A wind turbine yaw control method for coping with wind direction fluctuation

Abstract

This invention relates to a yaw control method for wind turbines to cope with wind direction fluctuations, comprising the following steps: S1. Acquiring historical SCADA data of the wind farm and filtering the wind direction data to obtain a wind direction sequence; S2. Constructing a short-term wind direction prediction model; S3. Determining the wind direction stability P based on the filtered wind direction sequence. st Based on the average wind speed and wind direction stability P within a set time period st S4. Based on the real-time average wind speed and real-time wind direction stability P, the current ideal yaw speed v is obtained from the yaw speed training model; st By adjusting the control parameter threshold, the yaw control of the wind turbine is achieved. This invention overcomes the adverse effects of wind direction fluctuations on system performance. By adjusting the threshold of the yaw system in real time based on wind direction stability and wind speed level, the frequent start-up and shutdown of the yaw system under low and excessively high wind speeds are avoided.

Description

Technical Field

[0001] This invention relates to a yaw control method for wind turbines that addresses wind direction fluctuations, belonging to the field of wind power generation technology. Background Technology

[0002] With the gradual depletion of fossil fuels, the demand for renewable and clean energy has increased significantly. Wind power, as a renewable energy source, has received widespread attention in recent years, and countries around the world have continuously strengthened the development and utilization of wind energy, leading to considerable progress in wind power technology. However, with the saturation of onshore wind turbine installations, improving the utilization rate of wind energy and optimizing the control methods of wind turbine units have become the mainstream research directions.

[0003] The yaw system is a crucial component of wind turbines. When the wind speed vector changes to a certain extent, the yaw system drives the wind turbine to align with the wind. However, wind is unpredictable, and wind direction fluctuates frequently. Therefore, the ability of the yaw system to align with the wind in a timely manner and at an appropriate yaw speed becomes a key factor in improving power generation and reducing turbine load. Currently, Model Predictive Controllers (MPCs) are commonly used controllers for wind turbines. However, these controllers require a large amount of computation and computing resources in the wind power field, placing high demands on computing resources and hardware equipment, thus increasing implementation costs. Furthermore, many wind farms use simple dead-zone control strategies for their yaw systems, which cannot specifically address changes in the local environment of the wind turbine, and the yaw control threshold parameters are constant, failing to cope with fluctuations in wind direction. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a yaw control method for wind turbines that can be used to predict the short-term optimal yaw speed by using real-time wind speed and wind direction stability.

[0005] The present invention adopts the following technical solution:

[0006] This invention provides a yaw control method for wind turbines to cope with wind direction fluctuations, comprising the following steps:

[0007] Step S1: Obtain the SCADA historical data of the wind farm, and filter the wind direction data in the historical data to obtain the wind direction sequence;

[0008] Step S2: Construct a short-term wind direction prediction model based on historical wind direction data;

[0009] Step S3: Determine the wind direction stability P based on the wind direction sequence obtained after the filtering process. st Based on the average wind speed and wind direction stability P within a set time period st By training the yaw speed model, the current ideal yaw speed v can be obtained;

[0010] Step S4: Based on the real-time average wind speed and real-time wind direction stability P st Adjust the control parameter threshold and use it as input to the model trained in step S3 to obtain the yaw speed v, and then perform yaw control on the wind turbine.

[0011] In step S2 of this invention, the short-term wind direction prediction model uses a combination of ARMA based on wavelet decomposition (WD) and Bayes-LSTM to predict short-term wind direction. The original wind direction data is preprocessed and decomposed by wavelet to obtain approximate trend components and high-frequency detail components. An ARMA algorithm is used to establish a prediction model for the trend components, and a Bayes-LSTM neural network is used to establish a prediction model for the detail components. The models are combined and reconstructed to obtain the final wind direction prediction value.

[0012] In step S3 of this invention, the fluctuation amplitude A and fluctuation duration T of the wind direction sequence are extracted to determine the wind direction stability P. st .

