TELKOM
NIKA Indonesia
n
Journal of
Electrical En
gineering
Vol.12, No.4, April 201
4, pp. 2497 ~ 2
5
0
5
DOI: http://dx.doi.org/10.11591/telkomni
ka.v12i4.4730
2497
Re
cei
v
ed Se
ptem
ber 11, 2013; Revi
se
d Octob
e
r 28,
2013; Accept
ed No
vem
b
e
r
15, 2013
ANN Based Modeling and Optimization of Large
Pumped Storage Station
Hong
tao Ze
n
g
*, Linlin Lin
,
Zhixin Wan
g
, Wenga
ng Tian, Cong F
e
ng, Zhihuai
Xiao
Schoo
l of Po
w
e
r and Mec
h
a
n
i
cal En
gin
eeri
n
g, W
uhan Un
iv
ersit
y
, W
u
h
an, Chin
a
*Corres
p
o
ndi
n
g
author, e-ma
i
l
: htzeng@
w
h
u
.
edu.cn
Abs
t
rac
t
Modeling of regul
ating system
of
large pum
p
ed stor
age
power
station with
l
ong pipeline
and
para
m
eters o
p
timi
z
a
ti
on i
n
transie
nt pro
c
esses ar
e
st
udi
ed i
n
this
pap
er. Accord
ing to
the
ac
tual
para
m
eters of a certain p
u
m
p
ed storag
e po
w
e
r station, MAT
L
AB/Simuli
n
k
toolbox is uti
l
i
z
e
d for mod
e
li
n
g
and s
i
mul
a
tio
n
of its subsys
tems. F
r
ictio
n
resistanc
e
c
oefficie
n
t an
d
w
a
ter elastic
i
ty are take
n i
n
to
consideration in
modeling of
pressur
e
div
e
r
s
ion syste
。
As
to simu
late
h
y
drau
lic vibr
ati
on ch
aracters,
BP
neur
al
netw
o
rk and
RBF
n
eura
l
n
e
tw
ork are a
d
o
p
te
d
in
mo
de
lin
g
of pu
mp
turbi
ne. Bas
ed
on
th
e
establ
ishe
d re
gul
ating syst
e
m
si
mulati
on
mo
de
l, im
prov
ed
orth
og
ona
l exper
iment me
thod
is ap
pli
e
d
i
n
para
m
eters o
p
t
imi
z
at
io
n of n
o
-lo
ad freq
ue
n
cy disturb
anc
e
,
load distur
ba
nce
a
nd loa
d
shed
din
g
trans
ient
process
e
s. Ac
cordi
ng to
the
results, th
e p
r
opos
ed
mo
d
e
l
reflect t
he
a
c
tual c
haracter
i
stics of
pu
mp
e
d
storage
units,
and i
m
prov
e
d
ortho
gon
al
exper
iment
me
thod
is effec
t
ive in fig
u
ri
n
g
out the
opt
imal
para
m
eters gr
oup w
i
thi
n
the
give
n ran
ge. This pa
per
pr
ovi
des g
u
id
anc
e for mode
lin
g of
regu
latin
g
syste
m
of larg
e p
u
m
p
ed stora
ge
un
i
t
s, and set ref
e
renc
es a
nd t
heor
etical
bas
i
s
for actua
l
o
p
timal c
ontrol
of
transie
nt proce
sses in pu
mpe
d
storage u
n
its
.
Ke
y
w
ords
:
regulating system
, artific
i
al
neural network,
modeling and
sim
u
lation
, transient
proc
ess,
para
m
eter opti
m
i
z
at
io
n
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
Due
to its
ra
pid respon
din
g
, stro
ng l
o
a
d
-fo
llo
wing
capability an
d
reserve
capa
city, the
pumpe
d sto
r
age po
we
r st
ation ha
s pla
y
ed an irrepl
ac
e
able a
nd
signifi
cant rol
e
in the ele
c
t
r
ic
system. Th
e
developm
ent
of pumpe
d st
orag
e po
we
r
st
ation i
s
not
only relate
d to daily ele
c
tri
c
ity
con
s
um
ption,
industri
a
l produ
ction and
soci
al ac
tivities, but also to stable op
eration and lon
g
-
term plan of e
l
ectri
c
sy
stem
.
