1

SMART GRID ANALYTICS:

SELECTED APPLICATIONS

Saikat Chakrabarti
Professor
Department of Electrical Engineering
Indian Institute of Technology Kanpur
UP 208016
Email: [email protected]
 2
Some important smart grid analytics
 Real-time visualization of power systems
 Improved state estimation
 Online estimation of load models
 Online monitoring of power system stability
 Analysis of the causes of a total or partial blackout
 Real-time congestion management
 Design of an adaptive protection system
 Detection of faults on transmission/distribution lines
Improved state estimation 3
Ensuring system observability (pre-requisite for state
estimation):
 Optimal placement for PMU-only measurement system
 Increasing measurement redundancy and avoiding critical
measurements
 Coordinating measurements in a multi-
area system
 Making an unobservable system
observable (while replacing old
conventional measurements)
 To enhance the performance of the
existing SCADA-based state estimator
Improved state estimation…contd. 4

Integration of PMUs into the state estimator (SE):

 The design and deployment of the hybrid SE is still in the
experimental stage
 Following are some of the benefits that can be derived
as a result of including PMU measurements in the SE:
 Improved accuracy of the SE
 Better observability
 Enhanced convergence characteristics
 Better handling of bad data
Hybrid state estimator 5
Two possible approaches to the inclusion of PMU
measurements in the SE:
 Post-processing:
 More practical at the moment, since PMUs are being installed
only in incremental fashion.
 Power companies are usually reluctant to go for expensive
modification of their existing state estimation software.
 Pre-processing:
 As the number of PMUs increases in power systems, it will be
wise to adopt the SE which can handle both conventional and
PMU measurements at the same time.
 Existing SEs at the control centres need to be replaced with this
type of hybrid SE (HSE).
Post-processing HSE 6

With limited number of PMUs, an obvious choice is to mix the

PMU measurements with the outputs of the existing SE, and
improve the estimate of the states.
Let 𝐳𝑃𝑀𝑈 be the vector containing the measurements obtained
from the PMUs.
𝑇
𝐳𝑃𝑀𝑈 = 𝑉෨𝑖 , 𝑉෨𝑗 , … , 𝐼ሚ𝑝𝑞 , 𝐼ሚ𝑟𝑠 , …

where 𝑉෨𝑖 = 𝑉𝑖 ∠𝜃𝑖 is the voltage phasor at bus i (where a PMU is

placed); and 𝐼ሚ𝑝𝑞 = 𝐼𝑝𝑞 ∠𝜃𝑝𝑞 is the current phasor between buses
p and q, as measured by the PMU placed at bus p.
Post-processing HSE…contd. 7

The output, 𝐱𝑐𝑜𝑛𝑣 of the conventional power system SE is the

minimal set of states, typically, the voltage magnitudes
and phase angles at all buses in the system, as shown
below for an N-bus system.
𝐱 = 𝜃1 , 𝜃2 , … , 𝜃𝑁 , 𝑉1 , 𝑉2 , … , 𝑉𝑁 𝑇
The new measurement vector for the HSE is,
𝑇
𝐳𝑛𝑒𝑤 = 𝐱𝑐𝑜𝑛𝑣 𝑇
𝐳𝑐𝑜𝑛𝑣 𝑇

The state estimation is carried out on the new measurement

vector 𝐳𝑛𝑒𝑤 to determine the new state vector 𝐱𝑛𝑒𝑤 for the
system.
Pre-processing HSE 8
 As the number of installed PMUs will increase in power
systems, a state estimator capable of handling both
conventional and synchronized measurements will become
the obvious choice.
 This will eliminate the additional computational burden and
reduction in estimation accuracy due to multiple processing
in the post-processing type estimator.
 In the pre-processing type HSE, the conventional and PMU
measurements will be processed together to obtain the final
estimates of the states.
 This will require modification or replacement of the existing
state estimation software in the EMS
Pre-processing HSE…contd. 9
 Inclusion of voltage phasor measurements by the PMUs in the
existing SE is straightforward.
 There are three possible ways in which the current measurements
by the PMUs can be directly incorporated in an SE:
1. Current phasor magnitude and phase angle (tend to pose
numerical ill-conditioning problem)
2. Real and imaginary part of the complex current measurement
(found most suitable, in terms of accuracy and convergence
characteristics)
3. Pseudo-voltage measurement with the help of current phasor
measurement and known line parameters (almost comparable
with the 2nd method above, slightly less accurate).
Challenges in HSE 10
 Refresh rate and accuracy of the conventional and PMU
measurements are widely different. Combining these two
measurements in practice is not an easy task.
 The accuracy of the PMU measurements are usually much higher
compared to the conventional measurements. This causes large
difference in the magnitude of the weights in a WLS state
estimator, and hence convergence problems may arise.
 Some of the steps taken to improve the numerical property of the
estimator are constrained formulation of the SE problem, optimal
ordering of the rows of the gain-matrix, robust numerical
techniques, etc.
 Further research is needed to arrive at a complete solution to this
problem
Dynamic estimator of state 11

