Genetic algorithms

Introduction
• Based on mechanics of natural genetics & natural
selection
• GA’s are difficult from classical optimization methods
– Deterministic tendency is replaced by probabilistic tendency to
evolve the solution
• One could get global solution using GA
– deterministic technique can give only for strictly convex or
concave problems
• First thing to learn is the working principle of GA
 Working principle
Problem statement:

Maximize f(x)
such that
x i  xi  x i
L U

Here xi is a vector for variables i numbered 1, 2, 3……….N

The first step in GA is to encode the continuous variable xi into binary

string. This process is called coding
 Working principle
Coding:

 Binary coded strings having 1’s and 0’s are used

 The length of the string is usually determined according to
the desired solution accuracy
 Illustration: If Four bits are used to code each variable in a
two variable problem, strings (0000, 0000) and (1111, 1111)
would represent the points
L L U U
[ x1 , x2 ]T [ x1 , x2 ]T

 Why ? Because (0000) and (1111) represent minimum and

maximum decoded values
 Working principle
Conversion from string to real value:
xi  xi
U L
xi  xi  Li
L
[ Decoded value( si )]
2 1
li 1
Decoded value (s i )   2i si
i0

where s i  [0 1]
string s represente d by (s l-1s l-2 ........s2 s1s0 )

First parameter (1100100000) decodes to 800. Corresponding

parameter value is 0 + (6-0)*800/1023 = 4.692
Second parameter (1110010000) decodes to 912. Corresponding
parameter value is 0+ (6-0)*912/1023 = 5.349
 Working principle
Conversion from string to real value: Example

Substring s i 0111 :
Decoded value (s i )  [(1)2 0  (1)21  (1)2 2  (0)23 ]
 7
Remarks

 Not all variables in the problem need same string length

 String length for a particular variable depends on the level of accuracy
desired in the solution
[ xi  xi ]
U L
 Accuracy obtained in li bit string is given as
2 li
 Working principle
Coding: Summary

 Each variable is defined with a discrete set of binary set of sub-strings.

These sub-strings can be called as population of sub-strings
 Each sub-string corresponds to a precision value of the corresponding actual
value of the variable
 The length of the sub-string defines the accuracy in the solution
 All we are saying here is that we are discretizing all the vector in which
solution vector is embedded
 The string is therefore, a combination of different sub-strings. It is the string
that we will try to manipulate in future with different genetic operators.
 The length of sub-string is dependent on the accuracy required in the variable
that has to be optimized.
 Working principle
Fitness function

 GA principle: Survival of the fitttest

Therefore, we would maximize a function only in GA never minimize it
 The function that has to be maximized is called as Fitness function F(x)
 If f(x) is to be minimized, then F(x) is developed from f(x) as

1
F ( x) 
1  f ( x)
Fitness value of a string is called string’s fitness
 Working principle
Genetic algorithm: Summary

Generate Initial Population

Fitness Function
Evaluate Fitness

Yes
Termination Condition? Best Individual

No

Select Parents
Crossover, Mutation

Generate New Offspring

 Working principle
Genetic algorithm: Explanation

 String manipulation is done using three operators

Reproduction : Rejects many worst strings and selects few
strings whose fitness is well above average
Cross over: Select a couple of parents strings and perform
changes by crossing over to get two offspring strings
When reproduction is done between two intelligent genes,
the probability to produce an intelligent gene is high
Mutation: Select a particular string and randomly perform
changes for one or more characters in the string
Change 0 to 1 or 1 to 0 in a random fashion for one of the
character.
 Working principle
Genetic algorithm: Terminology

 Population: The number of strings available at any time that could

be subject to genetic operation
 Termination criteria: Based on
Maximum number of generations (one combination of three
operators applied to population)
No improvement in the fitness value after a fixed number of
iterations
 Reproduction allows move randomly out of the neighbourhood.
Mutation allows refining the location of the string in the
neighbourhood.
 Working principle
Genetic operator: Reproduction

Also known as selection operator

 Selects good strings in the population and forms a mating pool
Above average strings are picked from the current population
Multiple copies are inserted in the mating pool in a probabilistic
manner
Proportionate reduction operator:
String selection based on the probability evaluated from fitness
Probability for selecting ith string is given as
F
p  n
i

F
i

i
i 1
n is the population size
 Working principle
Genetic operator: Reproduction

 In order to choose n strings, n random numbers between 0 to 1 are

created random
The cumulative probability is calculated based on the individual string
probabilities
A string that represents chosen random number in the cumulative
probability range (calculated from fitness values) is copied to the mating
pool
String with higher fitness represents larger range in the cumulative
probability values.
String with lower fitness represents smaller range in the cumulative
probability values.

