METHOD FOR PREDICTING A CHARACTERISTIC RESULTING FROM A SWELL ON THE BASIS OF A SPECTRAL MODEL OF THE SWELL

Abstract

The present invention is a method for predicting a swell-resulting characteristic for a floating system. The method updates (MAJ) a spectral model (MSH) of the swell to form a swell prediction model (MPR). This model is applied to real-time measurements (MES) to predict the swell-resulting characteristic (pred).

Claims

1-14. (canceled)

15. A method of obtaining a characteristic of a water swell acting on a floating system subjected to the water swell, by using at least one sensor to measure variation in the water swell, the method obtaining the characteristic of the water swell by using a transfer function relating the characteristic of the water swell to a measurement provided by the at least one sensor, comprising steps of: a) measuring the variation in the water swell in real time during a first time interval by using the at least one sensor; b) providing a spectral model of the water swell and updating the spectral model of the water swell in a second time interval longer than the first time interval based on at least one of meteorological data and at least one measurement of the at least one sensor; c) determining a water swell prediction model by using the transfer function and the updated spectral model of the water swell; and d) obtaining the characteristic of the water swell in real time for a future time period by using the swell prediction model applied to real-time measurements.

16. The method for predicting the characteristic recited in claim 15, wherein the floating system is a wave-energy converter for converting energy of the water swell into electrical, pneumatic or hydraulic energy acting on a vessel, a floating platform, a floating wind turbine, an amphibious vehicle or a plane.

17. The method for predicting the characteristic as claimed in claim 15, wherein the at least one sensor is a sensor chosen from one of radar, lidar, a sensor of deformation of at least one deformable portion of the floating system, a sensor of movement of at least one mobile portion of the floating system, an accelerometer placed on at least one mobile portion of the floating system, and a pressure sensor within at least one of a pneumatic or a hydraulic portion of the floating system.

18. The method for predicting the characteristic as claimed in claim 16, wherein the at least one sensor is a sensor chosen from one of a radar, lidar, a sensor of deformation of at least one deformable portion of the floating system, a sensor of movement of at least one mobile portion of the floating system, an accelerometer placed on at least one mobile portion of the floating system, and a pressure sensor within at least one of a pneumatic or a hydraulic portion of the floating system.

19. The method for predicting the characteristic as claimed in claim 15, wherein the first time interval comprises between 0.01 s and 10 min.

20. The method for predicting the characteristic as claimed in claim 15, wherein the second time interval comprises between 10 min and 24 h.

21. The method for predicting the characteristic as claimed in claim 16, wherein the second time interval comprises between 10 min and 24 h.

22. The method for predicting the characteristic as claimed in claim 17, wherein the second time interval comprises between 10 min and 24 h.

23. The method for predicting the characteristic as claimed in claim 18, wherein the second time interval comprises between 10 min and 24 h.

24. The method for predicting the characteristic as claimed in claim 15, wherein the method comprises a prior step of constructing the transfer function.

25. The method for predicting the characteristic as claimed in claim 15, wherein the characteristic comprises filtering the measurements of the at least one sensor.

26. The method for predicting the characteristic as claimed in claim 16, wherein the characteristic comprises filtering the measurements of the at least one sensor.

27. The method for predicting the characteristic as claimed in claim 17, wherein the characteristic comprises filtering measurements of the at least one sensor.

28. The method for predicting the characteristic as claimed in claim 18, wherein the characteristic comprises filtering the measurements of the at least one sensor.

29. The method for predicting the characteristic as claimed in claim 19, wherein the characteristic comprises filtering the measurements of the at least one sensor.

30. The method for predicting the characteristic as claimed in claim 20, wherein the characteristic comprises filtering the measurements of the at least one sensor.

31. The method for predicting the characteristic as claimed in claim 21, wherein the characteristic comprises filtering the measurements of the at least one sensor.

32. The method for predicting the characteristic as claimed in claim 22, wherein the characteristic comprises filtering the measurements of the at least one sensor.

33. The method for predicting the characteristic as claimed in claim 15, comprising determining a degree of confidence from obtaining a characteristic by use of the water swell prediction model.

34. The method for predicting the characteristic as claimed in claim 15, wherein the characteristic is elevation of the water swell at at least one of one point of the water swell as measured by the at least one sensor.

35. The method for predicting the characteristic as claimed in claim 15, wherein the floating system has sensors and the variation in the water swell is measured by each sensor.

36. The method for predicting the characteristic as claimed in claim 35, wherein, for a future time period, a future value of a signal produced by each sensor is determined by accounting for measurements from each sensor.

37. The method for predicting the characteristic as claimed in claim 35, wherein, for a future period, a future value of a signal provided by each sensor is determined by accounting for measurements from all of the sensors.

