This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2017.2720621, IEEE
Transactions on Power Systems
1

Minimum Synchronous Inertia Requirement of

Renewable Power Systems
Huajie Gu, Ruifeng Yan, Member, IEEE and Tapan Kumar Saha, Senior Member, IEEE

AbstractThe installed capacity of wind generation and Number of synchronous generators registered
photovoltaics (PV) in many countries is going to dominate N reg* for a type of regulation FCAS, where * is R
generation fleets in a bid to meet growing renewable energy (raise) or L (lower)
targets. Synchronous inertia has never been problematic as there
Number of synchronous generators registered
was more available than needed, but it is being significantly
reduced due to the increasing integration of non-synchronous for a type of contingency FCAS, where ** is R6
renewable generation. When the low bidding priced generation of N ctg** (6 second raise) or L6 (6 second lower) or R60
wind and PV becomes considerably large, conventional economic (60 second raise) or L60 (60 second lower) or R5
dispatch algorithms can result in less online synchronous inertia (5 minute raise) or L5 (5 minute lower)
and put power system security at risk. However, the compromise NSC Number of registered synchronous condensers
of power system security due to synchronous inertia shortage is N WF Number of wind farms having power reserve
not well studied in literature. This paper develops a synchronous
Psync i t
inertia constrained economic dispatch algorithm to satisfy the Dispatched power output of the ith synchronous
minimum required synchronous inertia of frequency control. generator in the dispatch cycle t
Synchronous condensers and wind reserve are economically Dispatched amount of a type of regulation FCAS
allocated to alleviate any shortage of synchronous inertia and
Preg* i t
from the ith registered synchronous generator in
frequency control ancillary services (FCAS). A Gaussian particle the dispatch cycle t, where * is for R (raise) or L
swarm optimization algorithm is introduced to simultaneously
(lower)
co-optimize the dispatch of synchronous generators and their
FCAS, wind reserve and synchronous condensers. Dispatched amount of a type of contingency
FCAS from the ith registered synchronous
Pctg** i t
Index Terms Synchronous inertia, economic dispatch, wind generator in the dispatch cycle t, where ** is R6
turbine, renewable energy, power system. (6 second raise) or L6 (6 second lower) or R60
(60 second raise) or L60 (60 second lower) or R5
(5 minute raise) or L5 (5 minute lower)
NOMENCLATURE
Pwrsv i t
Dispatched amount of power reserve from the ith
registered wind farm in the dispatch cycle t
C t
Overall power system operational cost in the
economic dispatch cycle t in terms of net load i Penalty coefficient of power reserve
Csync i Generation cost of the ith synchronous generator unavailability assigned to the ith wind farm
Synchronous inertia of the ith synchronous
Cost of a type of regulation FCAS from the ith Ii
Creg* i registered synchronous generator, where * is R
generator or condenser
Total synchronous inertia of dispatched
(raise) or L (lower)
I t synchronous generators and condensers in the
Cost of a type of contingency FCAS from the ith
dispatch cycle t
registered synchronous generator, where ** is
Cctg** i R6 (6 second raise) or L6 (6 second lower) or f Rate of change of frequency (RoCoF)
R60 (60 second raise) or L60 (60 second lower) f cstr RoCoF constraint
or R5 (5 minute raise) or L5 (5 minute lower) f nom Normal frequency deviation
Csc i
Cost of synchronous inertia provision services f cstr1 Normal frequency deviation constraint
from the ith registered synchronous condenser f ctg Contingency frequency deviation
Cwrsv i
Cost of power reserve from the ith registered
f cstr2 Contingency frequency deviation constraint
wind farm
Nsync Number of synchronous generators Pctg t Contingency size in the dispatch cycle t
Perr t
Power mismatch in the dispatch cycle t caused
by forecast errors
Corresponding author: Ruifeng Yan (email: ruifeng@itee.uq.edu.au). He is
with Global Change Institute, University of Queensland, Brisbane St Lucia, Current wind speed blowing across the ith wind
vi
QLD 4072 Australia. farm
Huajie Gu and Tapan Kumar Saha are with power and energy systems Minimum wind speed of the ith wind farm to
research group, School of Information Technology and Electrical Engineering, vmin i
University of Queensland, Brisbane St Lucia, QLD 4072, Australia (emails: enable power reserve
h.gu@uq.edu.au, saha@itee.uq.edu.au).

0885-8950 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2017.2720621, IEEE
Transactions on Power Systems
2

Scale parameter of Weibull distribution for wind non-synchronous renewable generation.

