Optimization Model for Refinery Hydrogen Networks Part I

Enrique E. Tarifa et al. Int. Journal of Engineering Research and Application www.ijera.com
ISSN : 2248-9622, Vol. 6, Issue 10, ( Part -2) October 2016, pp.84-88
www.ijera.com 84|P a g e
Optimization Model for Refinery Hydrogen Networks
Part I
Enrique E. Tarifa1*
, Carlos R. Vera2
, Samuel FrancoDomínguez3
,
Leonel A. Benitez4
(1)*
Consejo Nacional de InvestigacionesCientíficas y Técnicas (CONICET), Facultad de Ingeniería, Universidad
Nacional de Jujuy (UNJu). ÍtaloPalancaN°10 - San Salvador de Jujuy (4600), Argentina.
(2)
Consejo Nacional de InvestigacionesCientíficas y Técnicas (CONICET). INCAPE, Facultad de Ingeniería
Química, Universidad Nacional delLitoral (UNL). Santa Fe(3000), Argentina.
(3)
Facultad de Ingeniería Universidad Nacional de Jujuy (UNJu). ÍtaloPalancaN°10 - San Salvador de Jujuy
(4600), Argentina.
(4)
Instituto de Investigaciones para la IndustriaQuímica - INIQUI (CONICET-UNSa), Facultad de Ingeniería,
Universidad Nacional de Salta (UNSa). Avda. Bolivia Nº 5150 - Salta (4400), Argentina.
ABSTRACT
Petroleum refineries have many process units that consume hydrogen.These process units are distributed in
different places everywhere in the refinery.In order to feed them, it is necessary to have sources capable of
supplying, in amount and quality, the hydrogen that every consuming unit needs.It is also needed to have a
distribution network that it is correctly designed and which operation is adjusted in an optimal manner to the
changing conditions of the refinery.This involves the minimization of the cost of installation and operation of
the hydrogen network.The installation cost is dominated by the amount of pipelines, compressors and purifying
units; while the cost of operation is dominated by the amount of fresh hydrogen that the plant consumes.In this
work a mathematical model is developed for a hydrogen network,which is adapted to the different information
levels available in the different stages of design of that system.The model is currently in use in the YPFLuján de
Cuyo refinery (Mendoza, Argentina). In this first part, the basic model is presented; whereas in a second part,
the model is enlarged to accommodate the incorporation of purifying units and new compressors.
Keywords–Integration in Hydrogen Networks, LINGO, Optimization, Refinery Hydrogen
Management,Refinery Hydrogen Networks, RefineryHydrogen Pinch.
I. INTRODUCTION
In the petroleum refineries and the
petrochemical complexes, a great amount of units
consuming hydrogen exist, such as hydrotreaters,
hydrocrackers, isomerization units and lube refining
units.Hydrogen production units also exist, such as
the catalytic naphtha reformers and the hydrogen
plants.In order to take the hydrogen from the source
points to the point where hydrogen is consumed, it is
necessary to have a distribution network.This
distribution network must be adequately designed
and must also be adequately operated in order to
supply the amount and quality of hydrogen required
by every consuming unit.An optimally designed and
operated network will demand a minimum amount
of fresh hydrogen (make-up). With this purpose, it
will minimize the amount of hydrogen leaving the
network (off-gas) and it will maximize the amount
of recycled hydrogen.
Any optimization study of a hydrogen
network must begin with an analysis of the hydrogen
pinch.Through the systematic analysis of offer
(sources) and demand (sinks), the hydrogen pinch
analysis tries to minimize the flowrates of the make-
up of fresh hydrogen and the discharges of off-
gas.For this purpose, the study maximizes the
amount of recovered and reused hydrogen, though
the recovery might demand the purification of off-
gas hydrogen.
The first step in the determination of the
hydrogen pinch is the calculation of the mass
balance of the hydrogen sinks and sources of the
network. The hydrogen sources comprise the sources
of fresh hydrogen (make-up), recycle streams,
streams issuing from hydrogen produces (e.g. the
naphtha reformer), product and residue streams of
hydrogen purifiers (e.g. membrane separators and
PSA units), off-gas streams of low and high pressure
separators and off-gas streams of consuming units
(e.g. hydrotreaters and hydrocrackers). The
hydrogen demand comprises the streams entering the
hydrogen consuming units and the purifying
units.For each stream the flowrate and purity are
indicated.Molar or STP flowrates must be used for
unambiguous calculations.
RESEARCH ARTICLE OPEN ACCESS

