You are on page 1of 9

This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2017.2776741, IEEE
Transactions on Power Systems
1

An Optimal Thévenin Equivalent Estimation


Method and its Application to the Voltage Stability
Analysis of a Wind Hub
Stephen Burchett, Senior Member, Daniel Douglas, Student Member, Scott G. Ghiocel, Member,
Maximilian Liehr, Student Member, Joe H. Chow, Life Fellow, Dmitry Kosterev, Anthony Faris, Eric Heredia,
Gordon H. Matthews

Abstract—A simplified voltage stability analysis method is factor loads. The maximum power transfer is achieved at the
to use a static Thévenin equivalent to represent the electrical tip of the P V curve, also known as the voltage collapse point.
connection to a load center. Assuming fixed values of the Thévenin
voltage and reactance, a power-voltage (P V ) curve analysis can
be performed to find the voltage collapse point and stability
margin. This paper proposes a method to compute the static 1.2
Thévenin equivalent voltage and reactance of a power system
using measured data. The method is validated with simulated 1 data points

Normalized Voltage, pu
PMU data and with 24-hour SCADA data at a wind hub on a
medium voltage transmission system in the US western system. 0.8
For the wind hub, the Thévenin equivalent parameters are used to
compute the maximum power transfer capability of the wind hub 0.6
as constrained by voltage stability. Then the method is extended
to online application by using short data windows and forgetting 0.4
factors.
pf = -0.9
Index Terms—Voltage stability, stability margin, Thévenin 0.2 pf = 1
equivalent, AQ bus, PV curve pf = 0.9
0
0 0.2 0.4 0.6 0.8
I. I NTRODUCTION Normalized Power, pu

T HÉVENIN equivalent is a simple circuit analysis tool


to calculate the voltage and current supplying a load
or subsystem. It is often incorporated in the computation of
Fig. 2: P V curves for a stiff-source single-load system with
various constant power factor loads

voltage stability margins, as shown by the system in Figure 1.


The basic assumption in a P V -curve analysis is that the
Thévenin voltage source Eth and reactance Xth are constant
Eth  V 0o during the analysis process [1], [2], [3]. For real power
jX th systems this assumption would be true for some short period
of time, as long as the system operating condition does not
change drastically. This paper is interested in calculating such
I  P + jQ
Thévenin equivalents using measured data only, without the
Thevenin Thevenin
voltage reactance need for power system line and generator data.
This paper demonstrates the systematic computation of
Fig. 1: Thevenin equivalent model Thèvenin equivalent model using simulated PMU data and
measured SCADA data. The key contribution is a simple
In Figure 1, the load bus is connected to a fixed voltage method to reliably calculate the Thévenin voltage source
source Eth via a fixed reactance Xth . As the active and Eth and reactance Xth from measured power network data.
reactive power load (P, Q) increases, the voltage V at the Previous measurement-based Thévenin equivalent parameter
load bus decreases. This behavior is illustrated in Figure 2 computation methods (see for example, [4], [5], [6]) are mostly
with the well-known P V curves for several constant power based on fitting the measured data to a P V curve using meth-
ods such as least-squares approximation. The challenge with
S. Burchett, D. Douglas, M. Liehr, and J. Chow are with the Department of these approaches is that there may not be enough variations in
Electrical, Computer, and Systems Engineering, Rensselear Polytechnic Insti-
tute, Troy, NY 12180, USA, e-mail: (burchs4,dougld2,liehrm,chowj@rpi.edu) the (P, Q, V ) measurement set to cover a significant portion of
S, Ghiocel is with Mitsubishi Electric Power Products, Inc. (MEPPI), the P V curve to yield reliable results. For example, consider
Rexford, NY 12148, e-mail: (Scott.Ghiocel@meppi.com) the situation shown in Figure 2 in which the measured data is
D. Kosterev, A. Faris, R. Heredia, and G. Matthews are with
Bonneville Power Administration, Vancouver, WA 98666, USA, email: contained in a small circle on the P V curve. This approach
(dnkosterev,ajfaris,emheredia,ghmatthews@bpa.gov) may be acceptable using simulated data with no measurement

