Sie sind auf Seite 1von 6

An Application of Model Predictive Control to

a Wave Energy Point Absorber


J. Cretel*. A. W. Lewis*
G. Lightbody**. G. P. Thomas***

*Hydraulics and Maritime Research Centre, University College Cork,
Cork, Ireland (Tel: (+353)(0)21 4250031; e-mail: j.cretel@student.ucc.ie).
**Department of Electrical and Electronic Engineering,
University College Cork, Cork, Ireland.
***Department of Applied Mathematics,
University College Cork, Cork, Ireland.

Abstract: Optimal performance of wave energy converters requires appropriate control strategies. This is
especially true of wave energy point absorbers, which are relatively small oscillators excited by waves.
The two control methods for point absorbers which are most studied in the literature are reactive control
and latching, which have major deficiencies. This article outlines a time-domain control method based on
Model Predictive Control, which can be applied to any wave energy point absorber whose behaviour can
be described by a linear state-space model. The control method is applied here to a semi-immersed
vertical cylinder in deep water, excited by regular or irregular waves and oscillating in heave only; the
motion may be either free or subject to amplitude constraints. Preliminary results from numerical
simulations are presented and discussed. This control approach aims to obviate some of the limitations of
already existing control strategies and to pave the way towards better control methods for point
absorbers.
Keywords: wave energy, point absorber, latching, reactive control, model predictive control, oscillator,
convex optimization, quadratic programming.

