WAKE MONITORING, WAKE MANAGEMENT AND SENSORY ARRANGEMENTS TO SUCH

Abstract

Disclosed is a method of establishing a wake management of a wind farm. The method comprises acts of monitoring one or more wake conditions using one or more sensors from one or more wind turbine generators (WTGs); and establishing a wake management of the wind farm as a function of the wake conditions. Disclosed is also a method of optimising operation of a wind turbine park based on wake management and a system for generating wake management.

Claims

1. A method of establishing a wake management of a wind farm, the method including the following steps: monitoring one or more wake conditions using sensory data from one or more sensors; processing the sensory data to identify the respective one or more wake conditions; and establishing a wake management of the wind farm as a function of the identified one or more wake conditions; the one or more sensors are accelerometers arranged as vibration sensors placed in one or more of the blades of a rotor from one or more wind turbine generators of the wind farm and using machine learning to process the sensory data to identify the wake conditions.

2. The method according to claim 1, wherein the step of processing sensory data includes identifying aerodynamic conditions such as turbulence, rain, or hail, as wake conditions from the sensory data.

3. The method according to claim 1, wherein the step of monitoring further includes monitoring power output based on rotor sensory data.

4. The method according to claim 1, wherein the step of monitoring further includes monitoring acoustic sensory data from an acoustic sensor.

5. The method according to claim 1, wherein the step of processing the sensory data is based on rotary sensory data provided by high frequency sampling.

6. The method according to claim 1, wherein the step of processing the sensory data is based on timestamped and synchronized sensory data.

7. The method according to claim 1, wherein the step of monitoring is performed by further using a temporally actual wake conditions; and wherein the step of processing is performed by further calibrating processed sensory data against the temporally actual wake conditions.

8. The method according to claim 1, wherein the accelerometers are three-axis accelerometers.

9. A method of optimizing operation of a wind farm with multiple wind turbine generators, including the following steps: monitoring one or more wake conditions using sensory data from one or more sensors; processing the sensory data to identify one or more wake conditions; establishing wake management of the wind farm as a function of the identified one or more wake conditions; determining individual wind turbine generator control settings as an optimized power production function of the wake management and individual wind turbine generator parameters; and operating one or more wind turbine generators in the wind farm based on the individual wind turbine generator control setting.

10. The method according to claim 9, wherein the step of determining individual wind turbine generator control settings includes minimizing the total wake in the wake management as a function of the wake conditions.

11. The method according to claim 9, wherein the step of operating one or more wind turbine generators includes one or more of the following: pitching, yawing, regulating rotational speed of the one or more blades of a rotor.

12. A system for generating a wake management comprising: one or more sensors arranged on respective one or more wind turbine generators, wherein the one or more sensors are accelerometers arranged as vibration sensors placed in one or more of the blades of the respective one or more wind turbine generators; means adapted to execute the acts of the method according to claim 1.

13. A controller system for optimizing operation of a wind farm with multiple wind turbine generators, the controller system configured and arranged to: receive sensory data from accelerometers arranged as vibration sensors placed in one or more of blades of the multiple wind turbine generators; establishing a wake management of the wind farm as a function of the wake conditions; determining individual wind turbine generator control settings as an optimized power production function of the wake management and individual wind turbine generator parameters; and operating one or more wind turbine generators in the wind farm based on the individual wind turbine generator control settings.

14. The controller system according to claim 13, wherein the controller system is further configured and arranged to establish a wake field map from the sensory data, and based on the wake field map a management strategy is defined and applied to the wind farm.

15. The controller system according to claim 13, wherein the controller system is further configured and arranged to determine GPS coordinates of the multiple wind turbine generators.

16. The controller system of claim 13, wherein the controller system is further configured and arranged to measure nacelle direction.

17. The controller system of claim 13, wherein the controller system is further configured and arranged to measure generator power output.

