Date of Award
Fall 2019
Document Type
Thesis
Degree Name
Master of Science (MS)
Department
Electrical and Computer Engineering
First Advisor
Yaz, Edwin E.
Second Advisor
Schneider, Susan
Third Advisor
Bonniwell, Jennifer
Abstract
In this thesis model order reduction techniques are applied to a set of systems representing the thermodynamics of a multi-zone building. The models are intended to be used in a model predictive control (MPC) application, with the individual zones defined using a simplified two temperature model. There has been an increased interest in model identification and reduction for MPC applications of building models that include multi-zone and non-linear models, but the most of this work has focused on models where individual zones are represented with a higher order model. Manual pole/zero removal, dominant eigenvalue, and balanced model reduction methods are presented, along with a proposed application specific method that takes advantage of the zone model’s simplified form. The proposed method treats a set of the zones as a common airspace with comparable control and reduces the underlying resistor/capacitor (RC) network. These methods are applied to a two zone and a six zone model with various coupling configuration tested. The general form of the multi-zone model proves difficult to reduce without making modification to the original form. The effects of reducing the inputs and outputs, through methods such as using a common temperature setpoint, are presented with significant improvements to reduction capabilities. Balanced model reduction the 18th order system down to 9th and 5th based on which inputs and outputs are reduced, and the application specific methods is able to reduce the same system down to a 3rd order model when the inputs and outputs are fully reduced.