Under-Frequency Load Shedding (UFLS) Schemes A Survey · E-mail: [email protected] D. H. A....

International Journal of Applied Engineering Research ISSN 0973-4562 Volume 11, Number 1 (2016) pp 456-472

© Research India Publications. http://www.ripublication.com

456

Under-Frequency Load Shedding (UFLS) Schemes – A Survey

M. Lu

PG Scholar, Department of Electrical and Electronic Engineering,

Universiti Malaysia Sarawak, Kuching, Malaysia.

E-mail: [email protected]

W. A. W. ZainalAbidin

Associate Professor, Department of Electrical and Electronic Engineering,

Universiti Malaysia Sarawak, Kuching, Malaysia.


T. Masri

Senior Lecturer, Department of Electrical and Electronic Engineering, Universiti Malaysia Sarawak, Kuching, Malaysia.


D. H. A. Lee, S. Chen

Sarawak Energy Berhad, Kuching, Malaysia.


Abstract The Under-frequency Load Shedding (UFLS) scheme has been

used by utility companies around the world to mitigate frequency drop caused by simultaneous or cascading tripping of

transmission lines and/or generators in a power system. In the

effort to devise an optimal load shedding scheme, it is imperative

that investigations are done on the many factors that may affect

the response of the scheme in the event of a system contingency.

This paper starts by analysing the implementation of UFLS in

various power utility companies in Asia, Europe, Australasia,

South Africa, Middle East and the Americas. It is observed that

UFLS in these continents differ in terms of total load shed,

number of UFLS blocks, average block size and trip frequency

deviation thresholds depending on their system size, system

inertia and generation mix. This paper also looked at the usage of System Frequency Response (SFR) models in the computation of

UFLS and system parameters. Analysis on the SFR model

showed that the impact of voltage dependence of loads was not

taken into consideration in the early implementation of the model

albeit load characteristics have significant influence on the

dynamic behaviour of power systems during low frequency

oscillation and severe faults. SFR model incorporating frequency

and voltage dependence load models was proposed later in

literature and used in the design of an optimal UFLS scheme.

Investigation was also conducted on UFLS operating philosophies

in terms of load shedding trigger, power imbalance estimation and distribution of load shedding. UFLS operating philosophies based

solely on frequency parameters is inadequate to determine the

stability of a power system especially for an islanded power

system following severe disturbances. The power system may be

susceptible to voltage collapse as well, which will lead to total

system blackout within shorter time duration as compared to a

frequency collapse phenomenon. Hence, trigger condition

considering voltage information and voltage stability criterions

were introduced and implemented in UFLS schemes. Distribution

of load shedding based on power flow tracing method catering for

both frequency and voltage instabilities was also introduced

and proven to give optimal system response.

Keywords: Under-frequency Load Shedding (UFLS),

frequency drop, load shed, System Frequency Response

(SFR) model, operating philosophies.

Introduction Under-frequency Load shedding (UFLS) is a common

demand reduction measure taken by most energy utilities to

mitigate frequency drop whenever there is dangerous

imbalance between loads and generation due to disturbances to the system such as loss of generation or major

transmission lines. UFLS is performed to force the perturbed

system to a new equilibrium state, balancing load and

generation, to minimize the risk of a further uncontrolled

system separation and loss of generation and to prevent

continuous frequency drop which may lead to total frequency

collapse and prolonged system outage [1].

UFLS has to be well-coordinated between interconnected

power systems and also with other system defense schemes

such as Under-Frequency Capacitor Shedding (UFCS),

Under-Frequency Generator Isolation (UFGI), Special

Protection Scheme (SPS), Under-Voltage Load Shedding (UVLS) and other automatic actions that will kick in to arrest

system from collapsing during abnormal frequency, voltage

and/or power flow conditions [2]. For power system that has

industrial and commercial customers with local generation

connected to it, UFLS can detect onset of disturbance, isolate

power systems by opening system ties and trip non-essential

industrial loads to match total loss of generation. However,

tripping these tie lines having active parallel generation

reduces the beneficial impacts of load shedding because the

sources of generation supporting system inertia are

eliminated.



457

Various UFLS load shedding schemes are used in the industry

worldwide. UFLS can be classified into two main categories –

fixed and adaptive. The fixed scheme sheds a pre-defined amount

of loads when the system frequency falls below a certain

threshold. Further load shedding is performed if the system

frequency continues to deteriorate after the activating the first stage of UFLS scheme. The total amount of load shed as a

function of time is expressed as a sum of incremental step

function [3]. The adaptive scheme sheds loads dynamically by

taking into consideration severity of disturbances and system

frequency-voltage characteristics and is proven to provide optimal

amount of load shed thereby giving better frequency recovery,

lower frequency overshoot and lesser minimum frequency

deviation [3]. However, simulation studies for system

contingencies for three load step increments also show that the

adaptive scheme only out-performs the fixed scheme for a

medium size disturbance [4].

Terjiza presented a new approach to adaptive UFLS in [5]. In the initial stage, the frequency and the rate of frequency change were

estimated using the non-recursive Newton-type algorithm and in

the latter stage, the magnitude of disturbance was determined

using the simplest expression of the generator swing equation.

In [6] and [7], two centralized adaptive load-shedding algorithms

called the response-based algorithm and the combination of

event-based and response-based, were introduced. These methods

were shown to be capable of mitigating instabilities during severe

or combinational power system contingencies.

Combinational load shedding schemes gain popularity when the

industry realizes the susceptibility of power systems to combinatorial frequency and voltage collapses. In [8], the authors

presented two combinatorial algorithms to combine UFLS and

UVLS schemes whereby locally measured frequency and voltage

signals were used to determine distribution of load curtailment

during severe disturbances. These methods known as V-F and

dV-F load shedding schemes, were shown to increase adaptability

of UFLS relay and enhance power system susceptibility to

voltage collapse by improving voltage stability margins whereby

loads with lower voltage level and greater voltage decline were

shed sooner. Voltage stability has become an issue in recent

decades as power systems are getting more interconnected and

heavily loaded [9], [10]. M. V. Suganyadevi and C. K. Babulal in [11] has defined voltage stability as "the ability of a power system

to maintain steady state voltages at all buses in the system after

being subjected to a disturbance from a given initial operating

condition".

Implementation of an integrated UFLS and UVLS scheme was

proposed in [12] to keep power islands stable after power

oscillations or out-of-step islanding. An interconnected grid

system such as China was susceptible to large disturbances and

has the tendency to go out-of-step. With the integrated scheme, a

certain proportion of load would be shed when system splitting

occurred whereby quantum of load shed was set according to the generation deficit of the area. UVLS would be activated if system

voltage was low but frequency was high at the same time. The

authors believed that this would improve the voltage profile,

increase the generators' output power, and limit island frequency.

Following system islanding, UFLS would be triggered when the

island frequency was lower than the UFLS setting threshold

whereas UVLS would work if the island voltage profile was

lower than its setting value. Simulation results showed that

the integrated measure could keep the island stable.

In [13] and [14], the conventional non-adaptive under-

frequency load-shedding scheme has been deemed

inadequate to provide sufficient protection against system

collapse. The authors in [13] proposed the use of three adaptive combinational load shedding methods to improve

the operation of a conventional UFLS scheme to enhance

power system stability following severe disturbances.

Operation of the conventional and the proposed load

shedding methods were simulated in an actual large network.

Obtained simulation results confirmed that the proposed

methods would provide considerable enhancement in the

power system voltage stability margin, and by using the

proposed algorithms, various power system blackouts could

be prevented.

Zhang and Wang proposed the use of a WAMS-based

adaptive UFLS scheme using real-time online power system measurements in [15]. The scheme was proven to be capable

of adapting its sensitivity to the current operating conditions

thereby eliminating the need for making many assumptions

necessary while tuning a conventional local scheme. In [16],

a coordinated UFLS and UFCS scheme combining with

automatic switching of shunt reactors was presented to

optimize the performance of the existing UFLS scheme

following major disturbances which resulted in large

mismatch between load and generation. Fixed and switched

shunt capacitors in service during normal operation for

maintaining system voltage and dynamic MVAR reserve could generate surplus reactive power post operation of

UFLS relays resulting in over voltage issues, generator under

excitation and other undesirable conditions such as

transformer saturation and ferro-resonance.

This paper is structured as follows: Survey on the application

of UFLS in 36 different power systems worldwide in terms

of total load shed, number of UFLS stages and average block

size followed by analysis on the System Frequency Response

(SFR) Models for computation of load-frequency response,

and lastly, study on the operating philosophy of UFLS

scheme in terms of trigger criteria, imbalance estimation and

load shedding distribution.

