Beruflich Dokumente
Kultur Dokumente
on DATA NETWORKS, COMMUNICATIONS & COMPUTERS, Bucharest, Romania, October 16-17, 2006 258
Abstract:- Ammonia production depends on temperature of feed gas at the top of the reactor (top
temperature), the partial pressures of the reactants and the reactor length. The optimal design problem
requires obtaining the optimal reactor length with maximum economic returns corresponding to a top
temperature of 694K. This paper presents an alternative approach in solving the boundary value problem and
at the same time determines the optimal solution. Shooting methods, namely single and multiple-shooting
methods are used. The software used for this modeling is MATLAB version 6.1. The ordinary differential
equation (ODE) integration routine technique used is ode45 and the optimization routine of FMINCON is
selected for multiple-shooting. The profiles of N N2 , Tf and Tg at a top temperature of 694K were generated.
From the results obtained, this temperature yields an objective function value of $5.015 x 10 6 per year at an
optimum reactor length of 6.695m. The objective function and reactor length values agree with the latest
literature work using Differential Evolution (DE) approach, and multiple-shooting performs better than
single-shooting under bad initial guesses of the reactor length.
1
Proceedings of the 5th WSEAS Int. Conf. on DATA NETWORKS, COMMUNICATIONS & COMPUTERS, Bucharest, Romania, October 16-17, 2006 259
temperature and gas flow rate profiles obtained. o The heat capacities of the reacting gas
Babu, B.V et. al [4] claimed that DE is an and feed gas remains constant
improved version of simple GA which is faster, o The catalytic activity is uniform along the
robust and more likely to a functions true global reactor and equal to unity
optimum. They used Nested DE in choosing the o The pressure drop across the reactor is
right combination of the DE key parameters. This negligible compared with the total
paper presents an alternative approach using pressure in the system
shooting methods to determine the optimal reactor
length and systems objective function subject to a 2.1 Objective Function
number of equality constraints involving solution The objective function depicts the economic return
of coupled differential equations. based on the difference between the value of the
product gas and the feed gas less the amortization
of reactor capital costs. Similar objective function
2. Problem Formulation for the process is adopted from Babu, B.V et. al
The formulation used in this paper is similar to [4];
Babu, B.V et. al [4] and Upreti, et. al [3]. In the f (x, N N2 ,T f ,Tg ) = 1.33563 107 1.70843 104 N N2 +
study by Upreti, et. al [3] by using GAs, it is found
that for the range of top temperature of the reactor 704.09(Tg T0 ) 699.27(T f T0 )
(from 668K to 706K), the profit corresponding to
[3.45663 x 107 + 1.98365 x 109 x]1/2
the optimal reactor length is increased
monotonically. The optimal reactor length at a top
temperature of 668K increases from 0.63m (1)
to5.31m. Typical economic return from the reactor 2.2 Energy Balance Equation
operation with a top temperature of 694K an d Equations (2)-(6) are the energy balances that need to
6
5.33m reactor length is about $4.23 x 10 per year. be satisfied in order to get the value of the three
Babu et. al [4] extended the study on the variables: Tf, Tg and N N2 .
optimization of an auto-thermal ammonia 1
synthesis reactor design by using DE method. dT f = US (Tg Tf ) (2)
They solved three-coupled differential equations dx WC pf
using NAG subroutine (D02EJF) in MATLAB. 2 2
The step sizes of 0.01 and 0.001 are used instead dTg US ( H )S dN N
of that used by Upreti et. al. [3] who used only
= WC (Tg Tf ) +
1
2
Proceedings of the 5th WSEAS Int. Conf. on DATA NETWORKS, COMMUNICATIONS & COMPUTERS, Bucharest, Romania, October 16-17, 2006 260
N2
p = 3p S0 S1
H2 N2
The boundary conditions are: Table 1: Initial Condition (S0, S1) and Initial Guessed (X0
Tf (x = 0) = T0 ; Tg (x = 0) = Tf ; and X1) Values for multiple shooting with 2 intervals
N (x = 0) = 701.2kmol S0 X0 S1 X1
N 701.2 500 510 450
N2
2 2
N2 m h (kmol/m h)
Tf (K) 694 540 550 410
Tg (K) 694 725 730 430
2.4 Inequality Constraints
The upper and lower bounds of the design
variables are as follows: x3
0 kmol N N2 3220 kmol ;
x2 x4
m2h m2h x1
400K Tf 800K ; 0m x 10m
Since the reaction gas temperature (Tg) depends on
the nitrogen mass flow rate ( N N2 ), feed gas
S0 S1 S2 S3
temperature (Tf) and reactor length ( x ), there is
no need to implicate any boundaries on Tg. Fig 1-3 Fig. 2: Illustration of multiple shooting with 4
show illustrations of multiple shooting method intervals
with 2, 4 and 8 intervals respectively. Table 1,2,
and 3 list the initial conditions and guesses for all
variables used in the modeling for multiple
shooting with 2, 4 and 8 intervals respectively.