38. The method for predicting a characteristic as claimed in claim 15, wherein the prediction model is determined by using a spectrum of the water swell.

39. A method for controlling a wave-energy converter, which converts energy from a water swell into electrical, pneumatic or hydraulic energy, wherein the obtained characteristic resulting from an effect of the water swell on the wave-energy converter is predicted by using the method of claim 15, with the wave-energy converter being controlled depending on the obtained characteristic resulting from the swell.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

[0035] The patent or application file contains at least one drawing executed in color. Copies of this patent or patent application publication with color drawing(s) will be provided by the Office upon request and payment of the necessary fee.

[0036] FIG. 1 illustrates the steps of the method for predicting a swell characteristic according to one embodiment of the invention.

[0037] FIG. 2 illustrates the steps of the method for predicting a swell characteristic according to a second embodiment of the invention.

[0038] FIG. 3 illustrates a wave-energy converter able to implement the method according to one embodiment of the invention.

[0039] FIG. 4 illustrates a vessel able to implement the method according to one embodiment of the invention.

[0040] FIG. 5 illustrates the correlation between measured and predicted values as a function of the prediction horizon, for a swell spectrum with a peak period of 2 s, the prediction being obtained using a method according to the prior art, and using two variant embodiments of the method according to the invention, respectively.

[0041] FIG. 6 illustrates the correlation between measured and predicted values as a function of the prediction horizon, for a swell spectrum with a peak period of 3 s, the prediction being obtained using a method according to the prior art, and using two variant embodiments of the method according to the invention, respectively.

[0042] FIG. 7 illustrates the correlation between measured and predicted values as a function of the prediction horizon, for a swell spectrum with a peak period of 4 s, the prediction being obtained using a method according to the prior art, and using two variant embodiments of the method according to the invention, respectively.

[0043] FIG. 8 illustrates the correlation between measured and predicted values as a function of the prediction horizon, for a swell spectrum with a peak period of 5 s, the prediction being obtained using a method according to the prior art, and using two variant embodiments of the method according to the invention, respectively.

DETAILED DESCRIPTION OF THE INVENTION

[0044] The present invention relates to a method for predicting a characteristic resulting from the effect of the sea’s swell (wave) on a floating system subjected to the movement of the swell (wave). The predicted resulting characteristic may in particular be the force exerted by the swell (wave) on the floating system, the elevation of the swell (wave) in the location of the floating system or in a location close to the floating system, the movement of the floating system, the value of a signal of a sensor (i.e. the measurement of a sensor), or any analogous characteristic, etc.

[0045] The floating system may be a wave-energy converter (of any envisionable form), a floating platform (for example a platform used in the petroleum industry), or an (offshore) floating wind turbine, a vessel, an amphibious vehicle, a seaplane, or any analogous floating system. In the remainder of the description, the predicting method is described, non-limitingly, in the context of a wave-energy converter. This wave-energy converter converts the energy of the swell (wave) into electrical, pneumatic or hydraulic energy. According to one design, the wave-energy converter may comprise a mobile device connected to an electrical, pneumatic or hydraulic machine, which harvests energy and controls the wave-energy converter. However, all the described embodiments are suitable for any floating or oscillating system.

[0046] The method according to the invention is a predictive method, because it allows the swell-resulting characteristic for a future horizon, to be determined. This future horizon may advantageously be comprised between 1 s and 5 min.

[0047] In the remainder of the description the terms waves, tides and swell are considered to be equivalent.

[0048] According to the invention, the floating system is equipped with at least one sensor that measures the variation in the swell or a characteristic resulting from the swell. In other words, the sensor is able to measure a parameter related to the swell, for example the elevation of the surface of the sea, a force related to the swell, the movement of the floating system, etc. Such a sensor may in particular be chosen from: [0049] A radar, which provides an image of the elevation of the surface of the sea in a region surrounding the floating system. Such a sensor may in particular be used when the floating system is a vessel or a floating platform; [0050] A sensor of deformation of at least one deformable portion of the floating system. For example, it may be a question of a sensor placed on a deformable wall of a flexible wave-energy converter in which the conversion of energy is distributed along the wall (and carried out through the very deformation of the wall) or localized at a specific location in the wave-energy converter (and carried out via a power take-off); [0051] A LiDAR sensor (LiDAR or lidar standing for light detection and ranging) capable of providing an image of the elevation of the surface of the sea in a region surrounding the floating system; [0052] An optical stereo-camera sensor, capable of providing an image of the elevation of the surface of the sea in a region surrounding the floating system; [0053] A sensor of movement of a mobile portion of the floating system, for example, in the case of a wave-energy converter with a mobile portion that oscillates with the swell; [0054] A sensor of pressure within at least one mobile portion of the floating system. It may for example be a question of a sensor of pressure within a pneumatic energy-converting system of a wave-energy converter; [0055] An acceleration sensor, capable of detecting one or more modifications in at least one of the three-dimensional inclination and elevation of the floating system, this elevation or inclination being due to the swell, and [0056] Wireless communication capable of communication with buoys equipped with swell-elevation sensors, etc.

