Note: This content is accessible to all versions of every browser. However, this browser does not seem to support current Web standards, preventing the display of our site's design details.


Statistical Learning of MPC Control Laws for Buildings


S. Deml

Automatic Control Laboratory, ETH Zürich, Semester Thesis, FS13 (10251)

This study investigates machine learning methods for model predictive control laws in the field of building control. It focuses on continuous input types of the HVAC system during a summer period with deterministic occupancy information. A general learning setup is introduced and an approach for sample generation is shown. Furthermore a simple algorithm for the optimal control inputs for blinds and electric lighting is derived, while for the cooling input a more generic approach is presented. Subsequently the closed-loop performance of the machine learned controller (MLC) is discussed and a comparison to other MPC implementations is made. In order to increase the need for predictive control the available cooling power has been limited, leading to significant pre-cooling in the morning hours. It is shown that even for this advanced control problem the presented learning approach leads to very convincing results which are close to optimal. Finally, a sensitivity analysis of the learned parameters shows that the controller also performs convincingly for changes in the weather data.


Type of Publication:

(13)Semester/Bachelor Thesis

G. Schildbach

File Download:

Request a copy of this publication.
(Uses JavaScript)
% Autogenerated BibTeX entry
@PhdThesis { Xxx:2013:IFA_4639
Permanent link