Stochastic Modeling of Tidal Generation for

Transient Stability Analysis: A Case Study based on

the All-Island Irish Transmission System
Gukrún Margrét Jónsdóttir and Federico Milano
School of Electrical & Electronic Engineering, University College Dublin, Ireland
[email protected], [email protected]

Abstract—The renewable energy currently generated in the (SDEs). SDEs are continuous with respect to time and are
Irish system is almost entirely supplied by wind power plants. therefore well equipped to reproduce transient fluctuations
However, in the sea around Ireland there is a significant tidal in the source of uncertainty. Power system models are typ-
energy potential. This paper provides a comparison of these two
renewable energy sources, namely wind and tidal, in terms of ically formalized as a set of Differential-Algebraic Equations
short-term variability and its impact on the dynamic behavior (DAEs). This allows for the SDEs to be readily incorporated
of the system. With this aim, stochastic models of the short-term into the system model and the resulting model is a set of
variability of these two energy sources are proposed. Simulation Stochastic Differential-Algebraic Equations (SDAEs).
results indicate that tidal generation leads to larger frequency The use of SDEs for power system studies has been a trend-
variations than those that are caused by wind generation. The
paper also shows that the inclusion of frequency control in tidal ing topic over the last few years with the increased uncertainty
power plants effectively mitigates such fluctuations. in the system, introduced with the increasing integration of
variable renewable energy sources. In [6], a systematic method
Index Terms—Irish power system, stochastic differential equa-
tions, Stokes wave model, tidal generation, wind generation to model power systems as SDAEs is presented. The method
is demonstrated through a case study of the IEEE 145-bus
50-machine system. A few more studies on multi-bus systems
I. I NTRODUCTION including uncertainties, modeled as SDEs, have been presented
in the literature. For example, in [7], the IEEE 145-bus test
A. Motivation
systems is modified to include wind generation, formulated
Ireland is ideally positioned to capitalize on ocean energy using SDEs and the systems stability is studied. The Icelandic
and has claims to one-third of the potential offshore renewable system is modeled with uncertain variations, modeled as a set
resources in northwestern Europe [1]. A relevant part of such of linearized SDEs and its frequency stability assessed in [8]
resources is tidal energy. In [2], the tidal resources in Ireland and in [9], the IEEE 39-bus test systems transient stability is
are studied and eleven sites are identified. Although there studied through SDAEs where wind is modeled as stochastic.
are multiple techno-economic issues still to be solved [3], Several SDE-based approaches, on the stochastic modeling
the recent success of some tidal generation projects, such as of specific energy sources such as wind [10]–[12] and tidal
MeyGen (Pentland Firth, Scotland) [4], has demonstrated that [13], as well as loads [10], [14], have been proposed in the
tidal stream generation is a viable technology. literature. The volatility of wind power is either modeled
Tidal currents have a high long-term predictability com- through the wind power directly, or through the wind speed.
pared to other prominent renewable energy sources, e.g. wind In [10], the wind power is modeled based on active power
and solar [5]. This makes tidal generation an excellent choice measurements, with a 1 Hz sampling rate, collected in Chile.
for supplying the base load of the system. However, short- In this case the wind power is modeled as an exponential
term fluctuations (seconds to minutes) in the current are less function of the Ornstein-Uhlenbeck (OU) process. However,
predictable and can negatively impact the power quality and for dynamic studies of the system, modeling the wind speed
stability of systems which include tidal generation. is to be preferred, as the dynamics associated with the turbine
B. Literature review itself can be considered.
In [11] the wind speed is modeled as an exponentially
Uncertainty, introduced through diverse sources, such as decaying autocorrelated Weibull distributed SDE. However,
loads, wind and solar, can negatively impact the reliability, more detailed models are required to properly capture the
safety and economy of power systems. The uncertainties in the autocorrelation of the wind speed. Thus, in [12], a method
system can be modeled using Stochastic Differential Equations to build SDEs with an arbitrary probability distribution and
autocorrelation is proposed, for wind speed modeling. The
This material is based upon works supported by the Science Foundation method consists in the superposition of OU stochastic pro-
Ireland, under Investigator Programme, Grant No. SFI/15/IA/3074. cesses. This method has since been extended to model the

