Artificial neural networks (ANNs) mimic a few aspects of the learning processes of the human brain. A method is presented for training ANNs to automatically update energy consumption predictors for commercial buildings. Approximately 100 ANNs have been trained and installed on a commercial building in Colorado as part of an HVAC system diagnostic expert system. Compares the accuracy of the ANN approach to more conventional regression-based methods, including singular valued decomposition, for this building. It will be shown that ANNs are more accurate with drastically less user input and operator knowledge. The lessons learned with this approach are described in the context of commercial buildings.

KEYWORDS: artificial intelligence, energy consumption, commercial, buildings, USA, building services, expert systems, comparing, accuracy, networks