SlideShare una empresa de Scribd logo
1 de 39
Descargar para leer sin conexión
New Multi-parent Crossover
based on Crossing Two
Segments Bounded by Selected
Parents
Atthaphon Ariyarit, Kanazaki Masahiro
Department of Aerospace Engineering,
Graduate School of System Design, Tokyo Metropolitan University
第6回進化計算学会研究会 2014/03/07
Outline
• Introduction
• Objective
• Overview of Popular Crossover Operators
• Blended Crossover (BLX)
• Unimodal Normal Distribution Crossover (UNDX)
• New Multi-parent Crossover Method
• Test Problems
• Single-objective optimization problems
• Multi-objective optimization problems
• Multi-objective airfoil optimization problem
• Results
• Conclusions
2
Introduction
• The Genetic Algorithms(GA) is popular optimization method to solve
single-objective and multi-objective optimization problem
• GA have three main operator, such as, Selection, Crossover and
mutation
• Crossover operator is the main operator for GA performance
• The popular Crossover operator is BLX and UNDX Algorithm[1]
• The GA is popular in aerospace engineering, such as, to increase the
efficiency of aircraft, to used for the navigation of aircraft, or to
reduce the weight of the aircraft weight, etc.
[1] Hajime K, Isao O and Shigenobu K. Theoretical Analysis of the Unimodal Normal Distribution Crossover for Real-coded
Genetic Algorithms. Transactions of the Society of Instrument & Control Engineers, 2002, 2(1), 187-194
3
Introduction
• Procedure of Genetic Algorithms
• Genetic operator such as selection, crossover and
mutation are applied to the parent to create the
offspring
• BLX and UNDX cannot always maintain
high diversity
• For real world problem, the algorithm that
can maintain higher diversity, good
convergence rate while it shows
4
Non-dominated Sorting Genetic Algorithm (NSGA-II)
• Step 1:
• Each individual is compared with another
randomly selected individual(niche comparison)
• The copy of the winner is placed in the mating
pool
• Step 2:
• Apply crossover rate for each individual in a
mating pool and select a parent
• Step 3:
• All non-dominated fronts of Pt and Qt are copied
to the parent population rank by rank
• Step 4:
• Stop adding the individuals in the rank when the
size of parent population is larger than the
population size
5
Non-dominated Sorting Genetic Algorithm (NSGA-II)
• Crowding distance assignment
• Individuals are sorted in each objective domain
• The first individual and the last individual in the rank are assigned
the crowding distance equal to infinity
• For the other individuals, the crowding distance is calculated by
𝑑𝑖 =
𝑚=1
𝑀
𝑓𝑚
𝐼𝑖+1
𝑚
− 𝑓𝑚
𝐼𝑖−1
𝑚
𝑓𝑚
𝑚𝑎𝑥
− 𝑓𝑚
𝑚𝑖𝑛
𝑖 ∈ [2, 𝑙 − 1]
• Niche comparison
𝑖 < 𝑛 𝑗 𝑖𝑓 𝑖 𝑟𝑎𝑛𝑘 < 𝑗 𝑟𝑎𝑛𝑘
𝑜𝑟 ((𝑖 𝑟𝑎𝑛𝑘 = 𝑗 𝑟𝑎𝑛𝑘) 𝑎𝑛𝑑 (𝑖 𝑑𝑖𝑠𝑡𝑎𝑛𝑐𝑒 > 𝑗 𝑑𝑖𝑠𝑡𝑎𝑛𝑐𝑒))
• Between two solution with differing non-domination ranks the
solution with the better rank is preferred
• If both solutions belong to the same front, the solution which is
located in lesser crowed region is preferred.
6
Objective
• Development efficient crossover operator for real world problem
• Investigation of the proposed crossover operator by solving the
single-objective optimization problem, multi-objective optimization
problem and for the real world problem
7
BLX Operator
• BLX is a crossover operator for real code GA
• BLX created two offspring from two parents by use following equation
𝑥𝑖
(1,𝑡+1)
= 𝛾𝑖 𝑥𝑖
1,𝑡
+ 1 − 𝛾𝑖 𝑥𝑖
(2,𝑡)
𝑥𝑖
(2,𝑡+1)
= 1 − 𝛾𝑖 𝑥𝑖
1,𝑡
+ 𝛾𝑖 𝑥𝑖
(2,𝑡)
where 𝛾𝑖 = 1 + 2α 𝑢𝑖 − 𝛼 and α = 0.5
• This is called BLX-0.5
Center pointOffspring1 Offspring2Parent1 Parent2
Possible crossover region
8
dα d α d
UNDX operator
• UNDX crossover is a crossover operator for real code Genetic
Algorithms
• UNDX is a multi-parent crossover operator
• UNDX is operator based on the normal distribution
9
Algorithm of the UNDX
𝑥 𝑐
𝑥2𝑥1
𝑥3
10
• Select Parent1 (𝑥1), Parent2 (𝑥2), Parent3(𝑥3)• Find the midpoint of Parent1 and Parent2
𝑥 𝑝 =
1
2
(𝑥1 + 𝑥2)
𝑥 𝑝
• Find difference vector of these parents
d = 𝑥2 − 𝑥1
d
• Next, find the distance between the third and forth parent and the line
connecting 𝑥1 and 𝑥2 by use the following equation
D = 𝑥3 − 𝑥1 × 1 −
𝑥3 − 𝑥1
𝑇
𝑥2 − 𝑥1
𝑥3 − 𝑥1 𝑥2 − 𝑥1
D
• The childs𝑥 𝑐 is yielded by the equation
𝑥 𝑐 = 𝑥 𝑝 + 𝜀𝑑 +
𝑖=1
𝑛−1
𝜂𝑖 𝑒𝑖 𝐷
New Multi-parent Crossover Operator
• The disadvantage of the UNDX is very hard to find the
optimal solution close to the boundary and low diversity
• The advantage of the UNDX is good convergence rate
• Maintenance of diversity by the UNDX result is required
• Definition of proper discover area can maintain high diversity
11
Algorithm of the New Multi-parent Crossover
• Find the midpoint of Parent1 and Parent2
𝑥 𝑝1
=
1
2
(𝑥1 + 𝑥2)
• Find the midpoint of Parent3 and Parent4
𝑥 𝑝2
=
1
2
(𝑥3 + 𝑥4)
12
• Find difference vector of these parents
𝑑1 = 𝑥2 − 𝑥1
𝑑2 = 𝑥4 − 𝑥3
• Next, find the distance between the third and forth parent and the
