The complexity of load flows in the distribution system requires self-learning algorithms.
The number of decentralized and fluctuating electricity producers, particularly photovoltaic installations, is rising. This is creating a growing challenge for distribution system stabilization. To provide better management of producers and consumers, creating grid stability and overcoming bottlenecks, artificial intelligence is becoming an attractive option that can help evaluate the flood of grid status data.
So far, it is just baby steps. Grid operators are testing artificial intelligence (AI) for various applications. The objective is to make the power system safer and more efficient, especially at the level of the distribution system. As decentralized power generation rises and load flows become more complex, this will become increasingly urgent.
Stadtwerk am See in Überlingen recently collaborated with HTWG Konstanz – University of Applied Sciences, Fraunhofer Institute for Solar Energy Systems (ISE) and the International Solar Energy Research Center Konstanz in a research project for the development and testing of prototypes of AI-based low voltage controllers.
“We had previously worked with algorithms without AI, with a similar intention,” says Jan Etzel, Head of Power Grid Operation at Stadtwerk am See. AI is now employed to optimize the system. According to Etzel, being able to use the novel algorithms was contingent on the extensive deployment of measurement technology within the low-voltage grid. The project used data from an industrial estate in Friedrichshafen to “design the power grid of the future and successfully simulate grid control,” says Etzel.
There is one limiting factor for AI, though: In many parts of the country – not just at Stadtwerk am See in the federal state of Baden-Württemberg – distribution system operators often know very little about the status of individual grid lines. This means that frequently, the measurement infrastructure must be expanded before sophisticated algorithms can be employed. AI depends on measurement data.
Once real time data is available, “the heart of a smart grid”, a smart control system based on AI, can be used. According to Stadtwerk am See, the control system accesses all relevant information from the low-voltage grid, such as up-to-date measurement data from transformer stations, consumers and producers. It also has data on annual consumption, weather, forecasts and much more at its disposal.
This allows AI to analyze the data sprawl in order to take quick and accurate decisions, explains Jan Etzel. Nowadays, load flows are too multidimensional for human operators to oversee – let alone control. This is where AI comes in, by flattening peak loads and preventing grid congestion, if necessary by curtailing generation. The fact that software learns from past events and mistakes means that, based on ever more detailed forecasts, it will be able to work more and more effectively to prevent critical situations in the grid.
The Smart Grids Platform Baden-Württemberg recently demonstrated the state of the art of AI for the electricity industry during the Smart Grids Dialog 2024 on “Artificial intelligence in grid operation” in Konstanz. Arno Ritzenthaler of Smart Grids Platform Baden-Württemberg stressed that AI is “more than a digital twin”. It can help stabilize the grids and reduce the need for grid expansion.
Professor Gunnar Schubert of HTWG Konstanz introduced the concept of “Digitainability”, a combination of digitalization and sustainability. Renewable sources of energy depend on intelligence and digitalization of the grid. Feeding 60 gigawatts of PV electricity into the medium or low-voltage grid will create major challenges, according to Schubert, who also sees this as an area of research into forecasting potential disruptions.
Manuela Linke, the leader of the AI4Grids research project at HTWG Konstanz, which was concluded at the end of 2023, explains how complex the power system of the future will be: There are expected to be 35 million electric vehicles on German roads by 2045, leading to the demand for electricity at least doubling. At the same time, the photovoltaics capacity will be expanded to more than 400 gigawatts, wind power to more than 300 gigawatts.
These fluctuations can only be managed with intelligent grid management involving controlling production and consumption as well as managing the local grid transformers. In some places, tap changers are installed to allow for variable turn ratios. This will prevent any overvoltage within a line from being transferred to the next voltage level.
This sounds simple in theory, but in practice it is often more difficult. Manuela Linke mentions the lack of grid data, such as valid information on operating status within the secondary substations. As soon as enough data and control options are available, AI can take on various tasks, such as load flow calculations based on the load forecast and the photovoltaic generation forecast. AI can then help to ensure grid stability and identify errors. It also permits grid planning based on time series. This means that it can predict where grid expansion will become necessary.
Manuela Linke explains that AI can be based on various algorithms. Some require additional training when the grid topology changes, some can be directly applied to a new topology. The main practical challenge for grid operation is not necessarily the quality of the algorithm, though. The quality of data collection and IT security can be more of a problem. After all, both IT and the electricity industry are affected by a lack of skilled workers, adds the researcher.
The processes in the electricity industry must be adapted to the changing grid situation, as Jann Binder of the Center for Solar Energy and Hydrogen Research Baden-Württemberg (ZSW) points out. Today, the electricity industry still applies the traditional standard load profile to residential customers. This is based on the assumption that all households have the same specific load profile over the course of a day. But households that use photovoltaic self-consumption or even photovoltaics in combination with battery storage, do not exhibit a standard load profile. “The currently used model produces serious errors,” says Jann Binder. This is why we need new load profiles, and these could be created with the help of AI.
The Fraunhofer Institute for Energy Economics and Energy System Technology (IEE) is conducting a research project on this topic. Standard load profiles are no longer able to cover a situation shaped by the rising number of photovoltaic installations, storage systems, heat pumps and EV chargers. “This is where our research project comes in: Our AI-supported processes generate the data basis needed for various optimization and forecasting tasks,” says Dominik Jost, project manager at IEE.
Karen auf der Horst, leader of the GenAI project at distribution system operator Netze BW, demonstrates that electricity companies can use AI for more than controlling the grid. She explains how Netze BW already uses AI to support the daily tasks of installation engineers. When they are in the field, they feed images documenting the installed technology into the system. Using image recognition, this provides valuable information. Based on standardized isolator switch colors, the insulating material used there can be automatically identified.
Photos of type labels enable the technical data to be automatically read out and stored in databases. While maintaining and repairing technical assets, workers can obtain the information they need from their own databases, which store AI-formatted, needs-based information. AI will also answer specific questions.
New documents generated in the field by installation engineers, such as photos or reports, are entered into the database directly. “We only use our own documents, which prevents AI from hallucinating,” says project manager Karen auf der Horst. Hallucinations are still a problem of generative AI – anyone using chatbots such as ChatGPT will have experienced this. Sometimes, the systems work on the basis of false information. Industries of systemic importance, such as power supply, have to avoid this at all cost.
