Building energy consumption modeling with Neural Ensembling Approaches for fault detection analysis

In the paper a fault detection analysis through neural ensembling approaches is presented. Experimentation was carried out over two months monitoring data sets for the lighting energy consumption of an actual office building located at ENEA ‘Casaccia’ Research Centre. Using a fault free data set for the training, the Artificial Neural Networks Ensembling (ANNE) were used for the estimation of hourly lighting energy consumption in normal operational conditions. The fault detection was performed through the analysis of the magnitude of residuals using a peak detection method. Moreover the peak detection method was applied directly to the testing data set. Finally a majority voting method to ensemble the results of different ANN classifiers was performed. Experimental results show the effectiveness of ensembling approaches in automatic detection of abnormal building lighting energy consumption.