combination of large amounts of energy flowing in and out of
1. INTRODUCTION
the system and losses in the conversion chain generally
Optimal control of wave energy converters is fundamental if prohibits the use of reactive control for wave energy harvest.
wave energy is to achieve its full potential and significantly
Latching control, as proposed by Budal & Falnes (1980),
contribute to the energy mix on a global scale. This is
does not involve any bidirectional flow of energy. Point
especially important for point absorbers (Falnes, 2002a),
absorbers generally have a natural frequency larger than
which can generally be described as oscillators, excited by
wave excitation frequencies and therefore tend to lead in
waves, whose horizontal dimensions are much smaller than
phase. By latching the oscillator, i.e. locking it into position
the prevailing wavelength. Such devices are intended to be
during part of the wave cycle, it is possible to approximately
deployed in arrays of several units, a few kilometres off the
achieve resonance, thus improving the performance over
coast.
passive control (fixed damping). Originally developed for the
Advanced control methods are expected to markedly improve case of sinusoidal excitation, latching control was extended to
the performance over passive control, without generating irregular waves by Hoskin (1988) by employing Pontryagin’s
significant additional cost. Control of point absorbers has Maximum Principle. The same technique was applied and
been extensively studied since the early days of wave energy, improved by Babarit & Clément (2006), who treated the
with two complementary reviews provided by Salter, Taylor radiation in a more rigorous manner. Eidsmoen (1998)
& Caldwell (2002) and Falnes (2002b). Prominent in the modified the basic principle of latching with the intent of
literature are reactive control and latching control, which are handling amplitude constraints. Falcão (2007) proposed a
briefly reviewed here. causal version of latching applicable to point absorbers
equipped with a conventional hydraulic PTO machinery.
Reactive control (Budal & Falnes, 1977) is generally
formulated in the frequency domain: the load impedance is Despite its popularity, latching still suffers from two major
chosen so as to verify some optimum amplitude and phase drawbacks: it may not be suitable for point absorbers
conditions (Falnes, 2002a) whereby the damping is optimal equipped with direct-drive linear generators and it is likely to
and the intrinsic reactance of the system is cancelled. Despite prove inadequate for arrays of WECs, as the optimum phase
being acausal, reactive control has been theoretically condition on which latching is founded is different if more
extended to the time domain, in which causal approximations than one oscillator are present, as stated by Falnes (1980) and
can be formulated (Naito & Nakamura, 1985). However, the Thomas & Evans (1981).
The development of new control methods for point absorbers expressed in terms of the product of the hydrostatic stiffness
is desirable, if wave energy is to succeed commercially 𝑘𝑕 and the vertical displacement, i.e.
(Ringwood, 2006). With the advent of more controllable,
𝑓𝑕 (𝑡) = −𝑘𝑕 𝑧(𝑡) . (2)
more efficient PTO systems for WECs, such as linear
generators (Baker, 2003) and smart variable-displacement Cummins (1962) expresses the radiation force as
hydraulics (Payne et al., 2005), it is becoming possible to 𝑡
explore control strategies that optimise the control force in 𝑓𝑟 (𝑡) = −𝜇𝑧(𝑡) − 0
𝑕𝑟 𝜏 𝑧 𝑡 − 𝜏 𝑑𝜏 , (3)
real time, as encouraged by Salter, Taylor & Caldwell (2002)
where 𝜇 is the added mass at infinity (representing the inertia
and Molinas et al. (2007).
of the surrounding fluid) and 𝑕𝑟 is the radiation kernel (also
Model Predictive Control (MPC) (Rossiter, 2003) appears as called retardation function). The convolution integral reflects
a promising candidate for this type of approach. MPC was the fluid memory effect, whereby the motion of the float at
first applied to a wave energy point absorber by Gieske time 𝑡 affects the motion of the surrounding fluid, not only at
(2007) and Hals (2009) recently applied MPC to a heaving time 𝑡 but also at subsequent times. The surrounding fluid
point absorber, subject to amplitude constraints, with a more effectively retains in memory the past trajectory of the float
rigorous treatment of the radiation. and in turn affects the motion of the float via the radiation
force. The added mass at infinity and the radiation kernel are
The MPC-based method outlined in this article remedies
connected to the frequency-domain hydrodynamic
some shortcomings of Hals’ approach:
coefficients and were obtained with WAMIT® for this study.
- Observability of the proposed state-space model, a
Introducing the substitutions
practical requirement of MPC, is assessed.
𝑓 𝑃𝑇𝑂 (𝑡) 𝑓𝑤 (𝑡)
- The objective function is chosen in such a way that 𝑢(𝑡) = , 𝑢𝑤 (𝑡) = , (4)
𝑚 +𝜇 𝑚 +𝜇
no energy is allowed to be stored in the system at the
end of the horizon, either in potential or kinetic and combining (1) - (4) gives the equation of motion in the
form. form
𝑡
The control method also lays the foundations for a strategy 𝑚 + 𝜇 𝑧(𝑡) + 0
𝑕𝑟 𝜏 𝑧 𝑡 − 𝜏 𝑑𝜏 + 𝑘𝑕 𝑧(𝑡)
