Machine learning helps to predict blackouts caused by storms
Thunderstorms are common all over the world in summer. As well as spoiling afternoons in the park, lightning, rain and strong winds can damage power grids and cause electricity blackouts. It鈥檚 easy to tell when a storm is coming, but electricity companies want to be able to predict which ones have the potential to damage their infrastructure.
Machine learning 鈥 when computers find patterns in existing data which enable them to make predictions for new data 鈥 is ideal for predicting which storms might cause blackouts. Roope Tervo, a software architect at the Finnish Meteorological Institute (FMI) and PhD researcher at Aalto university in Professor Alex Jung鈥檚 research group has developed a machine learning approach to predict the severity of storms.
The first step of teaching the computer how to categorise the storms was by providing them with data from power-outages. Three Finnish energy companies, J盲rvi-Suomen Energia, Loiste S盲hkoverkko, and Imatra Seudun S盲hk枚nsiirto, who have power grids through storm-prone central Finland, provided data about the amount of power disruptions to their network. Storms were sorted into 4 classes. A class 0 storm didn鈥檛 knock out electricity to any power transformers. A class 1 storm cut-off up to 10% of transformers, a class 2 up to 50%, and a class 3 storm cut power to over 50% of the transformers.
The next step was taking the data from the storms that FMI had, and making it easy for the computer to understand. 鈥淲e used a new object-based approach to preparing the data, which was makes this work exciting鈥 said Roope. 鈥淪torms are made up of many elements that can indicate how damaging they can be: surface area, wind speed, temperature and pressure, to name a few. By grouping 16 different features of each storm, we were able to train the computer to recognize when storms will be damaging鈥.
The results were promising: the algorithm was very good at predicting which storms would be a class 0 and cause no damage, and which storms would be at least a class 3 and cause lots of damage. The researchers are adding more data for storms into the model to help improve the ability to tell class 1 and 2 storms apart from each other, to make the prediction tools even more useful to the energy companies.
鈥淥ur next step is to try and refine the model so it works for more weather than just summer storms,鈥 said Roope, 鈥渁s we all know, there can be big storms in winter in Finland, but they work differently to summer storms so we need different methods to predict their potential damage鈥
Paper link:
R. Tervo, J. Karjalainen and A. Jung, "Short-Term Prediction of Electricity Outages Caused by Convective Storms," in IEEE Transactions on Geoscience and Remote Sensing.
doi: 10.1109/TGRS.2019.2921809 URL:
Read more news
Decoding the chemistry of space with machine learning
Astronomers can detect complex chemical fingerprints聽in stardust聽鈥 but many of them remain unidentified. The聽SpaceML聽project combines machine learning and computational chemistry to simulate how molecules form and evolve in space, helping researchers decode these signals.
Catalysis in a new light: Microscale interactions could enhance clean energy technologies
A new study provides a more detailed view of how catalysts function during chemical reactions. The discovery could help develop more efficient materials for applications such as green hydrogen production and a more sustainable chemical industry.
Physics Days 2026 gathered Finnish physicists 黑料网
The 2026 edition of the annual conference featured talks on moir茅 matter, women in physics and paper cuts.