UFLS in Asia, Europe, Australasia, Africa, Middle

East and the Americas A few countries within each continent were selected

randomly for this survey. For Asia, survey was done on the

Malaysia (East), Japan, Taiwan, Central China Power Grid, North China Power Grid, Eastern China Power Grid,

China/Hong Kong Interconnection and Thailand; for Europe,

Portugal, Spain, Union for the Coordination of Transmission

of Electricity (UCTE), Ireland, Western Europe, United

Kingdom (UK), Slovenia and Sweden; for Australasia, Africa

and Middle East, New Zealand (North and South Island),

South Africa, Libya, Bangladesh and Israel; for the

Americas, Florida Reliability Coordinating Council (FRCC),

Midwest Reliability Organization (MRO), Mid-Atlantic Area

Council (MAAC), Pennsylvania-New Jersey-Maryland

Interconnection (PJM), Electric Reliability Council of Texas

(ERCOT), Western Electricity Coordinating Council (WECC), Mid Continent Area Power Pool (MAPP), Mid



458

America Interconnected Network (MAIN), Southwest Power Pool

(SPP), East Central Area Realibility Coordination Agreement

(ECAR), Northeast Power Coordinating Council (NPCC),

Northwest Power Pool (NWPP), Brazil and Guam. Most of these

systems are operated at nominal frequency of 50Hz except for

Taiwan and the Americas. Parameters of load shedding algorithm differ by countries and

continents [17]. In this survey, various UFLS criterion such as

maximal capacity of a load connected to the load shedding

system, selection of average block sizes for load shedding,

selection of the number of necessary steps for the load shedding,

selection of upper frequency setting level, selection of intervals

between nearest frequency settings and selection of the tripping

time for a load shedding automatic relay were also investigated.

Figures 1 to 3 summarizes the survey on UFLS scheme in Asia in

terms of total load shed/system loading, number of blocks and

average block size [18]-[22]. Out of the eight power systems in

this survey, two are located in Southeast Asia namely Malaysia (East) and Thailand, four from the People’s Republic of China

and another two from Northeast Asia namely Japan and Taiwan.

Majority of these systems adopt customer load shedding schemes

except for Taiwan and China/Hong Kong Interconnection

whereby pumped-storages are also shed during system

disturbances.

Figure 1: Total Load Shed/System Loading (%) for UFLS

in Asia

Figure 2: Number of UFLS Blocks for UFLS in Asia

Figure 3: Average Block Size (%) for UFLS in Asia

A stark difference in terms of UFLS allocation is observed

among the surveyed power systems in Asia. China/Hong

Kong interconnection has the highest UFLS allocation of

54% of total UFLS load shed with respect to system loading

followed by Malaysia (East) and Thailand at 50%, North

China Power Grid at 47%, Central China Power Grid at 36%, Japan at 30%, Taiwan at 27% and lastly Eastern China Power

Grid at 15%.

A difference of 39% in terms of maximal loads allocated for

load shedding is observed between the power system with the

highest and lowest UFLS allocation in Asia with an overall

average of 39% as well for this continent. It is imperative to

have sufficient allocation in the UFLS scheme to arrest rapid

frequency decay and tripping of generating units due to

triggering of generating under-speed protection setting as this

may lead to total system collapse as experienced by the

power system in Sarawak on 9th August 2008 [23].

In Taiwan, 7% of total system loading has to be shed per every Hertz of frequency drop experienced by the power

system based on machine inertia, load characteristics,

spinning reserve and technical requirements for UFLS in

order to restore system frequency back to acceptable

operation limits [24]. Although sufficient loads have to be

shed to match the loss of generation, power grid operators

should also be wary of the possibility of creating over-

voltage in the power system caused by the increase of total

load shed exceeding the acceptable limits [23]. This is critical

because overvoltage conditions caused by simultaneous

unloading of transmission lines and uncontrolled load shedding in cable networks with high shunt capacitance may

render the load shedding scheme ineffective [2].

The number of load shedding steps and the size of each step

are influenced by factors such as system inertia constant,

percentage overload, islanding patterns and reactive support

by generation during system disturbances [2]. From Figure 2,

it is obvious that Japan has the least number of UFLS blocks

with one of the highest percentage of average block size.

UFLS in Japan consist of a total of three blocks with average

block size of 10% and with intervals of 0.5Hz between

nearest frequency settings. This setting increase the probability of over-loadshedding during light load condition

leading to over-frequency problems [2].



459

Albeit a responsive govenor in the system will help to limit the

over-frequency to reasonable level, the problem may still be a

threat to the system if the amount of imbalance is large enough to

trigger load rejection controls or cause problems in plants due to

the sudden decrease in steam flow through the turbines. Load

controllers may eventually override the govenor action and return power to original set point, sinking the system back to under-

frequency condition. In addition, shedding of a big block of loads

at one time may have a big impact on an already weakened

system.

Similar situation exists for China/Hong Kong interconnection

which has the largest average block sizes but among one with the

least number of load shedding blocks. However, in cases where

over-shedding of loads do not cause over-frequency, systems with

larger average block size will take lesser time to stabilize the

system frequency. This is critical especially when power system

is encountering severe contingencies and quick immediate

intervention is required to arrest system frequency decline and prevent a total system collapse. The remaining survey results for

UFLS in Europe, Australasia, Africa, Middle East and the

Americas are shown in Figures 4 to 12 [25]-[33].


in Europe

Figure 5: Number of UFLS Blocks for UFLS in Europe

Figure 6: Average Block Size (%) for UFLS in Europe

Figure 7: Total Load Shed/System Loading (%) for UFLS in

Australasia, Africa & Middle East

Figure 8: Number of UFLS Blocks for UFLS in Australasia,

Africa & Middle East

0

10

20

30

40

50

60

70

Total Load Shed/System Loading (%)

Portugal

Ireland

Western

EuropeUK

UCTE

0

1

2

3

4

5

6

7

8

9

10

Number of UFLS blocks

Portugal

Ireland

Western Europe

UK

UCTE

Slovenia

Spain

Sweden

0

5

10

15

20

25

30

Average Block Size (%)

Portugal

Ireland

Western Europe

UK

UCTE

Slovenia

Spain

Sweden



460

Figure 9: Average Block Size (%) for UFLS in Australasia,

Africa & Middle East


in the Americas

Figure 11: Number of UFLS Blocks for UFLS in The Americas

Figure 12: Average Block Size (%) for UFLS in

the Americas

From Figure 4, it is observed that the eight surveyed

European power systems have a reasonably similar allocation

for UFLS with total load shed size of within the range of

46.7% to 60% of total system loading. This is different from

other parts of the world such as Asia, Australasia, Africa,

Middle East and the Americas whereby distinct difference in

load shed allocation is observed. In Europe, UK tops the list in terms of total UFLS allocation with 60% total load

shed/system loading followed by Ireland and Slovenia with

par allocation of 55%, Portugal, Western Europe, UCTE and

Sweden at 50% and lastly Spain at 45%. The average total

UFLS allocation for Europe is 52%, which is highest among

all continents in this survey.

It is observed from Figures5 and 6 that UFLS in UK has

highest number of load shedding blocks in Europe but with

relatively small average block size whereas UFLS in

Portugal, on the other extreme, is split into just two blocks

but with largest average block sizes among power systems in all continents included in this survey. Splitting the UFLS

scheme into many small blocks is beneficial in reducing over

load-shedding but this may render the scheme less-effective

if larger amount of loads has to be shed to save the system

from frequency instability. The scheme will respond slower

and cause lower minimum transient frequency. On the other

hand, a power system like Portugal may be susceptible to

over-frequency due to tripping of large blocks of customer

loads during less critical contingencies. UFLS in Western

Europe and UCTE are similar in terms of total load

shed/system loading, number of UFLS blocks and average

block size. In Figures7 to 9, UFLS in Israel stood out among the other

power systems in Australasia, Africa and Middle East.

Within these three continents, Israel has the highest UFLS

allocation of 54.7% of total load shed/system loading

whereby more blocks with smaller average block sizes are

available for shedding followed by Libya at 52%, South

Africa at 50%, Bangladesh at 41% and New Zealand at 32%.

The average UFLS allocation for all three continents is 44%.

Israel has the overall highest number of UFLS blocks among

all countries surveyed in this study. UFLS adopted by South

Africa and Libya are similar in terms of total load



461

shed/system loading, number of UFLS blocks and average block

size. In addition, it is also noted that UFLS for North and South

Island in New Zealand are identical in terms of total load

shed/system loading, number of UFLS blocks and average block

size.

Figures10 to 12 show that FRCC has the highest total load shed of 56% relative to system loading with the most number

of UFLS blocks in the Americas. Brazil has the second highest

quantum of UFLS allocation at 55% with the second highest

average block size also. Next in line in terms of UFLS allocation

is Guam at 38%, followed by MRO at 34%. Guam is also ranked

fourth amongst other power utility companies in the Americas in

terms of number of UFLS blocks and average block size. MRO is

ranked second in terms of number of UFLS blocks but has one of

the smallest average block sizes which is similar to that of North

China Power Grid and Ireland.

MAAC, PJM, MAPP, MAIN and SPP has identical

UFLS allocation in terms of total load shed/system loading (30%), number of UFLS blocks (three) and average block size

(10%). Next in line are WECC with total UFLS allocation of

29.1% with respect to system loading, NWPP at 28% and lastly

ERCOT, ECAR and NPCC, with the lowest quantum of UFLS

allocation of 25% respectively. WECC, NWPP and ECAR have

relatively high number of UFLS blocks with small block sizes.

