Abstract
Vegetation management programs are some of the largest budget items for utility companies in their efforts to prevent power outages. Our work focuses on developing techniques to streamline vegetation management operations through detection of at-risk locations in a power distribution network. We have trained deep convolutional neural networks to segment vegetation and estimate the risk of outage using high resolution aerial imagery. The output of these networks helps in the optimization of resource allocation and vegetation management planning. Experimental results on the Bay Area Synthetic Network demonstrate high accuracy in segmentation and risk prediction, thus minimizing costs for vegetation management programs.
•A pipeline involving image processing and DCNNs to detect vegetation near power lines in a distribution network.•A new methodology for predicting risk of outage due to vegetation falling on power lines.•A new methodology for resource allocation and route planning using DCNNs.•Customized clustering algorithms to handle large scale distribution networks with thousands of electrical nodes and power lines.
