Professional Documents
Culture Documents
a r t i c l e i n f o a b s t r a c t
Article history: This paper utilizes data measured by phasor measurement units (PMUs) to extract a low-order dynamic
Received 16 May 2014 equivalent model for power system stability studies. The estimated model is a 2-order model for syn-
Received in revised form 27 May 2015 chronous machines. This model has the advantage of simplicity of classical model and considerably
Accepted 9 July 2015
reduces the oversimplifying error of classical model. This method offers an alternative approach to ana-
lytical model reduction techniques based on the detailed system models. The proposed method uses the
synchronized bus voltage and current phasors measured by PMUs. Using post disturbance data, electrical
Keywords:
and mechanical parameters of the equivalent generator are estimated sequentially. Furthermore, a new
Phasor measurement unit
Model estimation
approach for estimation of two-machine and single machine infinite bus (SMIB) equivalent systems are
Power system model reduction presented for analysis of electromechanical oscillations. To evaluate the performance of the proposed
Power system oscillations approach, simulations are performed on a two area 13-bus test system and real measured PMU data.
Simulation results show that the estimated model can represent the dynamic behavior of the studied
system with good approximation.
Ó 2015 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.ijepes.2015.07.015
0142-0615/Ó 2015 Elsevier Ltd. All rights reserved.
M. Shiroei et al. / Electrical Power and Energy Systems 74 (2016) 134–141 135
coherent or non-coherent groups of generators without requiring a Nonlinear Least Square-1 (NLS1)
large data set. According to Fig. 1, relationship between the internal voltage
In this paper, a new measurement based method using synchro- and the terminal voltage can be stated as follows:
nized phasor measurements is presented. Dynamic equivalent of 0
E\d jX d I ¼ U ð1Þ
the external system viewed from the measured bus is identified
in the form of classical or second-order model of synchronous gen- where, U and I are the generator terminal voltage and current pha-
erator. Estimation of the model parameters is done in two simple sors. By separating real and imaginary parts, (1) is divided into two
steps. At the first step, electrical parameters of equivalent machine equations as presented in the following.
are estimated. Mechanical parameters are estimated in the next
step. Estimation of electrical parameters is carried out by two U j E sinðdÞ
Ir ¼ ð2Þ
methods. In one method, using non-linear least square, electrical X 0d
parameters of the machine and its internal voltage angle are esti-
mated together. In the other method, electrical parameters are U r E cosðdÞ
Ij ¼ ð3Þ
estimated without estimation of the internal angle, which is calcu- X 0d
lated after estimation of the transient reactance. The latter method
is faster, while the former yields more accurate results. Machine where, U r ; U j ; Ir and I j are real and imaginary parts of terminal
modelling is then extended to model the prevailing tie-line oscilla- voltage and current, respectively. Considering the last m samples,
tions using SMIB and two-area equivalent. the above equations can be rewritten as following.
The remainder of the paper is organized as follows. In
E sinðdi Þ þ X 0d Iri U ij ¼ 0 i ¼ 1; . . . ; m ð4Þ
Section ‘The equivalent machine parameter estimation’, estimation
of the classical model is formulated and presented with two
methods. Section ‘Illustration using the SMIB system’ and E cosðdi Þ þ X 0d Iij U ri ¼ 0 ð5Þ
Section ‘Estimation of external power system dynamics for a
With 2m equations in (4) and (5) only m þ 2 variables, i.e. E; X d and
two-area system’ provide simulation results of these methods on
di ; i ¼ 1; . . . ; m are unknown. If m 2 then the set of equations will
the SMIB and two-area systems. In Section ‘Estimation of external
be over determined and can be calculated using the following non-
power system dynamics for a two-area system’, it is also shown
linear least square optimization:
how to identify a two-machine or SMIB equivalent model to inves-
tigate tie-line inter-area oscillations. The application of estimation E sinðd Þ þ X 0 Ir U j
1 d 1 1
approach on the real data is presented in Section ‘Application on
E cosðd1 Þ þ X 0 I j U r
the real PMU data’. Conclusions are given in Section ‘Conclusion’. d 1 1
Min .. ð6Þ
0
E;X d ;di .
