PPT file

advertisement
Constrained Optimization
by the e Constrained Differential Evolution
with an Archive and Gradient-Based Mutation
Tetsuyuki TAKAHAMA
(Hiroshima City University)
Setsuko SAKAI
(Hiroshima Shudo University)
Outline

Constrained optimization problems

The e constrained method


Constraint violation and e-level comparisons
The e constrained differential evolution (eDEag)

differential evolution (DE) with an archive

gradient-based mutation

control of the e-level

Experimental results

Conclusions
2010/07/21
T.Takahama and S.Sakai in CEC2010
2
Constrained Optimization Problems

objective function f , decision variables xi

inequality constraints gj, equality constraints hj

lower bound li, upper bound ui
2010/07/21
T.Takahama and S.Sakai in CEC2010
3
e constrained method

Algorithm transformation method
 algorithm
for unconstrained optimization
→ algorithm for constrained optimization
 e-level comparison
 comparison
between pairs of objective value
and constraint violation
replacing ordinary comparisons to e-level
comparisons in unconstrained optimization
algorithm
 by
2010/07/21
T.Takahama and S.Sakai in CEC2010
4
Constraint Violation

Constraint Violation f (x)
f ( x )  0 , if x is infeasible
f ( x )  0 , if x is feasible

max
f ( x )  max{ max { 0 , g j ( x )}, max | h j ( x ) |}

j
sum
f (x) 
 || max{
j
2010/07/21
j
0 , g j ( x )} ||   || h j ( x ) ||
p
p
j
T.Takahama and S.Sakai in CEC2010
5
e-level comparison

Function value and constraint violation(f ,f)
2010/07/21
 precedes
constraint violation usually
 precedes
function value if violation is small
T.Takahama and S.Sakai in CEC2010
6
Definition of e constrained method
∥

Constrained problems can be solved by replacing
ordinary comparisons with e level comparisons in
unconstrained optimization algorithm

2010/07/21
<→<e,  →
e
T.Takahama and S.Sakai in CEC2010
7
Differential Evolution (DE)

simple operation avoiding step size control
crossover
parent
population
base vector
-
F
difference vector
(CR)
+
trial vector

trial vector (child) will survive if the child is better

robust to non-convex, multi-modal problems
2010/07/21
T.Takahama and S.Sakai in CEC2010
8
eDEa: eDE with an archive (1)

A small population and a large archive are adopted


N
Generate M initial individuals


Small population is good for search efficiency
but is bad for diversity
A={ xk | k=1,2,...,M } (M=100n)
Select top N individuals from A
as an initial population
2010/07/21
P

P={ xi | i=1,2,...,N } (N=4n)

A=A-P
A
T.Takahama and S.Sakai in CEC2010
M-N
9
eDE with an archive (2)

DE/rand/1/exp operation
correction of Fig.2
 mutant
vector:

and
are selected from P

is selected from P A w.p. 0.95 or P w.p. 0.05
 exponential crossover

Uniform convergence of individuals
 When
a parent generates a child and
the child is not better than the parent,
the parent can generate another child
2010/07/21
T.Takahama and S.Sakai in CEC2010
10
eDE with an archive (3)

Direct replacement for efficiency
 Continuous
generation model
 If the child is better than the parent,
the parent is directly replaced by the child

(f(xtrial),f(xtrial)) <e (f(xi), f(xi))
Perturb scaling factor F in small probability
 to
escape from local minima
 F is a fixed value (0.5) w.p 0.95
 F=1+|C(0,0.05)| truncated to 1.1 w.p. 0.05
2010/07/21
T.Takahama and S.Sakai in CEC2010
11
Gradient-based mutation (1)


adopts the gradient of constraints to reach
feasible region
Constraint vector and constraint violation vector
T
C( x )  (g 1 ( x ) , , g q ( x ) ,h q  1 ( x ) , h m ( x ))
 C( x )  (  g 1 ( x ) , ,  g q ( x ) ,h q  1 ( x ) , h m ( x ))

T
 g j ( x )  max{ 0 , g j ( x )}
Gradient of constraint vector
 C( x )  x    C( x )
if satisfied,
2010/07/21
x   x will be feasible
T.Takahama and S.Sakai in CEC2010
12
Gradient-based mutation (2)


inverse  C( x )  1 cannot be defined generally
Moore-Penrose inverse (pseudoinverse)  C( x ) 
 approximate
or best (LSE) solution

x '  x   C( x )  C( x )