18. The method of claim 7, wherein the temporally actual wake conditions are obtained by LiDAR measurements.

Description

BRIEF DESCRIPTION OF DRAWINGS

[0084] Embodiments of the invention will be described in the figures, whereon:

[0085] FIG. 1 illustrates a method of establishing wake management;

[0086] FIG. 2 illustrates further aspect of identifying wake conditions;

[0087] FIG. 3 illustrates a method of operating a wind turbine generator using wake management;

[0088] FIG. 4 illustrates a wind turbine generator;

[0089] FIG. 5 illustrates a sensory arrangement inside blades;

[0090] FIG. 6 illustrates further aspects of a sensory arrangement;

[0091] FIG. 7 illustrates further optional or alternative aspects of sensory arrangement on a wind turbine generator, including a sensor node arrangement;

[0092] FIG. 8 illustrates a wind turbine generator with a sensory arrangement in interaction with a remote-/cloud-based processor;

[0093] FIG. 9 illustrates a configuration of a sensor node;

[0094] FIG. 10 illustrates a wind farm without wake management;

[0095] FIG. 11 illustrates a wind farm with wake management;

[0096] FIG. 12 illustrates a process of wake management;

[0097] FIG. 13 illustrates calculation of calculating turbulence intensity for a wake map; and

[0098] FIG. 14 illustrates a process of wake management.

DETAILED DESCRIPTION OF THE INVENTION

[0099]

TABLE-US-00001 Rotary device 10 Wind Turbine Generator (WTG) 12 Tower 13 Rotor 14 Rotor sector 18 Nacelle 19 Set of rotor blades 20 Rotor blade/blade 22 Generator 28 Dataset 30 Data 31 Timestamped data 32 Timestamp 34 Set of blade sensors 40 Sensor means 41 Blade sensor 42 Sensor node 45 Node Power 46 Vibration sensor/acceleration sensor 50 Acoustic sensor 60 Computational means/Processor 72 Communication 74 Storage 76 Wind farm 80 System for operating a wind turbine generator 90 Controller system 92 Monitoring 100 Identifying 110 Processing 120 Wake conditions 130 Wind direction 132 Sensory input 140 Establishing wake management 200 Wake management 230 Synchronizing 240 Determining 300 Control setting 350 Operating 400 Rotational speed/Power production 610 Establishing wake management 1000 Optimizing operation of wind farm 2000 Machine learning 3000 Time series data 3080 Labelled Time series data/Labelled data 3085 Unlabelled Time series data/Unlabelled data 3086 Supervised machine learning 3010 Supervised machine learning model (SML) 3015 Un-supervised machine learning 3020 Un-supervised machine learning model (USML) 3025 Training 3100 Grouping 3110 Associating 3120 Verifying 3200 Comparing 3220

[0100] FIG. 1 illustrates a method of establishing 1000 a wake management 230 of a wind farm 80, see FIGS. 10 to 14. The method 1000 comprises an act of: monitoring 100 one or more wake conditions 130, see FIGS. 12 and 13, using one or more sensors 40 from one or more wind turbine generators 12 (WTGs), see FIGS. 4 and 6.

[0101] There is an act of establishing 200 a wake management 230 of the wind farm 80 as a function of the wake conditions 130.

[0102] The act of monitoring is performed based on sensory input 140 from the wind turbine generator 12. The sensory input 140 may comprise data from sensor nodes placed in the wind turbine blades.

[0103] FIG. 2 illustrates further aspect of identifying wake conditions. It is noted that the act of monitoring 100 may be performed independently of the act of establishing 200 a wake management 230.

[0104] The act of monitoring 100 may be performed by identifying 110 the respective one or more wake conditions 130, see FIGS. 10 and 13, by an act of processing 120 sensory data 31 provided from the respective one or more wind turbine generators 12 (WTGs).

[0105] The act of processing 120 sensory data 31 involves identifying aerodynamic conditions as wake conditions 130 in the sensory data 31; see FIG. 13 as one example.

[0106] The act of monitoring 100 may be based on rotor sensory data 42, see FIGS. 5 to 7.

[0107] In particular, the act of monitoring may be performed based on one or more vibration sensors 50, see FIG. 7, placed in or on one or more of the blades 22 of the rotor 14, see FIG. 4.

[0108] Acts of monitoring 100, including acts of identifying 110 or processing 120, may be performed using machine learning, ML, and/or artificial intelligence, AI.

[0109] FIG. 3 illustrates a method of operating a wind turbine generator 12 using wake management 230.

[0110] Illustrated is a method of optimizing 2000 operation of a wind farm 80 with multiple wind turbine generators 12 (WTGs), see FIGS. 10, 11. The method comprises the following acts:

[0111] There is an act of monitoring 100, say using one or more sensors 40, one or more wake conditions 130 from one or more wind turbine generators 12 (WTGs).

[0112] There is an act of establishing 200 a wake management 230 of the wind farm 80 as a function of the wake conditions 130.

