Вы находитесь на странице: 1из 8

1

Identification of Short Transmission-Line


Parameters from Synchrophasor Measurements
Di Shi, Student Member, IEEE, Daniel J. Tylavsky, Senior Member, IEEE, Naim Logic, Member
IEEE, and Kristian M. Koellner, Member, IEEE

based on the distributed line model and the non-linear


Abstract—Accurate knowledge of transmission line impedance estimation theory. In [3], the authors proposed a method to
parameters helps to improve accuracy in relay settings, post- estimate the zero and positive sequence impedance from the
event fault location and transmission power flow modeling. Four synchronized measurements.
methods are presented in this paper to identify transmission line
All these methods assume that the synchrophasor
impedance parameters from synchronized measurements for
short transmission lines. Estimates of parameters for short measurements are obtained without noise and bias errors. In
transmission lines is more challenging than for long transmission reality, noise is present in all measurements. PMU
lines since measurement noise often causes large errors in the measurements however do have greater accuracy than other
estimates. The effectiveness of these methods is verified through measurements. PMU data are time tagged with an accuracy of
simulations. These simulations incorporate two types of better than 1 microsecond and magnitude accuracy that is
measurement errors: biased and non-biased noise. The different
better than 0.1%. However, this potential performance is not
effects of bias errors and random noise on the accuracy of the
calculated impedance parameters are quantified. Last, some achieved in an actual field installation due to errors from
complicating factors and challenges inherent in real world instrumentation channels and system imbalances [4].
measurements are discussed. Because bad data can be found in PMU measurement data
sets, it is advantageous to screen for bad data points. Some
Index Terms—Transmission line impedance parameters, bad data in the synchrophasor measurements can be removed
Phasor measurement unit (PMU), Synchrophasor, ATP by scanning the data for values obviously out of range. More
sophisticated techniques using statistical methods are also
important. Many times, however, it is difficult and may even
I. INTRODUCTION be impossible to filter out all data corrupted with noise and
bias errors. For long transmission lines, noise in the

T RANSIMSSION line impedance parameters are


traditionally calculated using the geometry of the
transmission line conductors. These calculations are
measurements is generally not a problem; however for short
transmission lines the noise in the measurements is more
problematic and can yield parameters that are erroneous. In
known to be approximate. Besides the inaccuracy brought this paper we develop methods that are robust with respect to
into the calculations by approximating the effect of sag, the measurements with random noise and bias errors.
impedance parameters themselves are changing with the This paper compares four methods that can be employed to
ambient temperature and load conditions. Presently there is estimate the transmission line impedance parameters:
potential to make the transmission line impedance parameters y Single measurement method
more accurate by directly measuring the parameters online y Double measurement method
using the synchronized phasor measurements. y Multiple measurement method using linear
Several methods have been developed in the past by regression
researchers to calculate the impedance parameters of the y Multiple measurement method using non-linear
transmission line [1]-[3] using synchronized measurements. R. regression
E. Wilson in [1] proposed that the ABCD chain parameters of Of these methods, the last two are novel.
the transmission line can be obtained by appropriate use of To test the methods, a simulation is created in ATP to
synchronized voltage and current measurements from both provide noise-free phasor measurements. Then reasonable
ends of a transmission line. The impedance parameters can noise and bias errors are added to the simulated synchrophasor
then be obtained from the chain parameters. In [2], the authors measurements. These noisy measurements are then used to
proposed estimating the impedance parameters per unit length estimate the impedance parameters of the transmission line for
each of the four methods. Influences of the inaccurate
This work was supported by funding provided by Salt River Project. measurements on the algorithms are quantified and the
Di Shi (Di.Shi@asu.edu) and D.J. Tylavsky (tylavsky@asu.edu) are with comparisons among these methods are made.
the Department of Electrical Engineering, Arizona State University, Tempe,
AZ 85287-5706 USA In addition to the inaccurate measurements problem, real
Naim Logic (Naim.Logic@srpnet.com) and Kristian M. Koellner world challenges are discussed in this paper. These challenges
(kmkoelln@srpnet.com) are with Salt River Project, Phoenix, AZ 85072- include phase angle shifting, untransposed lines with
2025
unbalanced loads and mutual coupling between transmission
978-1-4244-4283-6/08/$25.00 ©2008 IEEE
2

lines. transformation.
We have chosen in this work to add random and biased
II. SIMULATION METHOD noise to the sequence components.

