A Nonlinear State Observer Approach For A New Generation of Ultra High Speed Relay

A nonlinear state observer approach for a new
generation of ultra high speed relay

F. Cacace, S. Muzi, G. Palombo
April 22, 2013
1 Introduction
2 Problem Formulation
We consider a wire line in two dierent situations: before and after the damage
appears. In the rst situation the line appears as in Fig. 2 (left). The two generators
e
1
and e
2
are independent with equations e
1
= Asin(t), e
2
= Asin(t + ). The
circuit resistance and inductance can be expressed as R
tot
= L
R
and L
tot
= L
L
,
where L is the line length,
R
is the resistance density and
L
is the inductance
density which are intrinsic line parameters.
As soon as the damage happens, a new resistance appears in some intermediate
location of the line, as shown in Fig. 2 (right). The position of R
d
corresponds to
the damage location. We have that the original resistance and inductance are split
in two contributions,
R
tot
= R
1
+ R
2
(1)
L
tot
= L
1
+ L
2
(2)
R
1
= xR
tot
, R
2
= (1 x)R
tot
(3)
L
1
= xL
tot
, L
2
= (1 x)L
tot
, (4)
(5)
where the parameter x [0, 1] represents the fraction of L where the damage oc-
curred. The model of circuit after the damage in Fig. 2 (right) requires the values
of x and R
d
. The goal of out algorithm is to identify these unknown parameters.
Figure 1: Circuit without damage (left) and after the damage (right)
1
Since the two parameters are constrained, because x [0, 1] and R
d
0, we
introduce a pair of equivalent unconstrained parameters
1
,
2
R, where
x = s(
1
) =
1
2
+
arctan(
1
)
(6)
R
d
= q(
2
) =
2
2
. (7)
From (3)-(4) we have
R
1
=s(
1
)R
tot
, R
2
= 1 s(
1
)R
tot
. (8)
L
1
=s(
1
)L
tot
, L
2
= 1 s(
1
)L
tot
. (9)
The damage is therefore completely characterized when (
1
,
2
) are specied.
The model of the damaged circuit can be written
e
1
(t) = L
1
di
1
(t)
dt
+ R
1
i
1
(t) + R
d
(i
1
(t) + i
2
(t)) (10)
e
2
(t) = = L
2
di
2
(t)
dt
+ R
2
i
2
(t) + R
d
(i
1
(t) + i
2
(t)) (11)
where i
1
(t) and i
2
(t) are the currents. Using (
1
,
2
) the above equations become
di
1
(t)
dt
=
R
tot
L
tot
i
1
(t)

2
2
s(
1
)L
tot
(i
1
(t) + i
2
(t)) +
e
1
(t)
s(
1
)L
tot
(12)
di
2
(t)
dt
=
R
tot
L
tot
i
2
(t)

2
2
(1 s(
1
))L
tot
(i
1
(t) + i
2
(t)) +
e
2
(t)
(1 s(
1
))L
tot
. (13)
3 Proposed Approach
In order to evaluate the damage position and eect we aim at estimating (
1
,
2
)
from eqs. (12)-(13) and from the measurements of i
1
(t), i
2
(t), e
1
(t) and e
2
(t) that
are supposed to be available in real-time.
The approach proposed in this paper is a standard technique in the area of pa-
rameter identication of deterministic systems. It consists in adding the parameters
1
and
2
to the state variables and to use a high-gain observer to estimate the
state variables from the available measurements. Since
1
and
2
are constants, the
model equations will include the two additional equations

