You are on page 1of 14

Dynamic-State Estimation

Ali Abur
Northeastern University, Boston, MA, USA

1 Introduction
Power system static-state estimation was introduced by Schweppe, Wildes, and Rom (1969) and his cowork-
ers in the late 1960s and since then, it has become one of the crucial applications used in modern energy
management systems (EMS). Recent advances in monitoring capabilities of power grids motivated several
investigations concerned about tracking not only the quasi-static bus voltage phasors but also the dynamic
states of generators and loads via the use of dynamic-state estimators (DSEs).
The main objective of static-state estimation is to monitor the quasi-steady-state operating conditions of the
power grid based on the measurements provided by the supervisory control and data acquisition (SCADA)
system. These measurements are received typically every few seconds and carry an unknown but usually
bounded time-skew owing to the lack of synchronization between individual measurements. These mea-
surements are commonly referred as SCADA measurements and include real/reactive power injections such
as generator outputs and loads, real/reactive power flows measured at the terminals of transmission lines,
and transformers as well as voltage and current magnitude measurements taken at substations and incident
lines or transformers. The time-skew among SCADA measurements adds to the uncertainty already present
in measurement equations owing to the measurement noise, network model inaccuracies, and errors intro-
duced at various stages of the communication path between the sensors and the state estimator. Static-state
estimators have evolved over the past several decades, and various methods have been developed to address
such errors and uncertainty in network model and parameters.
These developments enabled the results of state estimation to be successfully used to execute real-time con-
tingency analysis for power systems. Thus, one of the significant roles of static-state estimation in the overall
EMS is facilitation of static security assessment, which involves execution of several other applications such
as bus load forecasting, external network modeling, topology processing, and security-constrained optimal
power flow. Having a static-state estimator helps initialize real-time contingency analysis by providing the
best estimate of the real-time network model.
Static (or quasi static) state of the system is defined by the vector containing magnitude and phase angles
of voltage phasors at all system buses. Use of phasors implicitly assumes constant frequency steady-state

Smart Grid Handbook, Online © 2016 John Wiley & Sons, Ltd.
This article is © 2016 John Wiley & Sons, Ltd.
This article was published in the Smart Grid Handbook in 2016 by John Wiley & Sons, Ltd.
DOI: 10.1002/9781118755471.sgd070
2 Smart Grid Handbook

operating conditions. Static-state estimation uses an algorithm that processes the measurements and other
information known or acquired about the network model and converts them into the best estimate of the
system state. This algorithm considers a single snap-shot in time and uses the measurements and other
information taken at that time instant in order to estimate the system state that will be valid for that time
instant.
Consider the following measurement model:

Z = h(X) + e (1)

where h(.) is a nonlinear vector function that relates state vector X to error free measurement vector, e the
vector of measurement errors, and Z the measurement vector. It is important to note that the state and error
vectors are unknown, and it is not possible to solve them. Instead, an estimate of the state vector is found
by minimizing an objective function expressed in terms of the residuals, which are defined as the difference
between the actual and estimated measurements.
Here, it is worthwhile to note that the type of nonlinearity in the measurement function h(.) of Equation 1
depends on the coordinate system chosen in formulating the measurement equations. If the system state
vector is expressed in rectangular coordinates in terms of real and imaginary parts of the voltage phasors,
the measurement function h(.) will be a quadratic function of the state variables. On the other hand, using
the polar coordinates will result in different order nonlinearity owing to the use of trigonometric functions
of the voltage phase angles.
In formulating the state estimation problem, one of the most commonly used objective functions is the
weighted sum of squares of the residuals. Use of this objective function will yield the so-called weighted
least squares (WLS) estimator for the system states. Such an estimator is known to be the best linear unbi-
ased estimator when the measurement model is linear and measurement errors have a zero mean Gaussian
distribution. Minimization of this objective function can be accomplished by an iterative process where an
initial guess for the system state is iterated in the following manner until convergence:

Xk+𝟏 = Xk + 𝚺 ⋅ HT ⋅ R−𝟏 ⋅ [Z − h(Xk )] (2)

where k is the iteration counter, H = 𝜕h/𝜕x the measurement Jacobian, 𝚺 = (HT ⋅ R− 1 ⋅ H)− 1 , R = E(e ⋅ eT )
the assumed covariance of measurement error e, and Xk the state vector at iteration k.
Inverse of the matrix 𝚺 is denoted by G and referred as the gain matrix. The diagonal entries of the matrix
𝚺 represent the variance of the estimated state variables and can be used as a metric of state estimation
accuracy in evaluating state estimators (Göl, Abur, and Galvan, 2012).
The above formulation is valid at a single point in time when a set of measurements are received as vector Z.
An iterative solution for X is obtained starting with an initial guess X0 and by iteratively updating X using
Equation 2 until the difference between two consecutive estimates falls below a convergence threshold,
|Xk+1 − Xk | < 𝜀. This solution will indicate the best estimate for the state vector at the given time instant.
Initial guess for the state vector is commonly assumed to be the so-called flat start value, which is where all
bus voltage magnitudes are assumed to be 1.0 per unit and all phase angles are set equal to zero.