[0057] FIG. 3 illustrates, schematically and non-limitingly, a wave-energy converter for implementation of the method for predicting a swell characteristic according to one embodiment of the invention. The wave-energy converter 1 is located below the sea’s surface 2 or on the sea’s surface (embodiment not shown). The wave-energy converter 1 is composed of flexible rings 3 hinged to each other, which form a flexible tube. The swell-related deformations generate energy by use of electroactive polymers or piezoelectric materials. Furthermore, the wave-energy converter 1 comprises deformation sensors 4. Non-limitingly, a deformation sensor 4 may measure the electrical signals generated by the electroactive polymers of the rings 3.

[0058] FIG. 4 illustrates, schematically and non-limitingly, a vessel 5 for implementing the method for predicting a swell characteristic according to one embodiment of the invention. The vessel 5 is equipped with a radar 4 for measuring the elevation of the sea in a region surrounding the vessel 5. In addition, the vessel 5 is equipped with wireless communications (not shown) which communicate with the buoy 7, that is equipped with a swell-elevation sensor. The method according to the invention allows use of lifting devices which are cranes 8 and 9 provided on the vessel 5 to be optimized, for example by accounting for future moments of cessation of the swell or by adapting the lift lengths to the swell predicted in real time.

[0059] Moreover, the method according to the invention implements a transfer function that relates the incident swell to measured and predicted swell-resulting characteristics. If the floating system comprises multiple of sensors, then the method according to the invention employs one transfer function per sensor. The transfer function expresses the relationship between the input incident swell, and the output sensor measurement. In other words, the transfer function may be considered to be a dynamic model that relates the input incident swell to the output sensor measurement. According to one embodiment of the invention, the transfer function may be known initially. Alternatively, the transfer function may be obtained in a prior step of identifying the model of the floating system and of the sensor.

[0060] According to one embodiment of the invention, for a rigid (non-deformable) floating system, for example for the vessel of FIG. 4, the transfer function may depend only on the frequency and on the angular direction of the incident swell. Such a transfer function H may be written: H(ω, θ) with ω frequency, and θ angular direction.

[0061] As a variant, for a flexible (deformable) floating system, for example for the wave-energy converter illustrated in FIG. 3, the transfer function may further depend on additional real coordinates. Such a transfer function H may be written H(χ, ω, θ) with χ being additional real coordinates, ω being frequency, and θ being angular direction. For example, for the wave-energy converter illustrated in FIG. 3, χ designates position along the wave-energy converter.

[0062] According to the invention, the method for predicting the swell-resulting characteristic comprises the following steps: [0063] measuring in real time [0064] updating a spectral model of the swell [0065] determining a swell prediction model [0066] predicting the swell-resulting characteristic.

[0067] These steps will be described in detail in the remainder of the description. The steps of updating a spectral model of the swell, of determining a prediction model, and of predicting may be implemented by computer, and in particular by a calculator or computer provided on the floating system, or in wireless communication with the floating system. The steps of measuring in real time and of predicting the swell-resulting characteristic are implemented in real time with a relatively short first time interval (i.e. at a relatively high first frequency), the time interval typically being from 0.01 s to 1 min. The steps of updating the spectral model of the swell and of determining a swell prediction model are implemented with a second time interval (i.e. at a second frequency), the second time interval is longer than the first time interval (the second frequency is lower than the first frequency), and typically from 10 min to 24 h, and preferably from 10 min to 6 h.

[0068] The method according to the invention implements what is referred to as an SPB approach (SPB standing for spectrum-based predictor), in this case the spectrum being the spectrum of the swell. Such an approach is based on the assumption that the physical variables form a stationary Gaussian process, this being a standard assumption in oceanography and marine engineering. Under this assumption, it is possible to determine a statistically optimal predictor, using the spectral model of the swell, and using transfer functions characterizing the observed and predicted variables.

[0069] In practice, it is possible to implement the steps with two different time intervals, since the swell spectrum, and therefore the optimal predictor, may be considered stationary to a horizon of a few tens of minutes. Thus, it is not necessary to update the spectral model of the swell in real time at a high frequency. By virtue of the implementation of the steps on different time scales, it is possible to limit the number of computations to be performed with the first time interval which allows the swell characteristic to be predicted with a low computation time, compatible with the first time interval.

