A review of numerical modelling of wave energy

converter arrays
Matt Folley, Aurélien Babarit, Ben Child, David Forehand, Louise O’Boyle,
Katie Silverthorne, Johannes Spinneken, Vasiliki Stratigaki, Peter Troch

To cite this version:

Matt Folley, Aurélien Babarit, Ben Child, David Forehand, Louise O’Boyle, et al.. A review of
numerical modelling of wave energy converter arrays. ASME 2012 International Conference on Ocean,
Offshore and Artic Engineering (OMAE2012), Jun 2012, Rio de Janeiro, Brazil. �10.1115/OMAE2012-
83807�. �hal-01202077�

HAL Id: hal-01202077

https://hal.archives-ouvertes.fr/hal-01202077
Submitted on 7 Mar 2019

HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est

archive for the deposit and dissemination of sci- destinée au dépôt et à la diffusion de documents
entific research documents, whether they are pub- scientifiques de niveau recherche, publiés ou non,
lished or not. The documents may come from émanant des établissements d’enseignement et de
teaching and research institutions in France or recherche français ou étrangers, des laboratoires
abroad, or from public or private research centers. publics ou privés.
 Proceedings of the ASME 2012 31st International Conference on Ocean, Offshore and Arctic Engineering
OMAE2012
July 1-6, 2012, Rio de Janeiro, Brazil

OMAE2012-83807

A REVIEW OF NUMERICAL MODELLING OF WAVE ENERGY CONVERTER ARRAYS

Matt Folley Aurélien Babarit Ben Child

Queen’s University Belfast LUNAM Université Garrad Hassan
Belfast, Northern Ireland Ecole Centrale de Nantes - CNRS Bristol, UK
Nantes, France

David Forehand Louise O’Boyle Katie Silverthorne

Edinburgh University Queen’s University Belfast Queen’s University Belfast
Edinburgh, UK Belfast, Northern Ireland Belfast, Northern Ireland

Johannes Spinneken Vasiliki Stratigaki Peter Troch

Imperial College, London Ghent University Ghent University
London, UK Ghent, Belgium Ghent, Belgium

ABSTRACT for modelling wave energy converter arrays are described in

Large-scale commercial exploitation of wave energy is this paper, using a common framework. This allows a
certain to require the deployment of wave energy converters qualitative comparative analysis of the different methods to be
(WECs) in arrays, creating ‘WEC farms’. An understanding of performed at the end of the paper. This includes consideration
the hydrodynamic interactions in such arrays is essential for of the conditions under which the models may be applied, the
determining optimum layouts of WECs, as well as calculating output of the models and the relationship between array size
the area of ocean that the farms will require. It is equally and computational effort. Guidance for developers is also
important to consider the potential impact of wave farms on the presented on the most suitable numerical method to use for
local and distal wave climates and coastal processes; a poor given aspects of WEC farm design. For instance, certain
understanding of the resulting environmental impact may models are more suitable for studying near-field effects, whilst
hamper progress, as it would make planning consents more others are preferable for investigating far-field effects of the
difficult to obtain. It is therefore clear that an understanding the WEC farms. Furthermore, the analysis presented in this paper
interactions between WECs within a farm is vital for the identifies areas in which the numerical modelling of WEC
continued development of the wave energy industry. arrays is relatively weak and thus highlights those in which
To support WEC farm design, a range of different future developments are required.
numerical models have been developed, with both wave phase-
resolving and wave phase-averaging models now available. 1. INTRODUCTION
Phase-resolving methods are primarily based on potential flow As development continues in WEC technology there is an
models and include semi-analytical techniques, boundary increasing interest in investigating how WECs interact with one
element methods and methods involving the mild-slope another and the environment when they are deployed in an
equations. Phase-averaging methods are all based around array. This understanding is vital to support wave farm design
spectral wave models, with supra-grid and sub-grid wave farm as commercialisation of WEC technologies progresses. In
models available as alternative implementations. recognition of this the Wave Energy Converter Array Network
The aims, underlying principles, strengths, weaknesses and (WECAN) was formed in 2010 as an international forum for
obtained results of the main numerical methods currently used researchers and developers active in the field of WEC arrays to

1
discuss relevant current research and development so as to WEC array numerical modelling is discussed and the
provide guidance and expert opinion on WEC arrays. This requirements for further research and development identified.
paper is the result of a review of WEC array numerical
modelling techniques held at the WECAN meeting in Ghent, 2. POTENTIAL FLOW MODELS
Belgium in October 2011.
A significant area of current research and development in 2.1 Boundary element methods
WEC arrays is their numerical modelling. The first numerical To model the wave-structure interaction of a single WEC
models of WEC arrays were developed in the late 1970s and with the incident wave field, the present state-of-the-art is to
were based on potential flow models. Based on the number of use Boundary Element Method (BEM) based numerical codes
publications, this numerical modelling technique remains the such as the well known WAMIT, ANSYS Aqwa, Aquaplus
most popular method for determining interactions between amongst others. When it comes to arrays of WECs, these
WECs; however, in the last 5 years a number of alternative numerical tools are theoretically able to deal with any number
numerical modelling techniques have been developed or are of devices without restrictions except the ones related to the use
being developed that provide alternatives for assessing the of linear potential theory.
WEC interactions as well as the potential environmental impact As a brief summary, potential flow methods are based upon
of a wave farm. These techniques include the use of the following assumptions:
Boussinesq or mild-slope wave models, the use of spectral • The fluid is inviscid.
wave models and also the use of nonlinear boundary element • The flow is irrotational. Therefore, there exists a velocity
and CFD models. potential φ(M,t) from which the velocity can be derived
It would be wrong to assume that these alternative r r
numerical modelling techniques for WEC arrays may be exact everywhere in the fluid domain: V ( M , t ) = ∇φ ( M , t ) .
replacements for potential flow models and that there is a single • The flow is incompressible. Adopting mass conservation,
best numerical modelling technique for WEC arrays. Each of this assumption leads to Laplace’s equation:
∆φ ( M , t ) = 0 everywhere in the fluid domain.
the four identified basic types of WEC array numerical
modelling techniques has a certain set of characteristics that
make it more or less suitable for particular modelling Formulating a set of boundary conditions that satisfy the
requirements. Even within these basic types there are sub-types Laplace equation results in a nonlinear Boundary Value
which offer further possibilities to optimise the numerical Problem which remains challenging to solve (nonlinear
modelling approach for a particular case study. potential flow formulations are discussed below). As a result,
To assist in the comparative analysis of the WEC array the problem is usually simplified further by being linearised.
numerical modelling techniques, each of the techniques is The two assumptions for linearisation are:
described and then assessed using a common set of defining • The ratios of wave height to wavelength (i.e. wave
characteristics. To assist analysis, these defining characteristics steepness) and wave height to water depth must both be
are separated into three basic types: fundamental modelling much smaller than 1.
characteristics, computational processing characteristics and • The motions of the body are small and around a fixed
usability characteristics. The fundamental modelling mean position: the ratio of the typical amplitude of motion
characteristics include the assumptions inherent in the model, to the typical dimension of the body is much smaller
together with the consequential limitations, strengths, than 1.
weaknesses and issues. The computational processing The next step is to transform the volumic problem into a
characteristics include the factors that define the computation surfacic problem by making use of Green’s second identity.
effort such as the model complexity, the number of WECs and Then, by using an adequate Green’s function, the problem can
spatial extent. Finally, the usability characteristics include the be discretised and solved numerically, usually in the frequency-
required skill of the user, the degree of ease of use and domain (which is possible as a consequence of the
availability of suitable software (including cost and user linearisation). Figure 1, taken from Borgarino et al. [1], shows
friendliness). an example of such a calculation for an array composed of two
This paper starts by reviewing all of the current WEC array clusters of 8 floating OWSCs (Oscillating Wave Surge
numerical modelling techniques and is followed by a Converters). The figure shows the normalised perturbation of
comparative analysis of the techniques identified. This the incident significant wave height.
comparative analysis uses the defining characteristics
mentioned above to identify which techniques are most suitable
for particular numerical modelling tasks. The tasks considered
include: evaluation of localised interactions and impacts,
evaluation of WEC array control strategy, estimation of power
productivity of a small WEC array (2-10 units), estimation of
power productivity of a wave farm (100+ units) and assessment
of distal environmental impact. Finally, the state-of-the-art in

