Publication Date


Advisor(s) - Committee Chair

Dr. Jonathan Quiton (Director), Dr. Melanie Autin, Dr. Mark Robinson

Degree Program

Department of Mathematics

Degree Type

Master of Science


In this study, we use a nonparametric technique, locally weighted robust least squares regression (LOESS), to forecast a 24 hour demand profile at the household level and compare it to existing aggregate demand models discussed in literature. Of these aggregate demand models, a quadratic autoregressive model was selected to be used as a basis for comparison with the LOESS forecasts. It was our goal to automate the forecasting process by using the goodness of fit metric, AICCI, for smoothing parameter selection. The statistical workflow was executed using SAS and data was provided by the Glasgow Electric Plant Board of Barren County, Kentucky. Results show that LOESS outperformed the autoregressive model in roughly 80% of all cases and than using LOESS alone or as part of an ensemble model is a feasible approach to automating future household demand profile for the purpose of generating different levels of power demand profile aggregation as needed by Glasgow Electronic Plant Board.


Mathematics | Statistics and Probability