Good strings are probabilistically assigned a large number of copies

 Working principle
Genetic operator: Crossover

 During reproduction, no new strings are generated

During crossover, new strings are created
By exchanging information among strings in the mating pool
General crossover mechanism
Two strings are picked up from the mating pool
Some portions of the strings are exchanged between the strings
Single point cross over
Randomly choose a crossing site
Exchange all the bits on the right side of the crossing site
 Working principle
Genetic operator: Crossover

 Example

00000 0 0 111
11111 1 1 000

Parent Children
strings strings

 Idea: Good parent strings are combined to form a better child string if
an appropriate site is chosen
 But good children strings may or may not be produced.
We do not worry much about this, because bad children strings
would be discarded in the future mating pool based on their fitness.
 Working principle
Genetic operator: Mutation
Mutation operator changes 0 to 1 and vice versa with a probability pm
Simulation
Choose a random number between 0 and 1
If the random number is less than pm perform mutation
Alter the chosen string at the random site
Other wise, keep the bit un altered
Mutation creates a point in the neighbourhood of the current point
A local search around the current solution is perormed
Mutation also maintains diversity in the population
 Working principle
Genetic operators: Summary
Reproduction: Good strings are selected
Cross over: Good strings are combined to create a better child stirng
Mutation: Alters a string locally to hopefully create a better child string
What happens to bad strings created ?
Eliminated in reproduction phase in future iterations
What happens to good strings created ?
Emphasized in an increasing manner due to the fitness probability
Solutions close to global optimality would be obtained due to
which GAs will be attractive.
 Algorithm
Step 1: i) Choose a coding to represent problem parameters, a selection
operator, a crossover operator, and a mutation operator.
ii) Choose population size n, crossover probability pc and mutation probability
pm.
iii) Initialize a random population of strings of size l.
iv) Choose a maximum allowable generation number tmax. Set t = 0
Step 2: Evaluate each string in the population
Step 3: If t > tmax or other termination criteria is satisfied, Terminate.
Step 4: Perform reproduction on the population
Step 5: Perform crossover on random pairs of strings
Step 6: Perform mutation on every string
Step 7: Evaluate strings in the new population. Set t = t +1 and go to Step 3
 Illustrative example
Problem statement

Minimize the function

f ( x , x )  (x  x  11)  ( x  x  7)
2 2 2 2
1 2 1 2 1 2

0  x ,x  6
1 2

True solution is (3, 2)T with f value of about zero at this point
 Illustrative example
Step 2 Evaluate each string in the population
 Evaluate each string in the population. Fitness of each string.
xi  xi
U L

First string: xi  xi 
L

2 Li  1
[ Decoded value( si )]

First parameter (1100100000) decodes to 800.

Corresponding parameter value is 0 + (6-0)*800/1023 = li 1
Decoded value (s i )   2i si
4.692 i0

Second parameter (1110010000) decodes to 912. where s i  [0 1]

Corresponding parameter value is 0 + (6-0)*912/1023 =

5.349 string s represente d by (s l-1s l-2 ........s2 s1s0 )

The function value is f(x(1)) = 959.689.

Fitness value F(x) = 1/(1+f(x)) = 1/(1+959.680) =
0.001
Similarly, evaluate other strings are evaluated and fitness
is calculated.
Note that the strings are chosen randomly initially and assigned to a string
 Illustrative example
Step 2: Summary of evaluations
 Illustrative example
Step 3: If t > tmax or other termination criteria is
satisfied, Terminate.