21st Power Systems Computation Conference Porto, Portugal — June 29 – July 3, 2020
PSCC 2020
short-term variations in tidal current speed [13]. between 2 − 3 m/s, compared to the rated wind speed of a
wind turbine. Thus, to obtain similar power ratings, the tidal
C. Contributions turbine rotor size is much smaller than that of a wind turbine.
The contributions of this paper are threefold, as follows. Another noticeable difference, when it comes to dynamic
• Provide a comparison of tidal and wind generation in analysis, is due to the very input signal, i.e. wind speed and
terms of technology and the uncertainty of the energy tidal current speed. For its relevance in the case study, the
source. following section is dedicated to this subject.
• Compare the impact of the short-term uncertainty of tidal
and wind generation on the system frequency through A. Wind Speed and Tidal Current Speed
a case study of the Irish system. The modeling of
If average values over some minutes are considered, the
the uncertainty of the two energy sources is based on
current speed is highly predictable many days or even years
measurements.
in advance. The mean current speed usually has roughly four
• Consider the use of frequency control of the tidal genera-
or eight peaks per day. This is illustrated in Fig. 1.a where
tion to mitigate the effect that waves have on the dynamic
an example of measured current speed for Port Mantee, US
behavior of the system.
is shown [16]. Wind power on the other hand will vary with
D. Organization more uncertainty in that minutely to hourly time frame. An
example of measured wind speed, collected in Mayo, Ireland,
The remainder of the paper is organized as follows. Section
is shown in Fig. 1.b [17].
II provides an overview of the current status of tidal generation
technology and its similarities to wind generation. Section II
also discusses the differences between the variations in the 14
wind speed and the tidal current. Section III outlines the 7 (a) (b)
12
stochastic models used in the case study. In Section IV, the test Tidal Current Speed [dm/s]
6
system and its modeling are presented and simulation results 10

Wind Speed [m/s]

are presented. Finally, in Section V conclusions are drawn and 5
8
future work is outlined. 4
6
II. T IDAL G ENERATION 3

2 4
The technology used to harvest the kinetic energy in tidal
current is relatively mature compared to other ocean energy 1 2
technologies. Tidal turbines extract energy from the ocean
0 0
movement due to the tidal phenomenon. Tides are a con- 0 12 24 36 48 0 12 24 36 48
sequence of the changing gravitational pull of the sun and Time [h]

moon with respect to the earth’s oceans. Tides consists in large Fig. 1: Wind speed and tidal current speed measurements averaged
bodies of water that move towards and away from the shore. every 10 minutes [16], [17].
The tidal movement is site specific. Each location experiences
diurnal tides (one high, one low in a tidal day), semi-diurnal Wind and tidal current speeds are variable on a shorter
tides (two high, two low in a tidal day) or a mixture of the time scale as well. In the case of wind speed these short-
two. Tides can be predicted far in advance and with a high term variations can be due to gusts and/or turbulence. The
degree of accuracy. This fact makes tidal generation one of intensity of the turbulence depends on the terrain surrounding
the more reliable sources of renewable energy and is one of the wind power plant. For a land-based wind turbines, the
the most relevant reasons for its exploitation. normal turbulence intensity is close to 20 % for wind speed
Numerous types of tidal generation devices have been of around 12 m/s [18], while for offshore locations, this
proposed in the literature. The main research and development percentage is expected to be slightly smaller. An example of
focus has been on horizontal-axis tidal stream turbines, with measured wind speed, with a sampling frequency of 1 Hz,
important tidal projects, such as SeaGen in Northern Ireland is shown in Fig. 2.b, measured in Tracy, California [19].
[15] and more recently MeyGen in Scotland [4] utilizing those. Further details on how these short-term variations in the wind
A contributing factor to their dominance is their similarity to speed are modeled for power system analysis are presented in
wind turbines. Section III-B.
The utilization of similar technology for tidal and wind Short-term variations in the tidal current are both due to
generation systems is not surprising, as both aim to capture the bottom and side friction and the surface waves. These
kinetic energy from one kind of a flowing mass. However, fluctuations are typically about 10 % of the mean speed
several important differences exist between the two energy [18]. An example of measured tidal current, with a sampling
sources, the density of the flowing mass being one of the most frequency of 1 Hz, is shown in Fig. 2.a, collected in the
relevant. The density of air is about 1/800 of the density of European Marine Energy Centre tidal test site in Orkney, UK
water. This means the rated current speed can be much lower, [20]. In this case the fluctuations are dominated by turbulence.

21st Power Systems Computation Conference Porto, Portugal — June 29 – July 3, 2020
PSCC 2020
 2.5 autocorrelation. Thus, ρ(t) is a stochastic process defined as
the weighted sum of n SDE processes:
2.4 12
n
X √
Tidal Current Speed [m/s]

2.3 ρ(t) = wi ηi (t), (2)

10

Wind Speed [m/s]