2
 these techniques are based upon linear wave theory and are
2000 1.1
1.08
therefore subject to the relevant assumptions of Section 2.1.
1.06 The principal assumption in the point absorber method [11,
1500 1.04 12] is that wave energy absorbers are sufficiently small
1.02 compared to their separation and to the incident wavelength
1000 1
that the far-field radiation pattern from each one is not
Y

0.98
0.96 significantly affected by the presence of the other absorbers.
500 0.94 Hence, the diffraction of radiated and diffracted waves from
0.92 each device by others in the array is neglected. Typically
0.9
0 further assumptions, that the devices are all identical bodies of
0 1000 2000 3000 4000 revolution, oscillating vertically and optimally controlled, are
X
Figure 1: Ratio of the significant height of the total wave field also made. The resulting solution for the quantity of maximum
around an array of two clusters of 8 floating OWSCs with power absorption by the array may be written in terms of the
respect to the significant height of the incident wave. Sea state inverse of a square matrix of order N (the number of absorbers
is Hs=1 m, Tp = 8 seconds. Picture is taken from [1] in the array).
The plane wave method [13] requires that the spacing
Usually, assumptions related to linear potential theory apply between axisymmetric absorbers is large compared to the
well for application to WECs in small to moderate sea states; incident wavelength. This allows the diverging wave coming
comparison of numerical predictions with experiments or more from one device to be approximated as a plane wave upon
complex CFD calculations agree well (see for example Durand reaching others in the array. Unlike the point absorber
et al. [2] or Thilleul et al. [3]). However, for larger sea states, approximation, the diffraction of all waves by each element in
significant discrepancies can be observed as the linearity the array is accounted for here, albeit using an approximation.
assumptions are violated and other effects such as vortex A version of the solution that includes the effects of evanescent
shedding become important. waves [14], requires the inversion of a square matrix of order
One attractive aspect of BEM based numerical codes is that N(N-1).
they are fast when compared to CFD techniques. However, Multiple body radiation and diffraction was dealt with by
with these codes, the computational resources increase rapidly Mavrakos [15] as a succession of distinct scattering events. In
with the number of WECs in the array, as numerical complexity principle, this multiple scattering method is capable of
is proportional to the square of the number of unknowns. For providing the exact linear wave theory solution by means of an
this reason, only recently did it become possible to consider infinite summation over horizontal angular and vertical
small arrays (typically 5-10 devices) with these numerical ‘eigenfunction modes’ as well as over successive ‘orders of
models [4-9]. For arrays larger than 10 devices, CPU time and interaction’. In practice all three constants must be truncated to
memory requirements become a real issue, even for higher- a finite number (for example, to integers M, P and S
order panel methods which are available in codes like WAMIT. respectively). Although matrix inversion is not required in the
However, it should be noted that there is ongoing research into main array calculation, the number and order of matrix
how to overcome this issue, using the Fast Multipole Algorithm multiplication operations increase with N as well as M, P and S
[10]. (which in turn depend on the physical scenario under
Another limitation of the BEM approach is that constant consideration).
water depth is usually assumed over the whole array. This Kagemoto and Yue [16] provided a ‘direct matrix’ method
might be very inaccurate in the case of large arrays of WECs, to address the same problem as the multiple scattering
for which bottom induced effects might be significant (such as technique, solving for the amplitudes of all scattered waves
shoaling or refraction, for example). Theoretically, one could simultaneously without the need for iteration. This interaction
take into account the bathymetry with BEM by meshing the sea procedure may be combined with a single body solution, such
bottom, but this would correspond to an enormous number of as the numerical scheme used in [16] or another eigenfunction
additional unknowns. In this case, computational resources expansion (see [17]). In common with the multiple scattering
would, again, be an issue. approach, the accuracy of this technique is dependent on the
number of horizontal angular and vertical eigenfunction modes
2.2 Semi-analytical techniques required. The solution of the hydrodynamic problem reduces to
Much of the existing research on the subject of the inversion of a square matrix of order MNP.
hydrodynamic interactions in arrays of wave energy converters In the latter three techniques, there are additional mild
has employed semi-analytical representations of a potential assumptions on the geometrical arrangement of the array
flow solution. These involve analytical expressions that either required. Due to the use of Graf’s addition theorem, the vertical
approximate or converge to an ‘exact’ solution in the limit of an projections of interacting bodies onto a horizontal plane must
infinite series. There are several main classes of such an not overlap and a circumscribed vertical cylinder around each
approach which are described here: the ‘point absorber’, ‘plane body centred on its imaginary origin must not contain the origin
wave’, ‘multiple scattering’ and ‘direct matrix’ methods. All of of any other body.