0885-8950 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2017.2776741, IEEE
Transactions on Power Systems
2

noise. However, with data measured from real power systems, The large generator in System 1 is connected to the bound-
during periods where the (P, Q, V ) values are relatively con- ary bus via a transmission line of reactance 0.2 pu. The
stant, the presence of measurement and quantization noise may small generator in System 2 is connected to the boundary
result in these methods yielding rather erratic results. Figure bus via a transformer with reactance 0.15 pu. The transient
3 shows the Thévenin equivalent reactance values estimated and subtransient reactances of the generator are Xd0 = 0.42
using 5 min (P, Q, V ) data windows, over a twenty-four pu and Xd00 = 0.31 pu, respectively, and the voltage regulator
period, for the wind hub system described in Section IV. For gain is KA = 141.
a practical power system, it would be unreasonable to expect
that the equivalent reactance would vary so substantially over V 0o
such short spans of time. P + jQ
System 1 System 2
I 
Boundary
bus
Fig. 4: A power system consisting of a small system (Sys-
tem 1) and a large system (System 2)

Disturbances (or control actions) are applied to both sys-


tems, but not simultaneously, to change the (P, Q) flow and
the voltage V . For a disturbance in System 2, one can compute
the Thévenin equivalent for System 1, and vice versa.
Suppose a disturbance is applied in System 2 such that the
values of (P, Q, V ) would vary. If System 1 is represented by
Fig. 3: Thevenin reactance calculation with each point on a Thévenin voltage Eth and a Thévenin reactance Xth , then
the horizontal axis representing 5 minutes the relationship
Ẽth = V + jXth (I 6 φ) = V + jXth (P + jQ)∗ /V (1)
The proposed optimal Thévenin equivalent parameter cal-
culation method is discussed in Section II. It is applied to holds for all (P, Q, V ) values, where Ẽth is a voltage phasor,
simulated PMU data from a test system in Section III. The with Eth as its amplitude. If the correct Xth is found, Eth is
method is then applied to day-long SCADA data at the wind a constant value.
hub on a median voltage transmission system in the US Thus a Thévenin equivalent computation method is to
western system in Section IV. In addition, Section IV also assume a range of reasonable values for Xth (as Xth is not
contains a voltage stability analysis using the AQ-bus method known ahead of time), and use them to calculate Eth . Then
[10]. Section V extends the algorithm proposed in Section II the value of a particular Xth yielding a mostly constant value
to an online method using short data windows augmented by for Eth can be selected as the Thévenin equivalent parameters
forgetting factors. (Eth ,Xth ).
To illustrate this concept, consider the simulation of the two-
system model in Figure 4 in which disturbances are applied
II. T H ÉVENIN E QUIVALENT C OMPUTATION FROM to System 2 to induce voltage steps in V . Figure 5 shows the
S IMULATED /M EASURED DATA simulated (P, Q, V ) responses.
For a Thévenin equivalent representation of a power system Using a range of values 0.05, 0.1, 0.15, 0.2, and 0.3 pu
in Figure 1, given the measured P, Q consumption of the load, for Xth , the resulting Eth responses are shown in Figure 6.
the voltage of the bus computed from the equivalent must Note that for smaller values of Xth , the calculated Eth follows
match closely the measured V for significant variations of closely the response of the measured voltage V . For the largest
P, Q. The load variations include all static load types. This value Xth = 0.3 pu, the calculated voltage diverges from V ,
property offers a simple and reliable approach to compute the that is, if V drops, Eth increases. However, if Xth = 0.2 pu
Thévenin equivalent voltage and reactance values. is selected, Eth is mostly constant, except at the instants of
To motivate the approach, consider a test system consisting the step changes in V .
of a large synchronous generator in System 1 and a small This process of selecting an optimal value of Xth such that
generator in System 2. The System 1 generator is modeled as Eth will be mostly constant is summarized in the proposed
a classical machine. The System 2 generator is modeled with algorithm:
subtransient dynamics and includes excitation and governor Optimal Thévenin equivalent estimation algorithm
models. The voltage V on the boundary bus and the (P, Q) 1) Obtain values of (P, Q, V ) by either simulation or
flow coming into the boundary bus are measured. Alterna- measurement
tively, the current measurement I 6 φ can be used instead of 2) Select a range of reactance values for Xth , as Xthi ,
(P, Q). i = 1, 2, ..., n

0885-8950 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2017.2776741, IEEE
Transactions on Power Systems
3