Since t = 0 < tmax = 30 we

proceed to step 4
 Illustrative example
Step 4:Perform reproduction on the population

 Select good strings and place them in the mating pool

 For proportionate selection, first calculate the average fitness of
the initial population
Average fitness = (Sum of fitness values)/Population size
= 0.008
 Expected count for each string is given as F(x)/Average
fitness
Probability of each string = Expected count/Population size
Evaluate cumulative probability
 Illustrative example
Step 4: Perform reproduction on the population
Population EXPECTED STRING CUMULATIVE

Tabulated string counter COUNT PROBABILITY PROBABILITY

1 0.13 0.007 0.007

probabilities 2

3
1.10

0.98
0.055

0.049
0.062

0.111

4 1.85 0.093 0.204

5 0.25 0.013 0.217

• Average fitness = (Sum of 6 1.71 0.086 0.303

fitness values)/Population size 7 1.34 0.067 0.370

= 0.008 8 0.12 0.006 0.376

• Expected count for each string 9 0.49 0.025 0.401

is given as F(x)/Average fitness 10 2.96 0.148 0.549

• Probability of each string = 11 0.14 0.007 0.556

Expected count/Population size 12 2.84 0.142 0.698

• Evaluate cumulative probability 13 0.36 0.018 0.716

14 0.24 0.012 0.728

15 0.37 0.019 0.747

16 0.61 0.030 0.777

17 1.22 0.060 0.837

18 1.09 0.054 0.891

19 2.08 0.103 0.994

20 0.13 0.006 1.000

 Illustrative example
Step 4:Perform reproduction on the population

 For forming good mating pool

 Create random numbers from 0 to 1
 Identify the particular string which is specified by each of
these random numbers
 If random number is 0.472 is created, tenth strings gets
a copy in the mating pool
 The selection procedure is repeated n times
 The number of selected copies for each string is
counted after the selection procedure is finished.
 Illustrative example
Step 4: Mating pool selection
Original Matting pool
papulation
pool

1 2

2 3

3 4

4 4 You could
5 6
see that
6 6

7 7
both
8 9 expected
9 10 count and
10 10

11 10
true count
12 12 agree more
13 12 or less.
14 12

15 13

16 14

17 17

18 18

19 19

20 20
 Illustrative example
Step 4: Figure of initial population and mating pool

Conclusion
Inferior points have
been probabilistically
eliminated for further
consideration.
No. Chrom- z f A B C D E F Mating
-osome function 1/(1+z) f/f avg A/Psize Bcumu randon( 0,1) pool
1 Kr1 5.320 0.158 0.940 0.063 0.063 0.356 6.000 0.000 Kr2
2 Kr2 4.904 0.169 1.010 0.067 0.130 0.871 14.000 2.000 Kr2
3 Kr3 4.691 0.176 1.050 0.070 0.200 0.128 2.000 1.000 Kr3
4 Kr4 5.013 0.166 0.990 0.066 0.266 0.325 5.000 1.000 Kr4
5 Kr5 4.955 0.168 1.000 0.067 0.333 0.249 4.000 3.000 Kr5
6 Kr6 4.762 0.174 1.030 0.069 0.402 0.642 10.000 2.000 Kr5
7 Kr7 5.215 0.161 0.960 0.064 0.466 0.333 5.000 0.000 Kr5
8 Kr8 4.838 0.171 1.020 0.068 0.534 0.864 13.000 0.000 Kr6
9 Kr9 4.866 0.171 1.010 0.067 0.601 0.371 6.000 0.000 Kr6
10 Kr10 5.115 0.164 0.970 0.065 0.666 0.275 5.000 2.000 Kr10
11 Kr11 5.352 0.157 0.930 0.062 0.728 0.712 11.000 1.000 Kr10
12 Kr12 4.821 0.172 1.020 0.068 0.796 0.170 3.000 0.000 Kr11
13 Kr13 4.691 0.176 1.050 0.070 0.866 0.910 14.000 1.000 Kr13
14 Kr14 5.111 0.164 0.970 0.065 0.931 0.125 2.000 2.000 Kr14
15 Kr15 4.745 0.174 1.040 0.069 1.000 0.605 10.000 0.000 Kr14
fAvg 0.167993 1.000 15.000
 Illustrative example
Step 5: Cross over operation

Single point cross-over: Two strings are selected at random and crossed at
random site
Pick pairs of strings from the top of the list
 (3, 10) first cross over, (2, 14) second cross over, (10, 19) third cross
over, (19, 17) fourth cross over etc.
 Recall total string length is 20.
For first cross over, Choose a random number and see if the random
number is less than pc = 0.8
Say this time it is true, and therefore cross over has to be carried out
 Choose a random site between (0, l-1) or ( 0 , 19). That turns out to be 11.
Cross the strings at the site 11 and create two new strings.
 Illustrative example
Step 5: Single point first cross over Illustration

Parent strings from mating pool

0010100100 1010101010 1010100001 0111001000

Crossover site No. 9.

