Вы находитесь на странице: 1из 12

352 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 63, NO.

2, FEBRUARY 2014

Power System Dynamic State Estimation With


Synchronized Phasor Measurements
Farrokh Aminifar, Member, IEEE, Mohammad Shahidehpour, Fellow, IEEE,
Mahmud Fotuhi-Firuzabad, Senior Member, IEEE, and Saeed Kamalinia, Member, IEEE

Abstract— The dynamic state estimation (DSE) applied to


power systems with synchrophasor measurements would estimate j Measurement index.
the system’s true state based on measurements and predictions. l Line index.
In this application, as phasor measurement units (PMUs) are p Uncertain parameter index.
not deployed at all power system buses, state predictions would t Time period index.
enhance the redundancy of DSE input data. The significance
of predicted and measured data in DSE is affected by their Sets
confidence levels, which are inversely proportional to the corre- P Set of uncertain parameters.
sponding variances. In practice, power system states may undergo
drastic changes during hourly load fluctuations, component out- Parameters
ages, or network switchings. In such conditions, the inclusion of Il  δ Il Current phasor of line l.
predicted values could degrade the power system state estimation. Mi Large positive constant associated
This paper presents a mixed-integer programming formulation
of DSE that is capable of simultaneously discarding predicted with state variable i .
values whenever sudden changes in the system state are detected. NB Number of buses.
This feature enhances the DSE computation and will not require NMi Number of measurements associated
iterative executions. The proposed model accommodates system- with state variable i .
wide synchronized measurements of PMUs, which could be NS Number of states.
of interest to smart grid applications in energy management
systems. The voltage phasors at buses without PMUs are cal- NT Number of time periods.
culated via voltage and current measurements of adjacent buses, rimax , rimin Upper and lower limits of r̃it .
which are referred to as indirect measurements. The guide Vb  δb Voltage phasor of bus b.
to the expression of uncertainty in measurement is used for Y, θ y , G, B Line admittance parameters.
computing the confidence level of indirect measurements based on
Z , θz , R, X Line impedance parameters.
uncertainties associated with PMU measurements as well as with
transmission line parameters. Simulation studies are conducted αi , βi Parameters of the prediction model
on an illustrative three-bus example and the IEEE 57-bus power associated with state variable i .
system, and the performance of the proposed model is thoroughly σ̄i j Standard deviation of measurement j
discussed. associated with state variable i .
Index Terms— Dynamic state estimation, mathematical σp Standard deviation of parameter p.
programming, power system monitoring, state prediction,
synchrophasor measurement, uncertainty propagation. Variables
Dit , Cit Coefficients of the prediction model
N OMENCLATURE associated with state variable i at period t.
Indexes Jt Objective function value at period t.
r̃it Residual of prediction associated with state
b Bus index.
variable i at period t.
i State variable index.
r̄i j t Residual of measurement j associated with
Manuscript received August 12, 2012; revised June 2, 2013; accepted state variable i at period t.
June 3, 2013. Date of publication August 29, 2013; date of current version wit , σit Fluctuation and its standard deviation
January 2, 2014. The Associate Editor coordinating the review process was
Dr. Carlo Muscas.
associated with state variable i at period t.
F. Aminifar is with the School of Electrical and Computer Engineering, x it True value of state variable i at period t.
College of Engineering, the University of Tehran, Tehran 14174, Iran (e-mail: x̄ i j t Measurement j associated with the state
aminifar@ece.ut.ac.ir).
M. Shahidehpour is with the Galvin Center for Electricity Innovation, the
variable i at period t.
Illinois Institute of Technology, Chicago, IL 60616 USA (e-mail: ms@iit.edu). x̃ it , σ̃it Prediction and its standard deviation
M. Fotuhi-Firuzabad is with the Center of Excellence in Power system Con- associated with state variable i at period t.
trol and Management, Electrical Engineering Department, Sharif University
of Technology, Tehran 145888, Iran (e-mail: fotuhi@sharif.edu).
x̂ it , σ̂it Estimation and its standard deviation
S. Kamalinia is with the Power System Solutions Division of S&C Electric associated with the state variable i at period t.
Company, Chicago, IL 60626 USA (e-mail: saeed.kamalinia@sandc.com). γit Binary variable to discard the prediction
Color versions of one or more of the figures in this paper are available
online at http://ieeexplore.ieee.org.
associated with state variable i at period t.
Digital Object Identifier 10.1109/TIM.2013.2278595 RMSD Sate estimator performance index.
0018-9456 © 2013 IEEE
AMINIFAR et al.: POWER SYSTEM DSE WITH SYNCHRONIZED PHASOR MEASUREMENTS 353

I. I NTRODUCTION 4) PMUs report measurements to phasor data concen-


trator at the rate of 50/60 samples/s for 50/60 Hz

S TATE estimation (SE) computes the best estimate of the


true operating state of a power system as a prerequisite for
an efficient and reliable operation [1]. Since SE serves several
systems.
5) Similar to digital fault recorders, PMU devices are usu-
ally responsible for recording sinusoidal signals which
crucial functions such as the detection of abnormal conditions, are downloadable through the direct access to PMUs.
generator correction actions, and contingency analysis [2],
The incorporation of PMU measurements in the conven-
its responsibility is increasing with the dimension of power
tional DSE was discussed in [10]–[15]. In addition, the Electric
systems and emerging uncertain conditions in the operation of
Power Research Institute had a research project relevant to this
restructured systems.
issue [16]. To avoid deficits associated with the conventional
In general, SE is categorized into the following three
metering system such as asynchronous measurement and time
classes: 1) static SE (SSE); 2) tracking SE (TSE); and
skew errors, a plenary set of WAMS is intended in future
3) dynamic SE (DSE) [3]. SSE estimates the power sys-
power systems [17]. The high reporting rate of PMUs, com-
tem state associated with a given time and based on the
missioning high speed communication facilities in substations,
measurement set corresponding to that moment of time. The
and developing fast and efficient DSE algorithms are three
SSE algorithm is iterative that is initialized as a flat start.
key forces for decreasing the time interval of successive
Hence, it uses heavy computations and cannot be executed in
estimations. The reduction of time interval would be very
short intervals. To mitigate this burden, TSE was introduced
desirable for near real-time applications.
in which the estimation starts from the last calculated state
The most common DSE algorithms are variants of weighted
variables instead of a flat point. In both SSE and TSE, the
least squares (WLSs) estimation. These models are solved
power system state is estimated based on a single set of
using iterative matrix-based calculations along with mathemat-
measurements. A more advanced class of estimators is DSE,
ical heuristics to reduce the computation [18]. An alternative
also referred to as forecasted-aided SE, which is becoming
way for DSE is mixed-integer programming (MIP) formula-
more practical in modern energy management systems. DSE
tion, which can efficiently be solved by commercially available
possesses the ability to predict the power system state pro-
solvers [19].
gressively in short time steps. Furthermore, at each time step,
This paper presents an MIP formulation of DSE that is based
the estimation uses both measurement and prediction data
on PMU measurements. The proposed method accelerates the
sets. A comprehensive survey of DSE techniques and models
DSE execution and is capable of simultaneously discarding
along with the results of DSE implementation in practical
predictions from the estimation process when estimated states
power systems have been presented in [4] and [5]. Other
exceed the estimation confidence bound. In practice, it is
reviews of DSE techniques are available in [6] and [7]. One
neither economical nor essential to equip all buses with PMUs.
of the challenging aspects of applying DSE is the estimation
Hence, the power system state variables would be calculated
error arising from the utilization of historical-based predic-
based on PMU measurements at adjacent buses. The calculated
tion data during drastic changes in the power system state
confidence level would be affected by uncertainties associated
variables.
with PMU measurements. In addition, transmission line para-
A new generation of power system monitoring schemes
meters are usually assumed to be exact while they are not
embedded in the wide-area measurement system (WAMS) is
indeed. The uncertainty of these parameters is considered in
enabled by escalating the deployment of phasor measurement
this paper for the calculation of PMU indirect measurements.
units (PMUs) [8]. PMUs directly measure the state variables,
An illustrative example and the IEEE 57-bus test system are
i.e., magnitude and phase angle of bus voltages, with a very
examined in this paper for investigating performance of the
high accuracy. They can consequently increase the robustness
proposed DSE model.
and precision of the estimation process. The superiorities of
The rest of this paper is organized as follows. Section II
WAMS over the conventional metering infrastructure are as
describes the error propagation phenomenon in the calculation
follows [9].
of voltage phasors associated with buses not having PMU.
1) PMUs are installed with specific guidelines, tested dur- A general introduction to DSE is described in Section III.
ing the commissioning process, and calibrated periodi- The proposed DSE model and formulation are presented in
cally. Section IV. Section V discusses the numerical examples.
2) PMU devices are furnished with advanced computation Concluding remarks are outlined in Section VI. The Appendix
algorithms and self-check/self-diagnostic capabilities. In presents formulas derived for the calculating error propagation.
addition, PMUs are equipped with 16 bit or upper A/D
converters providing an extra high sampling rate and II. S TANDARD D EVIATION OF I NDIRECT M EASUREMENTS
very accurate measurements. A PMU installed at any bus would measure the corre-
3) GPS receiver is added to the PMU structure for syn- sponding voltage and current phasors. Using the phasors at
chronizing the A/D phase locked loop (sampling clock) one end of a transmission line, the corresponding phasors
and accompanying the PMU data with the exact time can be calculated at the other end [9]. The DSE algorithm
of measurement. The quality of time synchronization incorporates the confidence level of all measurements either
associated with the data is sent as well. directly measured or indirectly calculated. Accordingly, the
354 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 63, NO. 2, FEBRUARY 2014