2 voltage Eth [8]. This approach has the extra complexity of


Power (pu) having to estimate the angle of the Thèvenin voltage.
1

0 III. T H ÉVENIN E QUIVALENT C OMPUTED USING


P S IMULATED DATA
Q
-1
The optimal estimation algorithm is applied to three simu-
1.1 lation cases, in which the (P, Q, V ) data on the boundary bus
Voltage (pu)

is sampled at 30 points per second. The first two cases involve


disturbances in System 1, to obtain the Thévenin equivalent
1.05
Vsim parameters for System 2. The third case involves a disturbance
Vcal in System 2, allowing the computation of the Thévenin equiv-
1 alent parameters for System 1. In all these cases, Gaussian
0 50 100 150 200
noise has been added to the simulated response.
Time (sec)
Fig. 5: Values of (P, Q, V ) for the voltage step disturbances
A. Voltage Step Disturbances
1.15 Applying the optimal estimation algorithm to the voltage
step disturbances data in Figure 5 yields X̂th = 0.2002
Voltage (pu)

1.1 pu and Êth = 1.0804 pu. The results are consistent with
System 1 data that X̂th is the sum of the transmission line
1.05
reactance (0.2 pu) and the machine transient reactance which
1
is small when scaled by the large machine base, and Êth
0 50 100 150 200 is essentially the large generator terminal bus voltage (1.08
Time (sec) pu). As a further verification, a sequence of 2-bus power flow
Vsim solutions is obtained with the large generator bus voltage set
Eth with X th =0.3 pu at Êth = 1.0804 pu, the boundary (load) bus power set at the
Eth with X th =0.2 pu simulated (P, Q) values, and X̂th = 0.2002 pu used as the
Eth with X th =0.15 pu reactance connecting the two buses. The calculated load bus
Eth with X th =0.1 pu voltage Vcal is shown with the simulated voltage Vsim in the
Eth with X th =0.05 pu lower plot of Figure 5. The root-mean-square (RMS) error in
the Thévenin voltage approximation is 0.000532 pu.
Fig. 6: Values of Eth for various values of Xth

B. Active-Power Step Disturbances


3) For each selected value of Xthi , calculate Ethi (k),
Next an active power step disturbance case is considered.
where k denotes the data point time index.
Figure 7 shows the simulated (P, Q, V ) data at the boundary
4) For each Ethi , calculate its average
! bus for a sequence of active-power step disturbances in System
XN 2, with System 1 at the same operating condition as the voltage
Ēthi = Ethi (k) /N (2) step disturbances.
k=1

where N is the number of sample points, and the root- 2


mean-square error as
Power (pu)

s
PN 1
2
k=1 (Ethi (k) − Ēthi )
εi = (3) 0
N P
Q
5) Select Ēthi with the lowest value of εi and its two adja- -1
cent neighboring points Ēth(i−1) and Ēth(i+1) . Apply
1.08
a 3-point quadratic fit algorithm in [7] to obtain the
Voltage (pu)

optimal value of X̂th .


6) Calculate Eth (k) using X̂th . 1.06
Vsim
7) Calculate Êth as the average of Eth (k)
Vcal
N
!
X 1.04
0 50 100 150 200
Êth = Eth (k) /N (4)
Time (sec)
k=1

It should be noted that a similar estimation idea is used to Fig. 7: Values of (P, Q, V ) for the active-power step dis-
estimate Xth by iteratively computing the Thèvenin equivalent turbances

0885-8950 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2017.2776741, IEEE
Transactions on Power Systems
4

Applying the optimal estimation algorithm results in X̂th = 2 type-3 wind turbines, with a combined maximum output of
0.2018 pu and Êth = 1.0807 pu, which for practical purposes, about 600 MW. The voltage at the wind farms is supported by
are identical to those obtained in the voltage step disturbances two 28.8 MVAR shunt capacitors in the substation and a wide
for estimating the Thévenin equivalent of System 1. To verify array of shunt capacitors and reactors, and two STATCOMs
the result, again a sequence of 2-bus power flow solutions is inside the wind farms. The WH bus is connected to the west
used to obtain the load bus voltage Vcal , which is shown with by a long 160-mile line (reactance is 0.0861 pu) which is
the simulated voltage Vsim in the lower plot of Figure 7. The eventually connected back to the 500 kV system. It is also
RMS error in the Thévenin voltage approximation is 0.000530. connected to the east via a 48-mile line (reactance is 0.02732
pu) to an area with several large generators and to the 500
C. Voltage Sag Disturbances kV system. Thus the west connection is regarded as weak,
whereas the east connection is regarded as strong.
For the third case, a voltage sag disturbance is applied in The objective is to assess the voltage stability margin to
System 1 to compute the Thévenin-equivalent for System 2. ensure safe transfer of the wind energy to the west and to the
Figure 8 shows the simulated (P, Q, V ) data at the boundary east. At first glance, a voltage stability analysis would require
bus for a sequence of voltage-sag disturbances in System 1, accounting for all the status of the reactive power supply
with System 2 also responding to this disturbance. at the WH Bus, which would be quite complicated as the
statuses of the shunt capacitors/reactors at the WH Substation
1 and the wind farms are not available from the SCADA data.
However, with valid Thévenin equivalent models, the WH Bus
Power (pu)

0 can simply be viewed as a P Q bus, with a negative load.


