Calculating Sequence Impedances of Transmission

Line Using PMU Measurements

ZHOU, Huafeng; ZHAO, Xuanyu; *SHI, Di; ZHAO, Huashi; *JING, Chaoyang
Grid Control Center, China Southern Power Grid, Guangzhou, China
*
eMIT, LLC, Pasadena, CA 91107
Email: c.jing@myemit.com

Abstract—An online transmission line (TL) impedance
parameter identification system (TPIS) using PMU
measurements was recently developed and implemented at
China Southern Power Grid Company (CSG). Traditional
approaches for TL impedance calculation only approximate the
effect of conductor sags and ignore the dependence of
impedances on temperature variation. Utilizing PMU
measurements may improve the accuracy of TL parameters
calculation. The challenge is that the parameters identified are
very sensitive to noise and errors in PMU measurements, which
are difficult to quantify and can be uncertain under different
system operating/loading condition. TPIS provides an innovative
yet practical problem formulation for TL sequence parameter
estimation based on least-squares with linear constraints. A
bootstrapping-based resampling technique is developed and a
new metric is proposed to determine the credibility of the
estimated sequence impedances. This paper discusses the
proposed methodologies, challenges, as well as implementation
issues identified during the development of TPIS.
Index Terms—Sequence impedance, Transmission line, Phasor
Measurement Unit (PMU), Constrained optimization.
I. INTRODUCTION
Accurate knowledge of transmission line (TL) impedance
parameters is of vital importance in power system operations
and planning, such as state estimation, transient stability etc.,
and is used as the basis for protective relay settings.
Traditionally, TL parameters have been estimated based on
the tower geometries, conductor dimensions, estimates of
actual line length, conductor sag, and other factors [1]. These
calculated parameters are based on assumptions and
approximations. With the development of the PMU
technology, synchronized phasors offer the possibility of
allowing accurate estimation of transmission line impedance
parameters. China Southern Power Grid Company (CSG)
recently developed and implemented an online PMU-based
TL parameter identification system (TPIS).
CSG operates one of the world’s most complex systems.
CSG has the world’s first ±800kV DC transmission project.
Hybrid operation of AC and DC, west-east corridor with long-
distance and huge-capacity, complicated network structure
with renewable energy integration, striking problem on grid
stability, and great challenge on operation control all are the
features of the CSG’s grid. Towards the development of smart
grid and with increasing renewable integration, CSG has put
tremendous efforts to maintain grid reliability, minimize costs
for customers, and to create an open infrastructure that can
take advantage of evolving tools and grid devices. A critical
portion of these efforts are to set up and maintain an accurate
model of the system. With a large number of PMUs widely
installed in its system, CSG plans to validate, improve, and
update the network parameters in its EMS database using the
newly developed TPIS.
Extensive research has been conducted in identifying TL
impedances using synchrophasor measurements [1]-[6]. Four
methods based on TL linear/nonlinear models were presented
and compared in [1]. One two-port ABCD parameter based
method is proposed in [2]. This method utilizes two samples
of PMU measurements from each terminal of the TL to
identify the ABCD parameters, from which the impedance
parameters can be calculated. Paper [3] proposes a method
based on the distributed TL model and uses nonlinear
estimation theory to generate an optimal estimator of the fault
location as well as TL parameters. Paper [4] proposed to use
the full TL model for parameter identification if the line is not
fully transposed and system is unbalanced. An extended
Kalman Filter based approach was proposed in [5]. As pointed
out by [6], PMU-based TL parameter identification is very
demanding on the quality of the measurements. In other word,
small errors in the phasor measurements can lead to huge
errors in the estimated impedance parameters. Recently,
people started to pay more attention to PMU data quality
issues and lots of efforts have been spent in developing
methods for PMU calibration [7]-[9]. Generally speaking, the
biggest challenge in this field still lies in the fact that TL
parameter identification requires highly accurate PMU data
while the accuracy of PMU measurements under different
working/system conditions can be uncertain.
This paper proposes a new method for calculating TL
sequence impedances using constrained optimization
considering various physical constraints. It is recognized that
existing methods neglect to include physical constraints into

