Topic: Physical model or Data model? The use cases of machine learning in building design and operation
Buildings are the most significant contributor to energy consumption in the United States. Because of this, the design team is encouraged to use building energy models (BEMs), a physical-based forecast model of the energy usage in the building, to inform building energy design decisions. However, only 20% of projects in the United States use BEMs, and 80% of these projects use BEM for code compliance when the building design is mostly completed. A national survey indicated that the long and complicated modeling process mostly diminishes the value of the doing BEMs. To address this issue, researchers proposed using a data model instead of a physical model at the early stage of building designs. This seminar will focus on introducing three use cases of machine learning in building energy design and operation, following by a discussion on the trade-offs between the physical model and data model.
Dr. Weili Xu is co-founder and Chief Product Officer of BuildSim, mainly responsible for the development and marketing of BuildSim Cloud, a cloud infrastructure of energy modeling. He received his Ph.D. degree in Building Performance and Diagnostics from Carnegie Mellon University in 2017. His research focuses on building design optimization for energy and cost. He also worked as a data scientist in Autodesk, help the team exploring the potential of use machine learning in assist design decision making. Besides software development, he also worked as a consultant for energy modeling projects and helped projects to achieve LEED certifications. Currently, he is managing and developing BuildSim machine learning module, which has successfully helped clients achieve fast design iterations and save 90% of the time.