Predicting residential energy and water demand using publicly available data

The overarching objective behind this work is to merge publicly available data with utility consumption histories and extract statistically significant insight on utility usage for a group of houses (n = 7022) in Gainesville, USA. This study investigates the statistical descriptive power of publicly available information for modeling utility usage. We first examine the deviations that arise from monthly utility usage reading dates as reading dates tend to shift and reading periods tend to vary across different months. Then we run regression models for individual months which in turn we compare to a yearly regression model which accounts for months as a dummy variable to understand whether a monthly model or a yearly model has a larger statistical power. It is shown that publicly available data can be used to model residential utility usage in the absence of highly private utility data. The obtained results are helpful for utilities for two reasons: (1) using the models to predict the monthly changes in demand; and (2) predicting utility usage can be translated into energy-use intensity as a first-cut metric for energy efficiency targeting in their service territory to meet their state demand reduction targets.