With the increasi
ng num
b
e
r, cap
a
city a
nd pro
p
o
r
tion
in state grid,
proble
m
s referred to
stability and
safety in pu
mped
st
orag
e power
stat
ion
are drawing incre
a
si
n
g
ly attention of
relevant
re
se
arche
r
s, th
us it is of g
r
e
a
t im
porta
nce t
o
stu
d
y on th
e sy
stem of
pumpe
d
storage
power
statio
n [1]. Simula
tion re
se
arch
on
pump
e
d
sto
r
age
po
wer statio
n i
s
a
b
le to
provide
effective tech
nical
sup
p
o
r
t to the control of
its norm
a
l ope
ration
and conditio
n
swit
ch [2], and
can
also red
u
ce th
e
co
sts and
risks of
variou
s te
sts
so a
s
to fin
d
out the featu
r
es of
pump
e
d
stora
ge u
n
its and imp
r
ove
their
safety, stability,
flexibility and e
c
o
nomical effici
ency via o
p
timal
control strate
gy. Therefo
r
e
,
it is quite essent
ial to st
udy the mod
e
ling, sim
u
lat
i
on and
opti
m
al
control of th
e re
gulatin
g
system
of p
u
mped
sto
r
a
ge po
we
r
station by n
u
m
eri
c
al
simul
a
tion
method
s.
In this pa
per,
MATLAB/Simulink i
s
u
s
e
d
to
model t
he re
gulatin
g
system of a
certai
n
pumpe
d
storage p
o
wer
st
ation in
Chin
a, on the
ba
sis of
whi
c
h p
a
ram
e
ters o
p
t
imization i
n
the
transitio
n p
r
o
c
e
s
ses
of no
-load f
r
eq
uen
cy dist
u
r
ba
nce, load di
stu
r
ban
ce
and l
oad
she
ddin
g
disturban
ce a
r
e sp
ecifi
c
ally
studied.
2.
Modeling of
Regula
t
ing Sy
stem of Pu
mped Storag
e Unit
The re
gulati
ng syste
m
of pumpe
d stora
ge
unit is similar to that of a common
hydrop
ower
station,
whi
c
h is a
com
p
lex no
n
lin
ea
r
c
l
os
ed
-
l
oo
p
c
o
n
t
r
o
l sys
te
m
inc
l
udin
g
hydrauli
c
, me
cha
n
ical and
power facto
r
s.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NIKA
Vol. 12, No
. 4, April 2014: 2497 – 25
05
2498
2.1.
Structure an
d Char
acter
istics o
f
Reg
u
lating Sy
st
em
Pumped
storage u
n
it con
s
ist of division
sy
stem, gove
r
no
r, pump
-
tu
rbine
and
ge
nerato
r
.
More specifically, the regulating
system compri
se
s four main
subsy
s
tems th
at are
p
r
e
s
sure
diversi
on sy
stem, electro
-
hydrauli
c
gov
erno
r,
pump turbine
a
nd motor-gen
era
t
or.The
st
ru
cture
of the system
is sho
w
n in F
i
gure 1.
Figure 1. Structure of
Regu
lati
ng System
of Pumped Storage
Unit
The pum
ped
stora
ge po
wer
station
studied in
this pape
r ha
s two natu
r
al rese
rvoi
r
basi
n
s
with th
e fall he
ad of
531 m
e
ters.
This
station
consi
s
ts of two
part
s
, i.e. A
and B, b
o
th
with
desi
gn h
ead
of 517.4
met
e
rs an
d rate
d flow
of 66.
2m
3
/s, and
i
n
stalle
d capa
city of 120
0
M
W
(4*3
00M
W).
2.2. Model of Div
e
rsion Sy
stem
Manifest
wat
e
r ham
mer ef
fect in the re
gulat
ing
syst
em of pumpe
d storage u
n
i
t
always
redu
ce
s the
output of pu
mp turbin
e to some ex
tent
. More impo
rt
antly, water
hamme
r effe
ct is
contrary to the accomm
od
ation of wicke
t
gate, wh
ich
seri
ou
sly affects the
regulatio
n
p
e
rform
of
units. Fo
r th
e modeli
ng
of diversi
on
system
in
p
u
mped
sto
r
a
ge po
we
r st
ation with lo
ng
pen
stocks, el
asticity of water, pipe
wall
and infl
ue
nce
of hydrauli
c
f
r
iction
sh
ould
be con
s
ide
r
e
d
.
