Pso Parameters Optimization For Ekf and Akf For Im Rotor Speed
Pso Parameters Optimization For Ekf and Akf For Im Rotor Speed
Pso Parameters Optimization For Ekf and Akf For Im Rotor Speed
2
R
_, o
2
=
I
m
oI
s
I
2
R
, o
3
=
I
m
oI
s
I
o = 1 -
L
m
2
L
s
L
r
, b
1
=
1
cL
s
, o
4
=
R
r
L
r
, o
5
=
R
r
L
r
I
m
,
o
6
=
I
m
[I
p, o
7
=
1
[
, o
8
=
{
[
, b
1
=
1
oI
s
T
l
i load torque. [ i the moment o) inertia.
A. Sampled model of the induction motor
A dicrete<time model o) the .M i o"tained "(
appl(ing a Ta(lor e5panion to equation 273. .t i
-ritten in the )ollo-ing )orm 2equation 2;33.
_
X(k +1) = A
d
X(k) + B
d
u(k) + H
d
w
d
(k)
(k) = C
d
X(k) + I
d
(k)
2;3
@here$
A
d
= c
At
c
I
4
+
At
c
1!
+
(At
c
)
2
2!
B
d
= _ c
At
c
B
t
c
0
Jt (I
4
t
c
+
At
c
2
2
)B
C
d
= C
w
d
= _ c
At
c
H
d
wH
d
1
c
A
T
t
t
c
0
Jt
I
d
=
:
t
c
I
4
i the identit( matri5 o) order :. w
d
(k) et I
d
(k) are
the dicrete tate and meaurement noie, t
e
i the
ampling period. Thee noie are conidered a
additive ignal and allo- o"taining the dicrete and
tochatic model o) the .M. .t i given "( equation 2;3.
The noie are uppoed peudo<-hite ignal, =ero<
mean and ?auian, then the covariance matrice w
d
and I
d
are given "($
E|w
d
(k)w
d
(k + 1)
1
] = w
d
o() 2A3
E|:
d
(k):
d
(k + 1)
1
] = I
d
o()
E|w
d
(k):
d
(k)
1
] = u
E| ] i the mathematical e5pectation, o()$ Dirac
ditri"ution.
.... KA,MA/ F.,T0E! 2KF, EKF, AKF3
The Kalman )ilter 2KF3 i an optimal tate o"erver
ued in a tochatic conte5t. .t ena"le the etimation o)
the tate vector )or a linear dicrete<time (tem )rom
input and output ignal. @hen the noie variance are
&no-n, it i a po-er)ul o"erver. Furthermore, i) the
noie can "e conidered a -hite and ?auian, it i
optimal, that i to a(, )or all o"erver, i that )or -hich
the o"ervation error variance i the lo-et 6B, A, C8. KF
i applied to .M )or the etimation o) the rotor )lu5 and
the tator current, under the aumption that the rotor
peed i meaured.
A. Kalman filter Algorithm
A)ter an initiali=ation phae, KF procedure i
recurive. .t algorithm i ummari=ed in )igure 7.