1.1 Tracking State Estimator


In the absence of sudden topology or load changes, power system states typically do not move abruptly,
resulting in a slowly changing system state. Recognizing this, Masiello and Schweppe (1970) observed that
the logical initial guess in a real-time implementation would be the estimated state from the previous state
estimation solution (Masiello and Schweppe, 1970). If state estimation can be carried out frequently enough

Smart Grid Handbook, Online © 2016 John Wiley & Sons, Ltd.
This article is © 2016 John Wiley & Sons, Ltd.
This article was published in the Smart Grid Handbook in 2016 by John Wiley & Sons, Ltd.
DOI: 10.1002/9781118755471.sgd070
Dynamic-State Estimation 3

so that the state does not move significantly between consecutive state estimation solutions, then only a single
evaluation of the update given by Equation 2 may be sufficient. This will then yield the following estimator,
which is also called the tracking state estimator:

X(tn+1 ) = X(tn ) + 𝚺(tn ) ⋅ HT (X(tn )) ⋅ R−𝟏 ⋅ [Z(tn ) − h(X(tn ))] (3)

where X(tn ) is the estimated state vector at time tn . The idea of tracking state estimator and its variations
were used by others to improve static-state estimator’s real-time performance. One example of this is where
the estimator is coupled with an exponential smoothing of individual measurements as a way to reject bad
data as described in Falcao, Cooke, and Brameller (1982).

1.2 Dynamic Estimator for the Network State


Given the dynamic nature of power system operation, it is also natural to incorporate information about
the time variation of the states in the estimation process. The idea of using information related to the time
variation of the system states and the initial concepts of dynamic-state estimation for power systems are first
described in Debs and Larson (1970). Dynamic power system state estimation problem is initially formulated
based on a quasi-static model for the system states. Discrete-time equations for the quasi-static system state
and the measurements are written as (Debs and Larson, 1970):

Xk = Xk − 𝟏 + wk − 𝟏 (4)
Zk = h(Xk ) + ek (5)

where:

Zk and Xk are the measurement and state vectors at discrete-time instant k,


w is a white Gaussian noise vector with zero mean and covariance Q,
e is the measurement error vector with zero mean and covariance R.

Note that the definition of the system state remains identical to the one used in static-state estimation in
this formulation, that is, the voltage phasor, typically expressed in polar coordinates as a vector of voltage
magnitudes and phase angles at all system buses. However, it is interesting to note that while not attempting
to pursue the option in their paper, Debs and Larson (1970) mention the possibility of using an augmented
state vector y which will include, in addition to the system bus voltage phasors, dynamics of the power plants,
loads and the effects of various controls. They indicate that the tracking dynamic estimator described in the
paper should still be applicable under such conditions.
Several researchers followed up on the work reported in Debs and Larson (1970). One of the challenges
in formulating the tracking dynamic estimator is the lack of a model to represent dynamics of the sys-
tem state. The model used in Equation 4 assumes a quasi-static condition where the system state remains
unchanged other than an additive Gaussian noise with known covariance. In fact, most of the time the system
state changes rather smoothly in response to changes in load and generation. It can also change abruptly in
response to changes in network topology and parameters owing to line or transformer switching, movement
of transformer tap positions, or switching of capacitor banks or shunt reactors. Studies have looked into
improved dynamic models for system states and proposed the use of state forecasting based on forecasted
bus loads and associated power flow solution in an attempt to improve the dynamic model of the system
state (Nishiya, Hasegawa, and Koike, 1982; Leite da Silva et al., 1983).

Smart Grid Handbook, Online © 2016 John Wiley & Sons, Ltd.
This article is © 2016 John Wiley & Sons, Ltd.
This article was published in the Smart Grid Handbook in 2016 by John Wiley & Sons, Ltd.
DOI: 10.1002/9781118755471.sgd070
4 Smart Grid Handbook

Another issue is related to the handling of bad measurements versus network anomalies that are caused
by sudden topology changes. Large and sudden changes that deviate from expected values of measurements
will trigger suspicion of bad data or a topology change. Detecting these anomalies and differentiating them
from valid changes in operating conditions is challenging. One approach to address this issue is proposed in
Nishiya, Hasegawa, and Koike (1982), where an innovation-based approach to detect and identify anomalies
is described.
Finally, it is recognized that implementation of DSEs for very large systems presents significant compu-
tational challenges owing to the updating of large covariance matrices at every time step. One approach to
address this challenge is proposed via the use of hierarchical algorithms whose details are given in Leite da
Silva, Do Coutto Filho, and Diego de Siqueira (1985) and Rousseaux et al. (1988).

