Date of Award

Spring 1-1-2019

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

First Advisor

Gregor P. Henze

Second Advisor

Michael J. Brandemuehl

Third Advisor

Balaji Rajagopalan

Fourth Advisor

Kyri A. Baker

Fifth Advisor

Robert H. Dodier

Abstract

Automation has proven indispensable to advancing human endeavors. Within the built environment its evolution and sophistication are on the cusp of moving beyond automatic control into automated prediction and diagnosis. A data-driven toolchain is developed so human efforts can be focused on high-value concerns. The research examines smart buildings as a cyberphysical construct and places the Bayesian perspective as paramount. Prior knowledge is leveraged through common building energy modeling and simulation tools, which are utilized and extended. An iterative, three-step process is developed to 1) classify building energy performance scenarios, 2) forecast dynamics over a planning horizon of interest, and 3) signal human decision-makers concerning deviations from ideal behavior. In the classification step, focus is placed on the discrete wavelet transformation of electrical demand profiles, producing energy and entropy feature extraction from the wavelet levels at definitive time frames, and Bayesian probabilistic hierarchical clustering. The process yields a categorized and manageable set of representative electrical demand profiles for smart grid applications. In the forecasting step, a cyclical two-stage model predictive control process of policy planning followed by execution is evaluated. The results show that even the most complicated nonlinear autoregressive neural network with exogenous input does not appear to warrant the additional efforts in forecasting model development and training in comparison to the simpler models. In the signaling step, a simulation study is considered to assess whole-building energy signaling accuracy in the presence of uncertainty and faults at the submetered level, which may lead to tradeoffs at the whole-building level that are not detectable without submetering. Together, the steps form a data-driven toolchain for the operational performance analysis and optimization of buildings.

Share

COinS