The proposed voltage stability analysis approach is to first
-1
P represent the east and west systems each with a Thévenin
Q equivalent, as shown in Figure 9, using the (P, Q, V ) SCADA
-2
data on the east and west connections. The second step is
1.1 to analyze the voltage stability margin of the 3-bus system
Voltage (pu)

consisting of the WH bus, the West Thévenin bus, and the


1.05 East Thévenin bus.
Vsim
Vcal
1 A. West Thévenin Equivalent
0 20 40 60 80 100 120
An advantage of estimating a Thévenin equivalent for a
Time (sec)
wind hub is that the power generation from the wind turbines is
Fig. 8: Values of (P, Q, V ) for the voltage sag disturbances not constant in a daily cycle. The variations in the (P, Q) flow
would provide more accurate Thévenin equivalent models.
Applying the optimal estimation algorithm results in X̂th = Figure 10 show a one-day data of the measured voltage at
0.1852 pu and Êth = 1.0768 pu. The results are consistent the WH Bus and the (P, Q) power flow to the west system.
with the parameters in System 2. The Thèvenin reactance Figure 11 shows the calculated voltage Vth at the west
X̂th is the sum of the transformer reactance and a fraction system using values of Xth from 0.05 to 0.3 pu. The calculated
of the transient reactance as impacted by the voltage regulator voltage stays mostly constant for a value of Xth = 0.2 pu.
gain, as the exciter is regulating the generator terminal bus. The optimal estimation algorithm yields X̂th = 0.1835 pu and
Also X̂th is higher than the terminal bus voltage (1.0 pu. Êth = 1.1181 pu. To verify the solution, the WH Bus voltage
This analysis shows that the fixed voltage point is inside the is computed using the Thévenin model and the (P, Q) flow
synchronous machine. To verify the result, a sequence of 2- values. Figure 12 shows that the resulting voltage matches the
bus power flow solutions is used to obtain the load bus voltage measured voltage at WH Bus, with a RMS error of 0.0025 pu.
Vcal , which is shown with the simulated voltage Vsim in the The average Thévenin equivalent parameters for 9 days are
lower plot of Figure 8. The RMS error in the Thévenin voltage shown in the top plot of Figure 13, indicating good consistency
approximation is 0.0023. Note that this RMS error is higher of results between Days 3 to 6. Note that there were not
than those obtained for System 1, because Vcal , computed from enough (P, Q, V ) variations in the Day 8 data to compute the
a static model, does not reproduce the oscillations occurring optimal parameters.
during the switching instants.
B. East Thévenin Equivalent
IV. VOLTAGE S TABILITY A NALYSIS OF A W IND H UB The optimal Thévenin equivalent estimation is repeated for
In this section, 2-sec SCADA data is used to derive two the east system. Figure 14 shows the measured voltage at the
Thévenin equivalents for a wind hub shown in Figure 9. Ten WH Bus, and the power flow to the east system.
days of 24-hour data were available for analysis, with each Figure 15 shows the calculated voltage Vth at the west sys-
data set consisting of 43,200 points. The wind hub (WH) bus tem using values of Xth from 0.03 to 0.11 pu. Here there is no
is a 230 kV bus serving as a wind hub/point of connection for fixed value of Xth that would result in an approximately fixed
six wind farms [9]. The wind farms consist of 4 type-2 and voltage Eth for the whole 24 hours. This reflects the situation

0885-8950 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2017.2776741, IEEE
Transactions on Power Systems
5

Fig. 9: Wind Hub Substation Layout

4 1.3
Voltage (pu)
Power (pu)

2 1.2

0 1.1
P
Q
-2 1
0 4 8 12 16 20 24
1.08 Time (hours)
Voltage (pu)