[0070] Preferably, the first time interval may be comprised between 0.01 s and 10 min. This time interval may for example be equal to 1 s. These values allow swell to be predicted in real time.

[0071] Preferably, the second time interval may be comprised between 10 min and 24 h, and preferably between 10 min and 6 h. This time interval may for example be equal to 1 h. These values make it possible to limit the frequency at which the spectral model is updated while keeping the spectral model satisfactorily representative of the swell, and therefore the reliability of prediction of the swell-resulting characteristic satisfactory.

[0072] FIG. 1 illustrates, schematically and non-limitingly, the steps of the predicting method according to one embodiment of the invention. The predicting method comprises a step MES of measuring in real time with a time interval T1 of the variation in the swell by use of at least one sensor. In this method, a spectral model MSH of the swell is further employed, which is updated MAJ with a time interval T2, on the basis of data DON, which may be meteorological data or data measured by at least one sensor. The time interval T2 is longer than the time interval T1. The method also employs a transfer function FT that relates the incident swell to the measurements of the one or more sensors in question, with a view to determining, with a time interval T2, a swell prediction model MPR. Next, this swell prediction model MPR is applied to the measurements MES with a time interval T1 in order to deduce therefrom the prediction Pred of the swell-resulting characteristic for a future period (for a future horizon).

[0073] According to one embodiment of the invention, the method for predicting a swell-resulting characteristic may further comprise a step of filtering the measurements. Filtering makes it possible in particular to reduce noise, to reduce measurement disparities between any sensors, etc. It may in particular be a question of FFT filtering (FFT standing for fast Fourier transform), spatial filtering of the data using, for example, polynomial functions such as Chebyshev polynomials, or any analogous filter.

[0074] In accordance with one implementation of the invention, the method for predicting a swell-resulting characteristic may further comprise a step for determining a degree of confidence in the prediction. This degree of confidence may be determined by use of the swell prediction model, with time interval T2. This step makes it possible to characterize the mean squared error for each predicted variable and each prediction horizon.

[0075] FIG. 2 illustrates, schematically and non-limitingly, the steps of the predicting method according to a second embodiment of the invention. The second embodiment of the invention comprises the two optional steps (filtering, degrees of confidence) described above. These two steps are independent. As a variant, the method according to the invention may comprise only one of these steps. The predicting method comprises a step MES of measuring in real time with a time interval T1 the variation in the swell by at least one sensor. This measuring step MES is followed by a step FIL of filtering the measurements. In this method, a spectral model MSH of the swell is further employed, which is updated MAJ with a time interval T2, on the basis of data DON, which may be meteorological data or data measured by at least one sensor. The time interval T2 is longer than the time interval T1. The method also employs a transfer function FT that relates the incident swell to the measurements of the one or more sensors in question, with a view to determining, in real time and with a time interval T2, a swell prediction model MPR and a degree of confidence ddc in the prediction of the swell-resulting characteristic. Next, this swell prediction model MPR is applied to the measurements MES with a time interval T1, in order to deduce therefrom the prediction Pred of the swell-resulting characteristic for a future period (for a future horizon).

1) Measuring in Real Time

[0076] In this step, the variation in the swell is measured in real time with a first time interval of at least one sensor. Thus, at least one variation in the swell is obtained in real time with a measurement frequency that is high (with respect to the frequency of update of the spectral model of the swell).

[0077] According to one embodiment of the invention, the measurements may be stored in memory, in particular in the computer, and for example in a memory of a computer or calculator. Thus, the prediction may take into account past measurements allowing a more accurate prediction of the swell-resulting characteristic.

2) Filtering the Measurement

[0078] In this optional step, the measurements are filtered. Filtering makes it possible in particular to reduce noise and to reduce any measurement disparities between sensors, etc. It may in particular be a question of FFT filtering (FFT standing for fast Fourier transform), spatial filtering of the data using, for example, polynomial functions such as Chebyshev polynomials, or any analogous filter.

[0079] This step may also be implemented by computing a computer or calculator.

3) Updating a Spectral Model of the Swell

[0080] In this step, a spectral model of the swell is updated with a second time interval (longer than the first time interval). The spectral model of the swell is updated on the basis of at least one of the meteorological models and on the basis of at least one measurement of at least one sensor, in particular a sensor used for step 1). Thus, spectral model of the swell makes possible accounting for the variability in the state of the sea.