line connecting 𝑥1and 𝑥2by use the following equation
𝐷1 = 𝑥3 − 𝑥1 × 1 −
𝑥3 − 𝑥1
𝑇 𝑥2 − 𝑥1
𝑥3 − 𝑥1 𝑥2 − 𝑥1
𝐷2 = |𝑥4 − 𝑥1| × (1 − (
𝑥4 − 𝑥1
𝑇(𝑥2 − 𝑥1)
𝑥4 − 𝑥1 |𝑥2 − 𝑥1|
))
• Then, find the distance between the first and second parent and the
line connecting 𝑥3and 𝑥4by use the following equation
𝐷3 = 𝑥1 − 𝑥3 × 1 −
𝑥1 − 𝑥3
𝑇 𝑥4 − 𝑥3
𝑥1 − 𝑥3 𝑥4 − 𝑥3
𝐷4 = |𝑥2 − 𝑥3| × (1 − (
𝑥2 − 𝑥3
𝑇(𝑥4 − 𝑥3)
𝑥2 − 𝑥3 |𝑥4 − 𝑥3|
))
• The childs𝑥 𝑐1
and 𝑥 𝑐2
are yielded by the equation
𝑥 𝑐1
= 𝑥 𝑝1
+ 𝜀𝑑 +
𝑖=1
𝑛−1
𝜂𝑖 𝑒𝑖 𝐷
𝑥 𝑐2
= 𝑥 𝑝1
+ 𝜀𝑑 +
𝑖=1
𝑛−1
𝜂𝑖 𝑒𝑖 𝐷
• The childs𝑥 𝑐3
and 𝑥 𝑐4
are yielded by the equation
𝑥 𝑐3
= 𝑥 𝑝2
+ 𝜀𝑑 +
𝑖=1
𝑛−1
𝜂𝑖 𝑒𝑖 𝐷
𝑥 𝑐4
= 𝑥 𝑝2
+ 𝜀𝑑 +
𝑖=1
𝑛−1
𝜂𝑖 𝑒𝑖 𝐷
• Select Parent1 (𝑥1), Parent2 (𝑥2), Parent3 (𝑥3) and Parent4 (𝑥4)
𝑥1
𝑥3
𝑥2
𝑥4
𝑥 𝑝1
𝑥 𝑝2
𝑑1
𝐷2
𝐷1
𝐷3
𝐷4
𝑥 𝑐3
𝑥 𝑐4
𝑥 𝑐2
𝑥 𝑐1
Single-Objective Test Function
• Sphere
𝑓𝑆𝑝ℎ𝑒𝑟𝑒 𝑥 =
𝑖=1
20
𝑥𝑖
2
.
−5.12 ≤ 𝑥𝑖 ≤ 5.12, 1 ≤ 𝑖 ≤ 20
• Exact solution is 0
• Rastrigin
𝑓𝑅𝑎𝑠𝑡𝑟𝑖𝑔𝑖𝑛 𝑥 = 200 +
𝑖=1
20
{𝑥𝑖
2
−10cos(2𝜋𝑥𝑖)}.
−5.12 ≤ 𝑥𝑖 ≤ 5.12, 1 ≤ 𝑖 ≤ 20
• Exact solution is 0
13
Single-Objective Test Function
• Rosenbrock
𝑓𝑅𝑜𝑠𝑒𝑛𝑏𝑟𝑜𝑐𝑘 𝐹𝑢𝑛𝑐𝑡𝑖𝑜𝑛 𝑥
=
𝑖=2
20
{100 𝑥1 − 𝑥𝑖
2 2
+ (𝑥𝑖 − 1)2
}.
−2.048 ≤ 𝑥𝑖 ≤ 2.048, 1 ≤ 𝑖 ≤ 20
• Exact solution is 0
• For Single-Objective optimization use 300
population and 500 generation
14
Multi-objective Test Function
• ZDT1
• convex
Minimize: 𝑓1(x) = 𝑥1
Minimize: 𝑓2 x = 1 −
𝑥1
𝑔 𝑥
g x = 1 + 9
𝑖=2
𝑛
𝑥𝑖
𝑛 − 1
0 ≤ 𝑥𝑖 ≤ 1, 1 ≤ 𝑖 ≤ 30
15
Multi-objective Test Function
• ZDT2
• Non-convex
Minimize: 𝑓1(x) = 𝑥1
Minimize: 𝑓2(x) = 1 −
𝑥1
𝑔(𝑥)
2
g x = 1 + 9
𝑖=2
𝑛
𝑥𝑖
𝑛 − 1
0 ≤ 𝑥𝑖 ≤ 1, 1 ≤ 𝑖 ≤ 30
16
• For Multi-Objective optimization use 300 population and 200 generation
Airfoil Optimization Problem
• Created airfoil my NACA 4 digit equation
• Maximum: Cl
• Minimize: Cd
• Subject to:
• 0.05 ≤ y1 ≤ 0.13; the thickness in upper side of airfoil (dv1)
• 0.05 ≤ y2 ≤ 0.13; the thickness in lower side of airfoil (dv2)
• 0 ≤ y3 ≤ 0.04; the maximum camber of airfoil (dv3)
• 0.2 ≤ y4 ≤ 0.6; the location of the maximum camber (dv4)
• Subject to:
• 0.6 ≤ cl ≤ 1.8 (dv5)
• Use 100 population and 70 generation
• Use Reynolds’ Number 106
17
NACA 4 digit Airfoil
• Equation for a symmetrical 4-digit NACA airfoil
𝑦𝑡 =
𝑡
0.2
𝑐 0.2969
𝑥
𝑐
− 0.1260
𝑥
𝑐
− 0.3516
𝑥
𝑐
2
+ 0.2843
𝑥
𝑐
3
− 0.1015
𝑥
𝑐
4
• Equation for a cambered 4-digit NACA airfoil
• 𝑦𝑐 =
𝑚
𝑥
𝑝2 2𝑝 −
𝑥
𝑐
, 0 ≤ 𝑥 ≤ 𝑝𝑐
𝑚
𝑐−𝑥
(1−𝑝)2 1 +
𝑥
𝑐
− 2𝑝 , 𝑝𝑐 ≤ 𝑥 ≤ 𝑐
• The example of airfoil when use t=0.12, m=0.04 and p=0.4
18
-0.04
-0.02
0
0.02
0.04
0.06
0.08
0.1
0.12
0 0.2 0.4 0.6 0.8 1 1.2
Results-Sphere Problem
• Sphere Problem
• This graph show the best solution in each generation
• The results show proposed method is better solution than BLX and
UNDX
19
Results-Sphere Problem
• This graph show the average solution in each generation
• This graph can explained the proposed method can maintain higher
diversity than BLX and UNDX
20
Results-Sphere
21
• This graph show the design parameter of the best solution in each
generation that can show the convergence rate
-5
-4
-3
-2
-1
0
1
2
3
4
5
6
0 100 200 300 400 500 600
New multi-parent crossover method
x_1 x_2 x_3 x_4 x_5
x_6 x_7 x_8 x_9 x_10
x_11 x_12 x_13 x_14 x_15
x_16 x_17 x_18 x_19 x_20
-5
-4
-3
-2
-1
0
1
2
3
4
5
6
0 100 200 300 400 500 600
UNDX
x_1 x_2 x_3 x_4 x_5
x_6 x_7 x_8 x_9 x_10
x_11 x_12 x_13 x_14 x_15
x_16 x_17 x_18 x_19 x_20
-5
-4
-3
-2
-1
0
1
2
3
4
5
6
0 100 200 300 400 500 600
BLX
x_1 x_2 x_3 x_4 x_5
x_6 x_7 x_8 x_9 x_10
x_11 x_12 x_13 x_14 x_15
x_16 x_17 x_18 x_19 x_20
Results-Sphere
22
• This graph show the close up view (from generation 1 to 200) of
design parameter of the best solution in each generation that can
show the convergence rate
-1
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
0 50 100 150 200
New multi-parent crossover method
x_1 x_2 x_3 x_4 x_5
x_6 x_7 x_8 x_9 x_10
x_11 x_12 x_13 x_14 x_15
x_16 x_17 x_18 x_19 x_20
-1
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
0 50 100 150 200
UNDX
x_1 x_2 x_3 x_4 x_5
x_6 x_7 x_8 x_9 x_10
x_11 x_12 x_13 x_14 x_15
x_16 x_17 x_18 x_19 x_20
-1
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
0 50 100 150 200
BLX
x_1 x_2 x_3 x_4 x_5
x_6 x_7 x_8 x_9 x_10
x_11 x_12 x_13 x_14 x_15
x_16 x_17 x_18 x_19 x_20
Results-Rastrigin Problem
• Rastrigin Problem
• This graph show the best solution in each generation
• The results show proposed method is better solution than BLX and
UNDX
23
Results-Rastrigin Problem
• This graph show the average solution in each generation
• This graph can explained the proposed method can maintain higher
diversity than BLX and UNDX
24
Results-Rastrigin
25
-5
-4
-3
-2
-1
0
1
2
3
4
5
6
0 100 200 300 400 500
New multi-parent crossover method
x_1 x_2 x_3 x_4 x_5
x_6 x_7 x_8 x_9 x_10
x_11 x_12 x_13 x_14 x_15
x_16 x_17 x_18 x_19 x_20
-5
-4
-3
-2
-1
0
1
2
3
4
5
6
0 100 200 300 400 500 600
UNDX
x_1 x_2 x_3 x_4 x_5
x_6 x_7 x_8 x_9 x_10