The transmission line studied is the Thunderstone-Santan, a


III. METHOD #1: SINGLE MEASUREMENT METHOD
11.97-mile, 230kV AC transmission line located in Arizona in
the United States. Two relay-based phasor measurement units
are installed on this transmission line, one at Thunderstone For short transmission lines, the nominal pi circuit, shown
and the other at Rogers. in the Fig. 2, is the accepted model. In this figure the phasor
This transmission system is modeled in ATP as shown in values VS ,V R and I S , I R denote the sending and receiving end
Fig. 1. The line is a short transmission line and is modeled by voltages and currents respectively.
the nominal pi circuit. In the following sections, the
Thunderstone terminal is referred to as the “sending end”
while the Santan terminal is labeled the “receiving end”. In the
ATP simulation, two sets of simulated PTs and CTs are placed
in both the sending and receiving ends to sample the voltages
and currents simultaneously. In this simulation, the output of
Fig. 2 Nominal Pi circuit for short transmission line
the CTs and PTs are real-time data, which are further
processed by a DFT to obtain the phasor components. Since
we desire the positive sequence impedance parameters of the For this circuit, through nodal analysis, the following
transmission line, those phasor components are further equations are derived:
processed to obtain the sequence components. Y Y
I S − VS ⋅ + I R − V R ⋅ = 0 (1)
2 2
Y
V S − Z ⋅ ( I S − VS ⋅ ) − V R = 0 (2)
2
where
Z = R + jX (3)
Y = jBC (4)
Fig. 1 Transmission system model in ATP
where BC is the shunt susceptance of the line.
Because some of the methods we propose require multiple Given the known quantities, VS ,VR , I S and I R measured by
measurements under different loading conditions, a time the PMUs installed at the two terminals of the line, we can
varying load is modeled on the receiving end. The load curve solve (1) and (2) for the two unknowns, Z and Y, and to get:
is set to be sinusoidal with a period of 24 hours. It varies VS2 − VR2
between one fifth and a half of the maximum line capacity. To Z= (5)
I S ⋅ V R − I R ⋅ VS
make sure that each sample represents a different load
condition, samples are taken every five minutes. IS + IR
Y = 2⋅ (6)
In [4], accuracy of the synchrophasor measurement system VS + VR
is characterized. This characterization shows that compared to This is the simplest method to implement among the four
the PMU, the cables and instrument transformers are the main methods presented in this paper. Applying this method to the
sources of random noise and bias errors. noise-free synchrophasor measurements generated by the ATP
In this work, two types of measurement noise are simulation, the results shown in TABLE I are obtained.
considered: the random noise and bias errors. By bias errors
we mean the measurements that are constantly and TABLE I
consistently in error by a fixed amount. Random noise, as the IMPEDANCE PARAMETERS CALCULATED BY METHOD #1
name implies, is random and in our simulations is normally
distributed with zero mean and1% deviation. The bias error in Quantity Actual values Calculated values Error
our simulations has a magnitude of 1% of the mean value of (%)
the current or voltage experienced over a 24 hour cycle. R 1.50765 1.5061 0.003%
There exist three places where noise can be injected into X 9.22047 9.22125 0.008%
our simulations: BC 6.76748E-05 6.76806E-05 0.009%
• noise added to the time domain samples from the
ATP simulation; As indicated by the simulation results, this method works
• noise added to the phasor values obtained from the well with noise-free measurements.
DFT; Both random noise and bias errors as discussed in part II
• or noise added to the sequence components we were then added to the noise-free synchrophasor
obtain from the symmetrical component measurements to test the performance of this method under
3

noisy conditions. Instead of giving absolute numbers, the VS = A ⋅ V R + B ⋅ I R (7)