E sinðd Þ þ X 0 Ir U j
m d m m
The equivalent machine parameter estimation
E cosðdm Þ þ X 0 I j U r
d m m
The classical or the second-order model of synchronous In (6) a suitable choice for initial conditions can be
machine is the simplest model that can be used in electromechan- ~ In the above least square formulation,
E ¼ 1; X 0d ¼ 0:2 and di ¼ \U.
ical dynamics analysis. This model offers considerable computa-
with m sets of measurements, the Jacobian matrix has the size of
tional simplicity; it allows the transient electrical performance of
ðm þ 2Þ ðm þ 2Þ. Therefore, by increasing the number of mea-
a machine to be represented by a simple voltage source with fixed
surement sets, computational effort will increase progressively.
magnitude behind an effective transient reactance as shown in
Fig. 1 [16]. This model has good performance in determining the
Nonlinear Least Square-2 (NLS2)
first swing stability [17]. When the system is subjected to a distur-
In the least square formulation of (6) with m sets of measure-
bance, parameters of the classical model can be estimated using
ments, m unknowns are the generator internal angle (d) corre-
post disturbance data. The electrical parameters of the model to
sponding to the individual samples. The following formulation
be estimated are the generator internal voltage (E), transient reac-
removes these variables from the optimization problem. By assum-
tance (X 0d ) and the variable rotor phase angle (d). The mechanical
ing the terminal voltage phasor as the angle reference, (1) is
parameters to be estimated are damping coefficient K d and inertia rewritten as:
constant (H). As stated above, the generator internal voltage is
0
assumed to be constant but the rotor phase angle varies from E\d0 ¼ U þ jX d I ð7Þ
one sample to the other.
By decomposing the current term (I) into real and imaginary parts,
Estimation of the classical model can be divided into two steps:
Eq. (7) can be written as follows:
estimation of the electrical parameters and estimation of the
mechanical parameters. 0 j
E\d0 ¼ U þ jX d ðIr þ jI Þ ¼ ðU X 0d I j Þ þ jðX 0d Ir Þ ð8Þ
accurate estimation of parameters using m samples (m > 2), the practice, the steady state value of Pelec is not constant and changes
following least square error optimization should be solved: continuously. In this case the moving average value of
2 pre-disturbance P elec can be used for determination of Pmech .
U þ ðX 0 Ir Þ2 þ 2X 0 Q ðE1 Þ2
1 d 1 d 1 Another solution for this problem is to assume Pmech as an unknown
min .. ð10Þ variable and to modify the least square formulation of (14) to
E;X 0d .
include P mech as an unknown variable to be estimated.
U 2 þ ðX 0 Ir Þ2 þ 2X 0 Q ðE Þ2
m d m d m m
Parameters X 0d (pu) H Kd
€di ¼ 4 ðdði þ 2Þ 2dði þ 1Þ þ dðiÞÞ
T2 ð15Þ
Original Value 0.3 6.5 2
2 Classic Model
d_ i ¼ ðdði þ 2Þ dðiÞÞ Estimated value (NLS1) 0.3 6.4801 1.9905
T
Estimation error (NLS1) % 0 0.306 0.2502
It should be mentioned that in the classical representation of Estimated value (NLS2) 0.3 6.4922 2.002
synchronous machine, Pmech is assumed to be constant, i.e. the gov- Estimation error (NLS2) % 0 0.12 0.1
ernor control is not taken to account. The steady-state or Detailed Model
pre-disturbance value of P elec can be used for Pmech to satisfy the Estimated value (NLS1) 0.2604 7.3 14.152
Estimated value (NLS2) 0.1745 8.044 15.388
equality of P elec and Pmech in pre-disturbance time window. In
M. Shiroei et al. / Electrical Power and Energy Systems 74 (2016) 134–141 137
7.15 Table 3
Detailed Model
NLS1 Model Parameters of the estimated equivalent machine from Bus 3, Bus 13 and SMIB of
Machine Active Power (p.u)
NLS2 Model
1.0001 ‘The equivalent machine parameter estimation’ is extended to esti-
1 mate the external power system dynamics for a two-area system.
1 Fig. 4 shows the single-line diagram of the two-area system. This
0.9999
system consists of two identical areas connected through a rela-
tively weak tie [20]. Each area includes two generating units that
0.9999
are strongly connected together (G1, G2 and G3, G4). As a result,
0.9998
G1 and G2 form a coherent group and G3 and G4 form another
0.9998
coherent group. When exposed to an external disturbance, the gen-
0 1 2 3 4 5 6 7 8 9 erators in one coherent group will swing in unison with each other
Time (sec) and against the other coherent group [15]. This system is a good
example to demonstrate application of the proposed method. The
Fig. 3. Generator speed oscillations. system data are given in [15].
