See

discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/251604637

Modeling and Optimization of an Industrial

Hydrocracker Plant

Article in Journal of Petroleum Science and Engineering September 2011

DOI: 10.1016/j.petrol.2011.07.019

CITATIONS READS

10 249

4 authors, including:

Gholamreza Zahedi Zainuddin A. Manan

Design Tanks LLC Universiti Teknologi Malaysia
97 PUBLICATIONS 1,443 CITATIONS 197 PUBLICATIONS 2,599 CITATIONS

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Sustainable Process Integration Laboratory (SPIL) View project

University 4.0 4th Industrial Revolution View project

All content following this page was uploaded by Gholamreza Zahedi on 01 April 2015.

The user has requested enhancement of the downloaded file.

 Journal of Petroleum Science and Engineering 78 (2011) 627636

Contents lists available at SciVerse ScienceDirect

Journal of Petroleum Science and Engineering

j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / p e t r o l

Modeling and optimization of an industrial hydrocracker plant

Ibrahim Alhajree b, Gholamreza Zahedi a,, Z.A. Manan a, Sasan Mohammad Zadeh c
a
Process Systems Engineering Centre (PROSPECT), Faculty of Chemical Engineering, Universiti Teknologi Malaysia, UTM Skudai, 81310 Johor Bahru, Johor, Malaysia
b
Department of Chemical Engineering, College of Technological Studies, Ofce: B7 144, P.O. Box: 42325, Shuwaikh 70654 - Kuwait
c
Department of chemical Engineering, Islamic Azad university of Ilkhchi, Ilkhchi, Iran

a r t i c l e i n f o a b s t r a c t

Article history: The main objective of this study is modeling and optimization of an industrial Hydrocracker Unit (HU) using
Received 8 April 2011 Articial Neural Network (ANN) model. In this case some data from an industrial hydrocracker plant were
Accepted 26 July 2011 collected. Two-thirds of the data points were used to train ANN model. Among the various networks and
Available online 9 August 2011
architectures, two multilayer feed forward networks with Back Propagation (BP) training algorithm were
found as the best model for the plant. Inputs of both ANNs include fresh feed and recycle hydrogen ow rate,
Keywords:
Hydrocracker unit
temperature of reactors, mole percentage of H2 and H2S, feed ow rate and temperature of debutanizer,
Renery pressure of debutanizer receiver, top and bottom temperature of fractionator column and pressure of
Modeling fractionator column. The rst network was employed to calculate the specic gravity of gas oil, kerosene, Light
Articial neural network Naphtha (LN), Heavy Naphtha (HN), gas oil and kerosene ash point and gas oil pour point. The second
network was used to calculate the volume percent of C4, LN, HN and kerosene, gas oil and fractionators
column residual (off test). Unseen data points were used to check generalization capability of the best
network. There were good overlap between network estimations and unseen data.
In the next step of study sensitivity analysis was carried out on plant to check the effect of input variables on
the plant performance. In this case temperature was found as the most affecting parameter in the plant.
Finally optimization was performed to maximize the volume percent of gas oil, kerosene, HN and LN
production and to identify the sets of optimum operating parameters to maximize these product yields.
Optimum conditions were found as feed ow rate of 113.2 m3/h, reactor temperature of 413 C, hydrogen
ow rate of 111.3 MSCM/h and LN () feed vol.% of 9.22.
2011 Elsevier B.V. All rights reserved.