B. Extended Kalman filter Algorithm (EKF)
KF can "e ued to etimate imultaneoul( the tate
vector and ome model parameter, and then -e o"tain
the e5tended Kalman )ilter 2EKF3. The tate model
"ecome nonlinear 2the varia"le to "e etimated appear
in "oth the e5tended tate vector and alo in the matrice
o) the tate model3. The (nthei i "aed on a local
lineari=ed model around a -or&ing point, and then
appl(ing the linear KF etimation algorithm. .n thi
ection, the KF i e5tended to the rotor peed, in order
to overcome the peed enor. The augmented model o)
the (tem i then conidered. The e5tended tate vector
i$
x
c
= (x
1
, x
2
, x
3
, x
4
, )
1
= (i
ds
, i
qs
,
d
,
q
, )
1
The e5tended model i -ritten in the dicrete<time non
linear )orm "elo-$
X
c
(k + 1) = [
A
d
u
u 1
X
c
(k) + [
B
d
u
u(k)
+ H
d
_
w
d
w
d5
] H
d
1
2A3
A = _
A
11
A
12
A
21
A
22
]
2B3
A
11
= _
o
1
u
u o
1
] A
12
= _
o
2
o
3
p
-o
3
p o
2
]
A
21
= _
o
5
u
u o
5
] A
22
= _
-o
4
-p
p -o
4
]
B = [
B
1
u
; B
1
= _
b
1
u
u b
1
] 4 u = [
u u
u u
C = [
1 u u u
u 1 u u
= o
6
(x
2
x
3
- x
1
x
4
) - o
7
I
I
o
8
2:3
D
D D D
3 E 2
3
F
, F 2 G
F
P k k P
s i
r
=
=
T
d d d
T
d d
d d
! A k k P A k k P
k " B k k # A k k #
+ = +
+ + = +
3 E 2 3 E 7 2
3 7 2 3 E 2
F
3 E 7 2
F
7
3 3 E 7 2 2 3 E 7 2 3 7 E 7 2
+ + + = + +
d
T T
$ % k k %P % k k P k k K
33 E 7 2
F
3 E 7 2 32 7 2 3 E 7 2
F
3 7 E 7 2
F
k k & k k & k K k k # k k # + + + + + = + +
3 E 7 2
F
3 E 7 2
F
k k # % k k & + = +
3 E 7 2 33 3 7 2 2 3 7 E 7 2 k k P % k K ' k k P + + = + +
Figure7$ algorithm o) KF
(k) = b(X
c
(k)) + I
d
(k)
2C3
The computation o) Haco"ean matrice o) equation
2A and C3 -ith repect to the tate vector #
e
(k) i
required to lineari=e thi model. And it i then
dicreti=ed to o"tain the e5preion "elo-$
X
c]
(k + 1) = A
]
X
c]
(k) + [
B
d
u
u(k)
+ H
d
_
w
d
w
d5
] H
d
1
2I3
]
(k) = E
]
X
c]
(k) +I
d
(k) 2J3
@ith
A
]
=
X
c
(k)
_[
A
d
u
u 1
X
c
(k)_, E
]
=
X
c
(k)
(k)
The KF algorithm mentioned a"ove i applied to the
e5tended and lineari=ed model given "( equation 2I3.
%. Adapti(e Kalman filter Algorithme (AKF)
A rotor peed adaptation "aed on ,(apunov
ta"ilit( theor( i ued. The adaptive KF alo allo-
etimating current and )lo- -ithout the rotor peed i
meaured a an alternative to EKF to reduce the
computational cot and enjo( the "ene)it o) KF
adaptive )iltering i generall( aociated -ith a real time
operation. .t i )ollo-ed "( a regulation 2*.3 that -ill
improve the per)ormance o) thi etimate 6B, A, C8.
The model 273 i ued to per)orm the (nthei o) the
peed adaptation la-. The diagram o) )igure 293 ho-
the principle o) rotor peed adaptation.
A
F
K
F
r
LF
i
F
i
Figure9 $ .llutration o) the adaptation o) AKF
The tate equation o) the o"erver, -hich i a cop(
o) the (tem model -ith corrective term, i given "(
the equation "elo-$
J
Jt
x = A
`
x (t) + Bu + K(y - y)
27D3
The d(namic o) the etimation error i given "($
c = x
`
(t) - x (t)
= [A
`
x +Bu + K(y - y) - Ax(t) + Bu(t)
= (A + KC)c +AAx
2773
@here
AA = A
`
- A = _
AA
11
AA
12
AA
21
AA
22
]
2793
AA
11
= [
u u
u u
AA
12
= _
u A0o
9
-A0o
9
u
]
o
9
=
I
m
oI
s
I
p
AA
21
= [
u u
u u
AA
22
= [
u -A
A u
@ith$ A0 = 0
`
- 0
To )ind the la- o) peed adaptation and prove the
ta"ilit( o) the etimation error, ,(apunov )unction
candidate i de)ined "($
I = c
1
c +
A)
2
z
, z > u
2793
$ i poitive. The computation o) it derivative give the
equation 7B.