x_11 x_12 x_13 x_14 x_15
x_16 x_17 x_18 x_19 x_20
-5
-4
-3
-2
-1
0
1
2
3
4
5
6
0 100 200 300 400 500 600
BLX
x_1 x_2 x_3 x_4 x_5
x_6 x_7 x_8 x_9 x_10
x_11 x_12 x_13 x_14 x_15
x_16 x_17 x_18 x_19 x_20
• This graph show the design parameter of the best solution in each
generation that can show the convergence rate
Results-Rastrigin
26
-1
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
0 50 100 150 200
New multi-parent crossover method
x_1 x_2 x_3 x_4 x_5
x_6 x_7 x_8 x_9 x_10
x_11 x_12 x_13 x_14 x_15
x_16 x_17 x_18 x_19 x_20
-1
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
0 50 100 150 200
UNDX
x_1 x_2 x_3 x_4 x_5
x_6 x_7 x_8 x_9 x_10
x_11 x_12 x_13 x_14 x_15
x_16 x_17 x_18 x_19 x_20
-1
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
0 50 100 150 200
BLX
x_1 x_2 x_3 x_4 x_5
x_6 x_7 x_8 x_9 x_10
x_11 x_12 x_13 x_14 x_15
x_16 x_17 x_18 x_19 x_20
• This graph show the close up view (from generation 1 to 200) of
design parameter of the best solution in each generation that can
show the convergence rate
Results-Rosenbrock Problem
• Rosenbrock Problem
• This graph show the best solution in each generation
• The results show proposed method is better solution than BLX and
UNDX
27
Results-Rosenbrock Problem
• This graph show the average solution in each generation
28
Results-Rosenbrock
29
-2
-1.5
-1
-0.5
0
0.5
1
1.5
2
0 100 200 300 400 500 600
New multi-parent crossover method
x_1 x_2 x_3 x_4 x_5
x_6 x_7 x_8 x_9 x_10
x_11 x_12 x_13 x_14 x_15
x_16 x_17 x_18 x_19 x_20
-2
-1.5
-1
-0.5
0
0.5
1
1.5
2
0 100 200 300 400 500 600
UNDX
x_1 x_2 x_3 x_4 x_5
x_6 x_7 x_8 x_9 x_10
x_11 x_12 x_13 x_14 x_15
x_16 x_17 x_18 x_19 x_20
-2
-1.5
-1
-0.5
0
0.5
1
1.5
2
0 100 200 300 400 500 600
BLX
x_1 x_2 x_3 x_4 x_5
x_6 x_7 x_8 x_9 x_10
x_11 x_12 x_13 x_14 x_15
x_16 x_17 x_18 x_19 x_20
• This graph show the design parameter of the best solution in each
generation that can show the convergence rate
Results-Rosenbrock
30
-1
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
0 50 100 150 200
New multi-parent crossover method
x_1 x_2 x_3 x_4 x_5
x_6 x_7 x_8 x_9 x_10
x_11 x_12 x_13 x_14 x_15
x_16 x_17 x_18 x_19 x_20
-1
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
0 50 100 150 200
UNDX
x_1 x_2 x_3 x_4 x_5
x_6 x_7 x_8 x_9 x_10
x_11 x_12 x_13 x_14 x_15
x_16 x_17 x_18 x_19 x_20
-1
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
0 50 100 150 200
BLX
x_1 x_2 x_3 x_4 x_5
x_6 x_7 x_8 x_9 x_10
x_11 x_12 x_13 x_14 x_15
x_16 x_17 x_18 x_19 x_20
• This graph show the close up view (from generation 1 to 200) of
design parameter of the best solution in each generation that can
show the convergence rate
Results-ZDT1
• This graph show the Pareto solution of ZDT1 problem
• This Pareto solution show the proposed method can get better
solution than BLX and UNDX
31
Results-ZDT1
• Metrices of ZDT1
• These graphs show the proposed method have good convergence
rate for ZDT1 problem
32
Hypervolume Maximum Spread Spacing
Results-ZDT2
• This graph show the Pareto solution of ZDT2 problem
• This Pareto solution show the proposed method can get better
solution than BLX and UNDX
33
Results-ZDT2
• Metrices of ZDT2
• These graphs show the proposed method have good convergence
rate for ZDT1 problem
34
Hypervolume Maximum Spread Spacing
Results-Airfoil Optimization Problem
• This graph show the Pareto solution of airfoil optimization problem
• These Pareto solution shows the proposed method can find the
optimal solution similar to the BLX algorithm and better than UNDX
35
30th generation 70th generation
Results-Airfoil Optimization Problem
• Metrices of Airfoil Optimization Problem
36
Hypervolume Maximum Spread Spacing
Results-Airfoil Optimization Problem
37
• PCP plot of the Pareto solution at 70th generation
• The plot shown the propose method have the solution similar to BLX
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
2
dv1 dv2 dv3 dv4 dv5 Cd Cl
BLX
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
dv1 dv2 dv3 dv4 dv5 Cd Cl
UNDX
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
2
dv1 dv2 dv3 dv4 dv5 cd cl
Proposed method
Results-Airfoil Optimization Problem
• Left figure show the airfoil comparison and the right figure show the
Cp(Pressure Distribution) comparison at design point(Cl=0.8 at 70th
generation)
• The plot shown the propose method have the solution similar to BLX
38
Conclusion
• Successful to development of new multi-parent crossover
method
• The proposed method successfully find out the better
solution of every test function compared with existent
crossover method
• The multi-objective test function are better compared with
other crossover method
• For the airfoil optimization problem the searching areas are
similar because the parameterizations is relatively smile so
the investigation by more complex problem is need
• This method can maintain higher diversity
39