results are given in the form of “A” for acceptable and “U” for I S = C ⋅ VR + D ⋅ I R (8)
unacceptable. The performance of the estimated parameters is
If the PMUs at the two terminal of the line generate two
determined to be “Acceptable” if the parameters are within a
different (independent) sets of measurements for different load
reasonable band (i.e. +/-10% for X and BC, and +/-20% for R);
conditions, the ABCD chain parameters can be determined by
otherwise the result were considered unacceptable. See TABLE
the following four equations:
II for the results.
VS1 = A ⋅ VR1 + B ⋅ I R1 (9)
TABLE II I S1 = C ⋅ VR1 + D ⋅ I R1 (10)
PERFORMANCE OF METHOD #1 WHEN NOISE AND BIAS ERROR
APPEAR IN THE MEASUREMENTS VS 2 = A ⋅ V R 2 + B ⋅ I R 2 (11)
I S 2 = C ⋅ VR 2 + D ⋅ I R 2 (12)
Calculate Calculated Calculated where
R X BC V S1 ,V R1 , I S1 , I R1 : Phasors from measurement 1
Bias VS U U A VS 2 , V R 2 , I S 2 , I R 2 : Phasors from measurement 2
error U U A
VR Solving these four complex equations in four unknown
in
IS A A U parameters using Cramer’s Rule gives:
I ⋅ V − I R 2 ⋅ V S1
IR A A U A = R1 S 2 (13)
det
Random VS U U A
noise V ⋅ V − VR1 ⋅ VS 2
VR U U A B = S 2 S1 (14)
in det
IS A A U I ⋅ I − I R 2 ⋅ I S1
C = R1 S 2 (15)
IR A A U det
U—unacceptable A—acceptable I ⋅ V − I S 2 ⋅ VR1
D = S1 R 2 (16)
det
As shown by the simulation results in TABLE II, some where
parameter estimates are acceptable for certain types of noise det = I R1 ⋅ VR 2 − I R 2 ⋅ VR1
and unacceptable for other types of noise. For example, in the
Once the chain parameters are calculated, the impedance
first row of Table II, with noise added to the measurements of
parameters can be calculated directly using the following
sending end voltage, the series resistance and series reactance
relationships:
calculated will be unacceptable while the shunt susceptance
will be acceptable. This indicates that the series resistance and A = 1 + 0.5 ⋅ Y ⋅ Z (17)
reactance are very sensitive to noise and bias errors in the B=Z (18)
sending end voltage while the shunt susceptance is not C = Y ⋅ (1 + 0.25 ⋅ Y ⋅ Z ) (19)
sensitive to VS. Similar observations can be made for the other
rows in the table. D = 1 + 0.5 ⋅ Y ⋅ Z (20)
Fig. 3 shows the performance in the algorithm when There are two redundant equations since we only have 2
random noise in the range of +/-1% is added to the sending unknowns, Z and Y. Combining the first two equations we get
end voltage measurement. This figure shows that this can lead the solution:
to a 7% error in the series reactance. Then for this same Z=B (21)
amount of noise the error in the R can be as large as 160%. 2 ⋅ ( A − 1)
Y= (22)
B
Appling this method to the simulated noise-free
synchrophasor measurements generated by the ATP
simulation, we get the impedance parameters shown in TABLE
III.

Fig. 3 Sensitivity analysis for calculated X to noise in sending end voltage TABLE III
using method #1 IMPEDANCE PARAMETERS CALCULATED BY METHOD #2

IV. METHOD #2: DOUBLE MEASUREMENT METHOD Quantity Actual values Calculated values Error
(%)
Two-port ABCD parameters can be used to represent the R 1.50765 1.50758 0.005%
transmission line in the most general form. The following two X 9.22047 9.22127 0.009%
equations can be derived according to Figure 2: BC 6.76748E-05 6.76559E-05 0.03%
4

Again the simulation results show that this method works Im[V S ] = Re[ A] ⋅ Im[V R ] + Im[ A] ⋅ Re[V R ]
well in the ideal conditions. We next added noise and bias + Re[ B] ⋅ Im[I R ] + Im[B ] ⋅ Re[ I R ] (24)
errors to the synchrophasor measurements to test the
Re[ I S ] = Re[C ] ⋅ Re[VR ] − Im[C ] ⋅ Im[VR ]
sensitivity of this method to noise. The results shown in TABLE
IV were obtained:
+ Re[D] ⋅ Re[I R ] − Im[D] ⋅ Im[I R ] (25)
Im[I S ] = Re[C ] ⋅ Im[VR ] + Im[C ] ⋅ Re[VR ]
TABLE IV
PERFORMANCE OF METHOD #2 WHEN NOISE AND BIAS ERROR + Re[D] ⋅ Re[I R ] + Im[D] ⋅ Re[ I R ] (26)
APPEAR IN THE MEASUREMENTS where Re(.) and Im(.) yield the real and imaginary part of the
input argument, respectively.
Calculated Calculated Calculated We collectively write these four equations into the matrix
R X BC format and get:
Bias VS A A U ⎡Re[VS ]⎤ ⎡Re[V R ] − Im[V R ] Re[ I R ] − Im[I R ]⎤
error A A U ⎢ Im[V ]⎥ = ⎢ Im[V ] Re[V ] Im[I ] Re[ I ] ⎥
VR ⎣ S ⎦ ⎣ R R R R ⎦
in
IS A A A ⎡Re[ A]⎤
⎢ Im[ A]⎥
A A A
IR ⋅⎢ ⎥ (27)
⎢Re[ B]⎥
Random VS U U U
⎢ ⎥
noise
VR U U U ⎣ Im[B]⎦
in
IS A A A ⎡Re[ I S ]⎤ ⎡Re[VR ] − Im[V R ] Re[ I R ] − Im[ I R ]⎤
⎢ Im[I ]⎥ = ⎢ Im[V ] Re[V ] Im[I ] Re[ I ] ⎥
IR U U U ⎣ S ⎦ ⎣ R R R R ⎦