Vmeas
Eth with X th =0.3 pu
1.06
Eth with X th =0.2 pu
Eth with X th =0.15 pu
1.04 Eth with X th =0.1 pu
0 4 8 12 16 20 24
Eth with X th =0.05 pu
Time (hours)
Fig. 10: 24-hour SCADA voltage at WH Bus and the Fig. 11: Calculated voltage Vth at west system using vari-
(P, Q) flow to the west system ous values of Xth

that the terminal voltages of the generators near the east system in the lower plot of Figure 13. Note that there were not enough
were adjusted during the day to accommodate load changes. (P, Q, V ) variations in the Day 8 data to compute the optimal
In particular, there is a noticeable voltage rise between Hours parameters.
7 to 8 (denoted by A in Figure 15) and a voltage drop between
Hours 21 to 22 (denoted by B). As a result, the optimal
Thévenin equivalent estimation is performed only for Hours 9 C. Voltage Stability Analysis
to 20, yielding the optimal parameters X̂th = 0.0498 pu and The AQ-bus method in [10] is used for developing the
Êth = 1.0704 pu. To verify the solution, the WH Bus voltage P V curve and computing the voltage stability margin, using
is computed using the Thévenin model and the (P, Q) flow the east and west system Thévenin model parameters. In
values. Figure 16 shows that the calculated WH Bus voltage this method, the bus that is vulnerable to voltage collapse
matches well the measured values from Hours 9 to 20. The is modeled as an AQ bus, where the bus voltage angle
RMS error of the Thévenin voltage during this Hours 9 to 20 and reactive power are specified. In the wind hub system,
period is 0.0016 pu. There is also a divergence between Hours the wind hub is selected as the AQ bus, even though it
0 to 8 and 22 to 24. is supplying power. However, with limited reactive power
The optimal parameters obtained for nine days of SCADA support, it encounters similar voltage collapse problems like a
data for the east system using Hours 9 to 20 measured data load bus. The power flow computation in the AQ-bus method
also show good consistency between Day 2 to Day 6, as shown uses a Jacobian matrix with one fewer dimension than a

0885-8950 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2017.2776741, IEEE
Transactions on Power Systems
6

in Figure 18. Note that the P V curve does not reach zero
1.075
voltage when the wind hub power is zero because of power
V
1.07 V
meas
transfers from east to west through the wind hub bus. Table
cal
I shows the maximum power transfer levels for the various
1.065 shunt compensation.
Voltage (pu)

1.06
A voltage stability analysis for the wind hub system with
the line to the East System disconnected, resulting in wind
1.055 curtailment, can be found in [11].
1.05
4
1.045

Power (pu)
2
1.04
0 4 8 12 16 20 24 0
Time (hours) P
Q
Fig. 12: Comparison of measured and west system -2
Thévenin equivalent calculated WH Bus voltage 1.08

Voltage (pu)
West 1.06
0.23 1.16
X TH (pu)

E TH (pu)

0.18 1.12
1.04
0 4 8 12 16 20 24
0.13 1.08
Time (hours)
0.08 1.04 Fig. 14: 24-hour SCADA voltage at WH Bus and the
East
0.09 1.1 (P, Q) flow to the east system
X TH (pu)

E TH (pu)

0.06 1.08

0.03 1.06 1.15


Voltage (pu)

A
0 1.04 1.1
1

9
10
ay

ay

ay

ay

ay

ay

ay

ay

ay
ay

1.05
D

D
D

B
Fig. 13: Optimal Thévenin equivalent parameters ob- 1
tained from the SCADA data (∇ denotes a voltage 0 4 8 12 16 20 24
data point, and × denotes a reactance value) Time (hours)
Vmeas
Eth with X th =0.11 pu
normal Jacobian matrix, because there are two buses with E with X
th th
=0.09 pu
fixed angles, allowing it to avoid the Jacobian matrix from E with X
th th
=0.07 pu

being singular at the collapse point. E with X


th th
=0.05 pu

To conduct the voltage stability analysis, the angle separa- E with X =0.03 pu
th th

tion between the wind hub (AQ bus) and the swing bus is
Fig. 15: Calculated voltage Vth at east system using various
increased to allow for the power exported by the wind hub to
values of Xth
the east and west systems. To solve for the power flow, the
split of the wind power exporting to the west and east systems
needs to be known. Here the incremental power transfer is Shunt compensation Pmax Vcrit
approximated by a linear relationship, shown in Figure 17. none 805 MW 0.879 p.u.
The east system is chosen as the swing bus because the S1 813 MW 0.890 p.u.
S1 and S2 822 MW 0.902 p.u.
power absorbed varies more than the west system. The west
system is modeled as a P V bus with its active power adjusted TABLE I: Maximum wind hub power transfer
in tandem with the east system based on the approximated
split. The voltages at these two buses are set to their Thévenin
equivalent values.
The P V -curve analysis was performed considering the two
shunt capacitors at the wind hub, which results in P V curves
for three levels of fixed shunt reactive power compensation:
no compensation, S1 only, and both S1 and S2 as shown

