TELKOM
NIKA
, Vol.14, No
.2, June 20
16
, pp. 647~6
5
4
ISSN: 1693-6
930,
accredited
A
by DIKTI, De
cree No: 58/DIK
T
I/Kep/2013
DOI
:
10.12928/TELKOMNIKA.v14i1.2753
647
Re
cei
v
ed
De
cem
ber 2
1
, 2015; Re
vi
sed
March 18, 20
16; Accepted
April 3, 2016
A Self-adaptive Multipeak Artificial Immune Genetic
Algorithm
Qingzh
a
o Li
1
,
Fei Jiang
2,3
*
1
School of F
o
reig
n La
ngu
ag
e
s
, Suzhou Un
iv
ersit
y
, Suz
hou
234
00
0, Anhui,
Chin
a
2
Labor
ator
y
of Intelli
ge
nt Information Proc
es
sing,
Suzh
ou U
n
iversit
y
, Suzh
ou 23
40
00, An
hui, Ch
ina
3
School of Infor
m
ation En
gi
ne
erin
g, Suzho
u
Univers
i
t
y
, Suz
hou 2
3
4
000, A
nhu
i, Chin
a
*Corres
p
o
ndi
n
g
author, e-ma
i
l
: fei
w
u
h
an@
1
26.com
A
b
st
r
a
ct
Genetic alg
o
rit
h
m
is a glo
b
a
l
prob
abi
lity
se
a
r
ch
al
gor
ith
m
d
e
vel
ope
d
by si
mu
lati
ng th
e
bi
olo
g
ica
l
natura
l
s
e
lecti
o
n a
n
d
g
enetic
evol
ut
ion
mec
h
anis
m
an
d
it h
a
s exc
e
ll
ent
gl
oba
l se
arch
a
b
ility, h
o
w
e
ver, i
n
practica
l ap
pl
ic
ations, pr
e
m
at
ure co
nver
gen
ce occurs
eas
il
y in the
ge
neti
c
alg
o
rith
m. T
h
is pa
per
prop
o
s
es
an se
lf-ada
ptiv
e multi-p
eak
i
m
mu
ne
gen
eti
c
alg
o
rith
m (S
MIGA) and thi
s
alg
o
rith
m i
n
tegrates
i
m
mun
i
ty
thoug
ht in the
biolo
g
y i
m
mu
ne system i
n
to the evol
utio
nary proc
ess of genetic
a
l
g
o
rith
m, uses self-
ada
ptive dy
na
mic vacc
in
atio
n and pr
ovi
des
a dow
ntime cr
i
t
erion, the se
le
ction st
rategy
of immu
ne vac
c
ine
and
the
co
nstruction
meth
od
of i
m
mu
ne
op
erators s
o
as t
o
pr
o
m
ote
the
pop
ulati
o
n
d
e
v
e
lo
p tow
a
rds
the
opti
m
i
z
at
ion
trend
an
d s
uppr
ess the
de
ge
n
e
racy
phe
no
meno
n i
n
th
e o
p
timi
z
a
ti
on
by
usin
g the
feat
ur
e
infor
m
ati
on
in
a sel
e
ctive
an
d
purp
o
siv
e
ma
nner. T
h
e si
mu
latio
n
ex
peri
m
e
n
t show
s that t
he
met
hod
of t
h
is
pap
er
ca
n better
solv
e
th
e o
p
timi
z
a
ti
on pro
b
le
m of mu
lti-
peak
functi
ons
, reali
z
e
glo
b
a
l
opti
m
u
m
se
ar
ch,
o
v
e
r
come
th
e p
r
e
m
a
t
u
r
i
t
y p
r
o
b
l
e
m
o
f
th
e
a
n
t
i
b
o
d
y
p
o
p
u
l
a
t
i
o
n
an
d
im
p
r
o
v
e
th
e
effe
cti
v
e
n
e
ss and
robustn
ess of opti
m
i
z
at
ion.
Ke
y
w
ords
: Genetic Alg
o
rith
m, Artificial Immune Syste
m
, Multip
eak F
uncti
on
Copy
right
©
2016 Un
ive
r
sita
s Ah
mad
Dah
l
an
. All rig
h
t
s r
ese
rved
.