U—unacceptable A—acceptable ⎡ Re[C ]⎤


⎢ Im[C ] ⎥
⋅⎢ ⎥ (28)
Compared with method 1, this method works much better ⎢Re[ D]⎥
for bias errors although it still is very sensitive to the random ⎢ ⎥
⎣ Im[D]⎦
noise as shown in Fig. 4. This figure shows that a 1% random
noise in the sending end voltage will cause the calculated Suppose N measurements have been collected from the
series reactance to be in error by 210%. phasor measurement units. Define the following:
⎡ Re[VS1 ] ⎤
⎢ Im[V ] ⎥
⎢ S1 ⎥
E = ⎢Re[VS 2 ]⎥ (29)
⎢ ⎥
⎢ Im[VS 2 ]⎥
⎢ # ⎥
⎣ ⎦
Fig. 4 Sensitivity analysis for calculated X to noise in sending end voltage ⎡ Re[VR1 ] − Im[VR1 ] Re[I S1 ] − Im[I S1 ] ⎤
using method #2
⎢ Im[V ] Re[V ] Im[I ] Re[I ] ⎥
⎢ R1 R1 S1 S1 ⎥
Note that the random noise introduced in the sending end H= ⎢ # # # # ⎥ (30)
current measurements does not affect our calculations because ⎢ ⎥
(13) and (14) are not functions of the measurement Is. The ⎢Re[VRN ] − Im[VRN ] Re[I SN ] − Im[I SN ]⎥
⎢Re[VRN ] Re[VRN ] Im[I SN ] Re[I SN ] ⎥
improved performance of method 2 over method 1 when the ⎣ ⎦
input data has bias errors suggests that redundancy in the ⎡Re[ A]⎤
measurements may improve the parameter estimation. ⎢ Im[ A]⎥
Consequently, we developed two multiple measurement F=⎢ ⎥ (31)
⎢Re[ B]⎥
methods for parameter estimation. These methods are ⎢ ⎥
described below. ⎣ Im[B]⎦
Using the unbiased least square estimator, the best
V. METHOD #3: MULTIPLE MEASUREMENT METHOD USING estimation of the chain parameters A and B are found to be:
LINEAR REGRESSION F = ( H T H ) −1 H T E (32)
Equations (7) and (8) are two complex equations with 6 Similarly, best estimated values of C and D can be found.
complex variables. Expanding them into four real equations Using equations (21), (22), (31) and (32), the impedance
we get: parameters can be calculated.
Re[V S ] = Re[ A] ⋅ Re[V R ] − Im[ A] ⋅ Im[V R ] Appling this method to the synchrophasor measurements
+ Re[ B] ⋅ Re[ I R ] − Im[B] ⋅ Im[I R ] (23) generated by our simulation (without added noise) we get the
impedance parameter results are shown in TABLE V. Under
5