ci
speed variation of the ith wind farm
Shape parameter of Weibull distribution for II. METHODOLOGY
ki
wind speed variation of the ith wind farm Australian energy market operator (AEMO) studied a 100%
Mean of Weibull distribution for wind speed
i RPS of the national electricity market (NEM) [1] and proposed
variation of the ith wind farm generation mixes for that scenario in 2030. One of them is
Autocorrelation of two successive wind speeds
i studied in this paper and listed in Table I. Fig. 1 shows the
of the ith wind farm predicted wind/PV generation and load profile of the NEM on a
hi Height above ground of the ith wind farm random day in 2030. The prediction is made based on the data
li Turbulence length scale of the ith wind farm from the NEM Review for the Australian power system [7].
Time lag between two successive wind speeds Net load is the actual load subtracted by non-dispatchable
Frequency in wind turbulence spectrum wind and PV generation [8]. As can be seen in Fig. 1, the net
Constant to tune the penalty coefficient load is significantly light and reaches its minima of 2.2 GW at
T0. Consequently, fewer synchronous generators are required
I. INTRODUCTION to power the net load. The light net load may cause a shortage
of synchronous inertia and degrade frequency control. At
R ENEWABLE energy is a promising technology to reduce
CO2 emission and therefore to mitigate global warming.
Within the next few decades, current power systems
present, successful frequency control in the NEM requires that
RoCoF should not exceed 0.5 Hz/s [8]. Tolerated frequency
predominantly supplied by synchronous generators will deviations for normal events are 0.15 Hz [8]. Maximum
gradually witness a structural change of generation portfolios allowed frequency deviations for contingency events are 0.5
and ultimately may even shift to a 100% renewable power Hz [8].
system (RPS). For an RPS with great asynchronous renewable
TABLE I. A 100% RPS GENERATION MIX PROJECTION OF THE NEM [1]
generation, although massive transmission upgrades [1] and
demand side management [2] are also crucial, the most basic Generation Capacity (MW)
and challenging problem lies in the electricity supply side. PV, rooftop 10,905
PV, utility 11,000
Economic dispatch algorithms proposed in literature have Wind, onshore 34,500
focused on the improvement of algorithm performance [3, 4] Geothermal 1,000
and consideration of more operational constraints [5], such as Bagasse 1,010
spinning reserve and generator ramp rates. They did not Biomass (wood) 7,000
Concentrating solar thermal (CST) 11,000
dynamically evaluate the synchronous inertia adequacy of a Hydro 7,330
dispatch result in terms of N-1 contingency [3-5]. This is due to Pumped hydro 740
the fact that synchronous inertia adequacy is not a problem for a Biogas 13,500
conventional power system with limited non-synchronous Total 97,985
Peak demand 43,000
generation.
However, synchronous inertia is being materially reduced
If there is no wind or PV generation in the power system,
due to ongoing displacement of synchronous generators by
conventional dispatch algorithms schedule a large amount of
asynchronous ones, such as wind turbine generators and
synchronous generators to meet the actual load. Consider an
photovoltaic (PV) panels. Electronic inverters used in wind
N-1 contingency event for the RPS, the largest online
turbines and PV panels lack the synchronous inertia that can
synchronous generator is suddenly tripped at T0 as highlighted
help an RPS survive a major disturbance.
in Fig. 1. The maximum post-contingency frequency deviation
If comprised of less online synchronous inertia, an RPS may
of the RPS is well within its required band (49.5~50.5 Hz) as
routinely suffer a rapid rate of change of frequency (RoCoF)
shown in Fig. 2. The largest RoCoF is 0.09 Hz/s which is less
and a large frequency deviation following a disturbance. A
than the threshold of 0.5 Hz/s. Therefore, the frequency control
great RoCoF of 6 Hz/s was recorded in the South Australian
security is conventionally guaranteed by a large fleet of
blackout on September 28, 2016 which was caused by the
synchronous generators.
lightness of the South Australian power system [6]. The
With a large amount of wind and PV integration, fewer
instantaneous penetration level of wind and PV generation was
synchronous generators are scheduled by conventional
over 50% before the blackout in South Australia and only three
economic dispatch algorithms, especially during the low load
thermal power plants were dispatched [6].
condition. For the same contingency event at T0 in Fig. 1, the
This paper therefore proposes a synchronous inertia
corresponding maximum frequency deviation rapidly breaches
constrained economic dispatch to keep the minimum amount of
its constraint (49.5 Hz) as in Fig. 2. The largest RoCoF is 1.43
synchronous inertia online. The proposed algorithm introduces
Hz/s which is much larger than the maximum tolerated value of
a feedback loop to conventional economic dispatch algorithms.
0.5 Hz/s. Therefore, the inadequacy of synchronous inertia
The feedback loop is used to evaluate the inertia adequacy of
should be taken into account when performing economic
dispatched synchronous generators and to ensure the dynamic
dispatch with a large amount of wind and PV generation.
security of frequency control for an RPS with high

0885-8950 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2017.2720621, IEEE
Transactions on Power Systems
Fig. 1. A projected daily wind/PV generation and load profile of the NEM in
2030 [7].
Fig. 2. A frequency control example with the current frequency regulation
standard used by AEMO.
In general, a rapid RoCoF may result in the cascading trip of
online synchronous generators [8-10]. A faster initial RoCoF
generally contributes a larger frequency deviation. A severe
frequency deviation can also trip synchronous generators and
load shedding is required to restore RPS frequency [8-10].
These two frequency control indices are secured by
synchronous inertia and frequency control ancillary services
(FCAS) in the NEM of Australia.
Synchronous inertia as a large damper is vitally important to
slow down frequency dropping so that sluggish droop
controllers can correct frequency later. As synchronous
generators are coupled with power system frequency, the
kinetic energy of their spinning rotors or inertia is
instantaneously released to address the power deficit in a fast
manner. Governor response of synchronous generators or
FCAS take time to further balance the power deficit.
A. Current Economic Dispatch and Frequency Control in the
NEM
AEMO uses a five minute central dispatch cycle consisting
0885-8950 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
3
of generation and FCAS dispatches to instantaneously balance
generation and load in the cheapest manner [11], as illustrated
in Fig. 3. Generation dispatch centrally schedules synchronous
generators based on their real-time bids to track a five minute
Generation and FCAS Dispatch
Generation and FCAS Dispatch
Generation and FCAS Dispatch
Generation and FCAS Dispatch
ahead load forecast [8].
... ... ...
... Load Tracking ... Load Tracking ...
12:05
12:10
19:35
19:40
Fig. 3. Current five minute central dispatch in the NEM.
Over periods shorter than five minutes, small power
mismatches caused by forecast errors of load, wind and PV
generation are controlled by regulation FCAS [8]. Regulation
FCAS are split into raise and lower regulation FCAS. Their
required amounts in the next five minutes are purchased from
registered synchronous generators.
Contingency FCAS are used to manage rapid and large
frequency deviations arising from sizable load-generation
unbalances caused by a sudden loss of large generation or a
load [8]. The contingency FCAS are split into six categories: 6
second raise (R6), 6 second lower (L6), 60 second raise (R60),
60 second lower (L60), 5 minute raise (R5) and 5 minute lower
(L5). The required amount of each of them in the next five
minutes is purchased from registered synchronous generators
with different ramp rates [8].
Regulation and contingency FCAS are required to restore
frequency to the normal frequency range within a dispatch
cycle. Therefore, the frequency control of every dispatch cycle
is relatively independent.
The blue dashed line in Fig. 2 illustrates a frequency control
example with the current frequency regulation standard used by
AEMO. A contingency event occurs at time zero, resulting in a
rapid drop in frequency. Synchronous inertia as a measurement
of stored rotational kinetic energy is instantaneously released to
slow down frequency dropping speed, in other words to reduce
RoCoF. When frequency further leaves the normal frequency
operating range at T1, R6 contingency FCAS are activated to
rapidly but temporarily balance the power deficit [11]. Then
R60 contingency FCAS start at T2 (T1 + 6s) to bring frequency
back to the normal frequency range [11]. Next, R5 contingency
FCAS as a delayed service enter at T3 (T1 + 60s) to further
restore the frequency to around the nominal value 50 Hz [11].
B. Synchronous Inertia Constrained Economic Dispatch
The flow chart of the proposed economic dispatch algorithm
is shown in Fig. 4. A feedback loop highlighted in red is added
to conventional economic dispatch. The feedback loop is
introduced to evaluate the synchronous inertia adequacy of a
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2017.2720621, IEEE
Transactions on Power Systems
4

