1680

IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 22, NO. 3, JULY 2007

An Approximate Method for

Transient State Estimation
Kent K. C. Yu and Neville R. Watson, Senior Member, IEEE

AbstractVoltage dip is a frequently occurring power-quality

problem that electricity consumers face. Knowledge of the source
and magnitude of the voltage dip is desirable; however, complete
monitoring of the power system is prohibitive. State estimation for
both fundamental and harmonic frequencies was presented previously. Its ability to utilize limited measurements for system-wide
estimation has proven to be very effective. In this paper, the
technique is extended further to estimate power system transients.
Transient state estimation (TSE) provides an effective and systematic approach in locating and identifying remote fault sources,
thus allowing the problem to be addressed more efficiently. The
proposed approach formulates the system using state-space theory
and results in a system of first-order differential equations which
are used to form the measurement system. The estimation is the
simultaneous solution of the measurement system equations based
on a time sequence of snapshots of the system measurements.
Additional measurement information utilizes the voltage and
current history measurements where they are approximated using
backward Euler integration formula, are also supplied to the TSE
algorithm. To evaluate the performance of the proposed TSE,
transients caused by a sudden lost of a system load have been simulated on the test system. This paper provides a new methodology
to estimate a complete transient profile of a power system.
Index TermsPower-quality assessment, state estimation, transient.

I. INTRODUCTION

LECTROMAGNETIC transients in power systems, such

as switching surges, overvoltages, or overcurrents, are
caused by disturbances (e.g., system faults, lightning strikes,
or switching events). The primary concern in transient assessment is not only to identify the cause of the transient, but
also to assess its effect on the surrounding system. Often a
transient event is recorded and the problem is to determine the
source and location of the disturbance. With current computer
technology and reliable mathematical models and numerical
techniques, the simulation of power system transients in the
time domain using the Electromagnetic Transient Program
(EMTP), Electromagnetic Transient Program for dc (EMTDC),
or alternative transient program (ATP) has become a standard
in non-real time simulation [1][5]. With simulation, a system
can be checked thoroughly to ensure satisfactory operation and
optimization of the component parameters and controller settings. The system response to a variety of transient phenomena
Manuscript received February 25, 2005; revised October 18, 2005. This work
was supported by the Foundation for Research Science and Technology (FRST)
Bright Future Scholarship. Paper no. TPWRD-00108-2005.
The authors are with the Department of Electrical and Computer Engineering,
University of Canterbury, Christchurch PB 4800, New Zealand.
Digital Object Identifier 10.1109/TPWRD.2007.901147

can also be checked. Although EMTP, EMTDC, and ATP are

effective tools in inspecting the system transient performance
in an unexpected fault event where the fault model is unknown,
the use of computer simulation to assess the problem becomes
inadequate.
Exhaustive search methods are commonly used to determine
the disturbance event. They require the possible causes of the
disturbance simulated in EMTP/EMTDC/ATP and by matching
the simulations with the actual system transient response, a good
indication of the cause may be achieved. These methods, however, are generally time consuming.
A better solution would be to combine partial measurements
at optimal locations with state estimation technique to derive
complete knowledge of the system states. The state estimation
technique has been applied in many power system applications,
such as fault detection, load-flow analysis, and harmonic assessments [6][13]. Its ability in estimating full system states with
limited measurements and minimizing errors and inconsistencies in measurement data would be ideal for transient state estimation (TSE).
In this paper, a TSE methodology is introduced. The proposed method models the power system using state-space theory
in which it allows for an unified treatment of both linear and
nonlinear circuit elements. The individual equivalent circuits
of plant components maintain their physical form (e.g., inductance, capacitance, etc., the value of which remains constant
within the integration step), but can be altered from one step to
the next to represent time-dependent parameters. The behavior
of the network (i.e., the current in the different branches and the
voltages at different nodes), is determined by the following constraints:
1) topological constraints: the way in which the branches are
interconnected;
2) algebraic constraints: basic electric network laws such as
Ohms law and Kirchhoffs law.
The power system can be described as a set of first-order differential equations by taking the busbar voltages and inductor
and
currents as the state variables. Past measurements
present measurements
are used to form the measurement
system. The proposed method is tested on the Lower South Island (LSI) of the New Zealand system. A simple case of overvoltages caused by a sudden lost of a load is simulated and the
results are compared with those obtained in PSCAD/EMTDC
[14].
II. STATE-SPACE FORMULATION
In the development of the state-space model for a general
power system, the relevant system components are modeled in

