TELKOM
NIKA Indonesia
n
Journal of
Electrical En
gineering
Vol.12, No.1, Jan
uary 20
14
, pp. 818 ~ 8
2
2
DOI: http://dx.doi.org/10.11591/telkomni
ka.v12i1.3342
818
Re
cei
v
ed Ma
y 27, 201
3; Revi
sed Aug
u
st 10, 2013; Accepted Aug
u
s
t 31, 2013
Study Oftest Data Generation Method Based on
Evolutionary Algorithm
NI Jian, Zha
ng Yunlong*
Heb
e
i Un
iversit
y
of Eng
i
ne
eri
n
g, School of Inf
o
rmatio
n
& Ele
c
tronic Eng
i
ne
erin
g
heb
ei h
and
an,
056
03
8
*Corres
p
o
ndi
n
g
author, em
ail
:
yun
l
on
gh
and
an@
163.com
A
b
st
r
a
ct
T
he study of automatica
lly ge
nerate test dat
a base
d
on ev
ol
uti
onary a
l
g
o
r
ithm
meth
od focuses
o
n
the path cov
e
rage d
i
rectio
n. T
he key
prob
le
m is how
to construct a suit
a
b
le a
nd has
a goo
d orie
ntatio
n o
f
the fitness fun
c
tion to eva
l
u
a
te the qu
al
ity of a te
st data. T
h
is paper i
n
troduc
es
sev
e
ral ev
ol
ution
tes
t
meth
ods, s
h
o
w
s its adva
n
t
ages
a
nd
pro
b
le
msit
can
e
ffectively so
lv
e, an
d
prop
o
s
esan
i
m
pr
ov
ed
evol
ution
a
ry te
st data gen
erat
ion
meth
od,
w
h
ich can g
e
n
e
ra
te better test data.
Ke
y
w
ords
:
aut
omatic test data gen
erati
on,
e
v
oluti
on al
gor
ithm, fitness, fun
c
tion structural
test
Copy
right
©
2014 In
stitu
t
e o
f
Ad
van
ced
En
g
i
n
eerin
g and
Scien
ce. All
rig
h
t
s reser
ve
d
.
1. Introduc
tion
Software te
st
ing is to
find
defect
s
in t
he so
ftwa
r
e,it is o
ne of th
e effective m
ean
s to
ensure th
e software
qualit
y, and is an i
ndispen
sabl
e
comp
one
nt of the softwa
r
e devel
opm
ent
life cycle. A
test prog
ram
is good or
bad dep
end
s
on the sele
ction of test ca
se
s. Howe
ver,
softwa
r
e te
sti
ng is
a
com
p
licate
d
process. It
throug
hout the
stag
es of
software
develo
p
me
nt
pro
c
e
ss, an
d requi
re
s a lot of manpowe
r
, material
re
sou
r
ces a
nd time. Usu
a
lly,
test automati
o
n
techn
o
logy
can p
r
ovide
a
method
to g
enerate
te
st
ca
se
s
with hi
gh q
uality, and imp
r
ove t
he
efficien
cy of testing.
Re
sea
r
ch
a
kind
of
effectiv
e test
automation
te
chn
o
logy, to redu
ce the
co
st
of softwa
r
e te
sting ha
s very important si
gnifica
nce.
Evolutionary
algorith
m
ha
s be
en
appli
ed to mai
n
f
i
elds,
su
ch
a
s
conting
ent
power
netwo
rk
[1]. Evolutionary Testing
m
a
k
e the
test
ca
se
s ge
neration
pro
c
e
ss i
n
to
method
s of u
s
ing
geneti
c
alg
o
rithm for n
u
m
eri
c
al o
p
timization
pr
o
b
lems,
usi
n
g
the fitness function
a
s
test
obje
c
tives, m
appe
d te
st sp
ace
to the
se
arch
sp
ace
of
the al
go
rith
m, gen
erate
test
ca
se
s
whi
c
h
meet the target of testing
aut
omatically and efficient
ly. The w
hol
e pro
c
e
ss
wi
thout too mu
ch
human i
n
terv
ention, thu
s
usin
g evoluti
onary te
stin
g
techn
o
logy t
o
gen
erate t
e
st cases
greatly
redu
ce
s t
he t
e
st
co
st
.