1.3 Power System Dynamic-State Estimator


While the above investigations provided improvements in tracking the bus voltages at faster rates and
improved capabilities of detecting and removing gross errors, the system model remained essentially
unchanged. It excluded modeling of loads, generators, and any other dynamics associated with control
devices connected to the high-voltage system. Gradually, the need to develop a modularized DSE capable
of estimating small and large dynamic fluctuations in phase angle and frequency is recognized in Modir
and Schlueter (1981) and Modir and Schlueter (1984), where the authors mentioned possible utilization of
such an estimator for future dynamic security assessment applications. More than three decades after the
publication of these papers, computer, communication, and monitoring technologies have reached desired
levels that allow real-time implementation and further advancement of these ideas.
One of the changes experienced by system operators in the recent years is the unusual power flow patterns
that are due to diverse power exchanges facilitated by power markets and increased penetration of renewable
energy sources with unpredictable and volatile power outputs. Thus, in addition to static security assessment,
system operators are faced with maintaining system stability under expected contingencies. This process is
generally referred as dynamic security assessment. This is a much more challenging task, requiring multiple
dynamic simulations of the system under considered contingencies. Initializing these simulations require
accurate knowledge of the dynamic state of the system, that is, those of the generators, loads, and any other
dynamic element connected to the system for control and protection purposes. Dynamic estimation of the
state of these devices will have to be carried out at refresh rates that are much higher than the static-state esti-
mation. This can be accomplished with the help of fast synchronized metering, wide-area communication,
and computation technologies available today.
One possible way to track the dynamic state of the overall generation, transmission, and distribu-
tion (modeled as dynamic loads) system using the above-mentioned technologies will be to design a
dynamic power system state estimator. Thus, formulation of the dynamic-state estimation problem first
requires identification of the mathematical model that will describe the time-varying behavior of the
dynamic-state variables in discrete time. This involves writing the differential equations, describing the sys-
tem dynamics and then transforming these equations into discrete-time domain using a robust discretization
method.
Consider that such a mathematical model resulted in the following set of nonlinear differential equations
representing the dynamic behavior of the system:

dy
= f (y, 𝝊) (6)
dt

Smart Grid Handbook, Online © 2016 John Wiley & Sons, Ltd.
This article is © 2016 John Wiley & Sons, Ltd.
This article was published in the Smart Grid Handbook in 2016 by John Wiley & Sons, Ltd.
DOI: 10.1002/9781118755471.sgd070
Dynamic-State Estimation 5

where f(.) represents the nonlinear system function, y the vector of state variables, and 𝝊 the vector of
modeling error. Equation 6 can be discretized and written as a difference equation in the following form:

yk + 𝟏 = f ( yk ) + 𝝊k (7)

where:

yk ∈ ℜn is the discrete-time state vector at time-step k,


𝝊k ∼ N(0, Qk ) is the process and modeling error at time-step k, which is assumed to have a Gaussian distri-
bution with zero mean and Qk covariance.

In the special case of a single synchronous generator, differential equations describing the dynamic behav-
ior of the generator represented by its classical model will have two state variables: rotor (or power) angle
of the generator, 𝛿 and angular frequency, 𝜔. Thus, when developed for a single generator, the state vector
y and the nonlinear function f will take the following form:
[ ]
𝛿
y= (8)
𝜔
[ ] [ ]
{ 𝜔 − 𝜔0 } 𝜈
f (y, 𝝊) = 𝜔0 D(𝜔−𝜔0 ) + 𝛿 (9)
PM − PG − 𝜔 𝜈𝜔
2H 0

where:

H is the generator inertia constant,


D is the damping coefficient,
PM is the mechanical power input to the generator,
PG is the net real power output of the synchronous generator,
𝜔0 is the synchronous angular frequency given by 2𝜋f0 ( f0 is the power frequency in hertz.)

In a large power grid with several synchronous generators, the state vector y will include a pair of state
variables per generator and therefore will have a dimension of 2NG , where NG represents the total number
of generators in the system. This dynamic-state vector for the generators can be augmented by a vector of
algebraic state variables of the transmission network, namely the vector of all bus voltage magnitudes and
phase angles excluding the slack bus. For a system with N buses, this augmented state vector at time instant
k can be expressed as follows:
xTk = [𝜹Tk 𝝎Tk V Tk 𝜽Tk ] (10)

where:

𝜹Tk = [𝛿k1 , 𝛿k2 , … , 𝛿kNG ]


𝝎Tk = [𝜔k1 , 𝜔k2 , … , 𝜔kNG ]
V Tk = [Vk1 , Vk2 , … , VkN ]
𝜽Tk = [𝜃k2 , 𝜃k3 , … , 𝜃kN ]
Vki , 𝜃 ki are the voltage magnitude and phase angle of bus i at time-step k,
𝛿 ki , 𝜔ki are the rotor angle and angular frequency of generator i.

