ARTICLE IN PRESS

Ocean Engineering 34 (2007) 2330–2344

www.elsevier.com/locate/oceaneng

Optimisation of wave energy extraction with the Archimedes

Wave Swing
Duarte Valérioa,, Pedro Beirãob, José Sá da Costaa
a
IDMEC/IST, TULisbon, Av. Rovisco Pais, 1, 1049-001 Lisboa, Portugal
b
Department of Mechanical Engineering, Instituto Superior de Engenharia de Coimbra, R. Pedro Nunes, 3030-199 Coimbra, Portugal
Received 29 November 2006; accepted 21 May 2007
Available online 2 June 2007

Abstract

This paper addresses the Archimedes Wave Swing (an offshore wave energy converter, which produces electricity from sea waves). It
compares the performances of latching control (a discrete, highly non-linear, intrinsically sub-optimum control strategy), of reactive
control, of phase and amplitude control (two optimum control strategies that involve non-causal transfer functions, which have to be
implemented with approximations, thus rending the control sub-optimum), and of feedback linearisation control (a non-linear control
strategy). From extensive simulations it is concluded that the latter performs clearly better irrespective of the sea state, and leads to a
significant increase of absorbed wave power.
r 2007 Elsevier Ltd. All rights reserved.

Keywords: Wave energy; Renewable energy; Archimedes Wave Swing; Reactive control; Phase and amplitude control; Latching control; Feedback
linearisation control

1. Introduction addressed in Section 3; control strategies are summarised in

Section 4; simulation results are given in Section 5;
Due both to the inexorable rise of oil prices and to conclusions are drawn in Section 6.
environmental concerns, renewable energies are receiving Due to industrial secrecy reasons several parameters of
increased attention. Among them, the energy of sea waves the AWS have been modified in the models given and
may play an important role in the near future. Wave energy employed.
converters (WECs) for turning it into electricity are not
commercially competitive—yet. But it is expected that they 2. The AWS
will soon be, as sufficiently efficient WECs are developed.
Control engineering plays an important role towards that The AWS is an offshore, fully submerged (43 m deep
objective. This paper contributes to that effort by compar- underwater), point absorber (that is to say, of neglectable
ing the performance of several control strategies suitable size compared to the wavelength) WEC. Its two main parts
for the Archimedes Wave Swing (AWS), a WEC of which a are the silo (a bottom-fixed air-filled cylindrical chamber)
2 MW prototype (Fig. 1) has already been built, tested at and the floater (a movable upper cylinder). Due to changes
the Portuguese northern coast during 2004, and then in wave pressure, the floater heaves (Fig. 1). When the
decommissioned. The aim is to see, by means of simula- AWS is under a wave top, the floater moves down
tions, which one leads to a better efficiency of that compressing the air inside the AWS. When the AWS is
prototype—and of the ones that will follow. under a wave trough, pressure decreases and consequently
The paper is organised as follows: Section 2 describes the the air expands and the floater moves up (Beirão et al.,
AWS; the identification of a linearised model thereof is 2006).
Within the air-filled space formed by the silo and the
Corresponding author. Tel.: +351 218419119; fax: +351 218498097. floater there are several components needed for the
E-mail address: [email protected] (D. Valério). functioning of the AWS, among which an electric linear

0029-8018/$ - see front matter r 2007 Elsevier Ltd. All rights reserved.
doi:10.1016/j.oceaneng.2007.05.009
 ARTICLE IN PRESS
D. Valério et al. / Ocean Engineering 34 (2007) 2330–2344 2331

Fig. 1. The AWS prototype and its working principle.

generator (ELG) to convert the floater’s heave motion into given by

electricity. The electric energy produced by the ELG is  
A B
transferred to shore via a 6 km long undersea cable; after SðoÞ ¼ 5 exp  4 , (1)
o o
that it passes through a converter, and following this
conversion is supplied to the electric grid. The AWS also R þ1 S is the wave energy spectrum (a function such that
where
has water dampers that are actuated when the floater 0 SðoÞ do is the mean-square value of the wave
approaches the mechanical end-stops, to reduce its velocity elevation). The numerical values A ¼ 0:780 (SI) and B ¼
and avoid a strong collision with them; they are also 3:11=H 2s were used. Values for H s and for T e (from which
actuated together with the ELG when the latter does not the limits of the frequency range were then found) were
suffice to supply the force required to adequately control provided by Table 1.
the movement of the AWS. These water dampers are A model of the AWS in the time domain is based on
otherwise inactive, so as not to hinder wave energy Newton’s law applied to the floater’s vertical acceleration
extraction. € The equation of motion of the floater is
x.
It is the force exerted by these two components, the ELG €
f pi  f pe  wf  f n  f v  f m  f wd  f lg ¼ ðmf þ mwt Þx.
and the water dampers, that we can control. Actually it is
also possible to change the mean value of air pressure (2)
within the AWS, by means of pumps that add or remove The total mass comprises the mass of the floater mf and the
sea water (thus changing the volume available for air); but water trapped inside the floater mwt (these two masses are
this is a slow process that takes several minutes, and so it is known to oscillate together). The total force acting on the
used to cope with the variations of sea level caused by the floater is the sum of the forces due to external water
tides (that are slower still). To control the AWS coping pressure f pe , to internal air pressure f pi , to the weight of
with different incoming waves and sea states, it is the ELG the floater wf , to a nitrogen cylinder extant inside the AWS
and the water dampers we have to resort to. This paper f n , to the hydrodynamic viscous drag f v , to mechanical
deals with this latter type of control. It will be assumed that friction f m , to the water dampers f wd , and to the ELG f lg ,
there will be no changes in tide, and no changes in the mean the last two being damping forces, and the ones we can
pressure of the air inside the AWS. control.
From the description above it is seen that the AWS Notice that the convention of signs assumes that positive
can be reasonably expected to behave much like a values are given to the most natural direction—hence
mass–spring–damper system, though with relevant non- among all forces the only one pointing upwards is f pi . Also
linearities. notice that lower-case letters are being used for variables in
The AWS was submerged 5 km offshore Leixões, the time-domain; their Laplace transforms (in the fre-
Portugal. Data for wave climate in several locations in quency domain) will be denoted using the corresponding
Portugal may be found with the ONDATLAS software def def
capitals. (Hence XðsÞ ¼ L½xðtÞ, F lg ðsÞ ¼ L½f lg ðtÞ, and so
(Pontes et al., 2005). The nearest location available is the on.) Both capitals and lower-case letters are used for
Leixões-buoy location ð41 12:20 N; 9 5:30 WÞ. The corre- constants.
sponding data on significant wave height H s (from trough Since the damping forces f wd and f lg are, as mentioned
to crest) and on maximum and minimum values of the above, the ones we can control, let us call f u (control force)
wave energy period T e is found in Table 1. to their sum:
Throughout this paper, simulations are performed using
regular and irregular waves. The first are sinusoidal; values f u ¼ f lg þ f wd . (3)
congruent with Table 1 for the amplitude and the period of The external pressure force can be decomposed thus:
the sinusoids are used. For the latter, 12 waves (one for
f pe ¼ f hs þ f rad  f exc . (4)
each month of the year) satisfying the Pierson–Mosko-
witz’s spectrum, which accurately models the behaviour of In (4), f exc (wave excitation force) is the force exerted on
real sea waves (Falnes, 2002), were used. This spectrum is the AWS by the incident sea waves assuming that the
 ARTICLE IN PRESS
2332 D. Valério et al. / Ocean Engineering 34 (2007) 2330–2344