these ideal conditions, this method generates very accurate Fig. 5 shows the error in the series reactance, X, as a
estimates of the impedance parameters. function of the number of redundant measurements. As
shown by Fig. 5, the calculated series reactance is fairly
TABLE V
IMPEDANCE PARAMETERS CALCULATED BY METHOD #3
accurate when 1% random noise exists in the sending end
voltage measurements provided we have a sufficient number
Quantity Actual values Calculated Error of samples. (Since sampling of PMU data is frequent and
values (%) continuous, a large number of samples are readily available.)
R 1.50765 1.5061 0.003% The calculated X values are well within the 10% error band
X 9.22047 9.22125 0.008% (the two straight lines in Fig. 5 denote the ± 10% error band).
BC 6.76748E-05 6.76806E-05 0.009% Compared with the double measurement method and the
single measurement method, this approach is clearly superior;
although there remains a problem with estimating the series
To test the performance of this method when noise and bias resistance.
errors exist in the synchrophasor measurements, such errors
were added to the measurements and the VI. METHOD #4: MULTIPLE MEASUREMENT METHOD USING
acceptability/unacceptability of the results is reported in NON-LINEAR REGRESSION
TABLE VI. The accuracy of the parameters estimated by this
method is a function of the number of data points sampled. To Method 3 uses the chain parameter description of the two-
obtain the results in TABLE VI 5600 observations sampled port transmission line model. A formulation of the problem
every 5 minutes were used. that includes the impedance parameters directly used the nodal
analysis description in (1)-(4). Combine equations (1) – (4)
TABLE VI and rewriting them into four real equations gives:
PERFORMANCE OF METHOD #3 WHEN NOISE AND BIAS ERROR 1
APPEAR IN THE MEASUREMENTS Re(VS ) − R ⋅ Re( I S ) + X ⋅ Im(I S ) − ⋅ BC ⋅ R ⋅ Im(VS )
2
Calculate Calculated Calculated 1
− ⋅ BC ⋅ X ⋅ Re(VS ) − Re(VS ) = 0 (33)
R X BC 2
Bias VS A A A 1
Im(VS ) − R ⋅ Im(I S ) − X ⋅ Re( I S ) + ⋅ BC ⋅ R ⋅ Re(VS )
error 2
VR A A A
in 1
IS A A A − ⋅ BC ⋅ X ⋅ Im(VS ) − Im(VR ) = 0 (34)
2
IR A A A 1
Re( I S ) + ⋅ BC ⋅ Im(VS ) + Re( I R )
Random VS U A A 2
noise U A A 1
VR + ⋅ BC ⋅ Im(VR ) = 0 (35)
in 2
IS A A A
1
IR A A A Im(I S ) − ⋅ BC ⋅ Re(VS ) + Im(I R )
2
U—unacceptable A—acceptable 1
− ⋅ BC ⋅ Re(VR ) = 0 (36)
2
As indicated by the table, this method works better than where Re(.) and Im(.) yield the real and imaginary part of the
either Method 1 or 2 for bias errors, which means that accurate input argument, respectively.
impedance parameters can be obtained even when there are
Let x = [ x1 , x2 , ⋅ ⋅ ⋅, x7 , x8 ] be the vector of measurements and
small bias errors in the synchrophasor measurements. When
random noise is present, estimates of the series reactance and define the following:
shunt susceptance are always acceptable; however estimates of x1 = Re(VS ) , x 2 = Im(VS ) ,
the series resistance are unacceptable when random noise is x3 = Re(I S ) , x 4 = Im(I S ) ,
present in the voltage measurements. x5 = Re(VR ) , x 6 = Im(V R ) ,
x7 = Re( I R ) , x8 = Im(I R ) .
Let θ = [θ1 ,θ 2 , θ 3 ] represent the vector of the unknown
parameters: R, X and BC , respectively.
Based on equations (33)-(36) and the definition of x and θ ,
we define the following functions respectively:

Fig. 5 Calculated series reactance when 1% noise exists in the sending end
voltage as a function of sample number using method #3
6