Vs =Vs∠δs Z∠θz = R+ jX Vr =Vr∠δr distribution function is assumed for the random transmission

s r parameter error in which the standard deviation is (2 3)−1
Is = Is∠δIs
PMU of the associated confidence intervals. The calculations in this
Y∠θy = G + jB Y∠θy = G + jB section correspond to buses without PMU. The PMU place-
ment scheme is assumed to be given by either deterministic
or probabilistic studies [25]–[28].
Fig. 1. Phasor measurement and calculation in a transmission line.
III. DYNAMIC S TATE E STIMATION
A. General Background
confidence in the indirect phasors measurement is required
which can be calculated by three approaches including the DSE has two objectives: 1) prediction of power system
uncertainty propagation, Monte Carlo simulation, and ran- state at the next time period and 2) SE based on both sets
dom fuzzy variables. Fixed transmission line parameters are of predicted and measured data. The first feature provides the
assumed in [20], which may not be the case in practice. power system operator an additional time for making control
For instance, the line resistance is dependent on the ambient decisions and analyzing the security of operating system.
temperature, and the asymmetry of three-phase currents could Talking specifically about WAMS, this feature is less important
impact the line reactance. A typical uncertainty bound for line since the measurement time instants are very close. The second
parameters is ±2% [21]. Hence, it is imperative to model line attribute would significantly improve the performance of DSE.
parameter uncertainties, which are discussed in this section. The consideration of predicted values in the estimation process
The inaccuracy of conventional measurement devices for would enhance the data redundancy and make the DSE more
measuring voltage, current, and power quantities, is usually robust as compared with SSE and TSE, which use real
over ±1% of the full scale [18]. In contrast, PMU mea- measurements alone.
surements are much more precise and time tagged with an Two consecutive stages are recognized in DSE which are
accuracy of better than 1 μs [9]. They are, however, still state prediction and SE (it is referred to as state filtering
prone to precision errors due to the inaccuracy of PMU in some documents). The power system state at the next
instruments including transformers, cables, and A/D convert- time period is predicted at the first stage, and upon receiving
ers [22]. The first two sets of random errors are mostly biased the measurement set, the system state is determined by the
and can be compensated by sophisticated PMU calibration estimation process.
techniques [23]. The A/D-induced errors are more random
and consequently difficult to eliminate. The total vector error B. Major DSE Challenges
(TVE) is a measure of the PMU measurement accuracy level Power systems follow a quasi-static regime as the daily
with acceptable ranges in steady-state and dynamic conditions, load profile is followed by small random fluctuations. This
which are provided in the IEEE C37.118.1 standard [24]. situation is referred to as normal operating condition in which
Fig. 1 shows the π model of a transmission line between the power system state varies slowly and the performance
buses s and r and a PMU located at bus s. The PMU measures of DSE is highly satisfactory. The major DSE challenge
the voltage phasor of bus s and the current phasor of the line occurs during sudden changes in power system conditions
connected to bus s. Hence including abrupt changes in load and generation and network
Vr  δr = Vs  ϕ1 − Is Z  ϕ2 + Vs Y Z  ϕ3 (1) reconfiguration [29]. When a sudden change occurs, state
variables that are geographically close to the affected area
where ϕ1 = δs , ϕ2 = δ I s + θz , and ϕ 3 = δs + θ y + θz . could experience large transitions. Hence, there could be
Since large differences between actual measurements and predictions
based on historical data. In such cases, the predictions could
P = {Vs , δs , Is , δ I s , Z , θz , Y, θ y } (2)
affect the final estimated values and render degradation of the
through applying the guide to the expression of uncertainty in overall performance of DSE. The proposed solutions to this
measurement (GUM), the standard deviations of Vr and δr are shortcoming are reviewed as follows.
calculated as The first proposed approach to detect the sudden change

  2  2 condition is based on the innovation analysis [3]–[5],
σVr = ∂ Vr ∂ p σ p (3) [29]–[32]. This method compares the predicted measurements
p∈P with the real measurement set and decides whether or not

  2  2 a sudden change condition has occurred. Two methods are
σδr = ∂δr ∂ p σ p (4)
offered to deal with this situation: 1) deemphasizing the
p∈P
importance of predicted values by reducing their weight (or
and the corresponding detailed formulation is presented in the analogously attach a higher weight to the measurements) and
Appendix. 2) labelling the prediction as unreliable and perform SSE or
We assume that the random measurement error is repre- TSE.
sented by a normal distribution. Thus, the standard deviation An algorithm to increase the robustness of DSE has been
of each PMU measurement, i.e., Vs , δs , Is , and δ I s , would be proposed in [33] and [34]. In contrast to the approach based
one sixth of the corresponding confidence intervals. A uniform on the innovation analysis, this technique does not need
AMINIFAR et al.: POWER SYSTEM DSE WITH SYNCHRONIZED PHASOR MEASUREMENTS 355

wi (t −1) equations were proposed by [36] for the Holt’s method


+ xit xi (t −1) Ci(t −1) = αi (1 + βi ) (8)
Di (t −1) ∑ z-1
+ Di(t −1) = (1 + βi )(1 − αi )x̃ i(t −1)
+

Ci (t −1) − βi Si(t −2) + (1 − βi )bi(t −2) (9)


