Entscheidungsfindung unter Unsicherheit durch Reinforcement Learning am Beispiel einer Microgrid-Steuerung

Translated title of the contribution: Decision making under uncertainty using reinforcement learning for a microgrid control system

Katharina Zeilinger

Research output: ThesisMaster's Thesis


Well-founded decision making requires knowledge about the decision situation and the potential consequences of alternative choices. If the information relevant to the decision is partially unknown, we call this decision making under uncertainty. Complex problems and uncertainties increasingly occur in strategic decision making. This leads to the question of how good decisions can be made in the presence of uncertainty about the relevant factors. The purpose of this master thesis is to evaluate whether reinforcement learning is suitable for decision making under uncertainty in a sequential decision problem. The investigated problem consideres decision making for the control of a diesel generator in a self-sufficient, hybrid microgrid. For this, a decision about whether the diesel generator should be switched on or not must be made at each time step. The aim is to reliably cover the energy demand of the consumers and to minimize the operating costs within the microgrid over a given period. In order to learn a decision strategy to control the diesel generator, a policy gradient algorithm is used. For the evaluation, the quality of the learned decision strategy is compared to the quality of a trivial strategy for different test cases under different circumstances. The operating costs of the microgrid are used as a quality criterion. The results of the evaluation show that the decision strategy learned through reinforcement learning is well suited for ensuring energy supply.
Translated title of the contributionDecision making under uncertainty using reinforcement learning for a microgrid control system
Original languageGerman
Awarding Institution
  • Montanuniversität
  • Auer, Peter, Supervisor (internal)
  • Ortner, Ronald, Co-Supervisor (internal)
Award date8 Apr 2022
Publication statusPublished - 2022

Bibliographical note

no embargo


  • decision making under uncertainty
  • reinforcement learning
  • microgrid
  • policy gradient

Cite this