Evolutionary
testing mai
n
l
y
include
s st
ructural
evolut
ion testing, f
unctio
nal test
ing and
perfo
rman
ce
testing, et
c.
At pre
s
ent th
e mo
st wi
del
y studie
d
of
evolutiona
ry testing
techno
logy
is
stru
ctural e
v
olution te
st tech
nol
o
g
y. Structu
r
al
evolu
t
ion test
ma
ke test ta
rg
et i
n
to the
cove
r
of
the pro
g
ram
stru
cture, su
ch as
state
m
e
n
t/bran
c
h
cov
e
rag
e
, path coverag
e
, etc.
[2], the mainly
use
d
fitness constructo
r are
di
st
an
ce o
r
iented m
e
tho
d
, so
rting
co
ver o
r
iented
method
and
the
control cove
r oriented m
e
thod [3]. Path orient
e
d
test data g
eneration is
one of the main
method
s of
test d
a
ta g
e
n
e
ration
ba
se
d
on th
e
stru
ct
ural te
st. Th
e
r
e a
r
e
a l
o
t o
f
re
sea
r
che
s
for
path ori
ented
method, e
s
peci
a
lly on e
v
olutionar
y a
l
gorithm to
g
enerate test
data got g
r
e
a
t
r
e
sear
ch pr
ogr
e
s
s
.
Evolution test
has th
e adv
antage
s of ef
ficien
t, dyna
mic an
d st
ron
g
guid
a
n
c
e; the key is
to design a reasona
ble fitness func
tio
n
.
If the fitness function is
n
o
t suitable, test efficiency
may
be lower tha
n
ran
dom te
sting. Fo
r different a
ppli
c
a
t
ions o
r
test
obje
c
ts, fitne
ss fu
nction
s
are
different. And in order to
get the bes
t effec
t,fi
tnes
s
func
tions
need to
be c
o
nstantly adjus
ted
according to t
he processin
g
of
evolution
a
ry testing.Th
ere a
r
e tw
o main problem
s existing i
n
the
curre
n
t evolutionary testin
g: the first one is
invalid
solution p
r
o
b
lem, whi
c
h
can't dete
r
mi
ne
wheth
e
r the
solutio
n
is u
s
ed in th
e e
ffective
sea
r
ch do
main; t
he othe
r is t
he problem
of
popul
ation d
e
g
rad
a
tion.Th
ere
are only
a
few
gen
otypes
i
n
the
pop
ulation,
which
dire
cts the l
o
cal
optimal soluti
on can al
way
s
survive [4]. The mai
n
ly
re
aso
n
is th
e fitness fun
c
tion
only acco
rdi
ng
to wh
ethe
r it
meet the
test
co
ndition
s
or is
clo
s
e
to t
he o
p
timal j
u
dgment,
rath
er th
an to
jud
ge
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 2302-4
046
TELKOM
NIKA
TELKOM
NIKA
Vol. 12, No
. 1, Janua
ry 2014: 818 – 8
2
2
819
wheth
e
r th
e
solutio
n
i
s
in
the effective
sea
r
ch
d
o
mai
n
, wh
ether it
lead
s to th
e f
a
ctors
su
ch
as
degradatio
n o
f
population o
r
not.