[0081] The spectral model of the swell is a power spectral density (PSD). By definition, the power spectral density is the square of the modulus of the Fourier transform divided by the width of the spectral band, itself equal to the inverse of the integration time. This spectral model characterizes the properties of swell as a random process. It is not a question of modes of oscillation as in a harmonic decomposition (which represents a deterministic system with a finite number of oscillatory modes), nor of an identification of a dominant frequency. The document: Athanasios Papoulis and S. Unnikrishna Pillai, Probability, Random Variables, and Stochastic Processes, Tata McGraw-Hill Education, 2002, is about random processes and their spectral representation.

[0082] This stochastic representation of the swell makes possible combining measured signals, by considering them to conjointly form a multi-dimensional random process. It is in this way that the method according to the invention makes it possible to process and combine at least two time series or at least two measurement points (for example the many observation points of a radar, or indeed of a plurality of different sensors, etc.) instead of a single time series, as is sometimes the case in the prior art.

[0083] The spectral model of the swell assumes that the swell is a zero-mean Gaussian process. Such an assumption is justified under most measurement conditions, excepting major storms, or in very shallow water. Therefore, application of such an assumption makes the prediction of the swell-resulting characteristic reliable. In addition, the wave field is considered, during the second time interval, to be stationary in which the statistical properties of the wave field are considered to not largely vary. Given the Gaussian-process assumption, the stationarity assumption may be reduced to the stationarity of the mean and of the auto-covariance function. Furthermore, the wave field may be considered to be uniform in the studied region (“in space” in light of the stationarity assumption). Thus, the covariance of the elevation of the free surface of the sea measured at two different positions depends only on the relative position of these two points. The swell field may be considered uniform over distances of a few kilometers. Under these assumptions, the sea’s state is fully characterized by the elevation spectrum of the free surface η which spectrum depends on frequency and on angular direction in the horizontal plane, and may therefore be written: S.sub.ηη(ω,θ) where ω is frequency, and θ is angular direction. When the spectral model of the swell is updated, it is S.sub.ηη that is updated. S.sub.ηη may be provided directly by weather-forecasting agencies such as the ECMWF (https://www.ecmwf.int/en/forecasts), or estimated from local measurements made by at least one sensor, for example a sensor used for step 1.

4) Determining a Swell Prediction Model

[0084] In this step, a swell prediction model is determined by the transfer function and of the updated spectral model of the swell. The prediction model relates the measurements generated by one or more sensors to a prediction of a swell-resulting characteristic. Because the spectral model of the swell is updated with a second time interval, the swell prediction model is determined with the second time interval. In other words, the swell prediction model remains valid for the duration of the second time interval.

[0085] The swell prediction model may depend on one or more sensors. With fewer sensors, the swell prediction model is simpler and requires less complex computations which facilitates its real-time implementation. With more sensors, the prediction may be more accurate.

[0086] Thus, the invention exploits the mathematical relationship between, on the one hand, the power spectral density, and, on the other hand, the statistical correlation between the various measured quantities, between the various predicted quantities, and between the measured and predicted quantities, according to the following schema: DSP — correlations — predictor.

[0087] According to one non-limiting exemplary embodiment, this step may be implemented by the operations described below.

[0088] First, the spectrum and cross-spectrum of all the observed and predicted variables are determined. Let y.sub.1 be the signal of a sensor used for the prediction or the swell-resulting characteristic that it is being sought to predict. Let y.sub.2 be the signal of a sensor used for the prediction, or the swell-resulting characteristic that it is being sought to predict. Under the stationary-Gaussian-swell assumptions, y.sub.1 and y.sub.2 are random zero-mean stationary Gaussian processes. The cross-spectrum of y.sub.1 and y.sub.2 is computed which is based on the spectral model of the swell and on the transfer functions H.sub.ηy1 and H.sub.ηy2 as relates the free-surface elevation to the variables y.sub.1 and y.sub.2, respectively, as follows:

[00001]${\text{S}}_{{\text{y}}_1{y}_2}\left(\omega \right)={\displaystyle {\int }_{\theta =0}^{2\pi }{H}_{\eta y_1}\left(\omega ,\theta \right)S_{\eta \eta }\left(\omega ,\theta \right)H_{\eta y_2}^{*}\left(\omega ,\theta \right)d\theta }$

with η being the free-surface elevation, S.sub.y1y2(ω) being the cross-spectrum of signals y.sub.1 and y.sub.2 (y.sub.1 and y.sub.2 possibly designate the same signal), S.sub.ηη being the spectrum of the swell, ω being frequency, θ being angular direction, H being the transfer function, and H* being its conjugate. This calculation is performed for each possible signal pair, that is each possible pair among the sensors used for the prediction and the swell-resulting characteristic.