[0013] In step S3 of this invention, the wind direction sequence determines the wind direction stability P. st The specific steps are as follows:

[0014] First, calculate the marginal probability density distributions of the fluctuation amplitude A and the fluctuation duration T using a Weibull distribution with offset. The formula for the Weibull distribution with offset is as follows:

[0015]

[0016] In the formula, α is the scale parameter, β is the shape parameter, and m is the length of the displacement;

[0017] The joint probability density of fluctuation amplitude A and fluctuation duration T is expressed using joint copulas, and its formula is as follows:

[0018] C(u, v)=ω1C_{Gum bel}(u, v)+ω2C_{Clayton}(u, v)+ω3C_{Frank}(u, v)

[0019] In the formula, C_{Gumbel}(u, v), C_{Clayton}(u, v), and C_{Frank}(u, v) are three Archimedes Copula functions, ω1, ω2, and ω3 are the weights corresponding to each copula, and u and v are the marginal probability densities of fluctuation amplitude A and fluctuation duration T.

[0020] Wind direction stability P st Calculation method:

[0021] In the formula, EA is a predefined fluctuation amplitude threshold, ET is a predefined fluctuation time threshold, and f(x, y) is the joint probability density of (A, T).

[0022] In step S1 of this invention, the historical wind direction data filtering specifically employs the following methods:

[0023] For data filtering where the fluctuation amplitude A is less than 2° or the fluctuation duration T is less than 2s, a sliding window of 2s is set. The data is traversed, and the data within the sliding window is treated as an independent signal segment. Each signal segment is analyzed, and the difference between its maximum and minimum values ​​is checked. If it is less than 2°, the signal segment is considered to meet the fluctuation amplitude filtering condition. At the same time, it is checked whether the fluctuation duration within the signal segment is less than 2s. If the fluctuation duration is less than 2s, the signal segment meets the fluctuation duration filtering condition. If the signal segment meets either the fluctuation amplitude filtering condition or the fluctuation duration filtering condition, the original data is retained. Signal segments that do not meet either condition can be attenuated. The processed signal segments are then spliced ​​together to obtain the filtered wind direction sequence.

[0024] In step S3 of this invention, a model is constructed based on a BP neural network, and its parameters are optimized using a Bayesian optimization algorithm.

[0025] In step S2 of this invention, a combination of ARMA based on wavelet decomposition (WD) and Bayes-LSTM is used to predict short-term wind direction.

[0026] The db4 wavelet is used to perform a three-level decomposition of the wind direction data. Wavelet decomposition extracts information of different frequencies from the original signal, where the approximate component represents the trend component of the original information, while the detail component represents the high-frequency fluctuation component of the original information. The preprocessed original wind direction data is used as the input signal, and the wind direction time series f(x) can be decomposed as follows:

[0027] f(x) = a n +d n-1 +...+d1

[0028] In the formula, a n The approximate trend component; n is the wavelet decomposition level; d n For high-frequency detail components.

[0029] After wavelet decomposition, the original wind direction data of this invention yields an approximate trend component a3 and high-frequency detail components d1, d2, and d3. For the approximate trend component, an ARMA model is used for prediction. For the high-frequency detail components d1, d2, and d3, a Bayes-LSTM model is used for training and prediction, respectively. The predicted components are then reconstructed using inverse wavelet to obtain the final wind direction prediction value.

[0030] This invention is based on real-time average wind speed and wind direction stability P st Adjust the yaw angle threshold EA and yaw delay threshold ET of the wind turbine, specifically:

[0031] Under normal conditions, when the wind speed v is 0-3m / s, the wind turbine does not operate. 3m / s is the cut-in wind speed. 3-11m / s is the cut-in wind speed and rated wind speed range. 11-16m / s is the rated wind speed range. 16-25m / s is the wind turbine in the power control stage. After 25-30m / s, the wind turbine stops operating.

[0032] The positive effects of this invention are as follows: This invention predicts the optimal short-term yaw speed by measuring real-time wind speed and wind direction stability, thus overcoming the adverse effects of wind direction fluctuations on system performance; at the same time, by adjusting the threshold of the yaw system in real time based on wind direction stability and wind speed level, it avoids frequent start-ups and shutdowns of the yaw system under low and excessively high wind speeds, and also avoids insufficient power output of wind turbines caused by the wind rotor not being able to align with the wind in time when the wind speed is suitable. Attached Figure Description

[0033] Appendix Figure 1 This is a flowchart of the method of the present invention;

