The Great Energy Predictor Shootout II : Measuring Retriofit Savings-Overview and Discussion of Results

Abstract

A second predictor shootout contest has been developed and conducted to evaluate the most effective empirical or inverse models for modeling hourly whole-building energy baselines for purposes of measuring savings from energy conservation retrofits. This second contest utilized two sets of measured hourly pre-retrofit and post-retrofit data from buildings participating in a revolving loan program in Texas. The accuracy of the contestants` models was evaluated by determining their ability to predict data that were carefully removed from the training (or pre-retrofit) period. A comparison of the savings predicted by the models is also presented. The results from the contest show that neural networks again provide the most accurate model of a building`s energy use. However, in contrast to the first contest, the second contest`s results show that cleverly assembled statistical models also appear to be as accurate or, in some cases, more accurate than some of the neural network entries. When these models were used to forecast the baseline use into the post-retrofit periods, large variations in the predicted savings occurred among the models, particularly for the cooling energy savings in one of the case study buildings. These variations appear to be due to the ability of the models (ormore » inability) to capture certain energy performance characteristics and the modeler`s assumptions about the post-retrofit energy use.« less