950 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 35, NO.

2, MARCH 2020

Continuous Under-Frequency Load Shedding Scheme

for Power System Adaptive Frequency Control
Changgang Li , Member, IEEE, Yue Wu, Yanli Sun, Hengxu Zhang , Member, IEEE,
Yutian Liu , Senior Member, IEEE, Yilu Liu , Fellow, IEEE, and Vladimir Terzija , Fellow, IEEE

Abstract—Frequency drop due to loss of massive generation is a to recover frequency in under-frequency scenarios is under-
threat to power system frequency stability. Under-frequency load frequency load shedding (UFLS) which rebalances generation
shedding (UFLS) is the principal measure to prevent successive and load by shedding appropriate amount of loads.
frequency declination and blackouts. Based on traditional stage-by-
stage UFLS scheme, a new continuous UFLS scheme is proposed UFLS can be generally classified into three categories: tra-
in this paper to shed loads proportional to frequency deviation. ditional stage-by-stage scheme, adaptive scheme, and semi-
The characteristic of the proposed scheme is analyzed with a adaptive scheme [6]. The major difference between the three
closed-form solution of frequency dynamics. Frequency threshold schemes lies in the method to determine load shedding amount.
and time delay are added to make the proposed scheme practical. For the stage-by-stage scheme, the loads to shed are predefined
A line-by-line scheme based on precise load control is introduced
to implement the continuous scheme for systems without enough and divided into several stages. Each stage is tripped when
continuously controllable loads. The load shedding scale factor of frequency drops beyond a certain threshold for a few seconds.
the proposed scheme is tuned with an analytical method to achieve It sheds loads based on local frequency and is widely used in
adaptability to different operating conditions. The adaptability of utilities [7]. Difference between the frequency thresholds of
the proposed scheme is validated with 39-bus New England model stages must be set to overcome oscillation between generators.
and simplified Shandong Power Grid of China.
The typical frequency difference between stages is 0.2∼0.25 Hz,
Index Terms—Power systems, frequency stability, under- and 5∼8 stages are usually set. The time delay is adaptively
frequency load shedding (UFLS), continuous control, precise load adjusted according to frequency dynamics in [8] to improve
control.
scheme adaptability. In [9] and [10], the stage-by-stage scheme
is optimized with trigger signal of both frequency deviation and
I. INTRODUCTION rate of change of frequency (ROCOF). Uncertainty of system
REQUENCY is an essential index of balance between parameters is considered in [11] to tune more adaptive scheme.
F active power generation and load. Frequency will drop if
active power generation is inadequate due to loss of generation in
Though the stage-by-stage scheme can be tuned for better adapt-
ability, the limited stages and inherent discreteness make it hard
large-scale power systems [1] or microgrids [2]. Large frequency to set up a scheme suitable for all possible operating condi-
excursion may lead to system blackouts and restoration failure tions. Over-shedding or under-shedding problem can hardly be
[3]. Systems with more integration of intermittent renewable avoided for the stage-by-stage scheme [12].
generation are more vulnerable to frequency deviation [4], [5]. For the adaptive scheme, the loads to shed are calculated
It is compulsory to arrest frequency declination for preventing online based on ROCOF of the center of inertia (COI). It is
power systems from severe consequences. The key technique derived from the fact that the initial ROCOF of COI is pro-
portional to power imbalance [13]. Therefore, loads equal to
the product of ROCOF and system inertia are shed to balance
Manuscript received October 20, 2018; revised April 19, 2019 and July 21, generation and load in one shot. In [14], load shedding amount
2019; accepted September 5, 2019. Date of publication September 26, 2019;
date of current version February 26, 2020. This work was supported in part by is distributed to different loads considering voltage stability
National Key R&D Program of China under Grant 2017YFB0902600, in part and power tracing criteria. The adaptive scheme is improved
by National Science Foundation of China under Grant 51407107, and in part by in [15] considering change of power generation during load
Young Scholars Program of Shandong University under Grant 2018WLJH31.
Paper no. TPWRS-01600-2018. (Corresponding author: Yutian Liu.) shedding process. The difference between measured ROCOF
C. Li, Y. Wu, Y. Sun, H. Zhang, and Y. Liu are with the Key Laboratory and threshold of ROCOF is adopted in [16] for determining
of Power System Intelligent Dispatch and Control of the Ministry of Educa- appropriate load shedding amount. The calculation of ROCOF
tion, Shandong University, Jinan 250061, China (e-mail: [email protected].
cn; [email protected]; [email protected]; [email protected]; of COI requires wide area measurement data from different
[email protected]). locations. Communication channels with high availability are
Yilu Liu is with the Department of Electrical Engineering and Computer Sci- prerequisite of the adaptive scheme for estimating power imbal-
ence, University of Tennessee, Knoxville, TN 37996 USA (e-mail: [email protected]).
V. Terzija is with the School of Electrical and Electronic Engineer- ance and sending load shedding command. Biased estimation
ing, The University of Manchester, Manchester, M60 1QD, U.K. (e-mail: of ROCOF or system inertia further prevents it from industrial
[email protected]). applications.
Color versions of one or more of the figures in this article are available online
at http://ieeexplore.ieee.org. Semi-adaptive scheme adopts both ROCOF and frequency
Digital Object Identifier 10.1109/TPWRS.2019.2943150 deviation to determine loads to shed [17]. It can be treated as a

0885-8950 © 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See https://www.ieee.org/publications/rights/index.html for more information.

Authorized licensed use limited to: National Institute of Technology Andhra Pradesh. Downloaded on September 15,2022 at 04:34:13 UTC from IEEE Xplore. Restrictions apply.
 LI et al.: CONTINUOUS UFLS SCHEME FOR POWER SYSTEM ADAPTIVE FREQUENCY CONTROL 951

Fig. 2. Modified SFR model with continuous UFLS scheme.