Más contenido relacionado

La actualidad más candente

Determination of Optimal Product Mix for Profit Maximization using Linear Pro...
Determination of Optimal Product Mix for Profit Maximization using Linear Pro...Determination of Optimal Product Mix for Profit Maximization using Linear Pro...
Determination of Optimal Product Mix for Profit Maximization using Linear Pro...IJERA Editor
 
Options Portfolio Selection
Options Portfolio SelectionOptions Portfolio Selection
Options Portfolio Selectionguasoni
 
Low Complexity Regularization of Inverse Problems - Course #3 Proximal Splitt...
Low Complexity Regularization of Inverse Problems - Course #3 Proximal Splitt...Low Complexity Regularization of Inverse Problems - Course #3 Proximal Splitt...
Low Complexity Regularization of Inverse Problems - Course #3 Proximal Splitt...Gabriel Peyré
 
Integer Programming, Goal Programming, and Nonlinear Programming
Integer Programming, Goal Programming, and Nonlinear ProgrammingInteger Programming, Goal Programming, and Nonlinear Programming
Integer Programming, Goal Programming, and Nonlinear ProgrammingSalah A. Skaik - MBA-PMP®
 
Problem Understanding through Landscape Theory
Problem Understanding through Landscape TheoryProblem Understanding through Landscape Theory
Problem Understanding through Landscape Theoryjfrchicanog
 
2012 mdsp pr12 k means mixture of gaussian
2012 mdsp pr12 k means mixture of gaussian2012 mdsp pr12 k means mixture of gaussian
2012 mdsp pr12 k means mixture of gaussiannozomuhamada
 
31350052 introductory-mathematical-analysis-textbook-solution-manual
31350052 introductory-mathematical-analysis-textbook-solution-manual31350052 introductory-mathematical-analysis-textbook-solution-manual
31350052 introductory-mathematical-analysis-textbook-solution-manualMahrukh Khalid
 
Model Selection with Piecewise Regular Gauges
Model Selection with Piecewise Regular GaugesModel Selection with Piecewise Regular Gauges
Model Selection with Piecewise Regular GaugesGabriel Peyré
 
07 cie552 image_mosaicing
07 cie552 image_mosaicing07 cie552 image_mosaicing
07 cie552 image_mosaicingElsayed Hemayed
 
beyond linear programming: mathematical programming extensions
beyond linear programming: mathematical programming extensionsbeyond linear programming: mathematical programming extensions
beyond linear programming: mathematical programming extensionsAngelica Angelo Ocon
 
Animashree Anandkumar, Electrical Engineering and CS Dept, UC Irvine at MLcon...
Animashree Anandkumar, Electrical Engineering and CS Dept, UC Irvine at MLcon...Animashree Anandkumar, Electrical Engineering and CS Dept, UC Irvine at MLcon...
Animashree Anandkumar, Electrical Engineering and CS Dept, UC Irvine at MLcon...MLconf
 
Ch 07 MATLAB Applications in Chemical Engineering_陳奇中教授教學投影片
Ch 07 MATLAB Applications in Chemical Engineering_陳奇中教授教學投影片Ch 07 MATLAB Applications in Chemical Engineering_陳奇中教授教學投影片
Ch 07 MATLAB Applications in Chemical Engineering_陳奇中教授教學投影片Chyi-Tsong Chen
 
Global optimization
Global optimizationGlobal optimization
Global optimizationbpenalver
 
MLHEP 2015: Introductory Lecture #3
MLHEP 2015: Introductory Lecture #3MLHEP 2015: Introductory Lecture #3
MLHEP 2015: Introductory Lecture #3arogozhnikov
 

La actualidad más candente (20)

Integer programming
Integer programmingInteger programming
Integer programming
 
Respose surface methods
Respose surface methodsRespose surface methods
Respose surface methods
 
QMC Opening Workshop, High Accuracy Algorithms for Interpolating and Integrat...
QMC Opening Workshop, High Accuracy Algorithms for Interpolating and Integrat...QMC Opening Workshop, High Accuracy Algorithms for Interpolating and Integrat...
QMC Opening Workshop, High Accuracy Algorithms for Interpolating and Integrat...
 
K-means and GMM
K-means and GMMK-means and GMM
K-means and GMM
 
Determination of Optimal Product Mix for Profit Maximization using Linear Pro...
Determination of Optimal Product Mix for Profit Maximization using Linear Pro...Determination of Optimal Product Mix for Profit Maximization using Linear Pro...
Determination of Optimal Product Mix for Profit Maximization using Linear Pro...
 
Options Portfolio Selection
Options Portfolio SelectionOptions Portfolio Selection
Options Portfolio Selection
 
Low Complexity Regularization of Inverse Problems - Course #3 Proximal Splitt...
Low Complexity Regularization of Inverse Problems - Course #3 Proximal Splitt...Low Complexity Regularization of Inverse Problems - Course #3 Proximal Splitt...
Low Complexity Regularization of Inverse Problems - Course #3 Proximal Splitt...
 
Integer Programming, Goal Programming, and Nonlinear Programming
Integer Programming, Goal Programming, and Nonlinear ProgrammingInteger Programming, Goal Programming, and Nonlinear Programming
Integer Programming, Goal Programming, and Nonlinear Programming
 
Distributed ADMM
Distributed ADMMDistributed ADMM
Distributed ADMM
 
Problem Understanding through Landscape Theory
Problem Understanding through Landscape TheoryProblem Understanding through Landscape Theory
Problem Understanding through Landscape Theory
 
"Let us talk about output features! by Florence d’Alché-Buc, LTCI & Full Prof...
"Let us talk about output features! by Florence d’Alché-Buc, LTCI & Full Prof..."Let us talk about output features! by Florence d’Alché-Buc, LTCI & Full Prof...
"Let us talk about output features! by Florence d’Alché-Buc, LTCI & Full Prof...
 
2012 mdsp pr12 k means mixture of gaussian
2012 mdsp pr12 k means mixture of gaussian2012 mdsp pr12 k means mixture of gaussian
2012 mdsp pr12 k means mixture of gaussian
 
31350052 introductory-mathematical-analysis-textbook-solution-manual
31350052 introductory-mathematical-analysis-textbook-solution-manual31350052 introductory-mathematical-analysis-textbook-solution-manual
31350052 introductory-mathematical-analysis-textbook-solution-manual
 
Model Selection with Piecewise Regular Gauges
Model Selection with Piecewise Regular GaugesModel Selection with Piecewise Regular Gauges
Model Selection with Piecewise Regular Gauges
 
07 cie552 image_mosaicing
07 cie552 image_mosaicing07 cie552 image_mosaicing
07 cie552 image_mosaicing
 
beyond linear programming: mathematical programming extensions
beyond linear programming: mathematical programming extensionsbeyond linear programming: mathematical programming extensions
beyond linear programming: mathematical programming extensions
 
Animashree Anandkumar, Electrical Engineering and CS Dept, UC Irvine at MLcon...
Animashree Anandkumar, Electrical Engineering and CS Dept, UC Irvine at MLcon...Animashree Anandkumar, Electrical Engineering and CS Dept, UC Irvine at MLcon...
Animashree Anandkumar, Electrical Engineering and CS Dept, UC Irvine at MLcon...
 