Un
steady flo
w
in pen
sto
c
k can be d
e
scribed by the following t
w
o di
fferential equ
ations [3].
Momentum e
quation i
s
bel
ow,
2
(,
)
(
,
)
(,
)
2
Q
L
t
H
Lt
f
Q
Lt
gA
tL
D
A
(1)
Contin
uity equation is,
2
(,
)
(
,
)
QL
t
H
L
t
ag
A
Lt
(2)
W
h
er
e
Q
i
s
the flo
w
in
the
pipeli
ne,
H
i
s
the
hea
d of
water,
L
i
s
th
e len
g
th
of pi
peline,
A
is the
cro
s
s-se
ctional
a
r
e
a
of
pipelin
e
,
D
i
s
th
e
di
ameter of
pi
peline
an
d
f
is friction l
o
ss
c
oeffic
i
ent.
Then the tran
sfer fun
c
tion of long pen
stock in
pumpe
d stora
ge po
wer
station wi
th elasti
c
water hammer is
as
follows.
()
2(
)
()
()
2
hW
r
Hs
ht
h
Qs
T
Gs
s
(3)
Whe
r
e
Tr
is p
hase length o
f
water ham
m
e
r and
hw
is
coeffici
ent of pipeline feat
u
r
es.
The above fu
nction
can b
e
transfo
rmed i
n
to the functi
on belo
w
in a
c
tual mod
e
lin
g.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
ANN Ba
sed
Modelin
g and
Optim
i
zation of Large Pum
ped Stora
ge
Station (Ho
n
g
t
ao Zeng
)
2499
33
22
1
24
1
1
8
()
()
()
rr
W
r
h
Ts
T
s
h
Ts
Hs
Qs
Gs
(4)
With c
o
effic
i
ent of fric
tion tak
e
n into c
o
n
s
ide
r
ation, it is de
scribe
d a
s
belo
w
.
33
22
1
4
24
11
1
28
()
()
()
rr
W
rr
h
f
Ts
T
s
h
f
Ts
T
s
Hs
Qs
Gs
(5)
In the dive
rsion
system
of
pumpe
d
stora
ge p
o
wer stat
ion with
lo
ng pipeline,
surg
e
tan
k
s
are
often a
d
opted in
the
pipeline
ne
ar the pla
n
t
ho
use
to imp
r
o
v
e the sta
b
ili
ty of regul
ating
system [4]. The station
stu
d
ied in this
p
aper i
s
eq
uip
ped with two
throttled surg
e tanks in bo
th
uppe
r sid
e
an
d down sid
e
, and corre
s
po
nding tra
n
sfe
r
function is a
s
follows [3].
()
()
()
1
h
T
T
Tq
Ts
qs
Gs
hs
T
s
(6)
Whe
r
e
Th
i
s
head time co
nstant an
d
Tq
is
flow time cons
tant.
For di
scu
ssi
o
n
cla
r
ity, this pape
r is
on
the ba
sis of
a sin
g
le pe
nsto
ck-single
turbine
-
gene
rato
r at an isol
ated n
e
twork. The d
i
vers
io
n syste
m
in Plant B is sh
own belo
w
.
Figure 2. Dia
g
ram of Dive
rsion System
of
Plant B (single pe
nsto
ck-singl
e turbi
ne)
Acco
rdi
ng to
relevant d
a
ta
of this po
we
r
stat
ion, pa
ra
meters of the
diversi
on
system ca
n
be obtain
ed a
nd sh
own in Table 1.
Table 1. Para
meters of the Diversion System
Tr
hW
1 3.329
0.114
2, 3
1.820
0.115
4, 5, 6
0.279
0.996
Th
Tq
Upper sur
ge tank
443.3
171.828
Acco
rdi
ng to
the tran
sfe
r
f
unctio
n
s and
param
ete
r
s,
the sim
u
latio
n
mod
e
l of di
versio
n
system of Pla
n
t B in MATLAB/Simulink is sh
own belo
w
.
paramete
r
Pipeline
number
paramete
r
Surge
tank
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NIKA
Vol. 12, No
. 4, April 2014: 2497 – 25
05
2500
Figure 3. Simulation Mod
e
l
of Diversio
n System of Plant B
2.3. Model of Electro-hy
d
raulic Gov
e
r
nor
The po
we
r station in this paper
adopt
s a
PID gov
erno
r man
u
factured by NEYRPIC
Fren
ch. Acco
rding to the
given st
ructu
r
e block, the
simulatio
n
m
odel of the g
o
verno
r
is a
s
in
Figure 4.