Basically, the hydrogen pinch is the purity at which
the hydrogen network has no surplus nor deficit of
hydrogen.The pinch represents the bottleneck of the
network or how much hydrogen can be recovered
and reused.The traditional approach to the hydrogen
pinch is graphic, and does not consider the pressure
of each current [1]-[4]. For this reason the
theoretical hydrogen consumption determined by
this pinch analysis is the theoretical minimum
consumption, whichcan be used to take decisions at
an early stage of the design.The changes needed in
the real network for achieving this minimum can be
as simple as the opening and closing of some valves,
or as costly as the installation of a multistage
compressor for connecting a low pressure source to a
high pressure sink.An intermediate change can be
the implementation of cascaded connections
between the purge of one unit and the make-up of
other one [5].
In order to take into account the pressure in
the management of the hydrogen network,
optimization techniques can be applied, which
minimize the costs of installation as well as the costs
of operation [6]-[8].
A model of optimization has the objective
of finding the best solution for a given problem.The
model of optimization is composed of decision
variables, an objective function and the
restrictions.The decision variables are the variables
that can be changed in order to find the best
solution.In this search, the decision variables must
respect the conditions imposed by the restrictions of
the problem.The goodness of the explored
alternatives is measured by the objective
function.The best alternative will be the one that
minimizes or maximizes the objective function.
In this work, a mathematical model is
presented for a hydrogen network. This model is
adapted to the different levels of information
available as a progress is made during the design of
a hydrogen network.The model is currently being
used by the YPFLuján de Cuyo refinery (Mendoza,
Argentina).The optimization model was
implementedby using Excel and LINGO.The Excel
spreadsheet enables entering the data in an easy way
and also shows the results obtained.The
mathematical model is solved by using LINGO, a
commercial optimization software package.
II. MODEL TO MINIMIZE THE
DEMAND OF HYDROGEN
2.1 Formulation of the model
In this first model, called Min Fg, the
objective function is the demand of hydrogen service
(amount of fresh or make-uphydrogen), and the
restrictions take into account the pressure levels and
the capacities of the compressors in order to
determine the feasibility of the flows in the network.
The model determines the connections to be made,
together with the flowrates and the purities of the
streams of the network that make minimum the
demand of hydrogen. In favorable conditions, the
demand can be reduced to the minimum level
determined by the pinch analysis.
The formulation of the model demands the definition
of the following sets of elements:
 N: set of source and sink nodes that belong to
the network.
Fig. 1 shows a simplified scheme of a
hydrogen consuming unit. In this diagram the right
location of the sink and source nodes is defined.
Reactor
Separator
Source
Liquid product
PurgeGas recycle
Feed
Make-up
Sink
Figure 1. Simplified scheme of a hydrogen consuming
unit showing the correct location of the source and sink
nodes.
Each source of the N set has a process
nucleusand an output splitter (Fig. 2), whereaseach
sink of the N set is formed by an input mixer and a
process nucleus (Fig. 3). Each node thus defined has
the following attributes:
 Type:
 FUE: source
 SUM: sink
 Unit: identifying code of the piece of equipment
to which the node belongs.
 Class: defines the type of unitto which the node
belongs, such as:
 UP: process unit.
 COM: compressor. It is model as a combination
of a sink and a source with only one input
stream and only one output stream.
 SEP: separator.
 GEN: pure generator, only has source nodes.
 CON: pure consumer, only has sink nodes.
 Pn: absolute pressure of the nucleus (kgf/cm2
).
 Fn: flowrate of the stream connected to the
nucleus (Nm3
/h).
 yn: hydrogen purity of the stream connected to
the nucleus (molar fraction).
 Femax: maximum input flowrate (maximum
compatible with the capacity of the piece of
equipment or the unit). This attribute is only
valid for sinks of units with COM class.