978-1-4673-8040-9/15/$31.00 ©2015 IEEE

the parameter identification process. During the development For a transmission line, both the impedance matrix Zabc
of TPIS, it is realized that including physical constraints can and Babc are symmetrical. Therefore, matrix of Gabc is also
be very helpful based on which we proposed a novel yet symmetrical, as shown below:
practical problem formulation and solution for such problems.
⎡ Ga Gab Gac ⎤
In addition, we propose a new metric that can be used to
determine the credibility of the calculated parameters, which is Gabc = ⎢⎢Gab Gb Gbc ⎥⎥ (10)
missing in the literature. This paper discusses the ⎢⎣Gac Gbc Gc ⎥⎦
methodologies, challenges, as well as several implementation
issues noticed during the development of TPIS. Rewriting equation (2) and (9) into matrix form:
⎡ΔVa ⎤ ⎡ Z a Z ab Z ac ⎤ ⎡ I a ⎤ ⎡ Ga Gab Gac ⎤ ⎡Va ⎤
S S
S Z abc R
⎢ S⎥ j ⎢ ⎢ S⎥
I abc I abc ⎢ΔV ⎥ = ⎢Z ⎥ ⎥
⎢ b ⎥ ⎢ ab Z b Z bc ⎥ ⎢ I b ⎥ − 2 ⎢Gab Gb Gbc ⎥ ⎢Vb ⎥ (11)
⎢⎣ΔVc ⎥⎦ ⎢⎣ Z ac Zbc Z c ⎥⎦ ⎢ I cS ⎥ ⎢⎣Gac Gbc Gc ⎥⎦ ⎢VcS ⎥
⎣ ⎦ ⎣ ⎦
1 1 1 R
V
S
Yabc ≈ j ⋅ Babc Yabc V abc ⎡∑ I a ⎤ ⎡ Ba Bab Bac ⎤ ⎡∑Va ⎤
abc 2 2 2 ⎢ I ⎥ = j ⎢B ⎥⎢ ⎥
⎢∑ b ⎥ 2 ⎢ ab Bb Bbc ⎥ ⎢ ∑Vb ⎥ (12)
⎢⎣ ∑ I c ⎥⎦ ⎢⎣ Bac Bbc Bc ⎥⎦ ⎢⎣ ∑Vc ⎥⎦

