The integration of distributed and supply-dependent primary energies represents a major challenge for the transformation of energy systems. Modern artificial intelligence and machine learning technologies make a significant contribution to meeting this challenge in many areas: in the semi-automatic management of electricity grids, in the insight-driven marketing of decentralised energy systems, in the forecasting of load and generation time series, to name but a few. The close linking of energy systems and ICT infrastructure in smart grids also requires an adaptive and autonomous "immune system" in order to deal with attacks against the infrastructure and failures of subsystems.
The Power Systems Intelligence Group therefore researches and develops solutions at the interface between the power grid, the energy market, artificial intelligence and a cyber-resilient system understanding. Our vision: the AI-empowered Smart Grid.
Every autonomous, pro-active system needs a model of its environment and a prediction of future system states. This can be the active power input of a wind or photovoltaic park, the demand in energetic neighbourhoods, the available flexibility for the provision of system services or the price development of the energy market. The digitalization of the energy supply opens up data volumes that represent a suitable basis for machine learning. In order to draw meaningful conclusions from these large amounts of data and to train systems for autonomous operation, approaches from deep learning have established themselves as a worthwhile research subject with impressive results. Structures such as deep recurrent networks are suitable for predicting time series or market behaviour. They also allow any systems to be simulated and can independently model complex issues using reinforcement learning. This technique, known as a surrogate model, makes it possible to simulate connections in the power grid, derive and correct sensor values or determine operating parameters for network management such as reactive power compensation factors almost in real time.
Our research in the field of machine learning therefore aims to contribute domain knowledge about the power grid and the energy market in the form of architectures for artificial neural networks to artificial intelligence and to make use of the methods of deep learning to enable prototypes for cyber-resilient network management and economic, intelligent action in energy markets. We work closely with the Computational Intelligence Department at the Carl von Ossietzky University of Oldenburg and play a key role in shaping the Deep Learning Competence Cluster.
Increasing decentralisation of energy supply is leading to increasing decentralisation of control and monitoring intelligence. We take the spatial and topological distribution of system components into account by developing distributed, (partially) autonomous agent systems. We work closely with the Department of Energy Informatics at Leibniz Universität Hannover and the OFFIS group Simulation and Agents in Multiple Domains. Our focus is on upgrading individual agents, who learn operating and trading strategies independently and bring the flexibility of the components they represent optimally into the overall system. The issues of network stability and the provision of regional system services also play a key role in this context. Learning agents turn decentralised energy systems into valuable, pro-active assets for distribution network operation. This contributes to a reliable and efficient energy supply and contributes to mastering the increasing complexity of the overall system. Furthermore, the detection of misconduct and other anomalies in power and communication networks is another part of our work.
Our research in the field of distributed and learning systems therefore aims at learning network and system characteristics and AI-supported analysis of possible attack scenarios. The unique research field of Adversarial Resilience Learning enables us to make the power grid a self-adapting, secure and cyber-resilient overall system, even in the case of strong digitalization. This goal connects us with the Automation, Communication and Control group, which researches the stable and reliable operation of dynamic systems.
Sonnenschein, Michael and Lünsdorf, Ontje and Bremer, Jörg and Tröschel, Martin; Environmental Impact Assessment Review ; 4 / 2015
Nieße, Astrid and Tröschel, Martin and Sonnenschein, Michael; Environmental Modelling & Software; 2013
Lehnhoff, Sebastian and Sonnenschein, Michael and Tröschel, Martin; Konferenz für Nachhaltige Energieversorgung und Integration von Speichern - NEIS 2013 - Tagungsband; 2013
Wissing, Carsten and Tröschel, Martin and Nieße, Astrid; 8. Internationale Konferenz und Ausstellung zur Speicherung Erneuerbarer Energien; 11 / 2013
Sonnenschein, Michael and Tröschel, Martin and Lünsdorf, Ontje; Environmental Informatics and Renewable Energies - 27th International Conference on Informatics for Environmental Protection; 2013
Sonnenschein, M. and Appelrath, H.-J. and Hofmann, L. and Kurrat, M. and Lehnhoff, S. and Mayer, C. and Mertens, A. and Uslar, M. and Nieße, A. and Tröschel, M.; Tagungsband VDE-Kongress 2012; 11 / 2012
Mayer, Christoph; Tröschel, Martin; Uslar, Mathias;; 2012
Nieße, Astrid and Lehnhoff, Sebastian and Tröschel, Martin and Uslar, Mathias and Wissing, Carsten and Appelrath, H.-Jürgen and Sonnenschein, Michael ; Proceedings of IEEE COMPENG2012; 6 / 2012