Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
Distributed Control of Large-scale Systems
Distributed MPC is a popular technique to control large-scale systems such as microgrids and vehicle platoons. In this project, we aim to develop a distributed MPC scheme with low computational complexity using distributed optimization techniques to control safety-critical large-scale systems.
Keywords: Model Predictive Control, Large-scale Systems, Distributed Optimization, Convex Optimization
Control of large-scale systems has attracted the attention of the control systems community due to its wide variety of applications which include, but are not limited to, power networks, DC microgrids, electric vehicle charging and vehicle platooning. Such systems can be controlled using centralized control schemes in which a central entity is used to collect measurements from the whole system and compute all the control inputs accordingly. However, centralized control schemes have their own drawbacks such as increased computational complexity and lack of robustness against failures. These schemes are also not well-suited for controlling varying-topology networks.
To overcome these issues, distributed control schemes are developed such that the large-scale system is decomposed into several smaller subsystems and a local controller is developed for each subsystem. Distributed model predictive control (MPC) in particular has gained attention due to its ability to handle the control requirements in a systematic manner, take state and input constraints into account and make use of the predicted behaviour. In our group, we have developed a novel distributed MPC scheme for large-scale systems. In this scheme, each local controller is required to solve an optimization problem within a predefined sampling period and is allowed to share information with the local controllers of a subset of other subsystems only.
The first aim of this project is to investigate methods for solving the resulting optimisation problem in a distributed fashion. Moreover, contrary to standard MPC schemes which solve quadratic programs, the developed scheme solves a semidefinite program (SDP). In this case, the solution becomes more time-consuming due to the existence of linear matrix inequalities in the optimization problem. The second aim is to explore methods to reduce the computational complexity of the developed scheme so that it becomes computationally comparable to standard MPC schemes.
Control of large-scale systems has attracted the attention of the control systems community due to its wide variety of applications which include, but are not limited to, power networks, DC microgrids, electric vehicle charging and vehicle platooning. Such systems can be controlled using centralized control schemes in which a central entity is used to collect measurements from the whole system and compute all the control inputs accordingly. However, centralized control schemes have their own drawbacks such as increased computational complexity and lack of robustness against failures. These schemes are also not well-suited for controlling varying-topology networks.
To overcome these issues, distributed control schemes are developed such that the large-scale system is decomposed into several smaller subsystems and a local controller is developed for each subsystem. Distributed model predictive control (MPC) in particular has gained attention due to its ability to handle the control requirements in a systematic manner, take state and input constraints into account and make use of the predicted behaviour. In our group, we have developed a novel distributed MPC scheme for large-scale systems. In this scheme, each local controller is required to solve an optimization problem within a predefined sampling period and is allowed to share information with the local controllers of a subset of other subsystems only.
The first aim of this project is to investigate methods for solving the resulting optimisation problem in a distributed fashion. Moreover, contrary to standard MPC schemes which solve quadratic programs, the developed scheme solves a semidefinite program (SDP). In this case, the solution becomes more time-consuming due to the existence of linear matrix inequalities in the optimization problem. The second aim is to explore methods to reduce the computational complexity of the developed scheme so that it becomes computationally comparable to standard MPC schemes.
We are looking for highly-motivated and talented students who are interested in working on distributed methods for convex optimization. Students are expected to perform the following tasks:
- Explore different distributed optimization methods (e.g. ADMM) which can be used to solve the distributed MPC problem and evaluate these methods numerically.
- Explore different SDP relaxation methods (e.g. Chordal Sparsity) which can be used to reduce the computational complexity of the developed MPC problems and evaluate these methods by comparing them to standard MPC schemes.
We are looking for highly-motivated and talented students who are interested in working on distributed methods for convex optimization. Students are expected to perform the following tasks: - Explore different distributed optimization methods (e.g. ADMM) which can be used to solve the distributed MPC problem and evaluate these methods numerically. - Explore different SDP relaxation methods (e.g. Chordal Sparsity) which can be used to reduce the computational complexity of the developed MPC problems and evaluate these methods by comparing them to standard MPC schemes.
Interested students should send an email with a short motivation statement, transcript of records (BSc and MSc) and updated CV to Ahmed Aboudonia at ahmedab@control.ee.ethz.ch and/or Goran Banjac at gbanjac@control.ee.ethz.ch
Interested students should send an email with a short motivation statement, transcript of records (BSc and MSc) and updated CV to Ahmed Aboudonia at ahmedab@control.ee.ethz.ch and/or Goran Banjac at gbanjac@control.ee.ethz.ch