1 ⎡ ∂f11 ∂f11 ∂f11 ⎤


f1 ( x,θ ) = x1 − θ1 ⋅ x5 + θ 2 ⋅ x6 − ⋅ θ1 ⋅ θ 3 ⋅ x2 ⎢ ⎥
2
⎢ ∂θ1 ∂θ 2 ∂θ 3 ⎥
1 ⎢ ∂f 21 ∂f 21 ∂f 21 ⎥
− ⋅ θ 3 ⋅ θ 2 ⋅ x1 − x3 (37)
2 ⎢ ⎥
1 ⎢ ∂θ1 ∂θ 2 ∂θ 3 ⎥
f 2 ( x,θ ) = x2 − θ1 ⋅ x6 − θ 2 ⋅ x5 + ⋅ θ1 ⋅ θ 3 ⋅ x1 ⎢ ∂f 31 ∂f 31 ∂f 31 ⎥
2 ⎢ ⎥
1 ⎢ ∂θ1 ∂θ 2 ∂θ 3 ⎥
− ⋅ θ 2 ⋅ θ 3 ⋅ x2 − x4 (38) ⎢ ∂f 41 ∂f 41 ∂f 41 ⎥
2 ⎢ ⎥
1 1 ∂F ( x,θ ) ⎢ ∂θ1 ∂θ 2 ∂θ 3 ⎥
f 3 ( x, θ ) = x5 + ⋅ θ 3 ⋅ x2 + x7 + ⋅ θ 3 ⋅ x4 (39) H= =⎢ # # # ⎥ (44)
2 2 ∂θ θ =θ k ⎢ ∂f1N ∂f1N ∂f1N ⎥
1 1 ⎢ ∂θ ∂θ 2 ∂θ 3 ⎥
f 4 ( x,θ ) = x6 − ⋅ θ 3 ⋅ x1 + x8 − ⋅ θ 3 ⋅ x3 (40) ⎢ 1 ⎥
2 2 ⎢ ∂f 2N ∂f 2N ∂f 2N ⎥
We define F ( x , θ ) to be a vector function which is ⎢ ∂θ ⎥
⎢ N1 ∂θ 2 ∂θ 3 ⎥
composed of f1 ~ f 4 . Assuming N samples are available, ⎢ ∂f 3 ∂f 3N ∂f 3N ⎥
F ( x ,θ ) is defined by: ⎢ ∂θ ∂θ 2 ∂θ 3 ⎥
⎢ N1 ⎥
⎡0⎤ ⎡ f11 ( x,θ ) ⎤ ⎡ ε11 ⎤ ⎢ ∂f 4 ∂f 4N ∂f 4N ⎥
⎢0⎥ ⎢ 1 ⎥ ⎢ 1⎥ ⎢ ∂θ ∂θ 2 ∂θ 3 ⎥
⎣ 1 ⎦
⎢ ⎥ ⎢ f 2 ( x,θ ) ⎥ ⎢ ε 2 ⎥
⎢0⎥ ⎢ f 31 ( x,θ ) ⎥ ⎢ ε 31 ⎥ where
⎢ ⎥ ⎢ 1 ⎥ ⎢ 1⎥ θ k ,θ k +1 are the variable vectors before and after the
⎢0⎥ ⎢ f 4 ( x,θ ) ⎥ ⎢ ε 4 ⎥ th
⎢#⎥ = ⎢ # ⎥ + ⎢ # ⎥ = F ( x,θ ) + ε (41) k iteration respectively;
⎢ ⎥ ⎢ N ⎥ ⎢ ⎥
N Δθ k is the variable vector update for the k th iteration.
⎢0⎥ ⎢ f1 ( x,θ )⎥ ⎢ε1 ⎥
⎢0⎥ ⎢ N ⎥ ⎢ N⎥ and H is the Jacobian matrix of F ( x,θ ) with respect to the
⎢ ⎥ ⎢ f 2 ( x,θ )⎥ ⎢ε 2 ⎥
⎢0⎥ ⎢ f 3N ( x,θ )⎥ ⎢ε 3N ⎥ unknown variables θ .
⎢ ⎥ ⎢ N ⎥ ⎢ N⎥ The iteration process was terminated once the variable
⎣0⎦ ⎣⎢ f 4 ( x,θ )⎦⎥ ⎣⎢ε 4 ⎦⎥ update is smaller than a specified tolerance. The initial values
where ε is a random error component that is assumed to have of the unknown parameters were set to the estimated value of
mean zero and unknown variance. the parameters of the transmission line calculated using
To solve (41) in a least squares sense, we used the non- classical techniques.
linear regression technique [5]. The solutions can be obtained With simulated measurements free of noise, we used the
following an iterative process. In the k th iteration, the iterative procedure described above to estimate the
unknowns are updated following the equations below: transmission line parameters and validate the approach. The
θ k +1 = θ k + Δθ k (42) results, which are given in Table VII below, show that the
actual and calculated values match well.
Δθ k = ( H T H ) −1 H T [− F ( x, θ k )] (43)
TABLE VII
IMPEDANCE PARAMETERS CALCULATED BY METHOD #4