On the contrary, ERCOT and NPCC have the least number of

UFLS blocks but with relatively high average block size. The

average total UFLS allocation for this continent is 33%.

Table 1 and Figure 13 show the trip frequencies for the

first and last UFLS block on the 36 power systems studied. FRCC and PJM have the least tolerance for frequency deviation whereby

first UFLS block for both systems operates once the system

frequency drop to 0.5% from the nominal frequency.

Table 1: Trip Frequency Deviation from Nominal Base (%)

Region Power System First UFLS

Block

Last UFLS

Block

Asia Malaysia (East)

Japan Taiwan Thailand Central China Power Grid North China Power Grid Eastern China Power

Grid China/Hong Kong Interconnection

1.2

2.0 1.0 2.0 2.0 1.5 2.0

2.4

3.6

4.0 1.5 4.2 4.0 4.5 4.0

4.0

Europe Portugal Spain UCTE Ireland Western Europe UK

Slovenia Sweden

2.0 1.5 1.0 1.2 1.5 2.0

2.0 2.0

3.0 2.7 1.6 1.5 2.3 2.2

4.0 4.0

Australasia, Africa & Middle East

New Zealand (North Island) New Zealand (South Island) South Africa Libya

2.2 2.5 0.8 0.6

2.5 4.5 2.1 1.4

Bangladesh Israel

NA 1.2

NA 4.4

The Americas FRCC MRO MAAC

PJM ERCOT WECC MAPP MAIN SPP ECAR NPCC

NWPP Brazil Guam

0.5 1.2 1.2

0.5 1.2 1.5 1.2 1.2 1.2 0.8 1.2

1.2 1.5 1.0

3.0 2.7 2.5

1.5 2.5 2.8 2.2 2.2 2.2 2.2 2.0

2.0 2.5 2.8

Figure 13: First and Last UFLS Blocks

The first step frequency is usually set be just below the

system normal operating or the frequency at which the

system could continue to operate [34]. In most systems, the

first block is set to trip at frequency deviation of slightly

lower than 1% from nominal as ±1% of nominal frequency is

the normal operating range whereby generating units can operate continuously [2]. Earlier initiation of load shedding

enables the system to be able to respond faster and allows

more time for generator’s Automatic Voltage Regulators

(AVRs) to respond so that more loads can be shed without

over-voltage problems. However, it is essential to ensure that

the first UFLS block is not too close to the nominal

frequency to avoid unnecessary tripping caused by frequency

swings caused by minor disturbances or sudden changes of

loads whereby the system is able to recover on its own.

Last UFLS block for North China Power Grid and New

Zealand (South Island) is triggered at frequency deviation of

4.5% from nominal and this marks the largest frequency deviation threshold tolerance for the last UFLS block. The

final UFLS block must be coordinated with equipment

operating limits during low frequency operation for instance,

operating limits of plant auxiliaries and turbine protection.

0.0

0.5

1.0

1.5

2.0

2.5

3.0

3.5

4.0

4.5

5.0

Fre

qu

ency

Dev

iati

on (

%)

First UFLS block Last UFLS block



462

The performance of the auxiliary equipment deteriorates at

frequencies below 48Hz and the under-speed turbine protection is

usually set to 47.5Hz for a system operating on 50Hz [23]. Relay

and breaker operating time should also be taken into

consideration when setting the UFLS stage limits.

In addition, it is also observed that the Israel has the widest range of distribution for UFLS blocks with a significant difference of

3.2% between the first and last load shedding. There is advantage

of having a wider frequency range over which loads can be shed

before system frequency reaches the minimum operating limits of

frequency-dependent electrical equipment such as AC motors,

turbo rotors and electrically operated clocks [1]. On the contrary,

UK has the smallest range of distribution for UFLS load shedding

blocks with a mere 0.2% difference between the first and last load

shedding block.

In terms of selection of intervals between nearest frequency

settings, UFLS in Guam has a significantly large interval

difference between two load shedding stages of 0.6Hz and UFLS in Taiwan has a distinctively small interval difference between

two load shedding stages of 0.08Hz. The intervals between two

load shedding blocks should not be set too close to each other to

prevent simultaneous triggering of two blocks when power

system experiences frequency decay. While relay manufacturers

provide set points in increments of 0.01Hz, there is typically 10 to

14 cycles of delay, including relay and circuit breaker operating

times, from the instant frequency reaches the set point to the

instant of actual load shed [2]. During this delay, the frequency

continues to drop, that is, the frequency at which load actually

sheds is below the set point. If the spacing between the shedding stages is too close, the load shed initiated by a stage could

actually shed the load while the frequency decline is continuing

through the following stages due to the inherent delay. Another

reason to space the frequency settings of the stages of load

shedding is that during a system disturbance the frequency

between different locations of the interconnected system will not

be the same. This variation in frequency has been observed to be

as great at 0.2Hz.

In terms of selection of the tripping time for a load shedding

excluding delay blocks, UFLS in Eastern China Power Grid has a

notably long delay time of one-half seconds for operation. On the

other extreme, UFLS in Taiwan operates without time delay upon trigger, disregarding delay caused by other factors such as circuit

breaker opening time and delay in operation of protection devices.

Time delay has been introduced in UFLS schemes to reduce

possibilities of “false triggering” or operation of the load

shedding scheme due to transient frequency dips and to provide

time for the load/frequency controls in the system to respond [2].

However, system stability may be jeopardized if time delays are

set to be longer than necessary. Time delay settings range from a

few cycles to several seconds, depending on the number of load

shedding stages and the expected rate of frequency decline.

Relatively long time delays are set to provide time for the system controls to respond where decline of system frequency is slow

and shorter time delays are used to cater for severe contingencies

where rapid decline of frequency is expected. Some power

systems such as Spain and Israel supplements the conventional

load shedding scheme with rate of change of frequency

monitoring elements to ensure optimal load shedding during

system contingencies.

System Frequency Response (SFR) Model

Early research works have concentrated on studying the

frequency-response of power systems in the event of power

disturbance. In 1971, the basic principles and philosophy of

frequency actuated load shedding and load restoration

program including its implementation on the American

Electric Power (AEP) System was presented by Maliszewski, Dunlop and Wilson [35]. A year later, Chan, Dunlop and

Schweppe modelled the effects of governor-turbine dynamics

on the average system frequency behaviour of a multi-

generator system after a major generation loss or load change

when the system remains in synchronism to obtain the

maximum frequency deviation and the time at which the

maximum occurs [36]. To reduce complexity while analysing

a multi-machine system, a nonlinear, high dimensional

closed-loop model was converted into a linear, approximated,

low dimensional open-loop model via delay and canonical

models. The delay model was used to model fast time

constants and valve motion using pure time delay thereby converting the closed loop model into an open loop one. The

canonical model portrayed the turbine reheats response as a

linear combination of a set of basis functions and combined

many machines into one simplified, low dimensional model.

The first System Frequency Response (SFR) model was

introduced by Anderson and Mirheydar in 1990 [37]. In this

work, a simplified low order SFR model was used to estimate

the frequency behaviour of a large power system or islanded

portion of the power system in the event of sudden load

disturbances whereby nonlinearities are assumed to be

negligible and only the largest time constant in the equations of the generating units of the power system are taken into

consideration. Assuming reheat steam turbine generators to

be the dominant source of generation and generating unit

inertia and reheat time constants pre-dominate the system

average frequency response, the resulting system frequency

response in terms of initial value of the rate of frequency

change and the size of the disturbance Pstep that caused the

frequency decline were computed in closed form:

df

dt⌋

t=0=

1

2H(Pstep) (1)

wheref is expressed in per unit on the base of the nominal

system frequency (50 or 60 Hz) and Pstep is in per unit on the

total apparent power of the whole system. The initial value of

the rate of frequency change is proportional to the size of the

disturbance through the inertia constant H.

In 1992, the SFR model was used indirectly in the

design of an adaptive UFLS for an interconnected power system during verification of acceptable power system

performance following operation of UFLS scheme [38] and a

year later, the SFR model was used directly used to design an

optimal and dynamic UFLS scheme for Israel which is a

small and isolated power system with respect to minimizing

the impacts arising from a host of credible contingencies

[39]. A simulation model of the SFR-UFLS was embedded

within the computation model of the optimization algorithm

to find the optimum load-shed size while other UFLS design

parameters were pre-determined.

An analytic model incorporating UFLS for a multi-machine power system was introduced by Lee in 2006 [3] and closed-



463

form expressions of load-frequency response, including the effect

of UFLS following system contingency were derived to compute

system and UFLS parameters such as minimum transient

frequency, steady-state frequency, number of stages and instants

of activation of UFLS scheme. These parameters were then

directly incorporated into constraints equations and fitness functions for finding optimal UFLS settings. The use of SFR-

UFLS model has eliminated the need for computation-intensive

time-domain simulation approach as results predicted by the

model were proven to be consistent with those of an actual system

incident and those derived from time-domain simulations of the

corresponding full-scale power system.

In [40], the SFR-UFLS model was redeveloped to generate an

optimal load shedding method based on disturbance estimation

that optimally sheds load following one single particular

contingency event taking into account power system operating

conditions. The proposed optimal load shedding scheme was

tested on the 39-bus New England test system to verify the performance against random load shedding scheme.