[0034] Appendix Figure 2 This is an example diagram of the fluctuation amplitude A and fluctuation duration T of the present invention;

[0035] Figure 3(a) is an example of the marginal probability density distribution curve of the fluctuation amplitude A of the present invention;

[0036] Figure 3(b) is an example of the marginal probability density distribution curve of the fluctuation duration T of the present invention;

[0037] Appendix Figure 4 This is a standard power curve diagram of the wind turbine generator set of the present invention;

[0038] Appendix Figure 5 This is a diagram of the short-term wind direction prediction model architecture of the present invention. Detailed Implementation

[0039] The invention will now be described in detail with reference to the accompanying drawings.

[0040] As attached Figure 1 As shown, the present invention provides a yaw control method for wind turbines to cope with wind direction fluctuations, characterized by the following steps:

[0041] Step S1: Obtain the SCADA historical data of the wind farm, and filter the wind direction data in the historical data to obtain the wind direction sequence;

[0042] As attached Figure 2As shown, the specific filtering method for historical wind direction data is as follows: For data with fluctuation amplitude A less than 2° or fluctuation duration T less than 2s, a sliding window of size 2s is set. The data is traversed, and the data within the sliding window is treated as an independent signal segment. Each signal segment is analyzed, and the difference between its maximum and minimum values ​​(i.e., fluctuation amplitude) is checked. If it is less than 2°, the signal segment is considered to meet the fluctuation amplitude filtering condition. At the same time, it is checked whether the fluctuation duration within the signal segment is less than 2s. If the fluctuation duration is less than 2s, the signal segment meets the fluctuation duration filtering condition. If the signal segment meets either the fluctuation amplitude filtering condition or the fluctuation duration filtering condition, the original data is retained. Signal segments that do not meet either condition can be attenuated (e.g., multiplied by a coefficient less than 1). The processed signal segments are then spliced ​​together to obtain the filtered wind direction sequence.

[0043] Step S2: Construct a short-term wind direction prediction model based on historical wind direction data;

[0044] The architecture of the short-term wind direction prediction model is as follows: Figure 4 As shown, this invention employs a combination of ARMA based on wavelet decomposition (WD) and Bayes-LSTM to predict short-term wind direction. In this embodiment, the db4 wavelet is used to perform a three-level decomposition of the wind direction data. Wavelet decomposition extracts information at different frequencies from the original signal, where the approximate component represents the trend component of the original information, while the detail component represents the high-frequency fluctuation component. The preprocessed original wind direction data is used as the input signal, and the wind direction time series... f(x) It can be broken down into:

[0045] f(x) = a n +d n-1 +...+d1

[0046] In the formula, a n The approximate trend component; n is the wavelet decomposition level; d n For high-frequency detail components;

[0047] like Figure 5 As shown, after wavelet decomposition, the original wind direction data yields an approximate trend component a3 and high-frequency detail components d1, d2, and d3. For the approximate trend component, the ARMA model is used for prediction, while for the high-frequency detail components d1, d2, and d3, the Bayes-LSTM model is used for training and prediction respectively. The predicted components are then reconstructed using inverse wavelet to obtain the final wind direction prediction value.

[0048] By constructing a short-term wind direction prediction model to predict the short-term wind direction during real-time control, the wind direction stability P is calculated. st Then, the real-time average wind speed is used as the input to the trained model in step S3.

[0049] Step S3: Determine the wind direction stability P based on the wind direction sequence obtained after filtering in step S1. st Based on the average wind speed and wind direction stability P within a set time period st By training the yaw speed model, the current ideal yaw speed v can be obtained;

[0050] The yaw speed training model is built based on a BP neural network, and its parameters are optimized using a Bayesian optimization algorithm. The input of the model is the average wind speed and wind direction stability Pst within a set time period, and the output is the yaw speed. The model is built with reference to the model training in the existing technology.

