Economic model predictive control applied to building climate control
using machine learning
Building climate control consists on keeping several physical
variables inside a building (like temperature
and humidity) within specified levels.
Traditional methods of building climate control
focus on tracking setpoints (e.g. specific values of temperature and humidity)
without taking into account the cost of operation.
Economic model predictive control (E-MPC) is an advanced control method
that explicitly minimizes the cost of operation of a process
while satisfying the physical constraints
determined by the process. Furthermore, E-MPC allows to include
predictions of disturbances (like weather) and predictions of operating costs (e.g. energy prices).
However, model-based control is challenging, due to the inherent complexity of
climate models.
Using machine learning techniques, like Gaussian processes, the modelling task
can be simplified.
This is an umbrella project that consist of several topics that are conceived to
be developed as several independent master's theses.
The aim of this umbrella project is to develop new methods for E-MPC for
building climate control using machine learning, to apply it to simulated and real plants,
and to analyse its closed-loop performance using hardware-in-the-loop simulation.
Parts of the implementation will rely on
μAO-MPC
as underlying embedded MPC software.
This bleeding-edge research project offers the possibility of writing a scientific article
to be published in a prestigious academic publication.
If you are interested in this topic, please discuss the details with
the
contact
persons.
Area:
Application of model predictive control
Helpful/Required Prerequisites:
Lectures: Regelungstechnik, Optimal Control.
Software: MATLAB, Python and C programming languages (previous experience not required).
Language: English.
Contact:
Markus Kögel
Sergio Lucia
Pablo Zometa