Back to search

ENERGIX-Stort program energi

AI-basert risikomodell for vegetasjon langs kraftlinjer

Alternative title: AI based risk model for vegetation along power lines

Awarded: NOK 4.4 mill.

Vegetation is the leading single cause of power outages. Climate change is expected to increase the frequency of power outages due to vegetation. The main objective of the project is therefore to develop an operational solution for local risk assessment of clearance issues that utilities can use for prioritizing and streamlining vegetation management near power lines and planning of emergency preparedness. The project has identified several research needs covering several topics, from forestry to computer science, from risk modelling to sensor technology. We have chosen to highlight risk modeling challenges, choice of optimal data sources and collection methods, use of artificial intelligence for local vegetation properties and how these data sources should be combined in a software. The hypothesis for the research project is that combination of existing data sources, such as digital terrain models, satellite-based forest maps and the like, combined with direct local measurements using drones and processing of these with artificial intelligence, will improve the precision and accuracy of risk models. The research needs emerge from the lack of knowledge of: - the use of artificial intelligence and image recognition for estimation of local vegetation-specific properties - how to best combine heterogeneous data sources to estimate the probability of clearance issues - how to combine probability and consequence of clearance issues into an operational risk model - how to develop architecture and computer models that can handle different and heterogeneous data sources for this purpose eSmart Systems AS, NIBIO and 5 utilities cooperate in this project. The project decided to use RGB sensors for 3D and 2D data collection. During summer 2020, data from more than 150 km of power lines, with possible clearance issues, has been collected. The data is now under processing and some initial analysis work has started on: - change analysis, using data from 2019 and 2020 to identify trees that has fallen during a season - treetop identification in 3D point clouds and link these to 2D images -development of deep learning models for identification of tree species and similar in 2D images The most likely data sources for the risk models are identified and the evaluation of the data source quality has started. During 2021 the project has - Finalized data collection and data processing for 2 seasons (about 150 km of power lines for each season) - From the data collected we have prepared large parts of the needed calibration dataset/validation dataset for our probability model of treefall. The full calibration dataset will be finalized Q1-2022 and we expect to have more than 200 trees in this dataset. - Developed a first version of an AI model that identifies the tree species from RGB images, with very good accuracy. Version 2 will be ready in Q1-2022 and will be trained on more data with the hope of improving accuracy even further. - Developed methods for deriving all needed insight about single trees and its neighbors/environment from point clouds and RGB images - Included external data such as soil data and snow load data in our probability models - Prepared a first dataset of about 50 000 trees that was run through a modified version of Forest Gales, adapted to Norwegian conditions (NIBIO). From the results we have developed a machine learning model that can predict the critical wind speed for each tree with very high accuracy compared to Forest Gales. The dataset used was based on more than 20 input variables. In Q1-2022 these models will be validated and calibrated using the calibration dataset as ground truth. - Started looking into how results of the final risk model should be presented visually to an end user The goal is to finalize the probability modelling, consequence modelling and data modelling during Q1/Q2-2022 and to implement these results into a visualization solution by Q3-2022. Testing of the solution will be done in Q4-2022.

Avbrudd i strømforsyningen har store samfunnsmessige konsekvenser og kostnader, og vegetasjon er den største enkeltårsaken til strømbrudd. Med klimaendringer forventes forekomsten av strømbrudd grunnet vegetasjon å øke. Hovedmålet med prosjektet er derfor å utvikle en operasjonell løsning for lokal risikovurdering av tresikkerhet som nettselskaper kan bruke til prioritering og effektivisering av skogrydding langs kraftlinjer og planlegging av beredskap. Ved å hensynta egenskaper ved enkelttrær og voksestedets egenskaper, kan risikotrær identifiseres, fremfor å gjennomføre tradisjonell og ressurskrevende breddehogst. Kombineres dette med avstand til kraftlinjene, nettspesifikk informasjon og ulike værscenarioer, vil nettselskapene få et bedre beslutningsgrunnlag for å planlegge vedlikehold av kraftlinjegatene. Løsningen vil bidra til effektivisering av nettdrift og sikrere strømleveranser. Prosjektet bidrar dermed til FNs bærekraftsmål nummer ni – å bygge robust infrastruktur. FoU utfordringene i prosjektet er tredelt: 1. Å utvikle metoder for å bestemme viktige egenskaper ved trær (tretype, høyde, diameter mm) langs kraftlinjer, basert på data innsamlet med droner og kunstig intelligens. Dette innebærer å utvikle nye metoder for å bestemme stabilitets-egenskaper i stående trær, egenskaper som i liten grad har vært i fokus i tidligere modeller. Prosjektet vil utforske bruk av dyp læring for å bestemme trevariabler. 2. Å utvikle en risikomodell for trefall hvor man sammenstiller kunnskap fra ulike fagområder (som kunstig intelligens, datavitenskap, skogbruksfag og statistikk), og kombinerer ulike datakilder (som lokale tresikkerhetsdata, terrengdata, løsmassedata, klima- og værdata og nettspesifikke NIS data). Sannsynlighet for trefall skal knyttes til konsekvens slik at resultatet blir en risikomodell. 3. Å utvikle arkitektur og datamodeller for håndtering av mange og heterogene datakilder til bruk i en prototype programvare.

Activity:

ENERGIX-Stort program energi