Particle Swarm optimization

Contents
 Introduction to Optimization
 Particle Swarm Optimization
 Illustration of Algorithm
 Mathematical Interpretation
 Example
 Dataset Illustration
Introduction to Optimization

 The optimization can be defined as a mechanism through which the maximum or minimum value of a
given function or process can be found.

 The function that we try to minimize or maximize is called as objective function.

 Variable and parameters.
 Statement of optimization problem
Minimize f(x)
subject to g(x)<=0
h(x)=0.
 Two main phases Exploration and Exploitation
Introduction to Optimization

Application to optimization: Particle Swarm Optimization

Proposed by James Kennedy & Russell Eberhart (1995)

Combines self-experiences with social experiences

Particle Swarm Optimization(PSO)

 Inspired from the nature social behavior and dynamic movements with communications of insects,
birds and fish.
 Particle Swarm Optimization(PSO)
 Uses a number of agents (particles) that constitute a
swarm moving around in the search space looking for the
best solution

 in search space adjusts its “flying” according to its

 Each particle
own flying experience as well as the flying experience of other
particles.

 Each particle has three parameters position, velocity, and previous

best position, particle with best fitness value is called as global best
position.
 Contd..

 Collection of flying particles (swarm) - Changing solutions

Search area - Possible solutions

 Movement towards a promising area to get the global optimum.

 Each particle adjusts its travelling speed dynamically corresponding to the flying
experiences of itself and its colleagues.

 Each particle keeps track:

its best solution, personal best, pbest.

the best value of any particle, global best, gbest.

 Each particle modifies its position according to:

• its current position

• its current velocity

• the distance between its current position and pbest.

• the distance between its current position and gbest.

Algorithm - Parameters
f : Objective function

Xi: Position of the particle or agent.

Vi: Velocity of the particle or agent.
A: Population of agents.
W: Inertia weight.
C1: cognitive constant.
R1, R2: random numbers.
C2: social constant.
Algorithm - Steps

1. Create a ‘population’ of agents (particles) uniformly distributed over X

2. Evaluate each particle’s position according to the objective function( say

Y=F(x) = -x^2+5x+20
y=f(x)= -x^2+5x+20

1. If a particle’s current position is better than its previous best position, update it.

2. Determine the best particle (according to the particle’s previous best positions).
 Contd..

5. Update particles’ velocities:

6. Move particles to their new positions:

7. Go to step 2 until stopping criteria are satisfied.

 Contd…

Particle’s velocity
:

1. Inertia • Makes the particle move in the same direction and

with the same velocity

• Improves the individual

2. Personal • Makes the particle return to a previous position,
Influence better than the current
• Conservative

3. Social
• Makes the particle follow the best neighbors
Influence direction
 Acceleration coefficients
• When , c1=c2=0 then all particles continue flying at their current speed until they hit the search space’s boundary.
Therefore, the velocity update equation is calculated as:
t 1
 vij
t
vij
• When c1>0 and c2=0 , all particles are independent. The velocity update equation will be:

vij
t 1
 vij
t
 c1r1 j
t
P t
best ,i  xij
t

• When c1>0 and c2=0 , all particles are attracted to a single point in the entire swarm and
the update velocity will become

vij
t 1
 vij
t
 c 2r 2 j
t
g best  xij
t

• When c1=c2, all particles are attracted towards the average of pbest and gbest.
 Contd…

 Intensification: explores the previous solutions, finds the best solution of a

given region

 Diversification: searches new solutions, finds the regions with potentially the
best solutions

 In PSO:


Example:
Contd..
Contd..
Contd..
Flow chart of Algorithm
 Mathematical Example and Interpretation

Fitness Function

 De Jong Function
minF(x,y) = x^2+y^2

Where x and y are the dimensions of the problem. The surface and contour plot of the De Jong
function is given as:
 Mathematical Example and Interpretation

Contd..

Rastrigin Function

D
t 1
( xi  10. cos(2.  .xi )

The surface and contour plot of the De Jong function is given as:
 Mathematical Example and Interpretation

Contd…

Banana Function

The surface and contour plot of the Rosenbrock function or 2nd De Jong function
Or valley or banana functions given as:
 Mathematical Example and Interpretation
Example 1

Find the minimum of the function

f ( x)   x 2  5 x  20
Using PSO algorithm. Use 9 particles with initial positions
x1  9.6, x2  6, x3  2.6, x4  1.1,
x5  0.6, x6  2.3, x7  2.8, x8  8.3, x9  10
Solution Choose the number of particles
x1  9.6, x2  6, x3  2.6, x4  1.1,
x5  0.6, x6  2.3, x7  2.8, x8  8.3, x9  10
Evaluate the objective function

 120 .16, f 2  46, f 3  0.24

0 0 0
f1
 13 .29, f 5  22 .64, f 6  26 .21,
0 0 0
f4
 26 .16, f 8  7.39, f 9  30
0 0 0
f7
 Mathematical Example and Interpretation
Contd..

Let c1=c2=1 and set initial velocities of the particles to zero.

v1  0, v1  v 2 , v3 , v 4  v5  v6  v7  v8  v9  0
0 0 0 0 0 0 0 0 0 0

Step2. Set the iteration no as t=0+1 and go to step 3

.
Step 3. Find the personal best for each particle by

P 1 best ,1  9.6, P 1 best , 2  6, P 1 best , 3  2.6

So
P 1 best , 4  1.1, P 1 best , 5  0.6, P 1 best , 6  2.3
P 1 best , 7  2.8, P 1 best ,8  8.3, P 1 best , 9  10
 Mathematical Example and Interpretation
Contd..

