The expansion of renewable energy is fundamentally changing the demands placed on the electricity grid and its operators. Companies such as Amprion face the challenge of efficiently managing growing volumes of data and making them usable for operational decisions. Together with Materna, Amprion is trialling an AI assistant that makes heterogeneous data sources accessible via a dialogue-based interface.
Amprion operates one of Germany’s largest high-voltage transmission grids, stretching from the North Sea to the Swiss border, and forms a critical piece of infrastructure for security of supply in the western part of the country. As the energy system continues to transform – with large power stations being taken off the grid and wind power and hundreds of thousands of small solar installations being added – the dynamics of the system are increasing significantly.
For grid operators, this means, on the one hand, that they must invest in the physical expansion of the grid, but on the other hand, it also entails a significantly greater need for information. Analyses must be more robust, scenarios more flexible, and technical information must be available at all times. The quality and accessibility of grid asset data are thus becoming increasingly important from a strategic perspective.
The grid asset data domain at Amprion is responsible for the quality-assured provision of technical master data on all transmission grid assets – including lines, pylons, transformers and switchgear. This data forms the basis for maintenance measures, grid models and design decisions.
In practice, however, the relevant information is scattered across various IT systems. It is supplemented by technical documentation and manufacturer’s documentation, which are structured in a heterogeneous manner and, in some cases, are difficult to analyse automatically. Anyone wishing to answer a specific question therefore often has to manually consult multiple sources, collate the information and interpret it. As the volume of data grows, this process becomes increasingly time-consuming.
However, the rapid advancement of generative AI models and multi-agent architectures has opened up a new scope for solutions. Amprion, in collaboration with Materna, initially launched a proof of concept centred on a key question: Can existing data sets be utilised in such a way that technical knowledge becomes available more quickly, without compromising requirements for security, traceability and data quality?
This initial proof of concept evolved into a comprehensive GenAI strategy for the network asset domain.
Materna has established a structured evaluation process in which agents are systematically tested using defined test cases. The aim is to ensure consistent and reproducible response quality – for example, free from hallucinations – so that correct results are guaranteed. The evaluation is based on clearly defined criteria such as accuracy, completeness, contextual relevance and the specific use of tools.
The results are continuously monitored and analysed via dedicated dashboards. This enables deviations to be identified at an early stage, trends to be analysed and targeted optimisation measures to be devised, in order to maintain the quality and performance of agents at a consistently high level.
Core functions were developed and tested in collaboration with the product team. This demonstrated the potential of the approach: when used systematically, AI can help to improve data consistency in the long term and make manual verification processes more efficient.
The speed with which the first usable results were achieved is largely attributable to the chosen process model. As early as the first development sprint – around three weeks after the project began – a functional interface was ready, enabling business users to interactively query real system data for the first time.
This early usability proved to be a springboard for further developments: new ideas for applications emerged through direct interaction with the system, whilst at the same time potential for improvement was identified – insights that would only have come to light much later in a purely concept-driven approach.
The project’s progress was facilitated by the existing collaboration between the two companies: Materna had already developed the IT system for substation data at Amprion and therefore possessed a deep understanding of the relevant data models, interfaces and system structures.