Fig. 1. Traditional stage-by-stage schemes with different stages and continu-
ous UFLS scheme.
where PLS is load shedding amount and Δf is frequency de-
viation. One is the solid line with five stages, and the other is
combination of the stage-by-stage scheme and adaptive scheme. the dot-dashed line with 12 stages. Scheme with more stages
In the first stage, loads to shed are calculated based on measured is more adaptive to different operating conditions than schemes
ROCOF of COI, like the adaptive scheme. For other stages, with fewer stages [9]. If the number of stages is increased to
loads are predefined and shed according to frequency deviation. infinite, the load shedding curve will become continuous as the
The semi-adaptive scheme is not immune to over-shedding or dashed line in Fig. 1.
under-shedding problem due to the similarly limited stages as In the continuous UFLS scheme, the load shedding amount
the stage-by-stage scheme. is determined online with frequency deviation as:
A new continuous UFLS scheme is proposed in this paper PLS = −KLS Δf (1)
to adaptively determine loads to shed while overcoming the
drawbacks of existing schemes. The main contributions of this where KLS is load shedding scale factor to tune.
paper are three-fold. Firstly, the concept of continuous UFLS is Equation (1) shows the basic idea of continuous UFLS scheme
proposed, and it is proved analytically that the proposed scheme in which loads are shed continuously. With loads continuously
is adaptive to different operating conditions and events, and im- shed with (1), it is supposed to be adaptive to different operating
mune to the over-shedding problem. Secondly, non-linear factors conditions and is analyzed as follows.
of frequency threshold and time delay are considered to improve
the continuous UFLS scheme for applicability. A line-by-line B. SFR Model With Continuous UFLS Scheme
scheme with precise load control is proposed to implement the System frequency is complicated in real power systems. To
continuous UFLS scheme. Thirdly, a tuning method is proposed investigate the general performance of the proposed continuous
to set up the load shedding scale factor of the continuous UFLS UFLS scheme, SFR model proposed in [18] is adopted in this
scheme. The proposed continuous UFLS scheme and tuning paper. Method to get the equivalent parameters for SFR can be
method are verified to be valid to bring system frequency back. found in [19]. It should be noted that the SFR model is simplified
The rest of this paper is organized as follows. The basic to examine the frequency dynamics of a system dominated by
idea of the continuous UFLS scheme is proposed in Section II, steam-turbine generators. Complex dynamics such as limited
and its property is analyzed with system frequency response spinning reserve and boiler dynamics are neglected. The results
(SFR) model. Impact of such nonlinear factors as spinning of the SFR model can only be used to check the general perfor-
reserve, frequency threshold, and time delay are analyzed in mance of the proposed UFLS scheme.
Section III, and the continuous UFLS scheme is improved with Fig. 2 shows the diagram of SFR model, where PD is the
frequency threshold and time delay. Practical implementation of amount of power imbalance due to event, H is system inertia,
the proposed scheme with precise load control is also discussed D is system damping consisting of both damping of generator
in Section III. A method to tune the proposed continuous UFLS and load frequency dependency, R is droop of governor, FH is
scheme is introduced in Section IV. With the 39-bus New Eng- the fraction of mechanical power generated by high-pressure
land model and simplified Shandong Power Grid of China, the cylinder, TR is time constant of reheater, and Km is gain of
performance of the continuous UFLS scheme is further verified mechanical power. The continuous UFLS scheme is added in
in Section V with different events and operating conditions. the dashed box of Fig. 2.
Discussions on the limitation and possible applications of the Though KLS works as feedback control and looks like the load
proposed scheme are made in Section VI, and conclusions are frequency control or load frequency dependency, it is different
drawn in Section VII. from the other two. The major difference lies in the fact that the
loads shed with KLS are disconnected by tripping transmission
II. THE BASIS OF CONTINUOUS UFLS SCHEME lines and will not be immediately restored when the frequency is
recovered. The dispatching command to recover the shed loads
A. From Discrete Scheme to Continuous Scheme
should be given according to guidelines only when the system
Some methods were proposed to increase the number of stages is successfully recovered from the severe disturbance.
for a smoother stage-by-stage scheme and better performance. Based on the analysis made in [18], the system frequency
Ignoring oscillation, Fig. 1 shows two stage-by-stage schemes deviation before reaching its nadir can be expressed as the

Authorized licensed use limited to: National Institute of Technology Andhra Pradesh. Downloaded on September 15,2022 at 04:34:13 UTC from IEEE Xplore. Restrictions apply.
 952 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 35, NO. 2, MARCH 2020

following frequency-domain equation: where M is a constant scale factor between Δfmin and PD :
 λ1
−RPD ωn2 1 + TR s R c1 −c1 λ2 −λ1
Δf (s) = (2) M= − −1
De (s2 + 2δωn s + ωn2 ) s De λ1 c2
where De is equivalent damping, ω n is natural oscillation fre-  λ2 
quency and δ is damping ratio. De , ω n , and δ are defined as: c2 −c1 λ2 −λ1
+ −1 (9)
⎧ λ2 c2
⎪
⎪ De = Km + (D + KLS ) R
⎨ Therefore, the total load shedding amount is:
ωn2 = 2HRT
De
(3)
⎪
⎪
R
PΣLS = −KLS Δfmin = −KLS M PD = CPD (10)
⎩ δ = 2HR+(D+KLS )RTR +Km FH TR ω
2De n
where C is:
In [18], the analytic time-domain expression of Δf was  λ1

built as two equations separately depending on δ. A uniform R c1 −c1 λ2 −λ1

C= −1
expression of system frequency is derived in this paper to make De λ1 c2
the model consistent as:  λ2 
     c2 −c1 λ2 −λ1
−RPD c1 eλ1 t − 1 c 2 e λ2 t − 1 + −1 KLS (11)
Δf (t) = + (4) λ2 c2
De λ1 λ2
C is a parameter determined by system configuration and
where KLS . It is independent of disturbance and is a constant for a
certain system with specific KLS .
λ1 = −δωn − ωn δ 2 − 1, λ2 = −δωn + ωn δ 2 − 1 (5)
It can be concluded from (10) and (11) that the total shed load
ωn2 (1 + TR λ1 ) ω 2 (1 + TR λ2 ) is proportional to the initial power imbalance. When system
c1 = , c2 = n (6) parameters and KLS are fixed, the continuous scheme will shed
λ1 − λ2 λ2 − λ1
loads proportional to the power imbalance, i.e., the continuous
If the system is over-damped with δ > 1.0, λ1 and λ2 are scheme is adaptive to different events.
real numbers and the system frequency response can be readily 2) Load shedding amount is always less than power imbal-
obtained. If the system is under-damped with δ < 1.0, λ1 and λ2 ance and C < 1.0. The continuous scheme is immune to
become complex conjugate numbers. It can be proved that the over-shedding problem.
two parts within the bracket of (4) are conjugate of each other. As discussed in (7), the derivative of Δf over t is 0 at tmin .
Therefore, equation (4) always give real frequency response in Therefore, the swing equation at tmin can be expressed as:
both under and over damped cases. 
When frequency reaches its minimum, the derivative of Δf dΔf  1
= (−PD + ΔPG + PΣLS − DΔfmin ) = 0
over t is 0. Therefore, it can be derived from (4) that the time dt tmin 2H
when frequency reaches its minimum is; (12)
and the expression of total load shedding amount is:
1 1 + TR λ1
tmin = ln (7) PΣLS = PD − ΔPG + DΔfmin (13)
λ2 − λ1 1 + TR λ2
where ΔPG is the change of mechanical power due to primary
C. Characteristic of Continuous UFLS Scheme frequency regulation.
With the frequency response of (4), some characteristic of the Since Δf is negative and ΔPG is positive in under-frequency
proposed continuous UFLS scheme can be found as follows. scenarios, it can be concluded that:
1) Load shedding amount is proportional to initial power im- PΣLS < PD (14)
balance, and the continuous scheme is adaptive to different
events. It indicates that the total shed load with the continuous UFLS
The total amount of shed load is the product of KLS and Δfmin , scheme is always less than power imbalance. In other words,
i.e., the minimum frequency at tmin . Δfmin can be expressed as: there is no over-shedding problem for the continuous UFLS
     scheme. It can also be concluded from (10) that the constant
RPD c1 eλ1 tmin − 1 c2 eλ2 tmin − 1 C is always less than 1.0.
Δfmin = − +
De λ1 λ2 Simulations are carried out on the SFR model to investigate
 the impact of KLS on load shedding amount with parameters
λ1
RPD c1 −c1 λ2 −λ1 from [18], i.e., H = 4 s, TR = 8 s, D = 1, FH = 0.3, Km = 0.95,
=− −1 R = 0.05. With an event of PD = 0.2, the relationship between
De λ1 c2
C and KLS is shown in Fig. 3.
 λ2 
c2 −c1 λ2 −λ1 It can be verified from Fig. 3 that the total load shedding
+ −1 = M PD (8) amount is increasing with KLS . Saturation can be observed
λ2 c2
in Fig. 3 that, C increases quickly when KLS increases in the

