Eventos Anais de eventos
COBEM 2017
24th ABCM International Congress of Mechanical Engineering
Nonlinear ARX Model Associated to Neural Networks to Predict Hygrothermal Behavior of Building Materials
Submission Author:
Cassiana Fagundes da Silva , PR
Co-Authors:
Roberto Zanetti Freire, joseph VIRGONE, Abdelkrim Trabelsi, Cassiana Fagundes da Silva
Presenter: Cassiana Fagundes da Silva
doi://10.26678/ABCM.COBEM2017.COB17-2673
Abstract
Moisture presence in building material can significantly affect heat exchanges between indoor and outdoor environments, influencing on both energy consumption and thermal comfort. With the objective of estimating the hygrothermal variations, computational tools, those based on analytical and numerical models, are being used to reduce the energy consumption of new and retrofitting buildings. However, when moisture presence is taken into account, especially when high hygroscopic materials are adopted in building projects, a nonlinear behavior may occur affecting temperature profiles within building structures. This situation is constantly discussed in the literature as a difficult task due to modeling difficulty and highly moisture-dependent properties. Based on these concepts, this work presents a black box system identification approach in order to reproduce the hygrothermal behavior of high hygroscopic materials that are commonly adopted as insulation on building envelopes. By assuming a mixed approach considering both linear and non-linear techniques, a NARX (Nonlinear AutoRegressive with eXogenous input) MIMO (multiple-input, multiple-output) model, where an Artificial Neural Network (ANN) was considered as nonlinear approximation tool, is presented in this work. By using a data set provided by experimental analysis, the model was validated and reasonable results in terms of approximation were obtained. To conclude this paper, remarks about the model performance are presented, including the computational cost.
Keywords
Building Simulation, hygroscopic materials, hygrothermal behavior, Neural Network, System Identification