1. Introduc
tion
With the rapi
d develop
me
nt of informat
i
on tech
nolo
g
y
, intelligent comp
utation
method
s
have b
een
ap
plied in
an
in
cre
a
si
ng
num
ber
of fi
eld
s
a
nd the
wid
e
ly-used i
n
tellig
ence al
gorith
m
s
inclu
de g
ene
tic alg
o
rithm,
ant
colo
ny algorith
m
an
d immu
ne al
gorithm
an
d
so
on. G
ene
ti
c
algorith
m
is a
n
evolutiona
ry algorithm with per
vasive
influen
ce formed at the very first begin
n
i
ng
and its ba
si
c princi
ple is
to exchan
ge
information
throug
h the geneti
c
ope
rations
su
ch
as
cro
s
sove
r, duplication an
d mutation in orde
r to obtain the n
ear optimu
m
solution to
the
optimizatio
n
probl
em [1]. Biology im
mune
syste
m
is a hi
ghl
y-parallel inf
o
rmatio
n lea
r
ning
system. It ca
n self-ad
aptively re
cog
n
ize
and
elimi
nat
e the a
n
tigen
ical fo
reig
n o
b
ject
s fro
m
the
body an
d it
has
adju
s
tm
ent ability su
ch a
s
m
e
mo
ry and l
earni
ng. Inspi
r
e
d
by the intelli
gent
behavio
rs of
the biology
immune sy
stem, artifi
ci
al immune a
l
gorithm i
s
a
rando
m se
arch
optimizatio
n algorith
m
int
egratin
g cert
aint
y and
randomi
c
ity [2]. At prese
n
t, intelligen
ce
algorith
m
s h
a
ve be
en
wi
dely u
s
ed i
n
machine
le
arnin
g
, patte
rn re
co
gnition
, robot
beh
a
v
ior
simulatio
n
, intelligent troub
le diagno
si
s, manag
em
e
n
t scien
ce a
n
d
social
sci
en
ce and they are
appli
c
able to
solve compli
cated non
-line
a
r and m
u
lti-d
i
mensi
onal
sp
ace o
p
timizati
on pro
b
lem
s
.
In as e
a
rly a
s
the 194
0s, t
he sch
o
lars h
a
ve
bee
n re
searchin
g ho
w to use th
e compute
r
to simulate
bi
ology technol
ogy. In the e
a
rly 197
0s, P
r
ofesso
r
Holl
and fro
m
Michigan
Unive
r
sity,
USA, ha
s
dra
w
n i
n
spiratio
n from
bi
ologi
cal
gen
etics
and
evolution
a
ry me
ch
ani
sm an
d p
r
op
o
s
e
d
geneti
c
algo
ri
thm, which i
s
a sea
r
ch m
e
cha
n
is
m b
a
sed on the th
eory of natural evolution a
nd
whi
c
h is an
optimization
method ap
plica
b
le
for
the comp
uta
t
ion of com
p
licate
d
syst
em;
however,
genetic algori
thm has
a poor local
search ability.
Therefore, how t
o
preserve the
excelle
nt indi
viduals
and t
he po
pulation
diversity
ha
s been
a difficulty in the ge
netic al
gorith
m
[3]. After the 197
0s, p
e
o
p
le have
g
r
a
dually r
eali
z
e
d
the e
n
light
enment
of bi
ology immu
n
e
mech
ani
sm i
n
developi
ng
new
comp
utational intelli
gen
ce. As a
new alg
o
rith
m inspi
r
ed from
biology immu
ne syste
m
, artificial immun
e
algorit
h
m
(AIA) focuses
on simul
a
ting
the immunol
ogy
with comp
utational and mathemati
c
al
model.
It si
mulates
the biology
imm
une system
an
d
c
o
ns
tr
uc
ts
th
e
c
o
rr
es
p
ond
in
g
in
fo
r
m
atio
n
p
r
oc
e
ssi
ng algo
rithm
s
by lear
nin
g
it
s st
ru
ct
u
r
e,
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 14, No. 2, June 20
16 : 647 – 65
4
648
function
s and
features. By integratin
g the ideas
of GA
and AIA, immune ge
netic algorithm hel
p
s
the antib
ody
to
continu
o
u
sly
sea
r
ch
and
evolve
in the
solu
tion spa
c
e t
h
rou
gh i
m
m
une
operation
s
an
d evaluate
s
the matchi
ng
degree bet
we
en the antibo
d
y and the an
tigen as
well as
the simil
a
rity
betwe
en the
antibodi
es
ba
sed
on th
e af
f
i
nity until the
optimal
soluti
on i
s
ge
ne
rat
ed.
Ho
wever, the
basic imm
u
n
e
genetic al
g
o
rithm ha
s sl
ow and
stagn
ant sea
r
ch proce
s
s bro
ugh
t by
the static de
si
gnated
cro
ssover, mutatio
n
prob
ability and vaccin
e [4].
This pa
per f
i
rstly descri
b
es the prin
ci
ples
of ba
si
c geneti
c
al
gorithm an
d
artificial
immune
algo
rithm. Then
based o
n
the sum
m
ary
of
the ch
ara
c
teri
stics an
d
sho
r
tco
m
ing
s
o
f
these
two
al
gorithm
s, it i
m
prove
s
th
e
algo
rithm by
co
mbinin
g
i
mmunol
ogy and gen
etics
and
prop
oses
an
self-adaptiv
e multi-p
e
a
k
immune
g
enetic
algo
ri
thm. Finally, the num
eri
c
al
simulatio
n
of the test function demo
n
stra
tes that
this algorithm can
obtain glob
al optimal sol
u
tion
effectively an
d verifies th
e
effectivene
ss and p
r
a
c
ti
cal
i
ty of this improved al
gorith
m
in the pa
p
e
r
so that it prov
ides a
solutio
n
to t
he multi-pea
k functio
n
optimization
probl
em
s.