Authorized licensed use limited to: National Institute of Technology Andhra Pradesh. Downloaded on September 15,2022 at 04:34:13 UTC from IEEE Xplore. Restrictions apply.
 LI et al.: CONTINUOUS UFLS SCHEME FOR POWER SYSTEM ADAPTIVE FREQUENCY CONTROL 953

Fig. 3. The relationship between load shedding factor C and KLS . Fig. 5. Frequency response with different frequency thresholds.

Comparing cases when the spinning reserve is 5% and 10%, it

can be found that more spinning reserve does not necessarily lead
to less frequency deviation due to nonlinearity. The minimum
frequency which determines the total load shedding amount is
greater in the case with 5% of spinning reserve. It then leads to
more shed loads than the case with 10% of spinning reserve and
less residual frequency deviation.

B. Impact of Frequency Threshold

Fig. 4. Frequency response with different levels of spinning reserve.

Under normal operating conditions, loads are continually
varying, and frequency cannot be exactly held at nominal value.
According to operating guides, frequency deviation under nor-
range of (0, 100), and C is less sensitive to KLS while KLS is mal operating conditions should be limited in a small range,
greater than 100. C is approaching to 1.0 when KLS is infinite. e.g., less than ±0.2 Hz, and regulated by primary and secondary
Therefore, the load shedding amount with the continuous UFLS frequency regulation. In such circumstance, loads should not be
scheme is always less than the power imbalance no matter how shed. Therefore, appropriate frequency threshold should be set
KLS is tuned. In other words, Fig. 3 is consistent with (14) and to avoid shedding loads unintendedly. The frequency threshold
there is no over-shedding problem for the continuous UFLS for UFLS is different among different balancing authorities. For
scheme. example, the frequency threshold is 59.5 Hz in Northeast Power
Coordinating Council (NPCC), 59.3 Hz in Mid Continent Area
Power Pool (MCAPP), and 59.1 Hz in Western Electricity Co-
III. IMPROVED CONTINUOUS UFLS SCHEME WITH
ordinating Council (WECC). In China, the frequency threshold
NONLINEAR FACTORS
is usually set as 49.5 Hz.
The SFR model is greatly simplified to be linear. However, The continuous UFLS scheme is updated to model the fre-
there are obvious nonlinear factors in practical cases where the quency threshold, and actual load shedding amount is deter-
analytical property in Section II is not applicable. In this section, mined in (15) where Δfth is the threshold of frequency deviation
nonlinear factors of spinning reserve, frequency threshold, and in per unit.
time delay are investigated, and an improved continuous UFLS
scheme is proposed. 0 if Δf > Δfth
PLS = (15)
−KLS (Δf − Δfth ) if Δf < Δfth
A. Impact of Spinning Reserve
With KLS = 20 and PD = 0.2, the frequency response with
Spinning reserve is the available generation that can be in- different frequency threshold is shown in Fig. 5. It can be found
creased in a short time to meet load variation. Spinning reserve from Fig. 5 that, with greater frequency threshold, the maxi-
is usually limited to a small portion of total generation to mum frequency deviation becomes greater, and lower residual
balance the requirements of stability and economy. Simulations frequency can be observed. System frequency performance is
are carried out with different spinning reserve levels as shown deteriorated with greater frequency threshold. It is necessary to
in Fig. 4 with KLS = 20 and PD = 0.2 to check the impact of tune KLS to be greater for systems with greater frequency thresh-
spinning reserve on frequency dynamics. old to prevent severe frequency drop and recover frequency to a
It can be found from Fig. 4 that frequency is stalling when high level.
there is no spinning reserve. It can be proved with (13) that, Furthermore, since the transient frequency nadir is always less
without change of mechanical power, the shed load will not than the frequency threshold, (13) can be modified as:
bring system frequency back and the system residual frequency
deviation is –PD /(D + KLS ). When the spinning reserve is 5%, PD − P LS = −DΔfmin + ΔPG  −DΔfmin > −DΔfth
system frequency deviation can be recovered to about –0.001 pu. (16)

Authorized licensed use limited to: National Institute of Technology Andhra Pradesh. Downloaded on September 15,2022 at 04:34:13 UTC from IEEE Xplore. Restrictions apply.
 954 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 35, NO. 2, MARCH 2020

Fig. 7. Frequency response with different time delay.

Fig. 6. The logic of the additional stage.

