UAV power line inspection strategy based on SAC algorithm

• Published in: Electric power systems research Vol. 248; p. 111925
• Main Authors: Xu, Cheng, Wang, Jiaxin, Ding, Yun, Zheng, Chunhou
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.11.2025
• Subjects: Deep reinforcement learning (DRL); Energy consumption optimization; Transmission lines; UAV inspection; Transmission lines; Deep reinforcement learning (DRL); UAV inspection; Energy consumption optimization
• ISSN: 0378-7796; 1873-2046
• DOI: 10.1016/j.epsr.2025.111925

With the evolution of the smart grid industry, transmission line inspection faces significant challenges. Traditional manual inspection methods suffer from low safety, poor accuracy, and high costs. In recent years, unmanned aerial vehicle (UAV) inspection technology has been widely adopted due to its advantages of low cost, ease of control, and strong scalability. However, the limited battery capacity of UAVs poses a critical challenge in efficiently completing inspection tasks under energy constraints. This paper comprehensively considers both flight and communication energy consumption, constructing a complete inspection environment model, and proposes a deep reinforcement learning (DRL)-based algorithm to minimize total energy consumption while completing inspection tasks. Experimental results show that the proposed algorithm outperforms traditional Deep Q-Network (DQN), Proximal Policy Optimization (PPO), and Deep Deterministic Policy Gradient (DDPG) algorithms on real-world maps and power grid topologies. •Development of a comprehensive energy consumption model for UAV-based power line inspection, which uniquely integrates both flight energy (based on steady state flight dynamics) and communication energy (considering data transmission requirements), forming the basis for a realistic inspection environment model.•Proposal of a SAC-based DRL algorithm specifically adapted for UAV power line inspection. This involves tailoring the SAC framework with a domain-specific reward function designed to balance inspection objectives with dual-energy minimization, and formulating the problem within a Partially Observable Markov Decision Process (POMDP) to handle environmental uncertainties.•Systematic validation of the proposed SAC-based approach using realistic UAV parameters (DJI Matrice 30) and power grid topologies. Experimental results demonstrate its superior performance in terms of total energy consumption optimization and inspection efficiency compared to other prominent DRL algorithms, namely DQN, DDPG, and PPO.•The significance of this work lies in addressing the critical energy constraints of UAVs by providing a robust DRL-based solution, thereby enhancing the feasibility and economic viability of automated power line inspections.