0885-8977/$25.00 2007 IEEE

YU AND WATSON: AN APPROXIMATE METHOD FOR TRANSIENT STATE ESTIMATION

1681

Fig. 2. Equivalent coupled PI model.

for star-g/star-g connection:

Fig. 1. Three-phase star-g/star-g-connected transformer.

their equivalents.
1) Generators are modeled by a voltage source with an equivalent R-L impedance.
2) Transformers are represented using a coupled coil model
depending on the type of magnetic circuit and on the connections of the terminals and the neutrals (i.e., delta or
wye). Core losses are represented internally with an equivalent shunt resistance across each winding in the transformer.
3) Transmission lines are modelled by the three-phase PI
nominal models and, hence, are capable of incorporating
a nonsymmetric condition.
4) The real and reactive power components of the loads are
represented by their equivalent resistance and inductance.
A. Transformer State-Space Model
The transformer is represented by using a coupled coil model
in terms of self inductance
; mutual inductance
,
; and
as in (1)
core losses

(2)

where
and

branch-node incidence (connection)

matrices for voltages and currents,
respectively;
,

primary currents and voltages;

secondary currents and voltages.

B. Transmission-Line State-Space Model

The transmission-line nominal coupled PI model is used for
short to medium length transmission lines. The capacitance and
inductance coupling between the phases are considered. The
lumped three-phase representation of a transmission line is illustrated in Fig. 2. By summing up the nodal currents and loop
voltages, the following set of differential equations can be derived for the transmission-line state-space model:

(1)
(3)

and
are the voltages across the first and second
where
winding, respectively. and are taken as positive when the
current flows through to their respective windings.
Similarly, for a three phase star-g/star-g connected transformer as shown in Fig. 1, neglecting core losses, the
single-phase state model can be expanded as

(4)
(5)
(6)
shown in (7), at the bottom of the next page, where , , ,
, , and
are the sending end and receiving end current,
inductor current, shunt current, and sending end and receiving
,
,
, and
end busbar voltages, respectively.
are resistance, inductance, shunt susceptance, and conductance
matrices, respectively.
C. Load State-Space Model
The real and reactive power components of the system load
are represented by the equivalent resistance and inductance.

1682

IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 22, NO. 3, JULY 2007

The system state equations expressed in (2), (7), and (8) can then
as
be rewritten in terms of the previous system state

(12)
where

Fig. 3. Equivalent load model.

Fig. 3 depicts the equivalent load model and its state equations
are shown
(13)

(8)
III. TRANSIENT STATE ESTIMATION
Consider the general mathematical model which is related to
the measurement vector to the state variable , expressed as
(9)
is the measurement matrix.
where is the error vector and
The error vector represents the difference between the measurements from their true values due to the presence of measurement noise or bad data. Measurement noise emanated from
transducers is usually difficult to filter and unlike gross error,
it degrades the quality of TSE and its effect needs to be considered. However, as an introductory paper on TSE, which focuses on the verification of the proposed concept and methodology, measurement noise is briefly mentioned in the scope of
this study.
Consider that the operator d/dt can be approximated as
(10)
(11)

as shown in (14) at the bottom of the next page.