It is indicated by (16) that, the shortage of load shedding

amount is always greater than –DΔfth . In order to make the load
shedding amount as close as the initial power imbalance, it is
necessary to shed additional loads greater than –DΔfth without
risk of over-shedding. The additional loads to shed are called
additional stage in this paper with Padd and tadd denoting the
load shedding amount and time delay of the additional stage.
An example is given in Fig. 6 to show the logic of the addi-
tional stage. The states of the additional stage, a.k.a. s in Fig. 6, Fig. 8. Extended SFR model with continuous UFLS scheme including fre-
are “wait”, “trigger” and “shed”. When frequency deviates from quency threshold, time delay and additional stage.
Δfth , a historic minimum frequency recorder is initiated to store
the minimum value of the following frequency dynamics. The
additional stage is in the “wait” state when the historic minimum where Pt LS is the loads that has already been shed corresponding
frequency is declining. It turns into “trigger” state and the timer to the historic minimum frequency, and Pc LS can be calculated
of the additional stage is started when the historic minimum by the following equation:
frequency stops changing. If the historic minimum frequency
holds for tadd , the timer is timed out and the additional stage
c
PLS = −KLS [Δf (t − td ) − Δfth ] Δf < Δfth (18)
will be in “shed” state to trip loads. Otherwise, the timer will be The recovery of frequency is an indicator of enough load
reset and the additional stage will return to “wait” state. Once shedding amount. In the improved logic (17), the loads are shed
frequency recovers above Δfth , the historic minimum frequency only when the timer is timed out and frequency is still declining.
record is cleared. The total shed amount is determined by the frequency at the time
which is td before tmin . In other words, the total shed amount
C. Impact of Time Delay is less than the one calculated with Δfmin . Therefore, the time
Traditional stage-by-stage scheme usually shed loads with delay will not introduce the over-shedding problem which can
appropriate time delay td to filter out temporary frequency drop be verified with (16).
and make sure different stages are shed in sequence. It is com- Simulation result with KLS = 20 and PD = 0.2 is shown in
pulsory for the frequency to drop beyond threshold frequency Fig. 7 to check the impact of time delay. It is clear that the
for the delayed time before loads are tripped. The typical time frequency nadir is dropped with a longer time delay. However,
delay is 0.1∼0.3 s. In this paper, td is selected as 0.2 s. the impact of time delay on residual frequency can be overlooked
An appropriate time delay is also necessary for the proposed when td < 0.2 s.
continuous UFLS scheme. A general way to implement the
time delay is to calculate the loads to shed with measured D. Improved Continuous UFLS Scheme and Examination
frequency and start a timer immediately. Loads are then shed With the previous discussion, an improved continuous UFLS
once the timer is timed out. However, the system may suffer the scheme is illustrated in the dashed box of Fig. 8 to include
over-shedding problem in cases that frequency has already been nonlinearity. In Fig. 8, Pmax and Pmin are upper and lower bound
recovered before the timer is timed out. The logic of time delay of generation. The output of recovered logic is 0 if frequency is
is then improved by adding logic as follows to overcome the declining and 1 if frequency is recovered.
over-shedding problem: An example is set with KLS = 35, Δfth = –0.01, td =
c
PLS if dΔf
< 0 and PLS
c t
> PLS 0.2 s, and PSR = 5% to compare the performance of the
PLS (t) = dt
(17) improved continuous UFLS scheme and the ideal scheme in
t
PLS if dΔf
dt ≥ 0 or PLS
c
≤ PLS
t
Section II. Additional stage is set as Padd = 1% and tadd = 10 s.

Authorized licensed use limited to: National Institute of Technology Andhra Pradesh. Downloaded on September 15,2022 at 04:34:13 UTC from IEEE Xplore. Restrictions apply.
 LI et al.: CONTINUOUS UFLS SCHEME FOR POWER SYSTEM ADAPTIVE FREQUENCY CONTROL 955

Fig. 11. Line-by-line UFLS scheme based on continuous UFLS scheme.

[20]. However, the amount of loads with such controllability

Fig. 9. Comparison between ideal and improved continuous UFLS schemes is not enough for emergency frequency control in most power
with different events.
systems nowadays.
Traditional stage-by-stage scheme sheds loads by tripping
lines of high voltage level, and massive loads will be shed if one
line is tripped. Current advancement of communication technol-
ogy enables such applications as fast demand side management
[21] and precise load control [22]. It provides a precise way to
trip lines of lower voltage level other than tripping lines of high
voltage level. The load shedding amount can be approximated
as continuous if lines of lower voltage level are tripped one by
one.
With precise load control, the proposed UFLS scheme can be
implemented with a line-by-line scheme in one of two modes:
Fig. 10. Comparison between ideal and improved continuous UFLS schemes
with different KLS.
local and remote, as shown in Fig. 11. In the local mode, UFLS
signal is generated in substation and then sent to circuit breakers
(CBs) inside the substation. In the remote mode, UFLS signal
is also generated in substation but sent to CBs of lower voltage
The relationship between total load shedding amount and power
level via additional communication channels. Load shedding
imbalance is shown in Fig. 9.
amount is calculated in the substation with local frequency
It can be seen from Fig. 9 that the total load shedding amount is
and a minimum amount of lines to provide enough load to
increasing with disturbance for the improved scheme. However,
shed is then determined according to line priority. When the
when disturbance is less than 0.044, no load is shed with the
frequency is declining, the increasing load shedding command is
improved scheme since frequency recovers before dropping
transformed into tripping signals for lines of lower voltage level
beyond Δfth . With limited spinning reserve, total load shedding
and sent to breakers via communication channels to shed loads
amount of the improved scheme is greater than the ideal scheme
gradually.
when disturbance is greater than 0.123 since the frequency nadir
If loads are shed in remote mode, the number of lines of lower
is greatly dropped.
voltage level is large and the load shedding is almost continuous.
Fig. 10 shows the relationship between total load shedding
However, in some systems with many large factories, a large
amount and KLS with PD = 0.2 and the same Δfth , td , PSR ,
portion of loads is directly connected to high-voltage lines.
Padd , and tadd as Fig. 9. It can be found from Fig. 10 that the
Tripping loads on lower voltage lines may be insufficient to
total load shedding amount is greater for the improved UFLS
bring frequency back. In this case, CBs on some high-voltage
scheme when KLS < 60 due to greater frequency deviation.
lines should be further tripped in local mode. Since the load on
With frequency threshold, the total load shedding amount of
the high-voltage line is heavy, tripping high-voltage lines may
the improved scheme will not reach power imbalance even KLS
deteriorate the continuity of the proposed UFLS scheme. The
approaches infinite. There is no over-shedding problem for the
performance of the line-by-line scheme is further examined in
improved continuous UFLS scheme.
Section V.
It can be summarized that the performance of improved
continuous UFLS scheme is nonlinear when spinning reserve,
IV. TUNING OF CONTINUOUS UFLS SCHEME
frequency threshold, and time delay are considered. However,
the dominant characteristic of the UFLS scheme is kept, e.g., A. The Requirement of Tuning KLS
the load shedding amount is increasing with greater KLS and is
Suppose the greatest active power deficiency is PDmax , the
almost proportional to power imbalance.
KLS should be tuned to make sure the residual frequency can be
successfully recovered above Δfth . Though greater KLS leads
E. Implementation With Precise Load Control to more load to shed and less frequency deviation, KLS cannot
The proposed continuous UFLS scheme is dependent on loads be tuned as infinite in field applications due to several reasons.
with high controllability which can be achieved with smart Firstly, frequency is always measured with bias. Greater KLS
power electronic controllers, such as energy storage systems leads to more shed loads with a small change of frequency.