0885-8950 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2017.2776741, IEEE
Transactions on Power Systems
7

1.075 1.1

1.07 1

Voltage magnitude (pu)


1.065
0.9 Measured data
Voltage (pu)

1.06 No shunt comp. (0 Mvar)


One shunt (S1)
0.8
1.055
Two shunts (S1,S2)

0.7
1.05
Vmeas
1.045 Vcal 0.6

1.04 0.5
0 4 8 12 16 20 24 0 2 4 6 8
Time (hours) Wind plant active power (pu)

Fig. 16: Comparison of measured and east system Fig. 18: Wind hub P V curves
Thévenin equivalent calculated WH Bus voltage

windows in which the new data is used with an equal or longer


3 duration of immediately past data. The method proposed here
will make use of forgetting factors [12].
2.5 The measured data is divided into periods of length T . Let
k
Xth be the estimated Thèvenin reactance at period k using the
West Power (p.u.)

k
2 optimal Thèvenin parameter method and X̂th be the estimated
value using the forgetting factor method. Then the proposed
1.5 estimation scheme is
−1
X 0
X
1 k k k+j
X̂th = α0 Xth + αk X̂th , αk = 1, (5)
j=−n k=−n
0.5 Measured data
k k
Power flow model where X̂th is a weighted sum of the current estimate Xth
0 and n immediately past optimal Thévenin reactance values. In
0 1 2 3 4 5
East Power (p.u.) addition, the forgetting factors αk satisfy
Fig. 17: Wind hub power export to east and west systems α−n < α−(n−1) < . . . < α0 . (6)
Once X̂th has been determined, V̂th is computed using steps
V. A N O NLINE E STIMATION A LGORITHM (6)-(7) of the algorithm discussed in Section II.
In the second strategy, there are some instances that the
As noted earlier, for significant changes in power system variation in the measured data in the new data window is so
operating conditions, including line and shunt switchings, gen- small that it would be more meaningful to keep the Thévenin
erator reference voltage adjustments, and control equipment parameters from the previous data window. From Steps (4)
reaching reactive power output limits, it would be necessary and (5) of the optimal estimation algorithm, define
to re-estimate the Thèvenin equivalent parameters. In this
section, the development of online algorithms is considered. εmax = max{εi }, εmin = min{εi } (7)
i i
In contrast to the use of day-long data records to estimate
and
the average Thèvenin equivalent parameters, the objective of
δ = εmax − εmin (8)
an online algorithm is to estimate the time varying Thèvenin
parameters by means of much shorter periods of measured If δ is less than a certain tolerance, new values of Thévenin
data. Such an algorithm will repeatedly compute the Thèvenin parameters will not be calculated and the previous Thévenin
parameters at regular intervals or on demand when new parameter values will be kept.
measured data is available. The main difficulty is that the The online algorithm is demonstrated using the 2-second
measured (P, Q, V ) data may not have adequate variations to SCADA data of east WH system for the same 24-hour mea-
provide reliable results. Thus two strategies are proposed to surement period in Figure 16. The length of the data window T
improve the reliability of the results. The core of the method and the choice of n have to be chosen carefully, often based on
will incorporate the optimal Thèvenin equivalent parameter the measured data. Ideally a short data window should be used,
estimation method developed earlier. but the resulting estimation can be noisy. Also, in order to get
The first strategy is to take into account not just the new fast updates, n should not be too large. For this illustration,
data, but also incorporate the most recent data in the estimation the following parameters were used in the Thévenin equivalent
process. There are many such methods such as overlapping estimation algorithm:

0885-8950 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2017.2776741, IEEE
Transactions on Power Systems
8

1) The duration of each period is T = 5 mins of each measurement window. Figure 21 shows the improved
2) The number of previous periods used is n = 4 match with the measured data of the bus voltage throughout
3) Linear varying forgetting factors are applied with the 24-hour period.
αk = {0.06̄, 0.13̄, 0.20, 0.26̄, 0.33̄}
4) The threshold δ = 0.003 pu 1.075

The results of the time-varying optimal Thévenin parameters 1.07


are shown in Figures 19 and 20. The average values of the
Thévenin parameters are also shown in these figures. The 1.065
online estimated Xth without forgetting factors is shown in
Figure 19. Note that without the forgetting factors, Xth varies 1.06
between 0.001 to 0.13 pu. Using the forgetting factors, the
variation of X̂th is reduced to 0.03 to 0.09 pu. Note also 1.055
that for several periods, there were not enough variation in
1.05
5 minutes for the algorithm to compute a new reactance. As
mentioned earlier, the voltage setpoints of the generators in
1.045
the East System are raised during peak load hours and lowered
during the off-peak hours. The estimated Vth shown in Figure 1.04
20 captures this operating characteristic. 0 2 4 6 8 10 12 14 16 18 20 22 24

