close

Вход

Забыли?

вход по аккаунту

?

Temperature control of a pilot plant reactor system using a genetic algorithm model-based control approach.

код для вставкиСкачать
ASIA-PACIFIC JOURNAL OF CHEMICAL ENGINEERING
Asia-Pac. J. Chem. Eng. 2007; 2: 526–535
Published online 1 November 2007 in Wiley InterScience
(www.interscience.wiley.com) DOI:10.1002/apj.097
Research Article
Temperature control of a pilot plant reactor system using a
genetic algorithm model-based control approach
Ahmad Khairi Abdul Wahab,1 * Mohamed Azlan Hussain1 and Rosli Omar2
1
2
Department of Chemical Engineering, Faculty of Engineering, University of Malaya, 50603 Kuala Lumpur, Malaysia
Department of Electrical Engineering, Faculty of Engineering, University of Malaya, 50603 Kuala Lumpur, Malaysia
Received 21 March 2007; Revised 24 July 2007; Accepted 13 August 2007
ABSTRACT: The work described in this paper aims at exploring the use of an artificial intelligence technique, i.e.
genetic algorithm (GA), for designing an optimal model-based controller to regulate the temperature of a reactor. GA
is utilized to identify the best control action for the system by creating possible solutions and thereby to propose the
correct control action to the reactor system. This value is then used as the set point for the closed loop control system of
the heat exchanger. A continuous stirred tank reactor is chosen as a case study, where the controller is then tested with
multiple set-point tracking and changes in its parameters. The GA model-based control (GAMBC) is then implemented
experimentally to control the reactor temperature of a pilot plant, where an irreversible exothermic chemical reaction
is simulated by using the calculated steam flow rate. The dynamic behavior of the pilot plant reactor during the online
control studies is highlighted, and comparison with the conventional tuned proportional integral derivative (PID) is
presented. It is found that both controllers are able to control the process with comparable performance.  2007 Curtin
University of Technology and John Wiley & Sons, Ltd.
KEYWORDS: genetic algorithm; model-based control; online control; chemical reactor
INTRODUCTION
Precise temperature regulation and control are crucial
in a wide-range of industries including pharmaceutical, petrochemical and chemical processing. In industries, proportional integral derivative (PID) controllers
are widely used because of their simplicity and effectiveness. Successful applications of the conventional
controllers require the satisfactory tuning of their
parameters according to the dynamics of the process.
However, these tuning methods may not produce satisfactory closed-loop responses in some cases, especially
when it has to deal with highly nonlinear processes
that take place in chemical reactors. Therefore, the PID
parameters obtained through the normally used methods, such as Cohen–Coon and Ziegler–Nichols, usually
need manual retuning before being used in real processes. In order to avoid human dependency and to
decrease the time taken in the tuning of PID controllers,
intelligent optimal control methods based on genetic
algorithm (GA) can be introduced that provide the optimized control action online.
*Correspondence to: Ahmad Khairi Abdul Wahab, Department of
Chemical Engineering, Faculty of Engineering, University of
Malaya, 50603 Kuala Lumpur, Malaysia. E-mail: khairi@um.edu.my
 2007 Curtin University of Technology and John Wiley & Sons, Ltd.
GAs are self-propelled search techniques that mimic
the process of evolution based on Darwin’s theory on
survival of the fittest.[1] GA employs multiple concurrent search points called chromosomes, which are
manipulated through three genetic operations, reproduction, crossover and mutations, to generate new search
points called offspring for the next iterations. Then,
some or all of the existing population of the current
solution set are replaced with the newly created population. The motivation behind the approach is that
the quality of the solution set should improve with the
increasing number of iterations. Thus, GA is a useful
approach to problems requiring effective and efficient
searching. GA has been successfully implemented in the
electronic industries, for example in parameter and system identification, control, robotics, pattern recognition
and classification.
Sarkar and Jayant[2] have implemented optimization
of fed-batch bioreactors using GA to determine the optimal feed substrate profile for the optimal operation of
their fed-batch bioreactor, with an optimal production of
secreted recombinant protein and a biphasic growth of
yeast as their case studies. Jose Maria Nougues et al .[3]
demonstrated the use of GA to get the best feeding
profile to minimize reaction time with temperature constraint to increase the productivity in their online batch
Asia-Pacific Journal of Chemical Engineering TEMPERATURE CONTROL OF REACTOR SYSTEM USING GA-BASED APPROACH
reactor. They have carried out online parameter estimation of the reactor mathematical model and then
successfully implemented optimal control for its temperature using GA.