Authorized licensed use limited to: National Institute of Technology Andhra Pradesh. Downloaded on September 15,2022 at 04:34:13 UTC from IEEE Xplore. Restrictions apply.
 956 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 35, NO. 2, MARCH 2020

To shed the precise amount of loads, it requires frequency To deeper look into the effect of varying damping on system
measurement devices with high accuracy. Secondly, results in frequency control, a load model with frequency dependency and
Fig. 3 indicate that the total load shedding amount is not sensitive load shedding is considered as:
to the change of KLS when KLS is great enough. It is reasonable
PL = PL0 (1 + DΔf ) (1 − PΣLS ) (26)
to select KLS as little as possible while preventing the system
from frequency instability. Thirdly, tmin is usually quite small for where PL is load power and PL0 is nominal load power. With
specific events. If the continuous UFLS scheme is implemented initial nominal load power as base power, PL0 is 1.0 p.u.
with controllable loads, loads need to change greatly in a short When there is no spinning reserve and frequency threshold
time if KLS is too great. It may be impossible for some devices of and additional stage are not considered, the power balance can
which the maximum power change rate is limited by controllers. be modeled as following which is similar to (19):
(1 − PD max ) − (1 + KLS Δfmin ) (1 + DΔfmin ) = 0 (27)
B. Tuning of KLS With Constant Damping
and KLS can be derived as:
It can be readily concluded that the frequency would drop
most if the power imbalance is PDmax when there is no spinning −PD max − DΔfmin
KLS = (28)
reserve. Without spinning reserve, the system minimum fre- Δfmin (1 + DΔfmin )
quency Δfmin is equal to the residual frequency Δf . Therefore, If frequency threshold is further considered when there is no
if no additional stage is considered, the power balance equation spinning reserve, power balance at tmin is:
when the system is stabilized can be modeled as:
(1 − PD max)−[1 + KLS (Δfmin − Δfth )](1 + DΔfmin ) = 0
(1 − PDmax ) − (1 + KLS Δfmin ) − DΔfmin = 0 (19) (29)
or If the additional stage is shed, power balance when the residual
PD max frequency is reached is modeled as:
KLS = − −D (20)
Δfmin (1 − PD max ) − [1 + KLS (Δfmin − Δfth ) + DΔfth ]
If the frequency threshold and additional stage are further (1 + DΔf∞ ) = 0 (30)
considered, the minimum frequency before the additional stage
is shed can be calculated with: From (29), the following relationship can be found:
1 − PD max
(1 − PD max ) − [1 + KLS (Δfmin − Δfth )] − DΔfmin = 0 1 + KLS (Δfmin − Δfth ) = (31)
(21) 1 + DΔfmin
After the additional stage is shed, the power balance when the Substitute (31) into (30), the minimum frequency is
residual frequency is reached is: obtained as:
(1 − PD max ) − [1 + KLS (Δfmin − Δfth ) + DΔfth ] Δf∞ (1 − PD max ) + Δfth (1+DΔf∞ )
Δfmin = (32)
(1 − PD max ) − DΔfth (1 + DΔf∞ )
− DΔf∞ = 0 (22)
Substitute (32) into (29), KLSmin can be obtained with
From (21) and (22), it can be calculated that, Δf = Δfth as:
Δfmin = Δfth + Δf∞ (23) PD max + DΔfmin
KLSmin = − (33)
Substitute (23) into (21), the KLS can be obtained as: (Δfmin − Δfth ) (1 + DΔfmin )

PD max + D (Δfth + Δf∞ ) If PDmax = 0.3, Δfth = –0.01, and D = 1, KLSmin is

KLS = − (24) 20.493 to make sure the system frequency can be successfully
Δf∞
brought back to Δfth and the maximum frequency deviation is
For the worst case with Δf = Δfth , the minimum KLS is: –0.0238 pu which is –1.190 Hz for 50 Hz systems or –1.428 Hz
PD max + 2DΔfth for 60 Hz systems.
KLS min = − (25) Since system damping is changing with operating conditions,
Δfth
it is hard to track the precise value of D. The D used for tuning
If PDmax = 0.3, Δfth = –0.01, and D = 1, KLSmin can be
KLS is usually chosen as typical value. It may be insufficient
readily calculated as 28 from (25). When KLS = 28, Δfmin is
to tune KLS as KLSmin if uncertainties of operating conditions,
–0.02 pu from (23) which is –1 Hz for 50 Hz systems or –1.2 Hz
voltage deviation and reduced boiler steam pressure are consid-
for 60 Hz systems. Tuning KLS to be greater than KLSmin will
ered. A small positive gain α should be added to tune KLS as
bring system frequency back to higher than Δfth .
follows to make the improved scheme adaptive to uncertainty:
C. Tuning of KLS With Varying Damping KLS = (1 + α) KLS min (34)
In most power systems, D is mainly contributed by load α should be set according to the frequency dynamics. For
frequency dependency, which is mostly determined by motors. different systems, α should be different and should be tuned
When part of loads is shed, the system damping is also re- to balance the frequency regulation performance and system
duced, and system frequency performance is deteriorated [23]. economy. In this paper, α is set as 0.05 with several numerical

Authorized licensed use limited to: National Institute of Technology Andhra Pradesh. Downloaded on September 15,2022 at 04:34:13 UTC from IEEE Xplore. Restrictions apply.
 LI et al.: CONTINUOUS UFLS SCHEME FOR POWER SYSTEM ADAPTIVE FREQUENCY CONTROL 957

TABLE I
STAGE-BY-STAGE UFLS SCHEME OF 39-BUS NEW ENGLAND MODEL

Fig. 13. Generator tripping events of New England model with 10% spinning
reserve.

