METHOD FOR CONTROLLING NOISE GENERATED BY A WIND FARM

Abstract

A method for controlling noise generated by a wind farm with a plurality of wind turbines is disclosed. In the event that a predicted noise level exceeds a predefined threshold noise value, one or more wind turbines are selected using a noise propagation model and respective wind turbine models for the selected one or more wind turbines, and by performing an optimisation process to reduce the predicted noise level at the predefined evaluation position to a level below the predefined threshold noise value while maximising the total power production of the wind farm.

Claims

1. A method for controlling noise generated by a wind farm, the wind farm comprising a plurality of wind turbines arranged at a wind farm site, the method comprising: providing a noise propagation model related to noise propagation across the wind farm and in the vicinity of the wind farm, under various operating conditions, the noise propagation model including the wind turbines of the wind farm as noise generators, and the noise propagation model taking interactions among the wind turbines into account, for each wind turbine of the wind farm, providing a wind turbine model, the wind turbine model relating to at least power production and noise generation of the wind turbine under various operating conditions, predicting a noise level at a predefined evaluation position, based on the noise propagation model, the wind turbine models and information regarding the current operating conditions, in the case that the predicted noise level exceeds a predefined threshold noise value, selecting one or more wind turbines among the wind turbines of the wind farm, and changing operation of the one or more selected wind turbines, wherein selecting one or more wind turbines and changing operation of the one or more selected wind turbines are performed using the noise propagation model and the wind turbine models, and by performing an optimisation process with the predefined threshold noise value at the predefined evaluation position as a constraint, the noise generation and the power production of the wind turbines as optimisation variables, and the total power production of the wind farm as an optimisation target, thereby reducing the predicted noise level at the predefined evaluation position to a level below the predefined threshold noise value while maximising the total power production of the wind farm.

2. A method according to claim 1, further comprising providing a site specific wind flow field across the wind farm site, based on the current operating conditions, and wherein predicting a noise level, selecting one or more wind turbines and changing operation of the one or more selected wind turbines is further performed based on the site specific wind flow field.

3. A method according to claim 2, wherein the site specific wind flow field is based at least partly on a high resolution weather model.

4. A method according to claim 2, wherein the site specific wind flow field is based at least partly on expected turbulence patterns at the wind farm site.

5. A method according to any of claim 2, wherein providing a site specific wind flow field is based at least partly on a wind flow model related to the wind farm site, and wherein the noise propagation model is at least partly based on the wind flow model.

6. A method according to claim 1, wherein at least one of providing a noise propagation model and providing wind turbine models comprises training an artificial intelligence (AI) model.

7. A method according to claim 6, wherein training the artificial intelligence (AI) model comprises applying reinforced learning.

8. A method according to claim 1, wherein changing operation of the one or more selected wind turbines comprises selecting one or more control parameter settings and adjusting the selected control parameter settings in a selected manner.

9. A method according to claim 1, wherein the noise propagation model further relates to frequency components of the generated noise.

10. A method according to claim 1, further comprising updating at least one of the noise propagation model and the wind turbine models during operation of the wind farm.

11. A method according to claim 1, wherein the noise propagation model is at least partly based on the wind turbine models.

12. A method for controlling noise generated by a wind farm, the wind farm comprising a plurality of wind turbines arranged at a wind farm site, the method comprising: providing a noise propagation model related to noise propagation across the wind farm and in the vicinity of the wind farm, under various operating conditions, the noise propagation model including the wind turbines of the wind farm as noise generators, and the noise propagation model taking interactions among the wind turbines into account; for each wind turbine of the plurality of wind turbines, providing a wind turbine model, the wind turbine model relating to at least power production and noise generation of the wind turbine under various operating conditions; predicting a noise level at a predefined evaluation position, based on the noise propagation model, the wind turbine models and information regarding the current operating conditions; when the predicted noise level exceeds a predefined threshold noise value, selecting one or more wind turbines among the wind turbines of the wind farm; and changing operation of the one or more selected wind turbines; wherein selecting one or more wind turbines and changing operation of the one or more selected wind turbines are performed using the noise propagation model and the wind turbine models, and by performing an optimisation process with the total power production of the wind farm as an optimisation target.

13. A method according to claim 12, further comprising providing a site specific wind flow field across the wind farm site, based on the current operating conditions, and wherein predicting a noise level, selecting one or more wind turbines and changing operation of the one or more selected wind turbines is further performed based on the site specific wind flow field.

14. A method according to claim 13, wherein the site specific wind flow field is based at least partly on a high resolution weather model.