Figure 4. Simulation Mod
e
l
of Electro-hydrauli
c
Gove
rnor in Plant B
As shown in
the figure, the
governin
g
syste
m
co
mbine
s
regul
ar pa
rall
el PID with
differential forward PID, by
whi
c
h the abil
i
ty of
overco
ming oversho
o
t is enhan
ce
d so as to rai
s
e
the cont
rol qu
ality.
2.4. ANN Ba
sed Model of Pump Turbine
As the flow in pump turbine can be quite
co
mpl
e
x, dynamic perform
ance is often
utilized
whe
n
a
nalyzi
ng transi
ent
pro
c
e
s
ses of
pump
ed
sto
r
age
po
wer st
ation. However, the
r
e
are
no
effective way
s
to acqui
re
relevant pa
ra
meters
until now, thu
s
dynamic p
e
rfo
r
mance is u
s
u
a
lly
repla
c
e
d
by steady-state
cha
r
a
c
teri
stics to
a
nalyze the dyna
mic processes. Steady-state
cha
r
a
c
teri
stics of
pum
p t
u
rbin
e
can
b
e
de
scri
bed
by
four-qu
a
d
r
ant com
p
let
e
cha
r
a
c
teri
stic
curve, yet th
e cu
rve is
with seri
ou
s o
v
erlap
se
ct
io
ns a
nd di
st
in
ct
“S
”
cha
r
a
c
t
e
ri
st
ic in
b
o
th
reverse
pum
p
zo
ne a
nd
breaki
ng
zon
e
,
whi
c
h al
way
s
re
sult in
co
n
s
ide
r
abl
e e
r
ro
r in i
n
terp
olati
o
n
cal
c
ulatio
n of the curve.
At pre
s
ent, th
ere
are mai
n
l
y
two meth
od
s a
pplied
in
modelin
g of
pump tu
rbin
e
.
One i
s
improve
d
SUTER curve m
e
thod, by
whi
c
h ove
r
lap
s
o
f
“S” zo
ne in
compl
e
te cha
r
acte
ri
stic
curve
can be
remo
ved thus re
d
u
cin
g
interp
ol
ation errors
, but the curve
conversi
on
p
r
oc
es
s
is
r
a
th
e
r
compli
cate
d [5, 6]. The other on
e is int
e
rnal
cha
r
a
c
t
e
risti
c
analy
s
i
s
method, wh
ich reli
es m
u
c
h
on stru
cture para
m
eters. Usi
ng
th
i
s
m
e
thod, a
ma
ss of
pa
ramet
e
rs in
pum
p t
u
rbin
e a
n
d
in
itial
boun
dary co
ndition
s are
need
ed, yet
whi
c
h are no
t easy to obtain. In consi
deratio
n of quite
stron
g
nonli
n
ear ap
proximation ability of artifici
al neu
ral network (ANN), this p
aper a
dopt
s two
mostly used
netwo
rks, i.e. BP
neural
netwo
rk
and
RBF neu
ral
netwo
rk, to
fit the nonlin
ear
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
ANN Ba
sed
Modelin
g and
Optim
i
zation of Large Pum
ped Stora
ge
Station (Ho
n
g
t
ao Zeng
)
2501
prop
ertie
s
of
pump turbine
accordi
ng to
the cu
rves o
f
unit discha
rge
Q11
-
un
it s
p
ee
d
n1
1
a
nd
unit torque
M
11-
unit spee
d
n11
.
Prope
rties of
pump turbi
n
e
can be expres
sed by five paramete
r
s, i.e. flow
Q
, to
rque
M
,
head
H
, rotat
i
onal
spe
ed
n
and
wi
cket gate op
enin
g
Y
, in whic
h
H
,
n
an
d
Y
are
in
dep
end
ent
while
Q
an
d
M
can
be p
r
e
s
ente
d
by the
s
e three. Hen
c
e, ANN ba
sed mod
e
ling
of pump tu
rbi
ne
is in fact mod
e
ling of a 3 in
puts- 2
output
s network.