Quantity Actual values Calculated values Error


(%)
R 1.50765 1.50760 0.003%
X 9.22047 9.22126 0.009%
BC 6.76748E-05 6.76806E-05 0.009%
7

TABLE VIII In addition to measurement noise, other real world


PERFORMANCE OF METHOD #4 WHEN NOISE AND BIAS ERROR APPEAR IN THE
MEASUREMENTS
challenges exist. These include: phase angle shifting, mutual
coupling problems between transmission lines, and
Calculate Calculated Calculated unbalanced load conditions on not fully transposed lines.
R X BC These challenges are discussed below.
Bias VS A A A A. Filtering Phase Angle Data
error A A A Fig. 7 shows recorded phase angle data from one PMU
VR
in installed at one terminal of the Thunderstone-Santan line.
IS A A A
IR A A A
Random VS U A U
noise U A U
VR
in
IS A A U
IR A A U
U—unacceptable A—acceptable

In the ideal case with perfect measurements, this method


generates accurate impedance parameters. When random noise
and bias errors are present in the measurements, as shown in
Table VIII, series reactance estimates are satisfactory. Series Fig. 7 Recorded phase angle data from PMU
resistance and shunt susceptance estimates are unacceptable From this figure, it is seen that the phase angle often makes
when random noise is present, that is, these parameters are abrupt jumps, jumps that cannot be explained by conditions
sensitive to random noise in the measurements. within the network. The phase angle shifting problem is
As with the linear regression method, the performance of known to occur and its cause is well understood as described
this method is a function of the level of redundancy in the in reference [7]. It is difficult to design a filter that can identify
measurements. Fig. 6 shows the error in the series reactance and eliminate these “jumps” in the data for a number of
estimate as a function of number of observations. This figure reasons.
shows that as the sample number increases, accurate
calculated series reactance is obtained when there is random B. Untransposed Lines
noise in the sending end voltage measurements. The two Transmission lines are often not transposed or not fully
straight lines denote a ± 10% error band of the true value of X. transposed. For fully transposed transmission lines,
unbalanced loads do not affect the calculations since the three
phase components are fully decoupled. However, for lines that
are untransposed or not fully transposed, there will be mutual
coupling between the three sequence components, which, as a
result, will cause the calculated sequence impedance
parameters to be inaccurate. Some of our elementary studies
and simulation results suggest that, for the untransposed
Thunderstone-Santan line, 10% unbalance in the currents may
Fig. 6 Calculated series reactance when 1% noise exists in the sending end cause more than 10% error in the calculation of series
voltage as a function of sample number using method #4
resistance and more than 1% percent error in series reactance.
VII. REAL WORLD CHALLENGES C. Mutual Coupling
Identification of short-transmission-line parameters from All of the research in the literature that aims at estimating
the synchrophasor measurements requires narrow bounds on parameters from measured PMU data assumes that no nearby
the accuracy of the data. Currently, the precision of the data lines exist. In many cases, such as the case with the
from PMUs installed at substations is affected by various Thunderstone-Santan line, a lower voltage line under built on
equipment and communication channels within the substation the transmission towers induces voltages into the line we wish
[6]. Some bad data in the measurements can always be to study. This induced voltage adds noise into our data which
detected and removed by statistically based methods while varies with time and is neither strictly random nor biased.
small amounts of noise and even bias error in the data are very Dealing with this type of measurement contamination is one
difficult to eliminate. And even small amounts of noise and goal of our future work.
bias errors may cause inaccuracy problems when estimating
VIII. CONCLUSIONS
the model parameters of short transmission line.
Methods for estimating the transmission line parameters
from PMU measurements for long lines are well known. For
8

