Sule Balkan & Tim Anderson
Energy Demand Forecasting
The purpose of this class was to develop an hourly demand forecast for 2017 based on on the previous demand and temperature data of the ISO NE region. There was no specified method to create the model, and the majority of the work was to be done outside of class. The class was supposed to coincide with the Global Energy Forecasting Competition (GefCOM) to drive innovations and competition. However, the class stated weeks into the competition leaving the students behind in both time and understanding. The forecasting method chosen for this report was a Support Vector Machine (SVM) based on the mean hourly demand that factored in various time-series information such as: day-of-month, weekday, week-of-month, and month. The mean demand was chosen to reduce the error introduced by attempting to predict the weather as well as the demand. While the final results for the SVM model were not extraordinary by any means. It was still shown to be an affective tool for time-series forecasting. This report does deviate from the majority of the class due to an error in the registration for the competition. Since this reports forecasts would not be seen by anyone other than in class. Nearly all of the competition deliverables were chosen to be either the previous years demand or it’s mean hourly demand. It was evident that many of the models developed through the term were not good, therefore it seemed reasonable to simple use the previous years data as a place holder until a better model could be developed.
Slay, Tylor, "Demand Forcasting:
Efficacy of Support Vector Machines" (2017). Engineering and Technology Management Student Projects. 105.