An intelligent modelling and control of pH value
using GA in a pH reactor has been simulated by
Mwembeshi et al .[4] where they optimized their controller parameters in order to regulate the pH value. In
their study, they determined the optimal values of the
transformed parameter in parallel with the model and
controller parameters. An online implementation of pH
control in a pilot-scale pH plant using GA for designing the controller was presented by Tan et al .[5] who
concluded that the controller, whose parameters were
evolved using GA, is able to regulate the pH value with
minimal overshoot under various feed disturbances.
In this work, we propose a mathematical model
optimized by GA to act directly as the controller
to control the temperature of a continuously stirred
tank chemical reactor. The case study described in
the next section is divided into two main sections,
i.e. the pilot plant reactor, and its cooling system to
remove heat energy from the process. The rest of
the paper is organized as follows. In the first section,
an introduction to GA and its control application are
introduced; then the experimental pilot plant to validate
the online performance of the controller is described.
A comparison of their performance with a tuned PID
controller is also highlighted. Finally, online results to
validate the simulation studies for the controllers are
presented to conclude the findings of this work.
MATHEMATICAL MODEL OF REACTOR
SYSTEM
The chemical reactor system consists of a reactor
and cooling jacket. It is assumed that a first-order
k1
(A1−−→B 1), irreversible exothermic reaction occurs in
the process. The dynamics of the process is described by
mass and energy balance equations[6] as shown below:
F
E
dCa
= (Caf − Ca ) − ko exp −
Ca
(1)
dt
V
RT
F
−H
dT
= (Tf − T ) +
ko exp
dt
V
ρCp
UA
E
Ca −
(T − Tj )
−
RT
V ρCp
Table 1. List of model variables and parameters used.
Variables
F , T , Tj , Tf
Ca , Caf
Measured from the process
Calculated from model
and defined by user
Computed from the reactor
fluid level
V
Parameters
UA
Cp , Cpj , (E /R), k0 , (−H )
Heat exchanger data
specification
Known a priori
Table 2. Process data.
Symbols
Values (units)
(−H )
k0
E /R
UA
R
ρ, ρj
Caf
Cp , Cpc
F
V
Tf
Tj
T0
a
20 (kcal/mol)
7.2 × 1010 × 60 (h−1 )
b
1 × 104 (K)
c
4200 (kcal.h−1 .K−1 )
1.987 (cal/(K.mol))
1000 (kg/m3 )
25 × 103 (kg.mol.m−3 )
1 (kcal/(g.K))
0.16 (m3 /h)
0.16 (m3 )
30 ◦ C (303 K)
35 ◦ C (308 K)
50 ◦ C (323 K)
b
a
From.[7,8]
b
From.[8]
c
From the heat exchanger data sheet.
its control action, u. The list of model parameters and
its operating values used in this study are tabulated in
Tables 1 and 2.
In the equations mentioned above, V is the reactor
volume; F is the feed flow rate; ko is the pre-exponential
factor; E /R is the thermodynamic properties of the
reactor; Tf is the feed temperature; Caf is the feed concentration; ρ, ρc are the densities for water and coolant
fluid, respectively; C , and Cp are the heat capacity for
water; (−H ) is the heat of reaction; and UA is the
heat transfer coefficient for the reactor and jacket.
PILOT PLANT DESCRIPTION
(2)
where, Ca and T represent the reactant concentration
and reactor temperature, respectively. The manipulated
variable for the system is the coolant jacket temperature,
Tj , which is directly manipulated by the controller as
 2007 Curtin University of Technology and John Wiley & Sons, Ltd.
Our constructed pilot plant consists of two main units:
a continuously well ‘stirred’ reactor with an operating
volume of 160 l and an Alfa Laval heat-exchanger unit.
The reactor is charged and fed continuously with feed
water. The feed water is controlled by a control valve
and measured by a flow meter. The process fluid from
the reactor is pumped through a plate heat-exchanger
cooling system and recycled back into the reactor. The
Asia-Pac. J. Chem. Eng. 2007; 2: 526–535
DOI: 10.1002/apj
527
528
A. K. A. WAHAB, M. A. HUSSAIN AND R. OMAR
changes in the temperature of the recycled process fluid
acting as coolant jacket regulate the average temperature
in the reactor. The average temperature is taken as the
average of three temperature sensors located at different
strategic locations in the reactor. The cooling through
the heat exchanger is done by controlling the supply of
fresh water pumped through the heat exchanger. The
reactor water level is monitored by the level transmitter
and regulated by manipulating the control valve in the
outlet line from the reactor. In order to simulate the
heat generated by the reaction, an equivalent amount
of energy from a known steam flow rate is supplied to
the reactor through a steam coil, which is regulated via
a control valve. A schematic for the described reactor
system is shown in Fig. 1.