1. Introduction 2006b). Heavy and time-consuming computations are sometimes

Hydro-cracking is a catalytic cracking process which has high
conversion. Cracking conversion of a hydrocracker unit (HU) is more
important than uid cat cracking units and hydrogen partial pressure
of the unit is higher than hydro-desulphurization processes (Maples,
2000).
In this process, hydrogen puries the hydrocarbon stream from
sulfur and nitrogen hetero-atoms. The process produces saturated
hydrocarbons. The main products for HU are jet fuel and diesel, while
also relatively high octane rating gasoline and LPG are produced. All
these products have a very low sulfur content and other contaminants
(Lyons, 2005).
The First Principle Models (FPMs) are common for the modeling
the HU, but the complete development of these models can be very
complex. Many industrial chemical and petrochemical processes are
complex in nature typically due to unknown reaction chemistry,
nonlinear relations and numerous involved variables (Bhutani et al.,
Corresponding author. Tel.: + 60 7 5535583; fax: + 60 7 5581463.
E-mail address: grzahedi@cheme.utm.my (G. Zahedi).
drawbacks of FPMs. Sometimes, in FPMs, complex partial differential
equations or complex algebraic equations appear which should be
solved analytically or numerically. Also, some phenomena, for
example, kinetics of HU reactions are still not well-understood to
develop an accurate mathematical model (Aguiar and Filho, 2001).
ANN modeling is a good alternative to FPMs to manage the
complexities mentioned since it only requires the input-output data
as opposed to a detailed knowledge of a system. In addition, ANN
requires less computational time and allows estimation of every
continuous and nonlinear function with high precision. because of
these features ANN is very popular for modeling, simulation,
optimization and control of processes in petrochemicals and reneries
(Al-Enzi and Elkamel, 2000; Bellos et al., 2005; Bollas et al., 2003;
Bollas et al., 2004; Falla et al., 2006; Shirvani et al., 2010; Zahedi et al.,
2006; Zahedi et al., 2010).
Al-Enez and Elkamel have focused on predicting the effect of
feedstock on the properties and the product yield for a Fluid Catalytic
Cracking (FCC) unit. They used feed forward ANN to predict the yield
of propane, butane, n-butane, iso-butane, propylene, butylene, light
gas, gasoline, light cycle oil, heavy cycle oil, coke and carbon
conradson number. Only four properties of feed including API,

0920-4105/$ see front matter 2011 Elsevier B.V. All rights reserved.
doi:10.1016/j.petrol.2011.07.019
628 I. Alhajree et al. / Journal of Petroleum Science and Engineering 78 (2011) 627636

Watson characterization factor, sulfur and volume conversion percent

Nomenclature
of liquid were introduced to NN as inputs. Their model gave better
prediction than models from non-linear regression and commercial
ak kth node activation
simulators (Al-Enzi and Elkamel, 2000).
BP Backpropagation
An ANN hybrid model was used by Bollas et al., 2003 to scale up a
CGF Fletcher-Powell Conjugate Gradient
FCC pilot plant into an industrial scale plant. The pilot model was able
f(x) The output of the node
to predict the weight percent conversion and the coke yield. The
fi Objective function for optimization
hybrid model was then compared with the pilot model and the pure
F Feed ow rate (make up and recycle)
ANN model. The results showed that the hybrid model has better
H Inlet ow rate of hydrogen
extrapolation capacity (Bollas et al., 2003).
HN Heavy Naphtha
Bellos et al., 2005 used a hybrid ANN model for hydrodesulphuriza-
HU Hydrocracker Unit
tion reactor modeling. They coupled the deterministic model which was
LN Light Naphtha
used to study the reactor performance and hydrogen consumption, with
Mea. Measured data
an ANN model which was able to evaluate the kinetic parameters. The
MLP Multi Layer Perceptron
obtained results showed that the hybrid model was capable of predicting
N Number of data
the reactor performance (Bellos et al., 2005).
Off test Residual Fractionator column
The crude oil description using the near infrared spectroscopy was
p The parameter vector (for sensitivity analysis)
performed by Falla et al., 2006. This analysis was called SimDis
RBF Radial Basis Function
(Simulation Distillation) and was faster than the true boiling point
SCG Scaled Conjugate Gradient
method. Forty oil samples with API of 1.31-36.4 were gathered. The ANN
Sim. Simulated data
was applied, which generated the SimDis curves accurately (Falla et al.,
tkp kth node target pattern
2006).
T Temperature of reactors
An ANN model was used by us for simulation of an industrial
wjk Weights
Hydrotreater Unit. (Zahedi et al., 2006). we used Radial Basis Function
x Sum of weighted inputs to the neuron
(RBF) modeling as an optimum architecture to predict hydrogen
y The state vector (for sensitivity analysis)
p demand, outlet API, and sulfur weight percent as a function of inlet
xki kth node input pattern
API and sulfur weight percent for seven different feed stocks. A
ykp kth node output pattern
comparison between our model with the obtained results of a
conventional simulator conrmed the superiority of the ANN model.
In another study we focused on enhancing the gasoline production
Greek symbols
of an industrial catalytic reformer unit. The ANN model anticipated
a Network learning rate
the unit outputs accurately and led to 2,38% increase in gasoline
k Hidden layer kth node error
production yield (Zahedi et al., 2008).
j Output jth node error
Aminian and Shahhosseini (2008) used ANN to predict the fouling
Transfer function
behavior of a crude oil preheat heat exchangers. They also used
wki The error gradient
sensitivity analysis known as the sequential zeroing of weights to
determine the effects of various parameters on fouling (Aminian and
Shahhosseini, 2008).