[0051] In this embodiment of the invention, the set time is 5 seconds;

[0052] As shown in Figures 3(a) and 3(b), the fluctuation amplitude A and fluctuation duration T of the wind direction sequence are extracted, and the wind direction stability P is determined from the wind direction sequence. st The specific steps are as follows:

[0053] First, calculate the marginal probability density distributions of the fluctuation amplitude A and the fluctuation duration T using a Weibull distribution with offset. The formula for the Weibull distribution with offset is as follows:

[0054]

[0055] In the formula, α is the scale parameter, β is the shape parameter, and m is the length of the displacement;

[0056] The joint probability density of fluctuation amplitude A and fluctuation duration T is expressed using joint copulas, and its formula is as follows:

[0057] C(u, v)=ω1C_{Gum bel}(u, v)+ω2C_{Clayton}(u, v)+ω3C_{Frank}(u, v)

[0058] In the formula, C_{Gumbel}(u, v), C_{Clayton}(u, v), and C_{Frank}(u, v) are three Archimedes Copula functions, ω1, ω2, and ω3 are the weights corresponding to each copula, and u and v are the marginal probability densities of fluctuation amplitude A and fluctuation duration T.

[0059] Wind direction stability P st Calculation method:

[0060] In the formula, EA is a predefined fluctuation amplitude threshold, ET is a predefined fluctuation time threshold, and f(x, y) is the joint probability density of (A, T).

[0061] Specifically, the output of the model is the yaw speed v, which is the stability P of the model under different wind speeds and directions. st The corresponding standard output power (standard wind power curve is shown in the figure) Figure 4 The lowest yaw speed corresponding to (as shown); it should be noted that the yaw speed v in the embodiment of the present invention is the ratio of the angle of the nacelle blade offset to the time period within a certain period. For example, if the offset angle is Δθ within the time period t1-t2, then the yaw speed v = Δθ / (t2-t1).

[0062] Step S4: Based on the real-time average wind speed and real-time wind direction stability P st Adjust the yaw angle threshold EA and yaw delay threshold ET of the wind turbine, and use them as input to the model trained in step S3 to obtain the yaw speed v, and then perform yaw control on the wind turbine.

[0063] In real-time, under normal conditions, when the wind speed v is 0-3m / s, the wind turbine does not operate. 3m / s is the cut-in wind speed. 3-11m / s is the cut-in wind speed and rated wind speed range. 11-16m / s is the rated wind speed range. 16-25m / s is the power control stage. After 25-30m / s, the wind turbine stops operating.

[0064] In this embodiment, the calculated wind direction stability P st Stability levels are classified according to their numerical range. In this embodiment, three levels are defined, with level 3 being the most stable. Stability decreases sequentially. Table 1 below shows the correspondence between wind speed range, stability level, and yaw threshold:

[0065] Table 1 Classification

[0066]

[0067]

[0068] It should be understood that the embodiments of the present invention provide the correspondence between wind speed ranges and three wind direction stability levels and yaw angle thresholds EA and yaw delay thresholds ET. In actual situations, wind speed ranges and wind direction stability levels can be divided more meticulously according to the actual site conditions.

[0069] In summary, under the existing fixed threshold parameters, based on the average wind speed and P... st The core of this invention lies in modifying the yaw angle and yaw delay thresholds according to the classification level, and then predicting the ideal yaw speed using a trained model. This invention effectively drives the yaw mechanism to work, better coping with wind direction fluctuations.

[0070] This invention predicts the optimal short-term yaw speed by analyzing real-time wind speed and direction stability, overcoming the adverse effects of wind direction fluctuations on system performance. Simultaneously, by adjusting the yaw system's threshold in real-time based on wind direction stability and wind speed level, it avoids frequent start-ups and shutdowns of the yaw system under low and excessively high wind speeds, and also prevents insufficient power output from wind turbines due to the rotor's inability to align with the wind in suitable wind conditions. The ideal yaw speed predicted by the trained model effectively drives the yaw mechanism and better copes with wind direction fluctuations.

[0071] Obviously, the above embodiments are merely one embodiment of the present invention and are not intended to limit the present invention; at the same time, those skilled in the art will recognize that the present invention can have various modifications and variations; any modifications, equivalent substitutions and improvements made within the spirit and scope of the present invention are within the protection scope of the present invention.

