You are on page 1of 9

Chapter 7 Integer

Programming – The
Method of Branch and
Bound
Shi-Shang Jang
Chemical Engineering Departme
nt
Introduction
 In many cases, engineering optimization
problems are taking some variables that only
have their values in a form of integer. For
instance:
 Maximize Z=x12+x2x3-x32
Subject to -2x1+3x2+x33
x1, x2, x3{0,1}
Heat Exchanger Network Superstructure
(Yee and Grossmann, 1991)

Stage 1 Stage 2
Q111 , Z111 Q112 , Z112 Qcu ,1 , Z cu ,1

H1-C1 H1-C1 CW
T11 T12 T13
Tin ,1 Tout ,1
Qhu ,1 , Z hu ,1 Q121 , Z121 Q122 , Z122
T11 T12 T13
Tout ,1 H1-C2 H1-C2
Tin ,1

S
Q211 , Z 211 Q212 , Z 212
Qcu , 2 , Z cu , 2
H2-C1 H2-C1
CW
T21 T22 T23
Tin , 2 Tout , 2
Qhu , 2 , Z hu , 2 Q221 , Z 221 Q222 , Z 222
T23 T22 T21
Tout , 2 H2-C2 H2-C2
Tin , 2
S

Temperature location Temperature location Temperature location


k=1 k=2 k=3
Objective Function— minimum annual cost

min  CCU QCU ,i   C HU QHU , j


iHP jCP

    CF z
iHP jCPkST
ij ijk   CF
iHP
i ,CU Z CU ,i   CF
jCP
j , HU Z HU , j

Cij Qijk
   U
iHP jCPkST [Tijk Tijk 1 (Tijk  Tijk 1 ) / 2]1/ 3
ij

CA, i, CUQCU ,i
 U
iHP [TCU ,i (Tout ,i  Tin ,CU )(TCU ,i  Tout ,i  Tin ,CU ) / 2]1/ 3
CU ,i

CA, j , HUQHU ,i
 U
jCP [ T (T  T )( T  T  T ) / 2]1/ 3
HU , j HU , j in , HU out , j HU ,i in , HU out , j

Tijk  Tik  T jk , Tijk 1  Tik 1  T jk 1 , TCU ,i  Tik  Tout ,CU , THU ,i  Tout , HU  T jk
5,000 selected samples

  Stage1 Stage2 Utilities  

Method
Q(1,1) Q(2,1) Q(1,2) Q(2,2,) Q(1,1) Q(2,1) Q(1,2) Q(2,2) Qc(1) Qc(2) Qh(1) Qh(2) Cost

Global 682 0 0 0 0 2426 1950 0 168 1973 491 0 155497

BP 1445 0 0 0 0 1203 0 1950 1354 1247 951 0 185531

Horse
Race+ 483 0 0 0 0 2536 1950 0 367 1863 580 0 163196
GRNN

SA 346 0 0 0 0 2525 1950 0 504 1874 728 0 173437

IDS 635 0 0 0 0 2445 1950 0 215 1954 519 0 155895


Example
 Maximize Z=3x1+2x2
Subject to x12
x22
x1+ x2 3.5
X1, x2 0 and integer
The Method of Branch and Bound –
Integer Linear Programming Case

Consider an MIP(Mixed integer :


Max Z=cx
Subject to Ax=b
x>0
xj is an inter for jI
Step 1: Solve the above LP (LP1) without integ
er constraints. The solution of Z1 is the upp
er bound on the maximum value of Z for the
MIP problem.
Branch and Bound-Continued
Step 2: Select the proper branching on one proposed integer variables
at a fraction value.
Step 3: Given xj selected from Step 2, solve LP2 such that xj  j , and
LP3 such that xj  j , where j , and j are two nearest integers.
In other words:
LP2 LP3
Max cx Max cx
s.t. Ax=b Ax=b
xj  j xj  j
x 0 x 0

If the solutions to the above are still fractional and hence are infeasible
to the MIP problem with integer restrictions. The next step is to sele
ct LP2 or LP3 and branch th from that bu adding a new constraint.
X1=2 , X2=1.5
Node 1
Z0=9
LP-1
X2  2
X2 1
(optimal) Node 3
X1=2 , X2=1 LP-3
Node 2 X1=1.5 , X2=2
Z0=8
LP-2 Z0=8.5
X1  1 X1  2

Node 4 Node 5
LP-4 LP-5

X1=1 , X2=2 Infeasible


Z0=7

You might also like