[0089] Next, the Wiener-Khinchin theorem may be applied to determine the covariance function of each pair of signals y.sub.1 and y.sub.2:

[00002]$\begin{array}{l}\left\{\begin{array}{c}{\text{S}}_{\text{zz}}\left(\omega \right)={\displaystyle {\int }_{-\infty }^{\infty }{\text{r}}_{\text{zz}}\left(\tau \right)e^{-i\omega t}d\tau }\\ {\text{r}}_{\text{zz}}\left(\omega \right)=\frac{1}{2\pi }{\displaystyle {\int }_{-\infty }^{\infty }{\text{S}}_{\text{zz}}\left(\omega \right)e^{i\omega t}d\omega }\end{array}\right)\\ r_{y_1{y}_2}\left(\tau \right)=\frac{1}{2\pi }{\displaystyle \underset{-\infty }{\overset{\infty }{\int }}{S}_{y_1{y}_2}\left(\omega \right)e^{i\omega t}d\omega }\end{array}$

with τ being one time interval, S.sub.y1y2(ω) being the cross-spectrum, r.sub.y1y.sub.2 being the variance-covariance function, and z being a signal (z may be y.sub.1 or y.sub.2).

[0090] In the remainder of the description, the index o refers to observed signals and the index p refers to predicted signals.

[0091] By virtue of this step, matrices r.sub.oo(τ), r.sub.op(τ) and r.sub.pp(τ) may be constructed: element (i,j) of the matrix r.sub.oo(τ) is the covariance function of the i-th and j-th observed signals; element (i,j) of the matrix r.sub.op(τ) is the covariance function of the i-th observed signal and the j-th predicted signal (swell-resulting characteristic); lastly, element (i,j) of the matrix r.sub.pp(τ) is the covariance function of the i-th and j-th predicted signals (swell-resulting characteristics).

[0092] The observed (measured) values form a random vector Zo(t), and the values that it is desired to predict form a random vector Zp(t). These two vectors are conjointly zero-mean Gaussian and entirely characterized by their covariance matrices, which are denoted .Math.oo, .Math.pp and

[00003]${\text{Σ}}_{op}={\text{Σ}}_{po}^T.$

Let

[00004]${\tau }_o={\left({\underset{¯}{\tau }}_o{\underset{¯}{\tau }}_o+\text{Δ}t.Math.{\overline{\tau }}_o\right)}^{\top }\in {ℝ}^{N_o}\text{and}{\tau }_p={\left({\underset{¯}{\tau }}_p{\underset{¯}{\tau }}_p+\text{Δ}t.Math.{\overline{\tau }}_p\right)}^{\top }\in {ℝ}^{N_p}$

be the set of observation and prediction times with respect to the present time t. A vector

[00005]$Z_o\left(t\right)=\left(\begin{array}{c}{z}_o\left(t-{\underset{¯}{\tau }}_o\right)\\ .Math.\\ z_o\left(t-{\overline{\tau }}_o\right)\end{array}\right)$

is defined that contains the data observed (measured) at all the past time increments that it is desired to take into account for the prediction, and a vector

[00006]$Z_p\left(t\right)=\left(\begin{array}{c}{z}_p\left(t-{\underset{¯}{\tau }}_p\right)\\ .Math.\\ z_p\left(t-{\overline{\tau }}_p\right)\end{array}\right)$

is defined that contains the data that it is sought to predict, to all the prediction horizons. Concretely, the prediction model must determine how to best compute Z.sub.p(t) on the basis of Z.sub.o(t).

[0093] Because stationarity is assumed, the covariance matrices corresponding to Zo(t) and Zp(t) do not depend on t. In addition, they are block Toeplitz, and may be structured in the following way:

[00007]${\text{Σ}}_{oo}=\left(\begin{array}{ccc}{\text{A}}_{1,1}& .Math.& {\text{A}}_{1,N_o}\\ .Math.& \ddots & .Math.\\ {\text{A}}_{N_{o,1}}& .Math.& {\text{A}}_{N_o,N_o}\end{array}\right)\in {ℝ}^{M_o{N}_o\times M_o{N}_o}$

[00008]${\text{Σ}}_{pp}=\left(\begin{array}{ccc}{\text{B}}_{1,1}& .Math.& {\text{B}}_{1,N_p}\\ .Math.& \ddots & .Math.\\ {\text{B}}_{N_{p,1}}& .Math.& {\text{B}}_{N_p,N_p}\end{array}\right)\in {ℝ}^{M_p{N}_p\times M_p{N}_p}$

