TransnetBW: Mit KI durch den Vorgabendschungel. 51nodes entwickelt Knowledge Assistenten als Generative KI-Anwendung.

TransnetBW: Navigating regulatory jungle with AI

Published:

Category:

Uncategorized

Tags:

Energy, GenAI, Generative AI, News, Project

Exploring the benefits of generative AI through specific use cases instead of lengthy experimentation – transmission system operator TransnetBW knew from the outset how it wanted to enter the world of generative AI. Together with 51nodes, TransnetBW identified a specific Gen-AI use case that enabled TransnetBW to quickly get started with the technology and gain tangible experience.

The problem

Operators of electricity transmission grids must comply with numerous EU regulations and specifications issued by the Association of Transmission System Operators. For example for the exchange of electricity between different transmission grids. These regulations are difficult to understand even for experienced employees, and new colleagues find it difficult to navigate the jungle of regulations without the support of experts.

The solution

Together with 51nodes, TransnetBW has developed a Knowledge Assistant that provides system operations staff with quick and easy-to-understand answers to regulatory questions. “EU regulations and the associated specifications are confusing and often difficult to understand,” explains Pavel Zolotarev, Head of Energy and Balancing Markets at TransnetBW. “The Knowledge Assistant finds the passages relevant to the specific application in various legal texts and summarizes them in an easy-to-understand form.”

51nodes developed the Knowledge Assistant as a generative AI application using a retrieval-augmented generation (RAG) structure. Regulatory documents and specification documents for the system operation of transmission grids were broken down by 51nodes into related paragraphs and sections using a computer script. The sections were vectorized and made available as a pinecone vector database for a generative AI application.

The mathematical formulas in the documents to be processed were particularly challenging for the project team. Simple text extraction tools can recognize the characters in PDF documents, but they cannot reproduce the formulas. 51nodes therefore used powerful tools such as pandoc to convert PDF content into a purely text-based Markdown format, including LaTeX formulas. Formulas in this format are easily accessible and understandable for generative AI.

The generative AI application created using eXation’s “promptFusion” platform allows users to easily implement complex multi-step processes with multiple prompts and different language models.

TransnetBW successfully evaluated the Knowledge Assistant and is looking for further areas of application in which generative AI can support and accelerate workflows.