short lines—and when the measurements are noisy— Daniel J. Tylavsky (SM’ 88) received the B.S., M.S.E.E., and Ph.D. degrees
in engineering science from the Pennsylvania State University, University
traditional methods do not work well. In this paper we have
Park, in 1974, 1978, and 1982, respectively.
examined two traditional methods for estimating the From 1974 to 1976, he was with Basic Technology, Inc., Pittsburgh, PA,
transmission line parameters and proposed two new methods. and from 1978 to 198, he was an Instructor of electrical engineering at
Simulations results have also shown that unbiased noise and Pennsylvania State. In 1982, he joined the Faculty in the Electrical
Engineering Department of Arizona State University, Tempe, where he is
bias errors are very critical to the accuracy of the parameters
currently Associate Professor of engineering.
calculated. Dr. Tylavsky is a member of IEEE Power Engineering Society and
Among the four methods presented in this paper, the Industry Applications Society, and is an RCA Fellow, NASA Fellow, and
multiple measurement method using linear regression is member of Phi Eta Sigma, Eta Kappa Nu, Tau Beta Pi, and Phi Kappa Phi.
superior. This method has acceptable performance when
Naim Logic graduated from the University of Sarajevo, received his M. Sc.
calculating the series reactance and shunt susceptance from degree from the University of Zagreb, and Ph. D. degree from Arizona State
phasor measurements that have random noise and bias errors University.
present. However, the series resistance calculated is still very Dr. Logic was with the Department of Electrical Engineering at the
University of Sarajevo and Arizona State University. He is currently with the
sensitive to the random noise, which means small random
Power System Operations Department – Computer Applications Group of Salt
noise in the voltage phasors will cause the series resistance to River Project (SRP), Phoenix, AZ.
be out of the acceptable range (+/-20%). His research interests are in the area of power system state estimation,
In the later part of the paper, some real world challenges application of synchronized phasor measurements, and power system
reliability.
are discussed. These challenges include problems with phase
Dr. Logic is SRP’s advisor for EPRI’s (Electric Power Research Institute)
angle shifting, accommodating the influence of untransposed Situational Awareness Task Force, and a member of the Western Electric
lines with unbalanced loads and accommodating the effects of Coordinating Council (WECC) Reliability Tools Working Group. He is a
mutual coupling. Work is ongoing to investigate strategies for member of the IEEE - Power and Energy Society (PES) and the Phoenix
Chapter treasurer.
mitigating these problem as well as strategies for improving
Dr. Logic is also a registered Professional Engineer in the State of
the estimates of series resistance. Arizona.

Kris Koellner is an Engineering Supervisor with Salt River Project (SRP) in


IX. ACKNOWLEDGEMENTS Phoenix, AZ. Kris has worked at SRP since 1994 in the areas of Distribution
Planning, Power Quality, and most recently, System Protection. Kris
graduated with a B.S.E. degree in Electrical Engineering from Arizona State
The work reported in this paper has been supported by Salt University and is registered as a Professional Engineer (PE) in the state of
Arizona. He currently serves as a co-chair of the North American
River Project. This support is gratefully acknowledged.
Synchrophasor Initiative (NASPI) Data & Network Management Task Team.

X. REFERENCES

[1] Robert E. Wilson, Gary A. Zevenbergen, and Daniel L. Mah,


“Calculation of transmission line parameters from synchronized
measurements,” Electric Machines and Power Systems, Vol. 27, No. 12,
December 1999, pp. 1269-1278
[2] Yuan Liao and Mladen Kezunovic, “Optimal Estimate of Transmission
Line Fault Location Considering Measurement Errors,” IEEE Trans.
Power Del., Vol. 22, No. 3, July 2007, pp. 1335-1341.
[3] I1-Dong Kim, Raj K. Aggarwal, “A Study on the On-line Measurement
of Transmission Line Impedances for Improved Relaying Protection,”
Electrical Power and Energy Systems, Vol. 28, No. 6, July 2006,
pp.359-366
[4] North American SynchroPhasor Initiative Performance & Standards
Task Team, “Synchrophasor Measurement Accuracy Characterization,”
November 2007.
[5] A. Abur and A. G. Exposito, Power System State Estimation—Theory
and Implementation. New York: Marcel Dekker, 2004.
[6] A. P. S. Meliopoulos, G. J. Cokkinides, F. Galvan, and B. Fardanesh,
“GPS-Synchronized Data Acquisition: Technology Assessment and
Research Issues,” Proceedings of the 39th Annual Hawaii International
Conference on System Science, Hawaii, January 4-7, 2006.
[7] E. Demeter, S. O. Faried and T. S. Sidhu, “Signal Phase Shifting during
Synchrophasor Measurements,” 2005 Canadian Conference on Electrical
and Computer Engineering, 2006, pp. 557-560.

Di Shi was born in Jiaozuo, China, on March 29, 1985. He received the B.S.
degree from Xi’an Jiaotong University, Xi’an, China, in 2007 and is currently
pursuing the M.S. degree in electrical power engineering at Arizona State
University, Tempe.

Вам также может понравиться