Takin
g
BP n
eural
net
work for
exampl
e,
newff
fun
c
tion and
LM
algorith
m
a
r
e
utilized
whe
n
e
s
tabli
s
hin
g
the
B
P
network d
ue to it
s fa
st conve
r
gen
ce
and
smal
l trainin
g
e
r
ror
comp
ari
ng wi
th other trai
ni
ng algo
rithm
s
. By compari
ng the differe
nce b
e
twe
en
target value a
nd
simulatio
n
re
sult of diffe
re
nt numb
e
r
(i.e. N) of
hid
d
e
n
layer
neu
ro
ns, it can
be
dra
w
n th
at m
odel
of
Q11
is bett
e
r with
N
= 9
.
The error value curve of
training is
sh
own in Fig
u
re
5. The traini
ng
pro
c
e
ss of
M11
is
simil
a
r to that of
Q11
and its model is
better with
N
= 8, a
n
d
the
corre
s
p
ondin
g
error value
curve i
s
sh
own in Figure 6.
F
i
gure 5. Error
Valu
e Curve of
Q
11
F
i
gure 6. Error
Valu
e Curve of
M
11
As for mod
e
li
ng usi
ng RB
F neural net
work, ne
wrb
function i
s
ch
ose
n
to esta
blish the
netwo
rk
be
ca
use it wo
rks
better with la
rge n
u
mbe
r
of sampl
e
as in the ca
se i
n
this pap
er.
By
comp
ari
ng th
e differen
c
e
betwe
en targ
et value and
simulatio
n
re
sult of differe
nt spread val
ues
of radial
ba
si
s fun
c
tion, it
can be d
r
a
w
n t
hat model
of
Q11
i
s
better
with spre
ad v
a
lue
= 0.49 a
nd
model
of
M11
is bette
r
wi
th sp
re
ad val
ue
=
0.78. T
he e
r
ror valu
e curve
s
of
Q11
and
M11
ar
e
respe
c
tively shown in Figu
re 7 and Figu
re 8.
Figure 7. Erro
r Value Curve
of
Q11
Figure 8. Erro
r Value Curve
of
M11
Thro
ugh th
e
com
p
a
r
iso
n
of mod
e
ling
of pum
p tu
rbine
u
s
ing
BP and
RB
F neu
ral
netwo
rk, it ca
n be found th
at the differe
nce b
e
twe
en
target value
and si
mulatio
n
re
sult is mu
ch
smalle
r wh
en
using RRF
netwo
rk, so that it
can be
tter approach
the nonlinea
r prop
ertie
s
o
f
pump turbin
e. Therefo
r
e,
model
s of
Q11
an
d
M1
1
usi
ng RBF netwo
rk
are
sele
cted
for
simulatio
n
te
sts
of the
re
gulating
sy
stem of Pla
n
t
B. Simulation
mod
e
l of th
e pu
mp tu
rbi
ne i
s
s
h
ow
n
as
fo
llo
w
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NIKA
Vol. 12, No
. 4, April 2014: 2497 – 25
05
2502
Figure 9. RBF Network Ba
sed Simul
a
tio
n
Model of Pump Tu
rbine
2.5. Model of Motor-gene
r
a
tor
Gene
rally, dynamic
pro
p
e
r
ties of gen
erator
can be
descri
bed
by a first-o
r
de
r
pro
c
e
s
s
and its tran
sf
er functio
n
is
as bel
ow [7].
1
()
()
g
ab
n
TT
s
e
Gs
(7)
Whe
r
e,
22
36
5
r
r
a
GD
n
P
T
, named unit ine
r
tia time const
ant, Tb is loa
d
inertia time
con
s
tant,
and en i
s
un
it synthetic self-re
gulation
coeffici
e
n
t (usu
ally between 0 an
d 2.0). According
to
actual d
a
ta of Plant B,
(Ta+Tb
)
= 8.4
53.
2.6.
Model of Re
gulating Sy
s
t
em of Plant
B
Basing
on
th
e ab
ove m
o
dels of dive
rsion
sy
stem,
ele
c
tro
-
hyd
r
aulic gove
r
n
o
r, p
u
mp
turbine
an
d
motor-gen
era
t
or, the
simu
lation mo
del
of regulatin
g sy
stem
of
Plant B can
be
acq
u
ire
d
by conne
cting the
sub
s
ystem
s
and is
sho
w
n
in Figure 10.