Table 1
Characteristics of several irregular waves according to ONDATLAS

Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec

H s (m) 3.2 3.0 2.6 2.5 1.8 1.7 1.5 1.6 1.9 2.3 2.8 3.1
T e;min (s) 5.8 5.8 5.2 5.5 5.0 4.7 4.6 5.0 5.2 5.3 5.5 5.3
T e;max (s) 16.1 14.5 13.7 14.8 12.2 9.7 11.1 10.5 12.0 12.6 13.3 14.2

floater is not moving, f hs is the hydrostatic force, and f rad is 4

the force exerted in the AWS by the wave that the floater 2.0m
creates by its movement. We will assume that the forces can 3
be superimposed on each other; actually, when wave
amplitudes are large compared with the wavelength, (4) is 2
no longer valid and a non-linear expression must be used,
1

ξ /m
but this approximation is suitable for our case.
Complete explicit expressions of all these terms cannot
0
be given here for lack of space. They may be found for
0.5m
instance in Pinto (2004); Sá da Costa et al. (2003, 2005).
-1
These references also describe an accurate, non-linear,
Simulink-based simulator of the AWS, the AWS Time- -2
Domain Model (TDM), that has already been developed
implementing the expressions above. The AWS TDM was 300 310 320 330 340 350
used for the simulations presented in this paper. time / s

3. AWS linear dynamic model identification Fig. 2. Output of the AWS TDM for regular waves with 10 s of period
and amplitudes of 0.5, 1.0, 1.5, and 2.0 m.

Before thinking about the extensive use of a non-linear

model of the AWS (such as the AWS TDM) for control output of the system and the orders of the numerator and
purposes, a linear model approximation of that same WEC the denominator of the function that it will provide.
should be identified in the first place (Beirão et al., 2007b). In what the input and the output of the identified linear
This is possible because, even though the AWS is a non- model are concerned, two possibilities were explored for
linear system, a sinusoidal input causes a fairly sinusoidal the AWS.
output, with an amplitude practically proportional to that The first one considers the wave excitation force F exc as
of the input, for all wave periods and amplitudes expected _ as the output.
the input and the floater’s vertical velocity X
to occur. Fig. 2 shows this for several regular waves of _
F exc and X data provided by the AWS TDM for regular
different amplitudes. (sinusoidal) waves with periods from 8 to 14 s (the range
the AWS was conceived for (Sá da Costa et al., 2003)) was
3.1. Identification procedure used. Following the ONDATLAS software, the most
frequent significant wave height H s (from trough to crest)
System identification deals with the construction of is admitted as being equal to 2 m. Hence several waves with
mathematical models of dynamical systems using measured a 1 m amplitude (half of H s ) and different periods were
data. In the particular situation of the AWS, very few assumed for the simulations. (Notice that an approxima-
experimental data are available. Thus, a different approach tion is involved here, since these waves used for identifica-
has been followed. The AWS TDM was used as an tion are regular, while those addressed by ONDATLAS are
emulator of the real non-linear AWS WEC; no controller real, irregular waves.) To apply the Levy identification
was employed, and hence both control forces (f lg and f wd ) method, Matlab’s function levy was used (Valério and Sá
were assumed to have only a minimum, unavoidable da Costa, 2007). The data found in Table 2 was used in that
residual value. Based on simulation results from the process.
AWS TDM, a linear dynamic model of the AWS was All combinations of values for the numerator and
estimated. Since sea waves are periodic oscillations (even if denominator orders m and n from 0 to 5 were tried. Only
not sinusoidal), an identification method in the frequency identified models with two poles or more and one zero or
domain, such as the classical method of Levy (1959), is the more reproduced the wave frequency behaviour correctly.
obvious choice. The identified model structure
Levy’s identification method provides us a transfer
_
XðsÞ 2:171  106 s  6:759  107
function relating the output data with the input data. ¼ (5)
The method receives frequency data for the input and the F exc ðsÞ 0:967s2 þ 0:5874s þ 1
 ARTICLE IN PRESS
D. Valério et al. / Ocean Engineering 34 (2007) 2330–2344 2333

Table 2
Data used in the identification

Period (s)

4 5 6 7 8 9 10 11 12 13 14

F exc ampl. (kN) 31.88 108.57 202.64 290.46 365.29 423.35 467.48 501.30 527.62 548.44 565.10
X amplitude (m) 0.1195 1.1359 1.1905 1.1939 1.2972 1.3573 1.5558 1.4766 1.4132 1.3803 1.4023
X gain (dB) 108.53 99.61 104.62 107.72 109.20 109.88 109.56 110.62 111.44 111.98 112.11
X phase ð Þ 160.20 85.44 26.80 15.77 14.10 27.20 30.72 16.08 9.00 8.31 11.31
_ ampl. ðm s1 Þ
X – – – – 1.0341 1.0535 1.0686 1.0071 0.8813 0.7999 0.7488
_ gain (dB)
X – – – – 110.96 112.08 112.82 113.94 115.54 116.72 117.56
_
Xphase ð Þ – – – – 111.60 112.00 97.92 81.82 78.60 78.09 76.11

-90 1.5
gain / dB

1
-120

0.5
-150 Imag
10-1 100 101
0
ω / rads-1
180 -0.5
phase /°

90
1
0
-1.5
-90 -0.5 0 0.5
10-1 100 101 Real

Fig. 3. Model (5); left: Bode diagram (dots mark data used for identification); right: pole-zero map.

with one (non-minimum phase) zero and two (stable, transfer function, was
complex conjugate) poles is the one that reproduces the
AWS TDM responses making use of as few parameters as XðsÞ 2:259  106
¼ . (7)
possible. By adding an extra pole at the origin, model F exc ðsÞ 0:6324s2 þ 0:1733s þ 1

XðsÞ 2:171  106 s  6:759  107 Fig. 4 shows the Bode diagram and the pole-zero map of
¼ (6) this last identified model.
F exc ðsÞ sð0:967s2 þ 0:5874s þ 1Þ
Even though the input of both transfer functions (5) and
relating the wave excitation force to the floater’s vertical (7) is the wave excitation force F exc , the outputs take into
position X is found. Fig. 3 shows the Bode diagram and the account the effects of the radiated force F rad as well, since
pole-zero map of model (5). This model’s drawbacks are its this latter force was included in the AWS TDM.
non-minimum phase zero (something that may be undesir-
able) and its unnecessary complexity (since a simpler one
can be got). So another solution was looked for. 3.2. Validation of identification results
The second possibility for input and output choice was
to consider the wave excitation force as the input and the Six hundred seconds (10 min) long simulations were
floater’s vertical position as the output and provide there carried out, employing the AWS TDM (for the non-linear
data (see Table 21) to Levy’s identification method. Since it case) and Simulink implementations of (5) and (7) (for the
was found that the former period range had insufficient linear cases). The root mean-square errors, given by
data to allow a good identification, it had to be enlarged to sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Z 600
4 s to 14 s in order to obtain an acceptable model. Under 1 ~ 2 dt
RMS ¼ ðX  XÞ (8)
this new assumption, the identified model, a second-order 600 0
1
In that table, a phase lead of 90 could be expected between the phases (X~ being the estimate of the floater’s vertical position), are
of X_ and X. Non-linearities, however, prevent this. given, for several significant simulations, in Tables 3 and 4.
 ARTICLE IN PRESS
2334 D. Valério et al. / Ocean Engineering 34 (2007) 2330–2344

-90 1.5
gain / dB

1
-120

0.5
-150

Imag
10-1 100 101
0
ω / rads-1
0 -0.5
phase /°

1
-90

-1.5
-180 -0.5 0 0.5
10-1 100 101 Real

Fig. 4. Model (7); left: Bode diagram (dots mark data used for identification); right: pole-zero map.

