Presents methods and results for the prediction of energy conservation retrofits for HVAC systems as part of the Great Building Energy Predictor Shootout II. The predictions are based on hourly whole-building electricity, motor control centre electricity, lights and equipment electricity, cooling and heating energy use and accompanying weather data. An autoassociative neural network was used as a preprocessor to replace the missing data and a standard feed-forward artificial neural network was utilised to predict building energy consumption. Finds that although the prediction results were acceptable for the verification of the sample networks, actual building energy prediction could have benefited by including the holiday and weekend information during the network training as well as information from previous time steps.

KEYWORDS: year 1996, Energy conservation, modernising, buildings, expert systems, accuracy, USA computer programs