Ch 07 MATLAB Applications in Chemical Engineering_陳奇中教授教學投影片
Ch 07 MATLAB Applications in Chemical Engineering_陳奇中教授教學投影片Ch 07 MATLAB Applications in Chemical Engineering_陳奇中教授教學投影片
Ch 07 MATLAB Applications in Chemical Engineering_陳奇中教授教學投影片
 
Global optimization
Global optimizationGlobal optimization
Global optimization
 
MLHEP 2015: Introductory Lecture #3
MLHEP 2015: Introductory Lecture #3MLHEP 2015: Introductory Lecture #3
MLHEP 2015: Introductory Lecture #3
 

Similar a Tokyo conference

Mathematical Understanding in Traffic Flow Modelling
Mathematical Understanding in Traffic Flow ModellingMathematical Understanding in Traffic Flow Modelling
Mathematical Understanding in Traffic Flow ModellingNikhil Chandra Sarkar
 
Paper study: Attention, learn to solve routing problems!
Paper study: Attention, learn to solve routing problems!Paper study: Attention, learn to solve routing problems!
Paper study: Attention, learn to solve routing problems!ChenYiHuang5
 
Paper Study: Melding the data decision pipeline
Paper Study: Melding the data decision pipelinePaper Study: Melding the data decision pipeline
Paper Study: Melding the data decision pipelineChenYiHuang5
 
Paper study: Learning to solve circuit sat
Paper study: Learning to solve circuit satPaper study: Learning to solve circuit sat
Paper study: Learning to solve circuit satChenYiHuang5
 
Optimization Techniques.pdf
Optimization Techniques.pdfOptimization Techniques.pdf
Optimization Techniques.pdfanandsimple
 
Multi-Domain Diversity Preservation to Mitigate Particle Stagnation and Enab...
Multi-Domain Diversity Preservation to Mitigate Particle Stagnation and Enab...Multi-Domain Diversity Preservation to Mitigate Particle Stagnation and Enab...
Multi-Domain Diversity Preservation to Mitigate Particle Stagnation and Enab...Weiyang Tong
 
Efficient anomaly detection via matrix sketching
Efficient anomaly detection via matrix sketchingEfficient anomaly detection via matrix sketching
Efficient anomaly detection via matrix sketchingHsing-chuan Hsieh
 
Elements of Statistical Learning 読み会 第2章
Elements of Statistical Learning 読み会 第2章Elements of Statistical Learning 読み会 第2章
Elements of Statistical Learning 読み会 第2章Tsuyoshi Sakama
 
Paper Study: OptNet: Differentiable Optimization as a Layer in Neural Networks
Paper Study: OptNet: Differentiable Optimization as a Layer in Neural NetworksPaper Study: OptNet: Differentiable Optimization as a Layer in Neural Networks
Paper Study: OptNet: Differentiable Optimization as a Layer in Neural NetworksChenYiHuang5
 
Genetic Algorithm
Genetic AlgorithmGenetic Algorithm
Genetic AlgorithmSHIMI S L
 
Unit-1 Basic Concept of Algorithm.pptx
Unit-1 Basic Concept of Algorithm.pptxUnit-1 Basic Concept of Algorithm.pptx
Unit-1 Basic Concept of Algorithm.pptxssuser01e301
 
On clusteredsteinertree slide-ver 1.1
On clusteredsteinertree slide-ver 1.1On clusteredsteinertree slide-ver 1.1
On clusteredsteinertree slide-ver 1.1VitAnhNguyn94
 
MOMDPSO_IDETC_2014_Weiyang
MOMDPSO_IDETC_2014_WeiyangMOMDPSO_IDETC_2014_Weiyang
MOMDPSO_IDETC_2014_WeiyangMDO_Lab
 
Computational Intelligence Assisted Engineering Design Optimization (using MA...
Computational Intelligence Assisted Engineering Design Optimization (using MA...Computational Intelligence Assisted Engineering Design Optimization (using MA...
Computational Intelligence Assisted Engineering Design Optimization (using MA...AmirParnianifard1
 
Clipping & Rasterization
Clipping & RasterizationClipping & Rasterization
Clipping & RasterizationAhmed Daoud
 
A compact zero knowledge proof to restrict message space in homomorphic encry...
A compact zero knowledge proof to restrict message space in homomorphic encry...A compact zero knowledge proof to restrict message space in homomorphic encry...
A compact zero knowledge proof to restrict message space in homomorphic encry...MITSUNARI Shigeo
 
Lecture slides week14-15
Lecture slides week14-15Lecture slides week14-15
Lecture slides week14-15Shani729
 

Similar a Tokyo conference (20)

Mathematical Understanding in Traffic Flow Modelling
Mathematical Understanding in Traffic Flow ModellingMathematical Understanding in Traffic Flow Modelling
Mathematical Understanding in Traffic Flow Modelling
 
Paper study: Attention, learn to solve routing problems!
Paper study: Attention, learn to solve routing problems!Paper study: Attention, learn to solve routing problems!
Paper study: Attention, learn to solve routing problems!
 
Paper Study: Melding the data decision pipeline
Paper Study: Melding the data decision pipelinePaper Study: Melding the data decision pipeline
Paper Study: Melding the data decision pipeline
 
Paper study: Learning to solve circuit sat
Paper study: Learning to solve circuit satPaper study: Learning to solve circuit sat
Paper study: Learning to solve circuit sat
 
Optimization Techniques.pdf
Optimization Techniques.pdfOptimization Techniques.pdf
Optimization Techniques.pdf
 
13 weightedresidual
13 weightedresidual13 weightedresidual
13 weightedresidual
 
Genetic algorithm
Genetic algorithmGenetic algorithm
Genetic algorithm
 
Multi-Domain Diversity Preservation to Mitigate Particle Stagnation and Enab...
Multi-Domain Diversity Preservation to Mitigate Particle Stagnation and Enab...Multi-Domain Diversity Preservation to Mitigate Particle Stagnation and Enab...
Multi-Domain Diversity Preservation to Mitigate Particle Stagnation and Enab...
 
Efficient anomaly detection via matrix sketching
Efficient anomaly detection via matrix sketchingEfficient anomaly detection via matrix sketching
Efficient anomaly detection via matrix sketching
 
Elements of Statistical Learning 読み会 第2章
Elements of Statistical Learning 読み会 第2章Elements of Statistical Learning 読み会 第2章
Elements of Statistical Learning 読み会 第2章
 
Paper Study: OptNet: Differentiable Optimization as a Layer in Neural Networks
Paper Study: OptNet: Differentiable Optimization as a Layer in Neural NetworksPaper Study: OptNet: Differentiable Optimization as a Layer in Neural Networks
Paper Study: OptNet: Differentiable Optimization as a Layer in Neural Networks
 
Genetic Algorithm
Genetic AlgorithmGenetic Algorithm
Genetic Algorithm
 
Unit-1 Basic Concept of Algorithm.pptx
Unit-1 Basic Concept of Algorithm.pptxUnit-1 Basic Concept of Algorithm.pptx
Unit-1 Basic Concept of Algorithm.pptx
 
On clusteredsteinertree slide-ver 1.1
On clusteredsteinertree slide-ver 1.1On clusteredsteinertree slide-ver 1.1
On clusteredsteinertree slide-ver 1.1
 
Extended Isolation Forest
Extended Isolation Forest Extended Isolation Forest
Extended Isolation Forest
 
MOMDPSO_IDETC_2014_Weiyang
MOMDPSO_IDETC_2014_WeiyangMOMDPSO_IDETC_2014_Weiyang
MOMDPSO_IDETC_2014_Weiyang
 
Computational Intelligence Assisted Engineering Design Optimization (using MA...
Computational Intelligence Assisted Engineering Design Optimization (using MA...Computational Intelligence Assisted Engineering Design Optimization (using MA...
Computational Intelligence Assisted Engineering Design Optimization (using MA...
 