1000 MW, and 1830 MW. The same events with 10% spinning
reserve are shown in Fig. 13. Nominal frequency and frequency
threshold are shown in the two figures as dotted lines.
Fig. 12. Generator tripping events of New England model with no spinning From Figs. 12 and 13, it can be found that both stage-by-stage
reserve. and continuous scheme can bring system frequency back. How-
ever, in the zero spinning reserve scenario, system frequency will
recover to greater than 60 Hz for the event of tripping 830 MW
simulations and is validated in Section V. A greater α can be generation with the stage-by-stage scheme. Therefore, though
selected to increase KLS if the frequency nadir is too low. the stage-by-stage scheme is well tuned, it still suffers from
the over-shedding problem. The proposed continuous UFLS
V. CASE STUDIES scheme, however, provides far better performance than the stage-
by-stage scheme without any over-shedding problems.
The general performance of the proposed UFLS scheme is
Fig. 12 shows that the recovery speed of the frequency is
analyzed in Sections II and III with the simplified model. It is
relatively slow for the event of tripping 1830 MW generation
worthwhile to check the actual performance of the proposed
with the proposed scheme. A greater KLS can be selected to avoid
scheme with detailed models. In this section, two systems are
starting under-frequency protections of some power plants. Re-
built with detailed models and tested with full time-domain
covery speed can also be increased by adding other action logic
simulation. One is 60 Hz 39-bus New England model, and the
to accelerate the operation of the additional stage according to
other is 50 Hz simplified Shandong Power Grid of China. System
the under-frequency protections of power plants.
damping is modeled as load frequency dependency in the two
To further check the voltage and angle stability, voltage and
systems. Therefore, KLS is tuned as Section IV-C.
angle dynamics of the event of tripping generator 38 in Fig. 13
are illustrated in Fig. 14. It is shown that the voltage deviation
A. Tests of New England Model
is small and system is angular stable.
In the New England model, the total load is 6150 MW and total The tuned continuous UFLS scheme is further implemented as
generating capacity is 7300 MVA. All generators are modeled three line-by-line schemes with 100, 50 and 20 low-voltage lines,
with 6-order GENROU generator model, IEEET1 exciter model, i.e., each line uniformly carries 1%, 2%, and 5% of substation
and IEEEG1 turbine-governor model [26]. Since load model is loads, respectively. The frequency response of the same events
critical for power system dynamics [27], static IEELAL model as Fig. 13 are illustrated in Fig. 15 to compare the line-by-line
with voltage and frequency dependency is used to model loads scheme with the continuous scheme. It can be found from Fig. 15
with typical parameters. The greatest power mismatch with N-2 that though the line-by-line scheme is discrete, the three schemes
contingency is 1830 MW when tripping equivalent generators implemented with precise load control can bring frequency back
at bus 38 and 39. With system total load as base power, system with similar performance as the continuous scheme.
D is 2.0, system R is 0.0421, and PDmax is 0.2975 pu. Δfth To further check the adaptability of the continuous scheme,
is set as –0.5 Hz which is the activating frequency of the first another continuous scheme is tuned as KLS = 28.790 with α =
stage of the stage-by-stage scheme. Thus, KLSmin is calculated 0.2, i.e., shedding 47.984% load per Hz. Results of the two
as 23.992 with (33). Final KLS is tuned as 25.192 with α = 0.05, schemes with KLS = 25.192 and KLS = 28.790 are shown in
i.e., shedding 41.987% load per Hz. The additional stage is set Fig. 16 with the same events as in Fig. 15. The scheme with
as Padd = –DΔfth = 0.0167 pu or 102.5 MW. Time delay for the KLS = 28.790 sheds loads slightly more than the scheme with
continuous part is td = 0.2 s, and time delay for the additional KLS = 25.192 and reduce maximum frequency deviation by
stage is tadd = 10 s. The traditional stage-by-stage UFLS scheme about 0.04 Hz. The performance of the proposed continuous
is listed in Table I [24] for comparison. scheme is tolerant to change of α.
COI frequency of tripping generator 38, generator 39, and The proposed UFLS scheme is dependent on frequency dy-
tripping the two generators is shown in Fig. 12 with no spinning namics which is affected by system inertia. To check the perfor-
reserve. Loss of generation of the three events are: 830 MW, mance of the proposed scheme with different inertia, the New

Authorized licensed use limited to: National Institute of Technology Andhra Pradesh. Downloaded on September 15,2022 at 04:34:13 UTC from IEEE Xplore. Restrictions apply.
 958 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 35, NO. 2, MARCH 2020

Fig. 17. Comparison of continuous schemes with different levels of system

inertia with 10% spinning reserve.

TABLE II
STAGE-BY-STAGE UFLS SCHEME OF SHANDONG POWER GRID

Fig. 14. Voltage and rotor angle dynamics of New England model with
continuous scheme when generator 38 is tripped.

Fig. 18. Dynamic frequency of HVDC blocking events of Shandong Power

Grid with 10% spinning reserve.

B. Tests of Shandong Power Grid

The simplified Shandong model consists of 118 buses with ten
1000 kV buses and eighty-one 500 kV buses, 24 generators, 31
transformers, 200 transmission lines, three high voltage direct
current (HVDC) links, and 65 loads. The total load is 58.9 GW.
Fig. 15. Comparison of continuous and line-by-line schemes for generator
tripping events of New England model with 10% spinning reserve. A ±660 kV HVDC link is operated at 4 GW and the other
two ±800 kV HVDC links are operated at 8 GW each. Except
for three generators modeled with classical GENCLS model
as sending units of the three HVDC links, other 21 generators
are all modeled with GENROU generator model, SEXS exciter
model, and TGOV turbine-governor model. All HVDC links
are modeled with CDC4T model [26]. All loads are modeled
with IEELAL model with voltage and frequency dependency.
The traditional stage-by-stage scheme used in Shandong Power
Grid is listed in Table II [25].
The maximum power deficiency is 20 GW or 33.97% when
all HVDC links are blocked. The system damping is 2.0 based
Fig. 16. Comparison of continuous schemes with different KLS for generator on system total load. With the proposed continuous UFLS
tripping events of New England model with 10% spinning reserve.
scheme, KLSmin is 14.130 when setting frequency threshold
as 49.25 Hz. With α = 0.05, the KLS is tuned as 14.837, i.e.,
shedding 29.674% load per Hz. The time delay of continuous
England model is tested with five cases, i.e., the base case, load shedding part is 0.2. The additional stage is set to shed 3%
increasing H by 15% and 30%, and decreasing H by 15% and of loads with a time delay of 10 s.
30%, and results are shown in Fig. 17. It can be found that the When the system spinning reserve is 10% of total loads, the
proposed continuous UFLS scheme is adaptive to the change of dynamic frequency is illustrated in Fig. 18 for events of blocking
system inertia. the ±600 kV HVDC link (4 GW), one ±800 kV HVDC link

Authorized licensed use limited to: National Institute of Technology Andhra Pradesh. Downloaded on September 15,2022 at 04:34:13 UTC from IEEE Xplore. Restrictions apply.
 LI et al.: CONTINUOUS UFLS SCHEME FOR POWER SYSTEM ADAPTIVE FREQUENCY CONTROL 959