Si(t −2) = αi x̂ i(t −2) + (1 − αi )x̃ i(t −2) (10)
Fig. 2. Dynamic system with linear characteristic. bi(t −2) = βi [Si(t −2) − Si(t −3) ] + (1 − βi )bi(t −3) (11)

where αi and βi are constant parameters determined by trial


to detect the anomaly conditions and remedy them at two and error. It is evident that the model parameters depend on
separate stages. In [33] and [34], it was illustrated that the previous states themselves. The above formulas were derived
proposed method, in spite of its effectiveness during bad data from the canonical representation of Holt’s method [36].
and topology error conditions, does not work well in sudden In the DSE problem, wi(t −1) presents the small random
change conditions. The application of fuzzy controllers to fluctuation in the power system state. It is a function of
intelligently guide the solution to a near-optimal trajectory has the power system load, generation schedule, and network
been proposed in [35]. This technique calculates the difference configuration associated with the corresponding time sample.
between predicted and estimated states. If the difference is The mean value of wi(t −1) is usually assumed to be zero;
greater than a small threshold, a compensation vector is however, its variance is determined by either historical records
generated and the estimation stage is repeated. Accordingly, of state variable changes or a series of off-line simulations.
the estimation stage of DSE will be longer.
B. State Estimation
IV. P ROPOSED DSE F ORMULATION AND M ETHODOLOGY SE (i.e., filtering) is the second stage of DSE that is executed
upon receiving the measured data set. At this stage, both
For a power system with NB buses, the state vector will
prediction and measurement sets are employed to reach to the
have NB bus voltage magnitudes and NB phase angles; hence, best estimation of the power system state. Note that this feature
the number of state variables, NS, would be the same as
is very desirable in WAMS in which the level of measurement
2NB. In the following, the two DSE stages are formulated redundancy is rather low since PMU devices are not deployed
and discussed. at all network buses [9].
Here, a new formulation of estimation process of DSE is
presented. The estimation process is formulated based on the
A. State Prediction
WLS model, which is widely used in the estimation problems.
In a linear dynamic power system, shown in Fig. 2, the The objective function of the estimation stage is
estimation process is presented as ⎧ ⎫
⎨ i (x̄ i j t − x̂ it )2 
NS NM NS
(x̃ it − x̂ it ) ⎬
2
x it = Ci(t −1) x i(t −1) + Di(t −1) + wi(t −1) (5) min Jt = + . (12)
⎩ σ̄i j
2 σ̃it2 ⎭
i=1 j =1 i=1
where x it is the system state, Ci(t −1) and Di(t −1) are model
coefficients, and wi(t −1) is the random change of the power The first term in (12) shows the weighted difference of the
estimation with measurement and the second term represents
system state. wi(t −1) is a white Gaussian sequence with zero
the weighted difference between the estimation and prediction.
mean and the standard deviation of σi(t −1) .
Equation (5) shows that the new power system state could In (12), the state variables are bus voltage phasor values
processed in the polar coordinate (namely magnitude and
be represented as a function of the past state (s). Accordingly,
the next power system state is readily predicted based on the phase angles). Note as well that PMU measurements are
synchronized with respect to the time reference provided by
information obtained in t − 1.
GPS satellites, which eliminate the need for a reference bus
Applying the conditional expectation operator on (5), the
power system state at the next time period is obtained as angle as in conventional state estimators [37].
As discussed earlier, we should discard the predicted values
x̃ it = Ci(t −1) x̂ i(t −1) + Di(t −1) (6) from the estimation process in the drastic change condition.
In this respect, two approaches are discussed as follows.
and the variance associated with the prediction would be 1) Discarding by Process: In this approach, the innovation
process [30] is employed. Upon receiving the new WAMS
σ̃it2 = Ci(t
2 2 2
−1) σ̂i(t −1) + σi(t −1) . (7) data set, the predicted values are compared with new mea-
surements, predictions which are far from the corresponding
The coefficients used in (5) are specified according to the measurements are identified, and associated terms are omitted
adopted prediction method. The commonly used prediction from (12). The remaining predictions along with all measure-
technique in DSE is the Holt’s linear exponential smooth- ments establish the objective function (12). This optimization
ing technique [35], which is employed here. The following problem can readily be solved by calculating derivative of (12)
356 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 63, NO. 2, FEBRUARY 2014

and equating it to zero. The solution would be as long as the estimated value of variable i is unknown.
Accordingly, σ̂it is a dependent variable, which is defined in
i
NM
x̄ i j t x̃ it
σ̄i2j
+ σ̃it2
terms of independent variables. Similar to the calculation of
j =1 (13), the solution of (14) is given as
x̂ it = , i = 1, . . . , Ns . (13)
i
NM
i
NM
1
σ̄i2j
+ 1
σ̃it2
x̄ i j t
+ (1 − γit ) σ̃x̃it2
j =1 σ̄i2j
j =1 it
x̂ it = . (17)
Using this approach, the solution of the optimization prob- i
NM
lem is readily computable and the time interval between
1
σ̄i2j
+ (1 − γit ) σ̃12
j =1 it
estimations would be significantly decreased. There are, how-
ever, disadvantages attributed to this approach. The first issue If we apply the variance operator to (17) and assume that
is in relation with the comparison criteria. For each state x̄ i j t , j = 1, 2, · · · , Nmi , and x̃ it are independent, we will have
variable, we might have several redundant measurements along 1
with one prediction. It could be challenging to decide which σ̂it =  . (18)
one of the measurements should be considered to measure i
NM
1
σ̄i2j
+ (1 − γit ) σ̃12
the distance of prediction. The other disadvantage of this j =1 it

approach is that when the prediction is compared with a The proposed model is summarized as follows: 1) (14)
measurement, or in general the equivalent of all measurements, is the objective function; 2) (16) expresses the constraints;
the effect of prediction is not considered in the reference and 3) (15) and (18) define auxiliary variables. The solution
point. Whereas the inclusion of the impact of prediction in the methodology is proposed next.
comparison reference values would make the rejection process The current representation of the model is highly nonlinear
more authentic. that includes the multiplication of a binary variable and a
2) Simultaneous Discarding: Another technique is proposed quadratic term in (14), absolute value function in (15), and
here to overcome the disadvantages associated with the first binary variable in denominator in (18). Thus, the current model
approach. With the considerations discussed earlier, the second is the mixed-integer nonlinear program (MINLP). The present
term of (12) should be omitted from the objective function MINLP solvers are not powerful enough to effectively solve
during the sudden change condition. This characteristic is this type of problems even with a long execution time.
achievable by accommodating binary variables γit in (14)
⎧ ⎫
⎨ i (x̄ i j t − x̂ it )2 
NS NM NS
(x̃ it − x̂ it ) ⎬
2 C. MIP-Based Formulation of SE
minJt = + (1 − γit ) .
⎩ σ̄i j
2 σ̃it2 ⎭ The efficient MIP solver can tackle very large-scale power
i=1 j =1 i=1 system problems with thousands of continuous and inte-
(14) ger variables within a tolerable time. It also guarantees a
This binary variable is an indicator of the sudden change solution that is globally optimal or one that is within an
condition corresponding to the state variable i . One of the acceptable tolerance. MIP proposes very flexible and accu-
advance topics in SSE is the bad data identification and rate modeling capabilities which are extremely desirable in
rejection. In [2], it has been pointed out that when the absolute practical problems. MIP includes two types of problems:
value of the measurement residual is greater than the triple of 1) mixed-integer linear programming (MILP) problems and
the estimation standard deviation, there is a good chance that 2) mixed-integer quadratic programming (MIQP) problems.
the corresponding measurement is bad. Comparing this subject MILP models comprise both linear objective function and
with the prediction elimination reveals the same. Accordingly, constraints; while MIQP models consist of quadratic objective
the rule is employed to detect the sudden change condition. function but linear constraints. The latter is the case in this
The prediction residual is represented by r̃it as paper which is denoted as a MIP model hereafter. Substituting
  the residuals of measurement and prediction with r̄i j t and
r̃it = x̃ it − x̂ it  . (15) r̃it , respectively, and (1 − γit )(r̃it )2 with ( R̃it )2 , the objective
function (14) is rewritten as
The following formulation is proposed for γit : ⎧ ⎫
⎨ i r̄i2j t
NS NM 
NS
R̃it ⎬
2
r̃it − 3σ̂it r̃it − 3σ̂it
≤ γit ≤ +1 (16) Min Jt = + (19)
Mi Mi ⎩ σ̄i2j σ̃it2 ⎭
i=1 j =1 i=1
where Mi is a positive constant number, which is always
larger than the numerator. In (16), when r̃it is <3σ̂it , the which is a quadratic function. To linearize R̃it in terms of γit
lower limit of γit would be a small negative number and its and r̃it , the following inequities are presented [38]:
upper limit would be a positive number smaller than unity. rimin (1 − γit ) ≤ R̃it ≤ rimax (1 − γit ) (20)
γit is consequently set to 0 and the prediction is included in
r̃it − rimax γit ≤ R̃it ≤ r̃it − rimin γit . (21)
objective function. Otherwise, for r̃it ≥ 3σ̂it , the variable γit
is 1 and the prediction is excluded. If x̂ it represents a voltage magnitude, and are rimin rimax
In (16), σ̂it is the standard deviation associated with the selected as 0 and 2 (p.u.), respectively, and if x̂ it represents
estimation of variable i at time t; however, it is unknown a voltage phase angle, rimin and rimax are set at –π and
AMINIFAR et al.: POWER SYSTEM DSE WITH SYNCHRONIZED PHASOR MEASUREMENTS 357

