A machine learning framework to estimate residential electricity demand based on smart meter electricity, climate, building characteristics, and socioeconomic datasets

McKenna Peplinski, Bistra Dilkina, Mo Chen, Sam J. Silva, George A. Ban-Weiss and Kelly T. Sanders

Applied Energy, 2024, vol. 357, issue C, No S0306261923017774

Abstract: Due to the substantial portion of total electricity use attributed to the residential sector and projected rises in demand, anticipating future energy needs in the context of a warming climate will be essential to maintain grid reliability and plan for future infrastructure investments. Machine learning has become a popular tool for forecasting residential electricity demand, but previous studies have been limited by lack of access to high spatiotemporal resolution at a regional scale, which reduces a model's ability to capture the relationship between electricity and its driving factors. In this study, we develop and execute a machine learning framework to predict residential electricity demand at varying temporal and spatial resolutions using hourly smart meter electricity records from roughly 58,000 homes provided by Southern California Edison as well as local weather data, building characteristics, and socioeconomic indicators. The best performing model at the household level, multilayer perceptron (MLP), was able to predict electricity demand most accurately at a monthly resolution, achieving an r2 of 0.45, while the most accurate annual and daily models (also MLP) had r2 values of 0.34 and 0.38, respectively. The results also show that models trained with data aggregated to the census tract level were more accurate (e.g., r2 = 0.82 for the monthly MLP model) than at the household level across all three temporal resolutions analyzed. Total square footage and various climate indicators had the highest feature importance values. Square footage was ranked first in feature importance for the annual and daily models, while the month of the year, which is strongly tied to temperature, was most important to the monthly model. Through this analysis we gain insight into factors that drive electricity demand and the usefulness of machine learning for predicting residential electricity use.

Keywords: Smart meter; Residential electricity; Machine learning; Climate change; Building energy; Energy forecasting (search for similar items in EconPapers)
Date: 2024
References: View references in EconPapers View complete reference list from CitEc
Citations: View citations in EconPapers (2)

Downloads: (external link)
http://www.sciencedirect.com/science/article/pii/S0306261923017774
Full text for ScienceDirect subscribers only

Related works:
This item may be available elsewhere in EconPapers: Search for items with the same title.

Export reference: BibTeX RIS (EndNote, ProCite, RefMan) HTML/Text

Persistent link: https://EconPapers.repec.org/RePEc:eee:appene:v:357:y:2024:i:c:s0306261923017774

Ordering information: This journal article can be ordered from
http://www.elsevier.com/wps/find/journaldescription.cws_home/405891/bibliographic
http://www.elsevier. ... 405891/bibliographic

DOI: 10.1016/j.apenergy.2023.122413

Access Statistics for this article

Applied Energy is currently edited by J. Yan

More articles in Applied Energy from Elsevier
Bibliographic data for series maintained by Catherine Liu ().