Modifications
 Numerical
gradient (costs n+1 FEs)
 Mutation is applied only in every n generations
 Skipped w.p. 0.5, if num. of violated constraints
is one
2010/07/21
T.Takahama and S.Sakai in CEC2010
13
Control of e-level

Small feasible region and e-level
small feasible region
f=0
f≦ e(Tf)
f≦ e(0)
2010/07/21
T.Takahama and S.Sakai in CEC2010
14
Control scheme of e-level
 e-level
should converge to 0 gradually
f ( x  )
cp




t
e (t )  e ( 0 ) 1 

Tc 


 0
e(t)
e0
(t  0 )
( 0  t  Tc )
(t  Tc )
x  is the top - th individual
0
2010/07/21
Tc
Tmax
in f
t
T.Takahama and S.Sakai in CEC2010
15
control of cp

instead of specifying cp, specify e-level at Tl
,


To search better objective value
 generation
 enlarge
2010/07/21
from Tl to Tc
e-level and scaling factor F
T.Takahama and S.Sakai in CEC2010
16
Effectiveness of e constrained method


The e level comparison does not need
objective values if one of the constraint
violations is larger than e-level
Lazy evaluation
 objective
function is evaluated only when
needed
 evaluation
of objective function can be often
omitted when feasible region is small
2010/07/21
T.Takahama and S.Sakai in CEC2010
17
Conditions of experiments

18 constrained problems, 25 trials per a problem

eDEag/rand/1/exp


Max. FEs: 20,000n

M=100n, N=4n, F=0.5, CR=0.9

e level control: =0.9, Tc=1,000, Tl=0.95Tc
Gradient-based mutation
2010/07/21

mutation rate: Pg=0.1, max. iterations: Rg=3

applied only in every n generations
T.Takahama and S.Sakai in CEC2010
18
Summary of Results



Feasible and stable solutions in all runs
 10D:
C01-C07, C09, C10, C12-C14, C18 (13)
 30D:
C01, C02, C05-C08, C10, C13-C16 (11)
Feasible solutions in all runs
 10D:
C08, C11, C15, C16, C17
(5)
 30D:
C03, C04, C09, C11, C17, C18 (6)
Often infeasible solutions
 30D:
2010/07/21
C12 (1)
T.Takahama and S.Sakai in CEC2010
19
10D (C01-C06)
2010/07/21
T.Takahama and S.Sakai in CEC2010
20
10D (C07-C012)
2010/07/21
T.Takahama and S.Sakai in CEC2010
21
10D (C13-C018)
2010/07/21
T.Takahama and S.Sakai in CEC2010
22
30D (C01-C06)
2010/07/21
T.Takahama and S.Sakai in CEC2010
23
30D (C07-C12)
2010/07/21
T.Takahama and S.Sakai in CEC2010
24
30D (C13-C18)
2010/07/21
T.Takahama and S.Sakai in CEC2010
25
Computational Complexity


T1: Time (seconds) of 10,000 function
evaluations for a problem on average
T2: Time (seconds) of 10,000 function
evaluation with algorithm for a problem
2010/07/21
T.Takahama and S.Sakai in CEC2010
26
Conclusions

eDE with a large archive and gradient-based
mutation
 can
find feasible solutions in all run and all
problems except for C12 of 30D
 can
often omit evaluation of objective values
and find solutions efficiently and very fast
2010/07/21
T.Takahama and S.Sakai in CEC2010
27
Future works

To find better objective values
 dynamic
control of e level
e level according to the number of
feasible points
 changing
 mechanism
for maintaining diversity
 subpopulations
or species to search various
regions
 adaptive
2010/07/21
control of F and CR
T.Takahama and S.Sakai in CEC2010
28

Thank you for your kind attention
2010/07/21
T.Takahama and S.Sakai in CEC2010
29
10D problems
2010/07/21
T.Takahama and S.Sakai in CEC2010
30
10D problems
2010/07/21
T.Takahama and S.Sakai in CEC2010
31
30D problems
2010/07/21
T.Takahama and S.Sakai in CEC2010
32
30D problems
2010/07/21
T.Takahama and S.Sakai in CEC2010
33
Moore-Penrose inverse

A : pseudo inverse
singlar va
lue decomposit
A  UV


T
 : inverting
non - zero elements
on the diagonal
2010/07/21
ion
T
A  V U

of A
of 
T.Takahama and S.Sakai in CEC2010
34
Download