[00009]${\text{Σ}}_{op}=\left(\begin{array}{ccc}{\text{C}}_{1,1}& .Math.& {\text{C}}_{1,N_p}\\ .Math.& \ddots & .Math.\\ {\text{C}}_{N_{o,1}}& .Math.& {\text{C}}_{N_o,N_p}\end{array}\right)\in {ℝ}^{M_o{N}_o\times M_p{N}_p}$

[00010]${\text{Σ}}_{po}={\text{Σ}}_{op}^{\top }$

with blocks A, B, C defined by:

[00011]$\left\{\begin{array}{l}{\text{A}}_{i,j}={\text{r}}_{oo}\left({\tau }_o\left(j\right)-{\tau }_o\left(i\right)\right),\forall \left(i,j\right)\in {\left[1\mathrm{..}{N}_o\right]}^2\hfill \\ {\text{B}}_{i,j}={\text{r}}_{pp}\left({\tau }_p\left(j\right)-{\tau }_p\left(i\right)\right),\forall \left(i,j\right)\in {\left[1\mathrm{..}{N}_p\right]}^2\hfill \\ {\text{C}}_{i,j}={\text{r}}_{op}\left(\tau {}_p\left(j\right)-{\tau }_o\left(i\right)\right),\forall \left(i,j\right)\in \left[1\mathrm{..}{N}_o\right]\times \left[1\mathrm{..}{N}_p\right]\hfill \end{array}\right)$

[0094] Because stationary Gaussian signals have been assumed, the best predictor relating Z.sub.p(t) at Z.sub.o(t) is a linear operation on the components of Z.sub.o(t), as detailed in step 6. This linear operation is performed by virtue of a matrix P, which may be obtained via the equation:

[00012]$\text{P}={\text{Σ}}_{po}{\text{Σ}}_{oo}^{†}$

[0095] In this equation, the “dagger” exponent designates the matrix inverse (if the matrix is invertible) or pseudo-inverse (if the matrix is not invertible this indicates that the invention contained in Zo(t) is statistically redundant). It will be noted that all the operations leading to the calculation of the prediction matrix P may be performed with a time interval T2, because they depend only on the spectral model of the swell and on the transfer functions.

[0096] These operations may be adapted to the assumptions in question, and may also be adapted to the transfer function for a flexible floating system, in particular by taking into account additional data.

5) Determining Degree of Confidence

[0097] In this optional step, a degree of confidence in the prediction of the swell-resulting characteristic may be determined. This degree of confidence may be determined by use of the swell prediction model. This step makes possible characterizing error. Because the spectral model of the swell is updated with a second time interval, the degree of confidence is determined with the second time interval. In other words, the degree of confidence remains valid for the duration of the second time interval.

[0098] This step may also be implemented by computing means (computer or calculator).

[0099] According to one non-limiting example of implementation of this step, the degree of .Math..sub.p|o confidence may be determined on the basis of the covariance matrix of the prediction error, which may be computed in the following way:

[00013]${\text{Σ}}_{p\left|o\right)}={\text{Σ}}_{pp}-{\text{Σ}}_{po}{\text{Σ}}_{oo}^{†}{\text{Σ}}_{op}$

In this equation, the “dagger” exponent designates the matrix inverse (if the matrix is invertible) or pseudo-inverse (if the matrix is not invertible this indicates that the invention contained in Z.sub.o(t) is statistically redundant). The matrix .Math..sub.p|o contains the values of the mean squared error for each pair of prediction horizons and each pair of predicted signals (swell-resulting characteristics).

6) Predicting the Swell-Resulting Characteristic

[0100] In this step, the swell-resulting characteristic is predicted, in real time, for a future horizon (for a future period) by applying the swell prediction model determined in step 4) to the measurements carried out in step 1), and optionally filtered in step 2). Thus, for a future horizon, the swell-resulting characteristic is obtained on the basis of a reliable model and of measurements. Therefore, the prediction of the swell-resulting characteristic is reliable.

[0101] This step is implemented with the first time interval (i.e. for each new measurement), the prediction model remaining identical during the second time interval. Thus, the same prediction model is used for a number of predictions.

[0102] For this step, the following equation may be employed:

[00014]${\tilde{Z}}_p\left(t\right)=P{Z}_o\left(t\right)$

with P being the prediction model determined in step 4), Z.sub.o(t) being the measured values, and Z̃̃̃.sub.p(t) being the predicted values of the resulting characteristic.

[0103] Furthermore, the invention relates to a method for controlling a wave-energy converter, which converts wave energy into electric, pneumatic or hydraulic energy. The control method comprises a step of predicting the swell according to one of the variants or combinations of variants described above, with at least the following steps: [0104] 1) Measuring in real time; [0105] 2) Updating a spectral model of the swell; [0106] 3) Determining a swell prediction model; and [0107] 4) Predicting the swell-resulting characteristic.

