Hydrogen is a highly versatile element and is considered to be the missing link in the energy transition. However, its production requires large amounts of energy. For this reason, excess power from solar and wind power plants is intended to be used predominantly for electrolysis.

To this end, power-to-gas plants for producing hydrogen are combined with photovoltaic and wind power plants and storage systems. For the industrial use of hydrogen, gas turbines can be used to cover the power and heat requirements of companies flexibly and according to need.

It is now important to make optimum use of all three segments. The Cognition²H2Force project investigates the application of deep reinforcement learning for this purpose. The aim is to develop an environment that allows for digital mapping of the plant portfolio, including the relevant physical and economic parameters, in which optimization algorithms are trained to recognize these relationships and independently optimize short-term power plant deployment planning.

The project is interesting to:

Energy managers, utility companies, energy suppliers, WPP operators, direct marketers, aggregators


  • Developing a training environment for self-learning energy management algorithms
  • Using deep reinforcement learning algorithms for deployment planning and control of electrolyzers
  • Taking short-term spot markets into account
  • Comparative analysis of the results of the AI-based energy management algorithms with a rule-based system as well as established conventional dynamic programming methods
  • Proof of concept in the simulation environment


  • Deep reinforcement learning
  • Proximal policy optimization (PPO) algorithms
  • Developing an environment in Python for mapping the plant portfolio and the relationships between the physical and economic parameters


Project partners

Research Field Energy Informatics, Fraunhofer IEE

The Energy Informatics research field of Fraunhofer IEE investigates the question of how information technology can support processes of energy economics and energy system technology in such a way that the energy system also works with a high proportion of distributed power generation. Alongside other research subjects, IT architectures, the automation of entire process chains, the design of resilient digital energy systems, and the potential of cognitive energy economics are examined to answer this question. Most recent research subjects include investigating the efficiency of artificial intelligence methods and machine learning algorithms in operational energy management systems. The Cognition²H2Force project focuses on the development of energy management systems that employ deep reinforcement learning algorithms, enabling them to learn physical and economic relationships of the distributed plant portfolio so that they can subsequently make independent decisions on the coordinated control of the plants.

Project participants: Alexander Dreher, Malte Lehna, Jasmin Pfeffer, Cyriana Roelofs, Jonathan Schütt, Wolfgang Slaby, Christoph Scholz

Institute of Power Plant Technology, Steam and Gas Turbines (IKDG), RWTH Aachen University

The subjects of the Cognition²H2Force projects were developed jointly with the Process Analysis and Systems group of the IKDG of RWTH Aachen University. The institute has many years of experience in the field of model-based analysis and optimization of energy transformation processes in power plants, placing a particular focus on the further development of hydrogen-fueled gas turbines.

Project participants: Thomas Bexten, Nils Petersen, Tobias Sieker

Project period

October 2020 – April 2021

Alexander Dreher


Fraunhofer IEE

+49 (0) 561 7294-1750

Share this Spotlight with your network.

HeatCast - Reducing CO2 emissions through efficient forecasts
Intelligent partial discharge evaluation of generator insulators using machine learning methods (SmartPD)