A Coordinated Active and Reactive Power Optimization Method for Integrated Energy System Based on an XGBoost Surrogate Model for Voltage Stability Margin

Conventional integrated energy system (IES) dispatch generally enforces bus-voltage limits but does not quantify proximity to the static voltage-instability boundary, while repeated physics-based stability calculations are computationally expensive for multi-period optimization. This study proposes a coordinated active- and reactive-power dispatch framework for an electricity–heat IES. The minimum singular value of the reduced power-flow Jacobian is used as a physics-based static voltage-stability margin. Monte Carlo operating states are labeled through AC power-flow and Jacobian calculations and used to train an XGBoost surrogate that maps loads, renewable output, electricity–heat coupling-device power, storage states, and reactive-power controls to the margin. SHAP analysis identifies influential buses and operating variables and supports feature reweighting during surrogate training. The predicted margin is converted into a risk penalty and optimized jointly with electricity-purchase, renewable-curtailment, and equipment operating costs. The framework coordinates active-power reshaping by heat pumps, electric boilers, storage, and flexible loads with reactive-power support from SVCs, SVGs, and converter-interfaced resources. The resulting schedule is finally verified by AC power-flow and reduced-Jacobian margin calculations. This approach makes a physics-based stability index practical for preventive, security-aware IES dispatch while retaining model interpretability and physical verification.