Data-driven Analysis of the Energy System
The new energy system brings with it a new complexity: instead of a few large, centralized power plants, there will be thousands of small, decentralized, mostly weather-dependent plants for renewable energy generation. In addition, new sectors such as heating and mobility, which were previously largely based on fossil fuels, will be largely powered by electricity. This development paves the way for a climate-neutral energy system, but requires greater coordination between energy generation and use.
Digital solutions are needed to master this increasing complexity and efficiently shape the transformation of the energy system: Smart metering systems, communication technologies, databases, data rooms and data analysis, artificial intelligence and cyber security.
Benjamin Schäfer's research group DRACOS at KIT investigates complex energy systems based on extensive data (big data). The research focuses on the development of new transparent AI models which, in contrast to so-called black box models, not only show the result, but also explain which factors influence the result. Benjamin Schäfer describes his research in the following profile.
What is researched?
We research complex energy systems, from the electricity grid and individual renewable energy systems to individual buildings. Complementary to model-based research, we dedicate ourselves to the increasingly comprehensive data of the energy system: from smart meters in households to data on electricity prices on the European markets. We use publicly available data, industry data and self-recorded data, e.g. from our research campus or from measuring equipment that we have installed in Mallorca, South Africa and Korea, among other places.
How is it researched?
We use the entire portfolio of “data science” and “artificial intelligence” (AI), combined with domain knowledge and existing models. Specifically, this means that we look at the frequency distributions of data in order to estimate how often extreme events (large deviations from the stable state, extreme weather...) occur. At the same time, we create machine learning models, e.g. to make predictions: How will the electricity consumption of a house develop over the next few hours? But also: How will the grid frequency, an indicator of the balance in the large-scale electricity grid, behave over the next few minutes and hours?
An important building block here is “explainable artificial intelligence”: this allows us to make opaque black boxes transparent and explainable. Many complex AI tools, such as deep learning, only show us the result, but not how they arrived at this result. With explainable AI, we learn which input data influenced the predictions and how. We then compare our explanations and data-based predictions with classic model-based approaches and examine where the two approaches differ. In this way, we aim to improve the model-based approaches.
What is te aim of the project?
Our overarching goal is to better describe the energy system with data and make it easier to understand. To this end, we are working on developing suitable AI methods for the energy system and making them transparent through explainable intelligence. This is the only way they will be used in practice. On the other hand, we are also trying to improve the model-based approaches with the interplay of data-based and model-based approaches.
What is the benefit for society?
We need a functioning, reliable and cost-effective energy system for our economy and society. Our work supports this in various ways:
Better forecasting models make it easier to coordinate electricity consumption and generation so that there are fewer bottlenecks and fewer expensive reserves, such as gas-fired power plants, need to be activated. However, such forecasts are not accepted by operators as a “black box”, but only as a transparent solution.
AI models and explanations are also important at the consumer level: for example, if we optimize the charging behavior of batteries (in the household or in electric cars), acceptance increases if we also provide an explanation as to why a certain behavior is used.