Identification of temperature and moisture content fields using a combined neural network and clustering method approach

Studies on the dynamics of temperature and moisture content distributions in porous soils have provided important insight on their effect on the building hygrothermal behavior, where the interaction between both building and soil can contribute to reduce building thermal gains or looses. Hygrothermal aspects can be related to many attributes such as energy consumption, occupants' thermal comfort and health, and material deterioration. Recently, a great variety of mathematical models to predict thermal and moisture content profiles in porous media have been presented in the literature. Most of those models are based on analysis of multilayer measurements or on Fourier analysis. The development and validation of such mathematical models facilitate the understanding of heat and moisture flows at different soil depths. In this research, a radial basis function neural network (RBF-NN) approach, combined with Gath–Geva clustering method in order to predict the temperature and moisture content profiles in soils, has been presented. A set of data obtained from the computation of the coupled heat and moisture transfer in porous soils for the Curitiba city (Paraná State, Brazil) weather data file has been used by the RBF-NN modeling method. Simulation results indicate the potentialities of the RBF-NNs to learn, for the one step ahead identification, the behavior of temperature and moisture content profiles in the media at various depths.