Suppose that we are interested in studying the dynamics of
Area-2, and thus represent G3 and G4 in detail. Therefore, Area-1
estimated and original model accounts the effect of higher order consisting of G1 and G2 can be viewed as an external system from
dynamics. Bus 3. The reduced system diagram is shown in Fig. 5.
In order to validate the estimation results, similar disturbance is At the first step, to estimate the external system dynamic
applied on the SMIB system with detailed model, with the original parameters (Geq parameters), post-disturbance data should be
second-order model, and with the second-order NLS1 and NLS2
acquired. A three-phase short circuit fault with duration of 0.05 s
estimated models. Active power and speed a oscillations of gener-
without any line switching is applied at Bus 101 to simulate a dis-
ator are shown in Figs. 2 and 3, respectively.
turbance on the two-area system. The post-disturbance voltage of
From Figs. 2 and 3 one can observe that NLS1 and NLS2 esti-
Bus 3 and current of the line 3–101 are used as the input data to
mated parameters are better approximations of the detailed gener-
NLS1. This means that at least one PMU must be installed at Bus
ator model with respect to the original classical model, and yield
3 or Bus 101. Table 3 shows the results of estimating the Geq
oscillations very close to the oscillations of detailed model. In other
parameters. The parameter estimation is performed at three levels
words, the estimated parameters are changed in such direction to
of loading.
compensate the difference between the second-order model and
Also, Table 4 compares the dominant oscillation mode (k) of the
the detailed model, and represent the effect of higher order
system reduced system at three levels of loading. From Table 3 it
dynamics on the oscillatory mode. Characteristics of the oscillatory
can be observed that, in the most of cases but not in all cases, when
modes with different models of the SMIB system are shown in
the loading level increases, X 0d decreases and H and K d increase.
Table 2. These modes are extracted by applying Prony method in
There is one case that the trend is not true. In this case, unlike to
the time response of 0 to 5 s on the post-disturbance oscillations
the other cases, that power is coming to the reference bus. In this
of active power [20]. As shown in the Table 3, damping ratios
situation, the estimation of X 0d tends to zero and be very small.
(DR) of NLS1 and NLS2 models are approximately equal, but the
oscillation damping frequency (DF) in the NLS1 model is more Since X 0d cannot be a negative parameter, it cannot change and
accurate than the other. decrease anymore. Therefore, X 0d will be estimated near zero. It
should be mentioned that in this case X 0d tends to be negative in
the optimization procedure, but we have limited it’s lower limit
Table 2 to zero. Therefore it cannot exceed its limit. This limitation changes
Oscillatory mode for different models of SMIB system.
the monotonic trend for estimation values of other parameters
Model k DR error (%) DF error (%) such as H and K d . In addition, as shown in Table 4, the final results
Detailed 0:435 5:784j – – show lower accuracy and performance compared with other cases.
Classic 0:086 5:916j 80.67 2.282 As can be observed from Table 4, even in the large error of the pro-
NLS1 0:497 5:779j 16.67 0.086 posed method in some cases, the provided results are more accept-
NLS2 0:493 5:828j 12.4 0.761
able with respect to INA technique. Consequently, in most cases,
138 M. Shiroei et al. / Electrical Power and Energy Systems 74 (2016) 134–141
3.25
Detailed Model
3.2 Estimated Reduced Model
Analytically Reduced Model
3.15
3.1
3.05
2.95
2.9
2.85
2.8
2.75
0 1 2 3 4 5 6 7 8 9
Fig. 6. Disturbance response of the detail model, analytically reduced model and
estimated reduced model.
Table 4
Disturbance response of the two models at three levels of loading to three-phase fault on Bus 101.
