METHOD FOR PREDICTING THE PROBABILITY OF THROUGHING AND / OR SHUTDOWN A WIND TURBINE BASED ON A PREDICTED ICE EFFECT

Description

TECHNICAL FIELD

[0001] Embodiments of the present disclosure relate to a method for creating a prediction model of ice accretion on wind turbines and the use of the method for creating a yield forecast. STATE OF THE ART

[0002] Wind turbine rotor blades are exposed to the surrounding weather conditions without protection. In certain locations, ice can build up on the rotor blades if the ambient temperature is sufficiently low and the humidity is sufficiently high, or if precipitation occurs. As wind turbine rotor blades grow larger, their surface area increases, thus increasing the risk of ice buildup, i.e., the formation of ice deposits on the rotor blades.

[0003] On the one hand, ice deposits pose a potential hazard to the area surrounding the wind turbine, as the ejected ice pieces can endanger people and objects within the ejection radius if they are released – whether during operation or when the turbine is shut down. On the other hand, uneven ice deposits, particularly, can cause an imbalance in the wind turbine's rotor, which can lead to damage during operation. Furthermore, persistent ice deposits can lead to the shutdown of the entire turbine. This is usually accompanied by yield losses and economic disadvantages.

[0004] It is known to evaluate data from a wind turbine to determine the risk of ice buildup. DE 10 2005 016 524 A1 discloses a method for detecting ice on a wind turbine, which involves monitoring both meteorological conditions related to icing conditions and one or more physical, characteristic variables of the wind turbine during operation that indicate a change in the mass of the wind turbine's rotor blades.

[0005] Furthermore, current yield forecasts are usually based on weather models. The yield of a wind turbine is determined based on expected wind speed, wind direction, temperature, humidity, etc., along with relevant turbine parameters (especially the power curve). Historical data is sometimes also used to determine a more accurate model for a wind turbine. Furthermore, planned maintenance work can be included in the yield forecast. Dedicated use of icing forecasts is generally not included.

[0006] US 2012 / 0226485 A1 describes a method for predicting the probability of ice formation or growth on wind turbine rotor blades. The method utilizes historical measurements of meteorological data such as wind speed, temperature, and relative humidity.

[0007] US2014 / 0244188 A1 describes a method and apparatus for predicting the performance of a wind turbine in a wind farm, wherein a corrected data set based on environmental data collected by a sensor in a wind farm is used to improve a weather forecast model.

[0008] CN 103 603 776 B discloses a method for predicting ice accretion on wind turbines, which is based on a machine learning method and uses both current and historical meteorological data and current and historical ice accretion data.

[0009] US 2017 / 058871 A1 discloses a method for preventing premature shutdown of a wind turbine due to ice buildup. Upon detection of ice buildup, the turbine is throttled back.

[0010] In wind farms, especially in winter, icing occurs frequently, for example, on the rotor blades of individual wind turbines. Often, wind turbines have to be shut down as a result. This leads to revenue losses for wind farm operators due to reduced turbine availability. SUMMARY OF THE INVENTION

[0011] Embodiments of the present disclosure provide a method of predicting ice accretion on wind turbines, as well as the use of the method to create a yield forecast.

[0012] According to one embodiment, a method for predicting ice buildup on wind turbines is specified, comprising: capturing current ice buildup data on at least one wind turbine, wherein the current ice buildup data is determined by means of at least one vibration sensor or at least one acceleration sensor; using current and historical meteorological data for the location of the at least one wind turbine; using the current ice buildup data and historical ice buildup data and predicting ice buildup based on a machine-learned model, wherein the prediction of ice buildup includes an uncertainty; and predicting a probability of throttling and / or shutting down the at least one wind turbine, wherein the probability of throttling and / or shutting down the at least one wind turbine is based on the ice buildup predicted for the at least one wind turbine.wherein the step of predicting the probability of throttling and / or shutting down the at least one wind turbine comprises throttling the at least one wind turbine when the predicted ice accretion on the at least one wind turbine reaches a threshold value SD, shutting down the at least one wind turbine when the predicted ice accretion on the at least one wind turbine reaches a threshold value SA, wherein the threshold value SA is greater than the threshold value SD, and including further parameters in predicting the probability of throttling and / or shutting down the at least one wind turbine, wherein the further parameters include material specifications of the rotor blade.

