A neural classifier for the optimal selection of conduction transfer functions

The transfer function method (TFM) is a very spread approach for the solution of heat transfer problems in building envelopes. However, the reliability of the simulation can be significantly affected by the choice of the set of coefficients for the conduction transfer functions (CTFs), especially in presence of very massive walls. This paper discusses a neural network classifier able to assess the reliability of CTFs on the basis of a specific evaluation parameter based on a comparison with the result obtained by the Fourier analysis. The model, trained with several thousands of samples obtained by using a software created by the authors, performs very well, as it is showed by the confusion matrix presented in the paper. It represents a very useful tool for the HVAC plants designers because it allows the selection of the best CTFs for the computation of the heat flows though multilayer walls before running simulations.