Reinforcement learning based energy system optimization
Optimized operation is a crucial requirement for smart energy systems, where diverse technologies can interact in a complex way. Data driven approaches like reinforcement learning may thereby provide fast and reliable control strategies based on trained scenarios, without relying on direct forecasting techniques and pre-defined rules.
Target of this master’s thesis is to implement a reinforcement learning based controller on the application case of a virtual power plant for a sustainable ship energy system. The simplest design may consist of one or two engines and a battery, which have to supply a ship’s propulsion demand profile in the most energy efficient way. Based on an existing simulation framework for energy systems at LEC (ENERsim), a basic concept to couple a Reinforcement learning module shall be elaborated, and potentials and limitations of the technique shall be analyzed.
Target of this master’s thesis is to implement a reinforcement learning based controller on the application case of a virtual power plant for a sustainable ship energy system. The simplest design may consist of one or two engines and a battery, which have to supply a ship’s propulsion demand profile in the most energy efficient way. Based on an existing simulation framework for energy systems at LEC (ENERsim), a basic concept to couple a Reinforcement learning module shall be elaborated, and potentials and limitations of the technique shall be analyzed.