Nucleus
Fn, yn, Pn
Figure 2. Structure of a source node.
Nucleus
Fn, yn, Pn
Figure 3. Structure of sink node.
From the N set, the following subsets are defined:
 CF: set of source nodes.
 CS: set of sink nodes.
The data required by the sources are the following:
 From UP class units: Pn, Fn, yn
 From COM class units: Pn
 From SEP class units: Pn, Fn, yn
 From GEN class units: Pn, yn
The data required by the sinks are the following:
 From UP class units: Pn, Fn, yn
 From COM class units: Pn, Femax
 From SEP class units: Pn, Fn, yn
 From CON class units: Pn
For convenience, the derived set FxS is
defined with elements (i, j), which are ordered pairs
of the source-sink type:
 , ,i j F xS i C F j C S    (1)
The elements of this set have the following
attributes:
 F: flowrate of the stream that goes from the
sourcei to the sinkj (Nm3
/h).
 y: hydrogen purity of the stream that goes from
the sourcei to the sinkj (molar fraction).
The derived FxSP set is also defined. Its elements
are the elements of the FxS set that represent
connections between nodes that do not belong to a
same unit and that are feasible due to the pressure
difference between the origin and the end of the
connection:
   
   
, ,
| i j i j
i j F xSP i j
F xS U nit U nit P n P n
  
  
(2)
That is, ordered pairs are defined of the
source-sink type, where the first condition prevents
the connection of the source of one units with the
sink of the same units. The last condition selects the
connections that are feasible from the point of view
of the pressures of the connected nodes.
The objective function is the total demand
of hydrogen service, Fg. The optimization problem
is written for the minimization of the objective
function by varying the hydrogen flowrates of the
existing streams among the nodes of the network.
The corresponding mathematical model is written
below:
 
   
, ,, , , , ,
M in
, , | G E N ,
| C O N C O M
i j i j k l lF y F n F n yn F g
k
l l
F g
i j F xS P k N C la ss
l N C la ss C la ss
  
   
(3)
The objective function is defined as:
| G E Ni
i i
i C F C la ss
F g F n yn
 
  (4)
On the other side, mass balances must be written
forthe splitter of each source k and for the mixer of
each sink l:
 
,
, |
i j k
i j F xSP i k
F F n k C F
 
  (5)
 
,
, |
i j l
i j F xS P j l
F F n l C S
 
  (6)
The purities of the streams must comply with the
following restriction:
 Definition of the molar fraction:
 ,
0 1 ,i j
y i j F xS P   (7)
 Hydrogen balance in the source splitters:
 ,
,i i j
yn y i j F xS P  (8)
 Balance of hydrogen in the mixers of the
sinks:
 
, ,
, |
i j i j l l
i j F xSP j l
F y F n yn l C S
 
  (9)
For the compressors these additional restrictions are
written:
   
 
, |
C O M
i j
i j
i
F n F n
i j F xS U n it U n it
C la ss

 
 
(10)
   
 
, |
C O M
i j
i j
i
yn yn
i j F xS U n it U n it
C la ss

 
 
(11)
| C O Mj j j
F n F em ax j C S C lass   (12)
Also, all the unfeasible connections must be
eliminated:
   ,
0 , | ,i j
F i j F xS i j F xS P   (13)
   ,
0 , | ,i j
y i j F xS i j F xS P   (14)
For the same reason, for all nodes that are not entries
of compressors, Femax must be annulled:
   
0
| C O M
k
k
F em a x
k N k C S C la ss

   
(15)

