Scribd
Upload
8 views

22 Multi Objective Optimization

Uploaded by

ZoycheN KenglanG
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
8 views

22 Multi Objective Optimization

Uploaded by

ZoycheN KenglanG
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 17

Multi Objective Optimization

Prof. (Dr) Arup Kr Sarma


Civil Engineering Department
Indian Institute of Technology Guwahati
1
Multi objective optimization
• In water resource optimization there may be several non commensurate
objectives for which multi objective optimization becomes necessary.
• For example lets consider a Multipurpose WR project with two
objectives:
• Power Maximization
• Flood reduction through flood control storage
• Effort to increase power would require water level to keep high whereas
effort to reduce flood would requires to keep the water level low to
increase flood cushioning provision
• Thus if equal weightage is given for both the objectives, then it become
conflicting for the decision maker. 2
Multi objective optimization
If we have k no. of objective functions say,
f1 ( x), f 2 ( x)..... f k ( x)
where X =[ x1, x2…. xn ] is a set of variables then multi objective
optimization problem can be stated as
Max/min Z = [ f1 ( x), f 2 ( x)..... f k ( x)] subject to
g j ( x)  0 where j= 1,2,3….m

X = x1 , x2 .....xi >=0 for i=1,2,3…..n

3
Multi Objective Optimization
• This sort of problem is also known as vector optimization problem.
• In general, no solution vector X exist that max/min all the k objective
function.
• Therefore a concept Pareto optimum solution is used in multi objective
programming.
• Another way to obtain solution for MOP is weighing method i.e.,
weightage is given to each of the objective functions and then optimize.
• Another approach is to optimize one objective function by putting other
objective function in the constraints. 4
Concept of Pareto Optimum
Vilfredo Pareto (1848–1923), Italian engineer and economist
Concept of Pareto optimal solution
A feasible solution X 1 is called Pareto optimal if there exist no other feasible
solution X 2 such that f i ( X 2 )  f i ( X 1 ) for i=1,2,3…..k, i j
with
f j ( X 2 )  f j ( X1 ) for at least one j, j=1,2,3….k

X 1 is called Pareto optimal solution if there exist no other feasible solution X 2


that reduce the value of some objective function without causing any simultaneous
increase in at least one other objective function
5
Graphical Representation of Pareto Optimum
Min Z=  f1 ( x), f 2 ( x)

Pareto Optimum
120

100

80

60

40

Function1 Function2
20

X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X 13 X14 X15

Pareto Optimum
6
Multi Objective Optimization: Solution Methods
• In the above curve any solution lying between X6 and X10 are Pareto
optimal solution. Such solution are also called non inferior solution or
efficient solution.
• Different method exist for solution of MOP. Most of these methods
generate a set of optimal solution that use some additional criteria to
select one optimal solution from the set of Pareto Optimum solutions.
• Some other methods of solving MOP are discussed in the subsequent
slides
7
1) Utility function method
Utility function U i ( f i ) for each of f i is decided depending on the importance
of that objective function as compared to the other objective functions and total
utility is computed as
k
U =  U i ( f i ) =  wi f i ( x)
i =1

Where wi = scaler weighing factor


Solution vector X is chosen as optimal solutiom which maximize U subject to all
constraints. This method is also known as weighing factor method.

8
2) Inverse utility function method

k
1
U −1
=
i =1 U ( f )i

The solution corresponds to minimum of U −1

9
3) Global criteria method
In this method the optimal solution is found by minimizing a pre selected global
criteria.
p =2
k
 fi ( x ) − fi ( xi ) 
* i th

F(x) =  
i =1 
i
* 
fi xi 
th
where xi* is the ideal solution for the i objective function

10
4) Bounded objective function method
Let Li and Ui are the minimum and maximum acceptable limit for the i th
objective function then X * is found by optimizing the most important
objective function( f r ) .Thus the problem become-
Optimize ,subject to
f r ( x)
g i ( x ) <=0
Li <= f i <= U i (for i=1,2,3…..k)
ir

11
5) Lexicographic Method
• Objective functions are ranked in order of importance by the
designer. The first ranking objective function is first optimized
then the optimal value of that objective function is kept as a
constraint while optimizing the second objective function. The
procedure is repeated till the last objective function is optimized.
Optimize, Z= f i (x) (for i=1,2,…..n)
subject to g j (x) = 0 for (j=1,2,….m )
ef ( x) = f * where e= rank (e=1,2,3…n-1)
e

f e* = optimal value of eth objective function Keeping


value of all higher ranking objective function value equal to their
optimal value
6) Goal Programming method
• In this simplest version the designer sets goal for each objective
function( f i ) which designer wishes to attain. Let bi is the goal for each i th
x1

objective function, then the optimal solution X( X 1 , X ,…… X n ) is that


2

which minimizes the deviation from the set goals.


Let di + is the under achievement and di − is the over achievement from
the set goal bi of the objective function fi x
then the mathematical statement can be written as
− p 1/ p
minimize (di + di )  where p>=1 and p depends on the utility
+
 
function chosen by the designer. Constraints are
+ − + −
g i ( x ) >=0 and fi ( x) + di − di = bi , di , di  0
13
7. Indifference curve method
• Let us consider two objective function with two decision variables
x1

• Subject to
x1

14
Plotting of the Constraints

Plotting the above constraints ,the apex values are (0,3),(1,4),(4,4) and (6,2)

15
Plotting of the Functions and
Determining Optimal Solution

In absence of an information it is difficult to choose one as best from these


non inferior solution. However if a preference curve is given in terms of a
indifference curve or surface then the solution can be found out as shown
in graph 16
17

You might also like