control. Besides the response-driven UFLS, event-driven emer-

gency control is another important method to prevent severe
frequency deviation, e.g., remedial action scheme (RAS) and
system integrity protection scheme (SIPS). RAS and SIPS are
set for specific contingencies and loads to shed are calculated
according to the event signal to balance power deficiency. How-
ever, they can only deal with preset N −1 or N −2 contingen-
cies. It is hard to set RAS and SIPS for other rare but severe
N − k (k > 2) contingencies due to the large amount of combi-
nation. For these contingencies, the response-driven UFLS can
be started to prevent system collapse. Therefore, UFLS is backup
of RAS and SIPS, and coordination between them need to be
further studied in the future.
For the line-by-line scheme in the remote mode, additional
communication channels are needed to transmit load informa-
tion and tripping signal. The great investment and maintenance
cost of communication infrastructures is the first obstacle for
implementing the proposed UFLS scheme. To reduce the cost,
communication channels can be shared with other applications.
For example, the proposed UFLS scheme can use the same com-
munication channels of RAS or SIPS to improve the utilization
ratio of communication infrastructures.
Fig. 19. Voltage and rotor angle dynamics of Shandong model with continuous For traditional stage-by-stage UFLS scheme, CBs on high-
scheme when all HVDC links are blocked. voltage lines are tripped when frequency drops beyond thresh-
old. Important loads powered by the tripped lines will be shed
and the power supply reliability of distribution system operators
(DSOs) is threatened. To improve the reliability of DSOs, the
proposed UFLS scheme can be implemented to trip lines of
lower-voltage level. Lines powering important loads can be
labeled with high priority by DSOs and tripped as late as
possible. It has potential to guarantee both frequency stability
of transmission system operators (TSOs) and power supply
reliability of DSOs. But the performance of the proposed scheme
need to be further verified.
The theoretical performance of the proposed UFLS scheme
Fig. 20. Comparison of continuous and line-by-line schemes for HVDC is derived in Section II based on the SFR model, in which only
blocking events of Shandong Power Grid with 10% spinning reserve. steam turbine is considered. However, there are other widely-
used turbine governors, e.g., hydraulic, gas, combined-cycle
wind, and etc. Dynamics of these turbine governors are different
(8 GW), and all HVDC links (20 GW). Fig. 18 verified that from that of steam turbine, and may affect the performance and
system frequency declination is arrested by both stage-by-stage tuning of the proposed UFLS scheme. Therefore, it is necessary
and continuous UFLS schemes. In the stage-by-stage scheme, to examine the adaptability of the propose UFLS scheme to the
the frequency of 20 GW event is reduced to 48.166 Hz. The pro- systems with other kinds of turbine governors in the future.
posed continuous scheme, however, prevents system frequency
from falling beyond 48.320 Hz.
Voltage and rotor angle dynamics when blocking all HVDC VII. CONCLUSION
links of Fig. 18 is shown in Fig. 19. The system is stable with UFLS is the key resort for preventing frequency collapse. With
control of the continuous scheme. the development of new controllable devices and precise load
The line-by-line scheme is also checked with the same events control technique, power system frequency can be controlled in
as Fig. 18, and results are shown in Fig. 20. The three line-by-line a smarter way than traditional stage-by-stage UFLS schemes. In
schemes in Fig. 20 are valid to recover system frequency to the this paper, a new continuous UFLS scheme is proposed to shed
allowed range. Therefore, the line-by-line scheme is applicable loads proportional frequency deviation. The proposed scheme
in large-scale power systems. is adaptive to power mismatch based on local measurement
of frequency dynamics and is immune to the over-shedding
problem. The improved continuous UFLS scheme incorporating
VI. DISCUSSIONS
frequency threshold and time delay is practical to implement
In this paper, a continuous UFLS scheme is proposed to and can be easily tuned with the proposed tuning method. The
reinforce the performance of the response-driven emergency proposed UFLS scheme is adaptive to events and operating

Authorized licensed use limited to: National Institute of Technology Andhra Pradesh. Downloaded on September 15,2022 at 04:34:13 UTC from IEEE Xplore. Restrictions apply.
 960 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 35, NO. 2, MARCH 2020