In 2010, Wu, Gao and Dai conducted a study on the impacts of

Primary Frequency Regulations (PFRs) and Over-speed

Protection Control (OPC) system actions of the generation units

on system responses of an islanded power system during

contingency situations [41]. The authors modelled a SFR-UFLS

model incorporating load shedding controls and OPCs on an

islanded power system with hydro and fossil powered generators

connected. Simulation results indicated that the load-frequency

response generated from SFR-UFLS model matches actual

system response recorded during disturbance. The authors also observed reduced frequency oscillation when the optimized

scheme was used and improved system frequency response when

the PFR of hydropower was not operated during the speed

governing process and when fossil fired turbine units were

dispatched more.

SFR model was also used to estimate frequency response of

power systems penetrated by distributed generation [42]. Global

concentration on distributed generation is on the increase due to

energy demand increase every year. Having distributed generation

within the power system improves power quality, reduces system

losses and helps in maintaining system voltage [43]-[45].

Shariatiet. al in [46] used a high-order multi-machine frequency response model for power system dynamic simulation.

Classification of modern power system components and using an

equal unit for each class was proposed in this work. Results

showed that Artificial Neural Network (ANN) models can be

implemented as a fast dynamic simulator of electric power

system. This assessment included a review of significant research

works on power system dynamic simulation and frequency

response model leading to an integrated UFLS system design.

In [47], an Implicit Enumeration with Adaptive Discretization

algorithm based on the SFR model was proposed to facilitate the

design of a WAMS-based adaptive UFLS scheme. Issue on generator mechanical power (saturation) limits was addressed and

simulations conducted on a real, small system weakly connected

to the Eastern China system have given optimal results for both

small and severe disturbances.

UFLS Operation Philosophy This section looks at UFLS operation philosophy in terms of load

shedding trigger criteria, estimation of load imbalance according

to load characteristics and distribution of load shedding. The

process flow is shown in Figure 14.

Figure 14: First and Last UFLS Blocks

A. Trigger Criteria

The trigger is a signal that starts the load shedding action

[48]. Variation of voltage and frequency following

disturbance is different. Frequency usually decreases

continually until enough amount of load is shed, while voltage usually increases gradually after its sharp decrease

following the disturbance [8].

In the early days of UFLS implementation, frequency

threshold Fhas always been used as the trigger parameter in

conventional UFLS schemes. However, this method is

ineffective/slow in handling severe contingencies with rapid

frequency decline. Hence, in 1992, an adaptive UFLS scheme

triggering methodology based on the initial rate of change of

the frequency was proposed to estimate the size of the step

change in load caused by the system separation in an

interconnected system and forecast the time and magnitude of maximum frequency deviation [38]. The trigger for first step

of the UFLS scheme was estimated to be one-half the static

load shed target, with additional increments of about 0.1 per

unit to be shed at 0.3 Hz increments until the dynamic load

shed amount has been reached.

A load shedding trigger based on rate of frequency change

was also used in [49] whereby load shedding would be

activated when power system vulnerability analysis indicated

that the system was approaching an extreme emergency state

during the occurrence of catastrophic disturbances. The rate

of frequency change was known to be an instantaneous indicator of system deficiency or power imbalance and

oscillatory in nature due to oscillation in the change of

generator machine speed, hence, was used with the frequency

function to provide a more selective and faster operation [2].

Load shedding trigger based solely on frequency

measurements is inadequate especially after islanding of an

interconnected power system as an islanded power system

following severe disturbances may experience low voltage

with high frequency when an area being split from the main

grid has large generation deficit [12],[14]. Generator over-

speed protection will be triggered thereby causing tripping of

generators simultaneously or in a cascaded manner. Terzija in [50] asserted that the trigger for first step of

conventional load shedding trigger has always been based on

frequency or voltage information independently. Various

load shedding trigger methods considering both voltage and

frequency jointly at each bus were proposed in literature after

that [6]-[8], [12]-[14].

In [6] and [7], two centralized adaptive load-shedding

algorithms were introduced namely the response-based

algorithm and the combination of event-based and response-

based. In the response-based algorithm, triggering of load

shedding scheme was based on response of the system

Step 1: Determine

Trigger

Step 2: Estimate

Power Imbalance

Step 3: Determine

Distribution of Load Shedding



464

disturbances for instance, voltage or frequency signals whereas

for an event-based algorithm, the state of important elements

within a system such as critical transmission lines or generators

were used as the load shed trigger. These methods were shown to

be capable of mitigating instabilities during severe or

combinational power system contingencies. Locally measured frequency and voltage signals were used jointly

as trigger for load shedding in [8] and to determine distribution of

load curtailment during severe disturbances in two combinatorial

algorithms consisting of a combination of UFLS and UVLS

schemes. These methods known as V-F and dV-F load shedding

schemes, increased the adaptability of UFLS relay and enhance

power system susceptibility to voltage collapse by improving

voltage stability margins whereby loads with lower voltage level

and greater voltage decline were shed sooner.

Using the V-F method, measured voltage and frequency levels at

locations equipped with load shedding relays was compared with

pre-set voltage and frequency blocks in each step as was the case with conventional UFLS scheme but total amount of load shed

was not divided among pre-defined load shedding steps. Voltage

level that was used as criteria in V-F load shedding scheme

indicated the voltage strength at each relay locations following

disturbance and this parameter depended on pre-disturbance

voltage level and amount of voltage decline following

disturbances. As the amount of voltage measured at the instant

when frequency reaches its threshold in each step was

unpredictable therefore making it difficult to set trigger threshold

for voltage, appropriate reset time was set for voltage blocks in

each step so that load shedding decision was made based on voltage level measured before the first load shedding step was

activated.

Using the dV-F method, measured frequency and the amount of

voltage decline was used as load shedding trigger criteria. The

amount of voltage decline was prepared by a filter which holds

the initial value of the input voltage in its state variable for 5

seconds and provides the difference between the instantaneous

input voltage and the value of its state variable. The amount of

voltage decline was calculated for a few seconds following

disturbance and this parameter was able to address location of

disturbance and reactive power deficit better than voltage level as

the effect of pre-disturbance voltage level was not used in this method.

The authors in [14] proposed the use of locally measured

frequency (F) and voltage index (VI) derived from the integral of

instantaneous value of positive-sequence voltage decline signals

as load shedding trigger in linear, parabolic and multi-step

adaptive combinational load shedding methods [14]. In order to

determine the load-shed region in the VI-F diagram for linear and

parabolic combinational load shedding methods, the range of F

and VI change at relay locations considering the worst

contingencies was determined followed by selection of

appropriate oblique line or parabolic curve as the load-shedding criterion in the load shedding framework.

For multi-step combinational scheme, the network was divided

into some local areas in which the bus voltages usually change

coherently. Then, a sufficient amount of load was taken for load

shedding in each area and the selected loads were divided

between some load shedding steps. The use of VI enabled faster

and localized load-shedding from locations experiencing low

voltage conditions for a longer time period. Simulation results on

an actual network and a standard test system using different

load shedding criterions confirmed that the proposed scheme

showed considerable enhancement in the power system

voltage stability margin and prevented various power system

blackouts.

In [51], the principles of a centralized UFLS scheme based on a frequency stability boundary curve defined within the

frequency-rate of frequency change (ω-dω/dt) phase plane

were presented for the small isolated power system in Spain.

The frequency stability boundary was derived from a

simplified power system model taking into account rotor and

turbine-governor system dynamics. Load shedding was

triggered when system trajectory fell outside the frequency

stability boundary or if its tendency in the ω-dω/dt phase

plane pointed to the boundary.

In [45], trigger based on frequency threshold, rate of

frequency change and minimum eigenvalue of the Jacobian

matrix φmin was proposed. Using modal analysis, the lth

eigenvalue of the Jacobian matrix, φl was defined as:

φl=

∆Qml

∆Vml (2)

where∆Vml and ∆Qmlwere the lth modal voltage and reactive power variation respectively. lth modal voltage would

collapse when φl ≤ 0, hence, φminindicated the condition

that was most prone to collapse and used as a reliable

indicator for voltage stability of a power system.

Apart from using frequency and voltage information, other

combinations of load shedding trigger parameters were also

proposed in literature. In [2], power system information such

as voltage, total system inertia, loads, system demand,

spinning reserve, system kinetic energy and the amount of

lower-priority load available for shedding were used in conjunction with locally measured rate of change of

frequency to determine the trigger for activating load

shedding for an adaptive load shedding scheme in an isolated

power system. Simulation results on a representative system

subjected to demands and outage constraints have proven the

ability of the proposed scheme to reduce load shedding to a

value close to the theoretical minimum while maintaining

satisfactory frequency response. The proposed scheme was

shown to be robust, insensitive to temporary SCADA failures

and individual relay malfunctions, and could be implemented

with currently available communications and microcontroller technology.