2.2 i=1

2.1 where ηi (t), with i ∈ {1, . . . , n}, are SDE processes with
8
autocorrelations Rηi (τ ), wi > 0 and
2.0
n
X
1.9 6
wi = 1. (3)
1.8 i=1
(a) (b)
4 If all n processes have an identical Gaussian probability
1.7
0 25 50 75 100 0 25 50 75 100 distribution N (µη , ση2 ), the stochastic process ρ(t) has the
Time [s]
same Gaussian probability distribution, N (µη , ση2 ), and an
Fig. 2: Wind speed and tidal current speed measurements with a 1 Hz
sampling frequency [19], [20].
autocorrelation which is a weighted sum of the autocorrelation
functions of the n SDE processes, that is:
n
X
Rρ (τ ) = wi Rηi (τ ). (4)
The tidal current speed can also be subject to wind waves
i=1
and ocean swells. Wind waves have a relatively short wave
length and high frequency. These are caused by local winds. If the n SDE processes in (2) are η(t) processes as in (1), the
Swells are long wavelength waves that originate in a remote resulting autocorrelation of ρ(t) is a weighted sum of decaying
region of the ocean and propagate out of their area of gener- exponential functions and (4) can be rewritten as:
ation. These are considered to have the greatest perturbation n
X
effect on the tidal current speed in a tidal turbine, as they travel Rρ (τ ) = wi exp(−αi τ ). (5)
deep under the ocean surface [21]. Further details on the tidal i=1
current models, representing turbulence and wave scenarios, Hence, the superposition of SDE processes enables the model-
for the case study, are presented in Section III-C. ing of any Gaussian stochastic process, with an autocorrelation
that can be modeled as a weighted sum of decaying exponen-
III. S TOCHASTIC M ODELING tials.
This section provides a quick introduction to the SDEs used
Traditionally, power systems are modeled as a set of in the case study of this paper. The interested reader can find
Differential-Algebraic Equations (DAEs). When stochastic further details on SDEs and the technique above in [12], [22].
perturbations are considered the system model becomes a set The individual stochastic models used to represent the loads,
of Stochastic Differential-Algebraic Equations (SDAEs), as wind speeds and tidal current speeds in the case study are
discussed in [6]. The stochastic processes for these simulations presented next.
are modeled using Stochastic Differential Equations (SDEs). A. Load modeling
SDEs have been utilized in previous power systems studies as
The stochastic load model considered in this paper is
outlined in Section I-B.
developed based on the well-known voltage dependent load
The Ornstein-Uhlenbeck (OU) process is one of the most
model coupled with OU processes, as follows [6]:
widely known SDEs because of its simplicity and versatility.
The stochastic processes for the simulations presented in the pL (t) = (pL0 + ηp (t))(v(t)/v0 )k ,
case study are built using OU SDEs. The OU process is a qL (t) = (qL0 + ηq (t))(v(t)/v0 )k ,
mean reverting Gaussian process. The general form of a SDE (6)
dηp (t) = αp (µp − ηp (t))dt + σp dW (t),
defining a OU process is:
dηq (t) = αq (µq − ηq (t))dt + σq dW (t),
dη(t) = α(µ − η(t))dt + σdW (t), (1) where pL (t) and qL (t) are the active and reactive power of the
load, respectively, and pL0 and qL0 are parameters representing
where α, σ > 0 and W (t) is a Wiener process. α is the mean active and reactive load powers at t = 0. v(t) is the voltage
reversion speed of the process, η(t), which defines the slope magnitude at the bus where the load is connected and v0 is
of its exponentially decaying autocorrelation. The process η(t) the value of this voltage magnitude at t = 0.
is Gaussian distributed with mean µ and variance σ 2 /(2α). The model (6) can, through the exponent k define whether
The technique proposed in [12] is used to model the the load is a constant power load (k = 0), a constant current
stochastic processes with an autocorrelation which is not load (k = 1) or a constant impedance load (k = 2). The
purely exponentially decaying. It consists in the superposition variability is modeled through the stochastic processes ηp (t)
of OU processes, as defined in (1), to capture the desired and ηq (t) which are formulated as OU processes, where the