π (rad), respectively. When γit is equal to 1, R̃it will vanish PMU locations could be, however, examined for enhancing
due to (20) and the bounds in (21) will be inactive. Otherwise, the performance of DSE.
when γit is equal to 0, (21) will enforce R̃it to be equal to r̃it Among performance indexes for quantifying the accuracy
and (20) will not bind R̃it . of state estimators, the root mean square deviation (RMSD)
Here, (22)–(25) represent a set of linear expressions for r̄i j t of estimated states is a commonly used indicator. Here, the
and r̃it in terms of x̂ it RMSD of states corresponding to period t would be averaged
over the entire time horizon
r̄i j t ≥ x̄ i j t − x̂ it (22) 
NT 
r̄i j t ≥ x̂ it − x̄ i j t 1  
(23) NS
RMSD =  1 (x̂ it − x it )2 . (27)
r̃it ≥ x̃ it − x̂ it (24) NT NS
t =1 i=1
r̃it ≥ x̂ it − x̃ it . (25)
The above index is computed separately for voltage mag-
Also, (18) is restated as a linear representation of σ̂it in nitudes and phase angles because these state variables have
terms of γit different scales. The estimator performance should be analyzed
γit (1 − γit ) statistically for enhancing the reliability and generality of the
σ̂it =  + . (26) obtained results. We consider 1000 executions with random
i
NM i
NM
measurement sets in which the performance indexes are aver-
1 1
+ 1
σ̄i2j σ̄i2j σ̃it2
j =1 j =1 aged over all executions. The variability range (or standard
deviation) of performance indexes is another index that would
Therefore, the SE problem is formulated in MIP. The
reflect the impact of measurement random errors on the overall
estimated states x̂ it and its corresponding standard deviation
performance of the estimator. The performance analysis is
σ̂it will be calculated accordingly.
usually carried out considering various network topologies,
An important point is to be emphasized with regards to
loading levels, and generation dispatch. This requirement is
the detection/rejection of bad measurement data in estimators.
satisfied in our analyzes since each execution has several time
WAMS is an accurate, reliable, viable, and secure mea-
instants with a variety of scenarios.
surement infrastructure. However, the inclusion of bad data
In the following studies, true system states are represented
detection/rejection in the proposed DSE, if intended, can be
by power flow solutions, measurements are simulated by
accommodated by adding a binary variable to the measurement
adding random noises within designated ranges to true values,
residual term in (14).
and prediction and estimated quantities are calculated based on
The other issue is that the proposed model is on the
the presentation in Section IV. Line parameter uncertainties
basis that the state variables are the same as measurement
are assumed to be typical as discussed in Section II, and
quantities, which is the case in WAMS. This assumption does
PMU measurement errors are adopted based on the maximum
not hold in conventional metering system in which a nonlinear
allowable TVE (1%) associated with steady-state and small
measurement function would correlate state variables with
change conditions [24]. Since the drastic changes considered
measurement quantities. Hence, the model proposed here
in this paper might correspond to dynamic conditions, the
is not applicable in a metering system composed of only
TVE associated with dynamic situations (3%) is also studied
conventional measurements. It should also be noted that due
as part of sensitivity analyzes. In practical cases, appropriate
to inherent differences of conventional measurements with
and realistic values could be employed which depend on
PMU data, which are being reported very fast and equipped
parameters specified by PMU manufacturers and experiences
with exact time stamps, adding conventional measurements
with periodical calibrations. The CPLEX solver in the general
to the WAMS is not a technically feasible choice. The
algebraic modeling system (GAMS) environment is employed
reverse relation in which some PMU data are, however, being
for the optimization purpose and the simulations were carried
added to a SCADA system does make sense and is reported
out using an Intel Core (TM) i7 at 1.60-GHz CPU with 4 GB
by several works. In such cases, the proposed DSE model
of RAM.
could be used. To do so, an old-fashioned SE with only
conventional measurements is first executed. Its output, which
includes voltage magnitude and phase angle, is then consid- A. Three-Bus System
ered as a set of phasor data for the proposed WAMS-based
DSE [9]. The three-bus power system is shown in Fig. 3 and the
corresponding data are given in the Appendix (Tables VI
and VII). The PMU located at bus 1 measures the voltage
V. N UMERICAL S TUDIES
phasor of bus 1 and current phasors of transmission lines
In this section, a three-bus power system is analyzed to 1 and 2. Accordingly, the voltage phasors at buses 2 and 3
illustrate the performance of the proposed method. In addition, would be measured indirectly. According to the power flow
the effectiveness of the proposed method is examined on the solution, unit 1 would generate 0.8 + j 0.25 (p.u. on the
IEEE 57-bus power system. Various scenarios are considered 100 MVA base) and the remaining load plus transmission
in each case for simulating sudden changes in the system state. system losses are generated by unit 2, which is located at
The PMU locations are assumed to be given here. Additional the slack bus.
358 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 63, NO. 2, FEBRUARY 2014