In [52], a load shedding strategy for islanded distribution

systems with DGs based on frequency information, rate of

change of frequency, customers’ willingness to pay and loads

histories to shed optimal quantum of loads in the islands was

proposed to stabilize power system frequency. Simulation

results showed that the proposed method was effective in

shedding optimal quantum of loads to stabilize frequency. A

summary for literature study done on trigger criteria for

UFLS is shown in Table 2.



465

Table 2: Literature Study on Trigger Criteria for UFLS

Year Trigger Criteria

Early Days

1992

2003

2006

2007

2008

2009

2011

2013

Frequency Threshold

Initial Rate of Frequency Change

Rate of Frequency Change

Frequency and Voltage Information Independently

Power System Information and Rate of Frequency

Change, Voltage/Frequency Information (Event-based),

Status of Critical Assets (Response-based)

Frequency and Voltage Information Jointly

Frequency Information, Customers’ Willingness to Pay,

Load Histories

VI-F Diagram

Frequency Boundary Curve, Frequency Threshold and Minimum Eigenvalue of Jacobian Matrix (Voltage

Stability Assessment)

B. Load Characteristics for Power Imbalance Estimation

Frequency of a power system is affected by components

connected to it especially bulk customer loads [46]. Past literature

has shown that load characteristics have significant influence on

the dynamic behaviour of power systems during low frequency

oscillation and severe faults.

In the classic SFR model, only inertia constant of generator and frequency information are considered and the impact of voltage

dependence of loads is not taken into account. However, as

suggested in IEEE standard for UFLS [53], the load models with

voltage and frequency dependence should also be included in the

design for UFLS in order to achieve accurate active power

imbalance estimation. Voltage deviation controls the variation of

the load active power in the initial one to two seconds following

disturbance after which the frequency takes precedence [54].

When a power system encounters a severe under-frequency

condition, generator governor will regulate the mechanical output

power by frequency variation and the loads connected to the

system will regulate its active power [55] to prevent system from collapsing. Loads can be assumed to be highly frequency and

voltage dependent in order to simulate the worst case contingency

scenario for the design of an optimal load shedding scheme [56].

There are three main types of loads usually used in power system

dynamic analysis namely static loads, dynamic loads and

composite loads. Characteristics of static load models can be

expressed using voltage and frequency dependent algebraic

models [57]. The exponential and polynomial representation of

static loads has been widely used in literature:

PL= ∑ PL,jMj=1 (3)

QL= ∑ Q

L,jMj=1 (4)

where PL and QL are the current total active and reactive power

load of all the load buses, PL,j and QL,j are the current active and

reactive power load of the jth load bus. Using an exponential model:

PL,j=PL0,j× (Vj

V0,j)

αj

(1+Kpf∆f) (5)

QL,j

=QL0,j

× (Vj

V0,j)

βj

(1+Kqf∆f) (6)

where PL0,j and QL0,j are the initial active and reactive power

load of the jth load bus before the disturbance, Vj and V0,j are

the current and initial voltage magnitude of the jth load bus

after the disturbance, αj and βj are the factors for active and

reactive power dependence of the load on voltage deviations,

Kpf and Kqf are the active and reactive load-frequency

characteristics, M is the total number of load buses and ∆f is

the frequency deviation. αjusually ranges between 0.5 and

1.8 whereas βj, which is a nonlinear function of voltage and

is caused by magnetic saturation of distribution transformers

and motors, usually ranges between 1.5 and 6. Kpfranges

from 0 to 3.0 and Kqf ranges from -2.0 to 0.

As voltage variation is believed to be much faster

and larger than the change of frequency whereby frequency

does not decay for more than 0.1% during measurement of

time delay, only voltage dependence of loads is considered in

power imbalance estimation in [47] whereby frequency dependence is deemed negligible. Turbine output is assumed

to be constant due to slow response of mechanical turbine

valves controlled by a turbine governor as compared to rate

of frequency decay [2]:

PL,j=PL0,j× (Vj

V0,j)

αj

(7)

QL,j

=QL0,j

× (Vj

V0,j)

βj

(8)

Hence, for an exponential load model, the amount of active

power changing of loads due to voltage fluctuation is:

Pdeficit= ∑2Hi

fn

dfi

dt+

Ng

i=1∑ PL0,j× [(

Vj

V0,j)

αj

-1]NL

i=1 (9)

The polynomial model, commonly known as the

"ZIP model" consists of constant impedance (Z), constant

current (I) and constant power (P) properties. This model can

be represented as:

PL,j=PL0,j× [p1(

Vj

V0,j)

2

+p2(

Vj

V0,j) +p

3] (1+Kpf∆f) (10)

QL,j

=QL0,j

× [q1(

Vj

V0,j)

2

+q2(

Vj

V0,j) +q

3] (1+Kqf∆f) (11)

where p1, p2 and p3 are the load parameters for constant

impedance, constant current and constant power. The

summation of p1, p2 and p3 is equivalent to one. Linearizing Equation (10), imbalance active power supplied by load at

bus j is determined as:

PL,j=PL0,j× [(2p1+p

2)(

Vj

V0,j)] (1+Kpf∆f) (12)



466

whereby ∆Vj = Vj − V0,j. As voltage variation is believed to be

much faster and larger than the change of frequency whereby

frequency does not decay for more than 0.1% during

measurement of time delay, only voltage dependence of loads is

considered in power imbalance estimation in whereby frequency

dependence is deemed negligible. Therefore, the total active

power imbalance for a polynomial model is:

Pdeficit= ∑2Hi

fn

dfi

dt+

Ng

i=1

∑ PL0,j× [(2p1+p

2)(

Vj

V0,j

)]

NL

i=1

=2Hc

fc

dfcoi

dt+ ∑ PL0,j× [(2p

1+p

2)(

Vj

V0,j)]NL

i=1 (13)

In [46], the EPRI static load model was used to represent

voltage and frequency dependence of this type of load

appropriately:

PL=P0 [Pα1 (V

V0)

kpv1

(1+Kpf1∆f)+(1-Pα1) (V

V0)

kpv2

] (14)

QL=P0 [Q

α1(

V

V0)

kqv1

(1+Kqf1∆f)+ (Q0

P0-Q

α1) (

V

V0)

kqv2

(1+Kqf2∆f)]

(15)

wherePα1 was the frequency dependent fraction of real load and

Qα1 was the reactive load coefficient of uncompensated reactive

load to real power load. kpv1andkpv2 were the voltage exponents

for frequency-dependent and frequency-independent real power

load whereas kqv1 and kqv2 were the voltage exponents for

uncompensated and compensated reactive power load. Kpf1was

the frequency sensitivity coefficient for real power load whereas

Kqf1 and Kqf2 were the frequency sensitivity coefficient for

uncompensated and compensated reactive power load. Hence, the

total active power imbalance for the model is:

Pdeficit= ∑2Hi

fn

dfi

dt+

Ng

i=1∑ PL0,j× [Pα1 (

V

V0)

kpv1

(1+Kpf1∆f)+(1-NL

i=1

Pα1) (V

V0)

kpv2

] (16)

Two commonly known types of dynamic loads are induction

motor loads and synchronous motor loads. The dynamics of loads

are usually modelled using EPRI LOADSYN program in studies

of inter-area oscillations and voltage stability or when running

studies on systems with large concentration of motors. The

program converts data on the dynamic aspects of load

components into the form required for stability studies. Dynamic

aspects of load components include load class mix, dynamics of

motors which consume more than 60% of total energy supplied

by a power system, operation of protective relays, thermostatic

control of loads such as water heaters and refrigerators which

operate longer during low voltages and response of ULTCs on

distribution transformers and voltage regulators.

The dynamic differential state equations were treated as algebraic

constraints in [58] after discretization thereby enabling the

transient behaviour of the system after fault clearing time to be

written in a compact way [58]-[60]. The consumer loads used in

the model were approximated by a resistive fraction of 60 %

and an inductive rotating fraction of 40% of the respective

maximal load. An induction motor load was used to represent

the 40% inductive fraction.

In [42], an equivalent induction motor was modelled as a

dynamic load model [61] at each load bus to represent the

induction motor portion of the load at that bus, in order to

consider the large impact of dynamic loads on the system

voltage stability. Figures15 and 16 show the equivalent

circuit of an induction motor [57] and a synchronous motor

where Rs is the stator resistance, Rr is the rotor resistance, Xr

is the rotor reactance, Xs is the stator reactance, S is the slip

and Xm is the exciting reactance.

Figure 15: Architecture of Sensor Node

Figure 16: Equivalent Circuit of Synchronous Motor

Composite load model consist of a combination of static and

dynamic load models. The equivalent circuit of a composite

model [52] is shown in Figure 17. Table 3 shows the

summary table for computation of power imbalance for

static, dynamic and composite loads. It is observed that

quantum of active power imbalance is affected by type or

characteristics of loads connected to the power system.