Clipping & Rasterization
Clipping & RasterizationClipping & Rasterization
Clipping & Rasterization
 
A compact zero knowledge proof to restrict message space in homomorphic encry...
A compact zero knowledge proof to restrict message space in homomorphic encry...A compact zero knowledge proof to restrict message space in homomorphic encry...
A compact zero knowledge proof to restrict message space in homomorphic encry...
 
Lecture slides week14-15
Lecture slides week14-15Lecture slides week14-15
Lecture slides week14-15
 

Tokyo conference

  • 1. New Multi-parent Crossover based on Crossing Two Segments Bounded by Selected Parents Atthaphon Ariyarit, Kanazaki Masahiro Department of Aerospace Engineering, Graduate School of System Design, Tokyo Metropolitan University 第6回進化計算学会研究会 2014/03/07
  • 2. Outline • Introduction • Objective • Overview of Popular Crossover Operators • Blended Crossover (BLX) • Unimodal Normal Distribution Crossover (UNDX) • New Multi-parent Crossover Method • Test Problems • Single-objective optimization problems • Multi-objective optimization problems • Multi-objective airfoil optimization problem • Results • Conclusions 2
  • 3. Introduction • The Genetic Algorithms(GA) is popular optimization method to solve single-objective and multi-objective optimization problem • GA have three main operator, such as, Selection, Crossover and mutation • Crossover operator is the main operator for GA performance • The popular Crossover operator is BLX and UNDX Algorithm[1] • The GA is popular in aerospace engineering, such as, to increase the efficiency of aircraft, to used for the navigation of aircraft, or to reduce the weight of the aircraft weight, etc. [1] Hajime K, Isao O and Shigenobu K. Theoretical Analysis of the Unimodal Normal Distribution Crossover for Real-coded Genetic Algorithms. Transactions of the Society of Instrument & Control Engineers, 2002, 2(1), 187-194 3
  • 4. Introduction • Procedure of Genetic Algorithms • Genetic operator such as selection, crossover and mutation are applied to the parent to create the offspring • BLX and UNDX cannot always maintain high diversity • For real world problem, the algorithm that can maintain higher diversity, good convergence rate while it shows 4
  • 5. Non-dominated Sorting Genetic Algorithm (NSGA-II) • Step 1: • Each individual is compared with another randomly selected individual(niche comparison) • The copy of the winner is placed in the mating pool • Step 2: • Apply crossover rate for each individual in a mating pool and select a parent • Step 3: • All non-dominated fronts of Pt and Qt are copied to the parent population rank by rank • Step 4: • Stop adding the individuals in the rank when the size of parent population is larger than the population size 5
  • 6. Non-dominated Sorting Genetic Algorithm (NSGA-II) • Crowding distance assignment • Individuals are sorted in each objective domain • The first individual and the last individual in the rank are assigned the crowding distance equal to infinity • For the other individuals, the crowding distance is calculated by 𝑑𝑖 = 𝑚=1 𝑀 𝑓𝑚 𝐼𝑖+1 𝑚 − 𝑓𝑚 𝐼𝑖−1 𝑚 𝑓𝑚 𝑚𝑎𝑥 − 𝑓𝑚 𝑚𝑖𝑛 𝑖 ∈ [2, 𝑙 − 1] • Niche comparison 𝑖 < 𝑛 𝑗 𝑖𝑓 𝑖 𝑟𝑎𝑛𝑘 < 𝑗 𝑟𝑎𝑛𝑘 𝑜𝑟 ((𝑖 𝑟𝑎𝑛𝑘 = 𝑗 𝑟𝑎𝑛𝑘) 𝑎𝑛𝑑 (𝑖 𝑑𝑖𝑠𝑡𝑎𝑛𝑐𝑒 > 𝑗 𝑑𝑖𝑠𝑡𝑎𝑛𝑐𝑒)) • Between two solution with differing non-domination ranks the solution with the better rank is preferred • If both solutions belong to the same front, the solution which is located in lesser crowed region is preferred. 6
  • 7. Objective • Development efficient crossover operator for real world problem • Investigation of the proposed crossover operator by solving the single-objective optimization problem, multi-objective optimization problem and for the real world problem 7
  • 8. BLX Operator • BLX is a crossover operator for real code GA • BLX created two offspring from two parents by use following equation 𝑥𝑖 (1,𝑡+1) = 𝛾𝑖 𝑥𝑖 1,𝑡 + 1 − 𝛾𝑖 𝑥𝑖 (2,𝑡) 𝑥𝑖 (2,𝑡+1) = 1 − 𝛾𝑖 𝑥𝑖 1,𝑡 + 𝛾𝑖 𝑥𝑖 (2,𝑡) where 𝛾𝑖 = 1 + 2α 𝑢𝑖 − 𝛼 and α = 0.5 • This is called BLX-0.5 Center pointOffspring1 Offspring2Parent1 Parent2 Possible crossover region 8 dα d α d
  • 9. UNDX operator • UNDX crossover is a crossover operator for real code Genetic Algorithms • UNDX is a multi-parent crossover operator • UNDX is operator based on the normal distribution 9
  • 10. Algorithm of the UNDX 𝑥 𝑐 𝑥2𝑥1 𝑥3 10 • Select Parent1 (𝑥1), Parent2 (𝑥2), Parent3(𝑥3)• Find the midpoint of Parent1 and Parent2 𝑥 𝑝 = 1 2 (𝑥1 + 𝑥2) 𝑥 𝑝 • Find difference vector of these parents d = 𝑥2 − 𝑥1 d • Next, find the distance between the third and forth parent and the line connecting 𝑥1 and 𝑥2 by use the following equation D = 𝑥3 − 𝑥1 × 1 − 𝑥3 − 𝑥1 𝑇 𝑥2 − 𝑥1 𝑥3 − 𝑥1 𝑥2 − 𝑥1 D • The childs𝑥 𝑐 is yielded by the equation 𝑥 𝑐 = 𝑥 𝑝 + 𝜀𝑑 + 𝑖=1 𝑛−1 𝜂𝑖 𝑒𝑖 𝐷
  • 11. New Multi-parent Crossover Operator • The disadvantage of the UNDX is very hard to find the optimal solution close to the boundary and low diversity • The advantage of the UNDX is good convergence rate • Maintenance of diversity by the UNDX result is required • Definition of proper discover area can maintain high diversity 11