[0113] There is an act of determining 300 individual wind turbine generator 12 (WTG) control settings 350 as an optimized power production function of the wake management 230 and individual wind turbine generator 12 (WTG) parameters, see FIGS. 10, 11, 12, and 14.

[0114] There is an act of operating 400 one or more wind turbine generators 12 (WTGs) in the wind farm 80 based on the individual wind turbine generator (12) (WTG) control setting 350.

[0115] The act of monitoring 100 may be performed as outlined in FIG. 2.

[0116] In one aspect of operating or optimizing 2000, the act of determining 300 individual wind turbine generator 12 (WTG) control settings 350 is performed by minimizing the total wake in the wake management 230 as a function of the wake conditions 130.

[0117] In one aspect the act of operating 400, one or more wind turbine generators 12 (WTGs) involve an act of pitching, yawing, regulating rotational speed, or combinations thereof; as is indicated in FIG. 11.

[0118] FIG. 4 illustrates a wind turbine generator 12 (WTG) comprising a tower 13 supporting a nacelle 19 with blades 22 rotably connected and forming a rotor 14.

[0119] FIG. 5 illustrates a sensory arrangement on blades 22A, 22B, 22C on a wind turbine generator 12 with a rotor 14. The blades 22A, 22B, 22C are a set of rotor blades 20. Each blade 22A, 22B, 22C comprises a set of blade sensors 40A, 40B, 40C as sensor means 41. In the present case, each set of blade sensors 40A, 40B, 40C comprises a blade sensor 42A, 42B, 42C. The blade sensor 42 may be a vibration 50 sensor or an acoustic sensor 60.

[0120] The sensory arrangement may be part of a system for detecting turbulent intensity to form input for a wake field map.

[0121] A blade sensor 42 is configured to be in communication 74 with a controller or computational means 72. The communication 74 may be wired or wireless as illustrated here.

[0122] The wind turbine generator 12 is disclosed with a set of blade sensors 40A, 40B, 40C on each blade 22A, 22B, 22C. However, each blade 22A, 22B, 22C experience the same conditions since the blades 22A, 22B, 22C move in a single common plane. Thus, the invention can be obtained by a wind turbine generator 12 having one blade 22 of the set of blades 20 with a set of blade sensors 40. The set of blade sensors 40 may be one, two or more blade sensors 42A, . . . , 42N.

[0123] FIG. 6 illustrates further aspects of sensory arrangements on a wind turbine generator 12. There may be a timer or clock configured to provide data 31 from the sensory means 41 with a timestamp 34, which thus provides timestamped data 32 or synchronized data for processing by the computational unit 72.

[0124] The wind turbine generator (WTG) 12 is with a rotor 14 and a set of rotor blades 20. The set of rotor blades 20 is with three rotor blades 22A, 22B, 22C.

[0125] Each blade 22A, 22B, 22C comprises a set of blade sensors 40A, 40B, 40C. In the present case each set of blade sensors 40A, 40B, 40C comprises a blade sensor 42A, 42B, 42C.

[0126] A further sensor means 41 is shown at the generator 28. In this embodiment, the further sensor is a rotary sensor (RPM-sensor or vibration sensor), such as a high sampling speed sensor measuring the rotational speed 610. The system may be configured for an act of synchronizing 240, as shown in FIG. 2 or variations thereof, and based on sensors 42ABC, and synchronization is performed against at least one other sensor 41 using the timestamp 34 to obtain timestamped data 32.

[0127] The sensor means 41 at the generator 28 may be a Rogowski coil arranged for precision measurements fluctuations in the generator output.

[0128] The sensory arrangement may be part of a system for operating a wind turbine generator. The computational means 72 or controller may be a single unit or distributed as illustrated here.

[0129] The wind turbine generator 12 is disclosed with a set of blade sensors 40A, 40B, 40C on each blade 22A, 22B, 22C. However, each blade 22A, 22B, 22C experiences the same conditions since the blades 22A, 22B, 22C move in a single common plane. Thus, the invention can be obtained by a wind turbine generator 12 having one blade 22 of the set of blades 20 with a set of blade sensors 40. The set of blade sensors 40 may be one, two or more blade sensors 42A, . . . , 42N.

[0130] FIG. 7 illustrates further optional or alternative aspects of a sensory arrangement on a wind turbine generator 12.