Finally, the purities of the streams that begin in
sources of constant purity must be fixed:
   ,
, | C O Mi j i i
y yn i j F xS P C la ss   (16)
2.2 Implementation
The model thus described was implemented
in the LINGO software environment. In order to
ease the data input and the reading of the results, an
Excel spreadsheet was developed that used many
sheets and a color coding. Blue sheets were
restricted to data input, whereas the sheet with
salmon color where used by LINGO to output the
optimization results.
2.3 Example I
Fig. 4 presents the initial configuration of
the plant to be analyzed in this example. The
example was taken from a work of Hallale and Liu
[6]. The pinch analysis for this plant reports that the
minimum production of the hydrogen plant is 182.8
MMscfd. This means a potential saving of 8.6 %
with respect to the initial 200 MMscfd. However, the
employed method only considers the flowrates and
the purities, leaving aside the pressures. Therefore,
the obtained result can be considered as an inferior
limit for the production of the hydrogen required by
the plant. Fig. 5 shows an implementation of the
solution reported by the pinch method. This solution
is not feasible because it suggests the connection
between the output of A and the input of B, which is
not possible due to the existing pressure difference.
A
B
H2 Plant Fuel
91%
1500 psi
300 psi
85%
1700 psi
40 MMscfd
10 MMscfd
310 MMscfd
490 MMscfd
1600 psi
2200 psi
90 MMscfd
110 MMscfd
99%
300 psi
200 MMscfd
Figure 4. Initial configuration of the plant of Example I.
A
B
H2 Plant
Fuel
91%
1500 psi
300 psi
85%
1700 psi
32.9 MMscfd
310 MMscfd
467.1 MMscfd
1600 psi
2200 psi
90 MMscfd
92.9 MMscfd
99%
300 psi
182.9 MMscfd
40 MMscfd
Figure 5. Scheme that implements the non-feasible
solution obtained by the pinch method (Example I).
Fig. 6 shows the superstructure to be
implemented in the optimization model presented in
this work. In this superstructure, the plant has been
decomposed into source nodes and sink nodes, and
all feasible connections between them are drawn.
The maximum flowrate Femax of each compressor
is supposed to be a 5 % superior to the operation
flowrate. The first solution obtained has a
consumption of 195.9 MMscfd. However, it
involves streams that have no practical meaning
(e.g., the connection between the output of the
make-up compressor and the inlet of the recycle
compressor). Those streams would be automatically
eliminated if the objective function considered the
cost of compressor and pipelines. An alternative to
eliminate those streams is to minimize the amount of
connections of the network. This alternative will be
presented in the following section.
BM
AM
Source A
Source B
H2 Plant300 psi
BR
AR
Sink B
Fuel
Sink A
BM
AM
BR
AR
1500 psi
1700 psi
1600 psi
2200 psi
300 psi
1600 psi
2200 psi
1600 psi
2200 psi
200 psi
200 psi
1500 psi
1700 psi
Surces Sinks
Figure 6. Superstructure for Example I.
M: make-up compressor; R: recycle compressor.
III. MODEL TO MINIMIZE
CONNECTIONS
As explained in the previous section, the
minimization of the hydrogen demand can lead to
the posing of streams withoutpractical sense. One
alternative for eliminating those streams is solving
again the optimization model of the previous section,
but this time adopting the following objective
function and additional restriction:
 
 
 
   
, ,
,
, , , , ,
,
M in sig n
, , | G E N ,
| C O N C O M
i j i j k l l
i j
F y F n F n yn F g
i j F xS P
k
l l
F
i j F xS P k N C la ss
l N C la ss C la ss

  
   