Fig. 21: Comparison of measured and east system


Thévenin equivalent calculated WH Bus voltage
using online estimation algorithm

The merits of the online method, as compared to analysis


based on detailed power system models, include (1) requiring
only measured data of (P, Q, V ) for a load center without
needing system data and state estimator solutions, (2) ac-
counting for the impact of changes in the voltage setpoints
of generators and the operation of the excitation systems,
instead of assuming the generator terminal bus voltages to
be constant, and (3) providing voltage sensitivity analysis of
load buses due to shunt reactor/capacitor bank switching. On
the other hand, the Thévenin model provides only a local view
of voltage stability, not an overall system view, and would be
Fig. 19: Optimal Thévenin reactance values obtained using only applicable to buses exhibiting large (P, Q, V ) variations.
a 5-minute window and n = 4
VI. C ONCLUSIONS
An optimal Thévenin model parameter estimation method
has been proposed. The method is based on the assumption
that the Thévenin voltage and reactance are constant during
some time periods. It is a reliable method as it does not solve
a set of nonlinear equations involving data with measurement
and quantization noise. In addition to calculating voltage
stability margins, the Thévenin parameters can also be used for
analysis of voltage related operational issues such as substation
voltage sensitivities with respect to shunt reactor/capacitor
bank switchings.
The methods proposed in this paper can be applied to
both off-line and online application. The techniques have been
successfully applied to simulated data as well as measured data
at a wind hub.
It should also be emphasized that the proposed
Fig. 20: Optimal Thévenin voltage values obtained using measurement-based methods are applicable to load buses that
a 5-minute window and n = 4 experience sufficient variations in active and reactive power
consumption and voltage amplitude. Load buses with voltage
To verify the online algorithm’s performance, the voltage at and power that remain mostly constant for long periods of
the boundary bus is computed using the optimal Thévenin pair time are not good candidates for the proposed methods.

0885-8950 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2017.2776741, IEEE
Transactions on Power Systems
9

ACKNOWLEDGMENT Scott G. Ghiocel (M’13) is a principal engineer at Mitsubishi Electric Power


Products, Inc. (MEPPI). He received his MS and PhD degrees from Rensselaer
The work described in this paper was supported in part by Polytechnic Institute in 2010 and 2013, respectively. His research interests
Bonneville Power Administration TIP 348, in part by the DOE include power system dynamics, voltage stability, reactive power control, and
CERTS program from Lawrence Berkeley National Labora- applications of synchronized phasor measurements.
tory, and in part by the Engineering Research Center Program
of the NSF and the DOE under NSF Award Number EEC-
1041877 and the CURENT Industry Partnership Program, and
in part by NSF Award Number EEC-1550029. Maximilian W. A. Liehr (S’14) received his BS and MENG in Electrical
Engineering from Rensselaer Polytechnic Institute. He is currently a Ph.D.
student at Rensselaer Polytechnic Institute. His research interests include
R EFERENCES power system dynamics, voltage stability, and renewable energy.

[1] C. W. Taylor, Power System Voltage Stability, McGraw Hill, 1994.


