A novel ANN-based approach to estimate heat transfer coefficients in radiant wall heating systems

This paper includes the validation of ANN solutions by reliable experiments to research the heat transfer characteristics in an actual size room in a laboratory. Experimental tests have been done in an experimental chamber at a constant height and a floor area. Heating through three different wall configurations is implemented during experiments. Furthermore, various ANN techniques in Matlab are employed to study the thermal behaviors of the problem with regard to the alterations of heat transfer coefficients. Backpropagation learning methods of Levenberg-Marquardt, Bayesian regularization, resilient backpropagation and scaled conjugate gradient with multilayer perceptron network are used in order to show the artificial intelligence's predictability area. Reference temperatures for corresponding heat transfer coefficients, heated wall temperatures and supply water temperatures are assigned as input variables, while convective, radiative and total heat transfer coefficients are defined as outputs. In conclusion, developed and detailed ANN model predicted heat transfer coefficients very successfully in tolerable deviation proportions from experimental findings. Also, the influence of supply water temperature on these coefficients was revealed. Moreover, the estimations of the ANN approach have been compared with the radiant heating and cooling data in the literature and a strong consistency has been noticed.