[0013] According to further embodiments, a probability of throttling and / or shutdown is calculated based on the prediction of ice buildup and on site-specific conditions. According to further embodiments, the machine-learned model is created according to a method for creating a prediction model of ice buildup on wind turbines, which method comprises the following: collecting current ice buildup data on at least one wind turbine, wherein the current ice buildup data is determined using at least one vibration sensor or at least one acceleration sensor; using current and historical meteorological data for the location of the at least one wind turbine, wherein the current and historical meteorological data comprise at least two of the following: wind speed, wind direction, temperature, air humidity, air pressure, and / or other weather-related parameters;Feeding the current ice accretion data and historical ice accretion data of the at least one wind turbine and feeding the current and historical meteorological data into a machine learning method; and creating a prediction model based on the machine learning method. According to further embodiments, the meteorological data contain predicted meteorological data. According to further embodiments, an icing severity is predicted using the prediction.

[0014] According to the invention, a use of the methods described herein for creating a yield forecast for at least one wind turbine, in particular for at least one wind turbine of a wind farm, is provided. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Examples of embodiments are illustrated in the drawings and explained in more detail in the following description. The drawings show: Fig. 1A shows schematically, by way of example, three wind farms according to the embodiments described here; Fig. 1B shows schematically a wind farm with, by way of example, three wind turbines according to the embodiments described here; Fig. 2A shows schematically a part of a wind turbine with rotor blades and sensors according to embodiments described here; Fig. 2B schematically shows a rotor blade of a wind turbine with a sensor according to embodiments described here; Fig. 3 shows a flowchart of a first method according to embodiments described here; Fig. 4 shows a flowchart of a second method according to embodiments described herein; Fig. 5A shows a flowchart of a method according to embodiments described herein; Fig. 5B shows a flowchart of a method according to embodiments described herein; Fig. 6 schematically shows an optical fiber with a fiber Bragg grating for use in sensors according to embodiments described herein; Fig. 7 shows schematically a measurement setup for a fiber optic sensor according to embodiments described here or for monitoring methods according to embodiments described here.

[0016] In the drawings, like reference symbols indicate like or functionally equivalent components or steps. WAYS OF IMPLEMENTING THE INVENTION

[0017] Reference will now be made in more detail to various embodiments of the invention, one or more examples of which are illustrated in the drawings.

[0018] The present invention is used to predict icing on wind turbines of one or more wind farms and to forecast yield losses due to icing. Ice formation on the blades of a wind turbine can result in the turbine having to be shut down or producing reduced yield. This can depend on the location (e.g., regulatory requirements often require shutdown at minimal ice buildup) and the severity of the icing (in the case of heavy ice buildup, the turbine must be throttled down or stopped to avoid severe loads). Predicting ice buildup enables a more accurate yield forecast, with all the resulting benefits.

[0019] Fig. 1A shows schematically, by way of example, three wind farms 10 according to the embodiments described here. According to the embodiments described, a method can relate to at least one or more wind farms 10, in particular to ten or more wind farms, such as twenty or more wind farms. The data of the wind farms can be documented centrally (dashed lines in Fig. 1A ).

[0020] Fig. 1B shows a wind farm 10 consisting of three wind turbines 200. The wind turbines 200 can, as in Fig. 1B represented by dashed lines, are interconnected. The interconnection can enable joint monitoring, control, and / or regulation of the wind turbines 200. In addition, the wind turbines 200 can also be individually monitored, controlled, and / or regulated. According to embodiments described here, a wind farm can include one or more wind turbines, in particular five or more wind turbines, such as ten or more wind turbines.

[0021] Fig. 2A shows, by way of example, a wind turbine 200 of a wind farm in which the method described herein can be used. The wind turbine 200 includes a tower 40 and a nacelle 42. The rotor is attached to the nacelle 42. The rotor includes a hub 44 to which the rotor blades 100 are attached. According to typical embodiments, the rotor has at least two rotor blades, in particular three rotor blades. During operation of the wind turbine, the rotor, ie the hub with the rotor blades, rotates around an axis. In the process, a generator is driven to generate electricity. As shown in Fig. 2A As shown, at least one sensor 110 is provided in a rotor blade 100. The sensor is connected to an evaluation unit 114 via a signal line. The evaluation unit 114 delivers a signal to a control and / or regulation system 50 of the wind turbine 200.