intended to estimate and forecast the wave force over a few = 𝑚 + 𝜇 (𝑢(𝑡) + 𝑢𝑤 (𝑡)) . (5)
seconds, which is the subject of ongoing research. It is
assumed in this study that the future wave excitation force is
known over the time horizon considered. 3. FORMULATION OF THE CONTROL
The control method outlined here is applicable to any wave A discrete-time state-space representation of the system must
energy point absorber whose behaviour can be described by a be obtained before MPC can be applied. As the convolution
linear state-space model. Furthermore, it is potentially integral present in (5) is not suitable for this application, the
extendable to arrays of such devices. The method is here Prony method described by Duclos, Clément & Chatry
applied to the classic example of a heaving, semi-immersed (2001) was used to approximate the convolution integral by
vertical cylinder in deep water, excited by regular or irregular an 𝑛𝑡𝑕 order linear model. This is represented by the
waves and either free or subject to amplitude constraints. controllable canonical form of a continuous-time SISO state-
Preliminary results from numerical simulations are presented space model, with state vector 𝒙𝒓 ∈ 𝑅𝑛 , giving the
and discussed. approximate form of the equations as
𝒙𝒓 𝑡 = 𝐴𝑟 𝒙𝒓 𝑡 + 𝐵𝑟 𝑧 𝑡 ,
2. MATHEMATICAL MODEL 𝑡
0
𝑕𝑟 𝜏 𝑧 𝑡 − 𝜏 𝑑𝜏 ≈ 𝐶𝑟 𝒙𝒓 (𝑡) .
In linear wave theory (Newman, 1977), which is assumed
throughout this paper, the total force acting in the vertical Although the components of 𝒙𝒓 do not have any physical
direction on a floating body can be expressed as a sum of 𝑓𝑕 meaning, the radiation state vector 𝒙𝒓 holds information
(the hydrostatic restoring force), 𝑓𝑤 (the wave exciting force, about the state of the surrounding fluid.
corresponding to the interactions of incident waves with the
The state and output vectors 𝒙 ∈ 𝑅𝑛+2 and 𝒚 ∈ 𝑅2
body fixed in position) and 𝑓𝑟 (the radiation force, due to
waves radiated by the oscillating body in an otherwise calm corresponding to the whole system (oscillator and
sea). The floating body of structural mass 𝑚 is constrained surrounding fluid) are respectively chosen as
to oscillate in heave only and, if 𝑓𝑃𝑇𝑂 denotes the force
applied by the PTO machinery, the time-domain equation of 𝑧
motion is 𝑧
𝒙= 𝑧 , 𝒚= .
𝒙𝒓 𝑧
𝑚𝑧(𝑡) = 𝑓𝑕 𝑡 + 𝑓𝑤 (𝑡) + 𝑓𝑟 (𝑡) + 𝑓𝑃𝑇𝑂 (𝑡) , (1)
where 𝑧(𝑡) is the vertical displacement of the float around the
equilibrium position. The hydrostatic restoring force can be
A linear time-invariant state-space representation of the
system is thus given by
𝒙 𝑡 = 𝐴𝑐 𝒙 𝑡 + 𝐵𝑐 𝑢 𝑡 + 𝐹𝑐 𝑢𝑤 𝑡 , By inspection of the observability matrix of this state-space
model, the system is fully observable. This means that
𝒚 𝑡 = 𝐶𝑐 𝒙(𝑡) , although the radiation states have no physical meaning and
cannot be measured, they can be inferred from the measured
outputs. Another important consequence of observability is
0 1 𝟎 that the wave input can be estimated, although it is not
−𝑘 𝑕 −1
𝐴𝑐 = 0 𝐶𝑟 ∈ 𝑅 (𝑛+2)×(𝑛+2) , measured. Furthermore, using this estimation, forecasting the
(𝑚 +𝜇 ) (𝑚 +𝜇 )
wave excitation force some seconds into the future may be
𝟎 𝐵𝑟 𝐴𝑟
possible. This possibility will be explored in a subsequent
article. It is assumed herein that the state of the system at the
0 current time is known, as is the future wave exciting force.
𝑛+2 ×1
𝐵𝑐 = 𝐹𝑐 = 1 ∈ 𝑅 , Introduce the notation … 𝑘 to denote the vertical
𝟎 concatenation of the considered column vector at times 𝑘 + 1
through to 𝑘 + 𝑁, where 𝑁 is the length of the horizon. The
open-loop prediction vector 𝒚(𝑘) can be expressed as a
1 0 𝟎
𝐶𝑐 = ∈ 𝑅2×(𝑛+2) . function of the current state and the future input increments,
0 1 𝟎 i.e.
𝒚 𝑘 = 𝑃𝒙 𝑘 + 𝐻𝜟𝒖 𝑘 +𝐻𝑤 𝜟𝒖𝒘 (𝑘) . (6)
A discrete-time approximation of this continuous system is
obtained by using a zero-order hold with sample time ∆𝑡 The form of the matrices 𝑃, 𝐻, and 𝐻𝑤 follow from Rossiter
(2003). The quantity sought to be maximised here is the
𝒙 𝑘 + 1 = 𝐴𝑑 𝒙 𝑘 + 𝐵𝑑 𝑢 𝑘 + 𝐹𝑑 𝑢𝑤 𝑘 , mechanical energy 𝐸𝑎𝑏𝑠 absorbed by the PTO system over
the time horizon 𝑇 and, using (5), corresponds to
𝒚 𝑘 = 𝐶𝑑 𝒙(𝑘) .
𝑡+𝑇
𝐸𝑎𝑏𝑠 = − 𝑚 + 𝜇 𝑢 𝜏 𝑧 𝜏 𝑑𝜏 .
The state and output vectors are then augmented, the input 𝑡
increments thus playing the role of inputs to the system, as Thus, the discrete-time objective function, to be minimised, is
follows chosen as
𝑁
𝐽 𝑘 = 𝑖=1 𝑢 𝑘+𝑖−1 𝑧 𝑘+𝑖 . (7)
𝒙(𝑘) It should be noted that the form of the objective function
𝒙 𝑘 = 𝑢(𝑘 − 1) ∈ 𝑅𝑛 +4 , departs from conventional MPC: since the future reference is
𝑢𝑤 (𝑘 − 1) not known in advance, the optimisation cannot be expressed
as a least-squares problem. The trajectory to be followed by
the system is instead only accessible after minimisation of
𝒚 𝑘 (7). This expression can be formulated as a quadratic function
𝒚 𝑘 = ∈ 𝑅3 , of 𝒚,
𝑢(𝑘 − 1)
1
𝐽(𝑘) = 𝒚𝑇 𝑘 𝑄𝒚 𝑘 , (8)
2
𝒙 𝑘 + 1 = 𝐴𝒙 𝑘 + 𝐵∆𝑢 𝑘 + 𝐹∆𝑢𝑤 (𝑘) , where 𝑄 ∈ 𝑅3𝑁×3𝑁 is a block diagonal matrix, whose
𝒚 𝑘 = 𝐶𝒙(𝑘) , building block 𝑀 is repeated 𝑁 times along the diagonal,
0 0 0
𝑀= 0 0 1 .
𝐴𝑑 𝐵𝑑 𝐹𝑑 0 1 0
𝑛+4 × 𝑛+4
𝐴= 𝟎 1 0 ∈𝑅 ,
Substituting (6) into (8) yields an expression of 𝐽 as a
𝟎 0 1
quadratic function of 𝜟𝒖. After removing the terms that have
no dependence in 𝜟𝒖, the following objective function is
𝐵𝑑 obtained as
𝑛+4 ×1
𝐵= 1 ∈𝑅 , 1
𝐽 = 𝜟𝒖𝑇 𝐻𝑇 𝑄𝐻𝜟𝒖 + 𝜟𝒖𝑇 𝐻𝑇 𝑄 𝑃𝒙+𝐻𝑤 𝜟𝒖𝒘 .
0 2