3
 Power absorption values are available from all methods (in These functions account for the effects that persist in the free-
the form of an upper bound for the point absorber technique). surface after body motion has occurred. The matrix k(t) is the
The methods that derive velocity potentials (plane wave, inverse Fourier transform of the radiation impedance matrix
multiple scattering and direct matrix) also give rise to K(ω) = B(ω) + iω(A(ω) - A(∞)) and k(t) can be obtained from
hydrodynamic forces on body surfaces (from which motions the added damping via:
may be calculated) whilst additional theory is needed to obtain ∞
these quantities using the point absorber approximation. The k (t ) = π2 ∫ B(ω ) cosωt dω. (3)
0
potential (from the multiple scattering and direct matrix Equation (2) represents the linear time-domain equations
method) may also be used to visualise the free surface elevation of motion and it is due to Cummins [19]. Very few papers have
in the vicinity of the array. actually applied it to arrays, however it has been extensively
All of the above methods are computationally efficient used for single devices and most of the issues are common to
(compared to, say, BEMs) for small to medium sized arrays and both. For example, there is not much difference between the
some are capable of approaching the ‘exact’ linear wave theory treatment of a 6×6 k(t) matrix for a single body and a 6N×6N
solution for the special cases to which they apply. This makes k(t) matrix for an array.
them most suited for use in optimisation routines and Normally, the hydrodynamic data A(ω), B(ω), C and F(ω)
preliminary array studies. In addition to the limitations of stated are obtained from hydrodynamic codes like WAMIT or
assumptions, the accuracy of the plane wave method may be Aquaplus, with k(t) then derived from equation (3). However,
reduced in the low frequency range [4] and so the more k(t) can also be computed directly from programs like Achil3D.
accurate direct matrix method or more efficient point absorber The main problem with equation (2) is the convolution
method are now more commonly used. term. This term is not well suited for studying and designing
All of the methods described in this section have typically WEC dynamics because it is computationally demanding to
been applied to heaving point absorber device types in the past, compute directly. Fortunately, because the convolution is
although the multiple scattering and direct matrix methods are linear, it is possible to replace it by other linear time-invariant
capable of being adapted for use with other geometries. The systems like transfer functions or a state-space system. System
algorithms for the methods described here are freely available identification is used for this approximation, as discussed by
in the literature and have been implemented at several academic Taghipour et al. [20], who compares system identification in
institutions over the years. However, currently there are no the time and frequency-domains. System identification in the
commercial software tools using these techniques. time-domain involves approximating the radiation impulse
response functions k(t) and system identification in the
2.3 Time-domain formulation frequency-domain involves approximating the radiation
This subsection focuses on the linear potential flow impedance K(ω). An example of system identification in the
problems mentioned in Section 2.1, i.e. it assumes small wave time-domain is Prony’s method [21]. Prony’s method uses a
steepness (linear waves) and small body motions. However, by sum of exponential functions to approximate k(t) and this
moving from the frequency-domain into the time-domain, it is works very well for a single device. However, based on our
possible to model transient phenomena and to include nonlinear experience, it does not work so well for arrays because of the
external forces such as nonlinear viscous damping, mooring cross coupled terms for which the maximum of the response
and power take-off forces. does not happen at the initial time. An example of system
For an array of N freely floating rigid body WECs, the identification in the frequency-domain can be found in McCabe
frequency-domain equations of motion are [18]: et al. [22].
[−ω 2 (M + A(ω )) + iω B(ω ) + C]X(ω ) = F(ω ), (1) An interesting extension to the above linear time-domain
where ω is the angular wave frequency and M, A(ω), B(ω) models is reported in Babarit et al. [23] and this could also be
and C are all 6N×6N matrices. The matrices M and C are block applied to arrays. In that paper, the radiation and diffraction
diagonal matrices and represent the global mass matrix and forces remain calculated by linear potential theory but the
matrix of hydrostatic and gravitational restoring coefficients, Froude-Krylov force (i.e. the sum of incident wave and
respectively. The 6×6 blocks on each diagonal represent the hydrostatic forces) is computed on the exact wetted surface.
corresponding body matrices. The matrices A(ω) and B(ω) are This means that important nonlinearities are taken into account
full matrices and are the added mass and added damping and although there is a moderate increase in computational
matrices respectively. The 6N×1 matrices (column vectors) time, the agreement with experimental results is improved.
X(ω) and F(ω) are the Fourier transforms of the body motions
x(t) and wave excitation forces f(t), respectively. 2.4 Nonlinear potential flow models
Taking the inverse Fourier transform of equation (1) yields: Like all potential flow models, nonlinear potential flow
t models rely on the assumptions of incompressible, irrotational
(M + A(∞))&x&(t ) + ∫ k (t − τ ) x& (τ ) dτ + Cx(t ) = f (t ), (2) and inviscid flow. Most commonly, these models are
0
implemented using a Boundary Element Method (BEM). In
where A(∞) is the added mass matrix at infinite frequency and contrast to the linear BEM frequency-domain models discussed
k(t) is the matrix of radiation impulse response functions.
above, the nonlinear counterparts operate in the time-domain.