[0022] Ice accretion 1 is schematically depicted in the area of ​​the rotor blade tip of the rotor blades 100. Vibrations or accelerations, for example, are detected in the rotor blade by means of sensor 110, which, according to the described embodiments, can be configured as a vibration sensor or acceleration sensor. Sensor 110 can be, for example, electrical and / or fiber-optic sensors. The sensors can be configured to measure, for example, ice masses.

[0023] Fig. 2B shows a rotor blade 100 of a wind turbine. The rotor blade 100 has an axis 101 along its longitudinal extent. The length 105 of the rotor blade extends from the blade flange 102 to the blade tip 104. According to embodiments described here, a sensor 110 is located in an axial or radial region, i.e., a region along the axis 101.

[0024] It is often desirable to predict ice buildup on wind turbines and / or wind farms. This can, for example, enable yield forecasting. Yield forecasts can be continuously improved and refined through more accurate predictions. Furthermore, yield forecasts offer advantages in energy trading and the control of conventional power plants. Furthermore, better planning of maintenance operations is possible.

[0025] In the event of icing, ice can accumulate on the rotor blades, among other things. The accumulation of ice or ice accretion can be measured on turbines. The amount or volume of ice accretion can vary between individual turbines.

[0026] Ice accumulation, as used herein, refers to an increase in ice mass on a rotor blade over time. Furthermore, a decrease in ice mass may occur. An increase in ice mass may be a buildup or increase in ice mass, i.e., be positive, or it may be a decrease in ice mass, i.e., be negative. An icing event, as used herein, refers to the occurrence of ice accumulation on at least one wind turbine in a wind farm.

[0027] In the event of icing, the amount of ice that accumulates on the rotor blades of wind turbines over a given period of time can be determined. The resulting data can be referred to as ice accumulation data. Determining the amount of ice requires the measurement of a suitable variable. The measurement is performed using sensors 110. The variable can be converted into a system variable S by transformation. The system variable S is determined indirectly. For example, the natural vibration of a rotor blade can be measured.

[0028] The system variable S is further related to the mass of the rotor blade. Alternatively or additionally, the system variable S is related to the mass of ice accretion on the rotor blade. Typically, the system variable S is obtained from measurement data from vibration or acceleration measurements in or on one or more of the rotor blades. In embodiments, the measurement variable is measured over the course of a detection period, preferably by measuring vibrations or accelerations over the course of the detection period T. The measurement is performed on or in the rotor blade.

[0029] The system variable S is derived from the measurement data, preferably by natural frequency analysis from the measurement data of the vibration or acceleration measurements. For example, a measuring system which determines the current ice buildup on the rotor blades of a wind turbine can be installed on several wind turbines (e.g. within one or more wind farms). The system variable S allows conclusions to be drawn about the mass of the respective rotor blade(s) and / or the ice buildup mass of the respective rotor blade(s). In some embodiments, the system variable S is proportional to the total mass of the rotor blade and / or proportional to an additional mass coating of the rotor blade. An additional mass coating of the rotor blade is a mass coating that is added to the dead mass of the rotor blade. Typically, the system variable is proportional to the ice mass.

[0030] The ice accumulation on a rotor blade is measured during the recording period T by sensors 110. The measurement data can be recorded continuously. The ice patterns of all wind turbines equipped with measuring systems, as well as current and forecast meteorological data for all wind turbine locations, for example, and other data on current and future weather conditions, can be recorded centrally. The recording period T is a measurement period that continuously increases over time.

[0031] A forecast period Pt is a period of time for which a forecast is made. For example, ice buildup can be forecast. The forecast period Pt can include or at least overlap the measurement period. Typically, the forecast period Pt is the measurement period plus an additional day (24 hours) into the future. This can mean that a measured value of the measured variable can be recorded at any time during a forecast. This can also enable a constant comparison of the forecast values ​​and the actually measured values. Recording a measured value for a forecast for a specific time can occur later than the forecast itself.