By inspection, the matrix 𝐻𝑇 𝑄𝐻 is positive-definite, so


𝐹𝑑 minimising 𝐽 amounts to solving a convex quadratic
𝐹= 0 ∈𝑅 𝑛+4 ×1
, programming problem (Boyd & Vandenberghe, 2004), which
is constrained if the case being considered is that of a point
1
absorber subject to amplitude constraints.
1 0 0 ⋯ 0 0 0
𝐶= 0 1 0 ⋯ 0 0 0 ∈ 𝑅3×(𝑛+4) . Only the first component of 𝜟𝒖∗ (𝑘) = 𝑎𝑟𝑔𝑚𝑖𝑛𝜟𝒖 {𝐽 𝑘 }, i.e.
0 0 0 ⋯ 0 1 0 the optimal control move at time 𝑘, is used. The same
optimisation procedure is then run again for time 𝑘 + 1 and significant performance degradation for low frequencies and
repeated for subsequent times. more so for shorter horizons, as would be expected from the
limited length of the horizon.

4. RESULTS
A semi-immersed vertical cylinder of radius 5m and draft 8m
is considered in deep water; a sample time of 0.05 s was used
for the simulations.

4.1 Open-loop system

The order 𝑛 of the system used for approximating the


radiation convolution integral was tested for a wide range of
values and 𝑛=6 was found to give the best compromise
between approximation accuracy and complexity of the
system (figure 1).
Figure 2 indicates a good agreement between displacement
amplitudes obtained from time-domain simulations of the
open-loop system undergoing monochromatic excitation and
response amplitude operators (RAOs) obtained from Figure 2. Open-loop system: comparison of the response amplitude
WAMIT®. It should be noted that in the case of an operators (RAOs) obtained from WAMIT® and from the time-
unconstrained system, linear wave theory is considered to be domain simulation, using a 6th order system.
invalid when the amplitude of motion becomes large
compared to the wave amplitude; a factor of 3 or more is
often regarded as a practical limit. This occurs around
resonance (figure 2) and is non-physical but figure 2 is
nonetheless useful to assess the agreement between the
frequency- and the time-domain models.

