OR-94-17-2 -- Predicting System Loads with Artificial Neural Networks -- Methods and Results from "The Great Energy Predictor Shootout" PDF

A feed-forward artificial neural network (ANN) procedure has been devised for predicting utility loads; the resulting predictions are presented for two test problems given by "The Great Energy Predictor Shootout - The First Building Data Analysis and Prediction Competition" (Kreider and Haberl 1994). Key ingredients in this approach are the multilayer perception and a method (delta test) for determining relevant inputs. These methods are briefly reviewed, together with comments on alternative schemes such as fitting polynomials and the use of recurrent networks.

KEYWORDS: artificial intelligence, expert systems, calculating, heat load, energy consumption, competition