Fig. 1 Transmission line PI model where ΔVx = VaS − VaR , ∑ I x = I xS + I xR , ∑Vx = VxS + VxR
II. PROBLEM FORMULATION AND SOLUTION x = a, b, or c .
Further expanding equations (11)-(12) yields the
A. Measurement Model following:
A three-phase PI model for a general transmission line is j
ΔVa = Z a I aS + Z ab I bS + Z ac I cS − (GaVaS + GabVbS + GacVcS ) (13)
S R S R 2
shown in Fig. 1, where V , V , I , I abc abc abc represent the
abc
j
three-phase voltage and current phasor vectors at both ends of ΔVb = Z ab I aS + Z b I bS + Z bc I cS − (GabVaS + GbVbS + GbcVcS ) (14)
the line while Zabc and Yabc are the series impedance matrix 2
and shunt admittance matrix. Based on nodal analysis, the j
ΔVc = Z ac I aS + Z bc I bS + Z c I cS − (GacVaS + GbcVbS + GcVcS ) (15)
following equations can be written: 2
j j
∑ I a = ⋅ [Ba ∑Va + Bab ∑Vb + Bac ∑Vc ]
S R S S
V abc − V abc = Z abc I abc − ⋅ Z abc ⋅ B abc ⋅ V abc (1) (16)
2 2
j ⋅ B abc j
S R
I abc + I abc =
S R
⋅ (V abc + V abc ) (2) ∑ I b = ⋅ [Bab ∑ Va + Bb ∑ Vb + Bbc ∑ Vc ] (17)
2 2
j
Voltage and current variables in the equations are all ∑ I c = ⋅ [Bac ∑Va + Bbc ∑Vb + Bc ∑Vc ] (18)
phase-frame-of-reference quantities. In order to transform 2
these quantities to sequence-frame-of-reference quantities, we In equations (13)-(18), it is noticed that Gx (x=a, b, c, ab,
need to apply the phase-to-sequence transformation matrix, bc, ac) is a complex number, and we define Gx=Sx+jTx. In
which is defined as: order to obtain an optimal estimate of the impedances, we
⎡1 1 1⎤ expand these 6 complex equations into 12 real equations. To
⎢ ⎥ j120 0
and a = e (3) simplify the expression, the following definitions are made.
2
A = ⎢1 a a⎥
⎢⎣1 a a ⎥⎦ 2 Define X to be the measurement vector, which is known
and can be obtained from the PMU measurements.
The following relationships hold between the phase X = [ x1 , x2 ,..., x24 ]T
quantities and sequence quantities:
S ( R) S ( R)
= [Re(VaS ), Im(VaS ), Re(VbS ),..., Re(VaR ), Im(VaR ), (19)
V abc = A ⋅ V 012 (4) S S R R T
..., Re( I a ), Im(I a ),..., Re( I a ), Im(I a )...]
S (R) S (R)
I abc = A ⋅ I 012 (5) where Re(•) and Im(•) yield the real and imaginary parts of the
Z 012 = A−1Z abc A input arguments, respectively.
(6)
Define β to be the unknown parameter vector, which is
B012 = A−1 Babc A (7) composed of the unknown impedance parameters.
In order to simplify equation (1), we define matrix Gabc as: β = [ β1 , β 2 ,..., β 30 ]T = [ Ra , S a , X a , Ta ,..., Rab , S ab , X ab ,
(20)
G abc = Z abc ⋅ B abc (8) ..., Tac , Ba , Bb , Bc , Bab , Bbc , Bac ]T
So that equation (1) can be rewritten as: With the definition given in (19), we can further define a
S R S j S measurement vector:
V abc − V abc = Z abc ⋅ I abc − ⋅ G abc ⋅ V abc (9)
2
 Z = [ x1 − x7 , x2 − x8 , x3 − x9 , x4 − x10 , x5 − x11 , x6 − x12 , where αR, αX, and αBc are constants that define the error bands
(21) of the TL impedance parameters. It should be noted that it is
x13 − x19 , x14 − x20 , x15 − x21 , x16 − x22 , x17 − x23 , x18 − x24 ]T not necessary to include all the above inequality constraints
Based on the definitions above, we further arrange these since some impedance parameters may not be immediately
12 equations into matrix form as: available from the EMS database. Generally speaking, the
most constraints are included, the better the results will be. To
Z = H ⋅β (22) simplify the expression, equation (29)-(31) can be collectively
where H is a matrix formulated from the 12 equations written as:
referenced above and contains measurements while the lb j ≤ β j ≤ ubj j=1, 2, 3… (32)
measurement vector Z contains PMU voltage and current
measurements. where lbj and ubj represent the lower and upper boundaries
identified.
B. Problem Formulation and Solution Third, resistances are usually much smaller as compared to
It has been demonstrated in [1], [4], and [6] that TL reactance for a transmission line:
parameter identification is very sensitivity to the noise/error in β1 ≤ β3 (33)
the PMU measurements. For example, as shown in [1], a 1%
error in the voltage measurements can lead to over 20% error β5 ≤ β 7 (34)
in the calculated series resistance and reactance. In this
project, even negative series resistance has been observed
⋮
many times. β 21 ≤ β 23 (35)
One thing we figured out when developing TPIS was that
if physical constraints can be considered in the parameter A total of 6 additional constraints can be set up. Again, to
identification process, accuracy of parameter estimation can simplify the expression, these equation s are collectively
be greatly improved. First, from equation (8), the following written as:
equality constraint can be identified: g k (β ) ≤ 0 k=1, 2,…, 6 (36)
β 2 = β1 ⋅ β 25 + β13 ⋅ β 28 + β 21 ⋅ β 30 (23) Therefore, the TL parameter identification problem can be
formulated as a least-squares curve-fitting problem subject to
β 4 = β3 ⋅ β 25 + β15 ⋅ β 28 + β 23 ⋅ β 30 (24) both equality and inequality constraints, as shown below:
β14 = β1 ⋅ β 28 + β13 ⋅ β 26 + β 21 ⋅ β 29 (25) 1 2
min ⋅ H ⋅β −Z 2
β 2
⋮
s.t. f i (β ) = 0 i=1, 2,…, 12 (37)
β10 = β 21 ⋅ β30 + β17 ⋅ β 29 + β 9 ⋅ β 27 (26) g k (β ) ≤ 0 k=1, 2,…,6
β12 = β 23 ⋅ β 30 + β19 ⋅ β 29 + β11 ⋅ β 27 (27) lb j ≤ β j ≤ ubj j=1, 2, 3…