 Static state estimator (SSE), for which WLS is the most popular
algorithm, estimates the state based on a snapshot of the
available measurements.
 Dynamic estimator of state (DES) utilizes past behavior of the
states along with the incoming new measurements to
estimate the new state.
 One attractive feature of the DES is its ability to predict the
states at future instants. This helps in taking preventive and
corrective control actions in the system.
 Different variations of Kalman filters are used to implement
DES algorithms
Dynamic estimator of state…contd. 12
 There are mainly two challenges to the implementation of DES at
the control centers:
 modelling the system dynamics
 computational complexity
 Utilizing the high refresh rate and high accuracy of the PMUs, the
system dynamics described by the so-called state transition
process can be better captured.
 Computational complexity of the DES is easily handled by the
state-of-the-art computing facilities.
Above two factors, in addition to potential usage in improving the SE
convergence, bad data processing, and topology processing, have
renewed the interest in DES
Few comments on DES 13
 The performance of DES can be further improved by using some
smoothing technique (e.g. Rauch-Tung-Striebel smoother), where
the estimated states at an earlier instant is re-estimated after the
arrival of new measurements.
 Variations of Kalman filters: unscented Kalman filter (UKF) (does not
need linearization of the system), cubature Kalman filter (CKF)
(better accuracy and computational efficiency).
 An efficiently designed DES can be executed within few
milliseconds even for a large power system. This may enable
estimation of system states at each arrival of complete PMU
measurement set at the control centre.
 DES has been used to estimate dynamic states (rotor angle,
speed, etc.) also for power systems.
Distribution system state estimator (DSSE) 14

 Deployment of SE at the distribution level needed for:

 handling massive penetration DGs
 managing demand response
 manage EVs and storage
 Challenges in DSSE:
 Unbalanced system – requiring 3-phase formulation
 High R/X ratio
 Large number of nodes
 Incomplete observability – possibility of utilizing smart meter
data along with SCADA measurements
PMU applications in stability monitoring 15

Voltage stability (VS) assessment using PMU

measurements
 Local measurement-based
 Uses local measurements of a PMU installed at a bus
 Synchronized measurements are not needed
 Algorithm may be implemented locally
 Wide-area measurement-based
 WAMS-based assessment of VS
 Better system-wide picture of VS
 Can be implemented in the control centre
VS monitoring with local measurements 16
Basic concept:
 Based on local measurements of voltage and current
phasors, transmission capacity of the network is determined.
 The closer the load demand is to the estimated capacity, the
more imminent is the voltage instability.
 For radial type of system, the method is equivalent to finding
the closeness of the load impedance to the conjugate of the
source impedance, using Thevenin equivalents, and the
principle of maximum power transfer.

K. Vu et.al., “Use of Local Measurements to Estimate Voltage-Stability

Margin”, IEEE trans. Power syst., Vol. 14., No. 3, Aug. 1999.
VS with local measurements…contd. 17
Tracing the Thevenin equivalent impedance:

Measurements from 2 or
more instants are needed
to solve for the Thevenin
equivalent impedance
Rotor angle estimation 18
 For the assessment of the transient stability of a power system,
knowledge of the rotor angle of the generators is essential.
 In the steady state, the rotor
angle of a synchronous generator
can be directly obtained using its
terminal voltage and current
measurements.
 During transient conditions, some of
the reactances of the machine
change their effective values.
 Under these conditions, estimation of
the rotor angle based on the terminal
measurements is not straight-forward.
Rotor angle estimation…contd. 19
 Based on the measurements
from a PMU, an observer for
the damper winding
currents is developed
 Using the observed damper
currents, known or estimated
parameters, and available
measurements, machine
reactances are found out
 Using these reactances and
detailed machine model,
rotor angle of the generator
is then estimated
Small-signal stability assessment 20
 Conventional method linearizes the system and finds the
Eigenvalues or modes of oscillation
 Computationally demanding task for a large system
 Needs detailed model of the system
 Model-free estimation of the modes is possible using
time-window of oscillation data (e.g. real power flow
through a line) captured by PMU
 Methods such as Prony, ESPRIT, matrix pencil can be used
to estimate the modes based on the captured data
 This process can be implemented for real-time
applications
Online mode estimation and PSS tuning 21