Figure 3. Control applied to the case of excitation by


monochromatic waves: mechanical power absorbed against angular
frequency, for three different horizon lengths.

4.3 Irregular waves with amplitude constraints

The more realistic case of a system with an amplitude


Figure 1. Approximation of the radiation impulse response function
by a 6th order system (𝑛=6). constraint of 5m and subject to irregular waves, defined by a
time series corresponding to a Bretschneider spectrum
4.2 Monochromatic waves without amplitude constraints (significant wave height: 3m; zero-crossing period: 8s), is
reported in figures 5-8. As the control tends to become too
Figure 3 confirms that, for an unconstrained system under aggressive in the constrained case, a penalty on the control
regular wave excitation, the control yields performance effort over the horizon was added to the objective function
results in accordance with those obtained by reactive control. (with 𝜆=1)
The power absorbed by passive control (with optimal
damping at the frequency considered) has been plotted for
1
comparison. The figure shows that the predictions from the 𝐽 = 𝒚𝑻 𝑄𝒚 + 𝜆𝜟𝒖𝑇 𝜟𝒖 .
2
longest horizon tested (6s) provides the best agreement with
theoretical optimum and figure 4 shows that the wave
exciting force and velocity are in phase, as is the case when
reactive control is applied. However, figure 3 indicates a
Figure 4. Control applied to the case of excitation by Figure 6. Control applied to the case of excitation by irregular waves
monochromatic waves (amplitude: 1m; period: 7s) – wave exciting – wave exciting force and device velocity.
force and device velocity.
Figure 5 indicates that the control is effective in keeping the
device within amplitude constraints. Figure 6 shows that the
wave exciting force and velocity are approximately in phase
as in the case of monochromatic excitation (figure 4).

Figure 7. Control applied to the case of excitation by irregular waves


– PTO force and device velocity.

Figure 5. Control applied to the case of excitation by irregular waves


– wave elevation and device displacement.