4
In early BEM schemes the free surface at each time step was few hours of wave input and can in some circumstances
mapped to a closed contour, providing a continuous boundary become unstable.
for the evaluation of the boundary integral equation. More Compared to the Boussinesq models, the models based on
recent nonlinear BEM models are often implemented as the linear mild-slope equations are considered to be fast
Numerical Wave Tanks (NWT) in physical space, where the solvers. The latter models describe the transformation of linear
geometry of the domain is that of the physical environment. In waves when propagating from deep to shallow water.
contrast to mapped solutions, physical space offers the benefit Limitations of these mild-slope models lie in the simplifying
of a non-uniform bathymetry, and enables the definition of assumptions. Nevertheless, they have proven to be an excellent
arbitrary body geometries within the domain. tool when investigating wave penetration in harbours,
Within a nonlinear NWT, the following boundary diffraction issues, wave transformations, etc [26].
conditions are imposed: (i) a wave generation condition on the
input boundary; (ii) a radiation condition on the outflow 3.1 Boussinesq models
boundary; (iii) a no-flow condition on the bed and (iv) the Boussinesq models are based on a set of nonlinear partial
dynamic free surface boundary condition and the kinematic free differential equations known as the Boussinesq Equations. The
surface boundary condition. Further details concerning classic equations basically approximate wave propagation by
nonlinear BEM formulations can be found in Hague and Swan eliminating the vertical component of velocity but still
[24] and many others. accounting for the vertical flow structure, assuming an
The two surface conditions, (iv) above, are computed on incompressible fluid and irrotational flow. As a result of this
the instantaneous position of the free surface. As a result, this depth averaging, the use of the classic equations is limited to
formulation retains the full nonlinearity of the underlying water depths less than 0.25 times the deep water wavelength.
hydrodynamics. Likewise, the forces acting on fixed or floating Boussinesq models are usually mathematically enhanced
bodies may be computed on the instantaneous position of the versions of the classic Boussinesq equations which include the
fluid-structure intersection, again retaining the full nonlinearity effects of, deeper water depths; varying bathymetry; frequency
of the problem. In the context of floating bodies this concept is dispersion; wave breaking and moving shore line to name a
described by Kashiwagi [25]. few. One such model is the Boussinesq Wave Editor
Nonlinear potential flow formulations have been used (MIKE21BW) provided as part of the MIKE 21 suite of
extensively for the computation of extreme loads on fixed software developed by the DHI Water and Environment [27].
offshore structures and in the simulation of large vessel This model is based on the enhanced Boussinesq equations
motions. Their high computational demand has hindered the formulated by Madsen and Sørensen [28], which calculates the
simulation of large domains and arrays of wave energy free surface elevation based on flux density, rather than velocity
converters. However, with the recent availability of parallel as is the normal method for other models, resulting in improved
computing (cluster computers) this limitation is expected to stability of the simulations. The formulation includes further
vanish in the near future and simulation of small arrays (5-10 improvements allowing the theory to be extended into deep
devices) is within practical reach. In fact, nonlinear potential water with a max depth limit of 0.5 times the deepwater
flow codes are now being applied to model wave energy wavelength. The model accounts for all important wave
converter arrays as part of the PerAWaT project. transformation processes including, shoaling, refraction,
An additional application of nonlinear BEM formulations diffraction, wave breaking, bottom friction, moving shoreline,
is the provision of nonlinear hydrodynamic data as the input to partial reflection and transmission, nonlinear wave-wave
other models. They could be coupled with Smoothed-Particle interaction and frequency and directional spreading. Other
Hydrodynamics (SPH), CFD (discussed below) or the nonlinear phenomena such as surf beats and generation of sub and super
time-domain simulations discussed above. This coupling may harmonics may also be modelled, making it an ideal tool for
provide the nonlinear fluid particle kinematics in the absence of studies of harbour resonance, seiching etc.
the structure, or the nonlinear loading on a single fixed or In general, Boussinesq models are not capable of
floating structure. modelling the hydrodynamics of a moving device. However,
they may be used to model device characteristics, such as wave
3. BOUSSINESQ / MILD-SLOPE WAVE MODELS transmission reflection and absorption. If radiation
This category includes phase-resolving models which can characteristics are known, these may be included by use of an
be subdivided into models based on the linear mild-slope internal generation line, although this may become
equations and models based on the nonlinear Boussinesq cumbersome when more than 1 WEC is considered and, for this
equations. Typical applications of phase-resolving models are at reason, their use warrants caution. Outputs are in the form of
the nearshore/local scale (harbours), using smaller grid cell surface elevation and flux/velocity components within the
sizes (down to 1.0 m). The Boussinesq models seem to be model domain. It is also possible to calculate the disturbance
accurate predictors of the nearshore hydrodynamic behaviour, coefficient which is the ratio of the significant wave height at a
such as the propagation of nonlinear waves in deep to shallow particular point relative to the significant wave height at the
water. The complexity of Boussinesq models makes them input; this is commonly used for port and harbour studies.
computationally very demanding when simulating more than a

5
 Venugopal and Smith [29], carried out an investigation into the absorption coefficients or the number of absorbing cells, the
the change in wave climate around a hypothetical array of 5 degree of reflection and transmission and therefore absorption
individual bottom mounted WECs at the European Marine of the porous structure can be changed [31]. When assuming a
Energy Centre (EMEC) in the Orkney Islands. This was constant absorption coefficient S for all cells of the WEC, the
achieved by utilising the capacity to model partial transmitting amount of reflection, transmission and absorption are coupled,
and reflecting obstacles in the MIKE21BW modelling tool, as seen in [29]. To avoid this coupling, the shape of the
with a domain size of 5km by 4.5km. Differing porosity values absorption function through the WEC is changed. This way, the
were used to simulate varying degrees of reflection, absorption degree of absorption (and consequently transmission) of the
and transmission ranging from 0 (i.e. 100% transmission – no WEC, given in the power matrix of the WEC, can be tuned for
WEC in place) to 1 (i.e. 100% reflection – no transmission). a fixed amount of reflection on the WEC as specified by the
The study shows that this method may be used to identify developer.
regions of reduced and augmented wave energy in the lee of the The power absorption of a WEC typically varies with
array for particular bathymetries, wave conditions and array frequency; however it is possible, using MILDwave, to
geometries, although there is currently no experimental nor represent the frequency dependent absorption by appropriate
field data to validate the success of this method. Venugopal and definition of the sponge layers. In this way, the wake behind a
Smith also identified that reductions in wave height vary WEC is studied for each frequency component separately, as
greatly depending on the values of porosity used; indicating the amount of absorption of the WEC in its lee depends on the
that, if this method is to be used confidently for future array remaining energy in the considered frequency components.
studies, great care should be taken that realistic device This is also the case for wave direction dependent WECs. The
characteristics are modelled. This will require calibration of the wake is then not only calculated for each frequency component
porosity values to match device specific transmission, but also for each wave direction.
reflection and absorption and validation with physical model
results. 4. SPECTRAL WAVE MODELS
Aside from the inclusion of nonlinearity and deep water Another category of model which has been used to
terms, one further advantage to using phase resolving models of simulate WEC arrays is the spectral wave model. Spectral
the Boussinesq type for modelling of wave farms is the realistic wave models are phase-averaging wave propagation models
representation of diffraction phenomena. Spectral models and which predict how the surface wave frequency and directional
some mild-slope models include only a parameterised spectrum will evolve as waves propagate through varying
representation, which does not accurately represent reality. It background currents and water depth. While the other models
remains to be seen whether this will have a significant effect on described in this paper solve an equation or set of equations to
studies of wave farm interactions or environmental impact. find the surface elevation of the waves, spectral wave models
solve what is essentially an energy conservation equation. In
3.2 Mild-slope models fact the quantity that is solved for is wave action, which is the
In general, mild-slope models are based on the linear form spectral energy density divided by the intrinsic frequency.
of Boussinesq shallow water equations and therefore linear Wave action is conserved even in the presence of varying
waves are generated, propagating over mildly varying background currents, and thus is the preferred quantity to solve
bathymetries. Nevertheless, they calculate velocity potential for. Spectral wave models are capable of representing
and surface elevations throughout the numerical domain with a numerous wave transformation processes. These include depth-
relatively low computational and accuracy cost and with a high and current-induced refraction, shoaling, wind forcing, white-
stability performance. capping and bottom friction dissipation, dissipation through
Recently, wake effects in the lee of a single and multiple bathymetric breaking, and nonlinear quadruplet and triad wave-
WECs and energy absorption have been studied [30-32] by wave interactions. Because spectral wave models are phase-
using the time-dependent mild-slope equation model averaging, they are unable to represent wave diffraction
MILDwave [33] and applying a sponge layer technique, by explicitly. However, a phase-decoupled refraction-diffraction
which the redistribution of wave power both within and behind representation has been developed that addresses this
each farm can be studied in detail. In this phase-resolving deficiency reasonably well [34]. There are currently two open
model each combination of reflection and transmission source spectral wave models that are readily available: the
characteristics, and consequently absorption characteristics, can SWAN model developed by the Delft University of Technology
be modelled for all individual WECs in a farm [30]. This results [35], and the TOMAWAC model developed by the Electricité
in a representation of the wake effects in the lee of a single de France [36].
WEC and in that of a farm of WECs. A WEC is implemented in Because spectral wave models solve for the conservation
MILDwave as an array of cells (covering the spatial extent of of wave energy, a representation of a WEC array in a spectral
the WEC) that have been assigned a given degree of absorption wave model must somehow account for the energy absorbed
using the sponge layer technique. Absorption functions define and the energy radiated by the WECs. There are a few existing
the absorption coefficient S attached to each cell of the WEC in methods which have been used to represent WEC arrays in a
the x-direction and the y-direction. By changing the values of spectral wave model. These can be divided into two categories:

6
supra-grid scale, in which the whole WEC array is represented 5. CFD MODELS
over several computational grid points, and sub-grid scale, in The term Computational Fluid Dynamics (CFD) model is
which each individual WEC in an array is represented at a commonly used for codes that seek to resolve the Navier-Stokes
single computational grid point. equations. The Navier-Stokes equations are derived from mass
and momentum conservation, and often regarded as the most
4.1 Supra-grid models fundamental set of fluid flow equations. Both viscous effects
There are two current examples of supra-grid scale and turbulence are accounted for.
methods. The first uses the built-in obstacle feature in the From a practical point of view, Navier-Stokes solvers may
SWAN model, for which an energy transmission coefficient can be classified into two distinct categories: (i) Direct Numerical
be set [37]. The WEC array is represented with a single Simulation (DNS), resolving turbulence at the smallest relevant
transmission coefficient and the effect on the coastline is length scale, and (ii) CFD codes, where turbulence is not
estimated after propagating the waves altered by the array to directly resolved but dealt with in a parametric representation.
the shore. However, this method does not allow the energy For the modelling of WEC arrays only the latter is relevant,
absorption of the array to depend on frequency. This DNS being prohibitively expensive in terms of computational
shortcoming in the method has recently been addressed in demand. CFD models are based on Finite Element or Finite
another SWAN model study which included the introduction of Volume implementations, and often referred to as such.
a frequency dependent transmission coefficient, and the use of In contrast to the potential flow models discussed above,
an obstacle to represent a single WEC, as opposed to the whole CFD models include viscous effects and two-phase flow (air
array [38]. Although the supra-grid scale methods can now entrainment in breaking waves) making them the ideal tool for
account for the frequency dependence of the energy absorption, the simulation of extreme wave loading. Furthermore, the CFD
they do not account for the radiation of energy by the WECs. approach retains the full nonlinearity of the underlying
hydrodynamics. An additional benefit of CFD codes
4.2 Sub-grid models (particularly when compared to potential flow models) is that
A sub-grid scale method of representing a WEC array is marine currents are easily described. In the near-shore
implemented in the TOMAWAC model and includes both environment, currents may add significantly to the overall
frequency dependent energy absorption and the radiation of loading and also affect the device dynamics.
energy by WECs [39]. This is done by treating each WEC Unfortunately, CFD models are often prone to internal
(located at a computation grid point) as a source and sink of dissipation, particularly when resolving gravity water waves
wave energy. The energy absorption and radiation, which can (free surface flow). Maguire examined the free-surface
be dependent on the incident wave, is therefore incorporated at modelling capability of a number of commercially available
each WEC location into the wave action conservation equation. codes [41]. The overall conclusion from this extensive study is
This technique is similar to how the existing wave processes that none of the tested tools may readily be used to model
such as wind generation and wave dissipation are treated in gravity water waves; however, more recently, a number of tools
spectral wave models. under development appear to be more reliable in terms of their
As the development of WECs matures, and the possibility free surface prediction [42, 43].
of deploying devices in large arrays becomes closer to a reality, To overcome this difficulty of internal dissipation, a
it is important to develop numerical tools that can be used to decomposition of variables can be used. This consists in
investigate both the annual power production of a WEC array splitting all unknowns of the problem (pressure, fluid velocity
and the potential impacts it may have downstream on the wave and free-surface elevation) into the sum of an incident term and
climate. It is advantageous to use a spectral wave model for a diffracted term. The incident terms are described explicitly
this task because it is possible to cover a relatively large using a linear or nonlinear potential flow model. Thus only the
domain (tens of kilometres square) with a large WEC array part of the grid in the vicinity of the structure needs to be
(tens of devices) in a relatively short computational time. The refined. The method is called SWENSE (Spectral Wave
parameterisations of WECs in spectral wave models which Explicit Navier Stokes Equations). It has been already
have been developed can represent the energy absorbed and successfully applied and validated in 3D cases [44-46]
radiated by individual WECs, and are also capable of The disadvantage common to all CFD codes is their
representing nonlinear processes [40]. Of course, phase- computational demand. To minimise computational demand,
averaging models cannot resolve phase-dependent processes, so many codes offer non-uniform (at times also adaptive) meshes,
near-field effects around each individual WEC are not where the grid in the area of interest (the free surface and the
explicitly modelled in a spectral wave model. It is therefore vicinity of the device) is defined with a finer spatial resolution.
important to carry out comparison studies between phase- Particularly in deep water this may offer significant benefits.
resolving numerical models, experimental results, and spectral The most extensive WEC CFD study to date has been
wave models in order to ensure the best possible representation reported by Westphalen et al. [47]. Recently, an array of 2
of a WEC in a spectral wave model. heaving WECs was considered by Agamloh et al. [48] using
CFD. A number of very recently funded research projects (UK
EPSRC funded SUPERGEN Marine Challenge - Accelerating