[0032] Historical data, as used herein, refers to data that was recorded in a past period at a time t xk , before a specific time tx or t x+n . Historical data is recorded and stored so that it can also be made available at a later time. Historical data can be data on ice formation. Historical data can be meteorological data. Historical data can be available for several wind farms, in particular for each wind farm, and furthermore for several wind turbines within a wind farm, e.g. one wind farm can be available for each wind turbine.

[0033] Current data, as used herein, refers to the respective data that represent the recorded data at a specific observation time tx. Current data can be the most recently recorded data at the observation time tx. Current data can be data on ice formation. Current data can be meteorological data. Current data can be available for several wind farms, in particular for each wind farm, and furthermore for several wind turbines within a wind farm, e.g., one wind farm can be available for each wind turbine.

[0034] Predicted data, as used herein, refers to data resulting from any forecast model or prediction model. This could be, for example, predicted meteorological data or predicted ice formation. Predicted data can relate to multiple wind farms, in particular, to each wind farm, and furthermore, to multiple wind turbines within a wind farm, for example, to each wind turbine within a wind farm. The predicted data can be regionally location-specific.

[0035] Meteorological data, as used herein, may include data on wind speed, wind direction, temperature, humidity, air pressure, and / or other weather-related parameters. Meteorological data may relate to one or more wind farm locations. Meteorological data may, in particular, include the general weather situation at one or more locations. Meteorological data may be obtained from a commercial service, in particular forecast meteorological data. Meteorological data may also be collected at a wind farm weather station.

[0036] Fig. 3 shows a flowchart of a first method 300 according to embodiments described herein.

[0037] In a step 310 of the first method, ice accretion data is acquired on at least one wind turbine. According to embodiments described here, an ice accretion measurement can be performed on one or more wind turbines. The measurement can be performed on all wind turbines of a wind farm. The measurement can be performed on turbines of one or more wind farms. The measurement can be performed indirectly. The measurement can be performed continuously. The acquired data can include historical data. The acquired data can include current data.

[0038] Furthermore, meteorological data from the wind farm in which the at least one wind turbine is located is used. Meteorological data from multiple wind farms can be used, particularly if data on ice accretion on wind turbines from these multiple wind farms is collected. The meteorological data used can include historical data. The meteorological data used can include current data.

[0039] In a step 320 of the first method, the acquired ice accretion data is fed into a machine learning process. The fed-in ice accretion data can be data from one wind turbine. It can also be data from multiple wind turbines. It can also be data from all wind turbines. Historical ice accretion data can be fed in. Current ice accretion data can be fed in.

[0040] The meteorological data used is also fed into the machine learning process. For example, all data collected over a longer period of time is used in a machine learning process. The meteorological data fed in can be data from one wind farm. It can also be data from multiple wind farms. Historical meteorological data can be fed in. Current meteorological data can be fed in. Forecasted meteorological data can be fed in. All collected and used data from all wind turbines in all wind farms can be fed into the machine learning process.

[0041] A machine learning method, as used herein, is an artificial system that learns from examples and generalizes these learning phases upon completion. This implies the recognition of patterns and regularities in the fed or input data. The practical implementation is achieved using algorithms. In particular, it is a supervised learning method. In a supervised learning method, a "teacher" specifies the correct or best corresponding output pattern(s) to the input pattern(s). Thus, after repeated presentation of the corresponding input and output patterns, the model can independently make this association and generalize, i.e., also do this for unknown, similar input patterns.

[0042] In a step 330, a prediction model is created. The prediction model is based on the machine learning method. The prediction model is based, for example, on a supervised learning method. The model can be a continuously improving, probabilistic model. The prediction model can, for example, predict the future icing severity for a wind turbine from the input data. For example, historical ice accretion measurements and meteorological data are used to learn a probabilistic model for predicting ice accretion on all turbines. This model can be used to determine the future ice accretion on all wind turbines under consideration based on current and past ice accretion measurements and current, past, and predicted meteorological data.