dispatch result via rapid dynamic simulation of an N-1 cost of its power reserve in the cost function. Therefore, power
contingency event. If the post-contingency frequency response reserve from a wind farm with greater uncertainty is less
of an RPS cannot satisfy the constraints of RoCoF and favored by the dispatch algorithm for sourcing FCAS.
frequency deviation, the dispatch algorithm will find a new Wind reserve price will not be out of control in the future,
solution until the simulated frequency response is acceptable. because FCAS market is a demand-driven and free-bidding
Synchronous generators and their FCAS are the only market. Many individual wind farms will partake in FCAS
resources available in the conventional economic dispatch. market and they will compete for selling their power reserve.
However, frequency control is compromised by the shortage of Whoever offers the cheapest wind reserve will be chosen.
synchronous inertia and FCAS in an RPS with high wind and PV theoretically can provide power reserve to support
PV penetration. Synchronous condensers and wind reserve are frequency control. However, the problem is its price. Levelised
therefore introduced to the proposed dispatch algorithm. Wind cost of electricity (LCoE) for PV is around 170 AUD/MWh,
reserve is dispatched to maintain the minimum amount of whilst for wind energy is 80-100 AUD/MWh [14]. Wind
power reserve needed by the studied power system to satisfy reserve is more financially attractive than PV reserve.
frequency control. Meanwhile, synchronous condensers are Therefore, the power reserve capability of PV is not taken into
scheduled to keep the adequacy of synchronous inertia. The account in this paper.
proposed economic dispatch algorithm schedules wind reserve
and synchronous condensers simultaneously with conventional Forecasted net load
FCAS from synchronous generators.
The economic dispatch problem in this paper is optimized by
Economic dispatch via
a Gaussian particle swarm optimization algorithm with

Proposed algorithm: a feedback loop to

evaluate synchronous inertia adequacy
DLGPSO
differential evolution local search (DLGPSO) [12, 13]. The
DLGPSO optimizer offers a self-tuning capability based on No

Conventional economic dispatch

Gaussian distribution so as to avoid the parameter sensitivity of
the original PSO algorithm. Furthermore, it is used with
Frequency response safe?
differential evolution to boost the algorithm performance.
Wind and PV generation is not dispatchable and the net load
should be powered by synchronous generators. The adequacy
of synchronous inertia and FCAS should also be maintained. N-1 contingency Yes
For the dispatch cycle t, the dispatches of synchronous rapid dynamic simulation
generators and their FCAS, synchronous condensers and wind
reserve are simultaneously co-optimized to minimize the total
cost of the next five minute RPS operation. The cost function is
Dispatch result Output dispatch result
the sum of the cost of synchronous generation and their FCAS,
the cost of synchronous inertia provision services from Fig. 4. Synchronous inertia constrained economic dispatch algorithm.
synchronous condensers and the cost of wind reserve. C. Minimum Synchronous Inertia
Therefore, the cost function to be minimized is formulated as
Nsync For the dispatch cycle t, the total synchronous inertia of
C t C sync i P t
sync i
dispatched synchronous generators and condensers is
Nsync NSC
i 1
N regR N regL I t Ii (2)
C regR i P t C
regR i regL i P t
regL i
i 1
The minimum amount of the total synchronous inertia should
i 1 i 1
N ctgR6 N ctgL6 ensure that RoCoF and frequency deviation meet their
C ctgR6 i P ctgR6 i t CctgL6 i PctgL6 i t constraints in case of a contingency event. They are
i 1 i 1
N ctgR60 N ctgL60
(1) f cstr f f cstr

C ctgR60 i P ctgR60 i t CctgL60 i PctgL60 i t f cstr1 f nom f cstr1 (3)
i 1 i 1 f f f
N ctgR5 N ctgL5 cstr2 ctg cstr2

C ctgR5 i P ctgR5 i t CctgL5 i PctgL5 i t Initial RoCoF is the largest after a contingency event, though
i 1 i 1
RoCoF is gradually reduced through the provision of R6
NSC N WF
Csc i I i Cwrsv i Pwrsv i t i vi vmin i contingency FCAS. The largest RoCoF is only determined by
i 1 i 1 the contingency size and synchronous inertia. The largest

where i vi vmin i is the penalty coefficient proposed to RoCoF of the dispatch cycle t can be derived as [15]
Pctg t
describe the power reserve uncertainty from the ith wind farm. f max t f cstr (4)
It is further explained in Section II-D. A wind farm with larger 2I t
power reserve uncertainty is assigned with a bigger penalty If only considering the RoCoF constraint, then the required
coefficient. This penalty coefficient intentionally scales up the minimum synchronous inertia according to (4) is