TABLE I
B INARY VARIABLES γit IN C ASE 0

1 1 2 is considered. γit associated with V2 is 0 as this state variable


G1 G2
is regulated by unit 2 and would inherently remain fixed
PMU
when volatile conditions are considered. γit associated with
2 3
V3 is 1 at time periods 9, 20, and 23, which correspond to
scenarios 2), 5), and 6), respectively. In addition, γV3 ,t is 1
at instants 10, 11, 21, and 24, which is due to the memory
3
of the prediction approach and large variations in the system
operating conditions at the indicated scenarios. According to
Fig. 3. Three-bus system. Table I, δ3 is more sensitive to sudden change conditions as
compared to other state variables, and its associated binary
variables are 1 corresponding to all scenarios except 1).
The reference bus angle is not defined because of direct Fig. 4 shows the DSE results in Case 0. The estimated
phase measurement of PMUs, and state variables are con- bus voltage errors are computed for both magnitudes and
sidered as { V1 , δ1 , V2 , δ2 , V3 , δ3 }. The challenging situation phase angles as the difference between estimated and true
for measuring the performance of DSE is a sudden change values in the base case of power flow solutions. In Fig. 4,
in system conditions. This condition is simulated via the measurement errors are presented for state variables which are
following scenarios. associated with PMU direct measurements at bus 1 and indi-
1) 20% load increase at time period 4. rect measurements at buses 2 and 3. Time instants with lower
2) Switch OFF generating unit 1 at time period 9. estimation error demonstrate the impact of data redundancy in
3) Switch ON generating unit 1 at time period 13. DSE. Time instants with identical estimation and measurement
4) 20% load decrease at time period 16. errors correspond to instants with prediction discarding, or
5) Switch out transmission line 3 at time period 20. γit = 1 in the mathematical sense. Fig. 4 shows that there is
6) Switch in transmission line 3 at time period 23. a similar error behavior for bus voltage magnitudes and phase
When unit 1 is on outage, the system load plus losses are fed angles in different buses. The reason is that only one PMU
by unit 2; thus, there would be no load curtailments. Note as exists in the three-bus network and state variables at buses
well that the observability of the power system is preserved 2 and 3 are dependent on the PMU voltage measurement at
when transmission line 3 is on outage. bus 1.
We consider the following three test cases. Here, the binary variables associated with switching OFF and
Case 0: Base case including the simultaneous discarding ON of generating unit 1 (or switching out and in of transmis-
capability. sion line 3) are not necessarily the same although the scenarios
Case 1: Exclude the discarding capability. have identical severities in terms of power systems. The reason
Case 2: Execute SSE with omitting the predictions. is that the scenarios are not identical from the DSEs viewpoint
These Cases are thoroughly discussed in the following. as measurement errors are randomly different in various time
Case 0: We have used the DSE formulation proposed in instants. This observation would be less likely as we increase
Section IV-C for 24 time periods. Although the performance the redundancy levels of measurement systems. The data set
of proposed DSE is generally drawn statistically, an execution would be prohibitively large if we present all state variables
of performance indexes around the average statistical values in the entire time horizon. Hence, Table II merely presents the
is adopted here for the sake of discussion. The binary vari- measured, predicted, estimated, and true values of V3 , and δ3
ables γit are shown in Table I, in which 1 and 0 denote to provide a set of specific numerical representation for the
the discarding and inclusion of predictions, respectively. Not DSE performance. These state variables would include 1 and
surprisingly, predictions associated with the state variables 0 values for γit which can clarify the impact of prediction
V1 and δ1 are discarded in most time instants. The reason inclusion and rejection. In Table II, as expected, estimations
is that V1 and δ1 have PMU direct measurements with higher are closer to the true state values as compared with the pure
accuracy as compared with indirect measurements at buses measurement data set. Since the measurement redundancy is
2 and 3 where the uncertainty of transmission line parameters zero in this case study, the analytical conclusions of the IEEE
AMINIFAR et al.: POWER SYSTEM DSE WITH SYNCHRONIZED PHASOR MEASUREMENTS 359

Bus 1 Bus 1
0.01 0.01
Estimation. Estimation.
Vol. Mag. Error [p.u.]

Vol. Ang. Error [rad.]


0.005 Measurment 0.005 Measurment

0 0

-0.005 -0.005

-0.01 -0.01
0 5 10 15 20 25 0 5 10 15 20 25
Time Instant Time Instant
Bus 2 Bus 2
0.01 0.01
Estimation. Estimation.
Vol. Mag. Error [p.u.]

Vol. Ang. Error [rad.]


Measurment 0.005 Measurment
0.005
0
0
-0.005
-0.005
-0.01

-0.01 -0.015
0 5 10 15 20 25 0 5 10 15 20 25
Time Instant Time Instant
Bus 3 Bus 3
0.015 0.02
Estimation. Estimation.
Vol. Mag. Error [p.u.]

Vol. Ang. Error [rad.]


0.01 Measurment Measurment
0.01
0.005
0
0
-0.01
-0.005

-0.01 -0.02
0 5 10 15 20 25 0 5 10 15 20 25
Time Instant Time Instant

Fig. 4. Error in state variable estimation and measurement.

TABLE II
C OMPARISON OF M EASUREMENT, P REDICTION , E STIMATION , AND T RUE VALUES

57-bus system are expected to be more general. The index referred to as SSE. There is, however, only one measurement
defined in (27) considered for assessing the performance of associated with each state variable and the SE leads to the
the proposed DSE is calculated as RMSDm = 0.0038 (p.u.) for measurement data set. Thus, the performance indexes of
voltage magnitudes and RMSD p = 0.0045 (rad) for voltage RMSDm = 0.0049 (p.u.) and RMSD p = 0.0057 (rad) are
phase angles. These indexes are calculated statistically with identical to the average values of RMSD of measurements.
1000 executions. The comparison of performance indexes in Cases 1 and 2
Case 1: In this case, the discarding capability is not consid- with those in Case 0 verifies the advantage of using DSE for
ered and all predictions are included in the estimation process. enhancing the estimation accuracy.
Hence, the objective function (12) is considered with both 1) Sensitivity Analysis: Here, Case 3 is considered to extract
measurement and prediction residuals, with a solution derived the sensitivity of DSE performance in preserving or discarding
already in (13). This attribute shows that an optimization the prediction terms with respect to measurement uncertainties.
would not be required as the solution is obtained with a In this case, PMU measurement uncertainties are multiplied
simple calculation. This simplicity is achieved at the cost of by three; thus, the confidence bounds of uncertain values are
not omitting the drastic changes. The performance indexes wider and equal to 3% which is the maximum allowable
associated with this case are RMSDm = 0.0046 (p.u.) and TVE of PMU measurements in dynamic conditions [24].
RMSD p = 0.0056 (rad). The measurements of V2 , δ2 , V3 , and δ3 are more volatile
Case 2: In this case, the state prediction is disregarded. By and it is expected that the role of prediction would become
definition, this situation corresponds to the conventional SE, more significant. Table III presents the binary variables in
360 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 63, NO. 2, FEBRUARY 2014

TABLE III
B INARY VARIABLES γit IN C ASE 3

which the predicted values associated with designated state G2


2
variables are discarded. The performance indexes are
17
RMSDm = 0.0160 (p.u.) and RMSD p = 0.0177 (rad). This
G3
research is repeated when the predictions are not considered, 5 3 G12