Table 3
Root mean-square errors for the simulations with regular waves

Wave amplitude (m) Model Wave period (s)

8.0 10.0 12.0 14.0

0.5 (5) 0.2001 0.2417 0.3528 0.4005

(7) 0.0886 0.1203 0.0674 0.0604
0.75 (5) 0.2717 0.3376 0.4912 0.5642
(7) 0.1919 0.2523 0.1476 0.1326
1.0 (5) 0.3380 0.4293 0.6100 0.7059
(7) 0.3319 0.4211 0.2639 0.2400

1.25 (5) 0.4105 0.5206 0.7139 0.8283

(7) 0.5181 0.6376 0.4259 0.3911

Table 4
Root mean-square errors for the simulations with irregular waves

Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec

Model (5) 0.5490 0.5039 0.5096 0.5316 0.3677 0.4161 0.3296 0.3283 0.3776 0.4576 0.4490 0.6602
Model (7) 0.4227 0.3472 0.2622 0.2386 0.1357 0.1375 0.1177 0.1229 0.1423 0.1991 0.2986 0.3828

A 100 s slice corresponding to March (a significant month) the transfer function from the wave excitation force to the
is highlighted for illustration purposes in Fig. 5. floater’s vertical velocity:
From these results, it is seen that model (7) reproduces  
the AWS TDM behaviour more accurately; it also F exc ðjoÞ
RðoÞ ¼ Re , (9)
requires less parameters than (5), and its structure is _
XðjoÞ
similar to the one normally assumed in the literature
(Falnes, 2002, e.g.). Actually, (5) performs slightly better R may be frequency dependent, but it is physically
than (7) for regular waves of low period and high impossible that it should be negative (Falnes, 2002).
amplitude. But these cases are a minority, and simulations Indeed, R is always positive for (7), and actually it does
with irregular waves (with which (7) is systematically not even depend on o, since, by definition (9),
better) are deemed more important since they are expected  
0:6324ðjoÞ2 þ 0:1733jo þ 1
to reproduce the behaviour of real sea waves more R ¼ Re
accurately. 2:259  106 jo
There is an additional reason to prefer model (7), related 0:1733
¼ ¼ 7:6715  104 . ð10Þ
to the resistance R, which is the real part of the inverse of 2:259  106
 ARTICLE IN PRESS
D. Valério et al. / Ocean Engineering 34 (2007) 2330–2344 2335

Irregular wave for March practice, they can only be implemented with approxima-
2 tions that cause a decrease in their efficiency—to an extent
1.5 that justifies considering other control strategies, that are
intrinsically sub-optimum because they can never (even
1 under the most favourable conditions) attain the efficiency
0.5 that optimum control has in theory. The two sub-optimum
position / m

strategies considered below are latching control and

0 feedback linearisation control.
-0.5
4.1. Optimum control
1

-1.5 This section closely follows Falnes (2002).

non-linear force-position force-velocity
Let us rewrite (7) as
-2
300 320 340 360 380 400 XðsÞ 1
time / s
¼ , (12)
F exc ðsÞ ms2 þ Rs þ S
Fig. 5. Floater’s vertical position for AWS TDM and identified linear where mass m, resistance R, and stiffness S are given by
models (100 s long period out of 600 s). m ¼ 2:8017  105 , R ¼ 7:6715  104 , and S ¼ 4:4267  105
(in SI units). The transfer function in (12) is the same as
€ þ RxðtÞ
mxðtÞ _ þ SxðtÞ ¼ f ðtÞ. (13)
x 105 exc
5
Defining complex-valued phasors f^exc ^
and x_ for f exc and
_x, respectively,
model (7)

0 f^exc iot f^exc iot
f exc ðtÞ ¼ e þ e , ð14Þ
2 2
R / N·s·m-1

^_ ^_ 
-5 _xðtÞ ¼ x eiot þ x eiot , ð15Þ
2 2
we can rewrite (13) as
-10    
S ^_
eiot f^exc  R þ iom þ x
io
model (5)    
-15  S ^_ 
10-2 100 102
þ eiot f^exc  R  iom  x ¼ 0, ð16Þ
io
ω / rad·s-1
where o is the frequency. Defining an impedance Z
Fig. 6. Evolution of R for both models.  
S
Z ¼ R þ i om  , (17)
o
But, for some frequencies, (5) leads to a negative value of it is possible to rewrite (16) as
R, since     
  eiot f^exc  Zx^_ þ eiot f^exc  Z  x^_ ¼ 0. (18)
0:967ðjoÞ2 þ 0:5874jo þ 1
RðoÞ ¼ Re
2:171  106 jo  6:759  107 To satisfy (18) for all values of time t the following
2
ð4:2221o  1:4795Þ10 6 condition must be verified:
¼ . ð11Þ
10:3170o2 þ 1 f^ ^_ ¼ jf^exc j .
x^_ ¼ exc ) jxj (19)
Both (10) and (11) are plotted in Fig. 6. This seems to Z jZj
denote an inappropriate structure of the model identified in
Definition (17) can be rewritten as Z ¼ R þ iX . The
the case of (5).
real part R ¼ Re½Z is called resistance and was
For these reasons, model (7) was the one chosen.
already mentioned above in (9); the imaginary part
X ¼ Im½Z ¼ om  ðS=oÞ is called reactance.
4. Control strategies for maximising absorbed wave energy As was already mentioned above, Z takes into account
both the mechanical and the radiation effects (since the
Four control strategies will be addressed in this paper. radiation force was included in the AWS TDM). Hence
The first two are optimum control strategies, in the sense these R and X are the intrinsic resistance and the intrinsic
that they maximise (in theory) the absorption of energy. In reactance, respectively. Thus they shall henceforth be
 ARTICLE IN PRESS
2336 D. Valério et al. / Ocean Engineering 34 (2007) 2330–2344

denoted Ri and X i , and Z will be similarly denoted Z i never negative, i.e., aðoÞX0. Hence, it is when a ¼ 0 that
(intrinsic impedance). W u is maximal.
Suppose now that the control force f u is applied to the
AWS, to ensure that the conditions leading to maximum 4.1.1. Reactive control
wave energy absorption are met (or at least approached). A first way of proving that a is never negative is as
Then follows. Solving (20) in order to F exc , multiplying it by its
_
complex conjugate, F exc ðoÞ ¼ Z i ðoÞXðoÞ  F u ðoÞ, and
Z i x_ ¼ f exc þ f u 3 Z i ðoÞXðoÞ
_ ¼ F exc ðoÞ þ F u ðoÞ. (20)
omitting the frequency argument, the following expression
Let us define a control impedance Z u by is obtained:

f^u ¼ Z u x^_ 3 F u ðoÞ ¼ Z u ðoÞXðoÞ.