Figure 17: Equivalent Circuit of Composite Model

Rr

s

Rs Xs Xr

Xm V Rr

s

Rs Xs Xr

Xm Rc Rr

s

Rr

s Xm

Xr Rs Xs

V Static Model

Rr

s



467

Table 3: Summary on Computation of Power Imbalance for

Static, Dynamic and Composite Loads

Type of Loads Power Imbalance Calculation

Static Exponential

Polynomial

(ZIP)

EPRI

Pdeficit= ∑2Hi

fn

dfi

dt+

Ng

i=1

∑ PL0,j× [(Vj

V0,j

)

αj

-1]

NL

i=1

Pdeficit= ∑2Hi

fn

dfi

dt+

Ng

i=1

∑ PL0,j× [(2p1+p

2) (

Vj

V0,j

)]

NL

i=1

Pdeficit= ∑2Hi

fn

dfi

dt+

Ng

i=1

∑ PL0,j×

[ Pα1 (

V

V0

)kpv1

(1+Kpf1∆f)+

(1-Pα1) (V

V0

)kpv2

] NL

i=1

Dyna

mic

Induction

Motor

Synchronous

Motor

Parameters Computed with EPRI

LOADSYN Program

Parameters Computed with EPRI

LOADSYN Program

Composite Parameters Computed with EPRI

LOADSYN Program

Distribution of Load Shedding In conventional schemes, fixed quantum of loads is shed at pre-

determined locations when frequency drops below frequency

threshold regardless of type of disturbance and all load buses are

involved in sharing of total power imbalance without selection.

This type of load shedding scheme is not practical and lacks

flexibility to execute load shedding fit for different type of

instabilities.

Other form of frequency-related information are also used to

determine distribution of load shedding for instance rate of

frequency change, df/dt and average rate of frequency change, Δf,

although the latter is usually used [53] due to the oscillatory

nature of the former which may provide misleading information on frequency variation during system contingencies. In [62], the

quantum of active power loads to be shed at jth load bus was

estimated as follows:

∆PLj=∆fLj∙PL0,j

∑ (∆fLj∙PL0,j)Mj=1

∙∆P (17)

where∆PL0,j is the quantum of active power load on jth bus before

disturbance, ∆fLj is the frequency deviation at jth load bus

compared to nominal frequency and M is the number of load

buses.

Load shedding distribution relying solely on frequency

information is inadequate because frequency decline during major

contingency may also lead to voltage instability. The concept of centralized adaptive UFLS algorithm was proposed in [6] and [7]

to mitigate cascading events due to frequency and voltage

instability following combinational disturbances. The authors

asserted that power systems were prone to cascading events due

to high dependence of loads on voltage, low frequency settings of

UFLS relays and inappropriate selection of quantum and location

of load shed. The authors proposed two algorithms to determine

the distribution of load shedding namely magnitude of sub-

transmission bus voltages and static voltage stability V-Q margins

of buses. Simulation studies performed on the Khorasan’s

network within the Iran interconnected grid showed that

these algorithms could effectively preserve system stability

following severe contingencies and successfully optimize the

quantum of load shed.

In [8], two combinatorial algorithms to combine UFLS and UVLS schemes were presented whereby parameters such as

voltage sensitivity of loads, voltage profile and location of

disturbance, determined from locally measured frequency

and voltage signals were used to determine distribution of

load curtailment during severe disturbances. These methods

known as V-F and dV-F load shedding schemes, was proven

to increase adaptability of UFLS relay and enhance power

system susceptibility to voltage collapse by improving

voltage stability margins especially for vulnerable points of

the system whereby loads with greater voltage decline and

higher reactive power demand were shed sooner.

With the V-F method, loads were classified into different steps according to the measured voltage level following

disturbance whereby loads with lower voltage and more

reactive power shortage were shed sooner. The scheme

comprised of a logical AND combination of voltage and

frequency blocks which was activated when the measured

voltage and frequency at the relay location were lesser than

predetermined thresholds of the voltage and frequency

blocks.

In the dV-F scheme, the amount of voltage decline was used

as a criterion in the voltage blocks instead of the voltage

level. The amount of voltage decline could address location of disturbance and reactive power deficit better than voltage

level, because the effect of pre-disturbance voltage level was

excluded in this method. Loads with greater voltage decline

or more reactive power shortage were shed sooner and

therefore selected locations of load shedding became more

dependent to the location of disturbance.

In [63], distribution of load shedding for an adaptive scheme

designed for a system connected with distributed generation

was done through a distribution management system (DMS)

which provided fast and optimal load management by

utilizing system topology, DG power generation; load demand and actual operating conditions. The quantum of

loads to be disconnected was evaluated according to real-time

loading of the feeder to be cut off and the total should meet

the calculated load shedding amount required.

An improved method of load-shedding model based on

contribution factors and distribution factors between loads

and generators in power flow tracing was proposed in [64]

whereby nodes with larger contribution factors and

distribution factors are selected to be adjusted to ensure that

the adjustment on the overload lines is the most effective, while having little impact on the non-overload lines. This

method decreased the number of the nodes to be regulated,

which in turn improved calculation speed and accuracy. The

contribution factor, αg referred to the share of the power of

generator contributing to the target line ratio of overloaded

line whereas distribution factor, αf represented the share of

the power injections of generation buses to the target line

ratio of generator power:



468

αf=PGi→l

PGi (18)

αg=PGi→l

Pl (19)

wherePGi→l was the power of ith generator that carries inflow to

line l, PGi was the power of generator i and Pl was the power flow

of line l. The suggested method could be useful in the risk

assessment of generation and transmission system for regulating

power flow.

Liu, Zhang and Yu in [12] proposed that load shedding should be

started from the locations encountering higher voltage decay and for longer period of time. The speed, location, and amount of load

shedding have to be changed adaptively depending on the

disturbance location, voltage status of the system, and the rate of

frequency decline. Operation of the conventional and the

proposed adaptive combinatorial load shedding methods were

simulated in an actual large network and simulation results

confirmed that the proposed methods provide considerable

enhancement in the power system voltage stability margin, and by

using the proposed algorithms, various power system blackouts

could be prevented.

A WAMS-based load shedding scheme was proposed in [65] to

mitigate combinatorial and severe contingencies which can jeopardize both frequency and voltage stability. The quantum of

load shed was calculated based on computed disturbance power

and voltage stability conditions of the power system with the aid

of real-time data taken from synchrophasor-based wide area

monitoring and control system (WAMCS). Frequency stability

analysis was carried out based on the low-order SFR model in

[37] for estimating the magnitude and subsequent classification of

the disturbance. Voltage stability was determined based on a

dynamic voltage stability criterion, formulated using Voltage

Stability Risk Index (VSRI) and load curtailments on load buses

were done according to this index:

VSRIj= {

∑(dk+d

k-1)

2j

j

k=1 , j=1, 2, …, N if j≤N)

∑(dk+d

k-1)∆t/2N

2j

j

k=j-N+1, j=N+1,…, M if j>N

(20)

whereby moving average value of load bus voltage at jth instant

from N available PMU measurements was computed from:

vj= {∑ y

k/j

j

k=1 , j=1, 2,…,M if j≤N

∑ yk/N

j

k=1 , j=N+1, …,M if j>N (21)

versity between voltage measured at jth instant was taken to be:

dj= (yj-vj) ×

100

vj, j=1, 2, …, M (22)

Load buses with a smaller negative index has the higher risk of

the voltage instability, hence, would be shed earlier. Peak values

of the VSRIs at buses indicated the highest voltage deviation from

nominal values and would converge to zero during steady-state.

Distribution of load shedding was determined as follows:

∆Pj= {

∆vj

∑ (∆vj)Mj=1

×∆P, for large disturbances

VSRIj

∑ (VSRIj)Mj=1

×∆P, for small disturbances (23)

In [48], Tang et.al. identified two setbacks in existing load

shedding distribution methods namely negligence of reactive

power and voltage stability issues leading to ineffective load

shedding. To address these setbacks, the authors proposed the

use of two indices – the load shedding distribution factor for

active power (LSDFP) and load shedding distribution factor

for reactive power (LSDFQ) which represented the active and

reactive load imbalance. In the calculation of these indices,

power flow tracing method was used to estimate the initial

loading of each load buses:

LSDFPj=∆fLj∙Ptracing,j

∑ (Mj=1 ∆fLj∙Ptracing,j)

(24)

LSDFQj=

VQSj∙Qtracing,j

∑ (Mj=1 VQSj ∙Qtracing,j)

(25)

where ∆fLj was the frequency deviation to the rated

frequency of the jth load bus, VQSj was the sensitivity of

voltage variation to reactive power of the jth load bus, Ptracing,j

and Qtracing,j

were the tracing active and reactive power from

the generators and lines to load buses. Hence, the distribution

of active and reactive power load shed was computed as:

∆PLj=∆Pimprovement ∙LSDFPj (26)

∆QLj

=∆Qimprovement

∙LSDFQj (27)

where∆Pimprovement and ∆Qimprovement

were total active and

reactive power deficit for voltage-dependent load models.

The new method was tested on the IEEE 39-Bus system and

simulation results showed improvement in terms of

frequency and voltage stability and loadability. Table 4

shows the summary of literature study done on the

distribution of load shedding for UFLS.

