Towards a Generic Methodology to Model Solar Thermal Systems using Neural Networks Through a Short Dynamic Test
Abstract
Nowadays there is no global approach to model and characterize solar thermal systems for building application from experimental data. Results of the existing approaches are valid only for specific conditions (type of climate and thermal building properties). The aim of this study is to create a generic methodology to model such systems. Neural networks (NN) proved to be suitable to tackle similar problems particularly when the system to be modeled is compact and cannot be divided up during the testing stage. Reliable “black box” NN modeling is able to identify global models of the system without any advanced knowledge about its internal operating principle. The knowledge of the system global inputs and outputs is sufficient. Results concerning the solar combisystem modeling show that a dynamic NN model is efficient to learn the dynamic of the solar system especially due to the heat storage component. NN model developed is able to predict, with a good precision degree, the annual energy performance of the system based on a learning sequence of only 12 days. Because of the NN generalization ability, it is possible to predict the solar combisystem behavior when operating under environments different from the one used during learning stage and so to characterize its performances.
Keywords
Solar thermal systems, Neural networks, Characterization, Dynamic modeling, System testing