0885-8950 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2017.2720621, IEEE
Transactions on Power Systems
5

Pctg t
I t
2f cstr
As can be seen in Fig. 5, the relevant minimum required
synchronous inertia is determined proportionally by the
contingency size and RoCoF constraint. For example, in order
to ensure RoCoF is not larger than 0.5 Hz/s, the minimum
required synchronous inertia is 35 GWs for the contingency
size of 700 MW. However, such a level of synchronous inertia
alone is not enough to prevent the maximum frequency
deviation from violating its constraint.
unlikely to reduce to a level below 6 m/s for the same period of
(5) time. As a result, WF B has a better capability to provide power
reserve, and should be favored by the economic dispatch
algorithm. Meanwhile, if the wind speed variability is the same
for both wind farms, the higher the current wind speed of WF A
is, the more secure its power reserve will be in the next dispatch
interval. Consequently, WF A is more preferable under this
situation. Therefore, the probability of wind reserve
unavailability is related to two factors the location of a wind
farm and its current wind speed. The location determines wind
variability calculated based on long-term historical wind speed
profile. The current wind speed is associated with the ability of
a wind farm during a dispatch interval to contribute a certain
amount of power reserve.

Fig. 5. Minimum required synchronous inertia with respect to contingency

size (RoCoF constraint only).

Frequency deviations also depend on the amount and ramp

rate of contingency FCAS. A rapid dynamic security model of
RPS frequency control is developed in Section III-A to quantify
frequency deviations with respect to different amounts of
online synchronous inertia and FCAS. It is therefore used to
quickly evaluate the synchronous inertia adequacy for every
economic dispatch result.
D. Wind Reserve Uncertainty
The variation of nationwide wind generation may be
relatively smaller as more wind turbines are being installed
over a larger geographical area. However, the power output
Fig. 6. Daily generation of Boco Rock wind farm in NSW Australia [7].
The variation of wind speeds over a certain time interval on
the ith wind farm is conventionally assumed to be subject to
Weibull distribution with scale parameter ci and shape
parameter ki [12, 16, 17]. Scale and shape parameters are
derived from historical wind statistics. They are different for
wind farms in different locations. The probability density
function (PDF) of bivariate Weibull distribution describing two
successive wind speeds on the ith wind farm is given by [16]
1 x2
x x
2 2
1
1 x2 2 i x1 x2
2 2

variation of a single wind farm within a day can be considerably 1

2 1 i2 2
e
large as shown in Fig. 6. Its power output can become zero for 4 1 i 2

hours during a day. Power reserve of an individual wind farm is f ws i v1 , v2 1

1
not as reliable as that of a conventional power plant. x x ki
A minimum wind speed is required to enable power reserve, 2
1 erf 1 1 erf 2
ci
ln 2 ln 2 (6)
otherwise a wind turbine may stall. In order to maintain
ki2 2 2
promised power reserve, the incoming wind speed of a wind
farm should be greater than a required minimum level, e.g. 6
m/s in the next dispatch cycle. Therefore, the availability of x x
wind reserve is determined by wind speed variation. 1 erf 1 1 erf 2
2 2
For instance, both Wind Farm A (WF A) and Wind Farm B
where
(WF B) are facing the same wind speed of 8 m/s. According to

k
v i
the wind speed variability from historical data, the wind speed 1
1
x1 2erf 1 2e i
c (7)
of WF A has a high probability to become less than 6 m/s in the
next dispatch cycle. On the contrary, the wind speed of WF B is

0885-8950 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2017.2720621, IEEE
Transactions on Power Systems
6

and penalty coefficient is proportional to the probability of wind

reserve unavailability. A wind farm is given a larger penalty
ki
v
2
x2 2erf 1 1 2e ci
(8) coefficient if it has a higher probability that its power reserve is
going to be unavailable.

i is the autocorrelation of two successive wind speeds on the
same wind farm at time lag . So f ws i v1 , v2 is the PDF of the
event that wind speed v1 changes to v2 after seconds.
The autocorrelation i can be estimated from a turbulent
spectrum of wind speeds as [18]

i Speci cos 2 d (9)
0

where Speci is the Kaimal spectrum for wind turbulence

of the ith wind farm and is defined as [18]
i
0.164
0 i
Speci (10)
5


3
1 0.164 i
0 i


where i hi / i and 0 i 0.041 hi / li . Fig. 7. Probability of power reserve unavailability of a wind farm over the
Over a time period of , the cumulative distribution function next five minutes for different Weibull distribution parameters.
of a wind speed ramp starting in the range vs1 to vs 2 and ending
The penalty coefficient changes the competitiveness of wind
in the range ve1 to ve 2 for the ith wind farm is [18]
farms in FCAS market. Boco Rock and Gullen Range wind
Fws i vs1 v1 vs 2 , ve1 v2 ve 2 f ws i v1 , v2 dv1dv2 (11) farms in NSW Australia [7] are used as examples for further
vs 2 ve 2

vs1 ve1

Therefore, the probability of current wind speed vi decreases explanation. Assume that at the beginning of one economic
dispatch cycle, they are offering the same amount of 1 MW
to less than the minimum one vmin i can be estimated by
power reserve at the same price of 150 AUD/MWh. Boco Rock