When the discretized state equations are formed in terms of
state variables at the previous time step, it allows the history
to be used as virtual measurements (i.e., meavalues
surements do not need metering) as well as real-time measurements.
By inspecting (13), it is shown that the transmission-line inductor current can be used as a measurement to form the TSE
measurement system. However, given that they are usually not
well-defined variables (i.e., difficult to measure) in a practical
system, the sending end and receiving end branch currents are
used instead. Therefore, the state equations become the second
equation shown at the bottom of the page, where (15) is, also
shown at the bottom of the next page, where

(16)
The proposed state estimation algorithm takes the busbar
voltages and inductor currents as the state variables (i.e.,
transformer winding currents, transmission-line inductor currents, and load inductor currents). The measurement system
is linear when the corresponding measurement equations described in (12)(16) are used. Each measurement will add a

(7)

YU AND WATSON: AN APPROXIMATE METHOD FOR TRANSIENT STATE ESTIMATION

row to the measurement matrix [H]. Both previous time-step

and present time-step
measurements are used
to form the measurement system. This provides additional
measurement information that can significantly reduce the
number of measurement points required to solve the measurement system. At each time step, measurements are updated
and a new set of measurement equations is formed. The state
variables can then be solved numerically in the LS sense using
(17). Once the state variables are known, dependent variables,
such as branch currents, are calculated accordingly
(17)
Matrix R is diagonal and contains the covariance of the
measurements.
The system is said to be observable if the set of available
measurements is sufficient to calculate all of the state variables
uniquely. In the case where the system is partially observable,
only the state variables corresponding to the observable islands
are uniquely defined. Observability analysis (OA) can be performed prior to the estimation to ensure that the selected measurement placements are adequate.

1683

Fig. 4 summarizes
implementation.

the

proposed

state

estimation

IV. TEST SYSTEM AND ESTIMATION RESULTS

The test system used is the reduced three-phase 220-kV
network below Roxburgh in the South Island of New Zealand.
The system is shown in Fig. 5. The system consists of 27 nodes
and 87 branches. The system contains five star-delta-connected
two winding transformers; two single transmission lines between
Roxburgh and Invercargill; three double transmission lines
between Manapouri, Invercargill, and Tiwai; and three linear
loads at Tiwai 220 kV (480 MW, 130 MVAR), Invercargill 33
kV (135 MW, 36 MVAR), and Roxburgh 11 kV (90 MW, 54
MVAR). An arbitrary measurement set is chosen. The voltage
and current measurement placements are shown by dots and
diamonds, respectively. The observability of the system is
checked by inspecting the null-space vector of the measurement
system using singular value decomposition (SVD). For the
given set of measurements, observable and unobservable islands
are reviewed as a byproduct using SVD [15]. SVD produces
a nullspace vector for each singularity. The nonzero entries

(14)

(15)

1684

IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 22, NO. 3, JULY 2007

Fig. 5. Measurement placements for the test system.

Fig. 6. Three-phase busbar voltages at Tiwai 220 kV (both actual and estimated
are plotted).

Fig. 4. Electromagnetic transient state estimation.

of the nullspace vector indicate that the corresponding state

variables are unobservable.
To look at the transient state estimation performance, overvoltages caused by a sudden lost of a Tiwai load are simulated

in PSCAD/EMTDC. Half of the Tiwai load is disconnected for

100 ms at a simulation time of 0.8 s. Due to the lack of field
system data, the corresponding measurements in Fig. 5 are taken
from the PSCAD/EMTDC simulation and supplied to the TSE
algorithm. No information is given about the disturbance location; thus, all three load locations are treated as suspicious
sources and are separated from the backbone network. Furthermore, no measurements are taken at the disturbance area (i.e.,
Tiwai 220 kV). The estimation is performed using a 50- s time
step.
Figs. 610 show the three-phase load voltages. The estimated
and simulated results are shown as dotted and solid lines, respectively. Both estimated and actual results are plotted, but are
indistinguishable due to similarity. Examples of the difference

YU AND WATSON: AN APPROXIMATE METHOD FOR TRANSIENT STATE ESTIMATION

Fig. 7. Voltage difference at Tiwai 220 kV.

