www.ietdl.

org
Published in IET Generation, Transmission & Distribution
Received on 28th June 2010
Revised on 13th January 2011
doi: 10.1049/iet-gtd.2010.0459

ISSN 1751-8687

Innovation concept for measurement gross error

detection and identification in power system state
estimation
N.G. Bretas1 A.S. Bretas2 S.A. Piereti1
1

Electric Engineering Department, University of Sao Paulo, Av. Trabalhador Sao Carlense 400, Sao Carlos,
SP 13566-590, Brazil
2
Electric Engineering Department, Federal University of Rio Grande do Sul, Av. Osvaldo Aranha 103, Porto Alegre,
RS 90035-190, Brazil
E-mail: ngbretas@sc.usp.br

Abstract: In this study, the innovation approach is used to estimate the measurement total error associated with power system
state estimation. This is required because the power system equations are very much correlated with each other and as a
consequence part of the measurements errors is masked. For that purpose an index, innovation index (II), which provides the
quantity of new information a measurement contains is proposed. A critical measurement is the limit case of a measurement
with low II, it has a zero II index and its error is totally masked. In other words, that measurement does not bring any
innovation for the gross error test. Using the II of a measurement, the masked gross error by the state estimation is recovered;
then the total gross error of that measurement is composed. Instead of the classical normalised measurement residual
amplitude, the corresponding normalised composed measurement residual amplitude is used in the gross error detection and
identication test, but with m degrees of freedom. The gross error processing turns out to be very simple to implement,
requiring only few adaptations to the existing state estimation software. The IEEE-14 bus system is used to validate the
proposed gross error detection and identication test.

Introduction

The ability to detect and identify gross errors is one of the

most important attributes of the state estimation process in
power systems. This characteristic to deal with gross errors
makes the results of the state estimation process preferable
if compared with the SCADA raw data [1].
In order to mitigate the gross errors effect on the state
estimation results, some robust estimators have been
introduced in power systems [2]. One of the rst robust state
estimators applied to power systems was the weighted least
squares (WLS) estimator endowed with the largest normalised
residual test [3] for gross error detection and identication.
However, this combination is not robust in the presence of
single and multiple non-interacting gross errors [2]. Also, this
estimator is not able to identify multiple interacting gross
errors, especially when the errors are conforming and/or occur
in leverage points, which are highly inuential measurements
that attract the state estimation solution towards them [4].
Other alternative estimators, which are more robust than the
WLS estimator, have been proposed. The weighted least
absolute value estimator, for example, can deal better with
multiple gross errors, but it is prone to fail in the presence
of a single gross error at a leverage point [2].
The least median of squares estimator is another estimator
alternative [2, 4]. It is a member of the family of estimators
IET Gener. Transm. Distrib., 2011, Vol. 5, Iss. 6, pp. 603 608
doi: 10.1049/iet-gtd.2010.0459

known collectively as high-breakdown point estimators. It is

inherently resistant to outliers in leverage points and can
handle multiple interacting gross errors, even when they are
conforming. However, it requires excessive computing time
for online applications [1].
In [5], using of geometrical properties, the authors state that
gross errors in critical pairs of measurements are detectable
but not identiable. They even generalise the idea of critical
pair of measurements and even generalise that for a critical
k-tuple of measurements, k 2 2 gross errors are identiable
while k 2 1 of them are detectable but not identiable. No
mention of recovering those errors is made.
In [68], more insights related to the detectability of gross
errors in power system state estimation are provided. This is
achieved decomposing each measurement error into two
components: the component pertaining to the measurement
Jacobean space, the masked error component and the
component orthogonal to that space. The ratio between the
norms of those quantities, the undetectability index (UI),
provides a measure of how much of masked error a
measurement contains. That allows more insights related to
the masking errors in power system state estimation, beyond
offering a wider view of the concept of leverage measurements.
In this paper, using the WLS estimator, and based on
geometrical concepts as in references [7, 8], more insights
related to the detectability of gross errors in power system
603

& The Institution of Engineering and Technology 2011

www.ietdl.org
state estimation are provided. This is achieved decomposing
each measurement into two components: the component
orthogonal to the measurement Jacobean space and the
component pertaining to that space, that is, the component
which masks error. The ratio between the norms of those
quantities, the innovation index (II), provides a measure of
how much new information a measurement contains.
Although this index is numerically just the inverse of the UI,
it allows explaining in a very clear way, using of the wellknown concept of new information available for dynamic
systems, all the contents related to the masking effect in state
estimation. Moreover, based on the II index, a technique to
recover the error not reected in the state estimation residual,
that is, the masked error is proposed. The recovering of the
measurements errors, as proposed in this paper, contradicts
the statement that errors in critical pair of measurements or
in k 2 1 measurements of a critical k-tuple are detectable but
not identiable, as previously stated in [5]. In this paper,
instead of the classical normalised measurement residual
amplitude, the corresponding normalised composed
measurement residual amplitude is used in the gross error
detection and identication test.