So far, to overcome some drawbacks of estimation of power sys- Therefore, by studying the effect of line loading, the effect of differ-
tem from one side, a new technique is presented. In the new ent load allocations are also investigated, automatically.
approach, we have reduced the external system, viewed from the
measurement bus, to an equivalent classic machine model. It is
Reduction of the system into SMIB system
also possible to reduce and replace both sides of the tie-line, and
convert it to a two-machine or a SMIB system. This process is
It is well known that a two-machine model can be transformed
explained in the following.
to an equivalent SMIB model with similar dynamic characteristic
[22]. Here, we are going to take a different approach, and identify
a SMIB equivalent directly. For this purpose, the electrical equiva-
Reduction of the system into two-machine system
lents of either sides of the tie-line are first estimated as for the
two-machine equivalent explained earlier. Next, one of the internal
For model reduction from both sides of a tie-line, the terminal
buses is selected as the phase reference and assumed as infinite
bus voltages and the tie-line current must be known. In the case
bus. Mechanical parameters of the equivalent machine, located
of two-area system, to reduce the system from Buses 3 and 13,
on the other bus, are then estimated. Fig. 8 shows the proposed
U 3 ; U 13 and I313 must be measured. If impedance of the line 3–
estimation process.
13 is known, then only one PMU is required at Bus 3 or 13.
In Fig. 8, Bus 2 is selected as the reference or infinite bus and the
Otherwise, at least two PMUs should be installed. The required
internal angle of Bus 1 is referred to that bus. Then, mechanical
steps for reducing the two-area system to an equivalent
parameters of equivalent machine are estimated only for machine
two-machine system are shown in Fig. 7. As it is clear from this fig-
1, i.e.:
ure, part of the system that is connected to each terminal of the
tie-line is modelled by an equivalent machine. d0 1;...;n ¼ d1;...;n d1;...;n ð16Þ
1 2
Part 2 of Table 3 shows the implementation result of the
approach on the two-area system. Similar to the previous sections, 2
2HSMIB d d0 K SMIB dd0
simulation is performed for three-phase fault on Bus 101 at three þ ¼ P mech Pelec ð17Þ
levels of line loading. w0 dt2 w0 dt
Extraction of the dominant mode of oscillations in response to Part 3 of Table 3 also shows the implementation results of SMIB
the three phase fault on Bus 101 is done to examine accuracy of model reduction for the two-area system.
the estimated two-machine model. The results are presented in Part 3 of Table 4 shows characteristics of the oscillatory mode in
part 2 of Table 4. It can be observed that in this case, error of the the equivalent SMIB model, and compares it with that of the orig-
estimated frequency of inter-area oscillations is not low. An alter- inal system. Comparison of part 2 and 3 of Table 4 reveals that the
native approach is proposed in the following, to improve this SMIB equivalent is more accurate than the estimated two-machine
drawback. model. Thus, it is more suitable, and yet simpler, for evaluating
The standard two-area network data are used to perform simu- tie-line transients.
lation for this study. There are two loads on buses 4 and 14. With
100 MVA as system base, the allocations of loads are presented in
Application on the real PMU data
Table 5. As shown in this table, in this paper the allocation of load
on buses 3 and 13 are changed to study the effect of line loading.
In this section PMU data measured in a western interconnection
power system is used for simulation purpose. Current and power of
Malin-Round Mountain transmission line and bus voltage are mea-
Table 5
sured with the rate of 20 sample/s during the serious breakup of
Load allocation among the buses.
August 10, 1996 [23]. Application of estimation approach on the
Tie line loading Load-4 Load-14 real PMU data are carried out to assess the effectiveness of pre-
1.5 9.611 15.249 sented approach on the practical system data. Since, only PMU data
3 8.186 16.674 from one bus is available, estimation of reduced SMIB or reduced
4.5 6.826 18.034
two machine model is not possible. In this case, only external
140 M. Shiroei et al. / Electrical Power and Energy Systems 74 (2016) 134–141
Table A.6
Parameters of detailed generator model of SMIB.
Conclusion
system dynamic estimation is possible. Fig. 9 shows the active
power oscillation of Malin-Round Mountain transmission line dur- In this paper, estimation of a second order or classical model of
ing the breakup. external power system, viewed from an arbitrary bus, using PMU
It should be noticed that extraneous data outliers and high data was presented. Two different formulations with nonlinear
order components are filtered out before implementing the pro- least squares optimization were proposed to calculate the equiva-
posed method. Fourth order Butterworth filter with cut off fre- lent model parameters. One of these algorithms yields more accu-
quency of 5 Hz is used to eliminate measurement noise and high rate results, whereas the other is computationally less demanding.
order outliers. After filtering, the data are fed into NLS1 method Implementation of these methods on a SMIB and a two-area sys-
to extract external system dynamic. tem showed their performance and effectiveness. The paper also
The output of estimation process are X 0d ¼ 0:5367; H ¼ 55:02 presented the procedure for estimation of a two-machine or a
and K d ¼ 46:13. Since the external system viewed from PMU is a SMIB equivalent model to study electromechanical transients on
large power system, obtaining high value for H and K d is justifiable. a tie-line.