 A PMU at the terminal of

the generator captures
the real power output
over a time-window
 An event detection
algorithm triggers the
mode identification
process
 Estimated modes are
used to tune the PSS
connected to the
Part of the PSS tuning project with
generator NTPC NETRA for Dadri power plant
Under-frequency load-shedding 22
 Imbalance between the generation and the demand is the
major factor that causes frequency instability in a power
system.
 This power imbalance or the so-called disturbance power can
be estimated based on the rate of change of frequency
information.
 One of the most effective means of restoring the frequency
stability under severe power imbalance is to shed part of the
loads.
 How much load should be shed thereof and where, i.e, how
the total amount of load shedding is to be distributed, is an
important consideration.
Load-shedding…contd. 23
The disturbance power, Δ𝑃, can be estimated by using the
rate of change of frequency of the generators.

2 𝑑𝑓𝑖
Δ𝑃 = ෍ 𝐻𝑖
𝑓𝑛 𝑑𝑡
𝑖∈𝐺𝑛

where 𝑓𝑛 is the nominal frequency of the system, 𝐺𝑛 is the

set of generators. The rate of change of frequency, i.e.,
𝑑𝑓𝑖 /𝑑𝑡, is obtained from PMUs installed at the generator
buses
Load modelling 24

Challenges in modelling the load

 A load bus is typically connected to a large number of
different types of loads. It is practically impossible to
represent all these loads individually.
 Location of the devices are not always known to the
utilities.
 Lack of correct information regarding the composition of
the load.
 Uncertainties regarding the characteristics of loads,
especially at large deviations of voltage and frequency
 Aggregate load 25

Source: IEEE Task Force, “Load representation for dynamic performance

analysis,” IEEE Trans. on Power Systems, Vol. 8, No. 2, May 1993, pp. 472-482.
Current practice on load modelling 26
 About 70 percent of utilities use constant power model for
steady state analysis and ZIP model for dynamic studies.
 Only 30 percent utilities use some kind of detailed
dynamic models.
 For each load class, same kind of models are being used
and most of the utilities update the models once in five
years.

Source: Milanovic, J.V. et.al, “International Industry Practice on

Power System Load Modeling”, IEEE Tran. on Power Systems, vol.
28, No. 3, Aug. 2013, pp 3038-3046
Steps in the load modelling process 27
 Voltage and current measurements are obtained from
measurement devices (e.g. PMU or DFR)
 Load power is computed based on the above measurements
 Disturbance classification.
 Selection of load structure.
 Some background knowledge is needed regarding the
structure of the aggregate load
 This structure may be dependent on time of day, month of
year, season, etc.
 Parameter estimation: to estimate the most likely parameters
of the assumed load structure
Commonly used load models 28

Nonlinear dynamic model with recovery:

𝑑𝑃𝑑 𝑑𝑉
𝑇𝑝 + 𝑃𝑑 = 𝑃𝑠 𝑉 + 𝐾𝑝 (𝑉)
𝑑𝑡 𝑑𝑡

where 𝑇𝑝 is the time constant of

recovery to some specified percentage Typical load response with
of the original load; 𝑃𝑑 is the load voltage step change (Source:
demand; 𝑃𝑠 𝑉 and 𝐾𝑝 (𝑉) are called D. J. Hill, “Nonlinear dynamic
load models with recovery for
static load function and dynamic load voltage stability studies,” IEEE
function, respectively. Trans. Power Syst., vol. 8, no.
1, pp. 166-176, Feb. 1993).
Commonly used load models…contd. 29

Composite load model (induction motor + ZIP load)

The induction motor is usually
represented by third order model,
neglecting stator dynamics.

The algebraic equations for stator currents are given by,

Load modelling under large and small disturbances 30

 Measurement based load modelling is usually carried out with

the data recorded following a disturbance.
 Large disturbances are less frequent in power systems.
 Load model parameters change frequently due to the
changes in nature and composition of the load.
 Model parameters identified from the most recent large
disturbance may change before the next large disturbance
occurs.
 Modelling loads under small disturbances is therefore also
equally important.
Modelling under unbalanced disturbances 31

 Modelling must also be done under unbalanced

disturbances.
 Fifth order induction machine is considered, incorporating
the stator transients.
 The commonly used ZIP model can not be used to
represent the three phase load under unbalanced
disturbances.
 Existing ZIP model is modified for the unbalanced
disturbances
Field experiment 32
Field experiment: results 33
Conclusion 34

 Number of PMU installations are increasing. PMU

applications are going to be more common in control
centres.
 WAMS is going to play a major role in monitoring,
protection, and control of power systems.
 Distribution PMUs are also coming up.
 Managing the large amount of information and ensuring
its security will be a big challenge.
 Cyber-security and cloud computing applications are
bound to get increased importance.