21st Power Systems Computation Conference Porto, Portugal — June 29 – July 3, 2020
PSCC 2020
 TABLE I: The SDE parameters for the wind and tidal current speed
parameters α, µ and σ have the same meaning as in (1). In the
models used in the case study.
case study, the uncertainty is set as 10 % of the nominal load
power and the mean reversion speed is set to αp = αq = 0.02. SDE process Parameters
Wind speed w1 = 0.35 α1 = −0.3
B. Wind speed modeling w2 = 0.5 α2 = −0.03
The wind speed model considered in this paper consists of w3 = 0.15 α3 = −0.0001
two parts: a constant mean wind speed, vc and a Gaussian Tidal current speed w1 = 0.23 α1 = −5
stochastic process, ρw (t). The wind is modeled in this way w2 = 0.32 α2 = −0.2
since the wind speed variations within a 10 minute time frame w3 = 0.45 α3 = −0.04
can be assumed to be Gaussian distributed around a certain
mean vc [23]. The wind speed model used for each wind farm
in the Irish system is: b. The stochastic turbulence, ρt (t), in the current speed is
vwind (t) = vc + ρw (t), (7) modeled using a stochastic process as defined in (2). It
is defined in the same way as for the wind speed (see
where vwind (t) is the modeled wind speed time-series and Section III-B). The only difference is that the standard
ρw (t) is a stochastic process as defined in (2). Thus, ρw (t) deviation of the tidal current speed is set to be 10 % of
has the probability distribution N (µρw , σρ2w ) and an autocor- the current speed, vct , and the autocorrelation parameters
relation as in (5). are set based on data analysis in [13]1 . The parameters
The following assumptions are made for the wind farm are shown in Table I.
modeling: c. The third part represents the effect of waves on the tidal
• All wind farms are assumed to be uncorrelated. This is current, vwaves (t).
a fair assumption as short-term wind speed fluctuations The resulting tidal current speed model is the following:
are uncorrelated for larger distances.
• Wind speed is location dependent. However, since the vtide (t) = vct + ρt (t) + vwaves (t). (9)
details of the exact locations of wind generators are
The ocean sea state is affected by a range of waves at the
not available to the authors, the wind speed, for all
same time. These waves are generally modeled using the first
locations, is modeled using the same model parameters.
order Stokes model representing a random sea-state [25]:
The parameters for the autocorrelation are set as shown
in Table I based on previous data analysis in [12] of the N
X cosh[ki (h + d)]
data set available in [19]. vwaves (t) = ai ωi cos[ωi t−ki x+φi ], (10)
sinh(ki d)
• The standard deviation of the process is set to be 20 % of i=1
the mean wind speed vc as supported by [18] and analysis where h is the vertical distance from the sea surface to the hub
of the data set in [19]. height of the tidal turbine, positive upwards, and d is the sea
• The damping effect of the rotor blades is modeled through depth. φi are random phases, uniformly distributed between 0
a low-pass filter as presented in [24]. The time constant, and 2π, ωi is the frequency of the i-th frequency component,
Tb , of the filter is dependent on the radius of the rotor. ki is the wave number of the i-th frequency component. Finally
In the case study, Tb = 10 s for all wind power plants as p
the rotor size of individual turbines is unknown. ai = 2S(ωi )∆ωi (11)
• The aggregated wind speed is defined through the σρw
parameter: is the amplitude of the i-th frequency component, defined from
√ the frequency spectrum, S(ω), of the waves.
σρw = 0.2 · vc / 2nturb , (8)
The frequency spectrum considered in this paper is the
where nturb is the number of turbines in the wind farm. JONSWAP spectrum [25]. The wave angular frequency, ωi ,
Thus, the standard deviation of the modeled wind speed is within the frequency band, ∆ωi , and N different frequency
decreases in proportion to the number of turbines, as the components are considered to represent the random sea-state.
variability of wind speed averages out over a spread wind The JONSWAP spectrum is defined as:
farm.
mg 2
  ω 4 
p
S(ω) = 5 exp − 1.25 γY , (12)
C. Tidal current speed modeling ω ω

The proposed tidal current speed model consists of three where g is the acceleration due to gravity, ωp is the peak
parts. frequency of the spectrum and γ is the peak enhancement
a. The variations in the current due to the tidal phenomenon factor which controls the sharpness of the peak. m is the
are very slow and relatively small in the time-frame of 1 Both the wind and tidal current speed model parameters are set based on
seconds to minutes. Thus, the predicted tidal current measured data, where the fluctuations have been separated from the rolling
speed is modeled as a constant, vct . average of the data, over a 10 minute time frame.

21st Power Systems Computation Conference Porto, Portugal — June 29 – July 3, 2020
PSCC 2020
intensity of the spectrum and can be defined for North Sea Location 1

applications [25] as:

 2
Hs
m = 5.058 (1 − 0.287lnγ), (13)
Tp2

where Hs is the significant wave height, Tp = 2π/ωp is the

peak wave period.
Waves can be thought of as an intermittent disturbance.
They are not always present and are not likely to effect the Location 4
different locations at the same time, with the same intensity.
Thus, the Stokes model, coupled with the JONSWAP spec-
trum, is set to represent different wave scenarios, as discussed Location 2

in Section IV-C.
Note that to take into account the aggregation of the full
tidal farm in each location the aggregation model for the wave
component of the tidal current, presented in [26] is utilized. Location 3