Background

Consider a power system with n buses and m measurements,

modelled for state estimation purposes, as a set of non-linear
algebraic equations, that is
z = h(x) + e
m

Fig. 1 Geometric interpretation of the operator P acting on the

vector Dz

into a direct sum of two vector subspaces, that is, Rm =

<(H) (<(H)) in which the range of H, <(H ), is an
N-dimensional vector subspace into R m and <(H) is
its orthogonal complement, that is, if u [ <(H ) and
v [ <(H) , then ku, vl uTWv 0.
In the linear state estimation formulation (3), the state
estimation solution can be interpreted as a projection of the
measurement vector mismatch Dz onto the <(H ). Let P be
the linear operator which projects vector Dz onto <(H ),
that is, Dz = PDz and r = Dz Dz be the residual vector
for the linearised model. The projection operator P, which
minimises the norm J, is the one that projects Dz
orthogonally onto <(H ) in the sense of the inner product
ku, vl uTWv, that is, the vector Dz = HDx is orthogonal
to the residual vector. More precisely

(1)
kDz, rl = (HDx)T W (Dz HDx) = 0

where z [ R is the measurement vector, x [ R is the state

vector, h: R N  R m (m . N ) is a continuously non-linear
differentiable function, e is the (m 1) measurement error
vector with zero mean and Gaussian probability distribution
and N 2n 2 1 is the number of unknown state variables
to be estimated. Since the number m of measurements is
higher than the number N, a common solution to estimate
the states variables is the well-known WLS method, which
searches for state x, minimising the functional J(x)
(z 2 h(x))TW (z 2 h(x)), where W is a symmetric and
positive denite real matrix. In power system state
estimation, the weight matrix W is usually chosen as the
inverse of the measurement covariance matrix. Functional J
is a norm in the measurement vector space R m that is
induced by the inner product ku, vl uTWv, that is

Solving this equation for Dx, one obtains: Dx =

(H T WH)1 H T W Dz.
As Dz = HDx, the projection matrix P will be the
idempotent matrix
P = H(H T WH)1 H T W

r = Dz Dz = Dz PDz = (I P)Dz
(2)

Let x be the estimated state, that is, the solution of the

aforementioned minimisation problem, and dene the
estimated measurement vector as z = h(x). The residual
vector is dened as the difference between z and z , that is,
r = z z .
The linearisation of (1), at a certain point x , yields
Dz = HDx + e

(3)

with H h/x is the Jacobean of h calculated at x ,

Dz z 2 h(x) z 2 z is the measurement vector
mismatch, and Dx x 2 x .
If system (2) is observable, that is, rank(H ) N [8], then
the vector space of measurements R m can be decomposed
604
& The Institution of Engineering and Technology 2011

(5)

Remark 1: If W is a diagonal matrix given by W cI, where

c . 0 is a real number, and I is the identity matrix, then
P P T and P is said to be an orthogonal projection.
In power system literature, the matrix P is usually called
Hat matrix and it is also known as K matrix [4]. The
residual vector is found to be

J (x) = z h(x)2W = kz h(x), z h(x)l

= [z h(x)]T W [z h(x)]

(4)

(6)

where the idempotent matrix (I 2 P) is an operator that

projects Dz onto <(H) . Matrix (I 2 P) is given by
(I P) = I H(H T WH)1 H T W

(7)

and is usually called residual sensitivity matrix; it is also

known as S matrix [4] in power system literature. Fig. 1
illustrates the operator P acting on the vector Dz.