Figure 7 indicates that the PTO force is not in antiphase with


the velocity and a comparison of figures 6 and 7 shows that
the PTO force is much larger than the wave exciting force.
Consequently, a prohibitively large amount of energy must
flow back and forth through the conversion chain, as is
apparent on figure 8, which suggest that large excursions of
the absorbed power both in the negative and the positive
directions are occurring. This would require a high power
capacity for the PTO system and would be very detrimental
to performance, because a large fraction of the energy would
effectively be lost in the conversion process, as pointed out
by Hals, Bjarte-Larsson & Falnes (2002). It is therefore
desirable to limit the amount of mechanical reactive power
Figure 8. Control applied to the case of excitation by irregular waves
involved in the control.
– instantaneous and mean mechanical power absorbed.
5. CONCLUSION PTO system. Proc. 7th European Wave and Tidal
Energy Conference, Porto, Portugal.
The results presented in this paper confirm that Model
Falnes J. 1980. Radiation impedance matrix and optimum
Predictive Control is a method of interest for the control of a
power absorption for interacting oscillators in surface
particular wave energy converter, the heaving buoy, which is
waves. Applied Ocean Research, 2, 75-80.
classified as a point absorber and for which effective control
Falnes J. 2002a. Ocean waves and oscillating systems: linear
is considered essential if such devices are to operate
interaction including wave-energy extraction.
efficiently and profitably. The classical results are recovered
Cambridge University Press.
in the case of regular waves and unconstrained body motions,
Falnes J. 2002b. Optimum control of oscillation of wave-
under the condition that the prediction horizon is sufficiently
energy converters. International Journal of Offshore and
long. For the case of irregular waves with constrained
Polar Engineering, 12(2), 147-155.
motion, which is close to practical application, the method
Gieske P. 2007. Model Predictive Control of a Wave Energy
ensures that the constraint is satisfied and highlights an issue
Converter: Archimedes Wave Swing. M.Sc. thesis, Delft
with the PTO system that requires further consideration.
University of Technology, the Netherlands.
Subsequent work will focus on the inclusion of a state Hals J. 2009. Constrained optimal control of a heaving buoy
estimator (which opens the possibility of forecasting the wave energy converter. Norwegian University of Science
exciting wave force over the horizon) as well as on limiting and Technology, Trondheim, Norway. Accepted for
the amount of energy flowing back and forth through the publication in Journal of OMAE.
PTO system. These improvements would bring the control Hals J., Bjarte-Larsson T. & Falnes J. 2002. Optimum
method a step closer to practical application. Future research reactive control and control by latching of a wave-
is also expected to enhance the robustness of the method and absorbing semisubmerged heaving sphere. Proc. of
to cover the extension of the control strategy to nonlinear OMAE'02, 21st International Conference on Offshore
point absorbers and arrays of point absorbers. Mechanics and Arctic Engineering, Oslo, Norway.
Hoskin R.E. 1988. Optimal control techniques for water
power generation. Ph.D. thesis, University of Reading,
ACKNOWLEDGMENTS UK.
Molinas M. et al. 2007. Power electronics as grid interface
The first author is a research fellow funded by the Marie for actively controlled wave energy converters.
Curie WaveTrain 2 training network and a member of the International Conference on Clean Electrical Power,
International Network on Offshore Renewable Energy Capri, Italy.
(INORE). Naito S. and Nakamura S. 1986. Wave energy absorption in
irregular waves by feed-forward control system. In
Hydrodynamics of ocean wave-energy utilization, ed. by
REFERENCES D.V.Evans & A.F. de O. Falcão, 269-280. IUTAM
Babarit A. & Clément A.H. 2006. Optimal latching control of Symposium, Lisbon. Springer-Verlag, Berlin.
a wave energy device in regular and irregular waves. Newman J.N. 1977. Marine Hydrodynamics. MIT Press.
Applied Ocean Research, 28, 77-91. Payne G.S. et al. 2005. Potential of digital displacement™
Baker N.J. 2003. Linear generators for direct drive marine hydraulics for wave energy conversion. Proc. 6th
renewable energy converters. Ph.D. thesis, University of European Wave and Tidal Energy Conference,
Durham, UK. University of Strathclyde, Glasgow, UK.
Boyd S. & Vandenberghe L. 2004. Convex Optimization. Ringwood J. 2006. The dynamics of wave energy. Irish
Cambridge University Press. Signal and Systems Conference. Dublin, Ireland.
Budal K. & Falnes J. 1977. Optimum operation of improved Rossiter J.A. 2003. Model-Based Predictive Control: A
wave-power converter. Marine Science Communications, Practical Approach. CRC Press. Ch.3.
3, 133-159. Salter S.H., Taylor J.R.M. & Caldwell N.J. 2002. Power
Budal K. & Falnes, J. 1980. Interacting point absorbers with conversion mechanisms for wave energy. Proc. of the
controlled motion. In Power from sea waves, ed. By Institution of Mechanical Engineers, Part M—Journal of
B.M. Count, 381-399. Academic Press, London. Engineering for the Maritime Environment, 216, 1-27.
Cummins W.E. 1962. The impulse response function and Thomas G.P. & Evans D.V. 1981. Arrays of three-
ship motions. Schiffstechnik, 9, 101-109. dimensional wave-energy absorbers. J. Fluid Mech., 108,
Duclos G., Clément A.H. & Chatry G. 2001. Absorption of 67-88.
outgoing waves in a numerical wave tank using a self-
adaptive boundary condition. International Journal of
Offshore and Polar Engineering, 11(3), 168-175.
Eidsmoen H. 1998. Tight-moored amplitude-limited heaving
buoy wave energy converter with phase control. Applied
Ocean Research, 20, 157-161.
Falcão A.F. de O. 2007. Phase control through load control of
oscillating-body wave energy converters with hydraulic

Das könnte Ihnen auch gefallen