2. Genetic
Algorithm
Geneti
c
algo
rithm (GA)
si
mulates th
e repro
d
u
c
tion,
cro
s
sove
r an
d geneti
c
mu
tation in
natural
sele
ction an
d n
a
tural inhe
ritan
c
e
and it i
s
a ra
ndomi
z
ed
se
arch al
gorith
m
by referrin
g to
the natural se
lection a
nd g
enetic m
e
cha
n
ism in th
e bi
ologi
cal world
.
It takes the
encodin
g
of the
deci
s
io
n vari
able a
s
the o
pera
nd a
nd t
he obj
ective functio
n
value
as the
se
arch inform
ation
and
adopt
s m
u
ltiple
sea
r
ch
p
o
ints to
pe
rf
orm
po
ssi
bility sea
r
ch. Th
e g
r
oup
of
origin
al
soluti
on
rand
omly ge
neratin
g fro
m
solving
the
probl
em
s
wit
h
GA is calle
d pop
ulation
and the
probl
em-
solving
pro
c
e
ss
sta
r
ts fro
m
the sub-se
t. Each i
ndivi
dual of the
p
opulatio
n is
a sol
u
tion to
the
probl
em and
the enco
d
in
g string o
b
tai
ned from
p
e
rforming bi
nary coding to the individual
is
calle
d ch
rom
o
som
e
. The
chromo
som
e
is mea
s
ured
by the fitne
ss. Th
e fitness fun
c
tion i
s
a
function
to m
easure
the
e
n
vironm
ental
adaptatio
n of
every in
divid
ual a
nd it
s v
a
lue i
s
th
e m
a
in
basi
s
to
re
ali
z
e
“survival o
f
the fittest”.
Acco
rdi
ng to
the fitness,
select
ce
rtain i
ndividual
s fro
m
the pare
n
t g
eneration an
d the offspri
ng gen
eratio
n and ge
nerate the next generation
of
chromo
som
e
popul
ation
whi
c
h i
s
more
self
-ad
aptive to the enviro
n
me
nt. After ce
rtain
gene
ration
s,
the algo
rithm
conve
r
ge
s t
o
the optimal
solutio
n
or t
he second
-o
ptimal sol
u
tio
n
to
the probl
em [5]. To map the optimal sol
u
tion sp
ac
e to GA spa
c
e i
s
indi
cated a
s
Figure 1.
Figure 1. Map the optimization
p
r
ob
le
m s
p
ac
e
to
G
A
s
p
ac
e
Selection, cro
s
sover an
d m
u
tation are th
e ma
in thre
e operators in
GA and they simulate
the survival o
f
the fitt
est and the geneti
c
pro
c
e
ss.
(1) Sele
ction
The first
step
of selectio
n
is to
cal
c
ulat
e the fitne
ss.
Every indivi
dual in
the
set to be
selected has a selection probab
ility, whi
c
h depends on the fit
ness and di
stribution of the
individual in the pop
ulatio
n and u
s
e
s
the pro
babilit
y
ratio of the fitness of ev
ery individual
to
determi
ne th
e
po
ssi
bility to
pre
s
e
r
ve its o
ffspring.
A
s
sume th
at the
popul
ation
si
ze is
N
and
the
fitness of the
individual
i
is
i
f
, then the probability
s
P
for the individual
i
to be s
e
lec
t
ed is
:
1
(1
,
2
,
,
)
i
N
s
i
i
f
Pi
N
f
(1)
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
A Self-adapti
v
e Multipe
a
k
Artificial Immune Ge
netic
Algorithm
(Qi
ngzhao Li
)
649
Obviou
sly, th
e individual with a higher selectio
n pro
b
ability has a highe
r po
ssi
b
ility to
be
sele
cted
an
d
the
pro
babili
ty to be i
nhe
rited to th
e p
o
pulation
in th
e next
gene
ration i
s
dire
ct
ly
prop
ortio
nal to the fitness
of this individ
ual. Pr
opo
rtio
nal sel
e
ctio
n is also call
ed
Roulette Wh
eel
Selection. It divides the roulette into
N
portion
s ba
se
d on the sele
ction proba
bil
i
ty
s
P
. When
rotating the roulette in the
sele
ction, if a ce
rtai
n poi
nt falls into the ith se
ctor,
then sele
ct the
individual i. T
he di
stributio
n ratio
of the f
i
tness value o
f
the Roulette
Whe
e
l Sele
ction is indi
cat
e
d
as Figu
re 2.