The process is equipped with 10 temperature transmitters (YTA50, Yokogawa, JAPAN) to accept RTD
input values (mV) and convert it to 4–20 mA electrical
signals. Except for the steam temperature transmitter, all
the RTD sensors have ranges from 273.15 to 373.15 K
(0–100 ◦ C). The RTD steam temperature sensor is used
to measure the temperature range of 273.15–573.15 K
(0–200 ◦ C) instead.
The reactor system is equipped with the SCADA
software, where Paragon TNT 5.3 is used to carry
out the important measurements and control strategies
such as for the reactor level, feed-in and coolant
flow rates. These loops are vital to ensure that the
operation of the reactor is at the given nominal process
parameter set points, which results in the continuous,
smooth operation of the reactor system. MATLAB
is used to execute the advanced nonlinear control
program using the Matlab Toolboxes. Visual Basic is
used to provide the link between the Paragon TNT
and MATLAB software, and a graphical user interface
was also developed for easy process monitoring and
control. The developed software interface for process
Asia-Pacific Journal of Chemical Engineering
Figure 1. The pilot plant chemical reactor schematic diagram. 1 = Reactor; 2 = pump; 3 = control valve (coolant
flow rate); 4 = plate heat exchanger; 5 = pressure reducing valve; 6 = control valve (steam flow rate); 7 =
ventilation line; 8 = control valve (feed in flow rate); 9 =
control valve (feed out flow rate; 10 = steam trap; 11 =
coolant flow rate line; 12 = recycle line; 13 = steam line;
14 = feed water line.
monitoring and control used in this study is shown in
Fig. 2. A more detailed description of the pilot plant
and its design can be seen in Abdul Wahab.[9]
GENETIC ALGORITHM CONTROLLER
Introduction
GAs are search algorithms based on the mechanics of
natural selection and genetics. It has been developed
by John Holland, his colleagues and students at the
Figure 2. Monitoring and control software (left) and genetic algorithm model-based controller (right)
developed for the PARS-EX reactor. This figure is available in colour online at www.apjChemEng.com.
 2007 Curtin University of Technology and John Wiley & Sons, Ltd.
Asia-Pac. J. Chem. Eng. 2007; 2: 526–535
DOI: 10.1002/apj
Asia-Pacific Journal of Chemical Engineering TEMPERATURE CONTROL OF REACTOR SYSTEM USING GA-BASED APPROACH
University of Michigan in 1960. GAs are particularly
suitable for solving complex optimization problems, and
they are inherently parallel in implementation since their
search for the best solution is performed over genetic
structures that can represent a number of possible
solutions.
A GA operates through a simple but important
iteration in four main stages, i.e. the creation of
population of strings, the evaluation of each string, the
selection of best or fittest strings and finally the genetic
manipulation to create the new population of strings.
In order to search for solution/s for any problem/s, a
typical GA would do the following steps iteratively:
• Start with any initial random sets of individuals/strings/chromosomes, denoted by P (0), which is
called the initial population;
• Evaluate the objective function for each individuals
in P (0);
• On the basis of this evaluation, create a new set of
population, i.e. P (1) using genetic operators, such as
the selection, crossover and mutation;
• Repeat the procedure iteratively until a defined terminating condition is reached and satisfied.
The flowchart to illustrate the general generation
procedure described earlier is shown in Fig. 3:
Creation of population
In the first stage, an initial random population of
potential solutions is created as the starting point for
the intended search. Each element of the population
is encoded into strings, also known as chromosomes,
which consists of genes that are to be manipulated by
genetic operators. There are several methods of coding,
such as binary, integer, hexadecimal and alphabetical
encodings.
String evaluation
The objective function to be optimized provides the
mechanism for evaluating each chromosome. On the
basis of each individual’s fitness, a selection mechanism chooses candidates for the genetic manipulation
process. A fitness function is devised for each problem; given a particular chromosome, the fitness function
returns a single numerical fitness value, which is proportional to the ability or goodness of the individual
represented by that chromosome.
Selection mechanism
The selection operator models nature’s mechanism on
the survival-of-the-fittest mechanism. Fitter solutions
will survive, while weaker ones will die and be discarded away from the population. A fitter string has
a higher chance of surviving in the next generation
of iterations. One type of GA uses a roulette wheel
selection scheme to carry out this task. Each string is
allocated with a sector or slot of a roulette wheel, with
the angle subtended by the sector at the center of the
2π f
wheel equalling f i , where fi is the fitness value of
ave
the i th is string and fave the average fitness value of the
population. A string is allocated with an offspring if
a randomly generated number in the range 0–2π falls
in the sector corresponding to the strings. The algorithm continues to select the string in this fashion until it
has generated the entire population in the next potential
solutions.