15. A method according to claim 13, wherein the site specific wind flow field is based at least partly on expected turbulence patterns at the wind farm site.

16. A method according to any of claim 13, wherein providing a site specific wind flow field is based at least partly on a wind flow model related to the wind farm site, and wherein the noise propagation model is at least partly based on the wind flow model.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

[0051] The invention will now be described in further detail with reference to the accompanying drawings in which

[0052] FIG. 1 illustrates a wind farm in which a method according to an embodiment of the invention may be performed,

[0053] FIG. 2 is a diagram illustrating a method according to an embodiment of the invention, and

[0054] FIG. 3 is a flow chart illustrating a method according to an embodiment of the invention.

DETAILED DESCRIPTION OF THE DRAWINGS

[0055] FIG. 1 illustrates a wind farm 1 comprising a plurality of wind turbines 2, four of which are shown. The site of the wind farm 1 defines elevation variations in the form of hills 3 and valleys 4, which affect the wind flow through the wind farm 1. Two metmasts 5 are arranged within the wind farm 1 for measuring weather conditions, including wind conditions, such as wind speed and wind direction.

[0056] Noise generated by the wind farm 1 may be controlled in the following manner. A noise propagation model related to noise propagation across the wind farm 1 and in the vicinity of the wind farm 1, under various operating conditions, is provided. The wind turbines 2 form noise generators of the noise propagation model, and the wind flow across the wind farm 1, under the various operating conditions, is taken into account when providing the noise propagation model. Thereby the noise propagation model may be used for predicting how noise generated by the wind turbines 2 propagates across and immediately outside the wind farm 1, under the various operating conditions.

[0057] Furthermore, for each wind turbine 2 of the wind farm 1, a wind turbine model is provided. The wind turbine models relate to at least power production and noise generation of the respective wind turbines 2 under various operating conditions. There is a link between power production and noise generation of a wind turbine 2. For instance, lowering the power production of a wind turbine 2 will normally result in a reduction in noise generated by the wind turbine 2. Accordingly, the noise generated by a wind turbine 2 may be reduced by appropriately reducing the power production of the wind turbine 2. The wind turbine models reflect this relationship, and the penalty, in terms of lost power production, for a given noise reduction can therefore be derived from the wind turbine models.

[0058] The wind turbine models are applied into the noise propagation model, in the sense that they represent noise profiles of the wind turbines 2, and thereby the noise generators, of the noise propagation model. Furthermore, the noise propagation model takes interactions among the wind turbines 2 into account. Operation of the wind turbines 2 affects the operation of the other wind turbines 2 of the wind farm 1, in the sense that the operating conditions across the wind farm 1 change in response to the operation of the wind turbines 2. For instance, wake effects will be introduced by the wind turbines 2, which in particular affect the operating conditions applying to downstream wind turbines 2, but which may also affect wind turbines 2 arranged upstream.

[0059] Next a noise level at a predefined evaluation position is predicted, based on the noise propagation model, the wind turbine models and information regarding the current operating conditions. The predefined evaluation position may be a position within the wind farm 1, or it may be a position outside, but in the vicinity of, the wind farm 1.

[0060] In the case that the predicted noise level exceeds a predefined threshold noise value, it will be required to reduce the noise level at the predefined evaluation position. This is done by the reducing the noise generated by one or more of the noise generators, i.e. by one or more of the wind turbines 2. To this end, one or more wind turbines 2 must be selected in an appropriate manner, and it must be determined how to change the operation of the selected one or more wind turbines 2 in order to obtain the desired reduction in the noise level at the predefined evaluation position.

[0061] This is done by performing an optimisation process, using the noise propagation model and the wind turbine models. The optimisation process has the predefined threshold noise value as a constraint, the noise generation and the power production of the wind turbines 2 as optimisation variables, and the total power production of the wind farm 1 as an optimisation target.

[0062] Accordingly, the one or more wind turbines 2, as well as the manner in which their operation should be changed, are selected in such a manner that it is ensured that the noise level at the predefined evaluation position remains below the predefined threshold noise value, while the total power production of the wind farm is maximised, i.e. the total penalty, in terms of reduction in total power production, for obtaining the desired reduction in noise level, is minimised.

[0063] FIG. 2 is a diagram illustrating a method for controlling noise generated by a wind farm 1 according to an embodiment of the invention. At a remote computation facility 6, a noise propagation model 7 and wind turbine models 8 are trained in the following manner. Site specific data related to the wind farm 1, in the form of terrain, elevation and roughness data, configuration of the wind farm 1 and of the wind turbines 2 of the wind farm 1, and a weather model are applied as training data input 9, along with operational specifications of the wind turbines 2, such as noise specifications, power curves, load specifications, etc. Furthermore, a blockage and wake model related to the site of the wind farm 1 is provided.