_ (21) F exc ¼ Z i X_  F u ) F exc F exc
The real and imaginary parts of Z u are, respectively, _   F u Z X
¼ F u F  þ Zi Z X_ X _   F  Z i X.
_ ð29Þ
u i i u
termed control resistance Ru ¼ Re½Z u  and control reac- From (17) and its complex conjugate, one possible
tance X u ¼ Im½Z u . expression for Ri is
The time-averaged absorbed wave power Pu ðtÞ can be
given by Z i þ Zi
2Ri ¼ Zi þ Z i 3 Ri ¼ . (30)
2
_
Pu ðtÞ ¼ f u ðtÞxðtÞ (22) _  þ F X_
Multiplying both sides by F u X u
and the absorbed wave energy W u can be obtained 
2Ri ðF u X_ þ F u XÞ
_ ¼ Zi F u X  
_ þ Z i F  X_ þ Z  F u X_ þ Z  F  X.
_
u i i u
integrating (22):
Z þ1 (31)
Wu ¼  _ dt.
f u ðtÞxðtÞ (23) Inserting (29) and (31) in (28) and simplifying the
0 symmetric terms
Considering that f u and x_ are real functions, i.e., _X_   F u Z X
_   F  Z i X_ þ Z i F u X_ 
_  ðoÞ ¼ XðoÞ,
_ a ¼ F u F u þ Z i Z i X
F u ðoÞ ¼ F u ðoÞ and X W u can, by applying i u

Parseval’s theorem, be given by þ Zi F  X _ þ Z F u X _  þ Z F  X

_
u i i u
Z 
¼ ðF u þ Z i X_ ÞðF u þ Z i X
_ Þ.
1 þ1 _  ðoÞ do.
ð32Þ
Wu ¼  F u ðoÞX (24)
2p 1 To complete the proof about the non-negativity of a, (32)
Knowing that W u is real, can be rewritten recovering again the frequency argument
 
_ ðoÞ ¼ Re½F u ðoÞX_ ðoÞ _
aðoÞ ¼ jF u ðoÞ þ Z i ðoÞXðoÞj 2
X0. (33)
F u ðoÞX
_  ðoÞ þ F  ðoÞXðoÞ,
¼ 1½F u ðoÞX _ ð25Þ From (33), one particular optimum condition attained
2 u
when aðoÞ ¼ 0 is
it is possible to rewrite (24) as
Z F u ðoÞ
1 þ1 ¼ Zi ðoÞ. (34)
_  ðoÞ  F  ðoÞXðoÞ
_ _
XðoÞ
Wu ¼ ½F u ðoÞX u dt. (26)
2p 0
Comparing the optimum condition in (34) with (21), it is
Since Ri is positive, it will be convenient to add and possible to see that Z u is equal to the complex conjugate of
subtract the term F exc ðoÞF exc ðoÞ=2Ri to the integrand of the intrinsic impedance Z i , i.e., Z u ¼ Zi . Another addi-
(26), and finally W u is now given by (omitting the tional conclusion is that under the optimum condition in
frequency argument in order to simplify the notation) (34) the reactive component X i ðoÞ þ X u ðoÞ of the total
Z   impedance Zi ðoÞ þ Z u ðoÞ is cancelled. This is an inherent
1 þ1 jF exc j2 jF exc j2 
Wu ¼   F u X_  F u X
_ do consequence of resonance. Reactive control is represented
2p 0 2Ri 2Ri
Z þ1   by (34), since it is related to the fact that X u (the imaginary
2
1 jF exc j a part of Z u ) will cancel X i (the imaginary part of Zi ).
¼  do. ð27Þ
2p 0 2Ri 2Ri Alternatively, this control strategy is also called complex
In (27), aðoÞ is the so-called optimum condition coefficient, conjugate control since it is related to the fact that the
given by optimum control impedance Z u;OPT equals the complex

conjugate of the intrinsic impedance Zi , i.e.,

aðoÞ ¼ F exc ðoÞF exc ðoÞ þ 2Ri F u ðoÞX_ ðoÞ þ F u ðoÞXðoÞ
_ .
Zu;OPT ¼ Z i . (35)
(28)
Consequently, from (27), the theoretical maximum ab-
There are two strategies—reactive control (also called sorbed wave energy W u;MAX is
complex conjugate control) and phase and amplitude Z
control—to find the optimum conditions. Both of them 1 þ1 jF exc ðoÞj2
W u;MAX ¼ do. (36)
are based on the same requisite, which is to prove that a is 2p 0 2Ri
 ARTICLE IN PRESS
D. Valério et al. / Ocean Engineering 34 (2007) 2330–2344 2337

Notice that the integrand of (36) is positive since Ri is also similar to (12), exists. The controller will be designed using
positive. the linear model and then applied to the non-linear WEC.
Results cannot, of course, be expected to be as good as they
4.1.2. Phase and Amplitude control would with a linear plant.
Alternatively to the previous one, another proof that a is
never negative is as follows. We now solve (20) in order to 4.1.4. Implementation
F u , instead of F exc . Inserting the obtained result and its The material in this subsection allows conceiving two
complex conjugate, F u ðoÞ ¼ Z i ðoÞX_  ðoÞ  F  ðoÞ, in (28), different optimum control strategies suited to the AWS:
exc
and omitting once more the frequency argument, in order reactive control (or complex conjugated control) and phase
to simplify the notation, and amplitude control (Valério et al., 2007b).
_X_   F exc X
_  þ Z X_  X
_  F  XÞ_ In what the application of optimum control to the linear
a ¼ F exc F exc þ 2Ri ðZi X i exc second-order model of the AWS (7) is concerned, a
 
_X
¼ F exc F exc þ 2Ri ðZi þ Z i ÞX _  2Ri F exc X_  2Ri F  X. _ Simulink block diagram comprising the two control
exc
ð37Þ strategies mentioned above is shown in Fig. 7. The one
used can be chosen by means of switch 1 (set for reactive
Inserting (30) in (37) control in Fig. 7). The second switch allows choosing the
 
_ X_  2Ri F exc X
a ¼ F exc F exc ð2Ri Þ2 X _  2Ri F  X_ model of the AWS (the linear model (12) or the non-linear
exc
_  _  AWS TDM).
¼ ðF exc  2Ri XÞðF exc  2Ri X Þ. ð38Þ In that diagram, reactive control is implemented repla-
To complete the proof about the non-negativity of a, (38) cing the non-causal transfer function Zi ðsÞ with
can be rewritten recovering again the frequency argument Z i ðsÞ=ðs þ 1Þ, the extra pole placed at 1 ensuring
_ 2 causality. Several locations have been tested for the pole,
aðoÞ ¼ jF exc ðoÞ  2Ri XðoÞj X0. (39)
and the one leading to a larger absorption of wave energy
Since Ri is positive, from (39) an alternative optimum was kept. An alternative procedure would have been to
condition can be written as identify from the frequency response of Zi a causal,
_ stable, approximate transfer function with a similar
_ XðoÞ 1
F exc ðoÞ ¼ 2Ri XðoÞ 3 ¼ . (40) response in the frequency range of interest; this approach
F exc ðoÞ 2Ri
was pursued, but led to no acceptable results.
The optimum condition in (40) is called phase and A proportional controller is used together with phase
amplitude control since it means that X_ must be in phase and amplitude control to drive the floater’s vertical
with F exc , and also that the amplitude of the floater’s velocity to the optimum value reckoned by 1=2Ri . Notice
_ must be equal to F exc =2Ri .
vertical velocity jXj that in our case this transfer function is constant
(and hence causal). The controller was obtained maximis-
4.1.3. Feasibility of optimum control strategies ing the absorbed wave energy with the MatLab function
A serious problem with both optimum conditions is that fminsearch (simplex direct search method), the opti-
they include (in the general case) non-causal transfer mum being 5:1348  104 . Integral and derivative terms
functions: Z i ðoÞ in (34), and 1=2Ri in (40) (recall that in (forming a PID controller) did not improve results.
the general case Ri may vary with frequency). This last Absorbed wave energy is obtained with a variation of
optimum condition even requires foreknowledge of the (23), because f u corresponds to the force exerted both by
wave excitation force, which in practice is available by the ELG and the water dampers, but the energy absorbed
means of predictions based on data measured by a buoy (or by the latter is wasted. Only the energy absorbed by the
buoys) placed near the AWS. To make things worse, both ELG is used; hence we will have
conditions often lead to control forces that assist the wave Z þ1
excitation force; in other words, instead of extracting Wu ¼  _ dt.
f lg ðtÞxðtÞ (41)
energy from the waves, we will be supplying energy to the 0