[2] T. Van Cutsem and C. Vournas, Voltage Stability of Electric Power
Systems, Springer, 1998.
[3] V. Ajjarapu, Computational Techniques for Voltage Stability Assessment
and Control, Springer, 2006. Joe H. Chow (LF16) received his BS degrees in Electrical Engineering and
[4] M. Parniani, J. H. Chow, L. Vanfretti, B. Bhargava, and A. Salazar, “Volt- Mathematics from the University of Minnesota, Minneapolis, and MS and
age Stability Analysis of a Multiple-Infeed Load Center Using Phasor PhD degrees, both in Electrical Engineering, from the University of Illinois,
Measurement Data,” Proceedings of IEEE Power System Conference and Urbana-Champaign. After working in the General Electric power system
Exposition, October 2006. business in Schenectady, New York, he joined Rensselaer Polytechnic Institute
[5] S. M. Abdelkader and D. J. Morrow, “Online Tracking of Thv́enin in 1987, and is a professor of Electrical, Computer, and Systems Engineering.
Equivalent Parameters suing PMU Measurements,” IEEE Transactions His research interests include power system dynamics and control, voltage-
on Power Systems, vol. 27, no. 2, pp. 975-983, 2012. source converter-based FACTS controllers, voltage stability analysis, and
[6] S. G. Ghiocel, J. H. Chow, R. Quint, D. Kosterev, and D. J. Sobajic, synchronized phasor data. He is a member of the National Academy of
“Computing Measurement-Based Voltage Stability Margins for a Wind Engineering.
Power Hub using the AQ-Bus Method,” Proceedings of Power and
Energy Conference at Illinois (PECI), 2014.
[7] D. G. Luenberger, Introduction of Linear and Nonlinear Programming,
Addison Wesley, 1973.
[8] S. Corsi and G. N. Taranto, “A Real-Time Voltage Instability Identifi- Dmitry Kosterev received his B.S. and Ph.D. degrees in Electrical Engi-
cation Algorithm based on Local Phasor Measurements,” IEEE Trans- neering from Oregon State University in 1992 and 1996, respectively. Dr.
actions on Power Systems, vol. 23, no. 3, pp. 1271-1279, 2008. Kosterev is with the Bonneville Power Administration, where his responsibil-
[9] E. Heredia, D. Kosterev, and M. Donnelly, “Wind hub reactive resource ities include network planning, power system modeling, and power system
coordination and voltage control study by sequence power flow,” Power performance monitoring. He chaired the WECC Modeling and Validation
and Energy Society General Meeting (PES), 2013. Work Group from 2007 to 2011, and the WECC Load Modeling Task
[10] S. Ghiocel and J. H. Chow, “A Power Flow Method Using a New Force since 2002. Dr. Kosterev is actively involved with synchro-phasor
Bus Type for Computing Steady-State Voltage Stability Margins,” IEEE technologies, and has served as a co-chair of Planning Implementation Task
Transactions on Power Systems, vol. 29, no. 2, pp. 958-965, 2014. Team at North-American Synchro-Phasor Initiative.
[11] S. M. Burchett, M. Liehr, S. G. Ghiocel, J. H. Chow, A. Faris, and D.
Kosterev, “Voltage Stability Analysis of a Wind Hub using Thèvenin
Equivalent Models Estimated from Measured Data,” Proc. Power Tech
Conference, pp. 1-6, Manchester, 2017.
[12] K. J. Astrom and B. Wittenmark, Adaptive Control, Addison Wesley, Tony Faris is an engineer in the Measurements Systems group at the Bon-
1989. neville Power Administration. He began working with phasor measurement
systems in 2004 as a student at BPA. He holds a BSEE from the University
of Portland and an MSEE from the University of Washington.

Stephen M. Burchett (SM16) received the B.S. degree in mechanical


engineering from Union College, Schenectady, NY in 2005 and the M.S.
degree in mechanical engineering from Union Graduate College, Schenectady, Eric Heredia received his BS in Electrical Engineering from the University
NY in 2006. He is currently pursuing a Ph.D. degree in Electrical Engineering of Portland in 2002 and joined BPA in August 2002. He worked in BPAs
from Rensselaer Polytechnic Institute. From 2006 to 2010, he was with PSEG Transmission Planning group for 13 years, then moved to BPAs Operations
Power, Newark, NJ. From 2010 to 2013, he was with Green Charge Networks, Planning group in 2015. His work at BPA focuses on transmission system
Brooklyn, NY. From 2013 to 2016, he was with the New York Independent studies determining network capabilities and limitations.
System Operator (NYISO), Albany, NY. His research interests include renew-
able generation, energy storage, electricity markets, optimization, and voltage
stability. Steve is a Registered Professional Engineer (Electric Power) in New
York, Massachusetts, and California.
Gordon Matthews is a General Engineer at Bonneville Power Administration.
He previously managed the Mechanical and Civil Sections of the BPA
Laboratories, including outreach to the BPA Energy Efficiency Seattle Office.
Daniel Douglas (SM’13) received his BS degree in Electrical Engineering His other assignments included the Power Services Western Area Marketing
from Temple University, Philadelphia. He is currently pursuing a PhD degree Hub, Corporate Services, and Transmission Services to assist Scheduling
in Electrical Engineering at Rensselaer Polytechnic Institute. His research Automation, defining and refining Transmission Policy and business practices,
interests include power generation control, power system control, power NERC Reliability, and A123 Compliance. His current assignment is in the
engineering computing, ferroresonance, and inductive power transmission. Technology Innovation work group in Corporate Strategy.

0885-8950 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

You might also like