Escolar Documentos
Profissional Documentos
Cultura Documentos
20,No.617,
pp.703-709, 19%
Pergamon
$8
./:
0098-1354(95)00203-0
IERAPETRITOU,.~.ACEVEDO~~~ E.N.Pwmco~ou~os~
Abstract-The
problem of selecting an optimal design/plan for process models involving stochastic
parameters is addressed in this paper. A classification of uncertainty is introduced &pending on its
sources and mathematical model structure. A combined multiperiodlstochastic optimization formulation is then proposed along with a decomposition-based algorithmic piocedure for its solution. The
approach is illustrated with a process synthesis/planning example problem.
1. INTRODUCTION
Production
systems
typically
involve
significant
in their operation
2.CLASSIFICATIONOFUNCERTAINTY
We consider
form:
process
s.t.
models
of the following
h(d, z, x, g) =0
g(d,z,x,WsO
xEX,zEZ,dED
should be addressed.
703
M. G.
704
Problem (P)
)
&&I,
s.t.
i. -
IERAPETRITOU et
z, e)
f(d, z, B)sO
zEZ,deD
al.
z (D2)-design
with optimal degree of flexibility,
if the trade-offs between economics and flexibility are properly explored.
Note that, from a modelling point of view, all
types of uncertainty could in principle be included in
either design objective. Therefore, in general, some
uncertain parameters will be associated with design
objective Dl, i.e. considered as deterministic,
while some others will be treated in a stochastic
manner (linked to design objective D2).
3. COMBINED MULTIPERlOD/STOCHASTC
PROGRAMMING
f(dd,,W~o
zEZ,dED
edE7-={edIe+edse:),
e,EJ(e,)
s.t.
705
Problem (MS-P)
where:
p=l,...,
e,)
P
zPEZ,dED,fl,EJ(&)
where the evaluation of the expectancy is performed
within the corresponding
feasible region R(d),
R(d) ={~,IV&ER~Z~: f(d, Zp, ep,, e,>So)vp}.
Problem
(MS-P) mathematically
describes a
design strategy consisting of two stages. Once a
design has been selected at the first (design) stage,
the objective of the second (operating) stage is to
determine an optimal vector of control variables zP
for every possible realization of the uncertainty 0,
and t3P,.Note that the overall objective of simultaneous economic optimization and design feasibility
is enforced at the design stage, since the selection of
the design variables is based on the maximization of
the expected profit evaluated inside the corresponding feasible region.
The general formulation in (MS-P) captures the
various types of uncertainty and design objectives
for flexibility in a unified way. Consequently, most
proposed formulations for flexible design optimization can be viewed as special cases of this combined
multiperiod/stochastic
optimization
problem; for
example,multiperiod
design optimization problems
(Varvarezos et al., 1992) for the case when only
deterministic uncertain parameters are involved,
stochastic design optimization problems (Pistikopoulos and Ierapetritou, 1994) if stochastic uncertain parameters
are considered. _ An attractive
feature of the formulation in (MS-P) is that it preserves the block diagional structure (in the constraints) of- the conventional multiperiod optimization problems.
This property will be put to
advantage by properly employing a decompositionbased strategy, as will be discussed in the following
sections.
I ier - e3
s.t. f(d,z+,
es, e:)sovp
ey =o.qe:(i
IW
0.
WqpPqp(& Zqp,es,
efp).r(ofp)
QXP
From the dual information obtained from the
solution of the feasibility and profit subproblems,
and the application of the general correction factors
of Appendix A, the following master problem is
constructed, yielding an upper bound to the solution
of (MS-P) and a new set of design variables d.
s.t.
PG EPk(d)-
$[f(d,
zk@,8p,, @qP))]
QxP
I&
-c I c
&J,kf( .) +LJf( .)
I
i=2
-~~~kf(.)-~~zfL(.)
k=l,.
..,K
M. G.
706
.,
IERAPETRITOIJ
. .. .
;t$
etul..
:
0:
1
OPTIMIZE PROFIT FUNCTION
AT EACH PERIOD (es) AND
QUADRATURE POINT (B#
'..
:,/1
i
EVALUATE EXPECTED
information PROFIT EPk AND UPDATE
LOWER BOUND IF EPkbEPL
Dual
.
fOFTIMAL DESIGN/PLAN)
li
dk
-<._> ,_
Process
Process
Process
Process
Process
Prodhction constant
_
(MP)
18
20
15
100
80
130
150
1
2
3
4
i .~ -
707
Distribution
function
Mean
value
Positive
deviation
Negative
deviation
Demand of C (6,)
Availability of A (0,)
Cost of A (0,)
Cost of B (0,)
N(17.5. 2.5)
N(27.5, 2.5)
N(250,lO)
N(300, 15)
17.5
27.5
250
300
7.5
7.5
30
45
7.5
7.5
30
45
. ..
Periodskenarios
Kinetic
constants
Then
12
4. PROCESS SYNTIiESIShLANNING
EXAMPLE PROBLEM
Demand limitation
Availability
limitations
Production
limitations
Objective function
A?+A,+A,=A
&+B3+B,=Bs
C=O.9 (B,+BI)
_
Er= Ml, In(1 +A,/k,)
E,= MP3 In(1 +A,/k,)
B, = MP, In(1+ Ad/k,)
56
k&J
4(6,)
19
20
25
19
21
26
20
21
25
20
22
26
20
22
27
21
22
27
21
21
26
21
20
26
Weights
0.1
0.15
0.15
0.15
0.1
0.1
0.15
0.1
k2 (e,)
CONCLUSIONS
CSB,
E,C 15
Ace,
Als25yl
A3<20~3
Ads20yd
Profit = 55OC- (&A + f&B,)
- I(% + 15(81+ &)) - (PCZY~+ 5A2)
- (PGyr + 15~43)- (PCIY~+ 5,431
The utilization of raw material at
processes 2.3.4. respectively
The production of intermediate
product B at processes 2.3,4, respectively
The production of final product
The amount of material B purchased
(0.1) binary structural variables
representing the existence (or not) of
units 2. 3 and 4
2.20 i
2.80 1.80 1.60 1.40 1.20--,//
1.00 e
1.06
2.00
3.00
Iteration
M. G. IERAPETRITOU efal.
708
decomposition-bas@:algtirithm
has also been presented that prdperly exploits the structure of the
optimiza&ion model and illustrated with an example
problem.
i.
APPENDIX
Correction Factors
Feasibility subproblems
REFERENCES
SET
ON CCE364IX
I;wi+,w...
W&-l
e;;,-f-hq
2
b&&l
X(g%~PJ}~~.
I
CF;=
CF, - 2
pi{-0.5(1
+vi)}
9i
CFI=C~i+~~lr{-0.5(1-vi)}
9i
A; = CFyfi
1
I= CF!f!
#I
(1989).
Uncertain Cost
Property:
709
max[P,,(d. z, S:)]J(el)
dfJ:+b.rE{e:)
J(6):) de:
0;
=
The
P(d,
z94)l
and since the cost coefficient does not affect the feasible .:
region, the inner integration is evaluated in the whole.
space of 0:and is equal to 1.
The second integral can also be simplified as:
_r(e,) de,
As a result:
tion:
m=Pdd,
Je,
2,
~:wuw~,+
bTep(e,) de,
J es
the integrals
)
max[P,,(d. Z, eJ)]J(e:)J(e:de:de:
0; z
+
II
bPey(e;)qe:) de:de:
0; K
bE{e:}J(e:) de:
0; :
=
I K
m;x[P(,(d.
0;
Z,
J(e:) de: