Adaptive noise control of windfarm

Abstract

The invention provides a method for controlling noise of wind turbines in a windfarm. Measured weather data measured noise data are obtained (O_WND) from a plurality of sensor nodes arranged at different positions in relation to the windfarm. Further, weather conditions (O_W_WT) are obtained at least at one wind turbine, e.g. from sensors / predictors in the wind turbine. The noise and weather data from the sensor nodes and wind turbine(s) along with distances (D) between at least some of the wind turbines and at least some of the sensor nodes are used to train a learning algorithm. Further, the learning algorithm is used to predict noise (P_N) at a position of one of the sensor nodes based on measured or predicted weather conditions at the windfarm. Next, extrapolating (EX_N) the predicted noise at said position of one of the sensor nodes to a predicted noise at a point of interest in surroundings of the windfarm. Further, changing mode of operation (C_M_O) of one or more wind turbines of the windfarm based on the predicted noise at the point of interest, especially if the predicted noise exceeds a noise limit for the point of interest P_I. The method allows compliance with noise limits e.g. by slightly re-orienting one or more wind turbines to reduce noise, if it is predicted that a noise limit is exceeded, and this only costs a slightly lower than optimal energy generation in a limited time period.

Description

[0001] ADAPTIVE NOISE CONTROL OF WINDFARM

[0002] FIELD OF THE INVENTION

[0003] The present invention relates to the field of wind turbines, more specifically the invention relates to a windfarm with a plurality of wind turbines.

[0004] BACKGROUND OF THE INVENTION

[0005] Acoustic noise from wind turbines in a windfarm is a common cause for derating electric power production from the windfarm, thereby limiting the power generation capacity and prompting complaints from local residents. Existing noise suppression solutions are based on hard-coded settings and dependent on noise test campaigns which limit their capability to measure noise and perform adjustments to address noise compliances timely.

[0006] Furthermore, various conditions in the environments can change acoustic noise transmission properties significantly from a large windfarm to a point in the surroundings at the large distance from the windfarm. This causes problems controlling the windfarm for optimal power generation and still complying with noise limits in the surroundings of a windfarm.

[0007] SUMMARY OF THE INVENTION

[0008] Thus, according to the above description, it is an object of the present invention to provide a method and system for controlling wind turbines of a windfarm to allow an optimized operation with a high power generation and still allow the windfarm to comply with noise limits in the surroundings.

[0009] In a first aspect, the invention provides a method for controlling noise of wind turbines in a windfarm comprising a plurality of wind turbines, the method comprising :

[0010] - obtaining measured weather data and measured noise data from each of a plurality of sensor nodes arranged at different positions in relation to the windfarm, - obtaining weather conditions at one or more wind turbines of the windfarm,

[0011] - training a computer-based learning algorithm based on: a) the obtained measured noise data and weather data from the sensor nodes, b) the obtained weather conditions at the one or more wind turbines, and c) distances between at least some of the one or more wind turbines and at least some of the sensor nodes,

[0012] - predicting noise at a position of one of the sensor nodes in response to a measured or predicted weather condition at the windfarm based on the learning algorithm,

[0013] - extrapolating the predicted noise at said position of one of the sensor nodes to a predicted noise at a point of interest in surroundings of the windfarm, and

[0014] - changing mode of operation of one or more wind turbines of the windfarm based on the predicted noise at the point of interest.

[0015] Such method is advantageous, since it is possible to adaptively control the wind turbines of the windfarm for optimal power generation taking into account weather conditions which can significantly influence noise from the wind turbines in the surroundings at long distances from the wind turbines.

[0016] By training a machine learning or Artificial Intelligence (Al) based algorithm with measured noise data and weather condition data, it is possible to predict noise at a point of interest in the surroundings of the windfarm if the weather condition is known. By use of the learning algorithm, a rather precise noise prediction is possible without exact knowledge of the acoustic noise propagation from each wind turbine involved which are highly dependent on weather conditions in a complex manner involving many parameters.

[0017] In this way, it is possible to control mode of operation of the wind turbines to avoid exceeding noise limits and at the same time ensure efficient operation utilizing the available noise margin to avoid derating power generation of the wind turbines, or at least avoid derating power generation of wind turbines which do not contribute to the resulting noise at the point of interest.

[0018] Based on predicted noise levels in upcoming weather conditions, it is possible to control operation of the wind turbines to comply with noise limits with a minimal impact on the generated electric power from the windfarm. For example, it may be possible to predict only to be necessary to change mode of operation to reduce noise from one or a few wind turbines near a noise critical point in the surroundings of the windfarm to comply with the setup noise limit. Thereby, full power generation from the remaining wind turbines of the windfarm can be obtained. Thus, based on the invention, a windfarm can adapt to various weather conditions and comply with noise limits as well as optimize energy production.

[0019] In the following, preferred features and embodiments will be described.

[0020] The step of changing mode of operation of one or more wind turbines may comprises the step of changing at least one of: orientation of at least one wind turbine, and pitch of blades of at least one wind turbine, in response to the predicted noise at the point of interest. In this way, it is possible to reduce noise contribution from the wind turbine(s).

[0021] The method may comprise the step of comparing the predicted noise at the point of interest with a noise limit. Hereby, it is possible to determine if the wind turbines can be operated for full power generation or if noise reducing mode of operation should be initiated by at least some of the wind turbines. Especially, the method may comprise the step of changing mode of operation of at least one of the wind turbines in case the predicted noise at the point of interest exceeds the noise limit. In this way it is possible to reduce the noise in the point of interest to comply with the noise limit. More specifically, the method may comprise changing mode of operation of at least one of the wind turbines in advance of the predicted noise at the point of interest exceeds the noise limit, so as to avoid exceeding the noise limit at the point of interest. The method may comprise changing mode of operation of a plurality of the wind turbines in case the predicted noise at the point of interest exceeds the noise limit.

[0022] The method may comprise repeating predicting noise at the point of interest in case it is predicted that a noise limit will be exceeded, namely to test the effect of various changes of mode of operation of one or a subset of the wind turbines and thereby determine which wind turbine(s) to be involved in mode of operation change to reduce noise. Especially, the total power generation may be compared between various scenarios of mode of change of wind turbine(s).

[0023] The learning algorithm may be a machine learning algorithm, such as a learning algorithm involving an Al algorithm. Especially, the learning algorithm may involve a neural network based algorithm. The learning algorithm may especially comprise a Recurrent Neural Network, a Convolutional Neural Network, or a hybrid solution combined with Bayesian regression and Kalman Filters for the prediction. However, it is to be understood that the learning algorithm may involve other Al-solutions.

[0024] The weather data and noise data used as into the the learning algorithm may be sampled or averaged over a suitable period. For example, it may be preferred to use high resolution data with a sample rate of the order of 1-100 ms, such as 100-500 ms. However, it may be preferred to use data sampled or averaged over 1-10 second or sampled or averaged over 10-60 seconds, or data sampled or averaged over 1-30 minutes, for example sampled or averaged over 2-10 minutes.

[0025] In some embodiments, the learning algorithm is trained to determine individual noise contributions from the plurality of the wind turbines.

[0026] The method may comprise obtaining measured noise data from a plurality of sensor nodes arranged at different distances and different angular locations relative to one of the wind turbines. Especially, sensor nodes may be arranged at different distances and different angular locations of a subset of the wind turbines of the windfarm, or all of the wind turbines of the windfarm. Thereby, important noise data are provided to the learning algorithm which allows a precise determination of individual noise contributions from each of the wind turbines under various weather conditions.

[0027] The sensor nodes may especially be configured to generate data indicative of measured high-frequency noise levels, for example this may be defined as noise in the frequency range above 500 Hz, or above 1 kHz, e.g. in the frequency range 1-10 kHz, along with time stamps. It has been found that noise level in the frequency range 1-5 kHz is the most relevant with respect to wind turbine noise contribution in the surroundings, however the most dominant frequency range will in general depend on the type of wind turbine as well as distance to noise sensitive areas in the surroundings as well as other parameters.

[0028] The noise data may be communicated as a measured sound pressure level dB value or in another way. Especially, the sound pressure level value communicated and entered to the learning algorithm may be such as: LAeq or Lpeak or both. Further, the noise data to the learning algorithm may comprise spectral noise data, e.g. in the form of 1 / 3 or 1 / 1 octave values.

[0029] Especially, the sensor nodes may be configured to generate data indicative of measured wind speeds, wind directions, temperatures, and air pressure along with time stamps. Further the sensor nodes may have a sensor configured to measure precipitation.

[0030] The sensor nodes may especially be combined sensor nodes configured to generate noise and weather data at one position above ground level.

[0031] The training of the learning algorithm may be further based on d) operating conditions of the one or more wind turbines. In this way, it is possible to train the learning algorithm with the effect of noise reducing mode of operation of the wind turbines as well as other conditions which may influence noise contribution from the wind turbines.

[0032] The method may comprise obtaining weather conditions at one or more wind turbines of the windfarm based on data received from the one or more wind turbines. These weather conditions may be based on weather sensors (such as wind speed, wind direction, temperature, pressure, and humidity sensors) located in the nacelle of the wind turbines or based on a weather estimator located in the wind turbines. The weather condition data may further include sensed precipitation and lightning hits collected at the nacelle of the wind turbine(s). The method may comprise obtaining weather conditions from a plurality of the sensor nodes, for example based on a weather sensor or a weather estimator located in the sensor nodes.

[0033] Especially, the step of training may be performed based on obtained weather conditions at a first group of the plurality of wind turbines, and wherein the step of changing the mode of operation is performed on a second group of the plurality of wind turbines, wherein the first and second groups are non-overlapping.

[0034] The position above ground level of the sensor node may be in the range from 0.5 meters to 4 meters, for example it may be 1-3 meters, e.g. 1.8 meters. This has been found to be a suitable height for measuring noise as well as weather condition data.

[0035] The sensor nodes may comprise a microphone for detecting or measuring noise as well as one or more weather condition sensors, such as one or more of: wind speed, wind direction, pressure, temperature, and humidity sensors. The sensor node may have one single housing positioned at least 0.5-4.0 m above ground level, and the housing may comprise noise sensor(s) as well as weather condition sensors. Further, the sensor nodes may comprise a wireless RF transmitter to allow transmission of noise data and weather data to a remotely located receiver. The sensor nodes may be configured to transmit noise data and weather data along with a time stamp indicating when the noise and weather data were obtained.

[0036] In preferred embodiments, the method comprises individually controlling operation of the plurality of wind turbines of the windfarm according to the determined noise control strategy. Especially, the method may involve individually controlling operation of rotors of the two or more of the plurality of wind turbines, so as to reduce noise at a point of interest in the surroundings of the windfarm.

[0037] Especially, noise from the windfarm may be controlled by means of controlling pitch and / or rotor speed and / or yaw of one or more rotor blades of the one or more wind turbines. In some embodiments, the method comprises determining individual control of the wind turbines of the windfarm control strategy based on the predicted noise at a point of interest so as to optimize power production, e.g. Annual Energy Production (AEP) at the windfarm level and at the same time complying with a noise limit at the point of interest.

[0038] For example, it may be chosen to compromise power production at night time, if the wind turbine is located near dwelling with stricter noise requirements at night.

[0039] In a second aspect, the invention provides a sensor system for predicting noise in surroundings of a windfarm comprising a plurality of wind turbines, the sensor system comprises

[0040] - a plurality of sensor nodes arranged at different positions in relation to the windfarm, wherein each of the sensor nodes comprises a sensor for generating noise data and weather data at a position above ground level,

[0041] - a computer system comprising a processor, wherein the computer system is configured

[0042] - to receive measured weather data and measured noise data from the plurality of sensor nodes,

[0043] - to receive weather conditions at one or more wind turbines of the windfarm,

[0044] - to train a computer-based learning algorithm based on: a) the obtained measured noise data and weather data from the sensor nodes, b) the obtained weather conditions at the one or more wind turbines, and c) distances between at least some of the one or more wind turbines and at least some of the sensor nodes,

[0045] - to predict noise at a position of one of the sensor nodes in response to a measured or predicted weather condition at the windfarm based on the learning algorithm,

[0046] - to extrapolate the predicted noise at said position of one of the sensor nodes to a predicted noise at a point of interest in surroundings of the windfarm, and

[0047] - to generate an output indicative of the predicted noise at the point of interest. In preferred embodiments, the sensor system is cloud based, and thus data from the sensor nodes and wind turbines are communicated wirelessly to a cloud-based learning algorithm which will perform the prediction. Further, it is preferred that the learning algorithm is supplied with information about location of the wind turbines as well as the sensor nodes, or at least a relative distance between a wind turbine and some of the sensor nodes.

[0048] In a third aspect, the invention provides a windfarm comprising

[0049] - a plurality of wind turbines each comprising

[0050] - a rotor comprising a plurality of rotor blades,

[0051] - an electric generator coupled to the rotor, wherein the rotor is coupled to rotate the electric generator,

[0052] - a sensor system according to the second aspect, and

[0053] - a windfarm controller comprising a processor, wherein the windfarm controller is configured

[0054] - to receive the output indicative of the predicted noise at the point of interest from the sensor system,

[0055] - to determine to change a mode of operation of one or more wind turbines of the windfarm based on the predicted noise at the point of interest, and

[0056] - to send a control signal to one or more wind turbines of the windfarm to cause the one or more wind turbines to change mode of operation according to said determined change of mode of operation.

[0057] Especially, the windfarm controller may be configured to receive a noise limit for a point of interest in the surroundings of the windfarm, and to determine to change mode of operation of one or more wind turbines if the predicted noise at the point of interest exceeds said noise limit. In this way, the windfarm can be controlled to comply with noise limits at a point of interest in the surroundings of the windfarm. Especially, method may comprise performing the prediction of noise at the point of interest in advance to allow time to change mode of operation, e.g. orientation of one or more of the wind turbines to reduce noise and thereby avoiding that a noise limit at the point of interest is exceeded. It is to be understood that the same advantages and preferred embodiments and features described for the first aspect apply as well for the second and third aspects, and the aspects may be mixed in any way.

[0058] BRIEF DESCRIPTION OF THE FIGURES

[0059] The invention will now be described in more detail with regard to the accompanying figures of which

[0060] FIG. 1 illustrates an example of a wind turbine for a windfarm, where the wind turbine has weather condition and / or weather predictors arranged in the nacelle, FIG. 2 illustrates a first step of a method embodiment, where noise data and weather condition data are collected from sensor nodes and wind turbine and applied to a cloud-based learning algorithm,

[0061] FIG. 3 illustrates a second step of a method embodiment, where predicted noise at sensor nodes is extrapolated to predicted noise at a point of interest.

[0062] FIG. 4 illustrates a third step of a method embodiment, where noise is predicted at a point of interest in advance, and where the wind turbine orientation is tuned to reduce noise at the point of interest,

[0063] FIG. 5 illustrates a windfarm embodiment, and

[0064] FIG. 6 illustrates step of a method embodiment.

[0065] The figures illustrate specific ways of implementing the present invention and are not to be construed as being limiting to other possible embodiments falling within the scope of the attached claim set.

[0066] DETAILED DESCRIPTION OF THE INVENTION

[0067] FIG. 1 illustrates an example of a wind turbine with typically two or three rotor blades BL which serve to drive an electric generator GN located inside the nacelle NC on top of a tower TW. Such wind turbine may generate an electric power of at least 1 MW, such as 2-10 MW, or more than 10 MW. The wind turbine may have a controller serving to control operation of the wind turbine, such as rotor speed, pitch of the blades BL and yaw of the nacelle NC (orientation of the wind turbine). The wind turbine is suited for a windfarm of the present invention since it can be controlled with respect to its mode of operation in accordance with an external windfarm controller. Hereby, the windfarm controller may determine to change orientation (yaw) or other parameters relating to mode of operation, so as to reduce noise in a particular position where it has been predicted that the total noise from the windfarm exceeds a noise limit.

[0068] FIG. 2 illustrates first step of a method embodiment, where a wind turbine WT of a windfarm is positioned point of interest P_I, e.g. a dwelling near the windfarm. The wind turbine WT generates noise dependent on weather conditions, e.g. wind speed and wind direction, as well as other parameters. Depending on the various parameter, the noise propagates to the point of interest P_I where there may exist a noise limit which is expected to be complied with.

[0069] A sensor system in the form of sensor nodes SN are setup to measure noise at various known distances and angular positions relative to the wind turbine WT. Further, at least some of the sensor nodes have weather condition sensors, e.g. wind speed, wind direction, temperature, and air pressure sensors. The noise and weather data SND from the sensor nodes SN are transmitted via a wi-fi gateway to a cloud-based learning algorithm to train the learning algorithm along with weather condition data TD from the wind turbine WT, e.g. both sensed current weather parameters as well as predicted parameters. As illustrated, data for training the learning algorithm may include noise data measured at a location near the point of interest P_I.

[0070] The sensor nodes SN may have noise and weather sensors arranged at a height H of 0.5-4 meters above the ground, e.g. a height H of 1.8 meters above the ground.

[0071] Based on the mentioned data from sensor nodes SND and data TD from the wind turbine WT over a period of time, thus covering various operating and weather conditions, the learning algorithm has been trained to predict the noise level at a position of one of the sensor nodes SN. A computer system CS can access the learning algorithm and based on a weather forecast, the computer system CS is programmed to extrapolate predicted noise at a sensor node SN to a predicted noise level PN at the point of interest P_I, and the computer system CS can generate an output of the predicted noise level PN.

[0072] The predicted noise level PN can be communicated to a windfarm controller which can change mode of operation of the wind turbine WT, e.g. orientation (yaw), to reduce its noise contribution at the point of interest P_I in case it is predicted that a noise limit is exceeded.

[0073] FIG. 3 illustrates step two of the method embodiment, where the learning model has been trained as explained above. Based on weather forecast data produced by the wind turbine WT along with other relevant weather data, the learning model can predict a noise level at the positions of two sensor nodes SN1, SN2 in a direction relevant with respect to the point of interest P_I, along with distances D, DP from the wind turbine WT, it is possible to extrapolate the noise prediction to the point of interest P_I, even though this point P_I is at a distance DP from the wind turbine WT which is outside the distance range of the sensors SN1, SN2.

[0074] FIG. 4 illustrates step three of the method embodiment, where the noise level at the point of interest P_I is predicted, based on predicted weather conditions. If the predicted noise level exceeds the noise limit for the point of interest P_I, it is possible to change mode of operation of the wind turbine WT in time to avoid the noise level actually exceeding the noise limit at the point of interest P_I. Here, the curved arrows indicate that the wind turbine WT is re-oriented (change of yaw) to reduce its noise contribution at the point of interest P_I and thereby reduce the total noise at the point of interest P_I which may thereby be brought to a level below the noise limit.

[0075] Such re-orientation of the wind turbine WT may reduce electric power generated by the wind turbine WT, but this is a significantly smaller amount of power lost compared to a derating event. Furthermore, it can be predicted for how long time the change of mode of operation to reduce noise should be maintained to avoid exceeding the noise limit at the point of interest P_I. Therefore, the wind turbine WT can return to a normal mode of operation with high power generation, once it can be predicted that the weather conditions change, so the predicted noise level will no longer exceed the noise limit by the normal mode of operation.

[0076] FIG. 5 illustrates a sketch of a windfarm WF embodiment with a plurality of wind turbines, such as that illustrated in FIG. 1. These wind turbines are set up in a given configuration covering an area.

[0077] A windfarm controller WFC receives input from a sensor system SNS with a plurality of sensor nodes providing noise and weather data to a learning algorithm. The sensor system SNS predicts noise from the windfarm based on predicted weather condition, e.g. a weather forecast. The sensor system SNS outputs a predicted total noise from the windfarm at a point of interest. Along with a noise limit NL, the windfarm controller WFC can send control signals to change mode of operation of one or more of the wind turbines, e.g. change their orientation, to reduce noise at the point of interest, if it the predicted noise exceeds the noise limit NL. Thereby, it is possible to comply with the noise limit NL, since mode of operation can be changed in advance of the noise level actually reaches the predicted too high noise level.

[0078] FIG. 6 illustrates steps of a method embodiment, i.e. a method for controlling noise of wind turbines in a windfarm comprising a plurality of wind turbines. The method comprises obtaining measured weather data and measured noise data O_WND from each of a plurality of sensor nodes arranged at different positions in relation to the windfarm. Further, obtaining weather conditions 0_W_WT at one or more wind turbines of the windfarm. Next, training a computer-based learning algorithm T_LA based on: a) the obtained measured noise data and weather data from the sensor nodes, b) the obtained weather conditions at the one or more wind turbines, and c) distances between at least some of the one or more wind turbines and at least some of the sensor nodes. Next, based on the trained learning algorithm, predicting noise P_N at a position of one of the sensor nodes in response to a measured or predicted weather condition at the windfarm. Further, extrapolating EX_N the predicted noise at said position of one of the sensor nodes to a predicted noise at a point of interest in surroundings of the windfarm. Finally, changing mode of operation C_M_O of one or more wind turbines of the windfarm based on the predicted noise at the point of interest. Especially, changing mode of operation C_M_O such as orientation of one or more wind turbines to reduce noise, in case the predicted noise at the point of interest exceeds a noise limit setup for the point of interest.

[0079] To sum up, the invention provides a method for controlling noise of wind turbines in a windfarm. Measured weather data measured noise data are obtained (O_WND) from a plurality of sensor nodes arranged at different positions in relation to the windfarm. Further, weather conditions (0_W_WT) are obtained at least at one wind turbine, e.g. from sensors / predictors in the wind turbine. The noise and weather data from the sensor nodes and wind turbine(s) along with distances (D) between at least some of the wind turbines and at least some of the sensor nodes are used to train a learning algorithm. Further, the learning algorithm is used to predict noise (P_N) at a position of one of the sensor nodes based on measured or predicted weather conditions at the windfarm. Next, extrapolating (EX_N) the predicted noise at said position of one of the sensor nodes to a predicted noise at a point of interest in surroundings of the windfarm. Further, changing mode of operation (C_M_O) of one or more wind turbines of the windfarm based on the predicted noise at the point of interest, especially if the predicted noise exceeds a noise limit for the point of interest P_I. The method allows compliance with noise limits e.g. by slightly re-orienting one or more wind turbines to reduce noise, if it is predicted that a noise limit is exceeded, and this only costs a slightly lower than optimal energy generation in a limited time period.

[0080] Although the present invention has been described in connection with the specified embodiments, it should not be construed as being in any way limited to the presented examples. The scope of the present invention is to be interpreted in the light of the accompanying claim set. In the context of the claims, the terms "including" or "includes" do not exclude other possible elements or steps. Also, the mentioning of references such as "a" or "an" etc. should not be construed as excluding a plurality. The use of reference signs in the claims with respect to elements indicated in the figures shall also not be construed as limiting the scope of the invention. Furthermore, individual features mentioned in different claims, may possibly be advantageously combined, and the mentioning of these features in different claims does not exclude that a combination of features is not possible and advantageous.

Claims

CLAIMS1. A method for controlling noise of wind turbines in a windfarm comprising a plurality of wind turbines, the method comprising :- obtaining measured weather data (O_WND) and measured noise data from each of a plurality of sensor nodes arranged at different positions in relation to the windfarm,- obtaining weather conditions (0_W_WT) at one or more wind turbines of the windfarm,- training a computer-based learning algorithm (T_LA) based on: a) the obtained measured noise data and weather data (SND) from the sensor nodes, b) the obtained weather conditions (TD) at the one or more wind turbines, and c) distances (D) between at least some of the one or more wind turbines and at least some of the sensor nodes,- predicting noise (P_N) at a position of one of the sensor nodes in response to a measured or predicted weather condition at the windfarm based on the learning algorithm,- extrapolating (EX_N) the predicted noise at said position of one of the sensor nodes to a predicted noise at a point of interest in surroundings of the windfarm, and- changing mode of operation (C_M_O) of one or more wind turbines of the windfarm based on the predicted noise at the point of interest.

2. The method according to claim 1, wherein the step of changing mode of operation of one or more wind turbines comprises the step of changing at least one of: orientation of at least one wind turbine, and pitch of blades of at least one wind turbine, in response to the predicted noise at the point of interest.

3. The method according to claim 1 or 2, comprising the step of comparing the predicted noise at the point of interest with a noise limit.

4. The method according to claim 3, comprising the step of changing mode of operation of at least one of the wind turbines in case the predicted noise at the point of interest exceeds the noise limit.

5. The method according to claim 4, comprising the step of changing mode of operation of at least one of the wind turbines in advance of the predicted noise at the point of interest exceeds the noise limit, so as to avoid exceeding the noise limit at the point of interest.

6. The method according to claim 4 or 5, comprising changing mode of operation of a plurality of the wind turbines in case the predicted noise at the point of interest exceeds the noise limit.

7. The method according to any of the preceding claims, wherein the learning algorithm is trained to determine individual noise contributions from the plurality of the wind turbines.

8. The method according to any of the preceding claims, comprising obtaining measured noise data from a plurality of sensor nodes arranged at different distances and different angular locations relative to one of the wind turbines.

9. The method according to any of the preceding claims, wherein each of the sensor nodes is configured to generate data indicative of measured high- frequency noise levels along with time stamps, such as measured noise levels in a frequency range of 1-5 kHz along with time stamps.

10. The method according to any of the preceding claims, wherein each of the sensor nodes is configured to generate data indicative of measured wind speeds, wind directions, temperatures, and air pressure along with time stamps.

11. The method according to any of the preceding claims, wherein each of the sensor nodes are combined sensor nodes configured to generate noise and weather data at one position above ground level.

12. The method according to any of the preceding claims, wherein training of the learning algorithm is further based on d) operating conditions of the one or more wind turbines.

13. The method according to any of the preceding claims, comprising obtaining weather conditions at one or more wind turbines of the windfarm based on data received from the one or more wind turbines, such as based on weather sensors or a weather estimators located in the wind turbines.

14. The method according to any of the preceding claims, comprising obtaining weather conditions from a plurality of the sensor nodes, such as based on a weather sensor or a weather estimator located in the sensor nodes.

15. The method according to any of the preceding claims, wherein the step of training is performed based on obtained weather conditions at a first group of the plurality of wind turbines, and wherein the step of changing the mode of operation is performed on a second group of the plurality of wind turbines, wherein the first and second groups are non-overlapping.

16. A sensor system (SNS) for predicting noise in surroundings of a windfarm (WF) comprising a plurality of wind turbines, the sensor system comprises- a plurality of sensor nodes (SN) arranged at different positions in relation to the windfarm, wherein each of the sensor nodes (SN) comprises a sensor for generating noise data and weather data at a position (H) above ground level,- a computer system (CS) comprising a processor, wherein the computer system (CS) is configured- to receive measured weather data and measured noise data from the plurality of sensor nodes (SN),- to receive weather conditions at one or more wind turbines (WT) of the windfarm (WF),- to train a computer-based learning algorithm based on: a) the obtained measured noise data and weather data from the sensor nodes (SN), b) the obtained weather conditions at the one or more wind turbines (WT), and c) distances (D) between at least some of the one or more wind turbines (WT) and at least some of the sensor nodes (SN),- to predict noise at a position of one of the sensor nodes (SN) in response to a measured or predicted weather condition at the windfarm (WF) based on the learning algorithm,- to extrapolate the predicted noise at said position of one of the sensor nodes (SN) to a predicted noise at a point of interest (P_I) in surroundings of the windfarm (WF), and- to generate an output indicative of the predicted noise (PN) at the point of interest (P_I).

17. A windfarm (WF) comprising- a plurality of wind turbines (WT1, WT2) each comprising- a rotor comprising a plurality of rotor blades (BL),- an electric generator (GN) coupled to the rotor, wherein the rotor is coupled to rotate the electric generator (GN),- a sensor system (SNS) according to claim 16, and- a windfarm controller (WFC) comprising a processor, wherein the windfarm controller (WFC) is configured- to receive the output indicative of the predicted noise (PN) at the point of interest (P_I) from the sensor system (SNS),- to determine to change a mode of operation of one or more wind turbines (WT1, WT2) of the windfarm (WF) based on the predicted noise (PN) at the point of interest (P_I), and- to send a control signal to one or more wind turbines (WT1, WT2) of the windfarm (WF) to cause the one or more wind turbines (WT1, WT2) to change mode of operation according to said determined change of mode of operation.

18. The windfarm (WF) controller according to claim 17, wherein the windfarm controller (WFC) is configured to receive a noise limit (NL) for a point of interest (P_I) in the surroundings of the windfarm (WF), and to determine to change mode of operation of one or more wind turbines (WT1, WT2) if the predicted noise (PN) at the point of interest exceeds said noise limit (NL).

Images