Structured probabilistic models for capturing household and temporal variations in the internal electricity load

Abstract

This study developed a structured probabilistic statistical model that captures household and temporal variations at different time resolutions separately for predicting the electricity load of residential communities. We used hourly electricity data, including plug-in and lighting loads, of 26 households obtained from the public data of the Korea Energy Agency. The prediction model set consists of four models. Models 1 and 2 are bilinear regression models that can predict annual and daily average electricity loads on the basis of the household characteristics and variation in the daily electricity load, respectively, and Models 3 and 4 are based on multivariate normal distribution, and they provide average hourly electricity load profiles and temporal variations from the average profile, respectively. Six key parameters that characterize the residential building electricity load magnitude and timing over one-day profile were defined and used to compare the model predictions against actual measurements. The structured probabilistic models resulted in the coefficient of variation of root mean square error (CV(RMSE)) ranging between 2.5 and 14.9% for key load characterization predictions. In addition, the percent error of the standard deviation predicted by the model ranged between 2.5 and 10.6%. The validation results demonstrated that the structured probabilistic models with full consideration of household and temporal variations provide plausible variations in alignment with actual measurements.