  • 12. Algorithm of the New Multi-parent Crossover • Find the midpoint of Parent1 and Parent2 𝑥 𝑝1 = 1 2 (𝑥1 + 𝑥2) • Find the midpoint of Parent3 and Parent4 𝑥 𝑝2 = 1 2 (𝑥3 + 𝑥4) 12 • Find difference vector of these parents 𝑑1 = 𝑥2 − 𝑥1 𝑑2 = 𝑥4 − 𝑥3 • Next, find the distance between the third and forth parent and the line connecting 𝑥1and 𝑥2by use the following equation 𝐷1 = 𝑥3 − 𝑥1 × 1 − 𝑥3 − 𝑥1 𝑇 𝑥2 − 𝑥1 𝑥3 − 𝑥1 𝑥2 − 𝑥1 𝐷2 = |𝑥4 − 𝑥1| × (1 − ( 𝑥4 − 𝑥1 𝑇(𝑥2 − 𝑥1) 𝑥4 − 𝑥1 |𝑥2 − 𝑥1| )) • Then, find the distance between the first and second parent and the line connecting 𝑥3and 𝑥4by use the following equation 𝐷3 = 𝑥1 − 𝑥3 × 1 − 𝑥1 − 𝑥3 𝑇 𝑥4 − 𝑥3 𝑥1 − 𝑥3 𝑥4 − 𝑥3 𝐷4 = |𝑥2 − 𝑥3| × (1 − ( 𝑥2 − 𝑥3 𝑇(𝑥4 − 𝑥3) 𝑥2 − 𝑥3 |𝑥4 − 𝑥3| )) • The childs𝑥 𝑐1 and 𝑥 𝑐2 are yielded by the equation 𝑥 𝑐1 = 𝑥 𝑝1 + 𝜀𝑑 + 𝑖=1 𝑛−1 𝜂𝑖 𝑒𝑖 𝐷 𝑥 𝑐2 = 𝑥 𝑝1 + 𝜀𝑑 + 𝑖=1 𝑛−1 𝜂𝑖 𝑒𝑖 𝐷 • The childs𝑥 𝑐3 and 𝑥 𝑐4 are yielded by the equation 𝑥 𝑐3 = 𝑥 𝑝2 + 𝜀𝑑 + 𝑖=1 𝑛−1 𝜂𝑖 𝑒𝑖 𝐷 𝑥 𝑐4 = 𝑥 𝑝2 + 𝜀𝑑 + 𝑖=1 𝑛−1 𝜂𝑖 𝑒𝑖 𝐷 • Select Parent1 (𝑥1), Parent2 (𝑥2), Parent3 (𝑥3) and Parent4 (𝑥4) 𝑥1 𝑥3 𝑥2 𝑥4 𝑥 𝑝1 𝑥 𝑝2 𝑑1 𝐷2 𝐷1 𝐷3 𝐷4 𝑥 𝑐3 𝑥 𝑐4 𝑥 𝑐2 𝑥 𝑐1
  • 13. Single-Objective Test Function • Sphere 𝑓𝑆𝑝ℎ𝑒𝑟𝑒 𝑥 = 𝑖=1 20 𝑥𝑖 2 . −5.12 ≤ 𝑥𝑖 ≤ 5.12, 1 ≤ 𝑖 ≤ 20 • Exact solution is 0 • Rastrigin 𝑓𝑅𝑎𝑠𝑡𝑟𝑖𝑔𝑖𝑛 𝑥 = 200 + 𝑖=1 20 {𝑥𝑖 2 −10cos(2𝜋𝑥𝑖)}. −5.12 ≤ 𝑥𝑖 ≤ 5.12, 1 ≤ 𝑖 ≤ 20 • Exact solution is 0 13
  • 14. Single-Objective Test Function • Rosenbrock 𝑓𝑅𝑜𝑠𝑒𝑛𝑏𝑟𝑜𝑐𝑘 𝐹𝑢𝑛𝑐𝑡𝑖𝑜𝑛 𝑥 = 𝑖=2 20 {100 𝑥1 − 𝑥𝑖 2 2 + (𝑥𝑖 − 1)2 }. −2.048 ≤ 𝑥𝑖 ≤ 2.048, 1 ≤ 𝑖 ≤ 20 • Exact solution is 0 • For Single-Objective optimization use 300 population and 500 generation 14
  • 15. Multi-objective Test Function • ZDT1 • convex Minimize: 𝑓1(x) = 𝑥1 Minimize: 𝑓2 x = 1 − 𝑥1 𝑔 𝑥 g x = 1 + 9 𝑖=2 𝑛 𝑥𝑖 𝑛 − 1 0 ≤ 𝑥𝑖 ≤ 1, 1 ≤ 𝑖 ≤ 30 15
  • 16. Multi-objective Test Function • ZDT2 • Non-convex Minimize: 𝑓1(x) = 𝑥1 Minimize: 𝑓2(x) = 1 − 𝑥1 𝑔(𝑥) 2 g x = 1 + 9 𝑖=2 𝑛 𝑥𝑖 𝑛 − 1 0 ≤ 𝑥𝑖 ≤ 1, 1 ≤ 𝑖 ≤ 30 16 • For Multi-Objective optimization use 300 population and 200 generation
  • 17. Airfoil Optimization Problem • Created airfoil my NACA 4 digit equation • Maximum: Cl • Minimize: Cd • Subject to: • 0.05 ≤ y1 ≤ 0.13; the thickness in upper side of airfoil (dv1) • 0.05 ≤ y2 ≤ 0.13; the thickness in lower side of airfoil (dv2) • 0 ≤ y3 ≤ 0.04; the maximum camber of airfoil (dv3) • 0.2 ≤ y4 ≤ 0.6; the location of the maximum camber (dv4) • Subject to: • 0.6 ≤ cl ≤ 1.8 (dv5) • Use 100 population and 70 generation • Use Reynolds’ Number 106 17
  • 18. NACA 4 digit Airfoil • Equation for a symmetrical 4-digit NACA airfoil 𝑦𝑡 = 𝑡 0.2 𝑐 0.2969 𝑥 𝑐 − 0.1260 𝑥 𝑐 − 0.3516 𝑥 𝑐 2 + 0.2843 𝑥 𝑐 3 − 0.1015 𝑥 𝑐 4 • Equation for a cambered 4-digit NACA airfoil • 𝑦𝑐 = 𝑚 𝑥 𝑝2 2𝑝 − 𝑥 𝑐 , 0 ≤ 𝑥 ≤ 𝑝𝑐 𝑚 𝑐−𝑥 (1−𝑝)2 1 + 𝑥 𝑐 − 2𝑝 , 𝑝𝑐 ≤ 𝑥 ≤ 𝑐 • The example of airfoil when use t=0.12, m=0.04 and p=0.4 18 -0.04 -0.02 0 0.02 0.04 0.06 0.08 0.1 0.12 0 0.2 0.4 0.6 0.8 1 1.2
  • 19. Results-Sphere Problem • Sphere Problem • This graph show the best solution in each generation • The results show proposed method is better solution than BLX and UNDX 19
  • 20. Results-Sphere Problem • This graph show the average solution in each generation • This graph can explained the proposed method can maintain higher diversity than BLX and UNDX 20
  • 21. Results-Sphere 21 • This graph show the design parameter of the best solution in each generation that can show the convergence rate -5 -4 -3 -2 -1 0 1 2 3 4 5 6 0 100 200 300 400 500 600 New multi-parent crossover method x_1 x_2 x_3 x_4 x_5 x_6 x_7 x_8 x_9 x_10 x_11 x_12 x_13 x_14 x_15 x_16 x_17 x_18 x_19 x_20 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 0 100 200 300 400 500 600 UNDX x_1 x_2 x_3 x_4 x_5 x_6 x_7 x_8 x_9 x_10 x_11 x_12 x_13 x_14 x_15 x_16 x_17 x_18 x_19 x_20 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 0 100 200 300 400 500 600 BLX x_1 x_2 x_3 x_4 x_5 x_6 x_7 x_8 x_9 x_10 x_11 x_12 x_13 x_14 x_15 x_16 x_17 x_18 x_19 x_20
  • 22. Results-Sphere 22 • This graph show the close up view (from generation 1 to 200) of design parameter of the best solution in each generation that can show the convergence rate -1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1 0 50 100 150 200 New multi-parent crossover method x_1 x_2 x_3 x_4 x_5 x_6 x_7 x_8 x_9 x_10 x_11 x_12 x_13 x_14 x_15 x_16 x_17 x_18 x_19 x_20 -1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1 0 50 100 150 200 UNDX x_1 x_2 x_3 x_4 x_5 x_6 x_7 x_8 x_9 x_10 x_11 x_12 x_13 x_14 x_15 x_16 x_17 x_18 x_19 x_20 -1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1 0 50 100 150 200 BLX x_1 x_2 x_3 x_4 x_5 x_6 x_7 x_8 x_9 x_10 x_11 x_12 x_13 x_14 x_15 x_16 x_17 x_18 x_19 x_20
  • 23. Results-Rastrigin Problem • Rastrigin Problem • This graph show the best solution in each generation • The results show proposed method is better solution than BLX and UNDX 23
  • 24. Results-Rastrigin Problem • This graph show the average solution in each generation • This graph can explained the proposed method can maintain higher diversity than BLX and UNDX 24
