Anna Klyushina M.Sc.

Model Predictive Control

Contact

work +49 6151 16-25042
fax +49 6151 16-25058

Work S3|10 408
Landgraf-Georg Straße 4
64283 Darmstadt

Model Predictive Control

Model predictive control (MPC) is a control technique that enables the optimal control of complex, multivariable systems while taking constraints into account. The core concept of MPC is to solve an optimization problem in real time. At each sampling instant, an optimization problem is solved over a specified time horizon, based on the current state of the system. The system is then controlled using the optimal solution. After one time step, the current state is measured, and the optimization problem is solved again from the new state over a shifted horizon. This approach is known as receding horizon control.

Thanks to its numerous advantages, MPC has become a standard in the process industry. However, one significant limitation of this method is that the optimization problem must be solved “online” at each time step. This can be challenging for complex, especially nonlinear, processes or systems with very short sampling times.

To overcome this limitation, the so-called explicit model predictive control was developed. Explicit MPC shifts all necessary computations for solving the optimization problem to an “offline” phase, i.e., before the actual operation begins, while retaining the key advantages of MPC. In this approach, the feasible state space is partitioned into regions where the solution to the optimization problem, i.e., the control input, is constant. During operation, only the control input needs to be evaluated based on the current state.

Due to the reduced computational effort during operation, explicit MPC has found wide application in real-time systems. There are two main approaches to solving explicit MPC: exact and approximate methods. Exact methods aim to optimally partition the state space. These methods are based on system linearity and can result in a very large number of small regions in the state space, leading to high memory requirements. Approximate methods, on the other hand, approximate the optimal solution and reduce both the memory required for look-up tables and the computational effort for evaluating the control input.

My research focuses on further developing and adapting these methods to improve the efficiency and applicability of model predictive control, especially in real-time systems.

Lecture Semester Tasks
System Dynamics and Automatic Control Systems II SS 2024 Tutoring of the course
System Dynamics and Automatic Control Systems III WS 2024/25 Tutoring of the course

Do you find my research interesting and would you like to write a thesis on this topic?

Then write me an e-mail and briefly introduce yourself and your interests!