Data-driven methods for smart building AHU subsystem modelling

Modern, densely instrumented, smart buildings generate large amounts of raw data. This poses significant challenges from both the data management perspective as well as leveraging the associated information for enabling advanced energy management, fault detection and control strategies. Networks of intelligent sensors, controllers and actuators currently allow fine grained monitoring of the building state but shift the challenge to exploiting these large quantities of data in an efficient manner. We discuss methods for black-box modelling of input-output data stemming from buildings. Using exploratory analysis it is argued that data mining inspired approaches allow for fast and effective assessment of building state and associated predictions. These are illustrated using a case study on real data collected from commercial-grade air handling units of a research building. Conclusions point out to the feasibility of this approach as well as potential for data mining techniques in smart building control applications.