A probabilistic selection is made on the basis of the
individual’s fitness, such that the better individuals will
have an increased chance of being selected. Consider
a set of individuals for mating/breeding, denoted by
M (k ), formed on the basis of the current population,
P (k ). The elements of this set, M (k ) are selected as
follows:
• Generate z , i.e. z ∈ [0, 1], a random number between
0 and 1;
• For each individual x (i ) of P (k ), calculate Pi ;
Pi =
f (x (i ))
j
(3)
f (x (i ))
1
Figure 3. Generalized generation procedure.
 2007 Curtin University of Technology and John Wiley & Sons, Ltd.
where Pi is the probabilistic value of the individual i
to be selected, f is the fitness function and j is the size
of the chosen set M (k );
Asia-Pac. J. Chem. Eng. 2007; 2: 526–535
DOI: 10.1002/apj
529
530
A. K. A. WAHAB, M. A. HUSSAIN AND R. OMAR
• If Pi z , select x (i ) as a member of M (k );
• Repeat until M (k ) has the same size as P (k ).
Genetic manipulation
The manipulation process uses genetic operators to
produce a new population of individuals or offspring
by manipulating the genetic information, referred to as
genes, owned by members of the current population.
It comprises two operators: crossover and mutation.
Crossover recombines a population’s genetic material.
Pairs of chromosomes are selected from the current
population and are subjected to crossover operation.
Assuming that the length of the chromosomes selected
is l , it will randomly choose a crossover point that
can take a value in the range of 1 to l − 1. The
portions of two chromosomes beyond this crossover
point are exchanged to form the new chromosomes. The
crossover point may assume any of the (l − 1) possible
values with equal probability. Figure 4 shows how a
crossover takes place between two chromosomes. The
point of crossover is randomly selected and here it is at
the middle of the chromosomes with the length of 4 bits.
It is not necessary that a crossover should always take
place between two chromosomes. After choosing a pair
of chromosomes, the algorithm invokes crossover only
if a randomly generated number in the range of 0–1 is
greater than a crossover rate or crossover probability,
Pc , which lies in the values ranging from 0 to 1. After
the crossover, the algorithm manipulates further the
chromosomes using the mutation operator.
Mutation of a bit involves flipping it, i.e. changing
it from 0 to 1 or vice versa. An example given in
Fig. 5 shows how mutation takes place in a single
chromosome.
The GA uses mutation only as a secondary operator
with the role of restoring lost genetic materials. In the
Asia-Pacific Journal of Chemical Engineering
event that all the chromosomes in a population have
converged to an undesired value at a given position,
then crossover is no longer capable of regenerating new
chromosomes, while the mutation operator is.
Termination conditions
A GA will eventually run continuously if it has no
termination criteria. There are two common types of
termination condition. The ‘terminate-upon-generation’
condition stops the generation after a given number of
generations have been reached, at which point the best
existing solution is obtained. Secondly, the ‘terminateupon-convergence’ condition terminates the evolution
iteration if the best existing solution scores a fitness
level that is above or equal to a previously defined
threshold value. These two termination strategies can
also be combined to have more control upon the generation.
GENETIC ALGORITHM MODEL-BASED
CONTROL (GAMBC)
This section describes a model-based control scheme,
i.e. the GAMBC to provide the optimal control action
to the process by means of GA. The GAMBC control
structure is highlighted in Fig. 6. Under this scheme,
the following elements of the controller structure have
to be defined prior to its implementation: an objective
function, process model, parameter encoding to present
the possible solution space and sets of genetic operators as mentioned in earlier sections. The MATLAB
GA toolbox was used throughout this work to implement the genetic operation to provide control action for
the process.[10]
Objective function
Figure 4. Crossover procedure illustrated.
The objective function used in this work is given
by Eqn (4) below, where Tset is the desired reactor
temperature set point and T̂ is the reactor temperature
from the first-principle model optimized by GA. The
GA routine minimizes this fitness value, J , given by
J = Tset (k ) − T̂ (k )
(4)
Process model
Figure 5. Mutation procedure illustrated for a single
chromosome.
 2007 Curtin University of Technology and John Wiley & Sons, Ltd.
Mathematical model used to get the best optimal control
action value has been described in the previous section.