Figure 2. Dist
ribution
ratio
of fitne
ss valu
e of roulette
whe
e
l sel
e
cti
o
n
(2) Cro
s
sov
e
r
Cro
s
sove
r is
to excha
nge
the gen
es in
the
sa
me p
o
s
ition in
two
different in
dividual
s to
be u
s
ed to
repro
d
u
c
e the
next gene
rat
i
on an
d ge
ne
rate n
e
w i
ndi
viduals [6]. In
other
wo
rd
s, it
take
s the
two
individu
als from the
po
pul
ation afte
r
sel
e
ction
op
erati
on a
s
th
e
cro
s
sover obje
c
t
s
,
rand
omly ge
n
e
rate
s a
crossover point
a
nd pe
rfor
m crossover on
some g
ene
s in
the rig
h
t sid
e
of
the cro
s
sove
r point, as indi
cated in Fig
u
re 3.
Figure 3. Cro
s
sover op
erator
(3) Mut
a
tion
Mutation i
s
to re
place th
e gen
e value
of ce
rtain lo
cu
s in the
e
n
co
ding
strin
g
of th
e
individual chromosome
with other allel
e
gene in the
same lo
cu
s, prod
uce mut
a
tion at a sm
all
prob
ability or step length
and form a
new indivi
d
ual. The sim
p
lest way is to chang
e th
e
nume
r
ical value of a certai
n position in the st
rin
g
. For the binary en
codi
ng, interchang
e 0 and 1:
cha
nge 0 to 1
and 1 to 0.
It is an a
u
xiliary metho
d
t
o
gen
erate n
e
w in
dividual
s an
d it dete
r
mines the lo
cal se
arch
ability of GA. The
com
b
ination of
mutation operator, se
lection operator and
crossover operator
help
s
GA ha
ve the local
rando
m se
a
r
ch a
b
ility and preve
n
t local
conve
r
g
ence. Gene
rally
spe
a
ki
ng, low-frequ
en
cy mutation ca
n
avoid t
he possible mi
ssi
ng of the important an
d single
gene. Th
e co
mmonly-u
s
e
d
mutation op
e
r
ation in
clu
d
e
s
ba
si
c bit m
u
tation, uniform mutation a
nd
binary mutation [7].
Geneti
c
alg
o
rithm is e
a
sy
for pa
ralleli
za
tion pro
c
e
s
si
ng an
d its fit
ness fun
c
tion
is not
rest
ricte
d
by su
ch con
d
itio
ns
a
s
contin
u
i
ty
and
differe
ntiation an
d it has
an exten
s
ive ap
plication
scope;
ho
we
ver, GA i
s
ea
sy to
get trap
ped i
n
lo
ca
l
o
p
timal
solutio
n
an
d it
s gl
ob
al sea
r
ch a
b
il
ity
is also very limited.
3. Artificial Immune Algo
rithm
Biological im
mune
syste
m
ha
s the
feature
s
su
ch a
s
imm
u
n
e
mem
o
ry,
antigen
recognitio
n
a
nd preservati
on of antibo
d
y divers
ity. By learning
from the a
bove-m
ention
e
d
feature
s
, im
mune
algo
rithm ge
nerate
s
a
co
m
put
ation mo
del
by integ
r
ati
ng en
gine
eri
ng
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 14, No. 2, June 20
16 : 647 – 65
4
650
appli
c
ation.
Comp
ari
ng t
he imm
une
algorith
m
wit
h
the
comm
on
sea
r
ch m
e
thod
s to
so
lve
optimizatio
n
probl
em
s, the affinity betwee
n
antige
n
and antibo
d
y
as well a
s
betwe
en anti
body
and a
n
tigen
corre
s
p
ond t
o
the matchi
ng bet
wee
n
the obje
c
tive
function
of the optimization
probl
em a
nd
the optimal
solution a
s
we
ll as b
e
twe
e
n
the solution
and the
obje
c
tive functio
n
.
Artificial imm
une algo
rith
m includ
es
such o
perat
io
ns a
s
vaccin
e extraction,
vacci
ne ino
c
ulation
and va
ccine
sele
ction [
8
, 9]. The
step
s to solve opti
m
ization
prob
lems
with im
mune
algo
rith
m
are cl
assified
as follo
ws.
Step 1: Identify the proble
m
, take the
given
obje
c
ti
ve function a
nd co
nst
r
aint
s a
s
the
antigen of th
e algo
rithm, namely to e
x
tract t
he effective inform
ation of the
probl
em to b
e
optimize
d
, the obje
c
tive function of the
optimal pr
obl
em and the d
e
scriptio
n of the co
nstraint
s.
Step 2: Gen
e
rate the a
n
tibody pop
ula
t
ion,
rando
ml
y generate t
he origi
nal a
n
tibody
popul
ation of
a certain
scal
e. The
antibo
d
y pop
ulatio
n
is
rep
r
e
s
ente
d
with
bina
ry
encodin
g
an
d
it
define
s
the a
n
tibody as th
e feasibl
e
sol
u
tion of
the objective fun
c
tion in the co
n
s
traint
s.
