Key Technologies for Target-Driven Autonomous UAV Inspection in Power Grids Based on Reinforcement Learning

To address the challenges of low automation and poor adaptability to complex environments in power grid UAV inspections, this study analyzes the architecture and operational constraints of intelligent UAV inspection systems, clarifying optimization goals in terms of efficiency, safety, and intelligence. A closed-loop autonomous flight control method is proposed, integrating high-precision target recognition, dynamic path planning, and reinforcement learning–based optimization. Furthermore, a Transformer-based global obstacle perception and risk assessment approach is designed to construct a holistic “perception–decision–control” technical framework. Field experiments demonstrate that the proposed strategy significantly enhances flight path planning efficiency, obstacle recognition accuracy, and operational safety. The results provide scalable technical support for advancing the digital transformation of power grids and promoting the development of new-type power systems.