Comparison of the Behavior Tree Concept with other Control Concepts used for Decentralized Energy Systems

Abstract

Behavior trees are a proven concept in the creation of complex task-switching control for robotic systems and non-player characters in the computer games industry. Requirements such as flexibility, maintainability, reusability of functionalities or expandability also apply to the control of decentralised energy systems. Despite this, there is a noticeable research gap regarding the application of behavior trees in that sector, which motivates the explorations of their potential within the scope of this work. Based on the definition and modelling of three exemplary concise energy systems, general behavior tree structures are created; implementation is carried out using the Python library py_trees. The advantage of modularity compared to the development of finite state machines known from robotics is verified for these systems. With a view to minimising annual operational performance indicators such as electricity price and emissions, we conclude that behavior trees have no fundamental advantages over other methods of decision making, such as decision trees, and that the comparison group of linear optimisations performed with the oemof-solph package can only be partially approximated. It is shown that the proposed basic structures allow specific relevant use cases, such as component blocking or peak reduction, to be realised.