[0043] A forecast model or prediction model, as used herein, can be, for example, a numerical data-based forecast model. The prediction model is based on the machine learning method. With the help of the prediction model, the future ice formation on a wind turbine can be predicted from the input data, for example. Using the specific model, the future ice formation can be predicted from current and historical data (including meteorological forecasts). Since measurement data on the actual ice formation on the wind turbines is available, a supervised learning method can be used. The resulting probabilistic model allows, for example, a prediction of the icing severity and uncertainty for each wind turbine.

[0044] Fig. 4 shows a flowchart of a second method 400 according to the invention.

[0045] In a step 410 of the second method, ice accretion data is recorded on at least one wind turbine. According to the invention, ice accretion is measured on one or more wind turbines. The measurement can be performed on all wind turbines of a wind farm. The measurement can be performed on turbines of one or more wind farms. The measurement can be performed indirectly. The measurement can be performed continuously. The recorded data includes historical data. The recorded data includes current data.

[0046] Furthermore, meteorological data from the wind farm in which the at least one wind turbine is located is used. Meteorological data from multiple wind farms can be used, particularly if data on ice accretion on wind turbines from these multiple wind farms is collected. The meteorological data used includes historical data. The meteorological data used includes current data.

[0047] In a step 420, ice accretion is predicted. For example, the icing strength and / or the mass of expected ice can be predicted. The prediction is based on a prediction model. For example, the prediction of ice accretion can be based on the prediction model that is based on Fig. 3 Ice accretion can be predicted for a single location. The location can include multiple wind farms. Ice accretion can also be predicted for one wind farm. Ice accretion can also be predicted for multiple wind turbines within a wind farm. Ice accretion can be predicted for one wind turbine within a wind farm.

[0048] The prediction contains uncertainty. This uncertainty is a non-negative parameter that characterizes the variance of the values ​​associated with the prediction based on the information used.

[0049] In step 430, the probability of throttling and / or shutting down a wind turbine is calculated. Depending on the turbine location and the resulting conditions, the probability of stopping or throttling the turbine is determined. This depends, for example, on regulatory requirements, the turbine type, and the expected severity of icing. The probability that a turbine can be stopped or throttled can be used to adjust a yield forecast based on meteorological data accordingly.

[0050] The throttling and / or shutdown is based on the predicted ice buildup. A wind turbine is throttled back when ice buildup on a turbine reaches a threshold value SD. Furthermore, a wind turbine is shut down when ice buildup on a turbine reaches a threshold value SA.

[0051] The threshold value SD as used herein is a threshold value for the amount of accumulated ice that is suitable for the system size S and at which the turbine that reaches the threshold value is throttled. The threshold value SD is, for example, a defined limit for the amount of accumulated ice, e.g., the mass and / or volume of accumulated ice, above which the wind turbine is throttled. Throttling describes the negative acceleration of the rotor blades to a lower speed. The threshold value SD can, for example, be determined empirically. Throttling of rotor blades must occur, for example, if faster rotation of the rotor blades would lead to damage.

[0052] Furthermore, a condition for shutting down a wind turbine due to an amount of ice accumulating that impairs operation is defined by a threshold value SA that is suitable for the system size S. The threshold value SA is, for example, a defined limit for the amount of ice accumulating, e.g. the mass and / or volume of ice accumulating, above which an affected wind turbine is shut down. The threshold value SA is greater than the threshold value SD. A previously throttled turbine can subsequently reach the threshold value SA. The previously throttled turbine is then shut down. The threshold value SA can, for example, be determined empirically. The shutdown of a turbine must occur, for example, if continuing in a throttled or normal state of the wind turbine would lead to significant damage and / or safety problems.

[0053] The probability of future throttling and / or shutdown for one or more wind turbines depends, for example, on the ice buildup predicted for the turbines. Ice buildup can also be predicted for the wind farm.

[0054] In a step 490, further parameters or framework conditions influencing the system are recorded. Such parameters include material specifications of the rotor blade and can further include, for example, the location of the turbine and / or official regulations. The parameters can be incorporated into the system in the form of threshold values. These parameters are included in the calculation of the probability of throttling and / or shutting down the turbine. The probability of throttling and / or shutting down the turbine can be specified in this way, for example. For example, an official regulation could require the shutdown of a turbine if the icing severity is lower than the material specifications of the rotor blade.