Figure 10. Simulation Mo
d
e
l of Regul
ating System of Plant B
3.
Parameters
Optimiza
tion
of Transien
t Processe
s
3.1.
Common Tr
ansien
t Proc
esse
s of Pu
mped Storag
e Unit
Thre
e comm
on tran
sie
n
t pro
c
e
s
ses
of pum
pe
d sto
r
age regul
atin
g system
are
no-lo
ad
freque
ncy
disturban
ce, l
o
a
d
distu
r
b
ance
and lo
ad
sh
e
dding. F
o
r n
o
-
load
freq
uen
cy distu
r
b
a
n
c
e,
freque
ncy of
the units i
s
gi
ven and the t
a
sk of re
gula
t
ing system i
s
to avoid flu
c
tuation
ca
used
by the un
stea
dy flow. Fo
r l
oad di
stu
r
ba
n
c
e, a
s
p
u
mpe
d
sto
r
ag
e uni
ts ne
ed to tra
c
k the
cha
n
g
e
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
ANN Ba
sed
Modelin
g and
Optim
i
zation of Large Pum
ped Stora
ge
Station (Ho
n
g
t
ao Zeng
)
2503
of load
rapidl
y and
stea
dil
y
, the task
o
f
reg
u
la
ting
system i
s
to
keep th
e freq
uen
cy of u
n
its
steady. In load she
dding p
r
oce
s
s, spe
e
d
of the uni
ts raise
s
qui
ckly
while the wi
cket gate clo
s
i
ng
down. Due to
self
drag to
rque,
spe
ed
o
f
the u
n
it
s gradually de
clin
es and
fre
q
u
ency of
the
u
n
its
turns to b
e
at the given poi
nt.
3.2.
Impro
v
ed Orthogon
al Experiment Me
thod [8-1
1]
There a
r
e
th
ree
pa
ram
e
ters (
bt, Td
a
nd
Tn
) that
need
to b
e
optimize
d
in
transi
ent
pro
c
e
s
ses of
pumpe
d
storage u
n
its. T
a
ke
no-l
oad
f
r
eque
ncy di
st
urba
nce te
st
for exam
ple,
th
e
test is to figu
re out the op
timal grou
p o
f
thes
e thre
e
param
eters that can m
a
ke the reg
u
lati
ng
time
(Tp
)
a
s
short a
s
p
o
ssi
b
le an
d the o
v
ersh
oot
(M)
as small as p
o
ssible.
T
e
st can be
a
r
ran
ged
by orthogo
na
l experiment
table and extremum di
ffe
rence analysi
s
ca
n be use
d
to analyze
the
test re
sults [1
2]. Extremum difference an
alysis tabl
e is sho
w
n bel
ow.
Table 2. Extremum Differe
nce Analy
s
is
Table
Indicator
Fac
t
or
No.
A
T
d
b
t
T
n
I/3
A
11
A
21
A
31
II/3
A
12
A
22
A
32
III/3
A
13
A
23
A
33
Extremum
differe
nce
B
1
B
2
B
3
In the tabl
e,
A
ij
rep
r
e
s
ent
s the
statistica
l avera
ge
of
indicator
A
with the
i
th
fa
ctor of
j
level,
B
k
stands for the extremum differe
nce of
A
with
the
k
th
fac
t
or (i, j, k
=
1, 2, 3).
Traditio
nal orthogon
al exp
e
rime
nt meth
od ha
s
bee
n proved to b
e
effective in obtaining
the optimal
g
r
oup
of pa
ra
meters with
d
i
spe
r
sed
th
re
e given level
s
after
nine t
e
sts. T
o
gai
n
the
optimal group
within the given ran
ge of these
paramet
ers, the meth
od nee
ds to b
e
improve
d
.