IV. C ASE STUDY

Fig. 3: The top four locations of potential tidal generation as identified
A. All-Island Irish Transmission System
in [2].
The Irish system is a relatively-small isolated transmission
system. It consists of two separate 50 Hz grids which are
AC interconnected and operated by independent TSOs: System B. System Model
Operator for Northern Ireland (SONI) and Eirgrid Group. The The Irish transmission system model used for this case
current transmission peak of the system is about 5, 500 MW study consists of 1, 479 buses, 1, 851 transmission lines and
and the demand is forecasted to grow between 22 % and 53 % transformers, 245 loads, 22 conventional synchronous power
by 2030 [27]. The system has two HVDC interconnections, to plants, modeled with 6th order synchronous machine models,
Scotland and Wales, both with a capacity of 500 MW, flowing with AVRs and turbine governors, 6 PSSs, and 169 wind power
in both directions. plants, of which 159 are DFIGs and 10 are CSWTs. In the
The installed wind generation has been rapidly increasing system for this case study the total load of the system is
over the last 10 years. At present, the record of produced wind 2, 215 MW and 25 % of the total generated power is supplied
power in the Irish system is approximately 3, 500 MW and the by wind.
system is able to handle upto 65 % renewable energy, primarily The tidal power plants are assumed to be installed in the
composed of wind power [28]. four locations shown in Fig. 3. The combined generation of
Ireland had the world’s first large scale commercial tidal these four locations is set to be 10 % of the total generation
stream generators in operation from 2003 until 2017. Those in the system. The local wind generation in each location is
turbines were a part of the SeaGen tidal test project and were substituted with tidal generation. The set power of the tidal
located in Northern Ireland’s Strangford Lough, producing generation at each location is based on their potential energy
1.2 MW [15]. This project has provided valuable information and is shown in Table II.
for further development of tidal generation in Irish waters The system is modeled as a set of SDAEs where the
and has shown that tidal generation is a viable energy source loads, wind and tidal generation are modeled using SDEs, as
for the Irish system. In [2], Ireland’s tidal resources were presented in Section III. All simulations were carried out using
assessed. Eleven practical resource sites for tidal generation Dome, a Python-based software tool for power system analysis
were identified. The four sites with the greatest estimated [29]. Dome solves the SDEs using the Itô integral. It supports
capacity are shown in Fig. 3. Table II shows the potential solving the SDEs using either the Euler-Maruyama or Milstein
energy of the four locations. integration method. For this case study the Euler-Maruyama
is used [22].
TABLE II: The potential energy and installed power at the four
modeled tidal power locations [2].
C. Scenarios
Potential Installed
Location Name
Energy [GWh/y] Power [MW] Five scenarios for the Irish power system model are studied:
1 Inishtrahull Sound 514 67
2 Shannon Estuary 367 34 • Scenario 1: The load is stochastic and modeled as
3 Tuskar Rock 420 47 presented in Section III-A. This serves as a base scenario.
4 Codling 791 69 Thus, the same stochastic load models are used in the
remaining scenarios.

21st Power Systems Computation Conference Porto, Portugal — June 29 – July 3, 2020
PSCC 2020
 • Scenario 2: The onshore wind, 25 % of generation, is for time lags h = 0 − 100 s, where xt is the COI frequency at
stochastic, modeled as presented in Section III-B. time t. Then, the standard deviation of the ramp rates, ∆h xt ,
• Scenario 3: Stochastic onshore wind, 15 % of generation, for each time step h is computed.
and stochastic offshore wind, 10 % of generation.
• Scenario 4: Stochastic onshore wind, 15 % of generation,
D. Wind vs Tidal
and stochastic tidal, 10 % of generation, modeled as The Irish system with all the loads modeled as stochastic
presented in Section III-C, without waves. (Scenario 1), as presented in Section III-A, is studied first. This
• Scenario 5: Same as Scenario 4, that is stochastic scenario serves as a reference for the remaining scenarios. To
onshore wind, 15 % of generation, and stochastic tidal, validate the stochastic load scenario, frequency data from the
10 % of generation, but the tidal generation is disturbed Irish system, gathered in the AMPSAS project is used. The
by waves. frequency is measured with a 10 Hz sampling frequency. In
In Scenario 3 - 5 offshore wind and tidal generation has this case, the morning of a singular day, that is the 20th of
been installed in the locations identified in Fig. 3, replacing May in 2014, is considered. During this time period there was
10 % of the local onshore wind generation. almost no wind generation, hence the stochastic fluctuations in
Scenario 5 includes three cases, where waves effect the tidal the frequency are mainly due to loads, as well as dispatches.
current, presented in Table IV. Typically, the sea surface is Further details on these measurements are provided in [31].
assumed to be stationary for 20 minutes, upto a couple of Fig. 4 shows the standard deviation of the ramp rates of
hours. A stationary sea-state can be characterized by a set of the measured frequency data and the COI frequency, for the
parameters. These are the significant wave height, Hs , and the simulated Scenario 1. The model underestimates the frequency
peak wave period, Tp . The significant wave height is defined ramps in the initial 20 s, when compared to the measured data.
as the mean wave height (trough to crest) of the highest third However, for higher time steps, the model, for Scenario 1, and
of the waves. The peak wave period is the wave period with the measured data are in agreement.
the highest energy. The wave parameters for the considered
sea-states are based on those listed in Table 3.17 in [25]. The
Standard Deviation of Ramps in Frequency [Hz]

parameters utilized for different sea-states in the case study are 0.014

shown in Table III, where Hs refers to the significant wave 0.012

height and Tp to the peak wave period; and Tmin − Tmax 0.010
gives the range of wave periods, sampled from the JONSWAP 0.008
spectrum in (12), for the Stokes wave model in (10).
0.006
TABLE III: The three sea-states considered for Scenario 5. Measured Data
0.004 Scenario 1
Sea-state Hs [m] Tp [s] Tmin − Tmax [s]
0.002 Scenario 2
(S) Small waves 1 5 1.4 − 8.8
Scenario 3
(M) Moderate waves 2.5 7 2.8 − 13.5 0.000 Scenario 4
(L) Large waves 5 10 3.8 − 15.5
−0.002 Scenario 5 − a