namely SSE. The analysis results in performance indexes as PMU 12


1 16
RMSDm = 0.0271 (p.u.) and RMSD p = 0.0306 (rad). In G1
G6
this case, the relative improvements due to the inclusion of 4
PMU 6 PMU
predictions in the estimation process are 40.5% and 42.1%
18 15
for voltage magnitude and phase angle, respectively. The G8 PMU 14
same comparison for the studies with 1% TVE in Cases 0 8
19 45
and 2 leads to 22.4% and 21.0% improvements for voltage PMU 46
magnitude and phase angle, respectively. Hence, predictions 7 20
PMU 47
will have a higher positive impact on the estimation accuracy 29
21 44 13
as we increase the measurement uncertainty.
49
28 48
B. IEEE 57-Bus System 22 PMU
38 50 10
PMU PMU 51
The network topology is shown in Fig. 5 [39]. We dis-
23
regard the zero-injection effect and use the PMU placement 27 PMU
37
algorithm proposed in [25] which considers 17 PMUs at 26 39 57
buses {1, 4, 6, 9, 15, 20, 24, 28, 30, 32, 36, 38, 41, 46, 50, 53, 57}
52 24 PMU
to make the system observable. The following scenarios would
PMU 36 40 56
simulate conditions with sudden changes. 53 PMU 25
1) Load increase at buses 31–33 at time period 2. 42
35
2) Switch OFF generating unit 3 at time period 5. 54 PMU 30 PMU
41
3) Switch out transmission line 18–19 at time period 9. 34 43
55 PMU
We simulate 12 time periods with an active simultaneous 31 32 33
G9 11
discarding capability. Similar to the three bus example, buses 9
PMU
with PMU experience more accurate measurements with their
binary variables equal to 1 at most time periods. Hence, we
hereafter focus on the binary variables of non-PMU buses. Fig. 5. One-line diagram of IEEE 57-bus system.
Among these variables, those which are equal to 1 are pre-
sented in Table IV. At time period 2, the loads at buses
31 to 33 are increased suddenly. The simulation results reveal The observation is justified as the closely located generating
that γVb ,t associated with buses 25, 31, and 33 are equal to 1 at units would prevent a significant voltage drop during the
time periods 2 and 3. Similarly, γδb ,t associated with buses 31, outage of generating unit 3. On the contrary, line flows and bus
and 33 are equal to 1 at the same time periods. As expected, phase angles experience relatively large variations. Table IV
the buses experiencing sudden changes are geographically shows that several binary variables associated with bus voltage
located in the vicinity of the one with the initial sudden angles would become 1 at time periods 5 and 6.
change. We switch out the transmission line 18–19 at time period
At time period 5, the generating unit 3 is switched OFF. The 9. At time period 8, the active power flow from bus 18 to bus
generation shortage is compensated for by unit 1 which shifts 19 is only 2 MW. Hence, the line outage would not impose a
the state variables to a new operating point. The transition sever condition. Table IV shows that γVb ,t and γδb ,t associated
depends on the unit 5 dispatch which is 40 MW in this case. with bus 19 is equal to 1 only at the instant of change
This event is not identified as a sudden change according to occurrence. Hence, the impact of outage of line 18–19 is very
the binary variables associated with bus voltage magnitudes. limited.
AMINIFAR et al.: POWER SYSTEM DSE WITH SYNCHRONIZED PHASOR MEASUREMENTS 361

TABLE IV the line parameters. The proposed MIP-based DSE formu-


B INARY VARIABLES γit FOR THE IEEE 57-B US P OWER S YSTEM lation is equipped with simultaneous discarding capability
of predictions. This feature alleviates the need for multiple
executions of DSE for averting the performance degrada-
tion when predictions are included at periods with sudden
changes. Two test systems were thoroughly examined and the
effectiveness of the proposed model was verified by consid-
ering various conditions for loading and topology changes.
TABLE V As illustrated in this paper, the proposed formulation works
P ERFORMANCE I NDEXES AND E XECUTION T IMES effectively during both normal and sudden change conditions.
FOR THE I EEE 57-B US S YSTEM At normal conditions, the prediction promotes the estimation
accuracy and enhances the data redundancy. However, upon
the occurrence of a major power system event, the predicted
data set are discarded to preserve the estimation quality. The
buses equipped with PMUs have more exact measurements
in which the prediction does not play an important role. In
contrast, buses without PMUs exploit the predicted data set
throughout the normal condition and even during some small
disturbances.
We examine three case studies (Cases 0–2) of the three-bus
system on the 57-bus test system. Similarly, 1000 simulations A PPENDIX
were carried out and the indexes were averaged. Table V shows
A. Derivatives of GUM Application
the index (27) separately calculated for voltage magnitudes and
phase angles. Here, the comparison of indexes with those of The partial derivatives required for (3) and (4) are extracted
the three-bus system reveals that the utilization of prediction in this section. Converting Vr  δr into the rectangular coordi-
elimination capability would result in smaller improvements nation, we get
in system-wide indexes for larger systems. This observation is Vr  δr = VrR + j VrI (A.1)
accurate since contingencies in larger systems would have lim- 
ited geographically impacts with a smaller impact on the entire where Vr = (VrR )2 + (VrI )2 and δr = tan−1 (VrI /VrR ).
set of system state variables. Regarding the same RMSD p Referring to (1), we can formulate VrR and VrI as
in Cases 1 and 2, it is worth noting that the performance follows:
indexes are averaged over the entire time periods. For the
time instants without drastic changes, the estimation errors in VrR = Vs cos(ϕ1 ) − Is Z cos(ϕ2 ) + Vs Y Z cos(ϕ3 ) (A.2)
Case 1 are smaller. However, because the discarding capability VrI = Vs sin(ϕ1 ) − Is Z sin(ϕ2 ) + Vs Y Z sin(ϕ3 ). (A.3)
is excluded, the predictions impose greater estimation errors
Note that ϕ1 , ϕ2 , and ϕ3 are defined after (1).
for time periods associated with sudden changes. Furthermore,
Partial derivatives of Vr and δr are expressed in terms of
the predictions impose greater estimation errors for time
partial derivatives of VrR and VrI as follows:
periods associated with sudden changes because we exclude
the discarding capability. This observation is critical as it ∂ VrR ∂ VrI
∂ Vr VrR ∂p + VrI ∂p
reflects the necessity of accommodating discarding capability = , ∀p ∈ P (A.4)
in the DSE. ∂p Vr
∂ VrR ∂ VrI
Table V additionally presents a comparison of computation ∂δr −VrI ∂p + VrR ∂p
speed. As expected, Case 0 takes more time for optimization; = , ∀p ∈ P (A.5)
∂p (Vr )2
however, the process would likely be faster than that of Case 1
when outlier predictions are eliminated iteratively. The other and partial derivatives of VrR and VrI are obtained as follows:
important point is that the execution time of Case 1 is not twice ∂ Vr
that of Case 2. This observation is sensible as the number of = [(VrR cos ϕ1 + VrI sin ϕ1 ) (A.6)
∂ Vs
input data in Case 1 is not twice. In Case 2, 17 PMUs assigned
+ (VrR Y Z cos ϕ3 + VrI Y Z sin ϕ3 )]/Vr
to the IEEE 57-bus system offer many redundant indirect
measurements; while, in Case 1, only 2 × 57 predictions are ∂ Vr
= −Vs [(VrR sin ϕ1 − VrI cos ϕ1 ) (A.7)
added to the input data set. ∂δs
+ (VrR Y Z sin ϕ3 − VrI Y Z cos ϕ3 )]/Vr
VI. C ONCLUSION ∂ Vr
= −Z (VrR cos ϕ2 + VrI sin ϕ2 )/Vr (A.8)
The solution of DSE problem with the PMU measurements ∂ Is
was addressed in this paper. The confidence level of measure- ∂ Vr
= Is Z (VrR sin ϕ2 − VrI cos ϕ2 )/Vr (A.9)
ments associated with buses without PMU was computed by ∂δ I s
propagating the uncertainties of original measurements over ∂ Vr
= −[(VrR Is cos ϕ2 + VrI Is sin ϕ2 ) (A.10)
the transmission lines and considering a variation bound for ∂Z
362 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 63, NO. 2, FEBRUARY 2014