In addition, th
e evolution
a
ry testing al
go
rithms
mo
stly use
covera
g
e
[5] the indi
cators a
s
the feedb
ack
informatio
n of
testing p
r
o
c
e
dure
s
, te
st
executio
n is
not
con
s
id
ere
d
i
n
co
st, there i
s
no gu
ara
n
te
e that you g
e
t the test p
r
og
ram
set
o
f
execution
efficien
cy. For a
software, if
compl
e
tes all
the
req
u
ired
validation
works ne
ed
a l
o
t of te
sting
pro
c
ed
ures,
and
ho
wever th
e
test executio
n efficien
cy i
s
lo
w, thu
s
can g
r
eatly le
ngthen th
e te
st time, and
l
e
ssen
efficie
n
cy.
And in th
e
software
devel
opment
cycl
e
,
need
for
re
gre
ssi
on te
sti
ng for many
times, repe
ated
low exe
c
utio
n efficien
cy o
f
testing p
r
o
c
ess, w
ill
se
ri
ously affe
ct the p
r
oje
c
t schedul
e. To
solve
these
pro
b
le
ms, this p
a
p
e
r imp
r
oved
the adapt
ive
fitness fu
nction of an in
tegrated
de
si
gn
approa
ch, bo
th to ensure coverage, im
prove the
ex
ecutio
n effici
ency, and
ca
n effectively avoid
popul
ation de
grad
ation an
d
local optimal
solutio
n
of the probl
em.
2. Impro
v
ing
the Fitne
ss
Function
For invalid
solution a
nd popul
ation deg
rad
a
tion pro
b
lem
s
, resea
r
ch
ers p
u
t
forwa
r
d
ano
ptimization
met
hod ba
sed
on
p
uni
shme
n
tfunction
of
evolutiona
ry
testing [6], t
h
is
method ca
n
puni
sh po
pulation
i
ndi
viduals wh
i
c
h areout
side
ofthe sea
r
ch
domai
n, and
redu
ce
sthe
fitness value
of local
opti
m
al solu
tion,
make mo
re
solutio
n
s in
volve in the
next
iteration ofev
olution proce
ss.
Th
e fitness functio
n
is d
e
fined a
s
sh
o
w
n in form
ula
(1):
y
x
f
x
θ
g
x
θ
h
x
,
θ
,
θ
∈
0,
∞
(1)
In formula (1
), f (x) is the original d
e
finiti
on of fitness fun
c
tion, g
(
x) is the pu
n
i
shme
nt
function to d
eal with inva
lid solution
s,
θ
1
is invalid solution pu
nish pa
ram
e
ter, h(x) is the
puni
shme
nt functio
n
to de
al with lo
cal
optimal
soluti
ons,
θ
2
i
s
the
local
optimal
solutio
npu
ni
sh
para
m
eter.
θ
1
and
θ
2
is consta
nt which sco
pe is [0, +
∞
), th
eir values d
e
termin
es th
e
corre
s
p
ondin
g
penaltie
s
, and is ne
ede
d to be deter
mined befo
r
e
the test started. For inval
i
d
solutio
n
s a
n
d
local o
p
timal
solution
s, u
s
ing st
ati
c
met
hod an
d dyn
a
mic p
uni
sh
ment metho
d
to
dopu
nishmen
t, becau
se f
o
r invalid
so
lutions,effe
ctive input do
main of solution ha
s b
een
determi
ned
be
fore th
e
start
of the te
st, to
judg
e
wheth
e
r th
ey a
r
e v
a
lid o
n
ly
nee
d
to
acco
rdin
g
to
distan
ce of th
e solutio
n
to the effective input fi
elds. T
he greater of
the distan
ce,
the large
r
ofthe
puni
shme
nt functio
n
’s valu
e, for efficient
solutio
n
s, th
e puni
shm
ent
function
doe
s not
work. T
he
puni
shme
nt for local optimal solutio
n
s is requi
red
t
o
according t
o
the individu
al’s propo
rtio
n o
f
the pop
ulatio
n to dynami
c
ally adju
s
tme
n
t. Usu
a
lly
u
s
e sim
u
lated
a
nneali
ng p
uni
shme
nt fun
c
tion
or ad
aptive p
unishment fu
nction t
w
o
wa
ys to puni
sh
l
o
cal
optimal
solution. The l
onge
r ofthe ti
me
individual a
s
l
o
cal
optimal
solutio
n
, the l
a
rge
r
of
the
p
unishment fu
nction’
s valu
e
.