3 Innovation concept and innovation index

proposition for the recovery of masked errors
3.1

Definition

A measurement innovation [14], in static state estimation, is

dened as the new information that measurement contains
related to the measurements that compose the Jacobean of
IET Gener. Transm. Distrib., 2011, Vol. 5, Iss. 6, pp. 603 608
doi: 10.1049/iet-gtd.2010.0459

www.ietdl.org
the system. That is, it is an extension of the Gram Schmidt
transformation [9], in the sense that the measurement is
contained in the measurement set that compose the Jacobean.
The new information of a measurement, for state estimation
purposes, is contained in the part of the measurement that is
orthogonal to the Jacobean range space related to the
measurement set. In order to obtain the innovation of a
measurement it is only required to project that measurement
in that space. In case a measurement contains error, the
component orthogonal to the Jacobean range space will
show its error through the measurement residual [7] and the
other measurement component, in the R(H ), will have the
error completely masked, see Fig. 2.
In this way, the masked error of a measurement, associated
with the state estimation process, is the measurement error
that is contained in the Jacobean range space.
Since the measurements error vector emi R(H) is known, is
in the R(H) , being equal to the measurement residual vector
[7]; the other component of the measurements residual vector,
by other side, is in the R(H ), therefore they are orthogonal to
each other. Then it is easy to compose the measurements total
error vector. That is, the corrected measurement error vector
(CNE) will be
em = emR(H) + emR(H)

(8)

and for the measurement i

emi 2 = emiR(H) 2 + emiR(H) 2

(9)

emi being the ith component of the measurement residual

vector em .
For a particular case this can be illustrated in Fig. 2.
In order to obtain the measurement i masked error, from (9)
the following is proposed:

using (5) and (7) dene MiR(H ) P(Midi) and

M iR(H) = (I P)(Mi di ),
the
measurement
vector
component in the R(H ) and in the R(H) , respectively.
With that in mind, the innovation index for the ith
measurement is thus proposed
IIi = M iR(H) W /M iR(H) W

(10)

Remark 2: As can be seen, the proposed innovation index for

a measurement does not depend on the measurement
magnitude but on the measurement position related to the
Jacobean range space as well as on the projection matrix P.
Observe that the critical measurements will be the ones
with II equal to zero.
Index II is the inverse of the UI as dened in [7].
The reason the II is being proposed, instead the UI, is
because the rst index is related to the information
theory and as a consequence is well known in many
researches eld with known properties. The unique required
change was to adapt the denition of innovation, available
in the eld of dynamic systems, to the static situation.
This will facilitate to explain the results of the paper
simulations.
Step 2: Recovering the measurements masked errors. For the
purpose of estimating the measurement total error magnitude,
the following is proposed:
Process the state estimation and compute the residual
vector r, (r = emiR(H) ).
Using the index II for each measurement i compute the
masked error vector, that is: MiR(H ) P(Midi) and the
measurement i component on the R(H) , that is:
M iR(H) = (I P)(M i di ). With those two vectors in hands,
and using (10), compute the measurement i innovation
index IIi . As a consequence, (9) becomes
emi 2 = 1/IIi2 .emiR(H) 2 + emiR(H) 2

Step 1: Suppose the ith measurement vector, that is, mi Midi

with di [0 . . . 1i . . . 0]T and Mi the measurement i magnitude;

(11)

or
emi 2 = (1 + 1/IIi2 ).emiR(H) 2 = (1 + 1/IIi2 )ri2

(12)

where ri is the measurement i residual.

Step 3: Gross error detection and identication test.
1. Apply the gross error detection and identication test
(HTI), as in the Appendix, but using the corrected
normalised measurement error (CNE), as in (12).
2. Once a gross error was detected and identied, perform the
measurement correction using the estimated gross error, that
is the measurement CNE, and then again estimate the
power system states.

Fig. 2 Gross error in measurement z1; with z estimated of z,

zwtge measurement without gross error, r the measurement
residual and R(H) the range space of the matrix H
IET Gener. Transm. Distrib., 2011, Vol. 5, Iss. 6, pp. 603 608
doi: 10.1049/iet-gtd.2010.0459

Remark 3: The understanding of the authors is that

Schweppe et al. [10 12], and many others, have applied
the classical statistical hypothesis test (HTI) in power
systems for gross error detection and identication with
limited efciency. The reason is that the HTI as used in
power system state estimation requires the non-existence
of masking effect in the estimation process; however, in
power system state estimation the masking effect does
exist in a signicant way since the measurements residuals
are very much correlated.
605

