LAUSR

Abstract Considerable efforts have been made to find a reliable model able to accurately describe and predict the thermal behavior of the indoor environment of a building. Such a model is essential, in particular, for designing climate control strategies for optimizing both the comfort level and the energy consumption. However, a building is a complex system characterized by a nonlinear thermal behavior, so the task to find such a reliable model is rather difficult. This paper aims at overcoming some of these difficulties by presenting a data driven approach based on a switched system identification to detect and model the thermal behavior of a building zone during normal usage. The proposed technique relies on a PieceWise AutoRegressive model with eXogeneous inputs (PWARX) consisting of a set of sub-models with each one of them describing a certain configuration/state of the dwelling, e.g. turning the heating ON/OFF or opening/closing windows, doors and shutters. The approach is data-driven, easy to implement and its computational time is inferior to creating a detailed model under a specialized software. Therefore, it is particularly suitable for providing a quick description of the thermal behavior of existing buildings for which it is possible to install sensors and perform measurements. Using the available measurements, the algorithm is able to detect various configurations as will be shown by two numerical examples. Such a collection of sub-models provides a better temperature estimate than using ARX models, so it will eventually allow to select better strategies for improving energy efficiency.