J
Jt
I = c
1
{(A + KC)
1
+ (A + KC)]c
-2
A)
o
9
(c
x1
x
4
- c
x2
x
3
) + 2
A)
z
J
Jt
)
27B3
@e put$
[
1
= c
1
{(A + KC)
1
+(A +KC)]c
27:3
[
2
= -2
Au
o
9
(c
x1
x
4
- c
x2
x
3
) + 2
Au
z
J
Jt
)
27;3
The *
+
term i de)ined negative "( the ta"ilit( o) the
o"erver. The *
,
term determine the adaptation la- o)
the rotor peed, given "($
J
Jt
0
`
= zo
9
(c
x1
x
4
- c
x2
x
3
)
27A3
c
x1
= x
1
- x
1
, c
x2
= x
2
-x
2
-. The P' 1ontroller
A *. controller 2equation 7C3 i ued to improve the
per)ormance o) the adaptation. The goal o) the integral
action i to eliminate the tatic error.
J
Jt
0
`
= K
P
(c
x1
x
4
- c
x2
x
3
)
+K
I
_(c
x1
x
4
- c
x2
x
3
)Jt
27C3
!(nthei "( conventional method can "e per)ormed
to determine the o) controller parameter K
'
and K
P
.
Mo-ever, )urther adjutment "( trial and error ma( "e
necear(. For thi reaon, the( are determined uing a
*!+ algorithm.
.>. *!+ A,?+0.TMM
The trial and error method i a tediou olution )or
the determination o) controller parameter, or
covariance matrice o) noie value. The olution
propoed "( *!+ i an intereting alternative that
allo- a heuritic earch o) the latter. Kno-ledge o) the
covariance matrice o) noie on tate !
d
and
meaurement $
d
ma&e etimation "( KF, EKF and
AKF optimal. For thi reaon, optimi=ation "( *!+
algorithm i ued to calculate !
d
, $
d
, K
P
and K
'
.
The particle -arm optimi=ation 2*!+3 i "aed on
the metaphor o) ocial interaction and communication
2group o) "ird or )ih in earch o) )ood3. A -arm i
ditri"ute randoml( in the earch pace. Then at each
time, a recurive algorithm i reali=ed 6I, J8$
Each particle * 2!
d
, $
d
, K
P
, K
'
3
evaluate it
poition 2it coordinate3 and remem"er hi
"et per)ormance and qualit( 2the value in the
poition o) the cot )unction3
Each particle receive coordinate -ith the "et
qualitie o) a num"er o) it neigh"or.
Each particle chooe the "et per)ormance a
a detination in it poeion.
FigureB$Application o) *!+ to KF<.M
Figure B illutrate the application o) *!+ algorithm to
the -hole .M<AKF.
The *!+ algorithm )or EKF provide the )ollo-ing
reult$
I
d
= (2.9)
3
. [
1 u
u 1
4 w
d
=
`
79.1 u u u u
u 79.1 u u u
u u 49.9 u u
u u u 49.9 u
u u u u 1u
The *!+ algorithm )or AKF provide the )ollo-ing
reult$
I
d
= 1,. 1u
3
[
1 u
u 1
4 w
d
=
1uu u u u
u 1uu u u
u u 1u u
u u u 1u
K
P
= 2uu4 K
I
=
>. !.MU,AT.+/ 0E!U,T!
@e preent imulation reult on the .M<EKF and
.M<AKF. .M parameter ued )or imulation are given
in the appendi5.