Fig. 8. Three-phase busbar voltages at Roxburgh 11 kV (both actual and estimated are plotted).

Fig. 9. Closer look at the voltage waveform at Roxburgh 11 kV.

between the actual and estimated values are shown in Figs. 7

and 9 with reference to the Tiwai 220-kV load and Roxburgh
11-kV busbar voltages, respectively. Both figures showed that
some discrepancies occurred but decayed very quickly during
the disconnection and reconnection of the Tiwai load. This is
due to the very fast transient at the step change where the time

1685

Fig. 10. Three-phase busbar voltages at Invercargill 33 kV (both actual and

estimated are plotted).

Fig. 11. Three-phase load currents at Tiwai 220 kV (both actual and estimated
are plotted).

step used in the Eulers approximation [(14) and (15)] was inadequate. The error can be reduced if a smaller estimation time
step is used. Despite this, the estimates are a good approximation of the PSCAD/EMTDC simulation.
In this simulation, it is possible to determine the cause of the
transient from the estimation results. The load currents at the
three suspicious load locationsTiwai 220 kV, Roxburgh 11
kV, and Invercargill 33 kV provided by the estimator at the end
of TSE are shown in Figs. 1113.
The results clearly verify that there is a significant drop of
load current indicating a loss of load at Tiwai 220 kV. This
demonstrates that TSE provided a good estimate of the system
states and was capable of correctly identifying the origin and duration of the lost load, even when no measurements were made
at or near Tiwai 220 kV.
Estimation With Measurement Noise: Noise at 5%, which is
assumed to be normally distributed, is added to all of the measurements and the same scenario is resimulated. Due to the page
limitation, only the Tiwai 220-kV busbar voltages are shown
here (Fig. 14). A closer look at the two waveforms and the
voltage differences are shown in Figs. 15 and 16. A comparison can be made with reference to Fig. 6. Despite the presence

1686

Fig. 12. Three-phase load currents at Roxburgh 11 kV (both actual and estimated are plotted).

IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 22, NO. 3, JULY 2007

Fig. 15. Closer look at the voltage waveform at Tiwai 220 kV (estimated with
measurement noise).

Fig. 16. Voltage difference between the estimated and actual at Tiwai 220 kV.
Fig. 13. Three-phase load currents at Invercargill 33 kV (both actual and estimated are plotted).

V. CONCLUSION

Fig. 14. Three-phase busbar voltages at Tiwai 220 kV (estimated with measurement noise).

of measurement noise, the proposed TSE is still capable of providing a good estimate of the system states. In practice, the effect of the measurement noise can be minimized with the use of
redundant measurements and noise prefiltering stage.

The ability of TSE, using the proposed approximation method

to identify remote sources of disturbances and assessing the
system transient performances from limited measurements,
has been demonstrated. The disturbance location was correctly
identified from the three suspicious locations. The power flow
of the network loads was determined even when no information
on the load is supplied to the TSE (i.e., the loads are separated
from the network during TSE). This is particularly useful when
there are unexpected hidden disturbances within the network.
The proposed approximation method assumed that there were
minimal magnitude variations within the time step, thus for a
system experiencing very fast transients, the time step needs
to be sufficiently small enough for accurate estimation. TSE
offers an effective and systematic approach to assess the effect
of electromagnetic disturbances. The developed MATLAB
implementation of the TSE algorithm (using a 50- s time step)
is presently too slow to run in real time. The simulation takes
923 s of CPU (AMD 2000+) time to complete 1 s of simulation
and the computation time per time step is 46.15 ms. With
greater computer processing power and an optimized algorithm
using the compiled code, the computation time is expected to
reduce significantly.