[0131] The sensors 42 may be implanted as a sensor node 45 (see FIG. 9) as illustrated by sensor nodes 45A, 45B, 45C. A sensor node may have a vibration sensor 50, and an acoustic sensor 60. There may also be additional sensor means 41.

[0132] A sensor node 45 may comprise essential processing 72 and be adapted for performing the acts or at least part of the acts of measuring.

[0133] A set of sensors 40 may be understood as a sensor node 45 with one or more sensors. Such a sensor node 45 may comprise processors or means to configure, collect, store and process generated sensor data. A sensor node 45 may have communication means to communicate with a controller (not shown) or other sensor nodes. A sensor node 45 may have means to synchronize 240 (as illustrated previously) say sensors 50, 60 using a timestamp 34.

[0134] FIG. 8 illustrates a wind turbine generator 12 with a sensory arrangement in interaction with a remote/cloud-based processor 72 as part of a system for operating a wind turbine generator 90.

[0135] The rotary device 10 comprises a set of rotor blades 20. The set of rotor blades 20 consists of three rotor blades 22A, 22B, 22C.

[0136] Each blade 22A, 22B, 22C comprises a set of blade sensors 40A, 40B, 40C. In the present case, each set of blade sensors 40A, 40B, 40C comprises a blade sensor 42A, 42B, 42C.

[0137] The data sets 30 are processed by computational means 72. The wind turbine generator 12 may have a clock for generating a timestamp 34. In this case the time stamp is further synchronized and delivered from a global time server. Hence, the datasets 30 may be timestamped data 32. Alternatively, a sensor node 45 may be configured to generate data that is synchronized, and the timestamp 34 may be applied at sensor node 45 level.

[0138] The system 70 may interact with an operator system, a mobile device, a client server and a storage or database via a cloud/connection service. Further access or mirroring or monitoring may be available via the cloud for long term monitoring, alerts or service programmes.

[0139] The methods and acts of measuring sensory data disclosed herein may be performed in a single processor 72 device or be distributed as illustrated here.

[0140] FIG. 9 illustrates a configuration of a sensor node 45. There is a node power 46 management layout, which may include a source of energy, storage of energy, a controller of power management and an interface for configuration/control and possibly charging.

[0141] The sensor node 45 is illustrated with a processor or computational means 72. The sensor node 45 includes sensory means 41 generating data 31. Illustrated is a vibration sensor 50, which in this case has three lines of output and could e.g. be a tri-axial accelerometer. Optionally, there is an acoustic sensor 60. Optionally, there are further sensor(s) means 41.

[0142] The computational means 72 may be adapted to perform instructions to perform one or more, or all of the acts as outlined to perform measuring or sampling sensor data.

[0143] The sensor node 45 is configured with communication means 74 and here with storage means 76.

[0144] FIG. 10 illustrates a wind farm 80 without wake management 230.

[0145] The wind farm 80 comprises a set of wind turbine generators 12 experiencing a wind direction 132 resulting in, due to the positioning and the orientation of wind turbine generators 12, a specific wake conditions 130. The wind is coming as indicated by the two arrows, and it can be seen that the turbines from second and third row are in the wake of the turbines behind.

[0146] FIG. 11 illustrates a wind farm 80 with wake management 230; The wind is coming as indicated by the two arrows, and it can be seen that, due to the wake management, the turbines are now oriented in such a way that the impact of the wake field is reduced.

[0147] FIG. 12 illustrates a process of wake management 230. With reference to the flow chart and the previous description, the flow chart outlines operation of a wind turbine park taking a wake field map into account. [0148] A: There are start conditions and assessment; [0149] B: Determines if WTGs are in operation? [0150] C: Detecting/determining turbulence intensity on individual WTGs; [0151] D: Determines if the turbulence intensity level is acceptable for all WTGs? [0152] E: Transmit/send data/information of turbulence detection period to general data collection box; [0153] F: Calculate and define strategy for wake reduction according to defined wake management objective as a function of the established wake field of turbulence intensity; [0154] G: Send/transmit instructions and/or strategy from collection box to all WTG controllers of a wind farm.

[0155] As for D, then D can include supervised or unsupervised machine learning techniques, using the vibration data from the sensor nodes and the LiDAR unit for labeling the data (if supervised). The algorithms take as input the vibration measurements of one or more sensors, and they output the turbulence intensity, either expressed as a percentage (0 to 100%) or a unity number (0 to 1), depending on the use.