& The Institution of Engineering and Technology 2011

www.ietdl.org
4

Numerical tests

For the numerical tests, the IEEE-14 bus system whose data
are available at [13], and illustrated on Fig. 3, will be used.
The measurement values (z lf ) used in the tests, were
obtained from a load ow solution, to which normally
distributed noises were added. The measurements noise was
assumed to have zero mean and a standard deviation s
given by s ( pr |z lf|)/(3)|; with pr, the metre precision,
equal to 3%. The measurement weights are assumed for

Fig. 3 IEEE-14-bus system

Table 1

each measurement as (1)/(s2i ). For all the simulated cases it

was assumed the system as observable.
Scenario-I: In this measurement scenario the measurement
set shown in Table 1 will be used.
Remark 4: In the following tables, this nomenclature was
used: I:a injection measurement at bus a ; F:a 2 b
ow measurement from bus a to bus b ; A and R stand
for active and reactive power measurements, respectively;
and V:a voltage magnitude measurement at bus a .
For all the cases, the measurement scenario is so that the
network is assumed to be observable.
In Table 1, using the proposed formulation, the index II for
each measurement was calculated, as well as the
measurements gross error detection level by means of the
largest normalised residual test, considering a threshold
value of l 3 (see the Appendix).
As seen in that table, there is a close relation between the
measurement detection level and the corresponding index
II. Measurements with high IIs, in general above one, have
gross errors detected and identied, with the normalised
gross error detection level close to 3.0s (average values for
30 different cases). The meaning of II equal to one is that
the masked measurement error magnitude is equal to the
measurement residual.
On the other hand, measurements with low II require very
high gross errors in order to be detected and identied, that is,
the masked error because of the estimating process is high. As
a consequence, the results of Table 1 shows that the largest
normalised residual test, as performed in these days, may
fail signicantly when the gross errors are in those
measurements with low II.

Measurements scenario-I: gross error innovation level

Meas.

Active power

II (A)

Detec. level (s)

Reactive power

II (R)

Detec. level (s)

I:3
I:8
I:9
I:10
I:11
I:13
I:14
F:1-2
F:2-1
F:1-5
F:5-1
F:3-4
F:4-5
F:5-4
F:4-7
F:4-9
F:9-4
F:11-6
F:6-12
F:7-8
F:8-7
F:7-9
F:11-10
F:13-14

20.942
0.000
20.295
20.090
20.035
20.135
20.149
1.569
21.526
0.755
20.727
20.233
20.612
0.617
0.281
0.161
20.161
20.073
0.078
0.000
0.000
0.281
0.038
0.056
magnitude
1.060
1.010
1.090

1.577

0.917
0.279
0.429
0.228
0.952
1.316
1.351
2.564
2.564
0.437
1.042
1.020
1.111
2.326
2.326
1.493
0.134

0.990
0.575
0.855

3.5

4.4
10.7
7.80
11.7
4.8
3.9
3.7
3.4
3.2
7.8
4.4
4.6
4.0
3.2
3.4
3.7
22.7

4.2
5.9
5.1
II
4.167
4.000
4.348

0.061
0.0176
20.166
20.058
20.018
20.058
20.050
20.204
0.277
0.039
0.022
0.045
0.158
20.142
20.097
0.040
0.017
20.034
0.025
20.172
0.176
0.058
0.016
0.017

0.871
1.389
1.042
0.385
0.476
0.255
0.893
0.709
1.163
0.228
0.219
0.990
1.123
0.935
2.857
0.208
1.163
1.515
0.099
1.449
1.389
0.358
0.549
0.704
detection level (s)
3.2
3.2
3.2

5.9
3.8
4.2
7.7
6.7
11.7
5.1
5.2
3.9
12.4
13.6
4.4
4.2
4.5
3.7
14.4
4.2
3.7
31.5
3.9
3.8
10.1
5.8
5.9

V:1
V:3
V:8

606

& The Institution of Engineering and Technology 2011

IET Gener. Transm. Distrib., 2011, Vol. 5, Iss. 6, pp. 603 608
doi: 10.1049/iet-gtd.2010.0459

www.ietdl.org
Table 2

Measurements scenario-I: measurements normalised gross errors composition CNR

Meas.

