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

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