Vegetation is responsible for nearly half the outages on Hydro-Québec’s main grid—a rate that tends to increase with climate change. Indeed, the mounting intensity of weather events causes more trees and branches to fall on power lines. The government corporation has therefore joined forces with horoma AI, a young Québec start-up that specializes in artificial intelligence applications in vegetation management. Together, the partners are working toward the ambitious goal of reducing the number outages triggered by vegetation by 25%.

Maintaining over 100 000 kilometres of overhead power lines is no simple task. “We develop vegetation maintenance plans, but the vegetation grows faster than the tree pruners can work!” deplores Richard Chatigny, researcher at the Institut de recherche d’Hydro-Québec. To improve the efficiency of the pruning process, the experts must find a way to determine the locations in which the likelihood of vegetation interfering with power lines is highest.

horoma AI brings its expertise in artificial intelligence to the collaboration. Accustomed to working with aerial images, it had to adapt to a new type of data, since the proximity of residential homes prohibits overflying drones and the cost of flying over by plane is too limiting. To carry out the project, the team turned to ground-based LIDAR: a remote sensing system installed on the roof of a van that takes 360-degree 3D images of power lines and their environments.

Since September 2019, Hydro-Québec and horoma AI have been using the data to build an algorithm to optimize the work of maintenance teams by targeting priority lines and interventions. To accurately determine the vulnerability of each section of the grid, the team must compose with a wide range of criteria, including the size of the interfering branches and their position relative to the lines, as well as the different tree species.

It has been found that some species are more problematic than others. “Not all individuals break in the same way,” affirms Yvan Ouellet, cofounder and CEO of horoma AI. “A birch will bend without breaking but a balsam fir will snap because its physical structure is different. A branch of maple will crack and fall straight down.” The importance of accurate detection in risk calculation is therefore critical.

But what is the best way to distinguish tree species when the impact of the vegetation changes with the seasons and weather, especially when considering that the trees growing under power lines are pruned on a regular basis and therefore grow differently than they would in nature? The complex challenge is one on which the partners are focused on tackling together. “We currently have enough data to prove that the model works,” confirms Yvan Ouellet. “We’re confident that, with enough data, we’ll reach an excellent level of accuracy.”