© Patrick Pollmeier / FH Bielefeld

Power generation fluctuates, so does consumption: solution through artificial intelligence and edge computing

More and more photovoltaic systems on rooftops, more and more e-cars plugged into power outlets: this puts a strain on the power grid due to fluctuations in generation and consumption. The international research project “AI4DG” initiated by Bielefeld University of Applied Sciences is now investigating how these fluctuations can be balanced out locally. The idea is to use distributed AI to reliably and autonomously control the power supply.

Bielefeld (fhb). On the roof of the Bielefeld University of Applied Sciences, they glisten and shine in the sun in such a way that you can literally watch the modules of the photovoltaic (PV) system generate electricity. “Unfortunately, the sun doesn’t always shine so beautifully,” says Katrin Schulte regretfully. With this, the specialist for power grids from the University of Applied Sciences (FH) Bielefeld names one of the two challenges facing grid operators in the expansion of decentralized renewable energy generation: Unlike conventional power plants, a PV system does not generate a constant and plannable output. “Weather-dependent or volatile power generation is what we call it,” Schulte explains. The power generated in this way is then fed into the low-voltage grid to which private households are also connected. However, these increasingly require a lot of power, for example when charging an e-car. “That,” Schulte said, “is the other big challenge!”

A decentralized AI to control the power grid.

Energy generation by many small plants and consumption in the low-voltage range will therefore increase in the future and at the same time fluctuate more strongly. “Controlling the network safely becomes a complex and increasingly difficult task due to the initial situation,” Katrin Schulte describes the problem. That’s why the 26-year-old has helped launch an international research project to work out a promising approach to the solution: distributed, or decentralized, artificial intelligence (AI) to control the power grid. Or as the project title says in English: “Artificial Intelligence on the edge for a secure and autonomous distribution grid control with a high share of renewable energies (AI4DG)”.

Universities of Bielefeld and Grenoble, Atos Worldgrid and Westfalen Weser Netz also on board

The scientists involved started work in October. Katrin Schulte is involved as a doctoral student for the UAS. The project builds on her previous research: After completing her bachelor’s degree in regenerative energies at Bielefeld UAS, the 26-year-old had studied smart energy systems in depth in her master’s degree in electrical engineering. Since the beginning of 2020, she has been conducting research as a scientific UAS employee in projects on shaping the energy transition at the Institute for Technical Energy Systems (ITES) in the Grids and Energy Systems (AGNES) working group under the direction of Prof. Dr.-Ing.

As initiator and project leader, Prof. Haubrock supervises AI4DG, in which partners with different focal points collaborate in a digitally networked manner. While the idea, initiative and power grid expertise come from the Bielefeld University of Applied Sciences, the Université Grenoble Alpes contributes the AI know-how. Bielefeld University is taking care of the already mentioned decentralized distributed use of AI, the so-called edge computing. Finally, industry partners Atos Worldgrid from France and the “Innovation – Intelligent Grid Technology” division of Westfalen Weser Netz from Germany are helping with implementation in the field.

Edge computing: Generation and consumption data are processed and analyzed decentrally

AI is already being researched today in approaches to controlling the power supply. But the Bielefeld research team and its French colleagues are going even further and linking AI with edge computing. “Edge computing means that data is not processed centrally in the cloud, but decentrally right where it is generated,” explains Timon Jungh. The biomechatronics engineer (MA) is conducting research in the working group of Prof. Dr.-Ing. Ulrich Rückert at CITEC (Center for Cognitive Interaction Technology) and is also completing his doctorate in the project. “This is a real innovation,” emphasizes Katrin Schulte. “We are increasing security and data protection with it, because power supply is part of the critical infrastructure and must be guaranteed.” If one decentralized AI unit fails, another can immediately take over. And if data doesn’t have to be sent, it can’t be lost or hacked along the way. And there’s another advantage to the decentralized approach: “The data can be processed locally with less delay,” says Timon Jungh.

Better forecasts can be made – supply becomes more reliable

On site – that’s in Herford, for example: A large gray box with wide vents stands on the side of the road in a residential area. A yellow sticker with a black lightning bolt is emblazoned on the side: a local network station. It transforms incoming medium voltage into low voltage, which is then used to supply the surrounding houses. Marco Sawatzki opens the station. Cables, switches, counters and a measuring field are revealed. Sawatzki plugs in his laptop. “Now you can see how much electricity is currently being consumed by households. Nothing seems to be fed into the grid at the moment,” explains the employee of Westfalen Weser Netz, the industry partner on the German side of the project.

AI4DG provides that exactly at this point, directly in the local substation, AI processes the measurement data. “We then used the AI analyses for forecasts, for example. They are supposed to predict the consumption of individual households or the power generated by PVs,” explains Katrin Schulte. Depending on this, the batteries in the households with PVs can then be controlled to serve the grid and the energy generated can either be stored or fed into the grid, so that there is always a constant voltage and the power supply is secured and overloads in the grid are avoided.

Pure transfer: from the lab, to the field, to the marketable product

Before that can happen, however, Schulte and her colleagues first develop and optimize their control system on the computer and in the laboratory: special software is used to set up an electricity grid as a simulation, while Westfalen Weser Netz supplies the real data for it. “This is our playground, this is where we try out our self-programmed algorithms,” says Katrin Schulte. If they pass the tests, they go to the Smart Energy Applications (SEAp) grid simulation lab. Real currents flow here, and the control system is validated using various hardware components such as battery storage and electronic loads. “Only then will we test our system in a field trial in the real power grid together with Westfalen Weser Netz. And if it works well, the industry partners can develop our system into a marketable product.”

Cooperation with France enriching – Europe-wide benefits sought

Possibly also for Europe-wide use: AI4DG is an international research project. It is funded by the German Federal Ministry of Education and Research (BMBF) as part of the program for Franco-German projects on artificial intelligence with a total of around one million euros (the UAS will receive around 200,000 euros of this). Prof. Haubrock established the contact with the French partners and is pleased about the international cooperation: “Through the cooperation with France, different distribution grid structures can be considered for a transferability of the AI methods to the European energy system. This will enrich the established, national AI strategies.”