Fig. 1. Process ow diagram of a hydrocracking unit (Isomax unit).

 I. Alhajree et al. / Journal of Petroleum Science and Engineering 78 (2011) 627636 629

Table 1 in a petroleum rening process (Bhutani et al., 2006a, 2006b; Elkamel

The range of input variables of Tabriz HU. et al., 1999; Sullivan et al., 1989).
Quantity Value Modeling of the hydrocracking unit for light vacuum gas oil using
ANNs was studied by Elkamel et al. Their model was capable of
Fresh feed ow rate (m3/h) 90.1110.5
Recycle feed ow rate (m3/h) 22.27529.7 predicting the properties and the yield of iso-butane, normal butane,
Fresh hydrogen ow rate (m3/h) 2655035577 LN, HN, light ATK, heavy ATK, etc. In their proposed model, the mean
Recycle hydrogen ow rate (m3/h) 96390113220 percent of error between the obtained results and experimental data
Temperature of reactors for 8 point (C) 393430
was 8.71% (Elkamel et al., 1999).
H2S (Percent mol) 0.40.8
H2 (Percent mol) 7490 Bhutani et al. (2006a, 2006b) applied the FPM, data-based model
Debutanizer feed temperature (C) 169179 and three hybrid models including the series, parallel and series-
Debutanizer feed ow rate (m3/h) 105.6144 parallel models for the prediction of the steady state operation and
Pressure of debutanizer receiver (kg/cm2) 1213.5 optimization of an industrial HU. The data-based model and genetic
Temperature of top frctionator column (C) 115123
algorithm were used to optimize this unit. The results showed that
Temperature of bottom frctionator column (C) 296315
Pressure of frctionator column (kg/cm2) 0.50.7 yield of the desired products was improved by 416% by appropriate
changes in operating conditions (Bhutani et al., 2006b). In the same
year, Multi-objective optimization of an industrial HU was studied by
Table 2 Bhutani et al. The reactor model was simulated based on a discrete
The range of output variables of Tabriz HU. lumped model (Bhutani et al., 2006a).
Quantity Value This paper uses an ANN technique for modeling, sensitivity
C4 yield (LV %) 46
analysis and optimization of a HU process. The paper begins by the
HN yield (LV %)) 3340 development of a detailed industrial ANN model for HU. Next
Kerosene yield (LV %) 1015 sensitivity analysis is performed using ANN on HU to evaluate the
Gas oil yield (LV %) 3843 effect of input variables on outputs. Finally, optimization of the HU
LN yield (LV %) 810
process is performed with the objective of maximizing the volume
Off test yield (LV %) 35
Specic gravity of LN 0.6430.668 percent of gas oil, kerosene, HN and LN.
Specic gravity of HN 0.72250.7375
Specic gravity of kerosene 0.7820.808 2. HU (Isomax Unit)
Specic gravity of gas oil 0.8170.829
Flash point of kerosene (C) 4456
Fig. 1 shows process ow diagram of the Tabriz renery HU
Flash point of gas oil (C) 4860
Pour point of gas oil (C) 614 (Tabriz, 2009). The unit, which is commercially known as the Isomax
Unit has been designed to rene 18000 barrel/day of vacuum gas oil
and produce high quality middle-distillate products like naphtha,
kerosene and gas oil. This unit comprises of three sections: make-up
compressor section, reactor section and fractionation section. In
Design of the expert system for an industrial distillation column make-up compressor section, 96.5% hydrogen gas is compressed in
using NN and its optimization using genetic algorithm was performed three stages to increase the pressure of hydrogen gas from 15.5 kgf/
by Motlaghi Jalali and Nili Ahmadabadi, 2008. They used the operating cm 2 to 190 kgf/cm 2.
conditions and the product quality as ANN input and outputs The feed is mixed with bottom recycles and hydrogen gas and then
respectively. Then, the oil production cost function was minimized is introduced to the reactor. Reactions occur in a xed bed reactor
(Motlaghi Jalali and Nili Ahmadabadi, 2008). while after separation of light gases the products ow to the
In another attempt delayed coking unit was also modeled by us. fractionation section. The fractionation section includes debutanizer
we used both MLP and RBF NNs. The obtained results showed that the tower, fractionation tower, splitter and stripper. The reactor products
generalization capability of the RBF network in modeling of this unit are separated in this section to yield light gases (8255 kg/h) butane
was better than the MLP (Zahedi et al., 2009). (92 barrel/day), LN (1022 barrel/day), HN (1418 barrel/day), kero-
Recently o research group have successfully employed ANN in sour sene (9781 barrel/day) and gas oil (7456 barrel/day).
gas water content estimation and also efciency of sieve trays in
distillation column ( Shirvani et al., 2010; Zahedi et al., 2010). 3. Articial neural network
HU has been subject of some researcher on productivity
improvement due to the high demand for petroleum products, ANNs are inspired by biological nervous systems made up of
especially for HU products, which is one of the most important units simple elements operating in parallel manner. The network function

