(620o) Use of Surrogate Models In Water Network Synthesis

INTRODUCTION

The water is used in
large quantities by the chemical
industries, for being such a
strategically resource it is the
aim of severe restrictions
imposed by environmental
surveillance and control organisms, which should be
followed strictly, causing a
significant increase of its cost.

Several studies were
developed with the aim of reducing water
consumption through segregation
and integration of wastewater streams.
The water network
problem was initially
formulated by Takama et. al.(1980). After the initial problem formulation there were no major changes in the models for
wastewater treatment processes synthesis.
The research in this area focused mainly in the development of new methods for solving
the NLP (nonlinear programming) or MINLP
(mixer integer nonlinear programming) generated in the superstructure
model formulated by Takama. Although it remained
dormant, research in this area has raised up in the mid
90s, with the works of Alva-Argaéz (1998), Galan
and Grossmann (1998),
Huang et al.
(1999), Bagajewicz (2000) and continues to increase steadily from 2000.

According to Jezówski (2010) "Just a few studies considered the most detailed treatment process models, however
these approaches are often limited to
a single technology
for treating and of a fixed type."
In Jezówskixs opinion "the crux involves the
application of more rigorous
processes models: water use units, as well as, effluent process treatment." However, the use of phenomenological models in superstructure process synthesis could be unfeasible due
to its high computational time.

Oil refineries are large consumers of water,
producing wastewater with high concentrations of
H2S, oils and salts, because of the direct contact
processes between water and
oil (e.g., desalting, stream stripping and
many washing operations) (Wang and Smith, 1994).

The most used process for such
wastewater treatment is the steam stripper, where
the volatile compounds are removed
by distillation of
the heavy component, water. The working temperature of this type of
equipment is usually high, close
to the water bubble point. Due to the temperature, the steam stripping
allows the removal of compounds heavier than
air stripping processes
do.

The steam stripper
is a device that
consists of a
column with previously
heated wastewater feed at the top, and with a steam feed at the bottom. The
phase contact takes place down the column while
the wastewater becomes leaner and the steam more enriched with contaminants respect, Zygula T. M. (2008).

The objective of
this contribution is the synthesis of a water treatment
systems at petroleum refineries using a steam stripper surrogate model, for the. This
surrogate model includes equipment construction variables (number of stages and steam flow) and process variables (feed
conditions: flow, temperature and contaminant concentrations).
The data for the adjustment
of the surrogate model are
generated from process simulations
in, ASPEN PLUS.

METHODOLOGY

The mixture to be treated
contains H₂S, oils and salts. As
a simplifying assumption, we adopted
the n-hexane
to represent the oils and used NaCl as
the contaminant salt of the process.
The electrolytes dissociation equations
in aqueous medium, where the equilibrium constants are temperature function, are considered.

There are several thermodynamic
models capable of predicting the thermodynamic behavior of liquid vapor mixtures, but the
mixture to be treated is formed by a combination of electrolytes,
which generate non-idealities
in the system, requiring a thermodynamic model that
takes into account their behavior. The model selected for this system was the electrolyte-NRTL (Lee
S. Y. et. al., 2004)

A stripper can use
direct steam at  the plant boiler or a column bottom reboiler. Moreover, we can choose whether or not to use a condenser, to integrate energetically the feed and output streams
and to preheat the feed stream. Fourteen equipment variations are considered for the simulations.

It is usual to use the treatment
cost as the objective function in
the wastewater treatment synthesis so, the fixed and operating equipment cost are calculated using cost
graphics updated by the Marshall &
Swift's cost index in the third
quarter of 2010 (Peters,
2003). In the objective function, TAC (total annualized
cost), the fixed cost is multiplied by an annualized factor, that divides the
equipment fixed cost during its life cycle, taking into account the interest
rate.

The data are
obtained from process
simulators with operating
conditions varying. The input
variables that are changed
in the stripper operating
ranges are the flow rate, the
temperature and contaminants concentration in the feed stream, the number theoretical
stages and the flow of steam. The data obtained from the simulations are normalized, and then
used in the regression of a complete linear
(with respect to the parameters) model  for the
following output variables: TAC, contaminants
concentrations, temperature and flow
of output stream.

The regression is performed
using the least squares method. This complete model is then reduced using the
PCA method (principal component analysis). The validation of the surrogate
model is made by cross-validation.

The model for the stripper is a MINLP composed by the mass balance of a mixer, stripper surrogate model and splitter.
 Binary variables are
used to represent decision variables for the stripper
type and their sum
is imposed to be 1.

CONCLUSIONS

A steam stripper
surrogate model was formulated which decreases the number of variables of the system, reduces the great sensitivity to
the initial guess and its complexity, without affecting much its prediction in the specific
range. In comparison to the
usage of a phenomenological model there
is a large economy in computing
time in the water network synthesis, appart of the simplicity
of implementation in an optimization
platform such as GAMS.

REFERENCES

Alvaargaez A., ?Wastewater minimization of industrial systems using an
integrated approach?, Computers & Chemical Engineering 22 (1998):
S741-S744.

Bagajewicz M., ?A
review of recent design procedures for water networks in refineries and process
plants*1?, Computers & Chemical Engineering 24, n. 9-10 (2000):
2093-2113.

Galan B. e Grossmann I. E., ?Optimal Design of Distributed Wastewater
Treatment Networks?, Industrial & Engineering Chemistry Research 37,
n. 10 (1998): 4036-4048.

Huang C. H. et al., ?A Mathematical Programming Model for Water Usage
and Treatment Network Design?, Industrial & Engineering Chemistry
Research 38, n. 7 (1999): 2666-2679.

Jeżowski J., ?Review of Water Network Design Methods with
Literature Annotations?, Industrial & Engineering Chemistry Research
49, n. 10 (2010): 4475-4516.

Peters M. S.. Plant design
and economics for chemical engineers. 5 ed. New
York: McGraw-Hill, 2003, 988p.

Lee S. Y.  , J. M. Lee, D. Lee, I.
B. Lee, ?Improvement in Steam Stripping of Sour water through an Industrial-Scale
Simulation?, Korean J. of Chemical
Engineering, vol. 21, issue 3 (2004): 549-555.

Takama N., ?Optimal water allocation in a petroleum refinery?, Computers
& Chemical Engineering 4, n. 4 (1980): 251-258.

Wang Y.P. and Smith R., ?Wastewater minimization?, Chemical
Engineering Science 49, n. 7 (1994): 981-1006.

Zygula T. M., ?Designing steam stripping columns for wastewater?, Periodic
115, n. 5, Chemical engineering (May 2008): 52-57.