Step 3: G
e
n
e
rate
immu
n
e
mem
o
ry
cell, cal
c
ul
ate
the
dire
ct a
ffinity betwee
n
ea
ch
antibody a
nd
antigen
re
spe
c
tively as
wel
l
as th
e
si
milarity between
the antib
odie
s
an
d p
r
e
s
e
r
v
e
the antibody
with a bigg
er
fitness valu
e as the mem
o
ry cell.
Step 4: Sele
ct the antibo
d
y and
cal
c
ulat
e the a
ffinity, cal
c
ulate th
e
con
c
e
n
tration
of the
antibodi
es
wit
h
simil
a
r fitne
ss i
n
the
cu
rrent antibo
d
y popul
ation a
s
well a
s
th
e a
ffinity or fitness
of every solut
i
on. Decre
a
se the sele
ctio
n prob
ab
ility of the individual with high
con
c
e
n
tration
and
increa
se the
sele
ction p
r
o
bability of the individual wit
h
low con
c
ent
ration.
Step 5: Differentiate the m
e
mory
cell
s, sort
th
em b
a
s
ed
on the
af
finity, put the antibody
with high
affinity to the antigen into the
memory
cell
, take the first few antibod
ies to form t
h
e
memory ware
hou
se an
d prese
r
ve the cu
rre
nt optimal feasi
b
le solution.
Step 6: Evolve and upd
ate the antibody wi
th crossover a
n
d
mutation operato
r
s,
inclu
d
ing th
e
antibo
d
y a
c
cele
ration
an
d supp
re
ssi
o
n
an
d g
ene
rate ne
w
anti
body p
opulat
ion.
Repl
ace the i
ndividual
of low fitne
s
s value with
that o
f
high fitne
ss
value in the
memory
cell
an
d
form the next-gen
eration a
n
tibody
popul
ation. Eliminate the antib
odie
s
with lo
w affinity of the
antigen a
nd h
i
gh co
ncentra
tion and form
a new g
ene
ra
tion of popula
t
ion.
Step 7: Ju
dg
e wh
ethe
r th
e termin
ation
con
d
ition
s
a
r
e
satisfie
d, namely to fi
nd the
optimal
soluti
on o
r
to
rea
c
h the m
a
ximu
m iteration
s
. I
f
so, outp
u
t th
e optimal
sol
u
tion; othe
rwi
s
e,
contin
ue Step
3.
The ba
sic flo
w
chart of artif
i
cial immu
ne
algorith
m
is i
ndicated a
s
Figure 4.
Figure 4. Flowchart of ba
sic artificial im
mune alg
o
rith
m
4. Process o
f
Self-a
dap
t
iv
e
Multipeak
Immune Genetic Algo
rithm
Artificial imm
une sy
stem can more bett
e
r en
su
re tha
t
algorithm i
s
not easy to
fall into
local
optimal
solution due to its unique antib
o
d
y di
versity maint
a
ining
me
cha
n
ism
duri
ng t
h
e
sea
r
ch p
r
o
c
e
ss. T
he mai
n
idea i
s
to ad
opt the
dyna
mic extra
c
tio
n
vaccine
me
thod to avoid
the
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
A Self-adapti
v
e Multipe
a
k
Artificial Immune Ge
netic
Algorithm
(Qi
ngzhao Li
)
651
disa
dvantag
e
of slow conv
erge
nce spe
e
d
resulted
fr
o
m
the
effectivene
ss of
static va
cci
ne i
n
t
he
pro
c
e
ss
of evolution. The
selectio
n upd
a
t
e me
ch
ani
sm based on t
he co
ncentration is a
dopte
d
to
repla
c
e the
selectio
n repli
c
ation b
a
sed on the fit
ness in terms of the origi
nal g
enetic al
gorit
hm,
thus maintai
n
ing the diversity of solution grou
p.
The self-ada
ptive adjustin
g
crossover pro
babili
ty
and mutatio
n
prob
ability gu
ides the
se
arch p
r
o
c
e
ss to
wards th
e glo
bal optimi
z
ati
on [10, 11]. The
step
s of algorithm in this paper a
r
e a
s
follows:
Step 1: Encode in view of cha
r
a
c
teri
stics of
questio
n
s
to be solve
d
with the pu
rpo
s
e to
gene
rate th
e
origi
nal
pop
ulation. Set
su
ch
parame
t
ers a
s
the
crossove
r
pro
bability
c
p
, the
mutation probability
m
p
, the populatio
n
size
N
, the binary en
co
din
g
length
L
, the max
evolution ge
neratio
n
T
, th
e medium in
dividual num
ber of ge
net
ic ope
ration
M
, and the
con
c
e
n
tration
thresh
old val
ue
.
Step 2: Take
the given o
b
jective fun
c
tion
and
con
s
traint co
nditio
n
as the a
n
tigen of
que
stion
s
to
be solved to
gene
rate
ori
g
inal a
n
tigen
gro
up. Rand
omly gen
erat
e
N
individu
a
l
popul
ations a
nd extract
M
i
ndividual
s fro
m
mem
o
ry
wareh
o
u
s
e to
form the
o
r
ig
inal a
n
tigen
grou
p, in whi
c
h,
M
is the nu
mber of indivi
dual
s in the memory ware
hou
se.
