A comparison between state-of-the-art and neural network modelling of solar collectors

作者：Fischer Stephan^*; Frey Patrick; Druck Harald

来源：Solar Energy, 2012, 86(11): 3268-3277.

DOI：10.1016/j.solener.2012.09.002

• Solar collector
• TRNSYS
• Artificial neural network
• Collector testing
• Dynamic system simulation
• Parameter identification

The state-of-the-art modelling of solar collectors as described in the European Standard EN 12975-2 is based on equations describing the thermal behaviour of the collectors by characterising the physical phenomena, e.g. transmission of irradiance through transparent covers, absorption of irradiance by the absorber, temperature dependent heat losses and others. This approach leads to so called collector parameters that describe these phenomena, e.g. the zero-loss collector efficiency eta(0) or the heat loss coefficients a(1) and a(2). %26lt;br%26gt;Although the state-of-the-art approach in collector modelling and testing fits most of the collector types very well there are some collector designs (e.g. %26quot;Sydney%26quot; tubes using heat pipes and %26quot;water-in-glass%26quot; collectors) which cannot be modelled with the same accuracy than conventional collectors like flat plate or standard evacuated tubular collectors. The artificial neural network (ANN) approach could be an appropriate alternative to overcome this drawback. %26lt;br%26gt;To compare the different approaches of modelling investigations for a conventional flat plate collector and an evacuated %26quot;Sydney%26quot; tubular collector have been carried out based on performance measurements according to the European Standard EN 12975-2. The investigations include the parameter identification (training), the comparisons between measured and modelled collector output and the simulated yearly collector yield for a solar domestic hot water system for both models. %26lt;br%26gt;The obtained results show better agreement between measured and calculated collector output for the artificial neural network approach compared with the state-of-the-art modelling. The investigations also show that for the ANN approach special test sequences have to be designed and that the determination of the ANN that fits the thermal performance of the collector in the best way depends significantly on the expertise of the user. %26lt;br%26gt;Nevertheless artificial neural networks have the potential to become an interesting alternative to the state-of-the-art collector models used today.

• 出版日期2012-11