So, the fitness
function
can
dynamic a
d
j
ustment of the individu
al fitness valu
e acco
rding
to whethe
r the
individual i
s
the efficient solution or l
o
cal optimal sol
u
tion, avoidin
v
alid solutio
n
and deg
ra
da
tion
occurre
d
.
For evolution
a
ry
al
gorithm
usi
ng cove
rage
i
ndi
cators a
s
judg
me
nt of te
st progra
m
’
s
optimizatio
n goal, and lea
d
to the probl
em of low
efficien
cy, resea
r
ch
er p
r
op
osed test pro
g
ram
gene
ration
m
e
thod b
a
sed
on multiple
target optim
ization
s
[7]. The meth
od
con
s
i
s
ts of t
w
o
obje
c
tive functions: the first
one i
s
to a
c
h
i
eve te
st p
r
o
c
edure’s l
a
rge
s
t cove
ra
ge; the second i
s
to
redu
ce
test
e
x
ecution
cost
to a
minimu
m.In te
st p
r
o
g
ram
ge
neration p
r
obl
em
s, for
cove
rag
e
is
the main indi
catorfo
r
a
s
se
ssi
ng ho
w th
e test pro
g
ra
m is. Ho
weve
r, for a test p
r
og
ram with v
e
ry
low
cov
e
rag
e
,
while it
s ex
e
c
ut
ion
co
st
i
s
v
e
ry
sm
a
ll, e
x
ecution
effici
ency i
s
hi
gh,
but it is
not the
solutio
n
we n
eed. If not remove those
solutio
n
s
,they
will always
at the top for they have high
executio
n efficien
cy. So we need to
set
punishme
nt
function to eli
m
inate the solution of sm
all
coverage.
We
can u
s
e form
ula (2
) to determin
e
wh
eth
e
r a sol
u
tion
sho
u
ld be eli
m
inated:
y
x
100%
(2)
In formula (2), f
ma
x
and f
min
is the
maxim
u
m an
d mi
ni
mum valu
e of
cove
rag
e
in
d
i
viduals
allow in a
p
opulatio
n. Th
e form
ula
re
flects th
e p
o
pulation
indiv
i
dual
s’ deviat
i
on d
egree
of
minimum coverage rate relative to the ma
ximum value.If the deviation degree exc
e
eds
t
h
e
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 2302-4
046
Study oftest d
a
ta gene
ratio
n
m
e
thod based on e
v
oluti
onary alg
o
rith
m
(Zhang Yu
nlong
)
820
threshold
wh
ich i
s
set at
the
begin
n
i
ng of
test, t
he in
dividual
s
will b
e
reg
a
rde
d
as inv
a
lid
solutio
n
, and
can b
e
elimin
ated.
