C09 - Machine Learning to Calibrate MIT’s Energy Planning Tool

Wednesday, November 20 from 4:00 PM to 5:00 PM | B401 « Back

Abstract Text:

To reduce greenhouse gas emissions associated with their building energy use, owners frequently rely on building energy models that are calibrated to existing conditions for evaluation of potential energy efficiency retrofits. Large campuses include several diverse-use buildings that need to be evaluated individually to manage their greenhouse gas emissions targets. These buildings cannot be easily classified into programmatic archetypes rendering the traditional urban building energy modeling approaches insufficient by themselves. In addition, the complex nature of institutional building characteristics makes the development of calibrated energy models for individual buildings prohibitively time and effort intensive. Finally, even when substantial time and effort is invested, the models are typically designed to be only used once, to evaluate static proposals relevant for the unique set of existing conditions, at a single point in time. To address these limitations, this lecture will present new workflows that combines established urban energy model generation techniques with data-driven methods to reduce the manual and computational cost of developing calibrated baseline campus energy models, allow for real time evaluation of future building upgrades, display their consequences to decision makers on an ongoing basis, and enable administrators to manage their building related greenhouse gas emissions over time. As a proof of concept, the complete method has been implemented and tested for the MIT campus in Cambridge, MA. This session will present, through technical and usability perspectives, the collaboration between research (Sustainable Design Lab), administration (Department of Facilities), and industry (Elementa Engineering) that resulted in the development and implementation of a data-driven auto-calibration framework to generate campus energy models for real time evaluation of building upgrades at the MIT campus.

Learning Objectives:

Explain the role of campus-wide building energy models for long term planning.

Explain conceptually how machine learning algorithms can auto-calibrate individual building energy models based on historic data.

Describe how facility managers can use such methods to prioritize which buildings within their portfolio to energy retrofit next.

Define the constraints of such planning tools and how personal knowledge of facilities still plays a key role.

Learning Level:



GBCI Credit Hours:
AIA Credit Hours:

Education Tracks: Data|AI|Smart Systems for Sustainability

Event Type: Conference Session > Session