Metamodeling of Energy and Operational Carbon in Detached Accessory Dwelling Units

Abstract

The rapidly escalating cost of housing has created a crisis in the United States that stems from a lack of housing supply, which is exacerbated by single-family zoning. While revising this policy is a needed course of action, doing so is slow and politically arduous. In the interim, many architects and planners propose allowing a detached accessory dwelling unit, or DADU, in the rear yard of single family lots to provide more housing options. This strategy has increased housing supply in west coast cities such as Portland, Seattle, and Los Angeles. Leveraging this trend, this research involves the creation of a web-based application for predicting DADU energy consumption, utility costs, and carbon emissions in the early stages of design through the use of surrogate modeling. This enables both homeowners and designers to quickly estimate and improve energy performance to reduce long term costs and increase the production of DADUs. The dataset was created by simulating the entire examined design space using Rhinoceros, Grasshopper, Ladybug Tools, and EnergyPlus. Simulation data is stored in .CSV files which were preprocessed using Pandas. XGBoost was selected as the machine learning option, with hyperparameter optimization via bayesian search. The web app was then constructed using Streamlit and hosted on Heroku.