II (A)

r N (A) (s)

CNE(A) (s)

II (R)

r N (R) (s)

CNE(R) (s)

I:3
I:8
I:9
I:10
I:11
I:13
I:14
F:1-2
F:2-1
F:1-5
F:5-1
F:3-4
F:4-5
F:5-4
F:4-7
F:4-9
F:9-4
F:11-6
F:6-12
F:7-8
F:8-7
F:7-9
F:11-10
F:13-14

1.578

0.917
0.279
0.428
0.228
0.952
1.316
1.351
2.500
2.564
0.437
1.030
1.020
1.124
2.381
2.381
1.493
0.134

1.000
0.575
0.862
II
4.167
4.000
4.545
4.167

8.52

6.82
2.82
3.90
2.64
6.62
7.90
8.06
9.18
9.41
3.91
7.14
7.11
7.48
9.25
9.25
8.25
1.31

7.10
5.08
6.22

10.08

10.1
10.51
9.89
11.8
9.62
9.93
10.1
9.89
10.1
9.80
9.97
10.0
10.0
10.0
10.0
9.94
9.92

10.0
10.2
9.56
r N (s)
9.68
9.64
9.72
9.67

0.571
0.571
1.042
1.042
0.490
0.255
0.472
0.704
1.149
0.230
0.220
1.013
1.123
0.935
2.857
0.209
1.136
2.000
0.099
1.471
1.408
0.358
0.549
0.704

5.00
5.00
7.21
7.21
4.32
2.83
6.29
5.80
7.65
2.48
2.23
6.97
7.38
6.77
8.95
2.11
7.38
8.26
0.57
8.07
8.04
2.99
5.02
5.37
CNE (s)
9.95
9.95
9.97
9.96

10.0
10.0
10.0
10.0
9.84
11.4
9.45
10.1
10.1
11.1
10.4
9.93
9.90
9.93
9.51
10.3
9.88
9.90
5.86
9.79
9.90
8.89
10.4
9.35

V:1
V:3
V:8
V:13

As can be seen in Table 1, the masking measurement error

in power system state estimation is a signicant fact, that is,
the measurement gross error detection level is related to the
index II; or in another words, it is related to the
measurement masked gross error.
With that in mind, in the next test, Table 2, using the same
measurement set as in Table 1 but with a new set of random
numbers, a gross error of 10s was added, in an isolated way,
to each measurement; in order to compare the available
measurement gross error detection and identication test
and the proposition of this paper, that is, the measurement
normalised residual r N and the corresponding CNE was
computed.
As can be seen in Table 2, for the measurements with high
II, their normalised residual and the corresponding CNE were
close to each other; however, for measurements with low II
they are quite different. As a consequence, for those
measurements with low II, the classical measurement gross
error detection and identication will fail in a signicant
way. They will require rst the measurement masked error
estimation, then the composition of the total measurement
error as proposed in this paper through (12). Table 2 shows,
in a clear way, that when the step of recovering the gross
error is performed and the measurement gross error is
composed, the HTI procedure will be reliable.
Remark 5: The numerical tests have shown that further studies
must be made when gross errors occur in redundant
measurements. Suppose, for example, a case of a
measurement critical pair. Since they have the same
normalised residual, when their residuals are corrected the
IET Gener. Transm. Distrib., 2011, Vol. 5, Iss. 6, pp. 603 608
doi: 10.1049/iet-gtd.2010.0459

measurement with lower II will have larger CNE. When the

error is on the measurement with lower II, using the largest
corrected normalised residual test the conclusion of gross
error on that measurement will work correctly. However,
when the error is on the measurement of larger II, some extra
work, in order to identify correctly the measurement
containing gross error, must be made.

Conclusions

In this paper, more insights related to detection and

identication of measurements gross errors in power system
state estimation, using the innovation concept has been
provided. This is achieved decomposing each measurement
into two components: the component orthogonal to the
Jacobean range space and the one onto that space. The ratio
between the w-norm of those quantities, the innovation
index, gives the ratio of how much of new information a
measurement contains. In other words, a large II, larger
than one, gives more information than it masks. According
to the papers results, it is possible to conclude that the
higher an II of a measurement the lower the masked error.
Based on the II index, a way of recovering the masked
errors, resulting from the measurement residual estimating
process, has been proposed.
The total measurement residual is then corrected. Also
the proposed measurement gross error detection and
identication test, using the measurement corrected gross
error, as proposed in this paper, works very well. These
results show the gross error test as used in the existing
literature is not effective in nding the measurement error.
607

& The Institution of Engineering and Technology 2011

www.ietdl.org
It does not consider the masking effect existing in the power
system state estimation. So just for large errors it will detect
error in the measurement set, but always correcting in
wrong way the measurement error amplitude.
The critical measurements are easily identied as those
with II equal to zero.
Using the papers approach, we are working on a multiple
gross errors detection and identication test in power system
state estimation.

