Doctor of Philosophy (PHD)
Civil, Architectural and Environmental Engineering (Engineering)
Date of Defense
First Committee Member
Second Committee Member
Third Committee Member
Matthew Jacobs Trussoni
Fourth Committee Member
Michael D. Sohn
In the U.S., the building sector accounts for the largest portion of primary energy consumption and its energy consumption is expected to continuously increase in the coming decades. Many methods have been proposed to enhance building operations. Among these methods, the model predictive control and the regression-model-based-control are promising for large-scale applications. However, the model predictive control is difficult to implement due to the lack of appropriate modeling tools and thermal load prediction methods, while the regression-model-based-control has low accuracy. In this dissertation, a software environment for implementing the model predictive control is first presented. In this software environment, Modelica is used for system modeling while Python is used to automatize workflow, including state variable resetting. With this software environment, the study focuses on optimizing the design of the model predictive control for the purpose of resetting the condenser water return temperature set point (condenser water set point) and chiller staging. The results show that the speed and accuracy of the condenser water set point optimization can be improved by using the proposed method for selecting the initial point for searching. Results also reveal that the energy savings from the condenser water set point optimization is not sensitive to the reset frequency for the mild climate in which the study was conducted. Regarding the chiller staging optimization, results show that there is a trade-off among the energy use of chillers, pumps, and cooling towers. If the trade-off is considered in the design of the model predictive control, more energy savings can be achieved. A Bayesian network model for the cooling load prediction is then proposed. Compared to the existing methods, the Bayesian network model is easier to implement. A case study shows that the Bayesian network model can achieve comparable accuracy to the support vector machine method that has been recommended by previous studies. For both the Bayesian network model and the support vector machine model, the accuracy of the cooling load prediction is not always proportional to the amount of training data and may be significantly affected by the uncertainties in the inputs. The Bayesian network model is then applied in the regression-model-based-control for resetting the condenser water set point. The case study shows that the linear and polynomial models that were proposed in the previous studies sometimes even increase energy consumption, while the Bayesian network model can achieve nearly optimal energy savings. Finally, this dissertation demonstrates the preliminary work of implementing a model predictive control for an integrated community energy system that serves a net zero energy community. Suggestions for future work are also provided.
Model Predictive Control; Modelica; Building Operation; Net Zero Energy Community
Huang, Sen, "Model based Technologies for Enhancing Building Operation" (2016). Open Access Dissertations. 1637.