B. Three-Bus System Data


− (VrR Vs Y cos ϕ3 + VrI Vs Y sin ϕ3 )]/Vr
∂ Vr TABLE VI
= Z [(VrR Is sin ϕ2 − VrI Is cos ϕ2 ) (A.11)
∂θz T RANSMISSION L INE D ATA
− (VrR Vs Y sin ϕ3 − VrI Vs Y cos ϕ3 )/Vr
∂ Vr
= Vs Z (VrR cos ϕ3 + VrI sin ϕ3 )/Vr (A.12)
∂Y
∂ Vr
= −Vs Y Z (VrR sin ϕ3 − VrI cos ϕ3 )/Vr (A.13)
∂θ y
∂δr TABLE VII
= [(VrR sin ϕ1 − VrI cos ϕ1 ) (A.14)
∂ Vs P OWER S YSTEM L OAD IN 24 T IME P ERIODS
+ (VrR Y Z sin ϕ3 − VrI Y Z cos ϕ3 )]/(Vr )2
∂δr
= Vs [(VrR cos ϕ1 + VrI sin ϕ1 ) (A.15)
∂δs
+ (VrR Y Z cos ϕ3 + VrI Y Z sin ϕ3 )]/(Vr )2
∂δr
= −Z (VrR sin ϕ2 − VrI cos ϕ2 )/(Vr )2 (A.16)
∂ Is
∂δr
= −Is Z (VrR cos ϕ2 + VrI sin ϕ2 )/(Vr )2 (A.17)
∂δ I s
∂δr
= −[(VrR Is sin ϕ2 − VrI Is cos ϕ2 ) (A.18)
∂Z
− (VrR Vs Y sin ϕ3 − VrI Vs Y cos ϕ3 )]/(Vr )2 R EFERENCES
∂δr
= −Z [(VrR Is cos ϕ2 + VrI Is sin ϕ2 ) (A.19) [1] M. Shahidehpour, W. F. Tinney, and Y. Fu, “Impact of security on
∂θz power systems operation,” Proc. IEEE, vol. 93, no. 11, pp. 2013–2025,
− (VrR Vs Y cos ϕ3 + VrI Vs Y sin ϕ3 )]/(Vr )2 Nov. 2005.
[2] A. J. Wood and B. F. Wollenberg, Power Generation, Operation, and
∂δr
= Vs Z (VrR sin ϕ3 − VrI cos ϕ3 )/(Vr )2 (A.20) Control, 2nd ed. New York, NY, USA: Wiley, 1996.
∂Y [3] A. M. Leite da Silva, M. B. Do Coutto Filho, and J. M. C. Cantera,
∂δr “An efficient dynamic state estimation algorithm including bad data
= Vs Y Z (VrR cos ϕ3 + VrI sin ϕ3 )/(Vr )2 . (A.21) processing,” IEEE Trans. Power Syst., vol. 2, no. 4, pp. 1050–1058,
∂θ y Nov. 1987.
[4] M. B. Do Coutto Filho and J. C. Stacchini de Souza, “Forecasting-aided
The other parameters of (1) that should be calculated are state estimation—Part I: Panorama,” IEEE Trans. Power Syst., vol. 24,
σ Z , σθz , σ (Y ), and σ (θ y ). Referring to Fig. 1, the polar no. 4, pp. 1667–1677, Nov. 2009.
[5] M. B. Do Coutto Filho, J. C. S. de Souza, and R. S. Fre-
representation of the line impedance and admittance are as und, “Forecasting-aided state estimation—Part II: Implementation,”
follows: IEEE Trans. Power Syst., vol. 24, no. 4, pp. 1678–1685, Nov. 2009.
[6] N. R. Shivakumar and A. Jain, “A review of power system dynamic state
  estimation techniques,” in Proc. IEEE Int. Conf. Power Syst. Technol.,
Z = R 2 + X 2 , θz = tan−1 (X R) (A.22) Oct. 2008, pp. 1–6.
 
−1 [7] A. Jain and N. R. Shivakumar, “Power system tracking and dynamic
Y = G 2 + B 2 , θ y = tan (B G). (A.23) state estimation,” in Proc. Power Syst. Conf. Expos., Mar. 2009,
pp. 1–8.
Standard deviations of impedance and admittance amplitude [8] A. G. Phadke and R. M. de Moraes, “The wide world of wide-area
measurement,” IEEE Power Energy, vol. 6, no. 5, pp. 52–65, Oct. 2008.
and phase angle are function of line parameter variances, as [9] G. Phadke and J. S. Thorp, Synchronized Phasor Measurements
given below and Their Applications. New York, NY, USA: Springer-Verlag,
2008.

  2   2 [10] A. Jain and N. R. Shivakumar, “Impact of PMUs in dynamic state
σZ = ∂ Z ∂ R [σ R ]2 + ∂ Z ∂ X [σ X ]2 (A.24) estimation of power systems,” in Proc. 40th NAPS, Sep. 2008, pp. 1–8.

  2   2 [11] H. Xua, Q. Jia, N. Wang, Z. Bo, H. Wang, and H. Ma, “A dynamic
σθz = ∂θz ∂ R [σ R ]2 + ∂θz ∂ X [σ X ]2 (A.25) state estimation method with PMU and SCADA measurement for power
systems,” in Proc. IPEC, Dec. 2007, pp. 848–853.
[12] N. R. Shivakumar and A. Jain, “Including phasor measurements in
where ∂ Z /∂ R = R/Z ,∂ Z /∂ X = X/Z ,∂θz /∂ R = −X/Z 2 , dynamic state estimation of power systems,” in Proc. Int. Conf. Power
and ∂θz /∂ X = R/Z 2 . Syst. ACO, Mar. 2008, pp. 1–6.
[13] E. Farantatos, G. K. Stefopoulos, G. J. Cokkinides, and
And A. P. Meliopoulos, “PMU-based dynamic state estimation for electric
 power systems,” in Proc. Power Energy Soc. General Meeting, Jul. 2009,
  2   2 pp. 1–8.
σY = ∂Y ∂ G [σG ]2 + ∂Y ∂ B [σ B ]2 (A.26) [14] A. Jain and N. R. Shivakumar, “Phasor measurements in dynamic

  2   2 state estimation of power systems,” in Proc. IEEE Region 10th Conf.
σθ y = ∂θ y ∂ G [σG ]2 + ∂θ y ∂ B [σ B ]2 (A.27) TENCON, Nov. 2008, pp. 1–6.
[15] Z. Huang, K. Schneider, and J. Nieplocha, “Feasibility studies
of applying Kalman Filter techniques to power system dynamic
where ∂Y /∂ G = G/Y, ∂Y /∂ B = B/Y, ∂θ y /∂ G = −B/Y 2 , state estimation,” in Proc. Int. Power Eng. Conf., Dec. 2007,
and ∂θ y /∂ B = G/Y 2 . pp. 376–382.
AMINIFAR et al.: POWER SYSTEM DSE WITH SYNCHRONIZED PHASOR MEASUREMENTS 363