In addition, t
hetraditio
nal
evolutiona
ry test
data
ge
n
e
ration
meth
od exist
s
p
r
o
b
lems of
algorith
m
co
mplex, and coding difficult
ies and p
a
ra
meters are n
o
t easy to se
t up and so
on,
some
schol
ars put fo
rwa
r
d
a kin
d
ofsoftware
test d
a
ta automate
d
gene
ration
m
e
thod b
a
sed
on
differential ev
olution alg
o
rit
h
m [8]. Operator to
the o
peratio
n of the algo
rithm
was im
prov
ed,
solve
s
the problem of test
data
to generate di
scret
e
differential
evolution alg
o
rithm. The b
a
si
c
idea of
differential evoluti
on alg
o
rithm
is firstgen
erat
e an i
n
itial po
pulation i
n
th
e se
arch
sp
a
c
e,
then produ
ce ne
w indivi
dual by
weig
hting any tw
o individual
s’
vector diffe
rential a
nd t
hen
sup
e
rim
p
o
s
e
a third indivi
dual. The
n
compa
r
e the fi
tness value
of the new in
dividual an
d the
individual, if the new in
dividual’
s
fitness
value is
bette
r than that of the pare
n
t individual’s fitn
ess
value, then
repla
c
e
pare
n
t individual
with the
n
e
w in
dividual
, or pa
rent i
ndividual i
s
still
pre
s
e
r
ved. By iterative calcul
ation un
cea
s
in
gly, keep excellen
t
individuals surviving,
and
eliminate inf
e
rio
r
individu
als, thu
s
the
sea
r
ch p
r
o
c
ess was
gui
ded to a
ppro
a
ch th
e opti
m
al
s
o
lution.
Acco
rdi
ng to
the mention
e
d
evolutiona
ry te
sting tech
nology imp
r
o
v
ement idea,
we put
forwa
r
d th
e followin
g
met
hod to
con
s
truct of fitness
function: Fi
rst
of all, in ord
e
r
to mea
s
u
r
e
the
quality of the test program
, in addition to using
co
verage indi
cato
rs, at the same time also u
s
e
perfo
rman
ce
indicators an
d dista
n
ce in
dicato
rs.
Exe
c
ution
efficie
n
cy indi
cato
rs can g
uaran
tee
test sp
eed of
test prog
ram
in large
soft
ware, and g
u
a
rante
e
the test ca
se
s to
cover th
e cu
rrent
node o
r
state
m
ents to the
test set go
als at the
begin
n
ing of dista
n
c
e info
rmatio
n, and judg
e
the
curre
n
t nod
e’
s o
r
a
state
m
ent’s
dista
n
ce
inform
ation
with test
goal
while
covering
the t
e
st
ca
se.Th
r
ee
i
ndicators
evaluatea
test
prog
ram
at t
he
sam
e
tim
e
, whi
c
h
can
assu
re
the
high
coverage
rate of
evolutio
nary te
sting.I
n
ad
diti
on, in
order to
sol
v
e the
pro
b
le
m of p
opul
ation
degradatio
n and lo
cal opti
m
al solutio
n
, the introdu
cti
on of puni
sh
ment functio
n
and the fitness
function to
in
fluence po
pu
lation evoluti
on test
ca
se
s a
r
e eli
m
ina
t
ed. The fitn
ess fun
c
tion
as
sho
w
n in formula (3
):
y
x
f
x
θ
e
x
θ
d
x
θ
g
x
θ
h
x
θ
,
θ
,
θ
,
θ
∈
0
,
∞
(3)
The f(x) is th
e origin
al fitness functio
n
, e(x)
is the ex
ecutio
n effici
ency of the d
e
ci
sion
function, d
(
x) is the test
case
s to
cover the di
stan
ce
with the targ
et node fun
c
t
i
ons, g
(
x) is t
he
puni
shme
nt for invali
d
sol
u
tion fun
c
tion
, h(x) i
s
th
e lo
cal
optimal
solution
of the
penalty fun
c
ti
on,
θ
1
,
θ
2
,
θ
3
,
θ
4
is the fou
r
rel
e
vant param
et
ers
of the functio
n
s, the
y
ar
e setu
p b
e
fore the
start of
the test. The value of the four parame
t
ers det
e
r
min
e
s thefun
ctio
ns’ influen
ce
degre
e
oftest
ca
se
s.
3. Experimental Verifica
tion
In orde
r to ve
rify the validity of the desi
gn
of
fitness function,
the triangle pro
g
ram
test
experim
ent was
ca
rrie
d
o
u
t. Take
the th
ree si
de
s of th
e trian
g
le a,
b
,
c a
s
for the i
nput item
s, th
e
test goal i
s
to
achieve th
re
e side
s e
qual
node
s, t
hat is a
=
b
=
c.
