Neural network modeling of structured packing height equivalent to a theoretical plate

The height equivalent to a theoretical plate (HETP) of nine types of struct ured packing was successfully modeled using a neural network. The network w as trained on data similar to that used to develop semiempirical mass-trans fer models. The HETP was then predicted using the trained network. The neur al network model yields a very accurate prediction of experimentally determ ined HETP values, and it is more accurate than a traditional semiempirical model. Using the neural network, it is also possible to rank the relative i mportance of input variables in determining the HETP. In particular, this w ork shows that the roughness of the structured packing surface is a very im portant input parameter.