Background
In control theory, traditional model-based approaches like Model Predictive Control (MPC) have been the cornerstone for designing and implementing efficient control systems and are at the forefront of research. These approaches typically rely on precise mathematical models that describe the dynamics of the system. As systems grow more complex and more data becomes available, the research world in control theory increasingly opens up towards data-driven control methods. These new approaches are especially useful in situations where creating a precise model is difficult or where the system is too complex. They also come with the promise of being more adaptable, as they are based on data that can be updated accordingly. Furthermore, many of these methods are based on the highly researched MPC approach, inheriting their stability and robustness guarantees along with other advantages.
In recent years, many theoretical concepts for data-driven control theory have been published, leaving numerous open research questions. As with every newly discovered approach, they come with a few limitations:
Computational complexity
Limited interpretability
High data quality and quantity requirements
Sensitivity to unseen events
Nevertheless, these approaches offer some possibilities. First, theoretical approaches need to be applied to specific use cases to demonstrate their practicality and effectiveness and compared to existing approaches (like MPC), which would be the main task of the master thesis. Second, when applying to specific use cases, some of the limitations of data-driven control theories can be addressed, which would be the cherry on top of the master thesis.
Control approaches:
Below is a list of data-driven control approaches of our special interest. If you have any other relevant control approaches that you are interested in, please feel free to reach out to us.
Koopman Operator Control
Sindy-MPC (or MPC based on other Physics Informed Machine Learning approaches)
MPC with Gaussian Processes
Learning MPC
Intelligent reference planning approaches
DeePC and SPC
(All the approaches can be found in more detail at the link provided below)
Use-cases:
The use-cases for developing the control approaches are based on projects from MATLAB's Sustainability and Renewable Energy Challenge. This means the foundations and basic models are already provided and built to be adapted for further work. Furthermore, these projects come with background material and a basic project plan, which could serve as a great starting point for you to get the project running. This provides a well-structured master thesis with a solid foundation from your first day.
The preferred projects are mainly coming from the energy sector:
As the projects are part of one of MATLAB’s Challenges, you are also allowed to compete with your solutions and findings at the end. If you are interested in any other projects from the Sustainability and Renewable Energy Challenge or in a robotics project with a similar structure, feel free to reach out to us, and we will see if we can match it to our interests and expertise.
Links and Documents:
All important links and documents related to the use cases and control approaches can be found here: Master thesis overview
Your Background:
All the projects require knowledge in coding and simulation. It is highly preferable if you have experience with MATLAB and Simulink. Furthermore, knowledge of basic control theory is expected, and familiarity with advanced control (particularly basic MPC) would be an advantage. Lastly, and most importantly, a high level of interest and motivation in the topic is essential.
Credits: 60
Supervisor: Sebastian Zieglmeier
Possible co-supervisor: Mathias Hudoba de Badyn, Paal Engelstad, tbd