[0055] Fig. 5A shows a flowchart of a method 500 according to embodiments described herein. Fig. 5B shows a flowchart of a method 500-1 according to embodiments described herein.

[0056] In a step 510 of the method, ice accretion data is recorded on at least one wind turbine. According to embodiments described here, an ice accretion measurement can be performed on one or more wind turbines. For example, the measurement of ice accretion is performed on turbine 1 according to step 512-1. The measurement can be performed on n wind turbines according to step 512-2. "n" is the number of wind turbines. For example, "n" wind turbines of a wind farm or different wind farms can be included. Historical and current data for ice accretion of all wind turbines can be obtained according to step 512-3. A measurement can be performed on all wind turbines of a wind farm. The measurement can be performed on turbines of one or more wind farms. The measurement can be performed indirectly. The measurement can be performed continuously. The recorded data can include historical data. The recorded data can include current data.

[0057] Furthermore, meteorological data from the wind farm in which the at least one wind turbine is located is used. Meteorological data from multiple wind farms can be used, in particular if data on ice accretion on wind turbines of these multiple wind farms is recorded. Meteorological data relating to wind turbine 1, as shown in step 514-1, can be used. Meteorological data relating to n wind turbines, as shown in step 514-2, can be used. The data used can include historical data. The data used can include current data. The data used can include predicted meteorological data.

[0058] In a step 520, the collected and used data is fed into a machine learning process. According to embodiments, the input data can be both ice accretion data and meteorological data. Both data sets can contain historical, current, and predicted data. The machine learning process can, for example, be the learning process from a step 320 of the method 300. A continuously improving probabilistic model can be learned, such as according to step 532.

[0059] In step 530, a prediction model is created. The prediction model is based on the machine learning method. The learning method is a supervised learning method. Current measured data can serve as a "teacher" to input correct output values ​​for the input values. This can be ice accretion data, meteorological data, or both. Continuous improvement of the model can, for example, mean comparing previously predicted data with the data actually measured at the relevant point in the future.

[0060] In a step 540, ice accumulation is predicted using a machine-learned model. The ice accumulation can be predicted for all wind turbines according to step 542. The ice accumulation can be predicted for one wind farm. The ice accumulation can be predicted for multiple wind farms. An amount of ice accumulation can be predicted; this can be, for example, the mass and / or the volume of the ice. The predicted amount of ice can be integrated in step 520. This can be done, for example, in step 542. As described above, the predicted ice accumulation can contain an uncertainty. According to step 544-1, ice accumulation for wind turbine 1 can already be predicted. According to a step 544-2, ice accumulation for n wind turbines can be predicted.

[0061] In a step 550, the probability of throttling / shutting down the wind turbine is determined. For example, a probability of throttling / shutting down for wind turbine 1 can be determined according to a step 552-1. Additionally, according to a step 552-2, the probability of throttling / shutting down for n wind turbines can be determined. The throttling and / or shutting down of the wind turbines can be based on the current ice buildup. Furthermore, the throttling and / or shutting down can be based on the predicted ice buildup. The throttling and / or shutting down can be based on both.

[0062] In step 560-1, the probability of throttling / shutdown for wind turbine 1 is obtained. Additionally, in step 560-2, the probability of throttling / shutdown for n wind turbines is obtained. From these probabilities, a yield forecast can be made. An accurate yield forecast is important for predicting the performance of a wind turbine, a wind farm, and / or multiple wind farms.

[0063] The probability calculation from step 550 can be influenced by additional external parameters. From these, threshold values ​​for, for example, ice amounts can be derived. Additional parameters can be, for example, official regulations. According to a step 590-1, threshold values ​​can exist that influence the throttling / shutdown of wind turbine 1. According to a step 590-2, threshold values ​​can also exist that influence the throttling / shutdown of n wind turbines. These parameters can be incorporated into the learning process. Furthermore, these parameters can be incorporated into the prediction of ice buildup.

[0064] As already indicated in step 520, the predicted as well as the currently measured ice accretion data can in turn be fed into the supervised learning process via a feedback 542.