Fws i vi 0.5 v1 vi 0.5, 0 v2 vmin i (12) wind farm has a penalty factor of 1.3 and its power reserve
Since the integration at one point is zero, a range of price is lifted to 1501.3=195 AUD/MWh in the cost function
vi 0.5, vi 0.5 is introduced to facilitate the probability of the proposed economic dispatch algorithm. Meanwhile,
Gullen Range wind farm has a penalty factor of 1.2 and its
calculation.
power reserve price is increased to 1501.2=180 AUD/MWh in
The probability of power reserve unavailability of a wind
the cost function. The proposed economic dispatch algorithm
farm for the next five minutes (an economic dispatch cycle) is
prefers the power reserve of Gullen Range wind farm because it
shown in Fig. 7. The minimum required wind speed is set to be
is cheaper and more reliable in comparison with that of Boco
6 m/s. In general, the probability of wind reserve becoming
Rock wind farm. Note that the increased price is only used to
unavailable is increasing when mean wind speed is dropping.
quantify the wind reserve uncertainty for the cost function. The
The probability is further affected by Weibull scale and shape
actual price paid to a wind farm is the same as its bid.
parameters.
According to the probability of wind reserve unavailability E. Forecast Errors
for the next five minutes, a penalty coefficient is introduced to Forecast errors of load, wind and PV generation can be
quantify the impact of wind reserve uncertainty on economic experienced at all times. Forecast systems used by AEMO are
dispatch decision making. Since one objective of the proposed required to have a target that five minute ahead normalized
economic dispatch algorithm is to simultaneously minimize the mean absolute errors should not exceed tolerated thresholds [8].
cost of FCAS, whoever offers the cheaper wind reserve will be Regulation FCAS are used to correct small power mismatches
favored. If the original price of power reserve from the ith wind caused by forecast errors.
farm is multiplied by a penalty coefficient larger than one, then The amount of regulation FCAS should not be less than the
the price of this wind reserve is increased. Consequently, the worst power mismatch caused by forecast systems, though it
opportunity of this wind reserve being chosen for sourcing may not be fully activated due to the fact that the sum of wind
FCAS is disadvantageous. and PV generation may change less as compared with
The penalty coefficient assigned to the ith wind farm for its individual wind or PV generation. For instance, an unexpected
power reserve uncertainty is defined as increase of wind generation over a dispatch cycle occurs along
i vi vmin i 1
vi 0.5 vmin i
f ws i v1 , v2 dv1dv2 (13) with an unpredicted decrease of PV generation. For this case,
vi 0.5 0
the overall variation of wind and PV generation is smaller, and
where is a constant to tune the penalty coefficient. The

