Our contributions to AI-based control of power grids and other critical infrastructure — peer-reviewed papers, Horizon Europe, Digital Europe and FFG projects, and results from open competitions.
Castagna, Zahlner and Fuxjäger (EnliteAI) with SBB and the Flatland Association: a semi-hierarchical multi-agent RL framework that separates dispatching decisions from routing under disruption — nearly doubling successful train arrivals while keeping deadlock rates below 5% against monolithic RL and heuristic baselines. The same Maze stack, on a second critical network. Part of AI4REALNET. arXiv 2605.10257.
Seventeen authors from the AI4REALNET consortium — among them TenneT, RTE, SBB, IRT SystemX, INESC TEC, TU Delft, Fraunhofer IEE, the Flatland Association and EnliteAI — propose an interdisciplinary framework for human–AI decision systems in safety-critical infrastructure: transparency, trust, explainability and robust decision-making treated as one design problem. DOI 10.1109/SMC58881.2025.11342626.
How operators and AI recommendations actually work together in critical-infrastructure settings — the research basis for FLEX's human-in-the-loop design.
A scalable two-phase approach balancing congestion-cost reduction against operational complexity, evaluated on real TenneT grid data — day-ahead plans in 4–7 minutes.
The methodological groundwork behind the current platform generation.
Presented at the workshop dedicated to exactly our problem: making reinforcement learning survive contact with reality.
A Horizon Europe project combining emerging AI algorithms, open-source AI-friendly digital environments and human–machine interaction design to improve the real-time and predictive operation of critical network infrastructures — electricity grids, railways and air traffic management. Its goal: next-generation decision-making methods that ensure trustworthy AI-assisted human control, with resilience, safety and security as core requirements.
A Horizon Europe project establishing a European AI Testing and Experimentation Facility (TEF) for the energy sector — helping develop, test and validate AI solutions across multi-energy systems, congestion management, energy efficiency and DER integration. Real-world implementations run in Denmark, the Netherlands, Portugal and Germany, each tackling a distinct challenge such as grid congestion or local renewable integration.
An EU-funded project establishing the Common European Energy Data Space (CEEDS) for secure, interoperable and federated energy data exchange across Europe — integrating national and regional platforms to support grid management, energy efficiency and renewable integration. The 54-partner consortium runs until March 2028, advancing decarbonised and decentralised energy markets through 12 use cases.
An FFG exploratory project investigating how AI methods can help distribution system operators manage the growing share of decentralised renewables. Unlike transmission grids, distribution networks typically lack digital data on grid expansion and operation — so the project surveyed framework conditions and data availability with up to six DSOs, then identified, analysed and proof-of-concept tested AI methods for PV generation forecasting and distribution-grid planning and operation. Results feed into the follow-on AI-State-Estimation project.
An Austrian research project advancing AI-supported grid state estimation for the Stadtwerke Kapfenberg distribution network. Building on the AI4GriDs pilot, it combines physics-guided neural networks with reinforcement learning to deliver more precise, real-time visibility into grid conditions and to recommend adaptive control measures — targeting detection of at least 90% of critical grid states with under 10% false alarms by project end.
Austria’s highest award for innovative companies, won in 2022 for “Power Grid 4.0”: reinforcement learning that cuts congestion-management and redispatch costs while lowering CO₂ emissions in European grid operation. Winner in the Energy & Sustainability category.
First place in the international “Learning to Run a Power Network” competition — the benchmark contest for RL-based grid control, initiated by RTE. Won with Maze; the winning code is public.
Third place in the “L2RPN with Trust” competition, organised by RTE with ICAPS 2021 — the first time we put Maze to the test against the international field, one year before winning.
Selected for the Top 30 out of 1,084 applications to the 2024 cohort of Free Electrons, the global energy startup programme run by leading utilities including EDP, E.ON, ESB and CLP — the programme’s bootcamp stage, where startups and utilities scope pilots together.
Second place in the innovation category of Vienna’s Mercur 2022 for “Power Grid 4.0 — AI-based control and optimisation of the power grid to increase resilience and reduce costs”, out of 95 submissions — and with it the nomination for the Austrian State Award for Innovation, which we went on to win.
The methods in these papers are the same ones the proof of concept runs on your grid data.