1
= 0 and

2
= 0. The
correction term in the observer equation will drive the estimate of
1
and
2
from
an initial guess to the true value, in a time interval that depends on the gain of the
observer and on the dynamics of the system.
Since (12)-(13) are nonlinear with respect to
1
and
2
, we need a state observer
for nonlinear systems. Any such observer can be used in principle, but in this
paper we have used the Luenberger-like observer described in [1],[2] that has been
successfully applies to a large number of applications in many areas.
We dene an extended state that includes the currents i
1
(t), i
2
(t) together with
the unknown parameters
1
,
2
x
e
(t) =
x
1
(t)
x
2
(t)
x
3
(t)
x
4
(t)
i
1
(t)
i
2
(t)
. (14)
2
The evolution of x(t) and the measurement equations are
x
e
(t) =
R
tot
L
tot
x
1
(t)
x
4
(t)
2
s(x
3
(t))L
tot
(x
1
(t) + x
2
(t)) +
e
1
(t)
s(x
3
(t))L
tot
R
tot
L
tot
x
2
(t)
x
4
(t)
2
(1s(x
3
(t)))L
tot
(x
1
(t) + x
2
(t)) +
e
2
(t)
(1s(x
4
(t)))L
tot
0
0
(15)
y(t) =
x
1
(t)
x
2
(t)
= Cx
e
(t) (16)
C =
1 0 0 0
0 1 0 0
, (17)
The main step to design the nonlinear observer of [1] is to dene the drift-observability
map z(t) = (x(t)), a map from the state to the output and its time derivatives.
This is equivalent to re-write the system state and output equations using a change
of coordinates from x
t
(t) to a new variable z(t) of the same size that contains the
output and its time derivatives. In our case we dene the drift-observability map
z(t) = (x
e
(t)) =
x
1
(t)
x
1
(t)
x
2
(t)
x
2
(t)
. (18)
The observer includes the Jacobian of (x), which is the R
nn
matrix
Q(x
e
) =
d(x
e
)
dx
e
(19)
Denoting with x(t) the observer estimate of the state, the observer equation is
x(t) = x(t) + Q( x(t))

1
K (y(t) C x(t)) (20)
where K is the gain of the observer, which is chosen to have the desired rate of
convergence to zero of the estimation error in the z coordinates. Specically, K is
computed choosing the eigenvalues
i
< 0, i = 1, 2, 3, 4 of the linear part of the
error dynamics in z coordinates as
K =
(
1
+
2
) 0
2
0
0 (
3
+
4
)
0
3
. (21)
4 Example
In order to validate the proposed approach, the observer algorithm has been vali-
dated using realistic values for the model. Referring to Figure 2 (right), the param-
eters used in the example are summarized in Table 1 .
3
Parameter Value
R
tot
9.3
L
tot
0.110H
A 122000V
50Hz
/2
Table 1: Parameters used in the example
The algorithm has been validated for various values of R
g
: between 0.1 and
20, and for various damage position: from the beginning to the end of the line,
with steps of 0.005%, of the entire line itself.
For the the observer design, the following values have been used in eq. (21),
1
= 2 10
3
,
2
= 2.2 10
3
,
3
= 4 10
4
,
4
= 4.4 10
4
. For what concern the
initial condition, they worth to the measured currents at time t = 0 for x
1
(0) and
x
2
(0), and to 1 for x
3
(0) and x
4
(0).
As rst result, an intrinsic not-observability of the system arise from the middle
to the end. This problem is solved using two observers, one for each half of the line.
The second one has been developed using the same model but setting, referring to
(8) and (9), R
2
= s(
1
), so R
1
= 1 s(
1
) and L
2
= s(
1
), so L
1
= 1 s(
1
).
The algorithm has been completed with a main procedure that understand which
observer diverges, based on the reciprocal condition number of Q(t), and produce
only the desired output.
In Figure figura are shown the results for various damage position, setting
R
g
= 7.02.
References
[1] M. Dalla Mora, A. Germani, C. Manes, Design of state observers from a drift-
observability property, IEEE Trans. on Automatic Control, 45 (8), pp. 1536
1540, 2000.
[2] G. Ciccarella, M. Dalla Mora, A. Germani, A Luenberger-like observer for
nonlinear systems, Int. Journal of Control, 57, (3), pp. 537556, 1993.
4

A Nonlinear State Observer Approach For A New Generation of Ultra High Speed Relay

Uploaded by

Copyright:

Available Formats

A Nonlinear State Observer Approach For A New Generation of Ultra High Speed Relay

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

A Nonlinear State Observer Approach For A New Generation of Ultra High Speed Relay

Uploaded by

Copyright:

Available Formats

A nonlinear state observer approach for a new

generation of ultra high speed relay

x(t) = x(t) + Q( x(t))

You might also like