0885-8950 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2017.2720621, IEEE
Transactions on Power Systems
the amount of regulation FCAS is always enough to
compensate power mismatches due to forecast errors.
III. FREQUENCY CONTROL DYNAMIC MODELLING
A. Frequency Control
The RPS frequency control dynamic model is shown in Fig.
8. Simulated power unbalances are caused by both forecast
errors and the sudden loss of the largest online synchronous
generator. Regulation and contingency FCAS can be provided
by hydro, bio-fueled gas, steam power plants and wind farms.
Details of FCAS modelling can be found in [11].
Electrically decoupled wind turbines can participate in
frequency regulation via emulated inertia (EI) and active power
control (APC). EI extracts kinetic energy from a spinning rotor
by means of reducing its speed. APC enables wind turbines to
have a governor type response similar to synchronous
generators.
Perr (t)
Pctg t
Power system
Wind farms
Hydro power
plants
Gas power plants
Steam power
plants
Fig. 8. RPS frequency control dynamic model.
EI and APC both respond to under-frequency contingency
events. APC can de-load wind turbines to provide power
reserve at the expense of profit losses. APC is also used to
reduce power output of wind turbines for over-frequency
events. Wind reserve comes with a cost whilst EI response for a
short period around ten seconds raises negligible financial
problems. EI and APC can effectively reduce RoCoF and
frequency deviation in cooperation with the governor response
of conventional hydro, gas and steam turbine generators [19].
Synchronous generators equipped with synchronous
condensing clutches can be decoupled from their driving
turbines and temporarily run as synchronous condensers at the
expense of small parasitic load [20]. Such an operation mode
can cost-effectively increase the synchronous inertia of an RPS.
All gas power plants in the studied generation mix are assumed
to be retrofitted with synchronous condensing clutches, and
0885-8950 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
7
they are the source of synchronous inertia provision services.
B. Variable-speed Wind Turbine EI
Frequency deviation [21] or its combination of RoCoF [22]
can be used to trigger EI response. Considering the
susceptibility of a derivative controller to noise, the frequency
deviation triggered EI controller proposed by General Electric
(GE) is used in this paper [21]. However, an EI controller is a
replaceable component for a fast dynamic RPS model. Since
the aim of this paper is to develop a synchronous inertia
constrained economic dispatch algorithm, other EI controllers
can also be chosen.
As shown in Fig. 9, frequency is the input of the EI controller
and it contains information of both frequency deviation and
RoCoF. Frequency is first passed through a dead band. When
frequency deviation is larger than the preset threshold, EI
response is activated. Furthermore, a washout filter is applied to
ensure the inertia response will be ceased after the required
support period. As a result, the washout filter can generate an
aggressive power surge signal to counter faster RoCoF.
Therefore, EI response is affected by both frequency deviation
and RoCoF.
f
pmxwi
1 sTwowi urlwi
f K wi pwi
dbwi 1 sTlpwi 1 sTwowi drlwi
f0
pmnwi
Fig. 9. Variable-speed wind turbine EI control model [21].
Although an EI controller is designed to respond to
under-frequency events, it is technically capable of responding
to over-frequency events as well. It can be done by modifying
the dead band highlighted in red in Fig. 9.
EI response of a single wind turbine with respect to different
wind speeds is shown in Fig. 10. Fig. 10(a) shows the simulated
RPS frequency variation. As can be seen in Fig. 10(b), when
wind turbine generation is less than 20% of rated at a low wind
speed, power surge from EI response is limited to prevent rotor
stall. Above this power output level, the net power increase of
EI response is comparatively similar. Initial power surge in
Stage 1 of EI response is beneficial to frequency control but
following underproduction in Stage 2 is detrimental since it can
result in a second frequency drop [19].
When wind speed is over the rated, temporal overproduction
can be provided by decreasing pitch angle and power recovery
is not required [23]. However, rapid EI response of wind
turbines cannot be replaced by such governor like response
through reducing pitch angle. It takes more time to change pitch
angle due to pitch rate limitations, so the resultant power surge
is slower as compared with that of the EI response.
C. Variable-speed Wind Turbine APC
A wind turbine can be de-loaded either by raising pitch angle
[19, 21, 24] or controlling rotor speed [25, 26]. A combination
of speed and pitch control has the advantage of eliminating EI
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2017.2720621, IEEE
Transactions on Power Systems
power recovery period and delivering more energy upon calling
[26]. It is used in this paper to enable the wind turbine with
power reserve.
Fig. 10. EI response of a single wind turbine with respect to different wind
speeds. (GE 2.5 MW wind turbine [21, 23]).
When rotor speed is below the rated, over-speeding is
implemented to de-load a wind turbine as well as to store extra
kinetic energy. The reserved power and stored kinetic energy
can later be released to address RPS disturbances. After rotor
speed reaches the rated, the pitch angle is increased to further
curtail power output [26]. As shown in Fig. 11, much more
energy is available from a wind turbine using the over-speeding
control for the same RPS frequency drop shown in Fig. 10(a).
Meanwhile, there are no power recovery periods for the wind
turbine. As a result, the wind turbine can contribute
significantly more energy to frequency regulation with less
wind spillage.
Fig. 11. EI and APC response of a wind turbine (GE 2.5 MW wind turbine [21],
wind speed 8 m/s).
IV. SIMULATION RESULTS
The generation mix of studied 100% RPS is listed in Table I
and is divided into three categories. Non-dispatchable wind and
0885-8950 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
8
PV generation is treated as negative load but they can be
curtailed for security concerns [1]. Geothermal, biomass
(wood) and bagasse power plants have slow ramp up
capabilities and they are dispatched close to their maximum
output [1]. They are suitable for baseload operation. Turbines
designed for hydro, pumped hydro, CST and biogas have a
faster ramp up capability and they are suitable for supplying
rapid demand growth during peak times [1].
Based on predicted load profile and wind/PV generation of a
random day as shown in Fig. 1, the proposed economic dispatch
algorithm is performed to show that online synchronous inertia
should not be less than a certain amount in order to ensure the
security of frequency control. The proposed algorithm offers a
framework to address the shortage of synchronous inertia
caused by the integration of a large amount of renewable
energy. It can be directly applied to other power systems.
A. Synchronous Inertia
As shown in Fig. 12, when net load is dramatically small
between 12:00 and 15:00 as in Fig. 1, the synchronous inertia of
dispatched synchronous generators based on conventional
economic dispatch is significantly light. Such a dispatch cannot
guarantee frequency control security as illustrated in Fig. 2.
Fig. 12. Synchronous inertia of the studied RPS.
The proposed dispatch algorithm gives another curve of
online synchronous inertia depicted by the blue dashed line
shown in Fig. 12. The minimum amount of synchronous inertia
is maintained to ensure frequency control security. To achieve
this goal, synchronous condensers are scheduled together with
wind reserve and other FCAS to satisfy N-1 contingency
frequency control. As a result, the frequency response of the
same N-1 contingency event at T0 is now acceptable as shown
in Fig. 13. The maximum frequency deviation is now within its
band (49.5 Hz). The largest RoCoF is 0.49 Hz/s and it is less
than the limit of 0.5 Hz/s.
When the proposed dispatch algorithm requires that there is a
need to prepare wind reserve or synchronous condensers in the
next dispatch cycle, the dispatched amount of wind reserve or
the dispatched capacity of synchronous condensers is not zero.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2017.2720621, IEEE
Transactions on Power Systems
9

Such a need comes from the result of N-1 contingency rapid

dynamic simulation as shown in Fig. 4. If this dynamic
simulation conveys the dispatch algorithm that the current
solution cannot ensure frequency control security, then the
dispatch algorithm will find a new solution to satisfy frequency
control requirements.

Fig. 14. Dispatched R6/R60/R5 contingency FCAS from synchronous

generators and the minimum amount of required wind reserve.

D. Cost Comparison
A 100% RPS is assumed to be subsidy free in this paper. The
LCoE of each type of generation is shown in Table II. The
pricing of synchronous inertia provision services from power
plants running on synchronous condenser mode should
Fig. 13. Frequency response of N-1 contingency event at T0.
consider the investment of synchronous condensing clutches
and O&M cost. The cost of synchronous inertia provision
B. Wind Reserve services is assumed to be 2 AUD/MWs per hour in this paper
De-loading operation of wind farms with abundant wind [20]. Wind reserve cost is assumed to be 150 AUD/MWh [27].
resources is assumed to be compulsory in this paper. Wind The daily RPS five minute operational cost is shown in Fig.
farms should support frequency control as they are the largest 16. The opportunity cost of FCAS is only paid when there is an
energy supplier in the studied generation mix. actual contingency event. The normal RPS operational cost
Fig. 14 shows the dispatched R6/R60/R5 contingency FCAS based on the proposed economic dispatch algorithm is slightly
from synchronous generators and the minimum amount of lifted during the period when synchronous condensers are run
required wind reserve. Simulation results show that wind to address the shortage of synchronous inertia. As highlighted
reserve is scheduled during the period when there is a shortage in Fig. 16, the additional cost of the activation of synchronous
of R6 contingency FCAS. The reason why there is such a condensers is economically acceptable to maintain frequency
shortage is that scheduled synchronous generators cannot offer regulation security.
enough rapid power reserve and it is not economical to turn on
extra synchronous generators. Therefore, the raise contingency
FCAS have to be sourced from more expensive wind reserve.
C. Synchronous Condenser
Simulation results show that synchronous condensers are
scheduled during the period when the net load is significantly
light as can be seen in Fig. 15. When scheduled synchronous
generators can provide enough synchronous inertia, the
capacity of scheduled synchronous condensers is zero. Because
the synchronous inertia from synchronous generators is
naturally available with their power dispatch, and the running
of synchronous condensers increases the cost of power system
operation.