Fig. 2. Schematic representation of the multilayer feed forward neural network for (a) the rst network (SCG) and (b) the second network (CGF).
630 I. Alhajree et al. / Journal of Petroleum Science and Engineering 78 (2011) 627636

Fig. 3. Comparison of measurement and simulated results of the rst ANN model For (a) pour point of gas oil, (b) ash point of gas oil, (c) ash point of kerosene, (d) specic gravity
of gas oil, (e) specic gravity of kerosene (f) specic gravity of HN and (g) specic gravity of LN.

is determined by the connections between elements. An ANN can be
trained to perform as a particular function by adjusting the values of
the connections (weights) between elements, so that a particular
input leads to a specic target output. The network is adjusted, based
on a comparison between the output and the target, until the network
output matches the target. Many input-target pairs are needed to
train a network. NNs can be trained to solve problems that are difcult
for conventional computers or human beings (Choudhury et al., 1999;
Pastor-Brcenas et al., 2005; Kucuk and Derebasi, 2006).
ANNs are strong tools for complex and exible input-output
mappings (Pastor-Brcenas et al., 2005). Construction of a neural
network includes the determination of the network properties (the
 I. Alhajree et al. / Journal of Petroleum Science and Engineering 78 (2011) 627636 631

Fig. 4. Comparison of measurement and simulated results of the second ANN model For the volume percent of (a) C4, (b) LN, (c) HN, (d) kerosene, (e) gas oil and (f) off test.

network topology, the type of connections, the order of connections,

and the weight range), the node properties (activation range and the
Table 3 activation function), and the system dynamics (the weight initializa-
Comparison of simulated outputs of two ANN models and measured industrial plant tion scheme, the activation calculating formula and the learning rule)
data.
(Choudhury et al., 1999).
Output Relative error An ANN has one input layer, one output layer and one or more
(%) hidden layers. The weights are initialized to small random values
Specic gravity of LN 0.4488 (Kucuk and Derebasi, 2006).
Specic gravity of HN 0.2301 The process of determining the values for weights based on the
Specic gravity of kerosene 0.1489
experimental data is called learning or training. ANNs learn from
specic gravity of gas oil 0.1152
Kerosene ash point 0.3112 examples and exhibit generalization capability (Choudhury et al.,
Gas oil ash point 0.4816
Gas oil pour point 0.3958
C4 (LV %) 3.6349 Table 4
LN (LV %) 2.3865 The limits of operating parameters used in sensitivity analysis.
HN (LV %) 1.5409
Parameters Limits
Kerosene (LV %) 1.1866
Gas oil (LV %) 0.4847 Temperature of reactors 380440 (C)
Off test (LV %) 6.7332 Feed ow rate (make up & recycle) 93160 (m3/h)
Average error 1.3924% Inlet ow rate of hydrogen (make up & recycle) 90,000160,000 (m3/h)
632 I. Alhajree et al. / Journal of Petroleum Science and Engineering 78 (2011) 627636

Fig. 5. The effect of temperature on the volume percent of (a) gas oil, (b) kerosene, (c) HN and (d) LN.