x8
x6 x7
x4 x5
x1 x2 x3
S0 S1 S2 S3 S4 S5 S6 S7
3
Proceedings of the 5th WSEAS Int. Conf. on DATA NETWORKS, COMMUNICATIONS & COMPUTERS, Bucharest, Romania, October 16-17, 2006 261
Table 2: Initial Condition (S0, S1) and Initial Guessed (X0 and X1) Values for multiple shooting with 4 intervals
S0 X1 S1 X2 S2 X3 S3 X4
2 701.2 550 540 505 500 498 498 490
N N2 (kmol/m h)
Tf (K) 694 640 635 505 500 370 360 200
Tg (K) 694 800 790 730 725 600 590 420
Table 3: Initial Condition (S0, S1) and Initial Guessed (X0 and X1) Values for multiple shooting with 8 intervals
S0 X1 S1 X2 S2 X3 S3 X4 S4 X5 S5 X6 S6 X7 S7 X8
N 701.2 610 600 550 545 510 505 503 500 498 496 495 494 493 492 491
N2
(kmol
2
/m h)
Tf (K) 694 675 670 635 630 575 570 505 500 415 410 370 365 280 275 200
Tg 694 765 770 800 790 760 755 730 725 660 650 600 590 500 490 420
(K)
4
Proceedings of the 5th WSEAS Int. Conf. on DATA NETWORKS, COMMUNICATIONS & COMPUTERS, Bucharest, Romania, October 16-17, 2006 262
Differential Equation
Not optimized Solver (Ode45)
( N N , Tf, Tg, and x )
2
Optimized
values
Fig.5 : Computational procedures
5
y
Proceedings of the 5th WSEAS Int. Conf. on DATA NETWORKS, COMMUNICATIONS & COMPUTERS, Bucharest, Romania, October 16-17, 2006 263
]
ui = i ( , qi ) for [ i , i+1 (7)
general dynamic optimization problem is
considered:
Min (z(t f ), x(t f ), u(t f ), t f , ) (1)
using N local control parameter
z(t), x(t), u(t), t f ,
vectors qi, where q0, q1, , qN-1
Subject to DAE model: The functions i are given basic functions,
typically vectors of polynomials. If
piecewise constant approximation is
dx = f (z(t f ), x(t f ),u(t f ),t f , ) , t I (2) chosen, then i ( , qi ) = qi .
dt
where 0 = G(z(t), x(t),u(t),t, ) , t I Step 2: State discretization by multiple shooting
Introduced N+1 additional vectors s0, s1, ..,
sN, where si = combination of siz and six.
Initial conditions: The dimension is the same as the system
state. These are also known as multiple
shooting node values. All but the last
(3) value serve as initial values for the
x(0) = x0
resulting N independent decoupled IVP on
the intervals Ii:
dxi = fi (zi ( ), xi ( ), i ( , qi ), )
As well as general inequality constraints (state and d
control): (8)
0 h(z(t), x(t),u(t), ) for [ i , i+1 ]
(4)
(9)
0 = Gi (zi ( ), xi ( ), i ( , qi ), )
In addition, terminal constraints (if any):
0 = r(x(t f )) (5) x z
Gi (si , si , i ( , qi ),
x
The functions f, G, h and r are assumed to be twice xi ( i ) = si (10)
z
continuously differentiable, tf > t0, and the initial zi ( i ) = si (11)
conditions x0 and the model parameters are known
and fixed. The following paragraph shows the
steps involved in the application of multiple
shooting method.
Step 1: Control representation/ discretization Additional constraints:
Let the initial time horizon I = [t ,t ]
0 f 1- The decoupled IVP are connected by
Divide into N subintervals, continuity (or matching) conditions, i.e.
Ii = [ i , i+1 ] (6) each node value should equal the final
6
Proceedings of the 5th WSEAS Int. Conf. on DATA NETWORKS, COMMUNICATIONS & COMPUTERS, Bucharest, Romania, October 16-17, 2006 264
i = 0, 1, . N-1
To carry out both shooting strategies, MATLAB
2- The first multiple shooting node variable version 6.1 is chosen as the platform for the
s0 must be equal to the initial value x0 of implementation, utilizing the ODE integration
the optimization problem (this applies routine plus the optimization solver available.
only when we are considering optimal
trajectory problems, otherwise this is not a
necessity), sx0 = x0.
