A whole building fault detection using weather based pattern matching and feature based PCA method

Heating, ventilation and air conditioning (HVAC) systems in commercial buildings consume more than 14% energy in the U.S. Malfunctioning sensors, components, and control systems, as well as degrading systems in HVAC and lighting systems are main reasons for energy waste and unsatisfactory indoor environment. Studies have demonstrated that large energy saving can be achieved by automated fault diagnosis followed by corrections. Data-driven based methods have been widely adopted in component level fault detection and diagnosis in building sectors. For a whole building system in which various sub-systems are coupled together and have closely interactions, conventional data-driven based methods that have been successful for component level fault detection encounter many challenges such as the curse of dimensionality, difficulty to generate system baseline, and developing system model. A new data-driven based strategy which includes a weather based pattern matching method and feature based Principal Component Analysis (PCA) method is proposed for whole building level fault detection. Symbolic Aggregate approXimation (SAX) method is employed to find similar weather patterns in historical database to accurately and dynamically generate baseline datasets. In order to handle the issue of high dimensionality of a whole building's dataset, a feature selection process is performed using Partial Least Square Regression and Genetic Algorithm (PLSR-GA) method. Selected features are then used in a PCA modeling and fault detection process. Data from a real campus building are obtained to evaluate the effectiveness of the proposed strategy.