@e ho- the graph related to the tate vector 2the
tator current i
ds
. i
/s
. the rotor )lu5 0
dr
. 0
/r
and
mechanical peed K3 and their repective etimate and
the graph o) etimation error. The )igure :, ;, A, C
and I are relative to EKF. The )igure J, 7D, 77, 79
and 7B are relative to AKF and it *. controller.
The optimi=ation "( *!+ o) the *. controller permit a
good )unctioning o) AKF. A load torque i imulated at
tND.:.
2a3 2"3
Fig:. 2a3!uperpoition o) i
ds
and
ds
.2"3 error "et-een
them
2a3 2"3
Fig;. 2a3!uperpoition o) i
/s
and
/s
.2"3 error "et-een
them
2a3 2"3
FigA. 2a3!uperpoition o) 0
dr
and
dr
.2"3 error "et-een
them.
2a3 2"3
FigC. 2a3!uperpoition o) 0
/r
and
qr
, 2"3 error "et-een
them.
2a3 2"3
FigI.2a3!uperpoition o) K and K
.
The reult given "elo- are relative to AKF$
2a3 2"3
FigJ.2a3!uperpoition o) i
ds
et
ds
,2"3 error "et-een them
2a3 2"3
Fig7D.2a3!uperpoition o) i
/s
and
/s
.2"3 error "et-een
them.
2a3 2"3
Fig77. 2a3!uperpoition o) 0
dr
and
dr
.2"3 error "et-een
them.
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
-30
-20
-10
0
10
20
30
40
time(s)
i
d
s
/
i
d
s
e
s
t
i
m
a
t
e
d
(
A
)
ids
ids estimated
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
-1
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
,e temp 23
,
Oe
r
r
e
u
r
e
n
t
r
e
.
d
e
t
.
d
e
t
i
m
2
A
3
l'erreur
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
-30
-20
-10
0
10
20
30
40
time (s)
iq
s
/
i
q
s
e
s
t
i
m
a
t
e
d
(
w
b
)
iqs
iqs estimated
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
-1
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
time (s)
i
q
s
-
i
q
s
e
s
t
i
m
a
t
e
d
(
A
)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
-1.5
-1
-0.5
0
0.5
1
1.5
time(s)
d
r
/
d
r
e
s
t
i
m
a
t
e
d
(
w
b
)
dr
dr estimated
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
-0.1
-0.05
0
0.05
0.1
time (s)
d
r
-
d
r
e
s
t
im
a
t
e
d
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
-1.5
-1
-0.5
0
0.5
1
1.5
time (s)
q
r
/
q
r
e
s
t
i
m
a
t
e
d
(
w
b
)
qr
qr estimated
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
-0.15
-0.1
-0.05
0
0.05
0.1
0.15
time (s)
q
r
-
q
r
e
s
t
i
m
a
t
e
d
(
w
b
)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
0
20
40
60
80
100
120
140
160
time (s)
/
e
s
t
i
m
a
t
e
d
(
r
d
/
s
)
estimated
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
-4
-2
0
2
4
6
8
10
time (s)
s
p
e
e
d
e
r
r
o
r
(
r
d
/
s
)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
-30
-20
-10
0
10
20
30
40
time (s)
i
d
s
/
i
d
s
e
s
t
i
m
a
t
e
d
(
A
)
ids
ids estimated
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
-1.5
-1
-0.5
0
0.5
1
time (s)
c
u
r
r
e
t
e
r
r
o
r
(
id
s
-
id
s
e
s
t
i
m
a
t
e
d
)
(
A
)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
-30
-20
-10
0
10
20
30
40
time (s)
i
q
s
/
i
q
s
e
s
t
i
m
a
t
e
d
(
A
)
iqs
iqs estimated
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
-1.5
-1
-0.5
0
0.5
1
time (s)
c
u
r
r
e
t
e
r
r
o
r
(
i
q
s
-
i
q
s
e
s
t
i
m
a
t
e
d
)
(
A
)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
-1.5
-1
-0.5
0
0.5
1
1.5
time (s)
d
r
/
d
r
e
s
t
i
m
a
t
e
d
(
w
b
)
dr
dr estimated
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
-0.1
-0.05
0
0.05
0.1
0.15
0.2
0.25
time (s)
d
r
-
d
r
e
s
t
i
m
a
t
e
d
(
w
b
)
2a3 2"3
Fig79. 2a3!uperpoition o) 0
/r
and
qr
.2"3 error "et-een
them.
