Privacy-computing-based potential evaluation and model predictive control for user-side demand response

Résumé fourni par la source

User-side demand response (DR) is a key flexibility resource for renewable-energy integration, yet practical potential assessment requires private smart-meter, appliance, charging, and comfort information that should not be centralized. This paper proposes a privacy-computing-based framework for evaluating user-side DR potential and converting the evaluated potential into optimal regulation commands. A local feature encoder, differential-privacy gradient perturbation, and secure aggregation are combined to train a federated potential evaluator without uploading raw user profiles. The predicted response capacity, participation probability, and uncertainty interval are then embedded in a constrained model predictive control model for peak shaving, incentive allocation, and rebound suppression. The method is positioned for automated energy-service and charging facilities where advanced metering infrastructure can coexist with non-identifying optical queue or occupancy sensing. A 500-user, 96-interval numerical benchmark shows that the proposed method obtains a 6.1% potential-estimation MAPE and a 12.2% peak-load reduction under ε = 2 and δ = 10-5, while avoiding raw-data transmission and keeping the control discomfort score below 0.09. The results indicate that privacy computing can preserve operational accuracy when the potential model, privacy budget, and control constraints are co-designed rather than treated as independent modules.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.