Smart Grid Handbook, Online © 2016 John Wiley & Sons, Ltd.
This article is © 2016 John Wiley & Sons, Ltd.
This article was published in the Smart Grid Handbook in 2016 by John Wiley & Sons, Ltd.
DOI: 10.1002/9781118755471.sgd070
6 Smart Grid Handbook

Now, consider the case where a set of measurements can be obtained at fast enough rate that will allow
estimation of the dynamic machine variables defined by vector y. Given that net power output of a generator
is a widely available measurement, it will be assumed that every generator output will be measured. Then,
the state and measurement equations will be given by:
Hi 𝜔 − 𝜔0
𝜔̇ = PMi − PGi − Di i i = 1, … , NG (11)
𝜋f 0 i 𝜔0
𝛿̇ i = 𝜔i − 𝜔0 i = 1, … , NG (12)
IG = YG E (13)

PGi = Re(Ei IGi ) i = 1, … , NG (14)

where:

Hi , PMi , PGi , and Di are the inertia constant, mechanical input power, injected active power, and damping
coefficient for the ith generator, respectively,
YG is the reduced admittance matrix obtained from Ybus replacing the loads by equivalent constant admit-
tances and using generator subtransient reactances connected between generator terminal buses and inter-
nal excitation voltage buses (Kundur, 1994; Glover, Sarma, and Overbye, 2008).
IG is the vector of injected currents by all generators,
Ei = |Ei | ∠ 𝛿 i is the excitation voltage for generator i.

Using all available measurements to form the measurement vector z, and discretizing the dynamic-state
equations and the measurement equations using one of the available numerical integration methods such as
the second-order Runge–Kutta, the following set of discrete-time equations can be obtained:

xk + 𝟏 = f (xk , k) + 𝝊k (15)
zk = h(xk , k) + ek (16)

where:

zk is the measurement vector of dimension m recorded at time-step k,


xk is the augmented state vector defined in Equation 10,
𝝊k ∼ N(0, Qk ) is the process noise at time-step k, assumed to have a Gaussian distribution with zero mean
and covariance Qk .
ek ∼ N(0, Rk ) is the measurement noise at time-step k, assumed to have a Gaussian distribution with zero
mean and covariance Rk .

A dynamic estimator for the augmented state vector can be obtained by applying the extended Kalman
filter (EKF) to the discrete-time state and measurement equations of 15 and 16. EKF has to be initialized
from an initial state x0 and its associated covariance matrix P+
𝟎
as follows:

x̂ +
𝟎
= E(x𝟎 ), P+
𝟎
= E[(x𝟎 − x̂ +
𝟎
)(x𝟎 − x̂ +
𝟎
)T ] (17)

where E represents the “expected value” operator. Given the discrete-time state and measurement
equations 15 and 16, respectively, EKF can be implemented in two steps, prediction and correction, as
outlined below:

Smart Grid Handbook, Online © 2016 John Wiley & Sons, Ltd.
This article is © 2016 John Wiley & Sons, Ltd.
This article was published in the Smart Grid Handbook in 2016 by John Wiley & Sons, Ltd.
DOI: 10.1002/9781118755471.sgd070
Dynamic-State Estimation 7

Prediction:

x̂ − x+
k = f k − 𝟏 (̂ k−𝟏
, k − 1) (18)
P−
k = Fk − 𝟏 P+ FT
k−𝟏 k−𝟏
+ Lk − 𝟏 Qk − 𝟏 LTk − 𝟏 (19)

where:
𝜕f k − 𝟏
Fk − 𝟏 = |+ (20)
𝜕x ̂xk−1

𝜕f k − 𝟏
Lk − 𝟏 = |+ (21)
𝜕𝝊 ̂xk−1

Correction:

K k = P− T − T T −𝟏
k Hk (Hk Pk Hk + M k Rk M k ) (22)
x̂ +
k
= x̂ − x−
k + K k (zk − h(̂ k , k)) (23)
P+
k
= (I − K k Hk )P−
k (24)