Table 4: Literature Study on Distribution of Load Shedding

for UFLS

Year Distribution of Load Shedding

Early Days

2005 2007

2008

2009

2011

2013

Fixed quantum of load shed

Based on average rate of frequency change Based on V-Q margin

Based on V-F and dV-F

Distribution Management System, Based on

disturbance location, voltage status, rate of

frequency decline

Power flow tracing using contribution and

distribution factors, Frequency stability analysis

and voltage stability analysis (VSRI)

Power flow tracing using LSDFP and LSDFQ



469

Conclusion UFLS scheme is a popular mitigation method used to arrest a

power system from load-generation imbalance in various parts of

the world such as Asia, Europe, Australasia, Africa, Middle East and the Americas. UFLS in these continents differ in terms of

total load shed, number of UFLS blocks, average block size and

trip frequency deviation thresholds depending on their system

size, system inertia and generation mix.

The SFR model has been widely used in computation of load-

frequency response of a power system during system

contingency. However in the classic SFR model, only inertia

constant of generator and frequency information are considered

and the impact of voltage dependence of loads is not taken into

account although load characteristics have been proven to have

significant influence on the dynamic behaviour of power systems during low frequency oscillation and severe faults. In more recent

literature, SFR model incorporating frequency and voltage

dependence load models is proposed and used in the design of

optimal UFLS scheme.

UFLS operating philosophies consisting of three important areas

namely trigger criteria, load characteristics and load shedding

distribution have always been based solely on frequency

parameters in the early days of implementation. Load shedding

trigger based solely on frequency measurements is inadequate to

determine the stability and “health” of a power system especially

after islanding of an interconnected power system as an islanded

power system following severe disturbances may experience low voltage with high frequency when an area being split from the

main grid has large generation deficit. The power system may be

susceptible to voltage collapse as well, which will lead to total

system blackout within shorter time duration as compared to a

frequency collapse phenomenon. Hence, trigger condition

considering voltage information and voltage stability criterions

are introduced and implemented in UFLS schemes.

Distribution of load shedding affects the minimum frequency

deviation from nominal and steady-state frequency recovery

following disturbances. In the early days of UFLS

implementation, fixed quantum of loads is shed at pre-determined locations based solely on system frequency information

regardless of type of disturbance and all load buses are involved

in sharing of total power imbalance without selection.

Improvement in this area is introduced over the years to cater for

combined frequency and voltage instabilities. In recent years,

distribution of load shedding based on power flow tracing method

is used. This method caters for both frequency and voltage

instabilities and has been proven to give optimal system response.

Acknowledgement The authors would like to thank Universiti Malaysia Sarawak

who has funded this research work.

References

[1] A. A. MohdZin, Mohd Hafiz, and M. S. Aziz, “A Review

of Under-frequency Load Shedding Scheme on TNB

System”, National Power and Energy Conference, 2004.

[2] Power Systems Relaying Committee, IEEE Power

Engineering Society, “IEEE Guide for the Application of

Protective Relays Used for Abnormal Frequency Load

Shedding and Restoration”, IEEE Standard C37.117-

2007, 2007.

[3] D. H. A. Lee, “A General-Order System Frequency

Response Modelincorporating Load Shedding:

Analytic Modelling and Applications”, IEEE

Transactions on Power System, vol. 21, no.2, pp. 709

– 717, 2006.

[4] B. Delfino, S. Massucco, A. Morini, P. Scalera, and F.

Silvestro, “Implementation and comparison of

different under-frequency load shedding scheme”,

Power Engineering Society Summer Meeting 2001,

2001.

[5] V. V. Terzija, “Adaptive Underfrequency Load

Shedding Based on the Magnitude of the Disturbance

Estimation”, IEEE Transactions on Power Systems,

vol. 21, no.3, pp. 1260-1266, 2006.

[6] M. S. Pasand, and H. Seyedi, “New Centralized

Adaptive Under Frequency Load Shedding

Algorithms”, 2007 Large Engineering Systems

Conference on Power Engineering, 2007.

[7] H. Seyedi, and M. S. Pasand, “New Centralised

Adaptive Load-Shedding Algorithms to Mitigate

Power System Blackouts”, IET Generation,

Transmission & Distribution, vol. 3, no. 1, pp. 99-114,

2009.

[8] A. Saffarian, M. S. Pasand, and H. Asadi,

“Performance Investigation of New Combinational

Load Shedding Schemes”, Joint International

Conference on Power System Technology and 2008

IEEE Power India Conference (POWERCON 2008),

2008.

[9] Y. Welhazi, T. Guesmi, I. B. Jaoued, and H. H.

Abdallah, “Power System Stability Enhancement

Using FACTS Controllers in Multimachine Power

Systems”, Journal of Electrical Systems, vol. 10, no. 3,

pp. 276-291, 2014.

[10] R. Z. Davarani, and R. Ghazi, “Optimal Simultaneous

Coordination of PSS and TCSSC using Multi

Objective Genetic Algorithm”, Journal of Electrical

Systems, vol. 9, no. 4, pp. 410-421, 2013.

[11] M. V. Suganyadevi, and C. K. Babulal, “Fast

Assessment of Voltage Stability Margin of a Power

System”, Journal of Electrical Systems, vol. 20, no. 3,

pp. 305-316, 2014.

[12] Y. Liu, W. Zhang, and Z. Yu, “Measures to Keep

Islands Stable after Power Oscillation or Out of Step

Islanding”, 8th International Conference on Advances

in Power System Control, Operation and Management

(APSCOM 2009), 2009.

[13] A. Saffarian, and M. S. Pasand, “Enhancement of

Power System Stability Using Adaptive

Combinational Load Shedding Methods”, IEEE

Transactions on Power Systems, vol. 26, no. 3, pp.

1010-1020, 2011.



470

[14] A. P. Ghaleh, M. S. Pasand, and A. Saffarian, “Power

System Stability Enhancement using a New Combinational

Load-Shedding Algorithm”, IET Generation, Transmission

& Distribution, vol. 5, no. 5, pp. 551-560, 2011.

[15] Z. Zhang, and Z. C. X. Wang, “Study on Emergency

Control Based on Regional Frequency Response Model”,

2011 The International Conference on Advanced Power

System Automation and Protection, 2011.

[16] H. Samet, and M. Rashidi, “Coordinated Under Frequency

Load and Capacitor Shedding for Bulk Power Systems”,

IET Generation, Transmission & Distribution, vol. 7, no. 8,

pp. 799-812, 2013.

[17] A. Sauhats, V. Chuvychin, V. Strelkovs, R. Petrichenko,

and E. Antonov, “Underfrequency Load Shedding in Large

Interconnection”, IEEE Grenoble Power Tech

(POWERTECH), 2013.

[18] Y-W. Au, and K-M. Yung, “Emergency Control and

Restoration of Power System under Disturbance”,

Proceedings of the 4th International Conference on

Advances in Power System Control, Operation and

Management (APSCOM-97), 1997.

[19] L. E. Ling, and N. M. Vong, “Special Load Shedding

Scheme to Enhance System Stability of Small Power

System”, 2006 International Conference on Power System

Technology, 2006.

[20] W. Pimjaipong, T. Junrussameevilai, and N. Maneerat,

“Blackout Prevention Plan – The Stability, Reliability and

Security Enhancement in Thailand Power Grid”, 2005

IEEE/PES Transmission and Distribution Conference and

Exhibition: Asia and Pacific, 2005.

[21] S. Imai, and T. Yasuda, “UFLS Program to Ensure Stable

Island Operation”, IEEE PES Power Systems Conference

and Exposition, 2004.

[22] S.-J. Huang, and C.-C. Huang, “An Automatic Load

Shedding Scheme Including Pumped-Storage Units”, IEEE

Transactions on Energy Conversion, vo. 15, no. 4, pp. 427-

432, 2000.

[23] L. E. Ling, V. Wong, and N. M. Vong, “Wide Area

Protection (WAP) as an Enhanced Load Shedding Scheme

to Mitigate Loss of Supply to Load Center”, Conference of

the Electric Power Supply Industry (CEPSI) 2010, 2010.

[24] Y. Lu, W-S. Kao, and Y-T. Chen, “Study of Applying

Load Shedding Scheme with Dynamic D-Factor Values of

Various Dynamic Load Models to Taiwan Power System”,

IEEE Transactions on Power Systems, vol. 20, no. 4, pp.

1976-1984, 2005.

[25] Midwest Reliability Organization, “Midwest Reliability

Organization Under-frequency Load Shedding Program

Report and Recommendations”, 2007.

[26] K. U. Z. Mollah, M. Bahadornejad, and N. K. –C. Nair,

“Automatic Under-Frequency Load Shedding in New

Zealand Power System – A Systematic Review”, 21st

Australasian Universities Power Engineering Conference

(AUPEC), 2011.

[27] U. Rudez, and R. Mihalic, “Dynamic Analysis of

Transition into Island Conditions of Slovenian Power

System Applying Underfreuqency Load Shedding

Scheme”, 2009 IEEE Bucharest Power Tech

Conference, 2009.

[28] S. Lindahl, G. Runvik, and G. Stranne, “Operational

Experience of Load Shedding and New Requirements

on Frequency Relays”, Developments in Power

System Protection, 1997.

[29] R. V. Fernandes, S. A. B. de Almeida, F. P. Maciel

Barbosa, and RuiPestana, “Load Shedding –

Coordination between the Portuguese Transmission

Grid and the Distribution Grid with Minimization of

Loss of Distributed Generation”, 2009 IEEE Bucharest

Power Tech Conference, 2009.