[0065] The calculated probability of throttling and / or shutting down the wind turbine can be used to create a yield forecast. The yield forecast can apply to multiple wind farms, in particular, it can apply to a single wind farm. Furthermore, a yield forecast can be valid for multiple wind turbines, in particular, for a single wind turbine. A yield forecast can be continuously improved and specified through the predictions.

[0066] In comparison to yield forecasts based solely on meteorological data, ice buildup can be explicitly included in the analysis. This enables a significantly more accurate forecast of downtimes due to ice buildup on wind turbines. Compared to ice forecasts based solely on meteorological data, there is an advantage in measuring the actual icing on the wind turbines. On the one hand, this allows for a continuous improvement of the forecast. On the other hand, the current icing condition is precisely known, which allows for a more accurate forecast, since future changes in ice buildup can depend heavily on the current icing condition of the rotor blades. Furthermore, this method can be used to determine the relationship in ice formation between different turbines. For example,Within a wind farm, there are turbines that always ice up first because they are more exposed, or turbines in one wind farm ice up before turbines in another farm ice up (depending on the general weather situation).

[0067] Fig. 6 schematically shows an optical fiber with a fiber Bragg grating for use in sensors according to embodiments described herein.

[0068] Fig. 6 shows a sensor integrated into an optical waveguide or a fiber optic sensor 610, which has a fiber Bragg grating 606. Although in Fig. 5 only a single fiber Bragg grating 606 is shown, it is to be understood that the present invention is not limited to data acquisition from a single fiber Bragg grating 606, but that a plurality of fiber Bragg gratings 606 may be arranged along an optical fiber 212, a transmission fiber, a sensor fiber, or an optical fiber.

[0069] Fig. 6 thus shows only a section of an optical waveguide, which is designed as a sensor fiber, optical fiber, or light guide 212, whereby this sensor fiber is sensitive to fiber strain (see arrow 608). It should be noted here that the term "optical" or "light" refers to a wavelength range in the electromagnetic spectrum, which can extend from the ultraviolet spectral range through the visible spectral range to the infrared spectral range. A center wavelength of the fiber Bragg grating 606, i.e., a so-called Bragg wavelength λB, is obtained by the following equation: λ B = 2 ⋅ nk ⋅ Λ .

[0070] Here, nk is the effective refractive index of the fundamental mode of the core of the optical fiber and A is the spatial grating period (modulation period) of the fiber Bragg grating 606.

[0071] A spectral width, which is given by a half-width of the reflection response, depends on the extension of the fiber Bragg grating 606 along the sensor fiber. The light propagation within the sensor fiber or the light guide 212 is thus dependent, for example, on the effect of the fiber Bragg grating 606, on forces, moments, mechanical stresses, and temperatures applied to the sensor fiber, i.e., the optical fiber and, in particular, the fiber Bragg grating 606 within the sensor fiber.

[0072] As in Fig. 6 As shown, electromagnetic radiation 14 or primary light enters the optical fiber or light guide 212 from the left, with a portion of the electromagnetic radiation 14 exiting as transmitted light 16 with a modified wavelength distribution compared to the electromagnetic radiation 14. Furthermore, it is possible to receive reflected light 15 at the input end of the fiber (i.e., at the end at which the electromagnetic radiation 14 is also radiated in), with the reflected light 15 likewise having a modified wavelength distribution. The optical signal used for detection and evaluation can, according to the embodiments described here, be provided by the reflected light, by the transmitted light, or by a combination of the two.

[0073] In a case where the electromagnetic radiation 14 or the primary light is radiated across a broad spectral range, a transmission minimum results in the transmitted light 16 at the Bragg wavelength. A reflection maximum results in the reflected light at this point. Detecting and evaluating the intensities of the transmission minimum or the reflection maximum, or intensities in corresponding wavelength ranges, generates a signal that can be evaluated with regard to the change in length of the optical fiber or light guide 212 and thus provides information about forces or vibrations.