2a3 2"3
Fig7B. 2a3!uperpoition o) K and K
= 2.1K4 R
s
= 4.27K4 I
= u.4 p = 2,
I
s
= u.4u44 I
m
= u.4 [ = u.u24
{
= u.uu29 4
I
I
= 1u . . ;
The dicrete<time model o) .M e5tended to the rotor
peed i$
X
c
(k + 1) =
`
(1 -t
) u
1
r
u
u (1 - t
) p
u 1 -
u
u
1 -
u
-
X
c
(4)
X
c
() u u 1 -
X
c
(k)
+
`
1
oI
s
u
u
1
oI
s
u u
u u
u u
u(k) +H
d
_
w
d
w
d5
] H
d
1
(k +1) = [
1 u u u u
u 1 u u u
X
c
(k + 1) + I
d
=
L
m
cL
s
L
r
, =
1
cL
s
(R
s
+ R
L
m
2
L
r
2
3, t
c
N7D
<B
.
>... 0EFE0E/1E!
678 H.*.1aron et H.*.Mautier QModliation et commande
de la machine a(nchrone Edition technip.7JJ;.
698
R. RME/? Q1ommande S haute per)ormance et an
capteur mcanique du moteur (nchrone S aimant
permanent ThTe de doctorat, ./* de Touloue,
9DDI.
6B8 M. M.,A.0ET QApplication de outil du traitement
du ignal S la commande de machine tournante
ThTe de doctorat de lOUniverit de /ante.9DD7.
6:8 ,.A. 1A,>.,,+ 1orona QPuelque contri"ution
au5 o"ervateur non linaire S hori=on gliant
ThTe de doctorat ./* de ?reno"le 9DD9.
6;8 M. Ku"ota, K. Matue, T. /a&ano QD!*< #aed
peed adaptive )lu5 o"erver o) induction motor,
.EEE Tranaction .ndutr( Application, >+,. 9J,
/U9, 7JJB.
6A8 0.?una"alan, >.!u""iah and #.0.0edd( Q!enorle
1ontrol o) .nduction Motor -ith E5tended Kalman
Filter onTM!B9DF9I79proceor.A1EE./+>.9DDJ
6C8 *.@ira, H.*.U0#A/ Q*rdiction de trajectoire par
)iltrage de Kalman adaptati) S tat dicret multiple.
!eptem"re 9DD7.
6I8 H.Kenned(, 0. E"erhart, V*article -arm
optimi=ationW. .n$ *roc. o) .EEE .nternational
1on)erence on /eural /et-or&, .EEE *re 27JJ;3,
7J:9<7J:I.
6J8 H. Kenned(, !tereot(ping$ V.mproving particle -arm
per)ormance -ith cluter anal(iW. .n$ *roc. o) the
.EEE 1ongre on Evolutionar( 1omputation. 29DDD3
7;DC<7;79
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
-1.5
-1
-0.5
0
0.5
1
1.5
time(s)
q
r
/
q
r
e
s
t
i
m
a
t
e
d
(
w
b
)
qr
qr estimated
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
-0.25
-0.2
-0.15
-0.1
-0.05
0
0.05
0.1
time (s)
q
r
-
q
r
e
s
t
i
m
a
t
e
d
(
w
b
)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
-50
0
50
100
150
200
time(s)
/
e
s
t
i
m
a
t
e
d
(
r
d
/
s
)
estimated
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
-40
-20
0
20
40
60
time (s)
s
p
e
e
d
e
r
r
o
r
(
r
d
/
s
)