1999).The error between the actual and predicted outputs is used to
modify the weights. This training method which uses the set of input-
output (examples) is known as supervised training. Multi-layer
perception (MLP) is the most popular feed forward network for
prediction, so that all signals ow in a single direction from the input
to the output (Kucuk and Derebasi, 2006; Reddy et al., 2006).
In MLPs, the back propagation (BP) algorithm for supervised
training is often adopted to propagate the error from the output
neurons and modify the weight for the neurons in the hidden layers
(Kucuk and Derebasi, 2006).
The error gradient (wki) for the kth output or hidden node is
given by (Kucuk and Derebasi, 2006).
p obtained ANN.
wki = ak xki
p within each layer trial and error approach was used. The trial and
where, a, xki are learning rate and kth node input pattern respectively.
k is hidden layer kth node error:
k = ak j wj k
j = Ik
For output node error can be calculated based on the following
equation:
p p

k = ak tk yk
p p
where tk is kth node target pattern, yk is kth node output pattern
(Kucuk and Derebasi, 2006).
4. ANN modeling
4.1. Industrial data sets and modeling procedure
To build up an ANN for simulating the HU, necessary data were
provided. Among the available data from Tabriz renery HU, 108 data
were screened based on principle component analysis (Reddy et al.,
2006). MATLAB software was employed to build the ANN models
(Demuth and Beale, 2010).
Various architectures of MLP, RBF and BP were employed to model
the unit. For scaling purposes, each type of input and output data were
normalized between 0 to 1. Each ANN was trained with 2/3 of data set
and the remaining 1/3 of samples was employed for testing the
1
In order to nd the optimum number of hidden layers and nodes
error approach includes one, two and three hidden layers, and 40 to
120 nodes per each hidden layer. Tables 1 and 2 show HU variables
and their operating range.
2
After the trial and error it was found that a single ANN can
simulate the plant very well. Therefore based on our previous
experience (Zahedi et al., 2008), two ANNs in parallel were used to
model the plant.
In the rst ANN, a BP network with the SCG algorithm was found as
the best network for calculation of the specic gravity of gas oil,
3 kerosene, LN, HN and the ash point of gas oil and kerosene and gas oil
pour point (seven output vector).
In the next ANN, the volume percent of C4, LN, HN, kerosene, gas oil
and residual fractionator column (off test) (6 output vectors) were
 I. Alhajree et al. / Journal of Petroleum Science and Engineering 78 (2011) 627636
Fig. 6. The effect of total feed ow rate (make up and recycle) on the volume percent of (a) gas oil, (b) kerosene, (c) HN and (d) LN.
calculated using BP network with Fletcher-Powell Conjugate Gradient
(CGF) algorithm. Both networks had the same inputs (20 input).
It is important to nd the optimum number of nodes in the hidden
layer which provide good estimates of the outputs. The rst network
was designed with 74 and 74 neurons in the rst and the second
hidden layers, respectively, while, in the second ANN, there were
three hidden layers with 61, 71 and 61 neurons, respectively.
In training the networks, the sigmoid and simple linear activation
functions were used in the hidden layers and output layers:
1 models can model the unseen data with excellent precision and can be
f x = Logistic sigmoid function
1 + ex
f x = x Linear function
Where x is the sum of weighted inputs to the neuron and f(x) is the
output of the node. In test or generalization step the relative error was
calculated based on the following equations:
mean relative error percent =
N 1
j
1 N Measured dataSimulated data
Measured data
100 6
The schematic of both designed networks is depicted in Fig. 2.
633
4.2. ANN modeling results
As mentioned before, two ANN models were developed for HU
modeling. The generalization results of the rst and second networks
are illustrated in Figs. 3 and 4, respectively. These gures show the
ANN outputs versus the number of unseen industrial data.
These gures show good capability of both ANNs for estimation of
unseen data. Mean relative percent of error between unseen plant
data and network outputs was obtained as 1.3924%. Details are
summarized in Table 3. According to Table 3, the proposed ANN
4 employed for optimization and further assessment of the plant.
4.3. Sensitivity analysis
5
The purpose of this section is to perform sensitivity analysis approach
to evaluate the effects of input variables on HU plant production. In other
words, sensitivity analysis was used to determine how the model would
respond to changes in input parameters (Jaiswal et al., 2005). The
following equation can be used to quantify sensitivity:
j y
= 7
p
where, p is the parameter vector which is mapped into the state vector y
(p y). In current study the trained ANNs model are utilized as HU
634 I. Alhajree et al. / Journal of Petroleum Science and Engineering 78 (2011) 627636