[0074] Fig. 7 shows a typical measurement system for evaluating fiber optic sensors, including acceleration sensors. The system has a source 702 for electromagnetic radiation, for example, a primary light source. The source serves to provide optical radiation with which at least one fiber optic sensor element of a sensor, for example, an acceleration sensor, can be irradiated. For this purpose, an optical transmission fiber or light guide 703 is provided between the primary light source 702 and a first fiber coupler 704. The fiber coupler couples the primary light into the optical fiber or light guide 212. The source 702 can be, for example, a broadband light source, a laser, an LED (light emitting diode), an SLD (superluminescent diode), an ASE light source (amplified spontaneous emission light source), or an SOA (semiconductor optical amplifier).For the embodiments described here, several sources of the same or different types (see above) can also be used.

[0075] The fiber optic sensor element 710, such as a fiber Bragg grating (FBG) or an optical resonator, is integrated into a sensor fiber or optically coupled to the sensor fiber. The light reflected by the fiber optic sensor elements is in turn guided via the fiber coupler 704, which guides the light via the transmission fiber 705 to a beam splitter 706. The beam splitter 706 splits the reflected light for detection by a first detector 707 and a second detector 708. The signal detected by the second detector 708 is first filtered with an optical edge filter 709.

[0076] The edge filter 709 can detect a shift of the Bragg wavelength at the FBG or a wavelength change through the optical resonator. In general, a measurement system as described in Fig. 6 As shown, the acceleration sensor can be provided without the beam splitter 706 or the detector 707. However, the detector 707 enables normalization of the acceleration sensor's measurement signal with respect to other intensity fluctuations, such as fluctuations in the intensity of the source 702, fluctuations due to reflections at interfaces between individual optical fibers, or other intensity fluctuations. This normalization improves measurement accuracy and reduces the dependence of measurement systems on the length of the optical fibers provided between the evaluation unit and the fiber optic sensor.

[0077] Particularly when using multiple FBGs, additional optical filter devices (not shown) can be used to filter the optical signal or secondary light. An optical filter device 709 or additional optical filter devices can comprise an optical filter selected from the group consisting of a thin-film filter, a fiber Bragg grating, an LPG, an arrayed waveguide grating (AWG), an echelle grating, a grating array, a prism, an interferometer, and any combination thereof.

Claims

1. A method for forecasting ice buildup on wind turbines, comprising: recording current ice buildup data on at least one wind turbine, the current ice buildup data being determined by means of at least one vibration sensor or at least one acceleration sensor; using current and historical meteorological data for the location of at least one wind turbine; using current ice buildup data and historical ice buildup data; forecasting ice buildup based on a machine-learned model, where the forecasting of ice buildup includes uncertainty; characterized in that the method further comprises the steps: forecasting a probability of throttling and / or shutdown of the at least one wind turbine, wherein the probability of throttling and / or shutdown of the at least one wind turbine is based on the forecast ice buildup for the at least one wind turbine, wherein, in the step of forecasting the probability of throttling and / or shutdown of the at least one wind turbine, the at least one wind turbine is throttled when the forecast ice buildup on the at least one wind turbine reaches a threshold value SD, the at least one wind turbine is switched off when the forecast ice buildup on the at least one wind turbine reaches a threshold value SA, the threshold value SA being greater than the threshold value SD, and further parameters are included in the forecast of the probability of throttling and / or shutdown of the at least one wind turbine, the further parameters including material specifications of the rotor blade.

2. The method according to claim 1, wherein a probability of throttling and / or shutdown is calculated based on the prediction of ice buildup and on site-specific basic conditions.

3. The method according to claim 1, wherein the machine-learned model is created according to a method for creating a forecasting model of ice buildup on wind turbines, which method comprises the following: recording current ice buildup data on at least one wind turbine, the current ice buildup data being determined by means of at least one vibration sensor or at least one acceleration sensor; using current and historical meteorological data for the location of the at least one wind turbine, the current and historical meteorological data including at least two of the following: wind speed, wind direction, temperature, humidity, air pressure, and / or other weather-related parameters; feeding the current ice buildup data and historical ice buildup data of at least one wind turbine and feeding the current and historical meteorological data into a machine learning process; and creating a forecasting model based on the machine learning process.

4. The method according to any one of claims 1 to 3, wherein the meteorological data include forecast meteorological data.

5. The method according to claim 1, wherein an icing severity is predicted with the forecast.

6. A use of the method according to any one of claims 1 to 5 for preparing a yield forecast for wind farms, in particular for at least one wind turbine.