Yujie Xu will present the thesis proposal defense for the dissertation, "Using Machine Learning to Target Retrofits in Commercial Buildings under Alternative Climate Change Scenarios," as a candidate of the PhD in Building Performance & Diagnostics (PhD-BPD) on Tuesday 19 May.
Title: “Using Machine Learning to Target Retrofits in Commercial Buildings under Alternative Climate Change Scenarios”
By Yujie Xu, PhD-BPD Candidate
Date: Tuesday, 19 May 2020
Time: 9:00-11:00am EDT
Location: Virtual on Zoom
Thesis Committee:
Professor Vivian Loftness, FAIA, Chair
School of Architecture, Carnegie Mellon UniversityProfessor Ömer T. Karagüzel, PhD
School of Architecture, Carnegie Mellon UniversityEdson Severnini, PhD Assistant Professor
Department of Engineering and Public Policy (EPP), Carnegie Mellon University
Abstract
Buildings account for 40% of energy consumption, and 36% of CO2 emissions in the U.S. With a low rate of new construction, energy retrofits of existing buildings are recognized as an effective means to reduce building consumption and carbon footprint. One of the key components in retrofit planning is projecting the retrofit effect for an action. This is achieved mostly by simulation-based tools. These tools allow for detailed assessment of a much larger range of retrofit actions. However, they usually require detailed data inputs, high expertise, and extensive computation power, rendering it challenging to implement in large portfolios, or in large scale policy evaluations. Existing data-driven methods are generally easier to implement, faster to run, and require less building systems expertise. If done properly, they could have more realistic estimates due to the use of real-world data. However, they are almost solely used in evaluating past retrofits, reporting effects either too specifically (for one specific building), or too generally (the average effect of a large population). This lack in prescriptiveness limits their ability to generalize to unseen new buildings, thus providing no decision support for future retrofits. This thesis proposes a data-driven approach that generalizes heterogenous effects of past retrofits to future savings potentials (for buildings with certain characteristics), assisting targeted retrofit planning. This method will also have the capability to estimate retrofit effect under alternative climate scenarios, providing valuable information for long-term retrofit planning in the face of climate change which affects many aspects of human life.