where:
𝜕hk
Hk = |− (25)
𝜕x ̂xk

𝜕hk
Mk = |− (26)
𝜕e ̂xk
Note that the superscripts − and + indicate instants right before (a priori) and right after (a posteriori)
the update step. “Hat” over a variable implies “estimated value” and subscript k is used to indicate the
discrete-time instant. Considering the prediction and update equations, computational burden can become
prohibitively large when applying dynamic-state estimation to very large-scale power systems containing
several thousand generator and load buses. Each generator’s terminal voltage phasor and complex power
injection will be dependent on the operating conditions of the rest of the system. If these variables are per-
fectly known or can be sampled without any errors at the same sampling frequency as the desired output rate
of the dynamic estimator, then individual generators can use their individual EKFs to estimate their machine
variables without having to rely on a central integrated dynamic estimator of the entire system. However,
given the fact that terminal voltage and power measurements will always be noisy and may even have occa-
sional gross errors, it will not be possible to reliably run such individual EKFs solely based on generator
terminal measurements. A viable alternative may be to establish a fast and robust tracking estimator for the
network bus voltage phasors and use the results of this estimator to provide the necessary measurements
to individual generators at their terminal buses. Specifically, a two-stage DSE can be implemented where
the first stage involves a robust network static-state estimator providing bus voltage estimates and the sec-
ond stage involves independently and simultaneously executed dynamic estimators for individual generators
based on the estimated values of the measurements at generator terminals.

Smart Grid Handbook, Online © 2016 John Wiley & Sons, Ltd.
This article is © 2016 John Wiley & Sons, Ltd.
This article was published in the Smart Grid Handbook in 2016 by John Wiley & Sons, Ltd.
DOI: 10.1002/9781118755471.sgd070
8 Smart Grid Handbook

2 Two-Stage Dynamic-State Estimator


Consider a single generator and the corresponding discrete-time state and measurement equations as given
below:

Xk + 𝟏 = f (Xk , uk ) + 𝝊k (27)
Zk = h(Xk ) + ek (28)

where:

Xk is the discrete-state vector for the generator at discrete-time k,


Zk is the discrete measurement vector at discrete-time k,
uk is the vector of inputs at discrete-time k,
𝝊k ∼ N(0, Qk ) is the Gaussian process noise at discrete-time k,
ek ∼ N(0, Rk ) is the Gaussian measurement noise at discrete-time k,
Qk and Rk are the covariance matrices of 𝝊k and ek , respectively.

The dimension and contents of the state vector Xk will depend on the complexity of the state dynamics
being modeled. For a given synchronous generator, which is represented by a two-axis synchronous gener-
ator model that uses the commonly known IEEE-Type1 exciter, the dynamic-state vector will include the
following variables:
X = [𝛿, 𝜔, Eq′ , Ed′ , Efd , Vf , VR ]T (29)

where:

𝛿 and 𝜔 are the rotor angle and angular frequency,


Eq′ and Ed′ are the d- and q-axis transient voltages,
Efd , Vf , and VR are the field voltage, the scaled output of the stabilizing transformer, and the scaled output of
the amplifier (or pilot exciter), respectively (Sauer and Pai, 1998; Glover, Sarma, and Overbye, 2008).

Note that in this two-stage dynamic estimator, the quasi-static bus voltage phasors constitute the inputs
in Equation 27 and are represented by the vector uk at time-step k. These inputs include the terminal bus
voltage magnitude and phase angle as well as the net power output of the generator. In order to provide
these inputs without gross errors, a fast and robust phasor measurement unit (PMU)-based linear estimator
can be used. Such an estimator can be implemented and will constitute the first stage of the two-stage
dynamic estimation algorithm. A block diagram of the implementation for the two-stage DSE is shown in
Figure 1. While the linear estimator provides static-state estimation results for bus voltages based on the
synchronized phasor measurements, the second-stage dynamic estimator tracks the dynamic states of each

^ ^
V1,θ1
Robust linear u
phasor Distributed
x^
estimator ^ ^ DSE
P1,Q1
z

Figure 1 Two-stage dynamic-state estimator

Smart Grid Handbook, Online © 2016 John Wiley & Sons, Ltd.
This article is © 2016 John Wiley & Sons, Ltd.
This article was published in the Smart Grid Handbook in 2016 by John Wiley & Sons, Ltd.
DOI: 10.1002/9781118755471.sgd070
Dynamic-State Estimation 9

generator. In implementing the DSE, it is noted that the estimated terminal voltage phasor (magnitude and
phase) for the generator is used as the input and the estimated active and reactive power injections at the
generator terminals are used as the measurements for the DSE.

2.1 Stage 1: Quasi-Static Linear Phasor Estimator


It can be shown that full observability can be achieved by strategically placing PMUs at about a third of
the system buses (Xu and Abur, 2004). Furthermore, placement strategy can be modified not only to make
the entire system observable but also robust against loss of PMUs (Chen and Abur, 2006). Hence, a linear
phasor estimator can be implemented to estimate the quasi-static state of the system and it will rely solely
on installed PMUs and will implicitly assume full network observability by only PMUs. This estimator can
be executed at the rate or faster than the rate of PMU measurements. While the linear estimator can be
implemented using different algorithms such as WLS (Göl and Abur, 2015) or least absolute value (LAV),
owing to its robust nature, the latter method that minimizes the L1 norm of the measurement residuals
(Göl and Abur, 2014) will be assumed to be employed to obtain bus voltage estimates. More details on the
implementation of linear phasor estimators can be found in Göl and Abur (2015) and Göl and Abur (2014).

