Kurzfassung

This work presents a two-phase, risk-aware framework for preventive control in low-voltage distribution grids with high photovoltaic (PV) and electric-vehicle (EV) penetration. The approach separates offline risk quantification from online control: Monte Carlo simulations are used offline to identify critical operating conditions and train a calibrated machine-learning surrogate, while the online controller uses predicted violation risk to trigger preventive actions without requiring real-time sampling.

The framework was evaluated on a 49-bus German residential low-voltage grid under PV-only, EV-only, and combined operation, including biased and noisy PV forecasts. Results show that a small number of high-stress weeks accounts for a substantial share of annual violations, and that ML-based risk triggering can reduce remaining voltage and line-loading violations, especially during severe PV forecast stress. However, performance depends strongly on the operating and forecast regime, and earlier triggering alone does not always improve control outcomes. Overall, the offline–online separation enables computationally efficient preventive grid control while explicitly accounting for uncertainty.