[0156] As for F, the calculation may as an example of wake management optimization could be established as:

[00001]$\underset{i=1}{\overset{N}{.Math.}}\left(\right)$

[0157] subject to:

Wake Minimization:

[0158]
θ.sub.i∈(θ.sub.min,θ.sub.max)

t.sub.i∈(t.sub.min,t.sub.max)

r.sub.i∈(r.sub.min,r.sub.max)

p.sub.i∈(p.sub.min,p.sub.max) [0159] or others

Wake Minimization+Optimizing Turbine Lifetime:

[0160]
l.sub.i∈(l.sub.min,l.sub.max) [0161] or others
where: [0162] Power.sub.i—power output from WTG.sub.i [0163] N—number of turbines in the wind farm (included in the optimization) [0164] θ.sub.i—static yaw misalignment from WTG.sub.i [0165] t.sub.i—turbulence intensity from WTG.sub.i [0166] r.sub.i—rotor rpm from WTG.sub.i [0167] p.sub.i—blade pitch angle from WTG.sub.i [0168] l.sub.i—average yaw misalignment from WTG.sub.i

[0169] Output for each individual WTG.sub.i: [θ.sub.i, r.sub.i and or p.sub.i and/or others]

[0170] FIG. 13 illustrates calculation of turbulence intensity for a wake map.

[0171] With reference to previous figures, there is an example of calculations for turbulence intensity (TI) detection using accelerometer data only from sensors 42 located inside wind turbine blades 22.

[0172] The data is obtained as acceleration data of 3-axis accelerometers placed identically as possible in all blades of a of a wind turbine generator. The three axes of an accelerometer sensor (channel1, channel2 and channel3) are all perpendicular to each other, in order to record acceleration from every possible direction.

[0173] The length of the sensor data is defined by:

length=Sqrt(channel1{circumflex over ( )}2+channel2{circumflex over ( )}2+channel3{circumflex over ( )}2).

[0174] This length is used for the calculations since it is independent from the orientation of the sensor 42 (the way it is installed in the blades). No matter what orientation the sensor 42 is installed with, in the blade 22, the output of the length would always be the same.

[0175] FIG. 13A shows what the length of a sensor looks like inside a wind turbine blade for one and a half rotation. FIG. 13B, shows the feature extraction and metric calculation.

[0176] To get rid of the influence of the installation distance of the sensor to the centre of the rotor, in a pre-processing process we centre and reduce the output length of the data.

[0177] Direct calculations may be applied data from the accelerometers from the blades to determine the turbulence intensity. Machine learning algorithms may also be applied to data from accelerometers from the blades in order to detect the turbulence intensity of the wind hitting the blades.

[0178] In order to label the full data and train the system, LiDAR data from a temporary installed LiDAR providing information about the actual turbulence intensity may be used.

[0179] FIG. 13C—shows the feature comparison and turbulence intensity determination. The “- - - ” line yields a turbulence index of “0.5”, The “______” line yields a turbulence index of “0.2”. The “-.-.” line is an example of a turbulence index to be determined, which may be based on the known turbulence intensities. The turbulence intensity may be determined according to a metric as indicated. The input of the Machine Learning (ML) algorithm is metrics of the centred-reduced length of acceleration data.

[0180] The metrics may be chosen from a variety of well-known measures, e.g.: Kurtosis, Crest factor, frequency of RPM, etc. The output is a Turbulence Intensity value.

[0181] The method of operating using wake management 230 may be performed by defining two or three wind turbines of the wind farm enduring the highest wake effect. There may be a test communication/control system, optionally over the whole wind far, that addresses optimizing the two or three turbines defined.

[0182] There may be a test of autonomous calculation (decision making) system focusing on the two to three turbines.

[0183] There may be a test of an individual WTG load when operating, say with yaw misalignment.

[0184] Based on the wake field map established by the sensors, e.g. accelerometers, a wake management strategy is defined and selections/options are made as to what extent the management strategy is to be applied: globally or locally, all turbines/only a set of turbines, what shall be controlled (yaw, pitch, load, all), etc.

[0185] Before applying the wake management strategy, there may be a test communication/control system and autonomous calculation system for all the wind farm.

