Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
Fast calculation of data-enabled predictive control solutions
We focus on a novel data-enabled predictive control (DeePC) algorithm, which relies only on input/output data to learn the behavior of the unknown system and perform safe and optimal control to drive the system along a desired trajectory using real-time feedback. The DeePC algorithm relies on solving an on-line convex optimization problem to obtain the optimal control sequence, which is sometimes computationally expensive for micro-controllers. In this project, we will explore different methods to reduce the computational burden and achieve fast calculation for the optimal control sequence.
Data-driven control is attracting increasing interest from both academia and industry. Compared to conventional model-based control, data-driven control has the advantage that it can be applied in scenarios where data is readily available, but the system and uncertainty models are too complex to obtain or maintain. Here we concentrate on the data-enabled predictive control (DeePC) framework, which relies only on input/output data to learn the behavior of the unknown system and perform safe and optimal control to drive the system along a desired trajectory using real-time feedback. The DeePC algorithm requires solving a convex optimization problem on-line to obtain the optimal control sequence in a receding-horizon manner. However, this could sometimes result in high computational burden for micro-controllers, compromising the on-line implementation. In this project, we will explore different methods to reduce the computational burden. We will derive approximated but explicit solutions for DeePC to enable fast on-line calculation. We will also explore the option of using deep neural networks to approximate the solution of a convex optimization problem. The developed methods will be tested in control applications such as power systems and renewable energy generators.
Data-driven control is attracting increasing interest from both academia and industry. Compared to conventional model-based control, data-driven control has the advantage that it can be applied in scenarios where data is readily available, but the system and uncertainty models are too complex to obtain or maintain. Here we concentrate on the data-enabled predictive control (DeePC) framework, which relies only on input/output data to learn the behavior of the unknown system and perform safe and optimal control to drive the system along a desired trajectory using real-time feedback. The DeePC algorithm requires solving a convex optimization problem on-line to obtain the optimal control sequence in a receding-horizon manner. However, this could sometimes result in high computational burden for micro-controllers, compromising the on-line implementation. In this project, we will explore different methods to reduce the computational burden. We will derive approximated but explicit solutions for DeePC to enable fast on-line calculation. We will also explore the option of using deep neural networks to approximate the solution of a convex optimization problem. The developed methods will be tested in control applications such as power systems and renewable energy generators.
1. The student will be introduced to and taught in the concept of data-enabled predictive control.
2. The student will be introduced to Simulink-based modelling and simulations.
3. The student will possibly have access to real micro controllers (if she/he is interested).
4. The student will implement the data-driven controller (DeePC) in simulations.
6. The student will explore different methods for fast calculation of DeePC.
7. The student will write the report and prepare a presentation.
This project could be done in person at the Automatic Control Laboratory, hybrid, or completely remotely depending on the current ETH rules. Most importantly, we can change between these forms whenever needed.
The project can be adapted on the run if new interesting research directions arise.
Finally, if the results are promising, they can be turned into a publication.
1. The student will be introduced to and taught in the concept of data-enabled predictive control.
2. The student will be introduced to Simulink-based modelling and simulations.
3. The student will possibly have access to real micro controllers (if she/he is interested).
4. The student will implement the data-driven controller (DeePC) in simulations.
6. The student will explore different methods for fast calculation of DeePC.
7. The student will write the report and prepare a presentation.
This project could be done in person at the Automatic Control Laboratory, hybrid, or completely remotely depending on the current ETH rules. Most importantly, we can change between these forms whenever needed.
The project can be adapted on the run if new interesting research directions arise.
Finally, if the results are promising, they can be turned into a publication.