The model Eqns (1) and (2) are solved at every instant
of the sampling time of 1 s to give the corresponding
reactor temperature, T , manipulated using GA.
Asia-Pac. J. Chem. Eng. 2007; 2: 526–535
DOI: 10.1002/apj
Asia-Pacific Journal of Chemical Engineering TEMPERATURE CONTROL OF REACTOR SYSTEM USING GA-BASED APPROACH
T, Ca
Process
Disturbances
Genetic Operators
T, Ca
Process Model
Tset
+
CSTR PROCESS
u
e
_
Objective Function
GAMBC
CONTROLLER
T, Ca
Figure 6. Genetic algorithm model-based control (GAMBC) structure.
Parameter coding
Acquire Tset, T
from the
process
The control actions, Tj , i.e. the coolant jacket temperature, is the variable that contributes to minimize the fitness value derived from the objective function above in
Eqn (4), and therefore this is the variable to be encoded
into the chromosomes. In this work, the standard binary
coding method has been chosen. In order to incorporate
saturation limits in the control action applied to the process, the encoding is quantified between umin = 25 and
umax = 65 ◦ C in the control action.
Error=Tset-T
YES
Abs(Error) <
0.01?
NO
Create Random Binary
Initial Population for
Control Variable,Tj
Evaluate
Fitness Values
G. A. Operators
(Selection, Crossover and
Mutation)
Genetic operators
The genetic operators used in this work are those
described in the previous section, i.e. linear ranking,
roulette wheel selection, single point crossover with
rate Pc of 0.6 and mutation rate Pm of 0.05 were
used throughout the work. These genetic parameters
are selected from examples cited in Goldberg[1] and
also from simulation experiences. The GAMBC control algorithm, as shown in the flow chart in Fig. 7,
executes by continuously monitoring the error between
the reactor temperature set point, Tset , and the current
temperature, T , for each sampling iterations.
The controller working procedure starts by taking into
account the process error. If it is below the defined
value, i.e. 0.01 ◦ C, the system continues to monitor
the process. The controller starts to create the initial
randomized population only when the error is more than
0.01 ◦ C. The initial 100 individuals are then evaluated
using the objective function mentioned earlier to obtain
the fitness value, J . At this stage, GA operators will
perform the selection, crossover and mutation at the
assigned parameters as shown in Table 3. Using the
evolved chromosomes, which consist of a 10-bit binary
representation of the possible control action, Tj , the
chromosomes are then decoded into real values before
they are used to solve, using the process mathematical
 2007 Curtin University of Technology and John Wiley & Sons, Ltd.
Solve ODE Model Using Tj
proposed by GA
Performance
Index
Terminate at
Convergence
?
Use previous
control action, u
YES
NO
NO
Terminate at
MAXGEN
?
YES
Implement to
Process
YES
Next Sampling
Instant?
NO
END
Figure 7.
Genetic algorithm model-based
control (GAMBC) algorithm.
Asia-Pac. J. Chem. Eng. 2007; 2: 526–535
DOI: 10.1002/apj
531
A. K. A. WAHAB, M. A. HUSSAIN AND R. OMAR
Asia-Pacific Journal of Chemical Engineering
model, for the reactor temperature, T . The evolution is
then checked for convergence and maximum allowable
generation per sampling time, which is defined as 0.001
and 30, respectively. These terminating conditions are
important so as to ensure that complete generations
are achievable between the sampling instants. The
individual corresponding to the fittest value is then
passed to the process as the control action, u. Finally,
the algorithm monitors the user feedback to continue
the next iteration or otherwise stops the controller loop.
ONLINE IMPLEMENTATION
Heat of reaction estimation
Table 3. Parameters of GAMBC controller.
Initial population, P (0)
100
Max. generation per sampling time, (MAXGEN)
30
Control action limits
25–65 ◦ C
0.6
Crossover rate, Pc
0.05
Mutation rate, Pm
Length of encoding bits, L
10 bit
 2007 Curtin University of Technology and John Wiley & Sons, Ltd.
Figure 8. Steam flow rate controller menu. This figure is
available in colour online at www.apjChemEng.com.
