Documente Academic
Documente Profesional
Documente Cultură
Kalman Filter”
Thomas F. Edgar
Department of Chemical Engineering
University of Texas
Austin, TX 78712
Ex
turbulent flow,
temperature sensor
in boiling liquid
var( x ) ∆ E ( x − x )
2
∆ (x-x)(x-x)
∞
∫ − 2
( x x ) p( x ) dx
1
−∞
[ var( x )]
2
= σ (standard deviation)
cov( x, y ) = E [ xy ] − x y
cov( x, x ) = var x = E x 2 − ( x )2
P =P P ≥ 0 pii = var xi
T
ef = (I − F ) e p + F e r F : weighting matrix
The covariance matric of ef is
H = P − F P − P F + F (P + R ) F
T T
F opt = P (P + R )−1
scalar example:
P = q2 1
(a ) F=
R =q 2
2
xˆ (k ) = A(k − 1)xˆ (k − 1)
+ K (k ) [ y (k ) − CA (k − 1) xˆ (k − 1)]
ˆ"
"y
y − yˆ : difference between measurement and estimate
n.b. if we can’t measure xi (k), then xi (0) is unknown
Thomas F. Edgar (UT-Austin) Kalman Filter Virtual Control Book 12/06 11
State Estimation (cont’d)
Define covariance matrices
Qnxn : w (k ) process noise vector (white)
Rℓxℓ : v (k ) instrument noise vector (white)
P = ( x p − x )( x p − x )
T
x: true value
(x − xk )( x p − xk )
T
P (k ) =
p
[ ]
T
R O
cov(v1 ) R1 = γ →∞ (bad or missing information
O γ I
p(x) on n − ℓ
x components )
estimate xr (k ) = C1−1y1(k )
Ex C=[1 0]
1 0
R=1 R1 = N large
0 N
1 0 note
R1−1 = -1
0 N N −1 → 0
xˆ (k ) = x p (k ) + F (k ) xr (k ) − x p (k )
(F opt = P (P + R )−1
from earlier analysis)
ℓx ℓ
update
H (1) = P (1) − K (1)CP (1)
n
= ∑ Pii (t ) = tr (P )
i =1