As shown above, a total of 12 equality constraints can be where ||•||22denotes the square of the L2-norm of the
added to the parameter estimation problem. To simplify the corresponding vector.
expression, these 12 equations can be collectively written as: Once β is calculated, the sequence impedances of the
transmission line can be calculated using equations (6)-(7).
f i (β ) = 0 i=1, 2,…, 12 (28)
C. Bad Data Detection
Second, although hand-calculated parameters only
approximate the sag effect and neglect the temperature The classical method described in [10]-[11] is used for bad
variation, it is believed that the true line parameters still stay data detection after solving the constrained least-squares.
within certain error band of the hand-calculated ones under Bad data identification is achieved by checking the
different system operating conditions. The line parameters normalized residuals of each measurement, which proceeds
stored in the EMS database can be used as additional as follows:
inequality constraints for the impedance parameters. Step 1) Solve the curve-fitting problem described in (32)
Assuming we can obtain the TL impedances stored in the and obtain the residual for each measurement point:
EMS database, it is reasonable to assume that the calculated ri = zi − H i ⋅ β, i = 1,2,..., N (38)
impedance should follow the constraints below:
Step 2) Compute the normalized residual as:
(1 − α R )RaEMS ≤ β1 ≤ (1 + α R ) ⋅ RaEMS (29)
(1 − α X )⋅ X aEMS ≤ β 2 ≤ (1 + α X ) ⋅ X aEMS (30)
(r ) i norm
=
ri
, i = 1,2,..., N (39)
Ω ii
⋮ where Ωii is the diagonal element of the matrix Ω
(1 − α Bc ) ⋅ BcEMS ≤ β 27 ≤ (1 + α Bc ) ⋅ BcEMS (31) Ω = H ( H T H ) −1 H T (40)
Step 3) Find the largest normalized residual rmaxNorm
and check
whether it is larger than a predefined identification threshold
c, for example 3.0:
Norm
rmax >c (41)
Step 4) If equation (41) does not hold, then no bad data will
be suspected; otherwise, the data sample corresponding to the
largest normalized residual is the bad data and should be
removed from the data set.
Step 5) If bad data is detected and removed from the data
set, the algorithm flow must return to step 1) and the process
above must be repeated. Otherwise, this process ends and
solutions are found.
III. ONLINE IMPLEMENTATION AND CREDITILITY METRIC
It is generally very difficult to determine the credibility of
the calculated TL parameters, mainly because in reality their
corresponding true values are unknown and there is no way to
make comparison. Most of the validation in existing research
is based simulation in which the true values are assumed, a
priori, to be known values and used throughout the simulation.
Having a metric that can quantify the credibility of the
calculated parameters is of critical importance in practice for
applications like such and so.
(a) Statistics for the calculated series resistance
(b) Statistics for the calculated series reactance
Fig. 2 Credibility metric for the calculated impedance parameters
In TPIS, for every five seconds, one set of impedance
parameters is generated for the TL under consideration. PMU
data collected during the five seconds are used to conduct
bootstrapping. Bootstrapping is one type of re-sampling
techniques that can be used to estimate the properties of an
estimator by sampling from an approximating distribution. By
resampling with replacement, we assume that each sample of
PMU measurements is independent and identically distributed.
We solve the problem described by equation (37) for each set
of sampled PMU data to get one set of TL impedance
parameters. Conduct the sampling many times so that we can
calculate the variances of the calculated parameters. Ratio of
the variance of the parameters to the corresponding value
stored in EMS database is used as the credibility metric for the
parameter identification. That is, the parameter obtained is
credible if the corresponding metric is smaller than a pre-
determined threshold:
σ ( x) ≤ ξ x x = R, X , or Bc (42)
where σ(x) stands for the variance of variable x.
Fig. 3 Online implementation of TPIS
During the implementation, two parameters are identified
to be critical, as discussed below:
• Number of bootstrapping samples: this number defines how
many times the program will acquire bootstrap samples from
the set of available data points. It is strongly recommended
that this number be larger than 30. As the number of
bootstrap samples increases, the variances associated with
the parameters decrease; however, execution time increases
as the number of bootstrap samples increases and selecting
arbitrarily large numbers may not be desirable. Tradeoff is
needed.
• Number of data points in each bootstrap sample: this number
determines how many data points are selected with each
bootstrap sample. Increasing this number may lead to greater
variance in the estimated parameters but will also increase
the computation time. Tradeoff is needed.
 Fig. 2 serves as an example to show the output of the TPIS
during one 5-second interval. And Fig. 3 shows the flowchart
for the online implementation of TPIS.
IV. NUMERICAL EXAMPLE
The TPIS has been implemented to a 525kV transmission
line at CSG. This line is named as “Lai Bin-Wu Zhou #2”,
which is the second transmission line connecting two cities
(Lai Bin and Wu Zhou) in Southern China. During a one-hour
period, we conducted the experiment and the calculated
parameters for the line are shown below in Fig. 4-Fig. 6.
14
13
12
V. CONCLUSION
PMU has the potential to improve the accuracy of
transmission line impedances. More accurate parameter means
better power system modeling, faster and more accurate fault
location as well as more economic system operation. China
Southern Power Grid has developed an online TL parameter
identification system. It was found that if the physical
constraints associated with the TL were included, accuracy of
the calculation can be greatly improved. Using bootstrapping,
multiple sets of PMU measurements can be obtained so as to
identify the credibility of the calculated parameters based on a
proposed metric. A novel method is presented to calculate the
TL sequence impedances.
-4
x 10
8