4. Results and Discussions
The constraints are not necessarily satisfied during 4.1 Single Shooting
the optimization iterations. Multiple shooting can Four MATLAB programming files were
deal with infeasible initial guesses of the variables developed to model and optimize the ammonia
si and qi. reactor. From the profiles obtained, it is found out
that the optimum reactor length is 6.695m and the
corresponding objective function is $5.015 x 106
The algebraic equations (i.e. equation (9)) in each per year. The corresponding values of N N2 , Tf and
multiple shooting interval have an additional
subtrahend term. This is called as a relaxed version Tg are shown in Table-4.
of the original DAE and allows for inconsistent
x z 4.2 Multiple-Shooting
initial values s i and s i.
Summarizing, the finite dimensional NLP in the For multiple shooting, four MATLAB
direct multiple shooting parameterization is given programming files were developed to model and
by: optimize the ammonia reactor. Figures 8, 9 and 10
N 1 shows the resulting profiles at different number of
Min Li (si x , si z , i ( , qi ), ) + E(s x N , ) intervals. Data from the graphs are shown in
(12) Tables 5, 6 and 7 respectively.
i =0
From Table-8, it is clear that irrespective of the
q, s number of interval, the same optimized values for
The expression in (12) is further minimized objective function ($5.015 x 106 per year) and
subject to the following continuity conditions: reactor length (6.695m) were obtained. This
x consistency authenticates the robustness of the
s i+1 = xi ( i+1 ) , for i = 0,1,, N 1 (13)
multiple shooting method regardless the number of
initial conditions: intervals.
s
x
0 = x0 (14)
Various initial guesses of N N2 , Tf, Tg and reactor
consistency conditions: length, x have been used to test the robustness for
this multiple shooting method including initial
x z
0 = G(si , si , i ( , qi ), , for i = 0,1,, N guesses which are beyond the upper and lower
boundaries. It is found that for multiple shooting,
(15) regardless the initial guess values of reactor
length, the objective function and optimum reactor
Control and path constraints are imposed point- length remain the same. Hence, this shows that
wise at the multiple shooting nodes: multiple shooting is robust in solving non-linear
x z
h(si , si , i ( ,qi ), ) 0 , for i = 0,1,, N programming problems.
(16)
7
Proceedings of the 5th WSEAS Int. Conf. on DATA NETWORKS, COMMUNICATIONS & COMPUTERS, Bucharest, Romania, October 16-17, 2006 265
900
800
Tg
700
600
NN2, Tf and
NN2
500
Tf
400
300 Tg
200
100
0
0 2 4 6 8
Reactor Length (m)
900
900
800
800
Tg
700
Tg
700
600 NN2
600
NN2, Tf and
,NN2 Tf and
500 NN2
Tf 500
400 Tf
Tg 400
300 Tg
300
200
200
100
100
0
0
0 2 4 6 8
0 2 4 6 8
Reactor Length (m)
Reactor Length (m)
Fig.8 : Profiles for two intervals using multiple- Fig.10: Profiles for eight intervals using
shooting multiple-
shooting
8
Proceedings of the 5th WSEAS Int. Conf. on DATA NETWORKS, COMMUNICATIONS & COMPUTERS, Bucharest, Romania, October 16-17, 2006 266
Table-8: Summary of number of intervals, time taken, profit and optimal reactor length for initial
guess of 7m.
2 4 2.17 6 6.695
5.015 x 10
4 5 5.83 6 6.695
5.015 x 10
8 5 19.00 6 6.695
5.015 x 10
9
Proceedings of the 5th WSEAS Int. Conf. on DATA NETWORKS, COMMUNICATIONS & COMPUTERS, Bucharest, Romania, October 16-17, 2006 267
Varies initial guesses of N N2 , Tf, Tg and reactor NN2 Mole flow rate of N2 per area catalyst, kg mol/(m2 h)
length, x have been used to test the robustness for Tf Temperature of feed gas, K
both method including initial guesses which are Tg Temperature of reacting gas,K
10
Proceedings of the 5th WSEAS Int. Conf. on DATA NETWORKS, COMMUNICATIONS & COMPUTERS, Bucharest, Romania, October 16-17, 2006 268
7. References
[1] A. Murase, H. L. Roberts, and A. O.
Converse (1970). Optimal Thermal Design of an
Autothermal Ammonia Synthesis Reactor,
Industrial & Engineering Chemistry Research, 9,
503-513.
[2] Edgar, Thomas F.; Himmelblau, David M.
and Lasdon, Leon S. (2001). Optimization of
Chemical Processes. New York, McGraw-Hill.
11