Steam flow rate (FT3) profile
100
80
60
40
20
14:52
14:32
14:11
13:51
13:31
13:11
12:51
12:31
12:10
11:50
11:30
11:10
10:50
0
FT3 Setp
FT3
10:30
The first-order exothermic chemical reaction is experimentally simulated by injecting steam through a coil
into the reactor liquid. A similar strategy was also utilized by Hussain and Kershenbaum[11] in order to simulate a chemical reaction. Having an accurate amount of
the heat generated is important to obtain good and reliable experiment results. In this study, the amount of the
heat generated from the reaction is calculated, and this
simulated value is charged into the system using dry saturated steam flow, with the amount fed into the reactor
being controlled by a control valve through a supervisory computer control system. The steam enters the
reactor through a steam coil and exhausts out through
a steam trap with complete condensation. The actual
steam flow rate supply is monitored using a differential
pressure flow transmitter in the steam supply line. These
procedures are automated in a computer to continuously
deliver the correct amount of heat energy to the reactor
on the basis of the online measured temperature and
calculated reactant concentration. The graphical user
interface developed for this routine is shown in Fig. 8.
The control menu gives the operator flexibility to control and monitor the amount of energy supplied to the
reactor.
The online process temperature is sampled at every
second from the process, and reactant concentration, Ca ,
is computed from the mathematical model in order to
give the approximate required steam flow rate. A typical
steam profile supplied to the reactor to simulate the heat
of reaction profile during a typical experimental work
is shown in Fig. 9.
Steam flow rate (%)
532
Time (h)
Typical steam flow rate profile during setpoint tracking. This figure is available in colour online at
www.apjChemEng.com.
Figure 9.
Software development
The information flowchart showing the computer control setup for the pilot plant system can be seen in
Fig. 10. The Paragon software manages the overall data
acquisition and conventional control of the local control
loops. This software is interfaced with the main data
acquisition and the reactor simulation program using
Visual Basic, which is also interfaced to the GA control
program written in Matlab, running on an Intel Pentium
3 personal computer.
At high frequency (1 s), the reactor simulation program reads in the relevant measurements: reactor temperature, level and feed rates, through the Paragon
system. These are used to dynamically update the concentration of the simulated reactants and instantaneous
rate of reaction in the reactor by solving the process
Asia-Pac. J. Chem. Eng. 2007; 2: 526–535
DOI: 10.1002/apj
Asia-Pacific Journal of Chemical Engineering TEMPERATURE CONTROL OF REACTOR SYSTEM USING GA-BASED APPROACH
mass balance equation within the reactor. At a lower
frequency (1 min), the advanced or conventional control
algorithm employed as a master controller sends a setpoint signal to the slave controller, which then adjusts
the cooling jacket temperature through the manipulation
of a control valve. The setup of this cascade control
loop with GA control action as the master controller is
shown in Fig. 11.
Multiple set-point tracking
The GA controller is subjected to a multiple set-point
tracking beginning from the nominal temperature value
of 50 ◦ C and changing to 60, 45 and 55 ◦ C, respectively.
The demands were introduced at hourly intervals. The
GAMBC controller successfully tracked the demands as
shown in the Fig. 12, but oscillations were observed in
the control results after the reactor temperature reached
the set-point values of 50, 60 and 55 ◦ C, respectively.
Load disturbance tests
The controller was then tested further under an introduced disturbance in the feed to the flow rate, F , of
the reactor, where it was reduced to 85% from its
nominal value. After the start-up of the process, the
reactor temperature was left regulated by the GAMBC
controller at Tset = 50 ◦ C for about an hour. The disturbance was then introduced, and at the same time,
the control action was paused at its current values to
observe the process response to these changes. After
about an hour, the control action was applied back automatically to the system in its effort to bring back the
reactor temperature to its set point. When the nominal feed in flow rate was reduced by 15%, the reactor
temperature rose to about 60 ◦ C. When the controller
was brought back to action, it was observed that the
controller was able to bring back the process variable
to its initial set point of 50 ◦ C, as shown in Fig. 13.
The results showed that the GAMBC performed comparably with the PID controller. The control response
for changes in pre-exponential factor, k0 , and heat of
reaction, (−H ), with similar experimental procedure
is shown in next section.
Controller implementation
In order to test the robustness of the proposed controller,
the following tests have been implemented in this work.
• Multiple set-point tracking at the nominal temperatures of 50, 60, 45 and 55 ◦ C
• Two types of disturbances applied to this process
• −15% of changes in external process operating
condition, Feed flow rate, F
• +3% of changes in internal process parameter,
the pre-exponential factor, k0 and heat of reaction,
(H ).
