AI to Improve Data Collection and Address Data Scarcity in the Energy System – AI Data

Project description

The goal of the AI Data project is to develop and validate AI approaches for utilizing various types of data. The RLI researchers are testing methods for working with structured and unstructured data as well as images to obtain reproducible results. At the same time, they are evaluating the methods based on accuracy, potential applications, and time savings.

AI Methods: Agents and Workflows

The AI methods under consideration include self-developed Claude Skills as well as agent-based workflows. In a data pipeline, for example, one agent can be used to parse a PDF file, while another agent validates the results. The researchers implement these workflows with varying levels of upstream data preparation and with human oversight based on the “human-in-the-loop” principle.

Processing Open Datasets

The researchers use open datasets that are relevant to energy research as test data. For structured data, they access the Core Energy Market Data Register directly via the API and as a local database via the RLI tool Open MaStR. In doing so, they use methods such as MCP and AI-powered SQL commands.

The researchers convert unstructured data – published as PDFs and containing text, graphics, and tables – into Markdown files. These can then be more easily parsed by Large Language Models (LLMs) and other AI systems. Tools are being tested for this purpose. As part of the project, the project team has transformed the grid expansion plans from the distribution network operators and the BTB’s transformer plan for Berlin-Adlershof in this manner. In the next step, the grid expansion plans will be made available in a Retrieval-Augmented Generation (RAG) system, which LLMs can use to search through the plans.

In the field of image files, the goal is to derive GIS data from map images. The researchers are testing this using grid expansion plans from distribution network operators. They are also examining whether input data can be prepared for use in RLI simulation tools, such as SpiceEV and oemof.

Using and Sharing Results Sustainably

The methods developed and tested are intended to be usable over the long term. The researchers are comparing different methods for documenting these routines and sharing them in a simple and structured way.

Project Period: May 2026 – December 2026