A data-driven approach to discover hidden complicated relationships of energy variables and estimate energy consumption in U.S. homes

Abstract

The U.S. government has committed to improving building energy efficiency. In many buildings, residential homes are one of the largest end-users of energy consumption. Today, many U.S. homes have been in use for decades and they are now outdated, poorly insulated and equipped. Retrofitting existing homes is therefore urgent to improve the quality of Americans’ life and reduce environmental impact from energy waste. To support successful retrofits, this study proposes a decision tree-based analytical model to identify the complex relationships between residential energy variables of physical and socio-economic characteristics using the Residential Energy Consumption Survey (RECS). For this, a model-based recursive partitioning (MOB) algorithm was applied in the decision tree models for understanding energy consumption in residential buildings. The results discovered the most influential energy variables for retrofits and identified heterogeneous relationships on energy consumption for different climatic regions. Also, the findings from decision tree models offer estimations for residential energy consumption in different U.S. climate zones, depending on the combinations of design and operating energy variables. The proposed equations for the EUI estimations can be used to predict the impact of energy variables on primary residential load components (i.e., cooling, heating, domestic hot water loads) to support effective retrofits for architects and homeowners in the future.