7
the Deployment of Marine Energy) propose the use of CFD for suitable because they are based on phase-averaging, together
the modelling of small arrays (5-10 devices). Guidelines for the with the Linear BEM and semi-analytical techniques, because
use of CFD codes in the modelling of WEC arrays are expected these are based on frequency-domain representations, whilst
to emerge within the next 2-3 years. dynamic control requires a solution in the time-domain. The
Boussinesq/Mild-slope models are poorly suitable because
6. COMPARATIVE ANALYSIS whilst they provide a solution in the time-domain, it is not clear
The fundamental modelling characteristics, computational how to change the absorption layers dynamically to correctly
processing characteristics and usability characteristics are used model the control. CFD is poorly suitable as well because of
for the comparative analysis of the numerical techniques used the CPU time. The remaining methods (Time-domain
to model WEC arrays. The results of the analysis are formulation, nonlinear BEM) are suitable for modelling
summarised in Table 1, located at the end of the paper. These dynamic control, with the most suitable approach depending on
characteristics are then used to consider the suitability of each the particular WEC array being modelled.
numerical technique for four different modelling tasks; Modelling the annual energy production (AEP) requires
localised effects, dynamic control, annual energy production power capture to be calculated for a large number of irregular
(AEP) - separated into small and large arrays, and distal sea-states. This means that CFD and Nonlinear BEM are poorly
environmental impacts. Of course, other modelling tasks exist; suited because of their high computational requirements. In
however, these are considered to be representative of a range of addition, Linear BEM models and the Time-domain
tasks and demonstrate the comparative performances of the formulations derived from them, rapidly become unsuitable as
different numerical modelling techniques. Table 1 also includes the number of WECs increase due to the quadratic relationship
an estimation of suitability for each numerical modelling between the computation effort and the number of WECs. In
technique for these four modelling tasks. addition, the supra-grid spectral wave model is not suitable for
The designations of the modelling technique suitability are modelling the AEP because WEC array interactions are
discussed below; however, prior to this, it is important to subsumed within the explicit definition of the absorption layer;
recognise that the characteristics of the different modelling the model itself cannot calculate array power performance
techniques make them suitable for modelling particular WECs. directly. The Boussinesq/Mild-slope models and sub-grid
For example, a linear BEM model may be suitable for a large spectral wave models are all suitable for the calculation of the
WEC in deep water that sheds minimal vortices, but is less AEP. The most suitable method will depend on conditions (e.g.
suitable for a small WEC that sheds significant amount of water depth, marine currents, bathymetry, etc) and also the
vortices, whose motions are significantly nonlinear and with a accuracy with which the WEC and its interactions with the sea
complex control strategy. This aspect will not be investigated (e.g. WEC radiation, diffraction, nonlinear dynamics, etc.) can
further; however, it is clearly an additional consideration in be modelled.
determining the most suitable modelling technique for a Finally, suitability for determining the distal environmental
particular WEC array. impact is considered. Unfortunately, none of the potential flow
Returning to the modelling of WEC arrays, the suitability models are suitable because of the assumption of constant water
for modelling localised effects is considered first. This refers to depth, which makes them unsuitable for propagating the waves
the extent that near-field effects, such as evanescent waves and to the shoreline, where the environmental impact is typically
vortex shedding, from one WEC may influence another WEC most significant. Furthermore, the large propagation distances
nearby. In close proximity, phase correlations between two mean that CFD models are poorly suited due to their high
WECs are high and so phase-averaged models, i.e. those based computational requirements. The remaining modelling
on spectral wave models, are not suitable. In addition, the semi- techniques, Boussinesq/Mild-slope Models and Spectral Wave
analytical techniques based on further simplifying Models, are all suitable for determining environmental impact
approximations to linear wave theory (the point absorber and and have been used extensively for this task in applications
plane wave methods) are not suitable because these focus on other than WECs. In addition to the model differences
modelling the far-field. On the other hand, provided that the discussed above, the larger cell size in Spectral Wave Models
modelling and computational effort can be justified CFD is means that these are most suited for modelling more remote
highly suitable for modelling localised effects because the impacts, whilst Boussinesq/Mild-slope Models are most
model may include both evanescent waves and vortex shedding suitable for situations where reflections and resonances may be
implicitly. However, in many cases it is possible that potential significant.
flow models (excluding some semi-analytical techniques)
would be adequate, with significantly less computation effort. 7. DISCUSSION
Finally, whilst Boussinesq/Mild-slope models resolve phase, This review paper is a snap-shot of the currently available
they are unlikely to accurately model the near-field and so are numerical modelling techniques for WEC arrays. Whilst it is
poorly suitable. not expected that the results of this comparative analysis will
The suitability for modelling dynamic control, whereby the change in the short-term, there will be long-term changes. It is
motion of each WEC is controlled to maximise power capture, clear from the descriptions of the different numerical modelling
is now considered. Again, the spectral wave models are not techniques that in many cases, potential remains for