  • 25. Results-Rastrigin 25 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 0 100 200 300 400 500 New multi-parent crossover method x_1 x_2 x_3 x_4 x_5 x_6 x_7 x_8 x_9 x_10 x_11 x_12 x_13 x_14 x_15 x_16 x_17 x_18 x_19 x_20 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 0 100 200 300 400 500 600 UNDX x_1 x_2 x_3 x_4 x_5 x_6 x_7 x_8 x_9 x_10 x_11 x_12 x_13 x_14 x_15 x_16 x_17 x_18 x_19 x_20 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 0 100 200 300 400 500 600 BLX x_1 x_2 x_3 x_4 x_5 x_6 x_7 x_8 x_9 x_10 x_11 x_12 x_13 x_14 x_15 x_16 x_17 x_18 x_19 x_20 • This graph show the design parameter of the best solution in each generation that can show the convergence rate
  • 26. Results-Rastrigin 26 -1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1 0 50 100 150 200 New multi-parent crossover method x_1 x_2 x_3 x_4 x_5 x_6 x_7 x_8 x_9 x_10 x_11 x_12 x_13 x_14 x_15 x_16 x_17 x_18 x_19 x_20 -1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1 0 50 100 150 200 UNDX x_1 x_2 x_3 x_4 x_5 x_6 x_7 x_8 x_9 x_10 x_11 x_12 x_13 x_14 x_15 x_16 x_17 x_18 x_19 x_20 -1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1 0 50 100 150 200 BLX x_1 x_2 x_3 x_4 x_5 x_6 x_7 x_8 x_9 x_10 x_11 x_12 x_13 x_14 x_15 x_16 x_17 x_18 x_19 x_20 • This graph show the close up view (from generation 1 to 200) of design parameter of the best solution in each generation that can show the convergence rate
  • 27. Results-Rosenbrock Problem • Rosenbrock Problem • This graph show the best solution in each generation • The results show proposed method is better solution than BLX and UNDX 27
  • 28. Results-Rosenbrock Problem • This graph show the average solution in each generation 28
  • 29. Results-Rosenbrock 29 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2 0 100 200 300 400 500 600 New multi-parent crossover method x_1 x_2 x_3 x_4 x_5 x_6 x_7 x_8 x_9 x_10 x_11 x_12 x_13 x_14 x_15 x_16 x_17 x_18 x_19 x_20 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2 0 100 200 300 400 500 600 UNDX x_1 x_2 x_3 x_4 x_5 x_6 x_7 x_8 x_9 x_10 x_11 x_12 x_13 x_14 x_15 x_16 x_17 x_18 x_19 x_20 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2 0 100 200 300 400 500 600 BLX x_1 x_2 x_3 x_4 x_5 x_6 x_7 x_8 x_9 x_10 x_11 x_12 x_13 x_14 x_15 x_16 x_17 x_18 x_19 x_20 • This graph show the design parameter of the best solution in each generation that can show the convergence rate
  • 30. Results-Rosenbrock 30 -1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1 0 50 100 150 200 New multi-parent crossover method x_1 x_2 x_3 x_4 x_5 x_6 x_7 x_8 x_9 x_10 x_11 x_12 x_13 x_14 x_15 x_16 x_17 x_18 x_19 x_20 -1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1 0 50 100 150 200 UNDX x_1 x_2 x_3 x_4 x_5 x_6 x_7 x_8 x_9 x_10 x_11 x_12 x_13 x_14 x_15 x_16 x_17 x_18 x_19 x_20 -1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1 0 50 100 150 200 BLX x_1 x_2 x_3 x_4 x_5 x_6 x_7 x_8 x_9 x_10 x_11 x_12 x_13 x_14 x_15 x_16 x_17 x_18 x_19 x_20 • This graph show the close up view (from generation 1 to 200) of design parameter of the best solution in each generation that can show the convergence rate
  • 31. Results-ZDT1 • This graph show the Pareto solution of ZDT1 problem • This Pareto solution show the proposed method can get better solution than BLX and UNDX 31
  • 32. Results-ZDT1 • Metrices of ZDT1 • These graphs show the proposed method have good convergence rate for ZDT1 problem 32 Hypervolume Maximum Spread Spacing
  • 33. Results-ZDT2 • This graph show the Pareto solution of ZDT2 problem • This Pareto solution show the proposed method can get better solution than BLX and UNDX 33
  • 34. Results-ZDT2 • Metrices of ZDT2 • These graphs show the proposed method have good convergence rate for ZDT1 problem 34 Hypervolume Maximum Spread Spacing
  • 35. Results-Airfoil Optimization Problem • This graph show the Pareto solution of airfoil optimization problem • These Pareto solution shows the proposed method can find the optimal solution similar to the BLX algorithm and better than UNDX 35 30th generation 70th generation
  • 36. Results-Airfoil Optimization Problem • Metrices of Airfoil Optimization Problem 36 Hypervolume Maximum Spread Spacing
  • 37. Results-Airfoil Optimization Problem 37 • PCP plot of the Pareto solution at 70th generation • The plot shown the propose method have the solution similar to BLX 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 dv1 dv2 dv3 dv4 dv5 Cd Cl BLX 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 dv1 dv2 dv3 dv4 dv5 Cd Cl UNDX 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 dv1 dv2 dv3 dv4 dv5 cd cl Proposed method
  • 38. Results-Airfoil Optimization Problem • Left figure show the airfoil comparison and the right figure show the Cp(Pressure Distribution) comparison at design point(Cl=0.8 at 70th generation) • The plot shown the propose method have the solution similar to BLX 38
  • 39. Conclusion • Successful to development of new multi-parent crossover method • The proposed method successfully find out the better solution of every test function compared with existent crossover method • The multi-objective test function are better compared with other crossover method • For the airfoil optimization problem the searching areas are similar because the parameterizations is relatively smile so the investigation by more complex problem is need • This method can maintain higher diversity 39