conditions. With precise load control, the continuous scheme can [22] W. Yin et al., “Design of fast communication interface for precision load
be implemented as the line-by-line scheme which can achieve control system,” Autom. Elect. Power Syst., vol. 42, no. 10, pp. 143–149,
2018.
similar performance as the continuous scheme. [23] H. Huang and F. Li, “Sensitivity analysis of load-damping characteristic
in power system frequency regulation,” IEEE Trans. Power Syst., vol. 28,
no. 2, pp. 1324–1335, May 2013.
REFERENCES [24] Northeast Power Coordinating Council, “NPCC regional reliabil-
ity reference directory #12 under-frequency load shedding pro-
[1] Y. Wen, C. Y. Chung, and X. Ye, “Enhancing frequency stability of gram requirements,” Accessed: Oct. 17, 2018. [Online]. Avail-
asynchronous grids interconnected with HVDC links,” IEEE Trans. Power able: https://www.npcc.org/Standards/Directories/Directory12%20Full%
Syst., vol. 33, no. 2, pp. 1800–1810, Mar. 2018. 20Member%20clean%2020150330%20GJD.pdf
[2] P. Ferraro, E. Crisostomi, R. Shorten, and F. Milano, “Stochastic frequency [25] Y. Liu and C. Li, “Impact of large-scale wind penetration on transient
control of grid-connected microgrids,” IEEE Trans. Power Syst., vol. 33, frequency stability,” in Proc. IEEE Power Energy Soc. Gen. Meeting, San
no. 5, pp. 5704–5713, Sep. 2018. Diego, CA, USA, 2012, pp. 1–5.
[3] Y. Liu, R. Fan, and V. Terzija, “Power system restoration: A literature [26] “PSS/E model library V33.5,” Siemens, Munich, Germany, Oct. 2013.
review from 2006 to 2016,” J. Modern Power Syst. Clean Energy, vol. 4, [27] V. Milanovic et al., “CIGRE WG C4.605: Modelling and aggregation
no. 3, pp. 332–341, 2016. of loads in flexible power networks,” Feb. 2014. Accessed: Apr. 10,
[4] N. Nguyen and J. Mitra, “An analysis of the effects and dependency of 2019. [Online]. Available: https://e-cigre.org/publication/566-modelling-
wind power penetration on system frequency regulation,” IEEE Trans. and-aggregation-of-loads-in-flexible-power-networks
Sustain. Energy, vol. 7, no. 1, pp. 354–363, Jan. 2016.
[5] H. Golpîra, H. Seifi, A. R. Messina, and M. Haghifam, “Maximum penetra-
tion level of micro-grids in large-scale power systems: Frequency stability
viewpoint,” IEEE Trans. Power Syst., vol. 31, no. 6, pp. 5163–5171, Changgang Li (M’15) received the B.E. and Ph.D.
Nov. 2016. degrees in electrical engineering from Shandong Uni-
[6] B. Delfino, S. Massucco, A. Morini, and M. Haghifam, “Implementation versity, Jinan, China, in 2006 and 2012, respectively.
and comparison of different under frequency load-shedding schemes,” in He was a Research Scholar with the School of Electri-
Proc. Power Eng. Soc. Summer Meeting, Vancouver, BC, Canada, 2001, cal Engineering and Computer Science, The Univer-
pp. 307–312. sity of Tennessee, Knoxville, TN, USA, from 2012 to
[7] IEEE Guide for the Application of Protective Relays Used for Abnormal 2014. He is currently an Associate Research Fellow
Frequency Load Shedding and Restoration, IEEE Standard C37.117-2007, with the School of Electrical Engineering, Shandong
Aug. 24, 2007, pp. 1–55. University. His research interest includes power sys-
[8] X. Lin, H. Weng, Q. Zou, and P. Liu, “The frequency closed-loop control tem operation and control.
strategy of islanded power systems,” IEEE Trans. Power Syst., vol. 23,
no. 2, pp. 796–803, May 2008.
[9] L. Sigrist, I. Egido, and L. Rouco, “A method for the design of UFLS
schemes of small isolated power systems,” IEEE Trans. Power Syst.,
vol. 27, no. 2, pp. 951–958, May 2012
[10] L. Zhang and J. Zhong, “UFLS design by using F and integrating DF/DT,”
in Proc. IEEE PES Power Syst. Conf. Expo., Atlanta, GA, USA, 2006, Yue Wu received the B.E. degree in electrical en-
pp. 1840–1844. gineering from Zhengzhou University, Zhengzhou,
[11] T. Amraee, M. G. Darebaghi, A. Soroudi, and A. Keane, “Probabilistic China, in 2018. He is currently working toward the
under frequency load shedding considering RoCoF relays of distributed postgraduate degree with Shandong University, Jinan,
generators,” IEEE Trans. Power Syst., vol. 33, no. 4, pp. 3587–3598, China. His research interest includes power system
Jul. 2018. dynamic frequency control.
[12] L. Sigrist, I. Egido, and L. Rouco, “Performance analysis of UFLS schemes
of small isolated power systems,” IEEE Trans. Power Syst., vol. 27, no. 3,
pp. 1673–1680, Aug. 2012.
[13] V. V. Terzija, “Adaptive underfrequency load shedding based on the
magnitude of the disturbance estimation,” IEEE Trans. Power Syst., vol. 21,
no. 3, pp. 1260–1266, Aug. 2006.
[14] T. Shekari, F. Aminifar, and M. Sanaye-Pasand, “An analytical adaptive
load shedding scheme against severe combinational disturbances,” IEEE
Trans. Power Syst., vol. 31, no. 5, pp. 4135–4143, Sep. 2016. Yanli Sun received the B.E. and M.E. degrees in elec-
[15] A. Ketabi and M. H. Fini, “An underfrequency load shedding scheme for trical engineering from Shandong University, Jinan,
hybrid and multiarea power systems,” IEEE Trans. Smart Grid, vol. 6, China, in 2015 and 2018, respectively. She is currently
no. 1, pp. 82–91, Jan. 2015. an Engineer with State Grid Shandong Electric Power
[16] L. Sigrist, I. Egido, and L. Rouco, “Principles of a centralized UFLS Corporation Weifang Power Supply Company, China.
scheme for small isolated power systems,” IEEE Trans. Power Syst., Her research interests include power system analysis
vol. 28, no. 2, pp. 1779–1786, May 2013. and control.
[17] P. M. Anderson and M. Mirheydar, “An adaptive method for setting
underfrequency load shedding relays,” IEEE Trans. Power Syst., vol. 7,
no. 2, pp. 647–655, May 1992.
[18] P. M. Anderson and M. Mirheydar, “A low-order system frequency re-
sponse model,” IEEE Trans. Power Syst., vol. 5, no. 3, pp. 720–729,
Aug. 1990.
[19] Y. Xie et al., “Development approach of a programmable and open
software package for power system frequency response calculation,” Pro- Hengxu Zhang (M’06) received the B.E., M.S.,
tection Control Modern Power Syst., vol. 2, pp. 189–198, 2017. and Ph.D. degrees in electrical engineering from
[20] M. A. Torres L., L. A. C. Lopes, L. A. Morán T., and J. R. Espinoza C., Shandong University, Jinan, China, 1998, 2000, and
“Self-tuning virtual synchronous machine: A control strategy for energy 2003, respectively. He is currently a Professor with
storage systems to support dynamic frequency control,” IEEE Trans. the School of Electrical Engineering, Shandong Uni-
Energy Convers., vol. 29, no. 4, pp. 833–840, Dec. 2014. versity. His research interests include power system
[21] S. Acharya, M. S. E. Moursi, and A. Al-Hinai, “Coordinated frequency security and stability assessment, power system mon-
control strategy for an islanded microgrid with demand side management itoring, and numerical simulation.
capability,” IEEE Trans. Energy Convers., vol. 33, no. 2, pp. 639–651,
Jun. 2018.

Authorized licensed use limited to: National Institute of Technology Andhra Pradesh. Downloaded on September 15,2022 at 04:34:13 UTC from IEEE Xplore. Restrictions apply.
 LI et al.: CONTINUOUS UFLS SCHEME FOR POWER SYSTEM ADAPTIVE FREQUENCY CONTROL 961

Yutian Liu (SM’96) received the B.E. and M.S. de- Vladimir Terzija (M’95–SM’00–F’16) has been the
grees from the Shandong University of Technology, EPSRC Chair Professor in power system engineering
Jinan, China, in 1984 and 1990, respectively, and the with the School of Electrical and Electronic Engineer-
Ph.D. degree from Xi’an Jiaotong University, Xi’an, ing, The University of Manchester, Manchester, U.K.,
China, in 1994, all in electrical engineering. He is cur- since 2006. From 1997 to 1999, he was an Assistant
rently a Chair Professor with the School of Electrical Professor with the University of Belgrade, Serbia.
Engineering, Shandong University, Jinan, China. His From 2000 to 2006, he was with ABB AG, Germany,
research interests include power system analysis and working as an Expert for switchgear and distribution
control, renewable energy integration, and artificial automation. His research interests include smart grid,
intelligence application to power system. application of intelligent methods to power system
monitoring, control, and protection, switchgear and
fast transient processes, as well as DSP applications in power systems.

Yilu Liu (S’88–M’89–SM’99–F’04) received the

B.S. degree from Xi’an Jiaotong University, Xi’an,
China, and the M.S. and Ph.D. degrees from The
Ohio State University, Columbus, OH, USA, in 1986
and 1989, respectively. She is currently the Gover-
nor’s Chair with the Knoxville and Oak Ridge Na-
tional Laboratory (ORNL), University of Tennessee,
Knoxville, TN, USA. She is elected as the member
of National Academy of Engineering in 2016. She is
also the Deputy Director of the DOE/NSF cofunded
engineering research center CURENT. Prior to join-
ing UTK/ORNL, she was a Professor with Virginia Tech. She led the effort to
create the North American power grid frequency monitoring network (FNET) at
Virginia Tech, which is now operated at UTK and ORNL as GridEye. Her current
research interests include power system wide-area monitoring and control, large
interconnection-level dynamic simulations, electromagnetic transient analysis,
and power transformer modeling and diagnosis.

Authorized licensed use limited to: National Institute of Technology Andhra Pradesh. Downloaded on September 15,2022 at 04:34:13 UTC from IEEE Xplore. Restrictions apply.