waves. This will of course happen only in a small fraction The implementation of optimum control together with
of the time and is necessary to extract the maximum the non-linear AWS TDM is again that of Fig. 7. Absorbed
possible wave energy during the remaining time. But if it is wave power is obtained with a Simulink block simulating
impossible to do so we will be limited to a sub-optimum the ELG (with non-linear dynamics and saturations).
solution. Actually all approximations indulged in to make
transfer functions in (34) and (40) causal will also decrease 4.2. Phase control by latching
wave energy extraction and place us in a sub-optimum
solution. But being sub-optimum should not be seen as a Because of the difficulties associated with putting the
major drawback. At least it can be realisable, something floater’s vertical velocity in phase with the wave excitation
which optimum solutions cannot. force, a sub-optimum alternative we can resort to is to latch
Both control strategies can also be applied to a non- the floater during some periods of its oscillation and
linear WEC, provided that a valid linear model thereof, unlatch it so that it will be (as nearly as possible) in phase
 ARTICLE IN PRESS
2338 D. Valério et al. / Ocean Engineering 34 (2007) 2330–2344

Fig. 7. Block diagram for optimum control.

with the wave excitation force; more precisely, the floater will Table 5
be latched when its velocity vanishes, and released when it is Latching force in kN
predicted that its maximum (or minimum) will coincide (in Wave period (s) Wave amplitude (m)
time) with the maximum (or minimum) of the wave excitation
force (Falnes, 1993, 2002; Greenhow and White, 1997; 0.25 0.5 1.0 1.5 2.0
Babarit et al., 2004). This latching control is sub-optimum 8 95 285 570 855 1425
since it can never achieve the wave energy absorption 10 95 285 760 1520 1995
efficiency that the optimum control would achieve. 12 190 380 1235 1900 2185
Latching is clearly a discrete, highly non-linear form of 14 190 380 1520 2185 1995
control. In what concerns the AWS, latching is achieved by
actuating the water dampers so as to prevent (as much as
possible) the floater from moving; unlatching means amplitude and period of the incoming wave, larger waves
turning the water dampers off to let the floater go about requiring a larger force and smaller waves requiring a
(as much as possible) freely. smaller force. The forces for each wave amplitude and
The following unlatching strategies were implemented period are those necessary to latch effectively the floater
(Beirão et al., 2006): when the incident wave is regular and has the required
amplitude and period. Values are given in Table 5 and are
(1) The latching time is constant. This is only reasonable interpolated and extrapolated as needed.
when the incident wave is regular (sinusoidal). Hence (4) Same as above, but the duration of the next unlatched
this strategy was used for testing only, and will not be period is assumed to be equal to the last one corrected
addressed further. according to the ratio between the next wave amplitude
(2) When the floater is latched, the duration of the last and the previous.
unlatched period is obtained. The next unlatched (5) Same as above, but the duration of the next unlatched
period is assumed to be going to last the same as the period is assumed to be equal to half of the floater’s
previous one. The floater’s vertical velocity is assumed natural period of oscillation, which is 11 s. Thus, the floater
to have its maximum (or minimum) precisely at the is unlatched 2.75 s (one quarter of the natural period)
middle of that time interval. So the latching time is before the next maximum (or minimum) of the wave
reckoned for that velocity maximum (or minimum) to excitation force (Falnes, 1993, 1997; Eidsmoen, 1998).
coincide in time with the next maximum (or minimum)
of the wave excitation force. The force required from 4.3. Feedback linearisation control
the water dampers to latch the floater is constant.
(3) The same as above, say that the force required from the The dynamics described by (2) are far from being linear.
water dampers to latch the floater depends on the The expressions for nearly all the forces involved show that
 ARTICLE IN PRESS
D. Valério et al. / Ocean Engineering 34 (2007) 2330–2344 2339

they depend from variables such as floater’s vertical assumed (as seen below) that f exc is known beforehand,
position, velocity, or acceleration. Heavy non-linearities, max j f exc j will be a constant value.
both smooth (with continuous derivatives of any order) This control law was chosen because, as shown above in
and hard (without continuous derivatives; this refers to Section 4.1, the optimum control of an oscillating WEC is
such common non-linearities as saturations, dead-zones, or the one providing a velocity in phase with the wave
hysteresis), are always present. excitation force.
This makes the AWS a suitable candidate for a non-
linear control strategy called feedback linearisation control 4.3.3. Third control strategy
(Valério et al., 2007a). Its aims are to provide a control The original version of the AWS TDM was implemented
action judiciously chosen to cancel the non-linear dynamics together with a simple proportional controller that may easily
of the plant, so that the closed-loop dynamics will be (as be replaced by another one to test any desired control strategy.
much as possible) linear (Slotine and Li, 1991). This This original controller provides a control force given by
method can be applied to plants that can be put into
_ p jx_  x_ stp j,
jf u j ¼ jxjk ð49Þ
companion form; this is the case of (2), since it suffices to
€ 8 sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 2ffi
solve it in order to x. >
>
< 3:5 1  x
2p
From (2)–(4), we have _ if jxjo3:5 m;
jxstp j ¼ 10 3:5 ð50Þ
f u ¼ ðmf þ mwt Þx€  f pi þ f hs þ f rad  f exc þ f n >
>
:
0 if jxjX3:5 m:
þ f v þ f m þ wf . ð42Þ
In (49), kp is the adjustable gain of the proportional controller.
Let us provide a control action given by Constant 3.5 m shows up in (50) because it is the maximum
f u ¼ f pi  f hs  f rad  f n  f v  f m  wf þ n. (43) possible amplitude for the floater’s vertical oscillations, while
constant 10 s shows up because it is a reasonable value for the
This is possible because there are explicit (non-linear) period of an incoming wave (Pinto, 2004). Suppose that we
expressions for all the forces involved in the right-hand want to follow the set point reference x_ stp for the floater’s
member of (43). Here n is an additional term that will be velocity, using feedback linearisation control. Then we would
chosen to provide the desired (linear) closed-loop dy- like to have
namics. Replacing (43) into (42), we will have the following
dynamics: dx_ stp
ðmf þ mwt Þx€ ¼ ðmf þ mwt Þ . (51)
dt
n ¼ ðmf þ mwt Þx€  f exc . (44)
By comparison with (44), we see that we must have
Three different values for n were considered.
dx_ stp
n ¼ ðmf þ mwt Þ þ f exc . (52)
4.3.1. First control strategy dt
Let
n ¼ 0. (45) 4.3.4. Implementation
The three control laws (defined by (43) together with one
Then, from (44),
of (45), (47), or (52)) from the previous section were
€
f exc ¼ ðmf þ mwt Þx. (46) implemented in the AWS TDM. Concerning (3), whenever
the ELG was able to exert the control force f u all alone, the
In other words, the floater’s vertical acceleration will be
water dampers were not resorted to (f wd ¼ 0, f lg ¼ f u ).
proportional to the wave excitation force, according to
They were used only when f u was beyond the limits of
Newton’s law, as though there were no other effects at all.
the ELG.
4.3.2. Second control strategy
Let 5. Results