[30] WECC, “WECC Coordinated Off-Nominal Frequency

Load Shedding and Restoration Requirements”, 2003.

[31] H. H. Wong, J. C. Flores, Y. Fang, and R. P. Baldevia

Jr., “Guam Power Authority Automatic

Underfrequency Load Shedding Study”, International

Conference on Energy Management and Power

Delivery, 1995.

[32] FRCC, Florida Reliability Coordinating Council,

“FRCC Handbook”, 2008.

[33] MAAC, Mid Atlantic Area Council, “MAAC

Protective Relaying Philosophy and Design

Standards”, 1997.

[34] Y. R. Omar, I. Z. Abidin, S. Yusof, H. Hashim, H. A.

Abdul Rashid, “Under Frequency Load Shedding

(UFLS): Principles and Implementation”, Proceedings

of 2010 IEEE International Conference on Power and

Energy, 2010.

[35] R. M. Maliszewski, R. D. Dunlop, and G. L. Wilson,

“Frequency Actuated Load Shedding and Restoration

Part I – Philosophy”, IEEE Transactions on Power

Apparatus and Systems, vol. PAS-90, no. 4, pp. 1452-

1459, 1971.

[36] M. L. Chan, R. D. Dunlop, and F. Schweppe,

“Dynamic Equivalents for Average System Frequency

Behavior Following Major Disturbances”, IEEE

Transactions on Power Apparatus and Systems, vol.

PAS-91, no. 4, pp. 1637-1642, 1972.

[37] P. M. Anderson, and M. Mirheydar, “A Low-Order

System Frequency Response Model”, IEEE

Transactions on Power Systems, vol. 5, no. 3, pp. 720-

729, 1990.

[38] P. M. Anderson, and M. Mirheydar, “An Adaptive

Method for Setting Underfrequency Load Shedding

Relays”, IEEE Transactions on Power Systems, vol. 7,

no. 2, pp. 647-655, 1992.

[39] Y. Halevi, and D. Kottick, “Optimization of Load

Shedding System”, IEEE Transactions on Energy

Conversion, vol. 8, no. 2, pp. 207-213, 1993.

[40] V. Chin, Z. Y. Dong, T. K. Saha, J. Ford, and J.

Zhang, “Adaptive and Optimal Under Frequency Load



471

Shedding”, 2008 Australasian Universities Power

Engineering Conference (AUPEC’08), 2008.

[41] C. Wu, L. Gao, and Y. Dai, “Simulation and Optimization

of Load Shedding Scheme for Islanded Power System”,

2010 International Conference on Power System

Technology, 2010.

[42] H. Y. Yu, and Z. Y. Dong, “Optimal Decentralized Load

Curtailment by utilizing NMPA with Distributed

Generations”, IEEE PES ISGT ASIA 2012, 2012.

[43] Y. R. Bhavanam, and M. Shahrazad, “Impact of Increased

Distributed Generation on Fault Levels and the Network

Equipment”, Journal of Electrical Systems, vol. 10, no. 2,

pp. 209-222, 2014.

[44] S. Mirsaeidi, D. M. Said, M. W. Mustafa, M. H.

Habibuddin, and K. Ghaffari, “Review and Analysis of

Existing Protection Strategies for Micro-grids”, Journal of

Electrical Systems, vol. 10, no.1, pp. 1-10, 2014.

[45] S. C. Reddy, P. V. N. Prasad, and A. J. Laxmi, “Hybrid

Approach for Voltage Sag Mitigation in Distribution

System by Optimal Location and Size of Distributed

Generation”, Journal of Electrical Systems, vol. 9, no. 2,

pp. 150-166, 2013.

[46] O. Shariati, A. A. MohdZin, A. Khairuddin, M. Pesaran,

and M. R. Aghamohammadi, “An Integrated Method for

Under Frequency Load Shedding Based on Hybrid

Intelligent System-Part I: Dynamic Simulation”, 2012

Asia-Pacific Power and Energy Engineering Conference

(APPEEC), 2012.

[47] G. Wang, H. Xin, D. Gan, N. Li, and Z. Wang, “An

Investigation into WAMS-based Under-Frequency Load

Shedding”, 2012 IEEE Power and Energy Society General

Meeting, 2012.

[48] J. Tang, J. Liu, F.Ponci, and A. Monti, “Adaptive Load

Shedding Based on Combined Frequency and Voltage

Stability Assessment Using Synchrophasor

Measurements”, IEEE Transactions on Power Systems,

vol. 28, no. 2, pp. 2035-2047, 2013.

[49] H. You, V. Vittal, and Z. Yang, “Self-Healing in Power

Systems: An Approach Using Islanding and Rate of

Frequency Decline-Based Load Shedding”, IEEE

Transactions on Power Systems, vol. 18, no. 1, pp. 174-

181, 2003.

[50] V. V. Terzija, “Adaptive Underfrequency Load Shedding

Based on the Magnitude of the Disturbance Estimation”,

IEEE Transactions on Power Systems, vol. 21, no. 3, pp.

1260-1266, 2006.

[51] L. Sigrist, I. Egido, and L. Rouco, “Principles of a

Centralized UFLS Scheme for Small Isolated Power

Systems”, IEEE Transactions on Power Systems, vol. 28,

no. 2, pp. 1779-1786, 2013.

[52] P. Mahat, Z. Chen, and B. Bak-Jensen, “Underfrequency

Load Shedding for an Islanded Distribution System With

Distributed Generators”, IEEE Transactions on Power

Delivery, 2009.

[53] Power Systems Relaying Committee, IEEE Power

Engineering Society, “IEEE Guide for the Application

of Protective Relays Used for Abnormal Frequency

Load Shedding and Restoration”, 2007.

[54] W. Tan, C. Shen, X. Zhang, and J. Ni, “A New Under-

Frequency Load Shedding Scheme Based on OBDD”,

International Conference on Sustainable Power

Generation and Supply (SUPERGEN ‘09), 2009.

[55] A. Li, and Z. Cai, “A Method for Frequency Dynamics

Analysis and Load Shedding Assessment Based on the

Trajectory of Power System Simulation”, Third

International Conference on Electric Utility

Deregulation and Restructuring and Power

Technologies (DRPT2008), 2008.

[56] P. Du, and J. K. Nelson, “Two-step Solution to

Optimal Load Shedding in a Micro-grid”, IEEE/PES

Power Systems Conference and Exposition (PSCE

’09), 2009.

[57] Y. Lu, W-S. Kao, and Y-T. Chen, “Study of Applying

Load Shedding Scheme with Dynamic D-Factor

Values of Various Dynamic Load Models to Taiwan

Power System”, IEEE Transactions on Power

Systems, vol. 20, no. 4, pp. 1976-1984, 2005.

[58] E. E. Aponte, and J. Keith Nelson, “Time Optimal

Load Shedding for Distributed Power Systems”, IEEE

Transactions on Power Systems, 2000.

[59] E. D. Tuglie, M. Dicorato, M. La Scala, and P.

Scarpellini, “A Corrective Control for Angle and

Voltage Stability Enhancement on the Transient Time-

Scale”, IEEE Transactions on Power Systems, vol. 15,

no. 4, pp. 1345-1353, 2000.

[60] M. La Scala, M. Trovato, and C. Antonelli, “On-line

Dynamic Preventive Control: An Algorithm for

Transient Securirty Dispatch”, IEEE Transactions on

Power Systems, vol. 13, no. 2, pp. 601-610, 1998.

[61] Z. Zhong, “Power Systems Frequency Dynamic

Monitoring System Design and Applications”, PhD.

dissertation, Virginia Polytechnic Institute and State

University Blacksburg, 2005.

[62] Y. Zhao, N. Zho, J. Zhou, K. Xie, Y. Liu, and J.

Kuang, “A Heuristic Approach to Local Load

Shedding Scheme for Reliability Assessment of Bulk

Power Systems”, 2005 IEEE/PES Transmission and

Distribution Conference & Exhibition, 2005.

[63] H. Hirodontis, H. Li, and P. A. Crossley, “Load

Shedding in a Distribution Network”, International

Conference on Sustainable Power Generation and

Supply (SUPERGEN ’09), 2009.

[64] R. Niu, Y.Zeng, M. Cheng, and X. Wang, “Study on

Load-shedding Model Based on Improved Power Flow

Tracing Method in Power System Risk Assessment”,

2011 4th International Conference on Electric Utility

Deregulation and Restructuring and Power

Technologies (DRPT), 2011.



472

[65] K. Seethalekshmi, S. N. Singh, and S. C. Sriastava, “A

Synchrophasor Assisted Frequency and Voltage Stability

Based Load Shedding Scheme for Self-Healing of Power

System”, IEEE Transactions on Smart Grid, 2011.

Under-Frequency Load Shedding (UFLS) Schemes A Survey · E-mail: [email protected] D. H. A....

Documents

Transcript of Under-Frequency Load Shedding (UFLS) Schemes A Survey · E-mail: [email protected] D. H. A....