Step 4: Gbest =max(Pbest) so gbest =(2.3).

Step 5: updating the velocities of the particle by considering the value of random
numbers r1 = 0.213, r2= 0.876, c1=c2=1, w=1.

v1  0  0.213(9.6  9.6)  0.876(2.3  9.6)  10.4244

1

v2  7.2708, v3  4.2924, v5  1.4892, v6  0, v7  0.4380, v8  5.256, v9  6.7452

1 1 1 1 1 1 1

Step 6: update the values of positions as well

 Mathematical Example and Interpretation
Contd..
So
 0.8244 , x 2  1.2708 , x 3  1.6924
1 1 1
x1
 1.8784 , x 5  2.0892 , x 6  2.3
1 1 1
x4
 2.362 , x8  3.044 , x 9  3.2548
1 1 1
x7

Step7: Find the objective function values of

f 1  23.4424 , f 2  24.739 , f 3  25.5978

1 1 1

f 4  25.8636 , f 5  26.0812 , f 6  26.21

1 1 1

f 7  26.231, f 8  25.9541, f 9  25.6803

1 1 1

Step 8: Stopping criteria

if the terminal rule is satisfied , go to step 2.
Otherwise stop the iteration and output the results.
 Contd..
 Step2. Set the iteration no as t=1+1 =2 and go to step 3

Step 3. Find the personal best for each particle by

 0.8244 , P  1.2708, P 2 best , 3  1.6924

2
2
P best ,1 best , 2

P 2 best , 4  1.87884 , P 2 best , 5  2.0892 , P 2 best , 6  2.3

P 2 best , 7  2.362, P 2 best ,8  3.044, P 2 best , 9  3.2548

Step 4: find the global best

Gbest  2.362

Step 5: by considering the random numbers in range (0,1) as

 0.113, r2  0.706
2 2
r1
 Contd..
 Find the velocities of the particles :

v1  11.5099, v2  8.0412, v3  4.7651, v4  3.3198

2 2 2 2

v5  1.6818, v6  0.0438, v7  0.4380, v8  5.7375, v9  7.3755

2 2 2 2 2

Step 6: update the values of positions as well

 12 .3343, x 2  9.312 , x 3  6.4575

2 2 2
x1
 5.1982 , x 5  3.7710 , x 6  2.3438
2 2 1
x4
 1.9240 , x8  2.6935 , x9  4.12078
2 2 2
x7
 Contd…
 Step7: Find the objective function values of

 70 .4644 , f 2  20 .1532 , f 3  10 .5882

2 2 2
f1
 18 .9696 , f 5  24 .6346 , f 6  26.2256
2 2 2
f4
 25 .9182 , f 8  0.7224 , f 9  17 .5839
2 2 2
f7

Step 8: Stopping criteria

if the terminal rule is satisfied , go to step 2.

Otherwise stop the iteration and output the results
 Contd..
 Step2. Set the iteration no as t=1+2 =3 and go to step 3

Step 3. Find the personal best for each particle by

P 3 best ,1  0.8244 , P 3 best , 2  1.2708, P 3 best , 3  1.6924

P 3 best , 4  1.87884 , P 3 best , 5  2.0892 , P 3 best , 6  2.3
P 3 best , 7  2.362, P 3 best ,8  3.044, P 3 best , 9  3.2548
Step 4: find the global best

Gbest  2.362
Step 5: by considering the random numbers in range (0,1) as

r1  0.178, r2  0.507
3 3
 Find the velocities of the particles

v1  4.4052, v2  3.0862, v3  1.8405, v4  1.2909

3 3 3 3

v5  0.6681, v6  0.053, v7  0.1380, v8  2.1531, v9  2.7759

3 3 3 3 3

Step 6: update the values of positions as well

 16 .7395 , x 2  12 .3982 , x 3  8.298

3 3 3
x1
 6.4862 , x 5  4.4391, x 6  2.3968
3 3 3
x4
 1.786 , x8  4.8466 , x 9  6.8967
3 3 3
x7

Step7: Find the objective function values of

 Contd..

 176 .5145 , f 2  71.7244 , f 3  7.3673

3 3 3
f1
 10.3367 , f 5  22.49, f 6  26..2393
3 3 3
f4
 25.7402 , f 8  27.7222 , f 9  62.0471
3 3 3
f7

Step 8: Stopping criteria

if the terminal rule is satisfied , go to step 2.

Otherwise stop the iteration and output the results
 Mathematical Example and Interpretation
Example
Iteration First:

Fitness Function :De Jong function min F ( x , y ) x 2

 y 2

Where x and y are the dimensions of the problem , the velocities of all the particles are
initialized to zero and inertia (W) = 0.3, and the value of the cognitive and social constants
are
C1= 2 and C2 =2. The initial best solutions of all the particles are set to 1000
P1 fitness value = 12  12  2
 Mathematical Example and Interpretation
Example
Iteration 2nd:
 Mathematical Example and Interpretation
Example
Iteration 3rd:
 Mathematical Example and Interpretation
Example
Iteration 3rd:
 Psuedocode

1. P=particle initialization();
2. For I =1 to max
3 for each particle p in P do
fp=f(p)
4. If fp is better than f(pbest)
pbest =p;
5. end
6. end
7. gbest = best p in P.
8. for each particle p in P do
9.

10.

11. end
12. end
DataSet
 Advantages and Disadvantages of PSO
Advantages

 Insensitive to scaling of design variables.

 Simple implementation.
 Easily parallelized for concurrent processing.
 Derivative free.
 Very few algorithm parameters.
 Very efficient global search algorithm.

Disadvantages

 Slow convergence in refined search stage (weak local

search ability).