Self Organizing Map 기반 Cascade Neural Neworks를 이용한 건물 에너지 누수 검지 시스템

Anomaly Detection System(ADS) for a building electric energy consumption plays a crucial role in plugging electricity leakages and is vital in supporting continuous commissioning to resolve operating proplems and optimize energy use. In this paper, we introduce a novel anomaly detection system designed for building electricity usage developed using Cascade Neural Networks(NN). Cascade NN automatically generates neural structure ,which create robust and compact models with large data-sets in a time efficient manner, making it a perfect choice for building such a system. For the study Electric Energy consumption data was produced utilizing EnergyPlus simulation and Self- 0rganizing Map(SOM) which is an unsupersived learning of neural networks for feature extraction was used along with domain knowledge to investigate critical factors that influence the electricity consumption. In our experiments on DOE (Department of Energy) Reference Building for the June to August, the average Mean Bias Error(MBE) of current ADS was less than 3.0%