Rep
r
ese
n
tation for the evolution
of
popul
ation in
dividual
s is (a, b, c). The
experim
ent
o
f
the total number of evol
ution T=500,
the
size of the po
pulation
size P=200, the la
rge
s
t numb
e
r of iterations I
=
20. The in
p
u
t fields for th
e
input item a,
b, c i
s
[30, 5
0
]. For the
efficien
cy
of de
cisi
on fun
c
tio
n
e(x) u
s
ing
comp
uter
clo
ck
cycle
s
to
det
ermin
e
, the compute
r
's m
a
in fre
quen
cy
for exe
c
utin
g the te
st progra
m
is1.7G
Hz.
Therefore
ex
1
.
7
T
I
/1.7
, distan
ce functi
on
dx
m
in |30
x|,
|
50
x
|
, punishm
ent
function of in
effective solut
i
on
gx
1
/
0.
001
1
/dx
, punishm
ent function o
f
local optima
l
solut
i
o
n
hx
e
xp 2/ T
P
,
θ
1
,
θ
2
,
θ
3
,
θ
4
are
set to 0.5, 1,
0.5, 30. So th
e fitness fun
c
tion
y(x) as sho
w
n in the formu
l
a (5):
y
x
f
x
0
.
5
1
.
7
T
I
1.7
m
i
n
|
30
x
|
,
|
50
x
|
0
.
5
1/0
.
001
1
/ dx
3
0
e
xp 2/ T
P
(5)
In experim
en
t, each
different si
ze
of the
po
pulatio
n wa
s
se
parately run fo
r
10 time
s,
record ea
ch
time to find the optimal
solution of the
iteration n
u
mber
and
ru
nning time, a
nd
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 2302-4
046
TELKOM
NIKA
TELKOM
NIKA
Vol. 12, No
. 1, Janua
ry 2014: 818 – 8
2
2
821
cal
c
ulate th
e
avera
ge ite
r
ativ
e count
and ite
r
ation
time. Figure 1 an
d fig
u
re
2 sho
w
the
experim
ental results comp
arison
to
f(x) and
y(x).
From th
e exp
e
rime
ntal re
sults of figure
1 and fi
g
u
re
2
,
it can be
se
en that after
usin
g the
adaptive valu
e fun
c
tion y(x
)
, the tri
angle
prog
ram
hav
e obviou
s
i
m
prove
both ite
r
ation
co
unt a
nd
iteration time,
whi
c
h in
dica
tes after
add
ed the p
uni
shment fun
c
tio
n
and
efficien
cy functio
n
, y(x)
is ve
ry efficie
n
t to imp
r
ove
the
cove
rag
e
an
d p
uni
sh
ment ineffe
cti
v
e sol
u
tion
s
and l
o
cal o
p
timal
s
o
lution.
Figure 1. Gen
e
rate an e
quil
a
teral tria
ngle
of the iteration cou
n
t
Figure 2. Gen
e
rate an e
quil
a
teral tria
ngle
of the iteration time
4. Summar
y
In this pap
er,
the method o
f
adopting a
d
d
s a
se
ri
es
of control mea
n
s
on the b
a
si
s of the
origin
al fitness functio
n
to solve the p
r
o
b
lems
of evolutionary te
sting. And su
pe
rvise the p
r
o
c
ess
of populatio
n
evolution, b
o
th retain
ed the ev
olution
a
r
y test cove
rage, and
ca
n
guarantee t
e
s
t
prog
ram’
s ex
ecutio
n effici
ency, a
n
d
can al
so
avoi
d de
gradatio
n an
d i
n
valid
sol
u
tion
in t
he
pro
c
e
ss of po
pulation evol
ution, optimize t
he quality and efficie
n
cy
of test data generation.