2.2 Stage 2: Dynamic-State Estimator (DSE)


The second stage involves a distributed set of DSEs, each one assigned to a synchronous generator. Each
DSE will thus estimate the dynamic-state variables given in Equation 29 associated with the above consid-
ered two-axis synchronous generator with IEEE-Type1 exciter (Sauer and Pai, 1998; Glover, Sarma, and
Overbye, 2008) and it can be implemented as a discrete-time EKF.
Dynamic equations related to a two-axis synchronous generator with IEEE-Type1 exciter (Sauer and Pai,
1998) are given below. Stator series resistance (Ra ) and machine saturation are neglected to simplify the
formulation.

𝛿̇ = 𝜔 − 𝜔0 (30)
H 𝜔 − 𝜔0
𝜔̇ = PM − Pe − D (31)
πf0 𝜔0
′ ̇′ ′ ′
T Eq = −Eq − (Xd − X )Id + Efd
do d (32)
′ ̇′
Tqo Ed = −Ed′ + (Xq − Xq′ )Iq (33)

where:
[ ] [ ][ ]
Vd sin (𝛿) −cos(𝛿) V cos(𝜃)
= (34)
Vq cos(𝛿) sin(𝛿) V sin(𝜃)
Eq′ − Vq Vd − Ed′
Id = , Iq = (35)
Xd′ Xq′
Pe = Vd Id + Vq Iq , Qe = −Vd Iq + Vq Id (36)

𝛿 is the rotor angle,


𝜔 is the rotor speed,

Smart Grid Handbook, Online © 2016 John Wiley & Sons, Ltd.
This article is © 2016 John Wiley & Sons, Ltd.
This article was published in the Smart Grid Handbook in 2016 by John Wiley & Sons, Ltd.
DOI: 10.1002/9781118755471.sgd070
10 Smart Grid Handbook

PM is the mechanical input power,


Eq′ and Ed′ are the d- and q-axis transient voltages,
V and 𝜃 are the generator terminal bus voltage magnitude and the phase angle, respectively,
Pe and Qe are the active and the reactive power delivered by the generator.

Definitions of the other machine parameters can be found in Glover, Sarma, and Overbye (2008) and Sauer
and Pai (1998).
Dynamic equations of the IEEE-Type1 exciter are given by:
Te Ė fd = VR − Ke Efd (37)
Kf KK
Tf V̇ f = −Vf + V − e f Efd (38)
Te R Te
Ta V̇ R = −VR + Ka (Vref − V − Vf ) (39)
where Efd , Vf , and VR are the field voltage, the scaled output of the stabilizing transformer, and the scaled
output of the amplifier (or pilot exciter), respectively (Sauer and Pai, 1998).
Equations 30–39 can be compactly expressed as follows:
ẋ = f (x, u, 𝝊) (40)
where:

x = [𝛿, 𝜔, Eq′ , Ed′ , Efd , Vf , VR ]T is the state vector as defined in Equation 29,
u = [V, 𝜃]T is the input vector of generator terminal voltage magnitude and phase angle,
𝝊 is the vector representing the process error.

3 Observability
Network observability analysis can be carried out by topological (Krumpholz, Clements, and Davis, 1980;
Clements, Krumpholz, and Davis, 1983) or numerical (Monticelli and Wu, 1985a, 1985b) methods for the
static-state estimation problem. Given the static nature of the problem, observability is determined by check-
ing the column rank of the measurement Jacobian H. Extra measurements referred as pseudo-measurements
can be added to the measurement set in case of unobservability so that the system can be made observable
again.
When using a DSE, the observability of a system is similarly defined as the ability to uniquely determine
the state of the system from available measurements. However, in this case, state dynamic equations will
also play a role in system observability. Observability for linear systems does not depend on the system
state or its inputs. It is mainly a function of the structure of the measurement and state equations. Consider
a discrete linear time-invariant system given by:
xk + 𝟏 = A ⋅ xk + B ⋅ uk (41)
zk = H ⋅ xk (42)
The system will be observable if the following observability matrix “O” has full rank:
⎡ H ⎤
⎢ H⋅A ⎥
O=⎢ ⎥ (43)