Claims

1. A yaw control method for wind turbines to cope with wind direction fluctuations, characterized in that, The following steps are used: Step S1: Obtain historical SCADA data of the wind farm, and filter the wind direction data in the historical data to obtain the wind direction sequence; Step S2: Construct a short-term wind direction prediction model based on historical wind direction data; Step S3: Determine the wind direction stability P based on the wind direction sequence obtained after filtering. st Based on the average wind speed and wind direction stability P within a set time period st By training the yaw speed model, the current ideal yaw speed v can be obtained; Step S4: Based on the real-time average wind speed and real-time wind direction stability P st Adjust the control parameter threshold and use it as the input to the model trained in step S3 to obtain the yaw speed v, and then perform yaw control on the wind turbine. Step S2: The short-term wind direction prediction model uses a combination of ARMA and Bayes-LSTM based on wavelet decomposition (WD) to predict short-term wind direction. The original wind direction data is preprocessed and decomposed by wavelet to obtain approximate trend components and high-frequency detail components. The ARMA algorithm is used to build a prediction model for the trend components, and the Bayes-LSTM neural network is used to build a prediction model for the detail components. The models are combined and reconstructed to obtain the final wind direction prediction value. Step S3 extracts the fluctuation amplitude A and fluctuation duration T of the wind direction sequence to determine the wind direction stability P. st ; In step S3, the wind direction sequence determines the wind direction stability P. st The specific steps are as follows: First, calculate the marginal probability density distributions of the fluctuation amplitude A and the fluctuation duration T using a Weibull distribution with offset. The formula for the Weibull distribution with offset is as follows: In the formula, α is the scale parameter, β is the shape parameter, and m is the length of the displacement; The joint probability density of fluctuation amplitude A and fluctuation duration T is expressed using joint copulas, and its formula is as follows: In the official They are three Archimedes Copula functions. These are the weights corresponding to each copula, and u and v are the marginal probability densities of the fluctuation amplitude A and the fluctuation duration T. Wind direction stability P st Calculation method: In the formula, EA is a predefined fluctuation amplitude threshold, ET is a predefined fluctuation time threshold, and f(x,y) is the joint probability density of (A,T).

2. The yaw control method for wind turbines to cope with wind direction fluctuations according to claim 1, characterized in that, In step S1, the historical wind direction data filtering specifically employs the following methods: For data filtering where the fluctuation amplitude A is less than 2° or the fluctuation duration T is less than 2s, a sliding window of 2s is set. The data is traversed, and the data within the sliding window is treated as an independent signal segment. Each signal segment is analyzed, and the difference between its maximum and minimum values ​​is checked. If it is less than 2°, the signal segment is considered to meet the fluctuation amplitude filtering condition. At the same time, it is checked whether the fluctuation duration within the signal segment is less than 2s. If the fluctuation duration is less than 2s, the signal segment meets the fluctuation duration filtering condition. If the signal segment meets either the fluctuation amplitude filtering condition or the fluctuation duration filtering condition, the original data is retained. Signal segments that do not meet either condition are attenuated, and the processed signal segments are spliced ​​together to obtain the filtered wind direction sequence.

3. The yaw control method for wind turbines to cope with wind direction fluctuations according to claim 1, characterized in that, In step S3, a model is constructed based on a BP neural network, and its parameters are optimized using a Bayesian optimization algorithm.

4. The yaw control method for wind turbines to cope with wind direction fluctuations according to claim 1, characterized in that, In step S2, a combination of ARMA based on wavelet decomposition (WD) and Bayes-LSTM is used to predict short-term wind direction. The db4 wavelet is used to perform a three-level decomposition of the wind direction data. Wavelet decomposition extracts information of different frequencies from the original signal, where the approximate component represents the trend component of the original information, while the detail component represents the high-frequency fluctuation component of the original information. Using the preprocessed original wind direction data as the input signal, the wind direction time series f(x) can be decomposed as follows: In the formula, a n The approximate trend component; n is the wavelet decomposition level; d n For high-frequency detail components.

5. The yaw control method for wind turbines to cope with wind direction fluctuations according to claim 4, characterized in that, After wavelet decomposition, the original wind direction data yields an approximate trend component a3 and high-frequency detail components d1, d2, and d3. For the approximate trend component, an ARMA model is used for prediction. For the high-frequency detail components d1, d2, and d3, a Bayes-LSTM model is used for training and prediction, respectively. The predicted components are then reconstructed using inverse wavelet to obtain the final wind direction prediction value.

Images