(17)
m in
F g F g (18)
WhereFgmin is the minimum demand of
hydrogen determined by the model of the previous
section, and sign(x) is the sign function of x. As all
Fi,j flowrates are not negative, the objective function
represent the amount of streams employed by the
solution. The goal is to minimize this amount, while
keeping the minimum consumption obtained in the
first solution. For this reason, the additional
restriction is used. The new objective function tries

to minimize the amount of pipelines required by the
network. Although, it is does not account for the
length of the pipelines, it is a good approximation to
be used in an early stage of the design of the
hydrogen network. This new model is called Min F.
3.1 Example II
Continuing with the example of the
previous section,Fgmin takes the following value
195.9×0.99 MMscfd of H2. Fig. 7 shows the optimal
structure obtained with the model that considers the
pressures of the nodes, and that employs a minimum
amount of streams. This time, no spurious streams
exist.
Analyzing the solution, it can be concluded
that the B make-up compressor is the limiting one
for the recovery of hydrogen. This can be verified by
a study of sensitivity in which the minimum
consumption of hydrogen is determined for a given
range of capacities of the compressors. This study
demonstrates that the hydrogen consumption
reported by the pinch method (182.8 MMscdf) can
be reached if an increase of 21 % of the flowrate of
the B make-up compressor is performed.
A
B
H2 Plant Fuel
91%
1500 psi
300 psi
85%
1700 psi
30.4 MMscfd
15.5 MMscfd
310 MMscfd
1600 psi
2200 psi
90 MMscfd
105.9 MMscfd
99%
300 psi
195.9 MMscfd
9.6 MMscfd 484.5 MMscfd
Figure 7. Plant of Example II optimized for minimizing
the consumption of the service of hydrogen considering
the pressures of the nodes and minimizing the amount of
connections.
IV. CONCLUSIONS
In this first part of the work,an optimization
model was presented, which was developed for the
design of hydrogen networks of refineries. The
model has many variants that can be adjusted to the
levels of information available at the different stages
of the design or evaluation of a network. The basic
variant, the Min Fg model, minimizes the hydrogen
consumption considering the pressures of the nodes
of the network. The second variant, the model Min
F,minimizes the number of connections of the
network, keeping the minimum hydrogen
consumption determined by the first model.
The model was implemented in the LINGO software
environment. For data input and results output an
Excel spreadsheet was implemented that was
interfaced to LINGO.
In the second part of this work, the model will be
enhanced to take into account the length of the
pipelines, the addition of purifying units and the
installation of new compressors.
REFERENCES
[1] J.J. Alves and G.P. Towler, Analysis of
Refinery Hydrogen Distribution Systems, Ind.
Eng. Chem. Res., 41 (23), 2002, 5759-5769.
[2] C. Bealins and D. Hutton, Hydrogen-pinch
Analysis, Chem. Eng. J.,109 (5), 2002, 56-61.
[3] D.C.Y. Foo and Z.A. Manan, Setting the
Minimum Utility Gas Flowrate Targets Using
Cascade Analysis Technique, Ind. Eng. Chem.
Res., 45 (17), 2006,5986-5995.
[4] Z. Zhao, G. Liu and X. Feng, New Graphical
Method for the Integration of Hydrogen
Distribution Systems, Ind. Eng. Chem. Res.,
45 (19), 2006, 6512-6517.
[5] A. Zagoria and R. Huycke, Refinery
Hydrogen Management – The Big Picture,
Hydrocarbon Processing,2 (82), 2003, 41-46.
[6] N, Hallale and F. Liu, Refinery Hydrogen
Management for Clean Fuels Production, Adv.
Environ. Res., 6 (1), 2001, 81-98.
[7] F. Liu and N. Zhang, Strategy of Purifier
Selection and Integration in Hydrogen
Networks, Trans. IChemE, 82 (10), 2004,
1315-1330.
[8] V. Agrawal and U.V. Shenoy, Unified
Conceptual Approach to Targeting and
Design of Water and Hydrogen Networks,
AIChE J., 52 (3), 2006, 1071-1082.

Optimization Model for Refinery Hydrogen Networks Part I

More Related Content

Viewers also liked (12)

Similar to Optimization Model for Refinery Hydrogen Networks Part I (20)

Recently uploaded (20)

Optimization Model for Refinery Hydrogen Networks Part I