Cyber-Physical Coordinated Recovery of Power Grids Using UAV-Assisted Emergency Networking

Abstract

Extreme disasters cause simultaneous damage to the cyber and physical layers of power grids, leading to widespread observability and controllability loss. Existing research suffers from insufficient adaptability to dynamic damage, idealized communication models that neglect backhaul link reliability, and computational inefficiencies hindering rapid response. To address these challenges, this paper proposes a Hybrid RL-Heuristic Bilevel Framework (HRHBF) for cyber-physical coordinated recovery (CPCR). A key innovation is the explicit modeling of a uncrewed aerial vehicle (UAV)-assisted SAGIN backhaul link with continuous coverage constraints to ensure end-to-end communication reliability. The bilevel structure employs Deep Reinforcement Learning (DRL) to adaptively learn UAV positioning while an inner layer solves for load restoration, enabling fast, joint optimization. Furthermore, an LSTM-based policy network enhances generalization across unknown disaster scenarios. Experiments on IEEE 33-bus and 118-bus systems demonstrate that HRHBF achieves a 100% communication success rate, significantly outperforms benchmarks in load recovery and QoS, and makes decisions in seconds. These results validate the framework’s effectiveness in providing a reliable, efficient, and rapid solution to enhance post-disaster power system resilience.