[16] EPRI. (2008). Next Generation State Estimation (Predictive, Dynamic [37] J. Zhu and A. Abur, “Effect of phasor measurements on the choice of
With System Identification), Charlotte, NC, USA [Online]. Available: references bus for state estimation,” in Proc. IEEE Power Eng. Soc.
http://portfolio.epri.com/ General Meeting, Jun. 2007, pp. 1–5.
[17] S. Chakrabarti and E. Kyriakides, “PMU measurement uncertainty [38] C. A. Floudas, Nonlinear and Mixed-Integer Optimization: Fundamen-
considerations in WLS state estimation,” IEEE Trans. Power Syst., tals and Applications. New York, NY, USA: Oxford Univ. Press, 1995.
vol. 24, no. 2, pp. 1062–1071, May 2009. [39] (2013). MATPOWER: A MATLAB Power System Simulation Package
[18] A. Abur and A. G. Expósito, Power System State Estimation: Theory [Online]. Available: http://www.pserc.cornell.edu/matpower
and Implementation. New York, NY, USA: Marcel Dekker, 2004.
[19] (2013). High-Performance Mathematical Programming Solver for Lin-
ear Programming, Mixed Integer Programming, and Quadratic Pro-
gramming [Online]. Available: http://www.ilog.com/products/cplex/ Farrokh Aminifar (S’07–M’11) received the B.Sc. (Hons.) degree from the
[20] S. Chakrabarti, E. Kyriakides, and M. Albu, “Uncertainty in Iran University of Science and Technology, Tehran, Iran, in 2005, and the
power system state variables obtained through synchronized measure- M.Sc. (Hons.) and Ph.D. degrees from the Sharif University of Technology,
ments,” IEEE Trans. Instr. Meas., vol. 58, no. 8, pp. 2452–2458, Tehran, in 2007 and 2010, respectively, all in electrical engineering.
Aug. 2009. He has been collaborating with the Robert W. Galvin Center for Electricity
[21] J. K. Al-Othman and M. R. Irving, “Analysis of confidence bounds Innovation with the Illinois Institute of Technology, Chicago, IL, USA,
in power system state estimation with uncertainty in both measurements since March 2009. He is currently an Assistant Professor with the School
and parameters,” Elec. Power Syst. Res., vol. 76, no. 12, pp. 1011–1018, of Electrical and Computer Engineering, the University of Tehran, Tehran.
Aug. 2006. His research interests include wide-area measurement systems, reliability
[22] J. Zhu, A. Abur, M. J. Rice, G. T. Heydt, and S. Meliopoulos, “Enhanced modeling and assessment, and smart grid technologies.
state estimators,” PSERC, Tempe, AZ, USA, Tech. Rep. 06-45, Dr. Aminifar has served as the Guest Editor of two Special Issues (Micro-
Nov. 2006. grids and Smart DC Distribution Systems) of the IEEE T RANSACTIONS ON
[23] Performance Requirements Part II Targeted Applications: State Estima- S MART G RID. He received the IEEE Best Ph.D. Dissertation Award from the
tion, PRTT, New York, NY, USA, Nov. 2005. Iran Section for his research on the probabilistic schemes for the placement
[24] IEEE Standard for Synchrophasor Measurements for Power Systems, of pharos measurement units.
IEEE Standard C37.118.1-2011, Dec. 2011.
[25] F. Aminifar, A. Khodaei, M. Fotuhi-Firuzabad, and M. Shahideh-
pour, “Contingency-constrained PMU placement in power net-
works,” IEEE Trans. Power Syst., vol. 25, no. 1, pp. 516–523,
Feb. 2010. Mohammad Shahidehpour (F’01) is the Bodine Chair Professor and Director
[26] F. Aminifar, M. Fotuhi-Firuzabad, and A. Safdarian, “Optimal PMU of the Robert W. Galvin Center for Electricity Innovation with the Illinois
placement based on probabilistic cost/benefit analysis,” IEEE Trans. Institute of Technology, Chicago, IL, USA. He is a Research Professor with
Power Syst., vol. 28, no. 1, pp. 566–567, Feb. 2013. King Abdulaziz University, Jeddah, Saudi Arabia, North China Electric Power
[27] F. Aminifar, M. Fotuhi-Firuzabad, M. Shahidehpour, and A. Saf- University, Beijing, China, and the Sharif University, Tehran, Iran. He was
darian, “Impact of WAMS malfunction on power system reliability the general chair of the 2012 IEEE Innovative Smart Grid Technologies
Conference, and Chair of the 2012 Great Lakes Symposium on Smart Grid
assessment,” IEEE Trans. Smart Grid, vol. 3, no. 3, pp. 1302–1309,
Sep. 2012. and the New Energy Economy. He is the Editor-in-Chief of the IEEE
[28] F. Aminifar, M. Fotuhi-Firuzabad, M. Shahidehpour, and A. Khodaei, T RANSACTIONS ON S MART G RID and an IEEE Distinguished Lecturer.
“Observability enhancement by optimal PMU placement considering He is a recipient of the 2012 IEEE PES Outstanding Power Engineering
Educator Award. He was a recipient of the 2009 Honorary Doctorate from
random power system outages,” Energy Syst., vol. 2, no. 1, pp. 45–65,
Mar. 2011. the Polytechnic University of Bucharest, Romania.
[29] M. B. D. C. Filho, A. M. L. da Silva, J. M. C. Cantera, and
R. A. da Silva, “Information debugging for real-time power systems
monitoring,” IEE Proc. Generat. Trans. Distribut., vol. 136, no. 3,
pp. 145–152, May 1989. Mahmud Fotuhi-Firuzabad (SM’99) received the B.S. and M.S. degrees in
[30] K. Nishiya, J. Hasegawa, and T. Koike, “Dynamic state estimation electrical engineering from Sharif University of Technology, Tehran, Iran, and
including anomaly detection and identification for power systems,” Tehran University, Tehran, in 1986 and 1989, respectively, and the M.S. and
IEE Proc. Generat. Trans. Distribut., vol. 129, no. 5, pp. 192–198, Ph.D. degrees in electrical engineering from the University of Saskatchewan,
Sep. 1982. Saskatoon, SK, Canada, in 1993 and 1997, respectively.
[31] A. M. Leite da Silva, M. B. Do Coutto Filho, and J. F. He joined the Department of Electrical Engineering, the Sharif University
de Queiroz, “State forecasting in electric power systems,” IEE of Technology, in 2002. Currently, he is a Professor and the Head of the
Proc. Generat. Trans. Distribut., vol. 130, no. 5, pp. 237–244, Department of Electrical Engineering, Sharif University of Technology. He
Sep. 1983. is an Honorary Professor with the Universiti Teknologi Mara, Shah Alam,
[32] J. K. Mandal, A. K. Sinha, and L. Roy, “Incorporating nonlinearities Malaysia.
of measurement function in power system dynamic state estimation,” Dr. Fotuhi-Firuzabad is a member of the Center of Excellence in Power
IEE Proc. Generat Trans. Distribut., vol. 142, no. 3, pp. 289–296, System Management and Control. He serves as an Editor of the IEEE
May 1995. T RANSACTIONS ON S MART G RID.
[33] K. R. Shih and S. J. Huang, “Application of a robust algorithm for
dynamic state estimation of a power system,” IEEE Trans. Power Syst.,
vol. 17, no. 1, pp. 141–147, Feb. 2002.
[34] S. J. Huang and K. R. Shih, “Dynamic-state-estimation scheme including Saeed Kamalinia (S’07–M’11) received the B.S. and M.S. degrees in
nonlinear measurement-function considerations,” IEE Proc. Generat. electrical engineering from the University of Tehran, Tehran, Iran, in 2005
Trans. Distribut., vol. 149, no. 6, pp. 673–678, Nov. 2002. and 2007, respectively, and the Ph.D. degree in electrical engineering from
[35] J. M. Lin, S. J. Huang, and K. R. Shih, “Application of sliding the Illinois Institute of Technology, Chicago, IL, USA, in 2010. He is currently
surface-enhanced control for dynamic state estimation of a power a Project Engineer with S&C Electric Company-Power System Solutions,
system,” IEEE Trans. Power Syst., vol. 18, no. 2, pp. 570–577, Chicago. His research interests include operation and control of power systems
May 2003. and integration of renewable energy resources.
[36] S. G. Makridakis, S. C. Wheelwright, and V. E. McGee, Forecasting: Dr. Kamalinia is the Chair of Award Committee at IEEE-Chicago Section
Methods and Applications. New York, NY, USA: Wiley, 1983. and a member of two technical working groups in Cigré.

Вам также может понравиться