PARS-EX PILOT
PLANT
PARAGON 5.3
Lower speed
sampling at 1 min
High speed
sampling at 1 sec
Changes in process parameters
Online experiments were carried out for changes in
both the rate parameter, k0 , and the heat of reaction,
(−H ). The results for a change in pre-exponential
factor and heat of reaction using PID and comparison with the proposed GAMBC are shown next. A
conventional PID control loop was also tested on the
same disturbances. Here, the reactor temperature, T , is
used in a master PID loop tuned by the well-known
Cohen–Coon method to send a revised set point to the
slave controller which regulates the jacket temperature,
DATABASE
MATLAB 5.3
VISUAL BASIC
Control action
from Advanced
Control Algorithms
(Genetic Algorithm)
Reaction Simulation
Programming
(Steam flow rate)
Figure 10. Information flowchart of the pilot plant reactor
system.
Tset
%CV4
Tjset
GAMBC
+
PID
Tj
HE
T, Ca
PARS-EX
+
-
Tj
T
Figure 11.
Cascade control loop for genetic algorithm model-based control.
GAMBC – genetic algorithm model based controller; PID – slave PID controller; HE – heat
exchanger, PARS – EX-partially simulated exothermic reactor.
 2007 Curtin University of Technology and John Wiley & Sons, Ltd.
Asia-Pac. J. Chem. Eng. 2007; 2: 526–535
DOI: 10.1002/apj
533
A. K. A. WAHAB, M. A. HUSSAIN AND R. OMAR
Asia-Pacific Journal of Chemical Engineering
Change in k0 for GAMBC and PID Controller
GAMBC
70
60
Temp (C)
Tset
T(PID)
T(GAMBC)
PID
13:15
12:58
12:41
12:25
12:08
11:51
11:34
11:17
11:00
10:44
10:27
10:10
9:53
20
Time (h)
Comparison of load disturbance rejection
with GAMBC control and conventional tuned PID controller (change in +3% of nominal pre-exponential
factor). This figure is available in colour online at
www.apjChemEng.com.
Figure 14.
Change in Heat of Reaction for GAMBC and PID
Controllers
GAMBC
70
Temp (C)
60
GAMBC
60
50
40
Tset
T(PID)
T(GAMBC)
PID
30
50
13:33
13:17
13:01
12:45
12:29
12:13
11:57
11:41
11:25
11:09
10:53
30
PID
10:37
T(GAMBC)
Tset
T(PID)
10:21
20
40
10:05
Temp (C)
40
30
Multiple set point tracking using GAMBC
and PID controllers
70
50
9:36
Tj . The most rigorous test of the PID controller in a
nonlinear situation is to examine its performance when
one or more of the nonlinear parameters in the process
model have been changed. The results demonstrate that
the GAMBC-based method performed with comparable
control results to the conventional PID controller. In
the first disturbance test, i.e. change in pre-exponential
factor as shown in Fig. 14, the GAMBC controller successfully brought the reactor temperature to the set point
faster compared to the PID controller. For the second
test, as shown in Fig. 15, the PID controller exhibited
visible offset when the heat of reaction was changed
during the test. As the GAMBC uses the process mathematical model itself to compute the optimal control
action, the strategy has an advantage over the PID controller since the PID works only on the error detected.
One disadvantage of this computational intelligence
technique is that it is computationally very intensive,
but this is no longer a major drawback because of the
recent advances in computing technology.
Time (h)
12:45
12:26
12:08
11:49
11:31
11:12
10:54
10:35
10:17
9:52
9:34
9:15
8:57
8:38
20
Time (h)
Figure 12. Online implementation of GAMBC and PID
control (set-point tracking). This figure is available in colour
online at www.apjChemEng.com.
Figure 15. Comparison of load disturbance rejection with
GAMBC control and conventional tuned PID controller
(change in +3% of nominal heat of reaction). This figure is
available in colour online at www.apjChemEng.com.
CONCLUSIONS
Change in F for GAMBC and PID Controllers
70
GAMBC
PID
60
Temp (C)
50
40
T(GAMBC)
Tset
T(PID)
30
12:34
12:18
12:02
11:46
11:30
11:14
10:58
10:42
10:26
10:10
9:54
9:38
9:08
20
8:52
534
Time (h)
Figure 13. Comparison of load disturbance rejection with
GAMBC control and conventional tuned PID controller
(change in −15% of nominal feed flow rate). This figure is
available in colour online at www.apjChemEng.com.
 2007 Curtin University of Technology and John Wiley & Sons, Ltd.
In this work, an advanced control approach using
model-based control utilizing GA was introduced. Our
main objective was to control the temperature of a partially simulated exothermic reactor pilot plant system.
Experimental results reported in this paper indicate that
mathematical model-based controller, whose parameters
were evolved by GA, is able to regulate the reactor
temperature over a wide range with minimal overshoot.
