Development and Testing of Building Energy Model Using Non-Linear Auto Regression Neural Networks (ST-16-C007)

Buildings account for about 48 percent of the energy consumed in the United States. Of this energy, heating and cooling systems use about 55 percent, while lights and appliances use the other 35 percent of energy use of existing buildings. If energy-use trends continue, buildings will become the largest consumer of global energy by 2025. The development of building energy savings methods and models becomes apparently more necessary for a sustainable future. Most new buildings are equipped with modern building automation system BAS and direct digital control that have the ability to collect large amounts of data. However, even with modern technologies, those buildings are unfortunately still not operating optimally due to the absence of computational means and centralized solutions embedded into the BAS. Therefore, there is a significant need to investigate how modern computational techniques can help generate the analysis needed to gain full benefit from real-time data and at the same time perform many potential intelligent applications such as modelling, optimization, energy efficiency and energy assessment, and fault detection and diagnosis. This paper discusses the modeling methodologies for building energy system using time series auto regression artificial neural networks. The model can be integrated into energy solution tools for building energy assessment, optimization, fault detection and diagnosis, and many other applications. The model predicts whole building energy consumptions as function of four input variables, dry bulb and wet bulb outdoor air temperatures, hour of day and type of day. To train and test the models, data from twenty existing buildings and from simulations are used for the analysis. Advanced computational methods are used for data analysis and preprocessing. The wavelet basis is used to remove the noise and anomalies. Different neural network structures are tested along with various input delays to determine the one yielding the best results in term of mean square error. The results show that the developed model can provide results sufficiently accurate for its use in various energy efficiency and saving estimation applications.