Predicting System loads with Artificial Neural Networks : Method and Result from "the Great Energy Predictor Shootout"

We devise a feed-forward Artificial Neural Network (ANN) procedure for predicting utility loads and present the resulting predictions for two test problems given by ``The Great Energy Predictor Shootout - The First Building Data Analysis and Prediction Competition''. Key ingredients in our approach are a method ($\delta$ -test) for determiningrelevant inputs and the Multilayer Perceptron. These methods are briefly reviewed together with comments on alternative schemes like fitting to polynomials and the use of recurrent networks.