max j f exc j 5.1. Simulations

n ¼ ðmf þ mwt Þx€ þ x_ . (47)
2:2
Replacing this in (44) we get Six hundred seconds (10 min) long simulations were
carried out to test these control strategies. As mentioned
max j f exc j 2:2
f exc ¼ x_ 3 x_ ¼ f . (48) above, full knowledge of how the incident wave will behave
2:2 max j f exc j exc in the future is assumed, as done for instance in Falnes
In other words, the floater’s vertical velocity will be in (1997); Eidsmoen (1998); Babarit et al. (2004). This unreal
phase with the wave excitation force. In (47) and (48), assumption will have to be dropped in future research, but,
constant 2:2 appears because the nominal value for the for now, the independent problem of wave prediction,
floater’s velocity that the AWS should work with is 2.2 m/s either from past data or from measurements done around
(Polinder et al., 2004). Also notice that, since it will be the WEC (Naito and Nakamura, 1985), was not tackled,
 ARTICLE IN PRESS
2340 D. Valério et al. / Ocean Engineering 34 (2007) 2330–2344

Table 6
Power in kW obtained under several regular waves (figurative data)

Wave amplitude (m) 0.5 0.75 1.0 1.25

Wave period (s) 14.0 12.0 10.0 8.0

AWS TDM, no control 1.0 3.2 10.2 16.7

Linear model, reactive control 6.2 17.6 39.8 78.2
% increase from no control 520 450 290 368
AWS TDM, reactive control 23.8 43.0 93.7 124.1
% increase from no control 2280 1244 819 643
AWS TDM, phase and amplitude control 17.3 47.8 87.8 98.0
% increase from no control 1630 1394 761 487

AWS TDM, latching strategy 2 1.9 7.8 7.3 70.6

% increase from no control 90 144 28 323
AWS TDM, latching strategy 3 36.5 53.0 119.6 153.9
% increase from no control 3550 1556 1073 822
AWS TDM, latching strategy 4 36.5 53.0 119.6 153.3
% increase from no control 3550 1556 1073 822
AWS TDM, latching strategy 5 10.1 40.1 71.5 56.3
% increase from no control 910 1153 601 237
AWS TDM, feedback linearisation, strategy 1 19.2 51.7 97.2 122.3
% increase from no control 1820 1516 853 632
AWS TDM, feedback linearisation, strategy 2 18.5 65.5 137.3 174.5
% increase from no control 1750 1947 1246 945
AWS TDM, feedback linearisation, strategy 3 20.1 45.5 71.6 72.5
% increase from no control 1910 1322 602 334

but postponed to some later opportunity. In all cases, the significantly improve the performance of the AWS in what
absorbed wave energy is given by (41) with integration wave energy absorption is concerned. Since nowadays the
limits from 0 to 600 s. From these simulations, values for major problem of WECs is their low efficiency, these are
the absorbed wave power (time-averaged in the case of very satisfactory and promising results. But there are
irregular waves) are given in Tables 6 and 7. For significant differences between them (Beirão et al., 2007a).
comparison purposes, absorbed wave power when the Concerning the two optimum control strategies, it is seen
AWS has no control at all are also given.2 Notice that there that they do not work equally well in practice: reactive
is only one case, that of reactive control, in which it is control works best during the whole year, while phase and
possible to present results from simulations with the linear amplitude control always lags behind, and provides
model (7), since the pole added to make the controller disappointing results from May to September (when there
causal leads immediately to a sub-optimum strategy. When is less wave energy). Fig. 8 (showing 20 s highlights, from
no control is applied, there is no control force, and hence the simulations corresponding to March and June, of the
no energy extraction, in that case. And when phase and floater’s velocity, together with the wave excitation force,
amplitude control is employed, everything being linear and for the several control strategies) gives an insight into the
no saturations existing, control works too well and values reason why this is so (similar results are obtained for all
obtained have magnitudes absolutely impossible to obtain months). Phase and amplitude control manages to put the
with the AWS. Finally, by their very nature, latching floater’s velocity nearly in phase with the wave excitation
control and feedback linearisation control cannot be force; reactive control does the same but not so efficiently.
simulated with a linear model. Nevertheless, the magnitude of the floater’s velocity is
clearly larger in this latter case, and so the higher phase
difference is more than compensated. Results obtained (for
5.2. Comments irregular waves) with the linear model and those obtained
with the AWS TDM can be related using a linear
The main conclusion to take from the results obtained regression:
with the AWS TDM is that nearly all control strategies
power obtained with the AWS TDM ¼ 27:12 kW
2
Values given in this paper for absorbed wave power when the AWS has
no control follow Valério et al. (2007a, b); Beirão et al. (2007a) and are
þ 1:514  power obtained with the linear model. ð53Þ
higher than those given in Beirão et al. (2006). This is because a residual
force exerted by the ELG that had been neglected was now taken into Even though this regression has a significant error (the
account, which seems to be more correct. maximal residual being 14.5 kW), it shows that the trend
 ARTICLE IN PRESS
D. Valério et al. / Ocean Engineering 34 (2007) 2330–2344 2341

Table 7
Power in kW obtained under several irregular waves (figurative data)

Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec

AWS TDM, no control 14.3 11.4 9.1 7.7 4.5 4.5 3.5 3.4 4.5 6.7 9.9 13.6
Linear model, reactive control 60.6 51.9 59.1 42.1 30.1 36.8 28.5 23.8 33.1 43.2 50.9 73.5
% increase from no control 324 355 549 447 569 718 714 600 636 545 414 440
AWS TDM, reactive control 79.1 65.5 50.1 45.3 18.9 14.2 10.8 12.2 21.8 35.5 55.4 73.5
% increase from no control 453 475 451 488 320 216 209 259 384 430 460 440
AWS TDM, phase & ampl. ctrl. 77.1 61.6 38.7 35.2 10.2 7.2 4.9 6.0 12.4 24.7 47.3 67.6
% increase from no control 439 440 325 357 127 60 40 76 176 269 378 397
AWS TDM, latching strategy 2 42.3 41.0 33.5 32.7 17.2 14.5 11.0 13.9 19.8 27.3 37.7 42.2
% increase from no control 196 260 268 325 282 222 214 309 340 307 281 210
AWS TDM, latching strategy 3 108.5 87.9 67.6 64.0 22.6 15.7 10.4 15.6 26.9 48.5 76.8 87.7
% increase from no control 658 674 644 736 401 252 201 364 493 624 673 545
AWS TDM, latching strategy 4 91.4 85.2 62.7 61.4 22.1 15.2 9.6 15.2 26.0 45.7 72.8 85.9
% increase from no control 539 647 589 697 391 238 174 347 478 582 635 532
AWS TDM, latching strategy 5 50.8 35.0 18.7 15.3 3.3 2.4 1.5 1.7 4.0 10.2 25.3 39.6
% increase from no control 255 207 105 99 27 47 57 50 11 52 156 191
AWS TDM, feedback lin., strat. 1 85.2 69.5 47.0 43.5 14.6 10.6 7.4 9.0 17.3 31.8 56.2 75.6
% increase from no control 496 509 416 465 224 135 112 164 284 374 467 456
AWS TDM, feedback lin., strat. 2 115.1 96.2 73.6 68.9 27.7 20.4 15.0 18.3 32.6 53.2 81.8 106.4
% increase from no control 705 744 709 795 516 353 330 437 623 695 726 683
AWS TDM, feedback lin., strat. 3 57.1 45.0 28.4 26.2 8.0 5.7 3.9 4.8 9.6 18.5 35.1 49.0
% increase from no control 300 295 212 241 77 26 11 41 112 176 254 260

Irregular wave for March Irregular wave for June

2 1

1.5
velocity / m s-1; force / MN

0.5
velocity / m s-1; force / MN

0.5
0
0

-0.5
-0.5
1

-1.5 Fexc phase and amplitude -1

none latching, strat. 3
-2 reactive feedback lin., strat. 2

-2.5 -1.5
410 415 420 425 430 410 415 420 425 430
time / s time / s

Fig. 8. Wave excitation force and floater’s vertical velocity.

obtained with the (much simpler) linear simulations is in waves, strategies 2, 3, and 4 are comparable; 3 is
reliable and allows inferring the behaviour of the non- always the best (with one single exception, and that by a
linear model. narrow margin). During the rest of the year (the
Concerning latching control, results for regular waves October–April period), when there is more energy
show that latching strategies 3 and 4 have nearly the available, strategies 3 and 4 are clearly better than all
same performance (to no surprise: since the wave is regular, others. Strategy 5, though improving energy absorption
the duration of each heave motion is always the same, over the situation without control, never leads to accep-
and there is no need of correcting the duration of the last table results. It is clear that strategy 4 is not a good
unlatched period; the usefulness of such a correction—if improvement over 3. Its more complicated algorithm
any—is to be assessed with irregular waves). Results seldom leads to a better performance. From this analysis,
for irregular waves show that winter and summer months it is clear that strategy 3 is the one to choose if latching
are rather different. During summer (loosely defined as the control is resorted to, and will be the only one addressed in
May–September period), when there is less energy available further analysis.
 ARTICLE IN PRESS
2342 D. Valério et al. / Ocean Engineering 34 (2007) 2330–2344

Concerning feedback linearisation control, the energy is not perfect. The ELG and the water dampers do not
absorption increases observed with all strategies are often suffice to effectively prevent the floater from moving, but
achieved in spite of the desired linear closed-loop dynamics they can hinder it well enough.) When phase and amplitude
not being obtained. This is because, with strategy 1, the control is employed, the amplitude of the floater’s
floater’s acceleration only rarely is in phase with the wave oscillations is relatively small (as can be seen from the
excitation force, and with strategy 3 the set point for the small values of the velocity), and this accounts for the poor
floater’s velocity is always very far from being attained performance. At a first glance, latching control might seem
(consequently the energy absorption improvement is not as to have to be better than feedback linearisation control,
high as with the other strategies). Strategy 2, on the other since the floater’s vertical oscillations have a larger
hand, achieves (as seen in Fig. 8) better results in causing amplitude. The reason why this is not so is seen in the
floater’s velocity and wave excitation force to be in phase plots of Fig. 10. The absorbed wave energy (41) is
(especially in what irregular waves are concerned), and computed as the integral of the product of two oscillating
hence the better results achieved. The deviations from the variables (the force exerted by the ELG and the floater’s
desired behaviour are likely to derive from the simplifica- vertical velocity). If both these oscillations have the same
tions that had to be admitted when designing the controller, amplitude, the absorbed wave energy will increase as the
namely assuming that the actuators (the ELG and the water phase difference decreases, since the time period when the
dampers) respond immediately and have no saturations product is negative will diminish. For the same phase
(when this is not the case, for each has its own internal difference between the two variables, absorbed wave
dynamics, and saturation values; this is especially the case of energy will increase with the increase of the magnitude of
the water dampers, and so f wd will not always follow its set either of the variables. And, indeed, the force exerted by
point). Since results show that strategy 2 is the best, in what the ELG is clearly much larger with feedback linearisation
wave energy absorption is concerned, this is the one to than with latching control. That is why absorbed wave
choose if feedback linearisation control is resorted to, and energy attains its maximum with feedback linearisation.
will be the only one addressed in further analysis. Also notice that latching and feedback linearisation are
Overall, feedback linearisation is clearly the best control the control strategies requiring the larger efforts from the
strategy, in what absorbed wave energy is concerned, and water dampers, that nearly are not needed with reactive
this happens all over the year, for all sea states. (The case of control or phase and amplitude control. The energy they
the regular wave of 0.5 m of amplitude and period of 14 s absorb is dissipated, but nevertheless what the ELG
does not seem to be relevant.) This is clear from the receives is more than if they were not used.
evolution of wave energy absorption with time for the The two (supposedly) optimum control strategies are not
several control strategies given in Fig. 9. the best performing ones, likely because they are sub-
The analysis of the plots in Fig. 8 provides an insight optimally implemented. The approximations used to
into the reason why feedback linearsation is the best. The ensure causality prevent them from performing as well as
requirement that wave excitation force and floater’s they should in the ideal case (the only one really optimum).
velocity be in phase, as seen in (40), is reasonably fulfilled
in the case of feedback linearisation, of latching, and of 5.3. Variations around the year
phase and amplitude control, but no so well accomplished
by reactive control. (Notice that the latching of the floater Table 1 shows that the energetic content of waves is not the
same around the year. This is reflected in the variations of
x 107 Irregular wave for March wave power Pu absorbed by the AWS, whatever the control
5 strategy employed. Fig. 11 attempts to explain these
feedback linearisation, strategy 2
variations as functions of H s , that suffices to explain much
latching, strategy 3 of the monthly variations (the maximum and minimum
4
values of the wave energy period T e have also been tested as
absorbed energy / J

reactive possible explanations but seem to be irrelevant). It is possible

3 to use the least-squares method to find relations between Pu
and H s . These can be given by linear expressions, but it is
phase
2 and
more reasonable (as can be seen by visual inspection of the
amplitude plots) to use quadratic expressions, as follows:

1 No control:
none

Pu ¼ 2:71H 2s  6:55H s þ 7:30. (54)

0
0 100 200 300 400 500 600
Reactive control:
time / s

Fig. 9. Evolution of absorbed wave energy with time (figurative data). Pu ¼ 8:59H 2s  0:34H s  8:78. (55)
 ARTICLE IN PRESS
D. Valério et al. / Ocean Engineering 34 (2007) 2330–2344 2343

Reactive control Phase and amplitude control

1.5 1
Fwd
Flg
1

velocity / m·s-1; force / MN

velocity
velocity / m·s-1; force / MN

0.5

0.5
0
0

-5
-0.5

-1 -1
410 415 420 425 430 410 415 420 425 430
time / s time / s
Control by latching Control by feedback linearisation
2 6

1.5
4
1
velocity / m·s-1; force / MN

velocity / m·s-1; force / MN

0.5 2

0 0
-0.5
-2
-1

-1.5 -4

-2
-6
-2.5
410 415 420 425 430 410 415 420 425 430
time / s time / s

Fig. 10. Evolution of the control forces with time (figurative data) for a March irregular wave.