[0108] The control method according to the invention also comprises a step of controlling the wave-energy converter depending on the characteristic (force, elevation, etc.) of the swell, so as to optimize the harvest of energy. This control may control the mobile device of the wave-energy converter, for example by an electrical, pneumatic or hydraulic power take-off (PTO). This PTO influences the movement of the mobile device and allows mechanical energy to be transferred to the electrical, pneumatic or hydraulic network. Model predictive control (MPC) is one example of a method for controlling wave-energy converters requiring real-time wave prediction. The control method according to the invention may also be applied to a wave-energy converter belonging to the category of wave-energy converters employing oscillating water columns (OWCs), or to any other type of wave-energy converter.

[0109] The control method according to the invention is particularly suitable for a wave-energy converter such as described with reference to FIG. 3. Specifically, such a system is equipped with a high number of sensors. In addition, the method according to the invention allows charging and discharging of electro-active polymers or of piezoelectric materials to be properly synchronized.

[0110] Specifically, the control method according to the invention allows optimal control, because the predicting method according to the invention provides a method for predicting the force, or elevation, that the swell will exert on the mobile device to a future horizon, on the basis of values measured in the past and on the basis of a spectral wave model.

[0111] In addition, the present invention relates to a method for controlling landing or transfer of at least one of device on and from a vessel or a floating platform. In this method, the following steps may be carried out: [0112] a swell-resulting characteristic is predicted for the vessel or floating platform by use of the predicting method according to any one of the variant embodiments described above; [0113] a future time at which the swell-resulting characteristic will vary little is determined on the basis of the prediction of the swell-resulting characteristic; and [0114] landing or transfer is performed at the time determined in the preceding step.

[0115] The present invention also relates to a method for controlling a floating wind turbine, in which method the following steps are carried out: [0116] a swell-resulting characteristic is predicted for the floating wind turbine by use of the predicting method according to any one of the variant embodiments described above; [0117] the action of the swell on the float of the floating wind turbine is deduced therefrom; and [0118] the floating wind turbine is controlled (in particular the angle of inclination of its blades are controlled) so as to reduce the stresses on the structure of the wind turbine, depending on the prediction of the action of the swell on the float.

EXAMPLE

[0119] Features and advantages of the method according to the invention will become more clearly apparent on reading about the following example of application.

[0120] In this example, a wave-energy converter such as illustrated in FIG. 3 was employed. The wave-energy converter was a flexible floating tube of 30 m length and of 1.2 m diameter, which was aligned with the main direction of the swell. The pressure wave created by the passage of the swell caused a radial deformation of the tube, and this deformation was converted into electricity by virtue of rings of electroactive polymers or of piezoelectric materials placed along the tube. The electric signals of the rings of electroactive polymers or piezoelectric materials were recorded and served as a deformation sensor for the purposes of the prediction. In addition, swell probes were placed along the tube in order to measure the elevation of the free surface.

[0121] Four experiments were carried out with different swell spectra, corresponding to four different sea states, with periods ranging from 2 to 5 s.

[0122] The transfer functions of each sensor were constructed on the basis of some of the measurement signals of the sensors.

[0123] Subsequently, a characteristic of the swell was determined for a horizon of 10 seconds: [0124] by use of the prior-art method described in the patent application FR 3042889 (WO 2017/071946). This embodiment is designated by the reference AA, [0125] by use of a first variant embodiment of the invention, in which the swell prediction model depends solely on the sensor in question. This embodiment is designated by the reference INV1, [0126] by use of a second variant embodiment of the invention, in which the swell prediction model depends on the measurements of all the sensors. This embodiment is designated by the reference INV2.

[0127] FIGS. 5 to 8 illustrate the correlation curves C of these three embodiments, as a function of the future time horizon T in s. The correlation expresses the reliability of the measurement: the closer the correlation is to 1, the better the prediction. FIGS. 5 to 8 regard the prediction for a sensor located towards the downstream end of the tube in the direction of propagation of the swell. In these figures, theoretical predictions have been represented by solid lines, and experimental predictions have been represented by broken lines. FIG. 5 corresponds to swell with a period of 2 s; FIG. 6 corresponds to swell with a period of 3 s; FIG. 7 corresponds to swell with a period of 4 s and FIG. 8 corresponds to swell with a period of 5 s.

[0128] It will be noted that, in these figures, the method according to the invention (INV1 and INV2) allows better correlation than the prior-art method. The second variant embodiment of the invention (INV2) has a better correlation than the first variant embodiment (INV1).

[0129] The results are similar for other sensors placed in other positions on the tube. Therefore, the method according to the invention allows an accurate prediction of a characteristic of the swell.