Method development

In-depth methodological competence is required to deal with thematic challenges and to develop and implement innovative concepts. In addition to basic skills in data analysis (e.g. statistics), this also includes handling and processing large amounts of data and visualizing the results. In addition to model-based simulation (coupling of simulation elements known), data-driven methods (showing correlations between input variables and results) are increasingly becoming the focus of method development. Both approaches have their advantages and disadvantages and must be fundamentally understood in order to develop relevant solutions for the future air traffic system. Since the existing methodologies are also constantly being developed, the application-oriented concepts must also be continuously adapted to these new findings.


The research field is divided into three major areas: Building, coupling, and simulation of agent-based models, derivation of complex correlations in the air traffic system by machine learning (classification, regression, clustering), and digitalization and virtualization of workflows and workplaces.

Agent-based models

Agent-based models represent systems by specifying the properties of different participants and their interactions with each other and with the system in detail. Given the extensive experience with the system and knowledge of significant interactions, new insights can be acquired through simulation by varying the input parameters and transferred to real-world applications. In this context, an agent is an element within the simulation environment with the ability to make independent decisions. Agent-based models can generate new, previously not described (emergent) behaviors/ structures due to the complex interactions among agents. This could generate new insights about the considered system. For our research, we are developing environments for modeling airport and passenger processes (Airport in a Lab), and for simulating aircraft trajectories as well as a basis for optimization approaches.

Maschine learning

Data-driven methods are used to derive new insights into the correlations from observations of the considered system. However, these correlations (pattern recognition) are not necessarily causal, i.e. they do not provide a link between causes and effects (patterns). With the necessary domain expertise, however, these correlations could be translated into causal relationships. Machine learning is divided into the following approaches: classification (classifying data into categories - supervised learning), regression (determining a functionally quantitative relationship between input and output variables - supervised learning), and clustering (grouping unlabeled data through similarity analysis - unsupervised learning). Reinforcement learning is another approach in which agents independently learn a strategy to maximize utility. No strategies are specified, but certain system states are scored positively or negatively. For research in the field of the overall air traffic system, appropriate machine learning models are developed for the specific problem, e.g. analysis of flight tracks in the vicinity of airports to identify beneficial factors for increased efficiency or learning conflict resolution strategies for controller support.

Digitalization and virtualization

By using simulations, large amounts of data and full-state information are available, which is limited in its ability to be recorded and processed in the real world. However, current and future technologies will enable more comprehensive access to data. Operational systems must now to able to process this data appropriately, derive information, and develop efficient guidance for action. Digitalization not only means computer-aided data processing, but also the development of new operational concepts. With these new concepts, the common demand for "more and more data" becomes obsolete and the demand for the "right data" returns to the focus.