[0064] The noise propagation model 7 and the wind turbine models 8 are then trained, based on the provided training input data 9, and by means of a deep reinforcement learning configuration. The models 7, 8 are validated and certified, and are then supplied to a Reinforcement Learning (RL) agent 10 forming part of a control system 11 of the wind farm 1.

[0065] Real time data regarding operation of the wind turbines 2 and meteorological conditions is collected and supplied to the trained models 7, 8. A noise level at a predefined evaluation position is then predicted, using the trained models 7, 8, and based on the collected data. The predicted noise level is compared to a predefined threshold noise value, and in the case that the predicted noise level exceeds the predefined threshold noise value, the RL agent 10 initiates an optimisation process.

[0066] In the optimisation process, the RL agent 10 applies the trained models 7, 8 in order to select one or more wind turbines 2 and change the operation of the one or more selected wind turbines 2 in such a manner that the noise level at the predefined evaluation position is reduced to a level below the predefined threshold noise value, while the total power production of the wind farm 1 is optimised.

[0067] The RL agent 10 then dispatches control signals to the wind turbines 2 in order to cause the selected wind turbines 2 to change their operation in the specified manner, thereby causing the desired reduction in noise level at the predefined evaluation position.

[0068] FIG. 3 is a flow chart illustrating a method for controlling noise generated by a wind farm according to an embodiment of the invention.

[0069] The process is started at step 12. At step 13 a noise propagation model related to the wind farm and wind turbine models for the wind turbines of the wind farm are trained, e.g. by means of artificial intelligence (AI), such as reinforced learning. The noise propagation model includes the wind turbines of the wind farm as noise generators, and it takes interactions among the wind turbines into account.

[0070] At step 14, data regarding operating conditions prevailing at the wind farm is obtained, during operation of the wind farm. The data may include data related to the wind turbines, such as power production and/or local meteorological conditions, e.g. measured by appropriate sensors arranged in or near the wind turbines. Furthermore, the data may include meteorological data related to the wind farm, such as wind speed, wind direction, turbulence conditions, precipitation, air density, temperature, etc.

[0071] At step 15, a noise level at a predefined evaluation position is predicted. The predefined evaluation position may be a position within the wind farm, or it may be position outside, but in the vicinity of, the wind farm. The noise level is predicted based on the noise propagation model and the wind turbine models, and based on the obtained data regarding operating conditions prevailing at the wind farm. Thus, the noise level is predicted based on knowledge regarding how the wind turbines generate noise under the prevailing operating conditions, how the operation of the wind turbines affect each other under the prevailing operating conditions, and how the noise generated by the wind turbines propagates across the wind farm and in the vicinity of the wind farm under the prevailing operating conditions.

[0072] At step 16, the predicted noise level at the predefined evaluation position is compared to a predefined threshold noise value. In the case that the predicted noise level is below the predefined threshold noise value, it can be assumed that the noise generated by the wind farm fulfils noise requirements at the predefined evaluation position. Therefore, when this is the case, the wind turbines of the wind farm are all allowed to continue operating without additional noise constraints. Accordingly, the process is returned to step 14 for continued collection of data regarding the prevailing operating conditions.

[0073] In the case that step 16 reveals that the predicted noise level at the predefined evaluation position exceeds the predefined threshold noise value, this may be an indication that there is a risk that noise requirements at the predefined evaluation position may not be fulfilled. In this case it will therefore be required to reduce the noise generated by the wind turbines of the wind farm in order to reduce the noise level at the predefined evaluation position to a level below the predefined threshold noise value.

[0074] Accordingly, when this is the case, the process is forwarded to step 17, where one or more wind turbines are selected, and a change in operation of each of the selected wind turbines is also selected. This is done by means of the noise propagation model and the wind turbine models, and by performing an optimisation process. The optimisation process has the predefined threshold noise value at the predefined evaluation position as a constraint, the noise generation and the power production of the wind turbines as optimisation variables, and the total power production of the wind farm as an optimisation target. Thus, the optimisation process adjusts the noise generation and the power production of the wind turbines, in order to reduce the noise level at the predefined evaluation position to a level below the predefined threshold noise value, while maximising the total power production. Accordingly, the desired reduction in noise level is obtained with minimal penalty, in terms of reduced power production.

[0075] At step 18, new control settings are applied to the selected wind turbines, in order to change the operation of the selected wind turbines in accordance with the selected changes in operation. Thereby the desired reduction in noise level is actually obtained.