出版社：Academy & Industry Research Collaboration Center (AIRCC)
摘要：Vegetation is the major cause of overhead power line failures. According to a recent HydroQuebec analysis, more than 60% of the power outages are related to vegetation. Specifically, when branches/trees nearby the distribution network interact with extreme weather conditions, e.g., melting snow and heavy rain, they may bend and cause power outages. To ensure the reliability of our distribution network, millions of dollars are yearly spent for pruning trees and trimming branches. Aiming to reduce these costs, we recently adopted a new approach based on light detection and ranging (LiDAR) data. Indeed, we scanned 150 km of Hydro-Quebec’s network using a mobile LiDAR system. Through data analysis, we target automatic detection of hot spots, i.e., locations of threatening branches to distribution lines. However, such an operation cannot be accurately completed without a prior efficient detection of poles and lines locations, even for incomplete or missing LiDAR data. Hence, we propose here a low-complex and robust method for poles/distribution lines detection and lines modelling. Through customized filtering and detection, we efficiently detect poles and distribution lines with high accuracy and recall. Indeed, poles are detected with an accuracy of 94.5% and a recall of 89.7%, while lines are detected with an accuracy of 84% and a recall of 98.9%. Finally, our approach reconstructs power lines with a distance deviation from the real ones below 20 cm, in 89% of the cases. Such accuracy is required to automatically evaluate the closeness of vegetation to distribution lines and prevent power outages.