⎢ − 𝟏 ⎥
⎣H ⋅ A n ⎦

Smart Grid Handbook, Online © 2016 John Wiley & Sons, Ltd.
This article is © 2016 John Wiley & Sons, Ltd.
This article was published in the Smart Grid Handbook in 2016 by John Wiley & Sons, Ltd.
DOI: 10.1002/9781118755471.sgd070
Dynamic-State Estimation 11

where xk ∈ ℝn and Zk ∈ ℝm . In case of nonlinear systems, observability will no longer be a global property,
but will be determined locally around a given operating state or equilibrium point.
This can be done via the use of Lie derivatives of the nonlinear measurement function h with respect to the
nonlinear function describing system dynamics (Muske and Edgar, 1997). Consider the following nonlinear
state and measurement equation set:

Ẋ = f (x(t)) + u(x(t)) (44)


Z = h(x(t)) (45)

where f ∶ ℝn −→ ℝn and h ∶ ℝn −→ ℝm
Lie derivative of h with respect to f will be given by:

Lf h = ∇h ⋅ f (46)

By definition:
Lf0 h = h (47)

𝜕(Lfk−1 h)
Lfk h = ⋅f (48)
𝜕X

Define 𝛀 as:
⎡ Lf0 (h𝟏 ) · · · Lf0 (hm ) ⎤
⎢ L1 (h ) · · · Lf1 (hm ) ⎥⎥
𝛀=⎢ f 𝟏 (49)
⎢ ⋮ ··· ⋮ ⎥
⎢Ln−1 (h ) · · · Lfn−1 (hm )⎥⎦
⎣ f 𝟏

and a gradient operator as:


⎡ dLf0 (h𝟏 ) · · · dLf0 (hm ) ⎤
⎢ dL1 (h ) · · · dLf1 (hm ) ⎥⎥
O = d𝛀 = ⎢ f 𝟏 (50)
⎢ ⋮ ··· ⋮ ⎥
⎢dLn−1 (h ) · · · dLfn−1 (hm )⎥⎦
⎣ f 𝟏

The observability matrix “O” defined in Equation 50 must have rank “n” in order for the system to be
observable.
An alternative approach is, as done in small signal stability analysis, to replace dynamic state and measure-
ment equations by first-order (linear) approximations yielding a discrete linear system as in Equations 41
and 42. Then, a rank check on the observability matrix “O” defined in Equation 43 can be used as an approx-
imate test for observability. Such tests, however, have their limitations as evident from the simulation results
shown in Figures 2 and 3. They correspond to cases where Pe and both Pe and Qe are used as measurements
in dynamic estimation of a two-axis synchronous generator’s rotor speed. Note that first-order approxima-
tion yields full rank observability matrices for both cases despite the fact that the first case shows weak
observability especially after 15 s when only Pe is used as a measurement.

Smart Grid Handbook, Online © 2016 John Wiley & Sons, Ltd.
This article is © 2016 John Wiley & Sons, Ltd.
This article was published in the Smart Grid Handbook in 2016 by John Wiley & Sons, Ltd.
DOI: 10.1002/9781118755471.sgd070
12 Smart Grid Handbook

1.001

Rotor speed / sync. speed 1

0.999

0.998

0.997

0.996 ω/ωsync
ω+/ωsync

0.995
0 5 10 15 20 25 30
Time (s)

Figure 2 DSE results when using Pe as the only measurement

1.0005

1
Rotor speed / sync. speed

0.9995

0.999

0.9985

0.998 ω/ωsync

ω+/ωsync
0.9975
0 5 10 15 20 25 30
Time (s)

Figure 3 DSE results using both Pe and Qe as measurements

Smart Grid Handbook, Online © 2016 John Wiley & Sons, Ltd.
This article is © 2016 John Wiley & Sons, Ltd.
This article was published in the Smart Grid Handbook in 2016 by John Wiley & Sons, Ltd.
DOI: 10.1002/9781118755471.sgd070
Dynamic-State Estimation 13

4 Conclusions
This chapter provides a brief overview of power system dynamic-state estimation, which is being considered
as an extension or as a stand-alone application in power system control centers or generation substations. In
addition to monitoring and tracking real-time model of the generator-turbine system, it can facilitate effective
implementation of dynamic security assessment of large-scale power grids. As the systems are integrated
with advanced control devices and populated with loads with unconventional dynamic properties, initial-
ization of dynamic security assessment tools such as short-term dynamic simulations will become more
challenging. A scalable and computationally efficient dynamic estimator may readily address some of these
challenges. The chapter also presents issues of observability within the context of nonlinear dynamic-state
estimation and which cannot be addressed by direct extension of the observability analysis methods used in
static-state estimation problem.

Related Articles
Wide Area Monitoring through Synchrophasor Measurement;
Dynamic Security Assessment;
Grid Monitoring and State Estimation.

