Without energy, the Netherlands comes to a standstill. Companies like Alliander develop and manage energy networks. Households and businesses receive electricity and gas through their cables and pipes. Alliander manages this for more than three million customers.
'Due to the energy transition, digitalisation, housing construction and economic growth, the electricity grid is becoming increasingly congested. In some places and times, too congested', explains Jacco Heres, data scientist at Alliander.
'We are working hard to expand the power grid, but in more and more places, the demand for electricity is growing faster than we can build', continues Jacco Heres. 'In those locations, we cannot provide additional capacity to industry, offices and supermarkets until the grid is expanded. Besides expansion, we are working on smarter and more efficient use of the power grid to prevent congestion.'
Project STORM
To prevent this congestion, better insight into all flows across the network is needed. This is where STORM helps, a collaborative project led by Roel Bouman, in which Alliander and the Department of Computing Science at Radboud University participate. Within the project team, Roel Bouman manages the project and provides technical oversight.
He graduated in Chemistry and Computing Science, specialising in data science and machine learning. 'To accelerate the project, we organised a hackathon on campus in November 2021', says Roel Bouman. 'The goal was for participants to generate ideas about using algorithms to automatically predict when switching events occur in the data.'
Automatic Filtering
'Of course, Alliander also tries to predict when and where congestion will occur through measurements. But measurement data always contains errors and irregularities. Moreover, "switching" takes place - alternative routes when a cable breaks somewhere, for example. This makes other routes busier.
If you want to measure the actual load and make predictions for optimal use, you need to filter out incorrect data. Preferably automatically', explains Roel Bouman, 'because if people do it, it takes a lot of time and you depend on their expertise and availability.'
Mathematical Model
'With machine learning, you are always operational and have current data at any moment.' Roel Bouman's team developed a mathematical model suitable for analysing data using algorithms. 'It's like solving a puzzle, a kind of super sudoku. Fifty percent doing and fifty percent thinking. We created a demonstrator where the data analysis happens entirely digitally. It works well and is now being implemented at Alliander by Jacco and his team.'
Is this THE solution? 'Yes', says Roel Bouman. 'Although updates will always be needed. Every system requires maintenance.'