Step 3: Calcu
l
ate the con
c
entration of a
n
tibody
v
c
.
1
1
N
vv
w
w
ca
c
N
(2)
In whic
h:
1
0
vw
vw
ay
ac
ot
he
rwise
,
is the dete
r
min
ed threshold
value.
vw
ay
is
t
h
e affinity
betwe
en anti
body
v
ab
and ant
ibody
w
ab
.
Step 4: Extract va
cci
ne
s and
calculat
e ea
ch
indivi
dual’
s
fitne
s
s value i
n
the
cu
rrent
popul
ation to
find out the
optimal in
di
vidual, and
take it
s all
g
ene
s a
s
va
ccine
V
. Set
M
antibodi
es
wi
th fitness op
timization i
n
stored pop
ulation
of ea
ch gene
ratio
n
,
then
*
KM
antibodi
es
in
total are reta
ined in
recent
gen
eratio
n
K
to form
the o
p
t
imal antibo
d
y
grou
p. Th
e
prob
ability of each gen
e le
vel of each a
n
tibody is:
*
,
1
1
*
KM
ij
j
j
pa
KM
(3)
Take
j
k
with max pro
babil
i
ty
,1
,
2
,
,
i
pi
L
in
such allelic
g
e
ne a
s
th
e va
cc
ine
segm
ent in such all
e
lic ge
ne, t
hus finall
y
extracting the vaccine
12
(,
,
,
)
L
Vv
v
v
.
Step 5: Va
ccination,
sele
ct antibody
to
be
vaccinat
ed from th
e
previou
s
gen
eration
grou
p a
nd
g
enerate n
e
w immun
e
in
d
i
viduals by e
x
chan
ging
th
e ge
netic co
de valu
es. T
he
sele
ction ma
nner of the g
enetic
segm
e
n
t is as follo
ws:
1
i
i
L
j
j
p
q
p
(4)
i
q
is the prob
ability of each vaccine
se
gment sele
cted and
L
is the length
of the
geneti
c
se
gm
ent.
,1
,
2
,
,
ij
L
.
Step 6: Pop
u
lation u
pdat
e, the follo
wing way is
adopte
d
to i
m
pleme
n
t the group
updatin
g ba
sed on the con
c
entration:
()
()
()
1
()
()
sv
f
it
v
f
it
v
pv
c
M
ax
fit
v
Max
f
it
v
(5)
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 14, No. 2, June 20
16 : 647 – 65
4
652
In which,
an
d
belong
s to adju
s
table
para
m
eters b
e
twee
n 0 an
d 1, and
()
s
pv
rep
r
e
s
ent
s th
e sele
ction
p
r
oba
bility of
antibody
v
,
()
f
it
v
i
s
the
fitness value
of nu
mber
v
antibody,
()
M
a
x
fit
v
is the max fitness of the antib
ody, and
v
c
is th
e antibody
v
’s
c
o
nc
en
tr
a
t
io
n.
Selec
t
M
optima
l
individual
s a
c
cordi
ng to th
e co
ncentrati
on of antib
od
y from high to
low
to form a
ne
w po
pulatio
n
(1
)
Xt
, thus the
proportio
n
of o
p
timal individ
uals i
n
the p
opulatio
n
can be increased to improv
e the local
search ability.
Step 7: Self-ada
ptive cro
s
sover an
d muta
tion ope
ration, self
-a
daptively sel
e
ct the
cro
s
sove
r an
d mutation p
r
obability acco
rding to the
di
versity of pop
ulation. Imple
m
ent crossov
e
r
towards
M
indivi
dual
s ba
se
d
on the p
r
obability
c
p
to form the
popul
ation
2
()
X
t
, then
respec
tively implement mutation towards
M
individual
s in
the
pop
ulation
2
()
X
t
ba
se
d on
th
e
probability
m
p
to form popul
ation
3
()
X
t
.
Step 8: Te
rm
ination te
st, if termin
ation
cr
ite
r
ia
has b
een m
e
t, out
put the gl
oba
l optimal
s
o
lution s
o
far, otherwise, return to Step 3.
5. Numerical
Optimizatio
n
Experimen
t
and An
aly
s
is
To test the
effectivene
ss
o
f
the algo
rith
m pro
p
o
s
ed i
n
this p
ape
r i
n
term
s of th
e sol
u
tion
of to solve
Multipea
k opt
imization p
r
o
b
lem, t
he typical Multipe
a
k
functio
n
is optimize
d
. See
following formula for specific
desc
r
iption of tes
t
func
tion:
2
2
2
(
(
0.08
)
/
0.