The three cases for Scenario 5 shown in Table III also 0 20 40 60 80 100
Time [s]
take into account the locations of the four tidal power plants.
The largest significant wave heights, of about 5 meters, are Fig. 4: The standard deviation of the ramps in the frequency of the
COI for Scenario 1-5, as well as of the measured frequency [31].
seen on the west coast during winter. On the other hand, Ramp rates are computed as in (14).
the average significant wave heights, in the Irish sea, do not
exceed 2 meters, during any season [30]. Thus, Location 1 & The five different scenarios, as presented in Section IV-C,
2 are more likely to experience higher significant wave heights are compared in Fig. 4. The inclusion of stochastic wind, in
than Location 3 & 4, as they are facing the Atlantic ocean. Scenario 2, increases the standard deviation of the ramp rates
Specifically, Location 1 is likely to experience the worst wave by approximately 25 %. An even bigger change is seen for
conditions. Table IV shows the sea-state at each location, for Scenario 3-5. For these scenarios, however, the generation
the three cases of Scenario 5. profile of the system had to be modified to accommodate
TABLE IV: The sea-state at each location, for the three cases offshore wind or tidal power plants.
considered for Scenario 5. The wind generation in Ireland is mostly composed of small
Case\Location 1 2 3 4 farms spread across the system. For Scenario 3-5, these small
a S S S S
b M M S S farms are replaced by four much larger offshore wind or tidal
c L M S S farm. This causes relatively larger power fluctuations to be
introduced in four points in the system. Whereas, for Scenario
The ramp rates of the Center of Inertia (COI) frequency are 2, smaller power fluctuations are installed in locations spread
utilized to compare the scenarios. Ramp rates are computed across the area.
as: Fig. 4 shows that Scenario 4 has slightly smaller frequency
∆h xt = xt − xt−h (14) variations than Scenario 3. This indicates that offshore wind

21st Power Systems Computation Conference Porto, Portugal — June 29 – July 3, 2020
PSCC 2020
would introduce larger frequency variations than tidal genera- Maximum Power Point Tracking (MPPT) and set the power
tion. However, the set up of the farms, the number of turbines output based on the deviation of the measured frequency
and the turbine sizes also impact the stochastic variations in (droop control) and/or Rate of Change of Frequency (ROCOF)
the output power of the farm. Thus, it cannot be claimed control. The combination of the two strategies proposed in [32]
that either source introduces less frequency variations. It can for wind turbines is considered below.
only be stated that tidal generation, when compared to wind The droop controller, with gain 1/R, is comparable to
generation, if no waves are presents, does not necessarily the primary frequency controller of a synchronous machine.
introduce bigger system frequency variations. The ROCOF controller consists of a low-pass filter with time
The trajectories of the standard deviation of the frequency constant Tl , the time derivative of the frequency measurement
ramps for Scenario 5-a are shown in Fig. 4. In this case, all and a gain Kl . The two controllers are complementary. The
tidal farms are subject to relatively small waves simultane- ROCOF control is faster and has its main effect in the very first
ously. This is the most common wave scenario for the Irish instants after the frequency drop. However, the droop control
system. The inclusion of waves in the tidal current results in is slower and mitigates the frequency deviation [32].
oscillatory fluctuations in the power outputs of the tidal farms, Fig. 6 shows the comparison of the three different cases
with a time period of less than 10 seconds. This increases the for Scenario 5 with the inclusion of frequency control in the
frequency variability, for the initial 20 seconds and introduces four tidal power plants. Compared to the results presented in
oscillations in the system frequency. Fig. 5, frequency variations are approximately halved in size
Scenario 5-b represents moderate waves in both Location for all three cases. For comparison, Fig. 6 also shows Scenario
1 & 2, which are facing the Atlantic ocean. In this case, 2 and 3, i.e. the two scenarios with only wind generation. The
frequency ramps have a standard deviation of about 0.015 Hz, frequency control of the tidal turbines effectively mitigates the
within a few seconds, as shown in Fig. 5. This state is likely frequency variations for both Scenario 5-a and 5-b to be less
to last for 20 minutes upto a few hours at a time. than those for Scenario 3, excluding the first few seconds.
Scenario 5-c represents the worst case scenario. In this Therefore, coupling frequency control to tidal generation can
case Location 1 is experiencing large waves, with longer time reduce the frequency variations, due to the waves, to an
periods, that is swell waves. This case is characterized by acceptable level.
unacceptable frequency variations. Energy storage systems can
be installed along side tidal generation, as suggested in [21] to
Standard Deviation of Ramps in Frequency [Hz]

