Multivariate fault detection for residential HVAC systems using cloud-based thermostat data, part I: Methodology

Abstract

Fault detection and diagnosis (FDD) using smart thermostat data is a relatively new field of research, but one with immediate practical application to residential indoor climate control. This two-part article proposes a statistics-based method for using aggregated data from many residential HVAC systems to identify outliers. These outlier systems can then be related to operational problems, such as inadequate capacity, malfunctioning control strategies, or other soft faults. The first part of the article presents the analytical methodology. For this study, historical and real-time data from thousands of cloud-based thermostats are used to extract features during the cooling mode of operation. First, the data are parsed to identify time segments that are characteristic periods of pseudo-steady-state operation. Then, a multistage fault detection method compares the features among all systems to find statistical outliers. Each feature is analyzed independently using kernel density estimation (KDE) methods. Additionally, multiple features are analyzed simultaneously using multivariate statistics and the Mahalanobis distance () to determine outliers. Anomalous systems are sorted heuristically, and information from univariate and multivariate methods is combined to more precisely classify systems exhibiting specific faulty behavior. The second part of this article provides some typical case studies, illustrating how this method can be used to identify aberrant behavior across a large number of monitored systems.