854
)
6
3
/
4
sin
(
5
(
0.05
)
)
,
[
0
,
1
]
ln
x
fe
x
x
(6)
Functio
n
f
has five peak val
ues
with n
o
n
-
uniform
dist
ribution a
nd n
o
n
-unifo
rm altit
ude.
Optimal value
s
of these fun
c
tion
s are diff
icult to sea
r
ch with gen
etic algorithm.
Antigens in th
e expe
riment
are
seve
ral
p
eak value
s
of
the fun
c
tion t
o
be
solve
d
,
and the
antibody i
s
th
e inde
pen
den
t variable
x
of the obtai
ned
p
eak val
ue, a
n
d
x
is re
pre
s
e
n
t
ed by the
binary st
ring
with the lengt
h of 15. The para
m
eter
se
tting of two kinds of alg
o
rit
h
m are
sho
w
n in
Table 1. The
experim
ental
comp
utationa
l
result
s are shown in Figu
re 5 and 6.
Table 1. Para
meters of Two Algorithm
Algorithm
Cr
ossover
probabilit
y
c
p
Mutation
probabilit
y
m
p
Concentration
threshold value
Grou
p
size
N
Memory
cell
length
Genetic algorith
m
0.07
0.02
50
SMIGA
0.75
0.15
0.85
50
Variable
G=100
G=2
00
G=3
00
Figure 5. Re
sults of ge
net
ic algo
rithm
From a
bove
Figure 5 an
d
Figure 6, we ca
n
see th
at, this SMIGA algorithm
can
still
maintain
con
v
ergen
ce
re
sult in the ca
se that
the evolution ge
ne
ration is g
r
ad
ually increa
sed,
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
A Self-adapti
v
e Multipe
a
k
Artificial Immune Ge
netic
Algorithm
(Qi
ngzhao Li
)
653
and elem
ents in the memory warehou
se
eventually
converg
e
s to
each pea
k value. The rea
s
o
n
lies in that in
the process
of search, S
M
IGA
can ch
ange the le
n
g
th of the memory cells
and
store th
ose n
e
w antib
odie
s
non
-si
m
ilar to its orig
in
al antibodi
es,
thus ma
king
such algo
rithm
more
effectively find total extreme valu
es. Ba
sed o
n
the evaluatio
n and
analysi
s
in
su
ch term
s
as the functio
n
conve
r
ge
nce spe
ed, abili
ty to
search t
he extreme value and the
robu
stne
ss, with
better robu
st
ness, the
alg
o
rithm p
e
rfo
r
mance in
thi
s
pape
r
sho
w
s larg
er a
d
vant
age to
better
find
each pe
ak v
a
lue of Multi
pea
k fun
c
tion
, thus c
an b
e
su
ccessfully
applie
d to M
u
ltipea
k fun
c
tion
optimizatio
n probl
em
s.
G=100
G=200
G=300
Figure 6. Re
sults of SMIGA algorithm
6. Conclusio
n
No
wad
a
ys, i
n
telligen
ce
a
l
gorithm
s
ha
ve bee
n u
s
e
d
mo
re
and
more exten
s
ively in
nume
r
ou
s fiel
ds
of scientifi
c
resea
r
ch a
nd en
gine
eri
ng p
r
a
c
tice
s
and a
n
in
cre
a
sin
g
nu
mbe
r
of
peopl
e be
gin
to pay atten
t
ion to them.
By referring
to the immu
ne matu
ratio
n
me
chani
sm
of
biology immu
ne system a
nd int
egratin
g genetic al
g
o
rithm, this p
aper h
a
s p
r
o
posed an sel
f
-
adaptive m
u
lti-pea
k imm
une g
eneti
c
algo
rithm,
whi
c
h
signifi
cantly en
ha
nce
s
the
gl
obal
conve
r
ge
nce
of the al
go
rithm a
nd th
e a
c
cura
cy
of the glo
bal
extremum
by adju
s
ting
the
cro
s
sove
r an
d mutation p
r
obability and
dynamically gene
rating v
a
ccine. Th
e
perfo
rman
ce
test
function h
a
s
sho
w
n that th
is algo
rithm h
a
s a
c
hiev
e
d
better effect
s
in solving m
u
l
t
i-pea
k fun
c
tion
optimizatio
n probl
em
s.
Ackn
o
w
l
e
dg
ements
This work wa
s su
ppo
rted b
y
the Universi
ty
Natural Sci
ence Proje
c
t of Anhui Province
(Grant No. KJ2014Z
D3
1).
Referen
ces
[1]
Xu
eso
ng Y
an,
Qing
hua
W
u
, Can
Z
h
a
ng, e
t
al. An
Impr
o
v
ed Ge
netic A
l
gorithm
an
d Its Appl
icati
on.
T
E
LKOMNIKA Indon
esi
an Jou
r
nal of Electric
al Eng
i
ne
eri
n
g
.
2012; 1
0
(5): 1
081-
108
6.
[2]
Mlun
gisi D
u
ma
,
T
s
hilidzi Mar
w
a
l
a, Bh
ekisi
p
ho T
w
al
a, Fulufhel
o Nel
w
a
m
ond
o.