8
improvement, by either increasing their accuracy and/or Control". in 7th European Wave and Tidal Energy
reducing their computational requirements. Improvements in Conference, Porto, Portugal.
readily available computing power are also likely to change [9] Folley, M. and Whittaker, T.J.T., 2009, "The Effect of
what can be done practically. Sub-Optimal Control and the Spectral Wave Climate on
Finally, it is clear from the comparative analysis described the Performance of Wave Energy Converter Arrays".
above that there is no single best numerical modelling Applied Ocean Research, 31(4): pp. 260-266
techniques for WEC arrays. The most appropriate numerical [10] Borgarino, B., Babarit, A., and Ferrant, P., 2011,
modelling technique being that which best matches the required "Extension of Free-Surface Green's Function Multipole
characteristics of the particular modelling task. Unfortunately, it Expansion for Infinite Water Depth Case". International
is not always clear which modelling technique this may be, as Journal of Offshore and Polar Engineering, 21(3): pp.
each model has different strengths and weaknesses, which 161-168
rarely match the characteristics of the modelling task exactly. [11] Budal, K., 1977, "Theory for Absorption of Wave Power
However, it is expected that identification of the most by a System of Interacting Bodies". Journal of Ship
appropriate numerical modelling technique for a particular task Research, 21(4): pp. 248-253
will become clearer with experience and by experimental [12] Evans, D.V., 1980, "Some Analytical Results for Two and
validation. Although, the lack of suitable validation data for Three Dimensional Wave-Energy Absorbers". in Power
these numerical modelling techniques is a significant issue that from Sea Waves, Edinburgh, UK.
needs to be addressed urgently. [13] Simon, M.J., 1982, "Multiples Scattering in Arrays of
Axisymmetric Wave-Energy Devices - a Matrix Method
REFERENCES Using a Plane-Wave Approximation.". Journal of Fluid
[1] Borgarino, B., Babarit, A., and Ferrant, P., 2011, "Impact Mechanics, 120: pp. 1-25
of the Separating Distance between Interacting Wave [14] McIver, P., 1984, "Wave Forces on Arrays of Floating
Energy Converters on the Overall Energy Extraction of an Bodies". Journal of Engineering Mathematics, 18: pp.
Array". in 9th European Wave and Tidal Energy 273-285
Conference, Southampton, UK. [15] Mavrakos, S.A., 1991, "Hydrodynamic Coefficients for
[2] Durand, M., Babarit, A., Pettinotti, B., Quillard, O., Groups of Interacting Vertical Axisymmetric Bodies".
Toularastel, J.L., and Clément, A.H., 2007, "Experimental Ocean Engineering, 18(485-5125)
Validation of the Performances of the Searev Wave [16] Kagemoto, H. and Yue, D.K.P., 1986, "Interactions among
Energy Converter with Real Time Latching Control". in Multiple Three-Dimensional Bodies in Water Waves: An
7th European Wave and Tidal Energy Conference, Porto, Exact Algebraic Method". Journal of Fluid Mechanics,
Portugal. 166: pp. 189-209
[3] Thilleul, O., Baudry, V., Guilcher, P.-M., Jacquin, E., [17] Child, B.F.M. and Venugopal, V., 2010, "Optimal
Babarit, A., Locuratolo, P., and Larivain, A., 2011, Configurations of Wave Energy Device Arrays". Ocean
"Assessment of Sizing Parameters of a Wave Energy Engineering, 37(16): pp. 1402-1417
Converters through the Complementary Use of a Linear [18] Newman, J.N., 1977, Marine Hydrodynamics, Cambridge:
Potential Code, a Rans and a Sph Solver". in 9th MIT Press.
European Wave and Tidal Energy Conference, [19] Cummins, W.E., 1962, "The Impulse Response Function
Southampton, UK. and Ship Motions". Schiffstechnik: pp. 101-109
[4] Mavrakos, S.A. and McIver, P., 1997, "Comparison of [20] Taghipour, R., Perez, T., and Moan, T., 2008, "Hybrid
Methods for Computing Hydrodynamic Characteristics of Frequency-Time Domain Models for Dynamic Response
Arrays of Wave Power Devices". Applied Ocean Analysis of Marine Structures". Ocean Engineering,
Research, 19(5-6): pp. 283-291 35(7): pp. 685-705
[5] Justino, P. and Clement, A., 2003, "Hydrodynamic [21] Babarit, A. and Clement, A.H., 2006, "Optimal Latching
Performance for Small Arrays of Submerged Spheres". in Control of a Wave Energy Device in Regular and Irregular
5th European Wave Energy Conference, Cork, Ireland. Waves". Applied Ocean Research, 28(2): pp. 77
[6] Ricci, P., Saulnier, J.B., and O. Falcão, A.F.d., 2007, [22] McCabe, A.P., Bradshaw, A., and Widden, M.B., 2005, "A
"Point-Absorber Arrays : A Configuration Study Off the Time-Domain Model of a Floating Body Using
Portuguese West-Coast". in 7th European Wave and Tidal Transforms". in 6th European Wave and Tidal Energy
Energy Conference, Porto, Portugal. Conference, Glasgow, UK.
[7] De Backer, G., Vantorre, M., Beels, C., De Rouck, J., and [23] Babarit, A., Mouslim, H., Clément, A., and Laporte-
Frigaard, P., 2010, "Power Absorption by Closely Spaced Weywada, P., 2009, "On the Numerical Modelling of the
Point Absorbers in Constrained Conditions". IET Non Linear Behaviour of a Wave Energy Converter". in
Renewable Power Generation, 4(6): pp. 579-591 28th International Conference on Ocean, Offshore and
[8] Hals, J., Taghipour, R., and Moan, T., 2007, "Dynamics of Arctic Engineering, Hawaii, USA.
a Force-Compensated Two-Body Wave Energy Converter [24] Hague, C. and Swan, C., 2009, "A Multiple Flux
in Heave with Hydraulic Power Take-Off Subject to Phase Boundary Element Method Applied to the Description of

