Scholarlink

<p>This study presents an autonomous full-coverage inspection route planning algorithm for rotary wing unmanned aerial vehicles (UAVs) in photovoltaic power plants. The method begins by normalizing environmental point cloud data from photogrammetry and lidar, aligning them via rotation and translation. A Gaussian mixture model is then applied to complete the global environmental modeling. The modeled environment is segmented into independent regions using spectral clustering. The entropy weight method determines target inspection areas, and threshold-based positioning identifies specific targets and obstacles. A spatial attention mechanism transforms the positioning data into two dimensions to calculate the required number of inspection routes. A distance formula corrects these routes to maintain a safe clearance between the UAV’s rotors and the photovoltaic modules, yielding preliminary planned paths. Reinforcement learning is employed for autonomous route planning, and the results are further optimized using a search optimization algorithm based on artificial potential fields. Simulation results demonstrate that the global environment modeling achieves elevation and plane errors under 0.4&#xa0;m, with target and obstacle positioning errors within 20&#xa0;mm. The planned routes successfully cover all target waypoints, effectively avoid obstacles, maintain a safe rotor-to-array distance, and increase the inspection coverage rate to over 96%.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Autonomous planning algorithm for the full coverage inspection route of rotary wing unmanned aerial vehicles for photovoltaic power plants

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年6月22日

查看全文
  • Chenglong Zhou,
  • Linwei Rao,
  • Xingming Xu,
  • Yu Zhang,
  • Yuejing Li

摘要

This study presents an autonomous full-coverage inspection route planning algorithm for rotary wing unmanned aerial vehicles (UAVs) in photovoltaic power plants. The method begins by normalizing environmental point cloud data from photogrammetry and lidar, aligning them via rotation and translation. A Gaussian mixture model is then applied to complete the global environmental modeling. The modeled environment is segmented into independent regions using spectral clustering. The entropy weight method determines target inspection areas, and threshold-based positioning identifies specific targets and obstacles. A spatial attention mechanism transforms the positioning data into two dimensions to calculate the required number of inspection routes. A distance formula corrects these routes to maintain a safe clearance between the UAV’s rotors and the photovoltaic modules, yielding preliminary planned paths. Reinforcement learning is employed for autonomous route planning, and the results are further optimized using a search optimization algorithm based on artificial potential fields. Simulation results demonstrate that the global environment modeling achieves elevation and plane errors under 0.4 m, with target and obstacle positioning errors within 20 mm. The planned routes successfully cover all target waypoints, effectively avoid obstacles, maintain a safe rotor-to-array distance, and increase the inspection coverage rate to over 96%.