0.016
mitigate these fluctuations. An alternative solution is to include
primary frequency control in the tidal generation as discussed 0.014
in the next section. 0.012

0.010
0.030
Standard Deviation of Ramps in Frequency [Hz]

0.008
0.025 Scenario 2
0.006
Scenario 3
0.020 0.004 Scenario 5 − a
Scenario 5 − b
0.002
0.015 Scenario 5 − c
0.000
0 20 40 60 80 100
0.010 Time [s]

Scenario 5 − a Fig. 6: Standard deviation of the ramps in the frequency of the COI
0.005 Scenario 5 − b for Scenario 5, case a, b and c with frequency control. Scenarios 2
Scenario 5 − c and 3 are shown for comparison. Ramp rates are computed as in (14).
0.000
0 20 40 60 80 100
Time [s] As a final remark, it is important to note that the sea
Fig. 5: The standard deviation of the ramps in the frequency of the state is strongly dependent on the location, as discussed in
COI for Scenario 5, case a, b and c. Ramp rates are computed as in Section IV-C. Thus, the location is the first design parameter
(14). to be considered to reduce the impact of waves on tidal
generation. If the potential tidal locations are all prone to
extreme wave conditions, it is key to determine the optimal
E. Frequency control of tidal generation placement for the tidal power plants to minimize the impact
In this section, frequency control of tidal generation is of waves on the system. In the case of the Irish system,
considered, to reduce the frequency variability caused by the for Location 1 and 2, Scenario 5-c represents the worst case
waves of Scenario 5-c. The similarity between tidal and wind scenario, while for Location 3 and 4 the frequency variations
turbines allows for the the frequency control implemented for due to wave disturbance would not get bigger than those shown
wind turbines to be adapted for tidal turbines. A common for Scenario 5-a. Thus, it appears sensible to commission tidal
approach for wind turbine frequency control is to bypass the installments on the east coast first, e.g. Locations 3 and 4.