Fig. 7. The effect of total ow rate of inlet H2 (make up and recycle) on the volume percent of (a) gas oil, (b) kerosene, (c) HN and (d) LN.

simulator. Sensitivity analysis of a system was conducted using the trained from the trained ANN model by observing the change in the value of the
NNs to characterize the relationship between inputs and outputs. The kth output brought by a small change in the value of the jth input at any
sensitivity of the outputs to the inputs dened by Eq. (7) was estimated operating point (Reddy et al., 2006).
At each operating point P, the j th input was varied over its entire
Optimum operating parameters to range, and the corresponding variation in the kth output can be
maximize LN (volume %) plotted to produce a sensitivity curve (Reddy et al., 2006). In this way,
the contribution of that variable to the network response is excluded
(Aminian and Shahhosseini, 2008).
9.4 Using the second designed network, the effect of the operating
parameters of HU, like temperature of reactors, volume ow rate of
9.2
feed (make up and recycle) and volume ow rate of hydrogen (make
9 up and recycle), on volume percent of main outputs including gas oil,
kerosene, LN and HN at reactors pressure of 185 kg/cm 2 were studied.
8.8
Table 4 shows domain of change for operating parameters.
Light Naphtha %

8.6 Fig. 5 illustrates the effect of temperature on the gas oil, kerosene,
HN and LN percent while feed ow rate and inlet hydrogen ow rate
8.4
are set in their average values. As shown in this gure, temperature is
8.2 the most affecting input variable. As reactor temperature changes from
380 to 440C, the trend of the volume percent of gas oil, kerosene, HN
8
and LN changes according to Fig. 5 (a)(d), respectively.
7.8
Temperature+300
7.6 Hydrogen*1000 Table 5
Feed Optimum parameters.
7.4
100 105 110 115 120 125 130 Feed Temperature of reactors Hydrogen rate LN
Temperature, Hydrogen and Feed (m3/h) (C) (m3/h) (vol.%)