001010010 0 1010101010 101010000 1 0111001000

Children strings

0010100101 0111001000 1010100000 1010101010

 Illustrative example
Step 5: Single point Fourth cross over Illustration

Parent strings from mating pool

0011100010 1011000011 0111000010 1011000110

Random number < pc NO. This implies no cross-over.

Children strings (Same as parent strings)

0011100010 1011000011 0111000010 1011000110

 Illustrative example
Step 5: Cross over summary

Population size = 20,

pc = 0.8. Expected
no. of cross over =
0.8*20/2 = 8.
However, we only
performed seven
cross overs with 3
cases where parent
strings are child
strings.
 Illustrative example
Step 5: Cross over Contour

 Some good points are

created
Some bad points are
created
Points far away from
parent points are created
Points closer to parent
points are created.
Conclusion: All random
moves are included that
allow variegated
distribution.
 Illustrative example
Step 6: Mutation

 pm = 0.05
 If random number < pm then do mutation else copy the parent string
as child string
 For pm = 0.05, n = 20, l = 20, we can expect to alter 0.05*20*20 or 20
bits in the population.
Note: Here we carry out bit wise mutation. We can complicate
mutation process by allowing more sites to undergo mutation in a
single generation
Those where random numbers are less than pm mutation is not carried
out and parent strings are placed as children strings.
 Illustrative example
Step 6: Mutation summary
 Illustrative example
Step 6: Mutation contour

 Mutation operator changes a

point locally some times
Mutation also brings large
changes
 Like cross over, mutation
created both good and bad points.
Such flexibility is required to
explore the search space properly
 Illustrative example
Step 7: New population evaluation

 Resulting population becomes the new population

Evaluate each string as before by first identifying the
substrings of each variable and mapping the decoded values
of the substrings in the chosen intervals
This completes one iteration of the genetic algorithms
We increment the generation counter to t = 1 and proceed
to step 3 for the next iteration.
 Illustrative example
Step 7: New population summary

• New population shown in

empty boxes
•In one iteration some
good points have been
found
 Illustrative example
Step 7: New population summary

 Average fitness of new population = 0.015

 Average fitness of initial population = 0.008
 Best point fitness in new population 0.050
 Best point fitness in initial population 0.024

Do Step 3 until maximum

allowable generation is reached
or some other termination
criteria is met.
 Illustrative example
Population after 25 generations

 Best point (3.003, 1.994) with

function value 0.001
 Fitness value 0.999
 Average population fitness is
0.474
Figure shows that many points
are clustered around the true
minimum
 Some inferior points are also
there which are due to some
unsuccessful crossover events
Total number of function
evaluations required are
0.8*20*26 = 416
 Illustrative example
Variation of population best function value with generation
number

 Three different runs with different initial population

 It can be observed that for all the runs one could quickly converge to a solution close
to the true optimum
 Critical generation number is around 8 for the runs carried out.
 Conclusions
Genetic algorithms are robust and very much applicable for process systems
research
 For systems engineering, we need to depend on an NLP /LP solver for an MINLP
problem tried to solve using binary coded Gas.
 The combinations of the three operators is wonderful. We first try to eliminate
maximum junk points, then allow reproduction on good points and further allow
possibility for local search.
 GAs can be also applied for constrained optimization problems
Use penalty functions where constraints are coupled with the objective function
Real Coded GAs are also familiar
Instead of Binary strings use real numbers directly in the fitness evaluation and
perform reproduction, crossover and mutation on the real numbers only rather
than on the strings.
 End of the day, what GA did is some thing better than what you
learnt in your earlier lectures. GA is not a determinstic but a
stochastic and probabilistic technique for the evolution of
optimal solutions from initial population. GAs are very robust
and can be applied for optimization of any engineering system let
it be Chemical, Mechanical, Electrical or Control systems.

THANK YOU VERY MUCH FOR YOUR ATTENTION. IF YOU

UNDERSTOOD EVERY SLIDE OF THIS PRESENTATION, YOU CAN
EASILY GO AHEAD IMPLEMENTING GA’S FOR SOLVING SOME
OPTIMIZATION PROBLEMS YOU ENCOUNTER IN YOUR
ENGINEERING EDUCATION AND RESEARCH.