Suppo
se that
factor X has three initial levels
{x, 2x,
3x}, if the optimal value of X is 2x
after the first roun
d of tests, then the
new l
e
ve
ls o
f
X in the ne
xt round of t
e
sts
sh
ould
be
sele
cted
as
{
1
.5x, 2x, 2.5x}. That is to
say, se
t the
optimal valu
e of the first
roun
d a
s
cen
t
ral
level of the
next rou
nd
o
f
test an
d ta
ke
hal
f of th
e differe
nce
betwe
en lev
e
ls
as the n
e
w
differen
c
e in
the next roun
d, then optimal grou
p
withi
n
a smalle
r range
can be
acq
u
ire
d
in the
next round of
test. Tests
co
uld be re
peat
ed unt
il the re
sult meet
s the requi
rem
e
n
t
s.
3.3. Parameters
Optimiza
tion
of
Differen
t
Transien
t Pr
ocess
es
Acco
rdi
ng to
relevant d
a
t
a, simulatio
n
te
sts
of different tra
n
s
ient p
r
o
c
e
s
se
s are
perfo
rmed a
n
d
the re
sults
are sho
w
n a
s
below.
0
10
20
30
40
50
60
70
80
90
10
0
-0.
0
1
0
0.
01
0.
02
0.
03
0.
04
0.
05
0.
06
0.
07
0.
08
ti
m
e
/s
r
e
l
a
t
i
v
e
f
r
equen
c
y
Figure 11. Compa
r
ison of Optimizatio
n
Re
sults of No
-load F
r
eq
ue
ncy Di
sturb
a
n
c
e Pro
c
e
ss
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NIKA
Vol. 12, No
. 4, April 2014: 2497 – 25
05
2504
0
10
20
30
40
50
60
70
-
0
.
025
-0
.
0
2
-
0
.
015
-0
.
0
1
-
0
.
005
0
0.
005
ti
m
e
/
s
r
e
l
a
t
i
v
e
f
r
equ
enc
y
0
10
20
30
40
50
60
70
-0
.
0
5
0
0.
05
0.
1
0.
15
0.
2
0.
25
ti
m
e
/s
r
e
lat
i
v
e
s
peed
F
i
gure 1
2
. Co
mpariso
n
of op
timizatio
n
resul
t
s of
+
10% lo
ad dist
urba
nce pr
oce
s
s
F
i
gure 1
3
. Co
mpariso
n
of Optimizati
on Re
sults
of
Load She
ddi
ng Pro
c
e
s
s
Acco
rdi
ng to
these fig
u
re
s, it is evident
that improve
d
ortho
gon
al
experim
ent
method
perfo
rms
better than traditi
onal meth
od i
n
reg
u
lating p
a
ram
e
ters op
timization. Th
roug
h re
peat
ed
tests, the ran
ge of optimal
grou
p is le
ssen
ed
an
d th
e reg
u
lating
system g
a
in
s better dyna
mic
corre
s
p
ond.
4. Conclu
sion
The propo
se
d pape
r is ai
m to build the model
of la
rge pu
mpe
d
stora
ge po
we
r station
and optimi
z
e
its tran
sient p
r
ocesse
s.
Acco
rdi
ng to
the ch
ara
c
te
ristics of p
u
m
ped
st
orag
e po
wer
stati
on with l
ong
pipeline,
model of its d
i
versio
n syste
m
with elasti
c wate
r ha
mm
er is e
s
tabli
s
h
ed, and RBF
neural network
is ad
opted to
approa
ch th
e nonli
nea
r
prop
ertie
s
of
pump tu
rbin
e whi
c
h
prov
es to effe
ctively
reflect
cha
r
a
c
teristics of pu
mped sto
r
a
g
e
units.
Beside
s, it i
s
conve
n
ient
and
efficie
n
t
to figu
re
out o
p
timal
gro
up
of regulatin
g
para
m
eters
within the gi
ven rang
e b
y
applying
improve
d
ort
hogo
nal exp
e
rime
nt meth
od in
simulat
i
o
n
t
e
st
s.
T
h
ro
ugh
t
h
is
way
,
c
o
st
s
an
d exp
ense of field
tests could
be redu
ce
d
and
simulatio
n
result
s shall
provide th
eoreti
c
al b
a
s
is
and te
chnical supp
o
r
t for pa
ram
e
ters
adju
s
tment a
nd optimal co
ntrol of pump
ed storage u
n
its.
Ackn
o
w
l
e
dg
ements
This work was supp
orte
d by the National Natu
ral Scien
c
e
Found
ation
of China
(No.5
137
916
0) and Key L
aboratory for
Hydro
d
ynam
i
c
Tra
n
si
ents
of Ministry of Educatio
n.