Fig. 15. Minimum capacity of required synchronous condensers.

0885-8950 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2017.2720621, IEEE
Transactions on Power Systems
10

TABLE II. LEVELISED COST OF ELECTRICITY (LCOE) condensers into the economic dispatch of power systems with
Cost high penetration of non-synchronous renewable generation.
Generation type Prime mover Energy source
(AUD/MWh) The proposed algorithm economically dispatches wind reserve
Non Solar [14] 170 to address the deficit of conventional FCAS from synchronous
Asynchronous
Wind turbine Wind [14] 100
generators. Meanwhile, power plants retrofitted with
Geothermal [14] 150
Bagasse [14] 120 synchronous condensing clutches are scheduled to cover the
Steam turbine shortage of synchronous inertia.
Biomass (wood) [14] 125
Synchronous CST [14] 200
Gas turbine Biogas [14] 180 REFERENCE
Hydro [28] 20
Hydro turbine [1] Australian Energy Market Operator. (2013). 100 percent renewables
Pumped hydro [29] 250
study modelling outcomes report. Available:
https://www.environment.gov.au/climate-change/publications/aemo-mod
elling-outcomes
[2] J. A. Short, D. G. Infield, and L. L. Freris, "Stabilization of grid frequency
through dynamic demand control," IEEE Transactions on Power Systems,
vol. 22, pp. 1284-1293, Aug 2007.
[3] P. Jong-Bae, L. Ki-Song, S. Joong-Rin, and K. Y. Lee, "A particle swarm
optimization for economic dispatch with nonsmooth cost functions,"
IEEE Transactions on Power Systems, vol. 20, pp. 34-42, 2005.
[4] T. Ding and Z. Bie, "Parallel augmented lagrangian relaxation for
dynamic economic dispatch using diagonal quadratic approximation
method," IEEE Transactions on Power Systems, vol. 32, pp. 1115-1126,
2017.
[5] C. Wang and S. M. Shahidehpour, "Effects of ramp-rate limits on unit
commitment and economic dispatch," IEEE Transactions on Power
Systems, vol. 8, pp. 1341-1350, 1993.
[6] Australian Energy Market Operator. (2016). Black system South Australia
28 September 2016 third preliminary report. Available:
http://www.aemo.com.au/-/media/Files/Electricity/NEM/Security_and_R
eliability/Reports/Integrated-Third-Report-SA-Black-System-28-Septem
ber-2016.pdf
[7] NEM Review. (2016). Australian national electricity market review.
Available: http://www.nem-review.info/
Fig. 16. Studied RPS five minute operational cost during a day. [8] Australian Energy Market Operator. (2013). Integrating renewable
energy - wind integration studies report. Available:
V. DISCUSSION http://www.aemo.com.au/Electricity/Planning/~/media/Files/Electricity/
Planning/Reports/Integrating%20Renewable%20Energy%20-%20Wind
With less synchronous inertia, frequency variation of an RPS %20Integration%20Studies%20Report%202013.pdf.ashx
following disturbances can be radical, especially at times of [9] Siemens. (2011). Global blackouts lessons learned. Available:
http://m.energy.siemens.com/ru/pool/hq/power-transmission/HVDC/Glo
light load. However, hydro, steam and gas turbines and their bal_Blackouts.pdf
paired synchronous generators have instability problems and [10] Australian Energy Market Operator. (2014). Renewable Energy
cannot stay synchronized when facing a high RoCoF up to 2 Integration in South Australia. Available:
http://www.aemo.com.au/Electricity/Planning/~/media/Files/Electricity/
Hz/s [30]. A design improvement of conventional turbines and Planning/Reports/Renewable_Energy_Integration_in_South_Australia_
synchronous generators may be needed so as to tolerate higher AEMO_Electranet_Report_Oct_2014.ashx
RoCoFs in the future. [11] Intelligent Energy Systems. (2010). Raise contingency FCAS-price
control mechanism. Available:
http://www.energyregulator.tas.gov.au/domino/otter.nsf/LookupFiles/10
VI. CONCLUSION 3252_IES_Final_Report_Raise_Contingency_FCAS_Price_Control_Me
chanism_28_July_2010.pdf/$file/103252_IES_Final_Report_Raise_Con
Conventional economic dispatch algorithms schedule tingency_FCAS_Price_Control_Mechanism_28_July_2010.pdf
synchronous generators to follow load variation without [12] H. Gu and J. Wang, "Irregular-shape wind farm micro-siting
considering the compromised dynamic security of frequency optimization," Energy, vol. 57, pp. 535-544, 8/1/ 2013.
[13] C. Wan, J. Wang, G. Yang, H. Gu, and X. Zhang, "Wind farm
control caused by a high penetration level of wind and PV micro-siting by Gaussian particle swarm optimization with local search
generation. An RPS cannot be safely run without enough strategy," Renewable Energy, vol. 48, pp. 276-286, 12// 2012.
synchronous inertia so as to survive potential contingency [14] AECOM Australia. (2014). Australia's off-grid clean energy market
events. research paper. Available:
https://arena.gov.au/files/2014/12/ARENA_RAR-report-20141201.pdf
This paper developed a synchronous inertia constrained [15] A. Ulbig, T. Rinke, S. Chatzivasileiadis, and G. Andersson, "Predictive
economic dispatch algorithm that guarantees the dynamic control for real-time frequency regulation and rotational inertia provision
security of frequency control all the time. The requirement of in power systems," in IEEE 52nd Annual Conference on Decision and
Control (CDC), 2013, pp. 2946-2953.
synchronous inertia is determined by both RoCoF and [16] D. Villanueva, A. Feijo, and J. Pazos, "Multivariate Weibull distribution
frequency deviation constraints. The largest RoCoF is solely for wind speed and wind power behavior assessment," Resources, vol. 2,
decided by synchronous inertia and contingency size, whilst the p. 370, 2013.
[17] H. Gu, J. Wang, Q. Lin, and Q. Gong, "Automatic Contour-Based Road
maximum frequency deviation is further affected by the Network Design for Optimized Wind Farm Micrositing," IEEE
provision of contingency FCAS. Transactions on Sustainable Energy, vol. 6, pp. 281-289, 2015.
This paper introduces wind reserve and synchronous

