AI-Powered 3D Data Fusion System for Vegetation Risk Assessment in Distribution Networks

Vegetation is a leading cause of outages in distribution systems. Traditional vegetation management still relies on costly manual patrols and fixed trimming cycles. This paper develops an end-to-end, AI-driven framework to address vegetation-related outage risk in distribution networks by operating directly on three-dimensional light detection and ranging (LiDAR) point clouds. A RandLA-Net–based semantic segmentation model is trained to automatically separate vegetation, buildings, poles, and power lines. A canopy height model and watershed pipeline is then used to isolate individual trees and derive per-tree attributes such as height, crown radius, and minimum three-dimensional clearance to conductors and poles. On top of these geometric descriptors, we propose a tree-level risk-scoring framework that links the per-tree metrics to utility risk priorities. The framework is validated on an open airborne LiDAR dataset collected over a North American urban and suburban environment with vegetation patterns similar to many utility circuits. The resulting workflow integrates naturally with existing geographic information systems and vegetation-management processes, providing utilities with a scalable path from raw LiDAR data to prioritized, tree-level mitigation actions.