Load disturbances in the form of varying flow rate in
the process input stream could also be rejected by the
proposed control methodology. Since the performance
of the GAMBC controller is heavily dependent on the
model precision, obtaining the most accurate model of
the process is important, as this will affect the controller performance in issuing the optimal control action.
This work demonstrates the applicability of GAMBC
Asia-Pac. J. Chem. Eng. 2007; 2: 526–535
DOI: 10.1002/apj
Asia-Pacific Journal of Chemical Engineering TEMPERATURE CONTROL OF REACTOR SYSTEM USING GA-BASED APPROACH
method and its possible application to a real experimental plant. Online implementation of this proposed
controller is strong evidence to indicate the reliability
of such technique for controlling the temperature of a
pilot plant chemical reactor, such as the one available
in this department.
Acknowledgements
This work was fully supported by a research grant
from The Ministry of Science, Technology and Innovation (MOSTI), Malaysian Government, under the
IRPA grant (09-02-03-0475). The grant for A. K. Abdul
Wahab from the National Science Fellowship Award of
the Malaysian Government is also gratefully acknowledged.
NOMENCLATURE
A1
A
B1
Ca
Caf
Cp
CV1,2,3,4
e
E
F
t
IAE
k0
k1
J
M
Reactant
Heat transfer area, m2
Product
Reactant concentration, kg.mol/m3
Feed concentration, kg.mol/m3
Heat capacity, kcal./(g K)
Control valves
Error, ◦ C
Activation Energy, kcal/mol
Feed flow rate, m3 /h
Time, h
Integral Absolute Error
Pre-exponential factor, h−1
Reaction rate
Fitness value
Subset of initial population chosen for next
k th iteration
P
Population size
PARS-EX Partially Simulated Exothermic
PID
Proportional Integral Derivative
Crossover probability
Pc
Mutation probability
Pm
 2007 Curtin University of Technology and John Wiley & Sons, Ltd.
R
RTD
SCADA
T
Tj
Tf
Tset
U
u
V
ρ
Ideal gas constant, cal/(K.mol)
Resistance temperature detection
Supervisory control and data acquisition
Reactor temperature, ◦ C
Jacket temperature, ◦ C
Feed temperature, ◦ C
Reactor temperature set point, ◦ C
Overall heat transfer coefficient,
kcal.h−1 . ◦ C−1
Control action, ◦ C
Reactor volume, m3
Fluid density, kg/m3
GREEK LETTERS
−H
Heat of reaction, kcal/mol
REFERENCES
[1] D.E. Goldberg, Genetic Algorithms in Search, Optimization
and Machine Learning, Addison-Wesley Publishing Company
Inc.: Canada, 1989; pp.1–2, pp.125–143.
[2] D. Sarkar, M.M. Jayant, Chem. Eng. Sci., 2003; 58,
2283–2296.
[3] M. Jose Maria Nougues, D. Grau, L. Puigjaner, Chem. Eng.
Process., 2002; 41, 303–309.
[4] M.M. Mwembeshi, C.A. Kent, S. Salhi, Comput. Chem. Eng.,
2004; 28, 1743–1757.
[5] W.W. Tan, F. Lu, A.P. Loh, K.C. Tan, Eng. Appl. Artif. Intell.,
2005; 18, 485–494.
[6] B.W. Bequette, Process Dynamics-Modelling, Analysis and
Simulation, Prentice Hall Publisher: New Jersey, USA, 1999;
pp.559–596.
[7] G. Lightbody, G.W. Irwin, IEEE Proc. Control Theory Appl.,
1995; 142(1), 31–43.
[8] S.S. Ge, C.C. Hang, T. Zhang, J. Process Control, 1998; 9(4),
313–323.
[9] A.K. Abdul-Wahab, Temperature control of a partially
simulated exothermic chemical reactor using neural networks,
M. Eng. Science Dissertation, University of Malaya, Malaysia,
2002; pp.90–120.
[10] A. Chipperfield, P. Fleming, H. Puhlheim, C. Fonseca,
Genetic algorithm toolbox . For use with MATLAB user’s
guide, Version 1.2, Department of Automatic Control and System Engineering, Univ. of Sheffield, UK, 1994.
[11] M.A. Hussain, L.S. Kershenbaum, Trans. Inst. Chem. Eng.,
Part A, 2000; 78, 299–311.
Asia-Pac. J. Chem. Eng. 2007; 2: 526–535
DOI: 10.1002/apj
535
Документ
Категория
Без категории
Просмотров
0
Размер файла
766 Кб
Теги
base, using, mode, approach, temperature, reactor, pilot, system, algorithms, genetics, plan, control
1/--страниц
Пожаловаться на содержимое документа