Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
Performance of nonlinear data-driven control on smart buildings
Data-driven control of nonlinear systems is a hot, exciting new field. Can it perform as well as model-based control? In this thesis, you get to find out by doing experiments on a smart apartment.
Keywords: Model predictive control, DeePC, data-driven control, building climate control, smart buildings, smart cities, data-enabled predictive control
Energy consumption in buildings accounts for over 30% of worldwide energy consumption. Mitigating energy consumption is typically done by improving operation of heating and cooling systems in buildings. Efficient operation of heating and cooling is a part of climate control in buildings. The main task of climate control is to satisfy heating and cooling needs of the occupants of buildings using minimal amount of energy. Due to the inherent complexity of the requirements and the dynamics of buildings, developing efficient climate controllers based on mathematical models is a challenge. Model-free controllers, such as DeePC, have a great potential thanks to their simplicity and ability to work with limited knowledge of the controlled building. Or do they? The question to answer in this project is what kind of of complexities DeePC can handle.
Perfect models of buildings do not exist, and even if they did, their use for control purposes would be limited due to computational and numeric challenges. To overcome the challenges, existing models are typically simplified to capture only the most important characteristics of the building. The definition of what most important characteristics entail is up to the designer of the controller and the choice of the model will affect the performance of the resulting control system. Model-free approaches, such as DeePC, overcome the issues of choosing the model by relying exclusively on measured data. Whereas it can be shown that DeePC is equivalent to model predictive control for linear systems, it is still unclear what kind of nonlinearities can be handled by DeePC. Is it better to use model-based controllers using simplified models or is it better to ditch the models completely in favour of model-free approaches?
Energy consumption in buildings accounts for over 30% of worldwide energy consumption. Mitigating energy consumption is typically done by improving operation of heating and cooling systems in buildings. Efficient operation of heating and cooling is a part of climate control in buildings. The main task of climate control is to satisfy heating and cooling needs of the occupants of buildings using minimal amount of energy. Due to the inherent complexity of the requirements and the dynamics of buildings, developing efficient climate controllers based on mathematical models is a challenge. Model-free controllers, such as DeePC, have a great potential thanks to their simplicity and ability to work with limited knowledge of the controlled building. Or do they? The question to answer in this project is what kind of of complexities DeePC can handle.
Perfect models of buildings do not exist, and even if they did, their use for control purposes would be limited due to computational and numeric challenges. To overcome the challenges, existing models are typically simplified to capture only the most important characteristics of the building. The definition of what most important characteristics entail is up to the designer of the controller and the choice of the model will affect the performance of the resulting control system. Model-free approaches, such as DeePC, overcome the issues of choosing the model by relying exclusively on measured data. Whereas it can be shown that DeePC is equivalent to model predictive control for linear systems, it is still unclear what kind of nonlinearities can be handled by DeePC. Is it better to use model-based controllers using simplified models or is it better to ditch the models completely in favour of model-free approaches?
The project focuses on DeePC and applies it to building climate control, assuming that building dynamics are nonlinear. The controller has been shown to perform well in small nonlinear applications, but it is unclear what kind of nonlinearities can be handled by DeePC. This project aims to answer the following questions:
- How do the nonlinearities in the system affect the performance of DeePC?
- When is DeePC better than model predictive control?
- What would be the benefits, if any, of extending the model-free approach to include some information about the dynamics?
The project focuses on DeePC and applies it to building climate control, assuming that building dynamics are nonlinear. The controller has been shown to perform well in small nonlinear applications, but it is unclear what kind of nonlinearities can be handled by DeePC. This project aims to answer the following questions:
- How do the nonlinearities in the system affect the performance of DeePC?
- When is DeePC better than model predictive control?
- What would be the benefits, if any, of extending the model-free approach to include some information about the dynamics?
Please sent a CV and transcript of courses to:
Mathias Hudoba de Badyn (mbadyn@control.ee.ethz.ch)
Marta Zagorowska (mzagorowska@control.ee.ethz.ch)
Please sent a CV and transcript of courses to:
Mathias Hudoba de Badyn (mbadyn@control.ee.ethz.ch) Marta Zagorowska (mzagorowska@control.ee.ethz.ch)