YU AND WATSON: AN APPROXIMATE METHOD FOR TRANSIENT STATE ESTIMATION

REFERENCES
[1] H. W. Dommel, Digital computer solution of electromagnetic transients, IEEE Trans. Power App. Syst., vol. PAS-88, no. 4, pp. 388399,
Apr. 1969.
[2] J. J. Marti and L. R. Linares, Real-time EMTP- based transients simulation, IEEE Trans. Power Syst., vol. 9, no. 3, pp. 13091317, Aug.
1994.
[3] W. Long, D. Cotcher, D. Ruiu, P. Adam, S. Lee, and R. Adapa,
EMTPa powerful tool for analyzing power system transients,
IEEE Computer Appl. Power, vol. 3, no. 3, pp. 3641, Jul. 1990.
[4] P. Lehn, J. Rittiger, and B. Kulicke, Comparison of the ATP version
of the EMTP and the NETOMAC program for simulation of HVDC
systems, IEEE Trans. Power Del., vol. 10, no. 4, pp. 20382053, Oct.
1995.
[5] J. A. Martinez, Time domain electromagnetic transients programs,
in Proc. IEEE Power Eng. Soc. Summer Meeting, 2000, vol. 2, pp.
763764.
[6] F. C. Schweppe, J. C. Wildes, and D. Rom, Power system static
state estimation, Part I, II and III, IEEE Trans. Power App. Syst., vol.
PAS-89, no. 1, pp. 120135, Jan. 1970.
[7] G. T. Heydt, Identification of harmonic sources by a state estimation
technique, IEEE Trans. Power Del., vol. 4, no. 1, pp. 569576, Jan.
1989.
[8] J. E. Farach, W. M. Grady, and A. Arapostathis, An optimal procedure
for placing sensors and estimating the locations of harmonic sources in
power system, IEEE Trans. Power Del., vol. 8, no. 3, pp. 13031310,
Jul. 1993.
[9] A. P. S. Meliopoulos, F. Zhang, and S. Zelingher, Power system harmonic state estimation, presented at the IEEE Power Eng. Soc. Winter
Meeting, 1994.
[10] Z. P. Du, J. Arrillaga, and N. R. Watson, Continuous harmonic state
estimation of power systems, Proc. Inst. Elect. Eng., Gen., Transm.
Distrib., vol. 143, no. 4, pp. 329336, 1996.

1687

[11] S. Bousghiri-Kratz, O. Malasse, and W. Nuninger, Fault detection

using state estimation. Application to an electromechanical process,
in Proc. 34th IEEE Conf. Decision and Control, Dec. 1995, vol. 3, pp.
23972402.
[12] F. N. Choudhury, J. P. Christensen, and J. L. Aravena, Power system
fault detection and state estimation using Kalman filter with hypothesis
testing, IEEE Trans. Power Del., vol. 6, no. 3, pp. 10251030, Jul.
1991.
[13] M. E. Baran and A. W. Kelley, State estimation for real-time monitoring of distribution systems, IEEE Trans. Power Syst., vol. 9, no. 3,
pp. 16011609, Aug. 1994.
[14] J. Arrillaga, N. R. Watson, and S. Chen, Power System Quality Assessment. New York: Wiley, 2000, Appendix A.
[15] K. K. C. Yu and N. R. Watson, Three-phase harmonic state estimation
using SVD for partially observable systems, in Proc. Powercon Conf.,
Nov. 2004, vol. 1, pp. 2934.

Kent K. C. Yu received the B.E. (Hons.) degree from the University of Canterbury, Christchurch, New Zealand, in 2001, where he is currently pursuing the
Ph.D. degree.
His area of research is state estimation techniques.

Neville R. Watson (SM99) received the B.E. (Hons.) and Ph.D. degrees
in electrical and electronic engineering from the University of Canterbury,
Christchurch, New Zealand.
Currently, he is Associate Professor at the University of Canterbury. His interests include power quality, steady-state, and dynamic analysis of ac/dc power
systems.