10 Schweppe, F.C., Wildes, J.: Power system static-state estimation, part I:

exact model, IEEE Trans. Power Appar. Syst., 1970, 89, (1), pp. 120125
11 Schweppe, F.C., Douglas, B.R.: Power system static-state estimation,
part II: approximate model, IEEE Trans. Power Appar. Syst., 1970,
89, (1), pp. 125130
12 Schweppe, F.C.: Power system static-state estimation, part III:
implementation, IEEE Trans. Power Appar. Syst., 1970, 89, (1),
pp. 130135
13 http://www.ee.washington.edu/research/pstca/, accessed January 2010
14 Anderson, B.D.O., Moore, J.B.: Optimal ltering (EE Prentice Hall
Series, 1979).

Acknowledgments

The authors would like to thank FAPESP, Scientic

Foundation of the State of Sao Paulo, Brazil, for the
nancial aid given to this research. Also we would like to
thank the graduate student Raphael A.S. Benedito from the
Electrical Engineering Department of EESC, USP, for
simulations performed.

References

1 Falcao, D.M., Arias, M.A.: State estimation and observability analysis

based on echelon forms of the linearized measurement models, IEEE
Trans. Power Syst., 1994, 9, (2), pp. 979987
2 Monticelli, A.: Electric power system state estimation, Proc. IEEE,
2000, 88, (2), pp. 262271
3 Handschin, E., Schweppe, F.C., Kohlas, J., Fiechter, A.: Bad data
analysis for power system state estimation, IEEE Trans. Power
Appar. Syst., 1975, 94, (2), pp. 329 337
4 Abur, A., Esposito, A.G.: Power system state estimation: theory and
implementation (Marcel & Dekker Publishers, Nova York, EUA, 2004)
5 Clements, K.A., Davis, P.W.: Multiple bad data detectability and
identiability: a geometric approach, IEEE Trans. Power Deliv.,
1986, 1, (3), pp. 355 360
6 Bretas, N.G., London Jr J.B.A., Alberto, L.F.C., Benedito, R.A.S.:
Geometrical approaches for gross errors analysis in power system
state estimation. Proc. 2009 IEEE Bucharest Power Tech, Romania,
June 2009
7 Bretas, N.G., London Jr J.B.A., Alberto, L.F.C., Benedito, R.A.S.:
Geometrical approaches on masked gross errors for power
systems state estimation. Proc. IEEE PES General Meeting, Calgary,
July 2009
8 Bretas, N.G.: Network observability: theory and algorithms based on
triangular factorization and path graph concepts, IEE Proc. Gen.
Trans. Distrib., 1996, 143, (1), pp. 123128
9 Kailat, T., Sayed, A.H., Hassib, B.: Linear estimation (Prentice Hall
Inc., 2000)

608
& The Institution of Engineering and Technology 2011

Appendix

8.1 Detection and identification of measurements

gross errors
Given a measurement vector z, the estimated state x will
depend on the projection operator P, which depends only
on the inner product choice (W ) and matrix H.
Matrix W is given by the inverse of the measurement
covariance matrix. After nding x , it is interesting to verify
the existence of gross errors in the measurements. Therefore
a routine for error detection is required. At this point, only
statistic concepts are needed.
Assuming that measurement errors have a normal
distribution, it is easy to show that index J (x), that is, the
function to be minimised in (2), has a chi-square distribution
(x2) with (m 2 n) degrees of freedom. Choosing a
probability 1 2 a of false alarm and being a the
signicance level of the test, a number C is obtained (via
chi-square distribution table for x2m2n,12a) such that, in the
presence of gross errors J (x) . C.
Another way to detect the presence of gross errors is by
the largest normalised residual test. Based on the same
assumption about measurement errors, the vector of
residuals r is normalised and subjected to a validation test:
r (k)N |r (k)|/sr(k) l (threshold value), where pr (k)N is
the largest among all r (i)N, i 1, . . . , m; sr(k) V(k, k)
is the standard deviation of the kth component of the
residuals vector, and V is the residual covariance matrix
given by V W 21 2 H(H TWH)21H T. If r(k)N . l, the
measurement with gross error is detected and the kth
measurement will be the one with gross error (usually
l 3 [4]).

IET Gener. Transm. Distrib., 2011, Vol. 5, Iss. 6, pp. 603 608
doi: 10.1049/iet-gtd.2010.0459