Latching control, strategy 3:

Jan

120
Dec

none
Pu ¼ 5:21H 2s þ 30:07H s  47:65. (57)
Feb

reactive
100
Feedback linearisation control, strategy 2:
Nov

phase and amplitude

absorbed power / kW

Mar

latching, strat. 3
80 Pu ¼ 9:45H 2s þ 14:05H s  28:46.
Apr

(58)
feedback lin., strat. 2
Oct

60
6. Conclusions
Sep
May

40
The main conclusion to take is that phase and amplitude
Jun
Aug

control are better than no control, reactive control is better

Jul

20
than phase and amplitude control, latching control (with
0
the best strategy conceived) is better than reactive control,
1 1.5 2 2.5 3 3.5 and feedback linearisation control (with the best strategy
Hs / m conceived) is the best. This is true for all months of the
year, and also for most regular waves (though not for all).
Fig. 11. Evolution of the absorbed wave power with H s . Nevertheless, irregular waves are expected to be closer to
the real sea states the AWS will have to work in, and so
Phase and amplitude control: their results are deemed more significant.
In principle, reactive control and phase and amplitude
Pu ¼ 18:41H 2s  45:01H s þ 31:06. (56) control should perform better. But these two optimum
 ARTICLE IN PRESS
2344 D. Valério et al. / Ocean Engineering 34 (2007) 2330–2344

(non-causal) control strategies cannot be implemented in Falnes, J., 1993. Optimal control of oscillation of wave-energy converters.
practice without approximations and modifications (to Technical Report, Commission of the European Communities, DG
make sure that all transfer functions are causal) that make XII Joule Wave Energy Initiative. hhttp://www.phys.ntnu.no/instdef/
them sub-optimum. And so latching control and feedback prosjekter/bolgeenergi/optcontr.pdfi. Annex to ‘‘Wave energy con-
verters: generic technical evaluation study’’.
linearisation control (that are intrinsically sub-optimum)
Falnes, J., 1997. Principles for capture of energy from ocean waves: phase
are not in disadvantage.
control and optimum oscillation. Internet Web page. hhttp://
Further refinements in control algorithms may be www.phys.ntnu.no/instdef/prosjekter/bolgeenergi/phcontrl.pdfi.
possible. The influence of the control algorithm on the Falnes, J., 2002. Ocean Waves and Oscillating Systems. Cambridge
performance of the converter that supplies the grid is still University Press, Cambridge.
to be investigated. Algorithms for estimating the incident Greenhow, M., White, S.P., 1997. Optimal heave motion of some
wave and the consequent wave excitation force from data axisymmetric wave energy devices in sinusoidal waves. Applied Ocean
collected by buoys placed around the WEC or by some Research 19, 141–159.
other means have to be studied and improved. But the Levy, E., 1959. Complex curve fitting. IRE Transactions on Automatic
Control 4, 37–44.
main future task will certainly be the application of these
Naito, S., Nakamura, S., 1985. Wave energy absorption in irregular waves
control strategies to the second generation of AWS
by feedforward control system. In: Hydrodynamics of Ocean-wave
prototypes—first in simulation, then in hardware, to help Energy Utilization. Evans, D.V., Falcão, A.F.O. (Eds.), Springer,
making them economically viable alternative sources of Berlin, pp. 169–180.
electrical power. Pinto, P., 2004. Time domain simulation of the AWS. Master’s Thesis,
Technical University of Lisbon, IST, Lisbon.
Acknowledgements Polinder, H., Damen, M., Gardner, F., 2004. Linear PM generator system
for wave energy conversion in the AWS. IEEE Transactions on Energy
Duarte Valério was partially supported by grant SFRH/ Conversion 19 (3), 583–589.
BPD/20636/2004 of FCT, funded by POCI 2010, POS C, Pontes, M.T., Aguiar, R., Oliveira Pires, H., 2005. A nearshore wave
energy atlas for Portugal. Journal of Offshore Mechanics and Arctic
FSE, and MCTES. Pedro Beirão was partially supported
Engineering 127, 249–255.
by the ‘‘Programa do FSE-UE, PRODEP III, acc- ão 5.3, Sá da Costa, J., Pinto, P., Sarmento, A., Gardner, F., 2003. Modelling and
III QCA’’. Research for this paper was partially supported simulation of AWS: a wave energy extractor. In: Proceedings of the 4th
by grant PTDC/EME-CRO/70341/2006 of FCT, funded IMACS Symposium on Mathematical Modelling. Agersin, Vienna,
by POCI 2010, POS C, FSE and MCTES. pp. 161–170.
Sá da Costa, J., Sarmento, A., Gardner, F., Beirão, P., Brito-Melo, A.,
References 2005. Time domain model of the Archimedes Wave Swing wave energy
converter. In: Proceedings of the 6th European Wave and Tidal Energy
Babarit, A., Duclos, G., Clément, A., 2004. Comparison of latching Conference, Glasgow, pp. 91–97.
control strategies for a heaving wave energy device in random sea. Slotine, J., Li, W., 1991. Applied Nonlinear Control. Prentice-Hall,
Applied Ocean Research 26, 227–238. Englewood Cliffs, NJ.
Beirão, P., Valério, D., Sá da Costa, J., 2006. Phase control by latching Valério, D., Sá da Costa, J., 2007. Identification of fractional models from
applied to the Archimedes Wave Swing. In: Proceedings of the 7th frequency data. In: Tenreiro Machado, J., Sabatier, J., Agrawal, O.
Portuguese Conference on Automatic Control, Lisbon. (Eds.), Advances in Fractional Calculus: Theoretical Developments
Beirão, P., Valério, D., Sá da Costa, J., 2007a. Comparison of control
and Applications in Physics and Engineering. Springer, Berlin.
strategies applied to the Archimedes Wave Swing. In: European
Valério, D., Beirão, P., Sá da Costa, J., 2007a. Feedback linearisation
Control Conference, Kos.
control applied to the Archimedes Wave Swing. In: 15th IEEE
Beirão, P., Valério, D., Sá da Costa, J., 2007b. Linear model identification
of the Archimedes Wave Swing. In: IEEE International Conference on Mediterranean Conference on Control and Automation, Athens.
Power Engineering, Energy and Electrical Drives, Setúbal, pp. 660–665. Valério, D., Beirão, P., Sá da Costa, J., 2007b. Reactive control and phase
Eidsmoen, H., 1998. Tight-moored amplitude-limited heaving-buoy wave- and amplitude control applied to the Archimedes Wave Swing. In:
energy converter with phase control. Applied Ocean Reserach 20, 17th International Offshore (Ocean) and Polar Engineering Con-
157–161. ference & Exhibition, Lisbon.