Referen
ces
[1]
Gao Huim
in. N
u
cle
a
r Po
w
e
r P
l
ant an
d Pump
ed Stor
ag
e Sta
t
ion Mod
e
ll
in
g and
C
oord
i
n
a
ted Operati
o
n
.
PhD thesis. Z
h
ejia
ng U
n
ivers
i
t
y
. 200
6.
[2]
Liu Z
h
uqi
ng.
Stud
y on
the
F
u
ll-Sco
pe S
i
mu
lat
o
r of P
u
mpe
d
Stora
g
e
Un
it PhD t
hesis
,
Ch
i
n
a
Agricult
ural U
n
i
v
ersit
y
. 2
000.
[3]
Shen Z
o
n
g
sh
u
,
Z
hang Yong
chua
n.
Hydrop
ow
er Unit Stability an
d Co
ntrol.
HUST
Epress, W
uhan
,
198
8.
[4]
Yang F
a
n. Co
mplete N
u
meri
cal Ana
l
ysis a
n
d
T
heoretical
Rese
arch of S
u
rge T
ank. Ph
D thesis
,
Ho
ha
i
Univers
i
t
y
, 2
0
0
5
.
[5]
HM
Gao, C W
ang.
A Deta
iled
Pu
mpe
d
Storage
Statio
n
Model for
Power System Analysis.
In:
Procee
din
g
s of
Po
w
e
r an
d En
erg
y
Soci
et
y
G
ener
al Meeti
n
g
.
2006; 1-4.
[6]
Wang Yu
an. St
ud
y
o
n
Ap
plic
a
t
ion of El
astic
Mode
l in
Penst
o
cks Base
d o
n
H
y
dra
u
lic V
i
br
ation T
heor
y
.
Master T
hesis. Hoha
i Univ
ersi
t
y
. 200
8.
[7]
Kou Pa
nga
o. Stud
y
on Par
a
meters Identi
f
icat
ion Meth
o
d
s and C
ontr
o
l La
w
s
of W
a
ter T
u
rbin
e
Generator
and
Its Speed Govern
or S
y
st
e
m
. PhD
thesi
s
. Huazh
ong
Univers
i
t
y
of Scienc
e an
d
T
e
chnolog
y. 2
012.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
ANN Ba
sed
Modelin
g and
Optim
i
zation of Large Pum
ped Stora
ge
Station (Ho
n
g
t
ao Zeng
)
2505
[8]
Z
H
Li, OP Mali
k. An orthog
on
al test ap
proac
h
bas
ed co
ntro
l param
eter o
p
timizatio
n
a
nd it
s appl
icati
o
n
to a h
y
dro-turb
i
ne gov
ern
o
r
. IEEE Transactions on Energy
Conv
ersion
. 19
97; 12(4): 3
88-
393.
[9]
T
ang LF
, W
ang YM.
T
he PID Control for E
l
ectro-hydr
aul
ic
Se
rvo Syste
m
Based
on Orthog
on
al T
e
st.
Procee
din
g
s of 2009 Intern
ati
ona
l F
o
rum on
Comput
er Sci
ence-T
e
chn
o
lo
g
y
a
nd Ap
plic
a
t
ions. 200
9;
255-
258.
[10]
Ye Jia
n
h
e
, Li
Che
ngj
un, W
u
Di, etc.
Ap
p
licatio
n of I
m
p
r
oved Ortho
g
o
nal Ex
peri
m
en
t Method
i
n
Governor Param
e
ter Optim
i
z
a
tion.
H
y
dro
p
o
w
er and N
e
w
En
erg
y
.
201
1; 3: 6-8.
[11]
Jin Bo, Z
h
u Sh
iqia
ng,
Z
h
ang
Guang
qi
ong, e
t
c.
Use Orthog
ona
l Exp
e
ri
me
ntal Meth
od t
o
Reg
u
late
PID
Param
e
ters [J],
Chin
a Mecha
n
i
cal En
gin
eeri
n
g. 2000; 1
1
(6): 682-
703.
[12] W
e
i
Shou
pin
g
.
Hydra
u
lic T
u
rb
i
ne Co
ntrol En
g
i
ne
erin
g.
HUS
T
Express, W
uhan. 20
05.
Evaluation Warning : The document was created with Spire.PDF for Python.