Measurement and Verification Building Energy Prediction (MVBEP): An Interpretable Data-Driven Model Development and Analysis Framework

Alrobaie, Abdurahman Saad

Graduate Thesis Or Dissertation

Measurement and Verification Building Energy Prediction (MVBEP): An Interpretable Data-Driven Model Development and Analysis Framework Public Deposited

Analytics

Citations

Citeable URL: https://scholar.colorado.edu/concern/graduate_thesis_or_dissertations/v692t7690

Abstract

The operation of building energy systems including Heating, Ventilation, and Air Conditioning (HVAC), lighting, and equipment accounts for 85% of the global building energy consumption. With several countries pledging to achieve sustainability goals, building retrofit is becoming a crucial pillar in attaining most of the set energy efficiency targets. However, several obstacles remain that prevent retrofitting buildings to be economically feasible. An essential task in retrofitting a building is justifying the cost effectiveness of the installed Energy Conservation Measures (EMC), that is, Measurement and Verification (M&V) of the achieved energy and cost savings.

Due to the rapid development of Advanced Metering Infrastructure (AMI), data-driven approaches are becoming more effective than deterministic methods in developing M&V baseline energy models for existing buildings using historical energy consumption data. This thesis develops and applies a framework for M&V baseline data-driven modeling aimed at estimating energy savings from building retrofits. The framework employs the common data-driven modeling approaches used for building energy prediction, necessary data processing steps, and industry-used evaluation approaches. The considered modeling approaches in the developed framework include linear regression, ensemble models, support vector regression, neural networks, and kernel regression. The developed framework is applied to two case studies with different data sources of: energy consumption data generated via a simulated office building model and mostly measured energy use data collected by Building Data Genome 2 (BDG2), a publicly available dataset.

The framework application to data from over 208 buildings indicated that energy use predictions can be achieved with medians of Coefficient of Variation of Root Mean Squared Error (CV(RMSE)) that are less than 20% and %25 for daily and hourly frequency, respectively. The Median Normalized Mean Bias Error (NMBE) for these predictions has a range of ±|3%| for all modeling approaches and data frequencies. No significant impact on prediction accuracies was observed for the building typology and the site climate.

Creator