NAPHTHA CATALYTIC REFORMER HYBRID MODELING

Naphtha Catalytic reforming is a very important process because it is responsible for high 
percent of gasoline production in the petroleum refinery. Because of the complex nature of naphtha reforming 
process and its reaction chemistry, mathematical modeling of such a process with First Principle Method 
(FPM) will provide highly non linear relations. This nonlinearity of the developed model will reduce the 
prediction capability and the model results will deviate from the actual plant data.
A hybrid modeling (HYB) method adapted in this paper, which is the first principle model in parallel 
to neural network (ANN) model are used to simulate naphtha reformer. The aim of the HYB is to add the 
corrections from ANN to the deviated results obtained from FPM.
Two months of operational data of a Libyan refinery is used in this study to simulate the semi regenerative 
naphtha reforming reactor, each data point was properly prepared by HYSYS Software to be used as input 
to FPM model which is simulated by MATLAB Software. 
The paper focused on one output variable (reactor outlet temperature Tout) from the FPM which showed 
high deviation between the predicted and actual data.
Also same data is also used as input to ANN model, the data is normalized, outliers removed and verified 
with variance and correlation coefficient to select the proper input variables. Correlation coefficients between 
input and output showed high dependency between the proposed variables. Different architectures of ANN 
are carried out using MATLAB where mean square error (mse) was used as performance parameter. The 
results showed that neural network with structure [4-5-1] provided best performance. This ANN provided 
best residual predictions to reduce the deviations in reactor outlet temperature.