0885-8950 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPWRS.2017.2720621, IEEE
Transactions on Power Systems
11

[18] L. Wu and D. G. Infield, "Towards an assessment of power system

frequency support from wind plant - modeling aggregate inertial Huajie Gu received the B. Eng. degree in
response," IEEE Transactions on Power Systems, vol. 28, pp. 2283-2291, Automatic Control from Nanjing Technical
2013. University, Nanjing, China, in 2011, and the M.
[19] V. Gevorgian, Y. Zhang, and E. Ela, "Investigating the impacts of wind Eng. degree in Control Theory and Control
generation participation in interconnection frequency response," IEEE
Engineering from Tongji University, Shanghai,
Transactions on Sustainable Energy, vol. 6, pp. 1004-1012, Jul 2015.
China, in 2014. Currently, he is a PhD student in
[20] F. Li, J. Kueck, T. Rizy, and T. King. (2006). A preliminary analysis of the
the Power & Energy Systems Research Division at
economics of using distributed energy as a source of reactive power
supply. Available: the School of Information Technology and
http://info.ornl.gov/sites/publications/Files/Pub1771.pdf Electrical Engineering, the University of
[21] General Electric Energy, "Modeling of GE wind turbine-generators for Queensland, Australia. His research interests
grid studies," General Electric International, Inc.2010. include wind turbine control and renewable power
[22] J. Morren, S. W. H. d. Haan, W. L. Kling, and J. A. Ferreira, "Wind systems.
turbines emulating inertia and supporting primary frequency control,"
IEEE Transactions on Power Systems, vol. 21, pp. 433-434, 2006. Ruifeng Yan (S2009, M2012) received the B.
[23] L. Ruttledge, N. W. Miller, J. O'Sullivan, and D. Flynn, "Frequency Eng. (Hons.) degree in Automation from
response of power systems with variable speed wind turbines," IEEE University of Science and Technology, Beijing,
Transactions on Sustainable Energy, vol. 3, pp. 683-691, Oct 2012. China, in 2004, the M. Eng degree in Electrical
[24] E. Ela, V. Gevorgian, P. Fleming, Y. C. Zhang, M. Singh, E. Muljadi, et Engineering from the Australian National
al. (2014). Active power controls from wind power brigding the gaps. University, Canberra, Australia, in 2007, and
Available: http://www.nrel.gov/docs/fy14osti/60574.pdf Ph.D. degree in Power and Energy Systems from
[25] Y. Xue and N. Tai, "Review of contribution to frequency control through the University of Queensland, Brisbane, Australia,
variable speed wind turbine," Renewable Energy, vol. 36, pp. 1671-1677, in 2012. His research interests include power
2011. system operation and analysis, and renewable
[26] A. Zertek, G. Verbic, and M. Pantos, "Optimised control approach for energy integration into power networks.
frequency-control contribution of variable speed wind turbines," IET
Renewable Power Generation, vol. 6, pp. 17-23, 2012.
[27] S. Rose and J. Apt, "The cost of curtailing wind turbines for frequency
Tapan Kumar Saha (M93, SM97) was born in
regulation and ramp-rate limitation," in Proc. 29th USAEE/IAEE North
Bangladesh in 1959 and immigrated to Australia
American Conference on Energy and the Environment: Conventional and
in 1989. He received his B. Sc. Engineering
Unconventional Solutions, 2010, pp. 1-18.
[28] International Renewable Energy Agency. (2012). Renewable energy (electrical and electronic) in 1982 from the
technologies: cost analysis series. Available: Bangladesh University of Engineering &
https://www.irena.org/documentdownloads/publications/re_technologies Technology, Dhaka, Bangladesh, M. Tech
_cost_analysis-hydropower.pdf (electrical engineering) in1985 from the Indian
[29] Melbourne energy institute. (2014). Opportunities for pumped hydro Institute of Technology, New Delhi, India and
energy storage in Australia. Available: PhD in 1994 from the University of Queensland,
http://energy.unimelb.edu.au/__data/assets/pdf_file/0007/1526587/Opps- Brisbane, Australia. Tapan is currently Professor
for-pumped-hydro-in-Australia.pdf of Electrical Engineering in the School of
[30] DNV KEMA Energy & Sustainability. (2013). RoCoF an independent Information Technology and Electrical Engineering, University of Queensland,
analysis on the ability of generators to ride through rate of change of Australia. Previously he has had visiting appointments for a semester at both
frequency values up to 2hz/s. Available: the Royal Institute of Technology (KTH), Stockholm, Sweden and at the
http://www.eirgridgroup.com/site-files/library/EirGrid/DNV-KEMA_Re University of Newcastle (Australia). He is a Fellow of the Institution of
port_RoCoF_20130208final_.pdf Engineers, Australia. His research interests include condition monitoring of
electrical plants, power systems and power quality.

0885-8950 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.