Partial Imputatio
n of
Unse
en R
e
cor
d
s to Improve
Classific
a
tio
n
u
s
ing
a H
y
br
id
Multi-la
ye
re
d A
r
tificial Immun
e
S
y
stem a
n
d
Genetic Alg
o
rit
h
m.
Appli
ed So
ft Comp
uting
. 2
013; 13(
12): 44
61-4
480.
[3]
Soed
ib
yo So
ed
ib
yo, Heri Sur
y
oatmoj
o
,
Imam
Ro
ban
di,
et a
l
. Optimal
D
e
sig
n
of
H
y
dr
oge
n
Based
W
i
n
d
Microh
ydr
o
usi
ng Gen
e
tic Al
gorithm.
T
E
LK
OMNIKA Indo
nesi
an Jo
urna
l
of Electrical
Engi
neer
in
g
.
201
2; 10(6): 13
09-1
318.
[4]
VS Aragó
n, S
C
Esqu
ivel, C
A
Coe
llo. An
Immune
A
l
gor
ithm
w
i
th P
o
w
e
r Redistri
buti
o
n for Solv
in
g
Econom
ic Disp
a
tch Probl
ems.
Informati
on Sc
ienc
es
. 201
5; 295(2
0
): 609-
63
2.
[5]
Ketan Di
nkar
Sarod
e
, V Ra
vi Kumar, BD
Kulk
ar
ni. Emb
edd
ed Mu
ltipl
e
Shooti
ng Met
hod
olo
g
y
in
a
Genetic Alg
o
rit
h
m F
r
ame
w
ork
for Parameter
Estimati
on a
n
d
State Identific
ati
on of C
o
mpl
e
x S
y
stems.
Che
m
ic
al En
gi
neer
ing Sci
enc
e
. 2015; 1
34(2
9
): 605-6
18.
[6]
Karthik Ba
lasu
brama
n
ia
n, Ba
sil Jac
ob, K Pri
y
a, K
S
a
n
geet
ha, N R
a
jas
e
k
a
r, et
al. Critic
a
l
Eval
uatio
n of
Genetic Alg
o
rit
h
m Based F
u
e
l
Cell Par
a
mete
r Extractio
n
.
Energy Proc
edi
a
. 2015; 75(
8): 1975-
198
2.
[7]
P Victer Paul, N Moganar
an
gan, S. Samp
ath Kumar,
et al. Performanc
e Anal
ysis ov
e
r
Popul
ation
Seed
ing T
e
ch
niq
ues of the
Permutatio
n
-C
ode
d G
enetic
Algorit
hm: An empiric
a
l Stud
y Bas
ed o
n
T
r
aveling Sal
e
sman Prob
lem
s
.
Applie
d Soft Co
mp
uting
. 2
0
15; 32(7): 3
83-
402.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 14, No. 2, June 20
16 : 647 – 65
4
654
[8]
Mahmo
ud M
El-Sher
bin
y
, R
a
shi
d
M Al
ha
mali. A H
y
brid Particle S
w
arm Algo
rithm w
i
t
h
Artificial
Immune Le
arn
i
ng for Solvi
n
g the F
i
xe
d C
harg
e
T
r
ansportation Prob
le
m.
Comp
uters
& Industrial
Engi
neer
in
g
. 2013; 64(
2): 610
-620.
[9]
F
engq
ian
g
Z
h
ao, Guan
gq
ia
ng L
i
, Ch
ao
Yang,
Aj
ith A
b
rah
a
m, Hon
g
bo L
i
u. A H
u
man–c
omputer
Coo
perativ
e Particle S
w
arm
Optimization
Based Immu
ne Alg
o
rithm
for La
yout D
e
sig
n
.
Neur
o
Co
mp
uting
. 2
0
14; 132(
20): 68
-78.
[10]
Poon
am Savs
ani, R
L
Jh
al
a, Vimal
Savsa
n
i
. Effect
of h
y
b
r
idizi
ng B
i
o
geo
grap
h
y
-Base
d
Optimizatio
n
(BBO)
T
e
chniq
ue
w
i
th Artifici
al Immune Al
g
o
ri
thm (AIA) and Ant Col
o
n
y
Optimizatio
n
(ACO).
Applie
d
Soft Computin
g
. 2014; 2
1
(8): 542-
553.
[11]
Se
yed
Abb
a
s
T
aher, Muham
mad K
a
rim A
m
oosh
ahi.
Ne
w
ap
pro
a
ch fo
r Optimal
UPF
C
Plac
eme
n
t
Using
H
y
bri
d
Immune Al
gorit
hm in El
ectric Po
w
e
r S
y
stem
s.
Internation
a
l
Journa
l of Ele
c
trical Pow
e
r &
Energy Syste
m
s
. 2012; 43(
1): 899-
909.
Evaluation Warning : The document was created with Spire.PDF for Python.