In order to investigate the performance and accuracy of the pro- The proposed methods are applicable for online evaluation of
posed approach, the output of estimated system is compared with small signal stability, inter-area oscillations and transient stability
real system. Fig. 10 shows power oscillation of estimated and real margin of power system. However, while the basic approach is
straightforward and comprehensive, the parameter identification
algorithm may be improved to yield more accurate results.
Real System
Estimated Model
1.4
Appendix A. SMIB system parameters
1.39
Parameters of the SMIB system are presented in Table A.6.
1.38
Parameters of the generator are provided in PU in 900 MVA base.
1.37 The line reactance in 100 MVA base is X line ¼ 0:1 PU.
1.36
References
1.35
[4] Chow J, Galarza R, Accari P, Price W. Inertial and slow coherency aggregation [13] Ju P, Ni L, Wu F. Dynamic equivalents of power systems with online
algorithms for power system dynamic model reduction. IEEE Trans Power Syst measurements. Part 1: Theory. IEE Proc Gener Transm Distrib
1995;10:680–5. 2004;151:175–8.
[5] Rahim A, Al-Ramadhan A. Dynamic equivalent of external power system and [14] Yu Y nan, El-Sharkawi M, Wvong M. Estimation of unknown large power
its parameter estimation through artificial neural networks. Int J Elect Power system dynamics. IEEE Trans Power App Syst PAS-98 1979:279–89.
Energy Syst 2002;24:113–20. [15] Chow J, Chakrabortty A, Vanfretti L, Arcak M. Estimation of radial power
[6] Parniani M, Chow J, Vanfretti L, Bhargava B, Salazar A. Voltage stability analysis system transfer path dynamic parameters using synchronized phasor data.
of a multiple-infeed load center using phasor measurement data. In: Power IEEE Trans Power Syst 2008;23:564–71.
systems conference and exposition, PSCE ’06; 2006. p. 1299–1305. [16] Kundur P. Power system stability and control. McGraw-Hill; 1994.
[7] Ljung L. System identification: theory for user. Prentice Hall; 1987. [17] Sauer W, Pai M. Power system dynamics and stability. Prentice Hall; 1997.
[8] Dou C xia, Zhang X zhong, Guo S liang, Mao C-C. Delay-independent excitation [18] Franklin G, Powell J, Workman M. Digital control of dynamic
control for uncertain large power systems using wide-area measurement systems. Addison-Wesley; 1998.
signals. Int J Elect Power Energy Syst 2010;32:210–7. [19] Chow J., Rogers G. Power system toolbox for MATLAB, Cherry Tree Scientific
[9] Ju P, Qin C, Wu F, Xie H, Ning Y. Load modeling for wide area power system. Int Software; 2007.
J Elect Power Energy Syst 2011;33:909–17. [20] Smith J, Fatehi F, Woods C, Hauer J, Trudnowski D. Transfer function
[10] Nazari-Heris M, Mohammadi-Ivatloo B. Application of heuristic algorithms to identification in power system applications. IEEE Trans Power Syst
optimal PMU placement in electric power systems: an updated review. Renew 1993;8:1282–90.
Sustain Energy Rev 2015;50:214–28. [21] Chow J. Time-scale modeling of dynamic networks with applications to power
[11] Shakouri H, Radmanesh HR. Identification of a continuous time nonlinear state systems. Lect Notes Control Inf Sci, vol. 46. Germany: Springer; 1982.
space model for the external power system dynamic equivalent by neural [22] Kimbark E. Power system stability. Wiley IEEE Press; 1995.
networks. Int J Elect Power Energy Syst 2009;31:334–44. [23] Kosterev D, Taylor C, Mittelstadt W. Model validation for the August 10, 1996
[12] Shiri F, Mohammadi-ivatloo B. Identification of inter-area oscillations using WSCC system outage. IEEE Trans Power Syst 1999;14:967–79.
wavelet transform and phasor measurement unit data. Int Trans Elect Energy
Syst 2014. http://dx.doi.org/10.1002/etep.1994.