References
Chen, J. and Abur, A. (2006) Placement of PMUs to enable bad data detection in state estimation. IEEE Transactions on Power Systems,
21 (4), 1608–1615.
Clements, K.A., Krumpholz, G.R., and Davis, P.W. (1983) Power system state estimation with measurement deficiency: an observability
measurement placement algorithm. IEEE Transactions on Power Apparatus and Systems, 102 (7), 2012–2020.
Debs, A.S. and Larson, R.E. (1970) A dynamic estimator for tracking the state of a power system. IEEE Transactions on Power
Apparatus and Systems, PAS-89 (7), 1670–1678.
Falcao, D.M., Cooke, P.A., and Brameller, A. (1982) Power system tracking state estimation and bad data processing. IEEE Transac-
tions on Power Apparatus and Systems, PAS-101 (2), 325–333.
Glover, J.D., Sarma, M.S., and Overbye, T.J. (2008) Power Systems Analysis and Design, Thomson, Toronto.
Göl, M. and Abur, A. (2014) LAV based robust state estimation for systems measured by PMUs. IEEE Transactions on Smart Grid,
5 (4), 1808–1814.
Göl, M. and Abur, A. (2015) A fast decoupled state estimator for systems measured by PMUs. IEEE Transactions on Power Systems,
30 (5), 2766–2771.
Göl, M., Abur, A., and Galvan, F. (2012) Metrics for success: performance metrics for power system state estimators and measurement
designs. IEEE Power and Energy Magazine, 10 (5), 50–57.
Krumpholz, G.R., Clements, K.A., and Davis, P.W. (1980) Power system observability: a practical algorithm using network topology.
IEEE Transactions on Power Apparatus and Systems, PAS-99 (4), 1534–1542.
Kundur, P. (1994) Power System Stability and Control, McGraw Hill, New York.
Leite da Silva, A.M., Do Coutto Filho, M.B., and de Queiroz, J.F. (1983) State forecasting in electric power systems. IEE Proceedings
on Generation, Transmission and Distribution, Part-C, 130 (5), 237–244.
Leite da Silva, A.M., Do Coutto Filho, M.B., and Diego de Siqueira, M.S. (1985) Hierarchical Dynamic State Estimation in Electric
Power Systems. IFAC Conference on Electric Energy Systems, Rio de Janeiro, Brazil.
Masiello, R.D. and Schweppe, F.C. (1970) A Tracking Static State Estimator. Paper 70 TP 707-PWR, Proceedings of the IEEE Summer
Power Meeting and EHV Conference, Los Angeles, CA, July 12–17.
Modir, H. and Schlueter, R. (1981) A dynamic state estimator for dynamic security assessment. IEEE Transactions on Power Apparatus
and Systems, PAS-100 (11), 4644–4652.
Modir, H. and Schlueter, R.A. (1984) A dynamic state estimator for power system dynamic security assessment. Automatica, 20 (2),
189–199.
Monticelli, A. and Wu, F.F. (1985a) Network observability: theory. IEEE Transactions on Power Apparatus and Systems, PAS-104 (5),
1042–1048.

Smart Grid Handbook, Online © 2016 John Wiley & Sons, Ltd.
This article is © 2016 John Wiley & Sons, Ltd.
This article was published in the Smart Grid Handbook in 2016 by John Wiley & Sons, Ltd.
DOI: 10.1002/9781118755471.sgd070
14 Smart Grid Handbook

Monticelli, A. and Wu, F.F. (1985b) Network observability: identification of observable islands and measurement placement. IEEE
Transactions on Power Apparatus and Systems, PAS-104 (5), 1035–1041.
Muske, K. and Edgar, T. (1997) Nonlinear State Estimation (eds M.A. Henson and D.E. Seborg), Prentice-Hall, Upper Saddle River,
NJ.
Nishiya, K., Hasegawa, J., and Koike, T. (1982) Dynamic state estimation including anomaly detection and identification for power
systems. IEE Proceedings on Generation, Transmission and Distribution, Part-C, 129 (5), 192–198.
Rousseaux, P., Mallieu, D., Van Cutsem, T., et al. (1988) Dynamic state prediction and hierarchical filtering for power system state
estimation. Automatica, 24 (5), 595–618.
Sauer, P. and Pai, M. (1998) Power System Dynamics and Stability, Upper Saddle River, NJ, New Jersey.
Schweppe, F.C., Wildes, J., and Rom, D. (1969) Power System Static State Estimation: Parts I, II, III. Proceedings of the Power Industry
Computer Applications (PICA) Conference, Denver, CO, June.
Xu, B. and Abur, A. (2004) Observability Analysis and Measurement Placement for Systems with PMUs. Proceedings of the IEEE
Power Systems Conference and Exposition, NY, October 10–13.

Smart Grid Handbook, Online © 2016 John Wiley & Sons, Ltd.
This article is © 2016 John Wiley & Sons, Ltd.
This article was published in the Smart Grid Handbook in 2016 by John Wiley & Sons, Ltd.
DOI: 10.1002/9781118755471.sgd070

You might also like