100
12
0
140
16
0
180
20
0
0
2
4
6
8
10
12
14
16
18
Po
p
u
la
tio
n
S
i
z
e
I
t
e
r
a
t
e
C
ount
C
o
mmon F
i
tne
s
s
F
u
n
c
t
i
on
O
p
ti
mi
z
e
d
F
i
t
n
e
s
s
F
unc
ti
on
10
0
12
0
14
0
16
0
18
0
20
0
6
8
10
12
14
16
18
P
o
p
u
la
tio
n
S
i
z
e
I
t
er
at
e
T
i
m
e
(
m
s
)
C
o
m
m
o
n
Fi
t
n
e
s
s
Fu
n
c
t
i
o
n
Op
t
i
m
i
zed
F
i
t
n
es
s
F
u
n
c
t
i
o
n
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 2302-4
046
Study oftest d
a
ta gene
ratio
n
m
e
thod based on e
v
oluti
onary alg
o
rith
m
(Zhang Yu
nlong
)
822
Referen
ces
[1]
Sa
w
a
n Sen, Pri
y
ank
a Ro
y, Abhi
jit Chakr
a
varti,
Samarjit
Sengu
pta. Genet
ator C
ontr
i
buti
on Base
d
Con
gestio
n
Ma
nag
ement us
in
g Mult
io
bjectiv
e
Genetic Al
go
rithm.
T
E
LKOMNIKA Indon
e
s
ian Jo
urn
a
l of
Electrical E
ngi
neer
ing
. 2
011;
9(1): 1-8.
[2]
Sand
ip C
h
a
n
da a
nd Ab
hi
nan
da
n De.
Con
gestio
n
R
e
lief
of Conti
nge
nt Po
w
e
r
Net
w
ork
w
i
t
h
Evoluti
onar
y Optimizatio
n
Algorit
hm.
T
E
LKOMNIKA Indones
ian Jo
urn
a
l of
Electrical
Engin
eeri
n
g
.
201
2; 10(1): 1-
8.
[3]
XIE
Xia
o
y
u
an.
Surve
y
of E
v
oluti
onar
y T
e
sting.
Jour
nal
of F
r
ontiers
of
Co
mp
uter
Scienc
e an
d
T
e
chno
logy
. 2
008, 2(5):4
49-
466.
[4]
Shi Li
ang. R
e
s
earch o
n
T
e
st
Data Automati
c Generatio
n. Nanj
in
g: S
outh
east Univ
ersit
y
, 2007.
[5]
Z
hang Li
ang. Cover
age
M
a
t
r
ix bas
ed
Ev
o
l
utio
nar
y T
e
st Program Ge
nerati
on for M
i
croproc
esso
r
Verification.
Jo
urna
l of Co
mp
uter- Aide
d De
sign & Co
mput
er Graphics
. 2
011; 23(
3): 456
-462.
[6]
Z
hang
Na
n. R
e
searc
h
o
n
Ev
oluti
onar
y T
e
sting Optim
i
zatio
n
Base
d o
n
Pe
nalt
y
F
uncti
on.
Co
mp
uter &
D
i
g
i
t
al
En
gi
ne
eri
n
g
. 20
09; 37(
4): 1-3.
[7]
Z
hang
Li
an
g.
T
e
st Program
Generati
o
n
Ba
s
ed
on
Multi-
o
b
jectiv
e Evo
l
uti
onar
y
Alg
o
rith
m.
Journ
a
l of
F
r
ontiers of Co
mp
uter Scie
nc
e and T
e
ch
no
l
ogy
. 201
0; 22(
8): 1382-
13
89.
[8]
Hua
ng
Xia
o
ch
eng. App
licati
o
n of modifie
d
differenti
a
l evo
l
utio
n in
test data gen
erati
o
n
.
Journal o
f
Co
mp
uter Appl
icatio
ns
. 200
9; 29(6): 17
22-
17
54.
Evaluation Warning : The document was created with Spire.PDF for Python.