113.6 418 111300 9.22

Fig. 8. The effect of temperature, feed and H2 ow rate on the volume percent of LN.
 I. Alhajree et al. / Journal of Petroleum Science and Engineering 78 (2011) 627636
Table 6
Comparison between the optimum and the lower limit of the current operating
parameters.
Product Feed Temperature of reactors Hydrogen rate
name (%) (%) (%)
Gas oil 6.74 4.33 16.14
Kerosene 10.34 8.40 2.80
HN 4.78 4.83 14.51
LN 1.09 6.36 9.47
Similar to temperature, the effect of total feed ow rate (make up
and recycle) on the volume percent of gas oil, kerosene, HN and LN is
evaluated and the obtained results are shown in Fig. 6. In this gure, feed
ow rate (make up and recycle) is changed from 93 to 160 m 3/h, while
the temperature of reactors and hydrogen inlet ow rate are set in their
average values. As illustrated in Fig. 6, the more feed ow rate, the more
is kerosene and HN production while LN and gas oil show decrease.
These results will help to obtain the optimum values of these products.
Similarly, the effect of hydrogen ow rate (make up and recycle) on
the volume percent of gas oil, kerosene, HN and LN are shown in Fig. 7. The
variation of the hydrogen ow rate is between 90,000 and 160,000 m3/h.
As shown in this gure, increasing hydrogen ow rate increases kerosene
and LN production while gas oil and HN production decreases.
4.4. Optimization of HU
The study on thermodynamic and kinetic interactions indicated that
any increase in temperature and decrease of pressure have a positive
effect and an increase in space velocity (feed volume ow rate) has a
negative effect on the main reactions in HU (Maples, 2000) Volume ow
rate of hydrogen is pre-determined by operating conditions. The main
variables in HU are temperature and pressure of reactors, volume ow
rate of feed, feed molecular weight and ow rate of hydrogen.
Optimization problems consist of three basic steps: 1. Specifying
an objective function 2. Determining constraints in the system, which
include a set of unknowns or variables which control objective
function value. 3. A set of constraints that allow the unknowns to take
on certain values but exclude others (Edgar et al., 2001).
Many parameters like temperature, pressure and raw material
ow rates can be considered in optimization of processes. In order to
maximize production rate, the optimum values of these parameters
should be obtained. In this case, for HU production the pressure is
assumed to be constant. Objective function was set to maximize the
volume percent of gas oil, kerosene, HN and LN based on interest and
demand on by Tabriz renery.
Temperature of reactors, feed ow rate (makes up and recycles)
and inlet ow rate of hydrogen (make up and recycle) were
considered as optimization variables. The MATLAB optimization
toolbox was applied to optimize the volume percent of the main
products of HU, as shown in Eq. (8). The changes are limited because
of industrial constrains.
max f1 T; F; H i = 1; 2; 3 and 4
f1 = gas oilvol%
f2 = kerosenvol% 8 Optimization of an Industrial Hydrocracking Unit. Ind. Eng. Chem. Res. 45,
f3 = HNvol
f4 = LNvol%
In this study four objective functions have been introduced. fi is the
volume percent of gas oil, kerosene, HN and LN obtained from the
second ANN model. T is the temperature of reactors, F, feed ow rate
(make up and recycle) and H is the inlet ow rate of hydrogen (make
up and recycle). For maximizing each objective function, the optimum
operating parameters of T, F and H are dened. Fig. 8 illustrates the
effect of temperature, ow rate of feed and H2 on the volume percent
of gas oil, kerosene, HN and LN.
635
The optimum values in order to maximize gasoil, kerosene, HN and
LN, are tabulated in Table 5.
In order to know effect of optimization on the Plant production, a
comparison between the optimum operating parameters and current
industrial conditions was made. The results have been summarized in
Table 6.
According to Table 6, for the maximum production of gas oil, the
consumed feed ow rate and hydrogen ow rate have been reduced
by 6.74% and 16.14% respectively compared to the current values. On
the other hand, for the maximum production of HN, the consumed
feed ow rate and hydrogen ow rate values have been reduced by
4.78% and 14.51% respectively. The consumed value of hydrogen ow
rate has been reduced by 2.80% and 9.47% for the maximum
production of kerosene and LN, respectively. The obtained results
were reported to the plant for application.
5. Conclusion
In this paper, the ANN technique was used to simulate and
optimize an industrial HU. For this purpose, ANN model of the unit
was developed based on industrial data.
Using the designed networks, a sensitivity analysis was performed
to study the effect of the operating parameters on the volume percent
of output gas oil, kerosene, LN and HN at a reactor pressure of 185 kg/
cm 2. Increasing of hydrogen ow rate increases kerosene and LN.
Increase of feed ow rate causes the increase of kerosene and HN.
Finally, the obtained networks were applied to predict plant
optimum operating conditions to maximize the volume percent of gas
oil, kerosene, HN and LN as objective functions. This study again
emphasizes ANN capability to increase productions and keep plants