11
R (Ohms)

7
10

9 6

Bc (Siemens)
8
5

7
4
6
0 10 20 30 40 50 60
Time (min)
3
Fig. 4 Calculated series Resistance
2
60 0 10 20 30 40 50 60
Time (min)
59

58
Fig. 6 Calculated shunt susceptance
57

56 VI. REFERENCE
X (Ohms)

55
[1] D. Shi, D. J. Tylavsky, N. Logic, and K. M. Koellner, “Identification of
54
Short Transmission-line Parameters from Synchrophasor
53 Measurements,” 40th North American Power Symposium, Calgary,
52
2008.
[2] R. E. Wilson, G. A. Zevenbergen, etc., “Calculation of Transmission
51
Line Parameters from Synchronized Measurements,” Electric Machines
50
0 10 20 30 40 50 60
& Power Systems, vol. 27, no. 12, pp. 1269-1278, 1999.
Time (min) [3] Y. Liao, M. Kezunovic, “Optimal estimate of transmission line fault
Fig. 5 Calculated series reactance location considering measurement errors,” IEEE Transactions on
As can be seen from the plots, the impedance parameters Power Delivery, vol. 22, no. 3, pp. 1335-1341, 2007.
[4] D. Shi, D. J. Tylavsky, K. M. Kollner, etc., “Transmission Line
do not vary much during the experiment period. In addition, as Parameter Identification Using PMU Measurements,” European
shown in Table 1, the calculated series impedance differs Transactions on Electrical Power, vol. 21, no. 4, pp. 1574-1588, May
from the EMS database by about 7.8% while the shunt 2011.
[5] E. Janecek, P. Hering, P. Janecek, and A. Popelka, “Transmission Line
susceptance differs from the database by 7.7%. These Identification Using PMUs,” 10th International Conference on
differences could be due to the inaccurate estimate of the line Environment and Electrical Engineering (EEEIC), pp. 1-4, May 2011.
length as well as the specific loading conditions, and other [6] D. Shi, Utilizing Synchrophasor Technology to Determine
Transmission Line Impedance Parameters, MS Thesis, Arizona State
ambience conditions such as temperature, wind speed, etc. at University, 2009.
the moment when this experiment was conducted. [7] D. Shi and D. J. Tylavsky, “An Adaptive Method for Detection and
Correction of Errors in PMU Measurements,” IEEE Trans. Smart Grid,
TABLE 1 COMPARISON BETWEEN CALCULATION AND EMS DATABASE vol. 3, no. 4, pp. 1575-1583, Aug. 2012.
Quantity EMS Calculated Difference [8] Q. Zhang, et al., “The integrated calibration of synchronized phasor
Database (%) measurement data in power transmission systems,” IEEE Trans. Power
Delivery, vol. 26, no. 4, pp. 2573-2581, Oct. 2011.
Series Impedance |Z| 23.8 25.8 7.8%
[9] M. Zhou, et al., “Calibrating Instrument Transformers with Phasor
(Ohms) Measurements,” Electric Power Components and Systems, vol. 40, no.
Shunt Susceptance 3.6E-04 3.9E-04 7.7% 14, pp. 1605-1620, 2012.
Bc (S) [10] D. C. Montgomery, E. A. Peck, G. G. Vining, Introduction to Linear
Regression Analysis, New York: John Wiley & Sons, Inc., 2006.
[11] A. Abur and A. G. Exposito, Power System State Estimation—Theory
and Implementation, New York: Marcel Dekker, 2004.