9
 Surface Water Waves". Journal of Computational Physics, Change to a Wave Farm". Ocean Engineering, 34(5-6):
228(14): pp. 5111-5128 pp. 884-901
[25] Kashiwagi, M., 2000, "Non-Linear Simulations of Wave- [38] Smith, H.C.M., Pearce, C., and Millar, D.L., 2012,
Induced Motions of a Floating Body by Means of the "Further Analysis of Change in Nearshore Wave Climate
Mixed Eulerian-Lagrangian Method". Proceedings of the Due to an Offshore Wave Farm: An Enhanced Case Study
Institution of Mechanical Engineers, Part C:Journal of for the Wave Hub Site". Renewable Energy, 40(1): pp. 51-
Mechanical Engineering, 214(6): pp. 841-855 64
[26] Stratigaki, V., Vanneste, D., Troch, P., Gysens, S., and [39] Silverthorne, K. and Folley, M., 2011, "A New Numerical
Willems, M., 2010, "Numerical Modeling of Wave Representation of Wave Energy Converters in a Spectral
Penetration in Ostend Harbour". in International Wave Model". in 9th European Wave and Tidal Energy
Conference on Coastal Engineering, Shanghai, China. Conference, Southampton, UK.
[27] MIKE 21 BW, 2011, Boussinesq Waves Module User [40] Folley, M. and Whittaker, T., 2010, "Spectral Modelling
Guide: Danish Hydraulic Institute. of Wave Energy Converters". Coastal Engineering,
[28] Madsen, P.A. and Sorensen, O.R., 1992, "A New Form of 57(10): pp. 892-897
the Boussinesq Equations with Improved Linear [41] Maguire, A.E., 2011, Geometric Design Considerations
Dispersion Characteristics. Part 2: A Slowly-Varying and Control Methodologies for Absorbing Wavemakers,
Bathymetry". Coastal Engineering, 18: pp. 183-204 PhD, The University of Edinburgh
[29] Venugopal, V. and Smith, G.H., 2007, "Wave Climate [42] Qian, L., Causon, D.M., Mingham, C.G., and Ingram,
Investigation for an Array of Wave Power Devices". in 7th D.M., 2006, "A Free-Surface Capturing Method for Two
European Wave and Tidal Energy Conference, Porto, Fluid Flows with Moving Bodies". Proceedings Royal
Portugal. Society A, 462(2065): pp. 21-42
[30] Beels, C., Troch, P., De Backer, G., Vantorre, M., and De [43] Spinneken, J., Heller, V., Kramer, S., Piggott, M., and
Rouck, J., 2010, "Numerical Implementation and Vire, A., 2012, "Assessment of an Advanced Finite
Sensitivity Analysis of a Wave Energy Converter in a Element Tool for the Simulation of Fully-Nonlinear
Time-Dependent Mild-Slope Equation Model". Coastal Gravity Water Waves". in 22nd International Ocean and
Engineering, 57(5): pp. 471-492 Polar Engineering Conference, Rhodes, Greece.
[31] Troch, P., Beels, C., De Rouck, J., and De Backer, G., [44] Gentaz, L., Luquet, R., Alessandrini, B., and Ferrant, P.,
2010, "Wake Effects Behind a Farm of Wave Energy 2004, "Numerical Simulation of a 3d Viscous Flow
Converters for Irregular Long-Crested and Short-Crested around a Vertical Cylinder in Nonlinear Waves Using an
Waves". in Proceedings of the International Conference Explicit Wave Model". in 23rd International Conference
on Coastal Engineering, Shanghai, China. on Ocean, Offshore and Arctic Engineering, Vancouver,
[32] Beels, C., Troch, P., De Visch, K., Kofoed, J.P., and De Canada.
Backer, G., 2010, "Application of the Time-Dependent [45] Luquet, R., Alessandrini, B., Ferrant, P., and Gentaz, L.,
Mild-Slope Equations for the Simulation of Wake Effects 2004, "Simulation of the Viscous Flow Past a Ship in
in the Lee of a Farm of Wave Dragon Wave Energy Waves Using the Swense Approach". in 24th ONR
Converters". Renewable Energy, 35(8): pp. 1644-1661 Symposium on Naval Hydrodynamics, Newfoundland,
[33] Troch, P., 1998, Mildwave – a Numerical Model for Canada.
Propagation and Transformation of Linear Water Waves, [46] Luquet, R., Jacquin, E., Guillerm, P., Gentaz, L., Ferrant,
Department of Civil Engineering, Ghent University: P., and Alessandrini, B., 2005, "Ranse with Free Surface
Ghent. Computations around Fixed and Free Dtmb 5415 Model,
[34] Holthuijsen, L.H., Herman, A., and Booij, N., 2003, in Still Water and in Waves". in CFD Workshop, Tokyo,
"Phase-Decoupled Refraction-Diffraction for Spectral Japan.
Wave Models". Coastal Engineering, 49(4): pp. 291-305 [47] Westphalen, J., Greaves, D., Williams, C., Taylor, P.,
[35] Booij, N., Ris, R.C., and Holthuijsen, L.H., 1999, "A Causon, D., Mingham, C., Hu, Z., Stansby, P., B, R., and
Third-Generation Wave Model for Coastal Regions 1. Omidvar, P., 2009, "Extreme Wave Loading on Offshore
Model Description and Validation". Journal of Wave Energy Devices Using Cfd: A Hierarchical Team
Geophysical Research, 104(C4): pp. 7649-7666 Approach". in 8th European Wave and Tidal Energy
[36] Benoit, M., Marcos, F., and Becq, F., 1996, "Development Conference, Uppsala, Sweden.
of a Third-Generation Shallow-Water Wave Model with [48] Agamloh, E.B., Wallace, A.K., and von Jouanne, A.,
Unstructured Spatial Meshing". in 25th International 2008, "Application of Fluid-Structure Interaction
Conference on Coastal Engineering, Orlando, USA. Simulation of an Ocean Wave Energy Extraction Device".
[37] Millar, D.L., Smith, H.C.M., and Reeve, D.E., 2007, Renewable Energy, 33(4): pp. 748-757
"Modelling Analysis of the Sensitivity of Shoreline

10
 TABLE 1: COMPARATIVE ANALYSIS OF NUMERICAL MODELLING TECHIQUES FOR WEC ARRAYS

Potential flow models Spectral wave models

Semi-analytical Time-domain
Linear BEM Nonlinear BEM Boussinesq Mild-slope Supra-grid Sub-grid CFD
techniques formulation
Fundamental
Definition of Implicit body surfaces Explicit Explicit source Implicit fluid
Explicit absorption layers
hydrodynamics Explicit coefficients absorption layer strength flow
Implicit body surfaces Implicitly Implicitly Implicitly capable for phase- Implicitly
Nonlinear wave dynamics Not capable Not capable
capable capable averaged dynamics capable
Explicit Explicit source
Nonlinear dynamics Not capable Implicit solver Explicit absorption layers Implicit solver
absorption layer strength
Vortex shedding Explicit inclusion by linearisation Explicit inclusion Explicit inclusion Explicit inclusion Implicit inclusion
Explicitly Implicitly
WEC radiation Implicitly capable Explicitly capable Not capable
capable capable
Approximated by phase-decoupled Implicitly
Diffraction Implicitly capable Explicitly capable
refraction-diffraction capable
Variable bathymetry and Implicitly Implicitly Implicitly
Not capable Implicitly capable
marine currents capable† capable‡ capable
Computational
Number of panels
Complexity of
Primary dependent Number of panels and complexity Number of panels Number of cells Number of cells Number of cells
function
of equations
Number of frequencies and Number of frequencies and Number of time-
Secondary dependent Number of time-steps Number of time-steps
directions directions steps
Determinate of array Linear inc. with
Quadratic increase with number of WECs Linear increase with spatial area Linear increase with spatial area
“size” spatial volume
Simple and poss. Complex and Simple and poss. Complex and
Solver Simple and stable Simple and stable Simple and stable
unstable stable unstable poss. unstable
Usability
Required skill Low High Medium High Medium Low Low Medium High
Commercial and
Software availability in Commercial code Research code Commercial code Research code Commercial code available, WEC Open-source code available, WEC
open-source code
2012 available only available only model required model required
available
Suitability ( **** - highly suitable, *** - moderately suitable, ** - poorly suitable, * - not suitable )
Localised effects *** * to *** *** *** ** ** * * ****
Dynamic control * * **** **** * * * * **
AEP (small WEC array) *** *** ** ** *** *** ** *** **
AEP (large WEC array) ** *** ** ** *** *** ** *** **
Environmental impact * * * * *** *** **** **** **

†
Limited to shallow water
‡
Limited to mild-slopes

11