Environmental Engineering Reference
In-Depth Information
Figure 4.27: Experimental (dots) and simulated (continuous) data for hydrodemetallization
(HDM) catalyst (Boscan crude) [From ref.
290
. Reprinted with permission].
composed of many simulation elements interacting with bandwidth channels and/or
connections to process the information by responding to the external inputs. Initially, it can be
built using the historical plant data as well as the results of laboratory and pilot plant research
and those found in the literature. Gradually, the models can be upgraded by incorporating
more reliable and advanced information. The neural network architecture developed by Kam
et al.
[293]
consisted of the three neural layers and eight specified input nodes and nine hidden
nodes to predict the expanded bed height. A comparison of the normalized bed height data,
obtained from a commercial reactor with those predicted by the artificial neural network,
showed a good agreement. The absolute average deviation value indicated the applicability of
the model to the expanded bed systems as employed in the commercial H-Oil reactors.
Furthermore, the predicted slate of products and their yields were in the reasonable agreement
with the plant data.