running in more economical conditions. The ability of ANN in
developing models for very complicated plant and analyzing the
plant was conrmed. Based on current increasing fuel price the
developed model easily can give us optimum operating condition of
the HU while this is very difcult to achieve using traditional model.
Acknowledgement
The authors would like to acknowledge Tabriz renery R&D unit
for providing industrial data. The rst author would like to
acknowledge The Public Authority for Applied Education and Training
(PAAET), with Project No: TS-10-02.
References
Aguiar, H.C., Filho, R.M., 2001. Neural network and hybrid model: a discussion about
different modeling techniques to predict pulping degree with industrial data.
Chem. Eng. Sci. 56, 565570.
Al-Enzi, G., Elkamel, A., 2000. Predicting the effect of feedstock on product yields and
properties of the FCC process. Petrol. Sci. Technol. 18 (34), 407428.
Aminian, J., Shahhosseini, Sh., 2008. Evaluation of ANN modeling for prediction of crude
oil fouling behavior. Appl. Therm. Eng. 28, 668674.
Bellos, G.D., Kallinikos, L.A., Gounaris, C.E., Papayannakos, N.G., 2005. Modeling of the
performance of industrial HDS reactors using a hybrid neural network approach.
Chem. Eng. Process. 44, 505515.
Bhutani, N., Ray, A.K., Rangaiah, G.P., 2006a. Modeling, Simulation, and Multi-objective
13541372.
Bhutani, N., Rangaiah, G.P., Ray, A.K., 2006b. First-principles, data-Based, and hybrid
modeling and optimization of an industrial hydrocracking unit. Ind. Eng. Chem. Res.
45 (23), 78077816.
Bollas, G.M., Papadokonstantakis, S., Michalopoulos, J., Arampatzis, G., Lappas, A.A., Vasalos,
I.A., Lygeros, A., 2003. Using hybrid neural networks in scaling up an FCC model from a
pilot plant to an industrial unit. Chem. Eng. Process. 42, 697713.
Bollas, G.M., Papadokonstantakis, S., Michalopoulos, J., Arampatzis, G., Lappas, A.A., Vasalos,
I.A., Lygeros, A., 2004. A computer-aided tool for the simulation and optimization of
the combined HDS-FCC processes. Chem. Eng. Res. Des. 82 (A7), 881894.
Choudhury, S.K., Jain, V.K., Rama Rao, Ch.V.V., 1999. On-line monitoring of tool wear in
turning using a neural Network. Int. J. Mach. Tools Manuf. 39, 489504.
Demuth, H., Beale, M., 2010. User's Guide: Neural Network Toolbox for Use with Matlab,
the Mathworks Inc.
636 I. Alhajree et al. / Journal of Petroleum Science and Engineering 78 (2011) 627636
Edgar, T.F., Himmelblau, D.M., Lasdon, L.S., 2001. Optimization of Chemical Processes,
second ed. McGraw Hill, New York.
Elkamel, A., Al-Ajmi, A., Fahim, M., 1999. Modeling the hydrocracking process using
articial neural networks. Petrol. Sci. Technol. 17 (910), 931954.
Falla, F.S., Larini, C., Le Roux, G.A.C., Quina, F.H., Moro, L.F.L., Nascimento, C.A.O., 2006.
Characterization of crude petroleum by NIR. J. Petrol. Sci. Eng. 51, 127137.
Jaiswal, S., Benson, E.R., Bernard, J.C., Van Wicklen, J.L., 2005. Neural Network Modelling and
Sensitivity Analysis of a Mechanical Poultry Catching System. Biosystems Eng. 92 (1),
5968.
Kucuk, I., Derebasi, N., 2006. Sensitivity analysis for estimation of power losses in
magnetic cores using neural network. J. Phys. Chem. Solids. 67, 24732477.
Lyons William, C., 2005. second edition. Standard Handbook of Petroleum and Natural
Gas Engineering, Volume 2. GULF PROFESSIONAL PUBLISHING, USA.
Maples, R.E., 2000. Petroleum Renery Process Economics, second edition.
Motlaghi Jalali, F., Nili Ahmadabadi, M., 2008. An expert system design for a crude oil
distillation column with the neural networks model and the process optimization
using genetic algorithm framework. Expert Systems with Applications. 35,
15401545.
Pastor-Brcenas, O., Soria-Olivas, E., Martn-Guerrero, J.D., Camps-Valls, G., Carrasco-
Rodrgueza, J.L., del Valle-Tascn, S., 2005. Unbiased sensitivity analysis and
View publication stats
pruning techniques in neural networks for surface ozone modeling. Ecol. Model.
182, 149158.
Reddy, N.S., Lee, C.S., Kim, J.H., Semiatin, S.L., 2006. Determination of the beta-approach
curve and beta-transus temperature for titanium alloys using sensitivity analysis of
a trained neural network. Mater. Sci. Eng. A. 434, 218226.
Sullivan, R.F., Boduszynski, M.M., Fetzer, J.C., 1989. Molecular transformations in
hydrotreating and hydrocracking. Energy Fuels. 3 (5), 603612.
Tabriz, Renery, 2009. hydrocracker unit data.
Shirvani, Yazdan, Zahedi, Gholamreza, Bashiri, Mohsen, 2010. Estimation of Sour
Natural Gas Water Content. J. Pet. Sci. Eng. 73, 156160.
Zahedi, G., Lohi, A., Karami, Z., 2009. A neural network approach for identication and
modeling of delayed coking plant. Int. J. Chem. Reactor Eng. 7, A16.
Zahedi, G., Mohammadzadeh, S., Moradi, G., 2008. Enhancing gasoline production in an
industrial catalytic-reforming unit using articial neural networks. Energy Fuels.
22, 26712677.
Zahedi, G., Fgaier, H., Jahanmiri, A., Al-Enezi, G., 2006. Articial neural network identication
and evaluation of hydrotreater plant. Petrol. Sci. Technol. 24, 14471456.
Zahedi, G., Parvizian, F., Rahimi, M.R., 2010. An expert model for estimation of
distillation sieve tray efciency based on articial neural network approach.
Journal of Applied Science 10( (12), 10761082.