DRO-based resilience enhancement of distribution networks considering thermostatically controlled load response prediction

To address the restoration scheduling challenges faced by distribution networks in the aftermath of extreme events, this paper proposes a distributionally robust optimization (DRO) strategy that incorporates temperature-responsive load prediction to enhance system resilience. On the demand-side management front, an XGBoost-based model is developed to predict the willingness of thermostatically controlled loads (TCLs) to participate in demand response (DR), leveraging historical response data and incentive signals. This enables the precise delineation of their adjustable operating boundaries. To capture the joint uncertainty of distributed renewable generation and load variability, a composite norm-based ambiguity set is introduced. Building on this, a bi-level DRO framework is formulated: the upper-level problem identifies the worst-case probability distribution, while the lower-level problem determines the optimal load restoration strategy. Case studies conducted on a CCHP-integrated IEEE 69-bus distribution system validate the effectiveness of the proposed method. The results demonstrate that the strategy enhances restoration performance, achieves a balanced trade-off between robustness and conservativeness, and significantly improves overall system resilience. Moreover, incorporating incentive-aware response prediction markedly increases user engagement and leads to superior recovery outcomes under extreme conditions, all while maintaining cost-effectiveness.