21st Power Systems Computation Conference Porto, Portugal — June 29 – July 3, 2020
PSCC 2020
 V. C ONCLUSIONS [16] NOAA - Tides and currents, “Current data, February 15, 2019.”
[Online]. Available: https://tidesandcurrents.noaa.gov/cdata/ [Accessed
This paper discusses the frequency fluctuations of the Irish in May 2019]
[17] Met Éireann, “The Irish meterological service online, historical
system due to wind and/or tidal generation for different scenar- data,” 2019. [Online]. Available: https://www.met.ie/climate/available-
ios. Simulation results indicate that tidal generation does not data/historical-data/ [Accessed in May 2019]
introduce higher frequency variations than wind, except when [18] A. I. Winter, “Differences in fundamental design drivers for wind and
tidal turbines,” in IEEE OCEANS, Spain, 2011.
waves are present in the tidal current. In that case, the output [19] RE<C: Renewable Electricity Less Than Coal, “Surface Level Wind
of the tidal generation fluctuates periodically, which results Data Collection,” 2011. [Online]. Available: https://code.google.com/
in significant fluctuations of the frequency of the system. archive/p/google-rec-csp/ [Accessed in March 2019]
[20] The Reliable Data Acquisition Platform for Tidal energy (ReDAPT),
The paper also shows the such frequency fluctuations can be “Met-Ocean Data Science for Offshore Renewable Energy Applica-
mitigated by including proper frequency control in the tidal tions),” 2017. [Online]. Available: http://redapt.eng.ed.ac.uk/index.php
turbines. [Accessed in February 2019]
[21] Z. Zhou, M. Benbouzid, J. F. Charpentier, F. Scuiller, and T. Tang,
Future work will focus on other renewable energy sources “A review of energy storage technologies for marine current energy
for the Irish system, such as generation based on wave energy, systems,” Renewable and Sustainable Energy Reviews, vol. 18, pp. 390–
and compare their dynamic modeling and impact on the grid 400, 2013.
[22] P. E. Kloeden, E. Platen, and H. Schurz, Numerical solution of SDE
with respect to offshore wind and tidal generation. through computer experiments. Springer Science & Business Media,
2012.
R EFERENCES [23] B. C. Cochran, “The influence of atmospheric turbulence on the kinetic
energy available during small wind turbine power performance testing,”
Ceder-Ciemat. Spain April, 2002.
[1] E. Robinson and D. McDonnell, “Ireland’s potential role in the develop- [24] T. Petru and T. Thiringer, “Modeling of wind turbines for power system
ment of offshore renewables: Operations and maintenance,” Engineers studies,” IEEE Transactions on Power Systems, vol. 17, no. 4, pp. 1132–
Journal, May 1, 2018. 1139, 2002.
[2] F. O’Rourke, F. Boyle, and A. Reynolds, “Tidal current energy resource [25] S. Chakrabarti, Handbook of Offshore Engineering (2-volume set).
assessment in Ireland: Current status and future update,” Renewable and Elsevier, 2005.
Sustainable Energy Reviews, vol. 14, no. 9, pp. 3206–3212, 2010. [26] A. N. Einrı́, G. M. Jónsdóttir, and F. Milano, “Modeling and con-
[3] E. Segura, R. Morales, J. Somolinos, and A. López, “Techno-economic trol of marine current turbines and energy storage systems,” IFAC-
challenges of tidal energy conversion systems: Current status and PapersOnLine, vol. 52, no. 4, pp. 425–430, 2019.
trends,” Renewable and Sustainable Energy Reviews, vol. 77, pp. 536– [27] EirGrid Group, “All-Island Generation Capacity Statement 2018-
550, 2017. 2027,” 2018. [Online]. Available: http://www.eirgridgroup.com/site-
[4] L. Buchsbaum, “MeyGen Array Sets Global Record for Harnessing Tidal files/library/EirGrid/Generation Capacity Statement 2018.pdf [Ac-
Power,” September 2018. [Online]. Available: https://www.powermag. cessed in September 2019]
com/meygen-array-sets-global-records-for-harnessing-tidal-power/ [Ac- [28] ——, “EirGrid Group achieves record level of variable
cessed in September 2019] renewable energy on Irish electricity system,” April 13, 2018.
[5] J. Widén, N. Carpman, V. Castellucci, D. Lingfors, J. Olauson, F. Re- [Online]. Available: http://www.eirgridgroup.com/newsroom/record-
mouit, M. Bergkvist, M. Grabbe, and R. Waters, “Variability assessment renewable-energy-o/ [Accessed in September 2019]
and forecasting of renewables: A review for solar, wind, wave and tidal [29] F. Milano, “A Python-based software tool for power system analysis,”
resources,” Renewable and Sustainable Energy Reviews, vol. 44, pp. IEEE PES General Meeting, Vancouver, 2013.
356–375, 2015. [30] S. Gallagher, R. Tiron, and F. Dias, “A long-term nearshore wave
[6] F. Milano and R. Zárate-Miñano, “A systematic method to model hindcast for Ireland: Atlantic and Irish Sea coasts (1979–2012),” Ocean
power systems as stochastic differential algebraic equations,” IEEE Dynamics, vol. 64, no. 8, pp. 1163–1180, 2014.
Transactions on Power Systems, vol. 28, no. 4, pp. 4537–4544, 2013. [31] M. Adeen, G. M. Jónsdóttir, and F. Milano, “Statistical correlation
[7] B. Yuan, M. Zhou, G. Li, and X.-P. Zhang, “Stochastic small-signal sta- between wind penetration and grid frequency variations in the irish net-
bility of power systems with wind power generation,” IEEE Transactions work,” in IEEE International Conference on Environment and Electrical
on Power Systems, vol. 30, no. 4, pp. 1680–1689, 2014. Engineering, 2019.
[8] H. Li et al., “Analytic analysis for dynamic system frequency in [32] J. Cerqueira et al., “Comparison of the dynamic response of wind turbine
power systems under uncertain variability,” IEEE Transactions on Power primary frequency controllers,” in IEEE PES General Meeting, Chicago,
Systems, vol. 34, no. 2, pp. 982–993, 2018. 2017.
[9] W. Wu, K. Wang, G. Li, and Y. Hu, “A stochastic model for power
system transient stability with wind power,” in IEEE PES General
Meeting, Washington, 2014.
[10] H. Verdejo, A. Awerkin, W. Kliemann, and C. Becker, “Modelling
uncertainties in electrical power systems with stochastic differential
equations,” International Journal of Electrical Power & Energy Systems,
vol. 113, pp. 322–332, 2019.
[11] R. Zárate-Miñano, M. Anghel, and F. Milano, “Continuous wind speed
models based on stochastic differential equations,” Applied Energy, vol.
104, pp. 42–49, 2013.
[12] G. M. Jónsdóttir and F. Milano, “Data-based continuous wind speed
models with arbitrary probability distribution and autocorrelation,” Ren-
wable Energy, vol. 143, pp. 368–376, 2019.
[13] ——, “Modeling of short-term tidal power fluctuations,” IEEE Transac-
tions on Sustainable Energy, 2019.
[14] K. Wang and M. L. Crow, “Numerical simulation of stochastic differen-
tial algebraic equations for power system transient stability with random
loads,” IEEE PES General Meeting, Detroit, 2011.
[15] G. Savidge et al., “Strangford Lough and the SeaGen tidal turbine,”
in Marine renewable energy technology and environmental interactions.
Springer, 2014, pp. 153–172.

21st Power Systems Computation Conference Porto, Portugal — June 29 – July 3, 2020
PSCC 2020