[0186] FIG. 14 illustrates a process of wake management 230. With reference to the flow chart and the previous description, the flow chart outlines operation of a wind turbine park taking a wake field map into account. [0187] A: There are start conditions and assessments; [0188] B: Determine if the wind (aerodynamic) conditions have changed since last check? [0189] C: Determine if some/all WTGs in operation? [0190] D: Establish/detect turbulence intensity (aerodynamic conditions) on individual WTGs; [0191] E: Determine if turbulence intensity (aerodynamic condition) level is acceptable? [0192] F: Determine if the wake management/controllers know how to optimise operation with the turbulence intensity (aerodynamic condition)? [0193] G: Send/transmit some/all current data on the turbulence intensity (aerodynamic conditions) and/or power output of the current period to the general collection box; [0194] H: Perform calculations and define strategy for wake reduction according to the wake management objective;—an example may be according to step F in FIG. 12. [0195] I: Apply new optimisation of wind farm; send/transmit instructions/controls and strategy from collection box to some/all wind turbine controller of the wind farm; [0196] J: Determine if the wind condition (aerodynamic condition) changed? [0197] K: Determine if the optimisation is sufficient/satisfactory? [0198] L: Add new optimisation plan for new wind conditions (aerodynamic conditions) to a knowledge database (dictionary of optimizations) [0199] There may further be the following step: [0200] M: Apply optimization of wind farm from experience. Send out instructions and strategy from collection box to all wind turbine controllers in a farm.

[0201] As for step F: The wake management algorithm involves an optimization algorithm in order to speed up the process creates collections of experiences and every time when there is need for using a wake management, the procedure first looks at the collection. If the event is already registered in the collection, then the procedure does proceed with the same optimization algorithm again. Instead the procedure goes directly to Step M. If the event is not in the collection of experiences, then the procedure proceeds with the optimization, i.e., Step G.

[0202] FIG. 15 illustrates what a person skilled in the art will recognize as machine learning 3000 as a supervised machine learning 3010.

[0203] There is an act of training 3100 a supervised machine learning model (SML) 3015 with sensory data 31 of labeled time series data 3085 and building the supervised machine learning model (SML) 3015.

[0204] Actual training 3100 is based on sensory data 31 with data that are labelled 3085.

[0205] The training 3100 may associate data metrics with respect to wake conditions such as aerodynamic conditions such as turbulence, rain/hail etc. The training 3100 may associate data metrics with respect to turbulence intensity. Training results in a supervised machine learning model (SML) 3015.

[0206] There are one or more acts of verifying 3200 the supervised machine learning model (SML) 3015 by inputting a sensory data 31 of labelled time series data 3085 to the supervised machine learning model (SML) and outputting a calculated data label and comparing 3220 the output with known wake conditions 130.

[0207] The act of determining 1300 (not shown) may be the upper path of the verification is performed by inputting the sensory data 31 to the supervised machine learning model (SML) 3015 and outputting the wake condition 130 from the supervised machine learning model (SML) 3015.

[0208] The supervised machine learning model 3015 will train itself (“write its own algorithm”) on data with a label 3085. As is apparent and to verify the quality of the algorithm or model, data where the outcome result is know is used to test the labelled data 85 against the model3 to compare the outcome to the known label.

[0209] FIG. 15B illustrates the supervised machine learning model 3015 as outlined where the acts of training 3100 is performed on respective n-vibration signals 31.sub.i, . . . , n. (Not shown) generating respective 3015 supervised machine learning models 3015.sub.i, . . . , n.

[0210] The acts of verifying 3200.sub.i, . . . n comprises and is performed on respective n-multiple vibration signals 31.sub.1, . . . ,31.sub.n obtained by corresponding n-multiple vibration sensors 40.sub.1, . . . 40.sub.n. The act of training 3100 is performed based on n-multiple labels (Y.sub.1 . . . n).

[0211] The act of verifying 3200 comprises an is performed based on a predetermined average measure of the n-multiple labels (Y.sub.1 . . . n) and finally by comparing 3220 the average with labeled 3085 data. The averaging may be a mean-average or similar average measures.

[0212] In this scenario several sensors are used individually and the results of the outcome are combined to find the best possible value.

[0213] FIG. 16 illustrates a confusion matrix between predicted and measured data based on supervised machine learning algorithm of turbulence index (TI). On the X-axis there are predicted turbulence intensities using the vibrations of the sensors. On the Y-axis, there are the true turbulence intensities, i.e. wake conditions, measured by a nacelle-based LiDAR

[0214] Since the data are clustered around the diagonal, the algorithm has performed very well. Except for two or three values, all the others are exactly on the diagonal, which means that the algorithm is predicting very well the turbulence index (TI) for the wake management very well.