A method and device for calculating the azimuth angle of a wind turbine generator system based on measured load data

Abstract

The embodiments of the present application provide a method and device for calculating the azimuth angle of a wind turbine based on measured load data, and relate to the field of wind power technology. The method includes analyzing the correlation between SCADA system parameters, measured load data and azimuth angle based on the Pearson correlation coefficient to determine strong correlation parameters; training a pre-built ModernTCN neural network model according to the strong correlation parameters; predicting the difference between the azimuth angle and its initial position during the operation of the wind turbine according to the trained neural network model; fitting the load curve through Fourier series according to the difference to obtain a predicted azimuth angle-load model; obtaining the unknown azimuth angle of the initial position of the wind turbine according to the predicted azimuth angle-load model and the measured azimuth angle-load model, thereby determining the azimuth angle time series during the entire operation of the wind turbine. It makes up for the deficiencies of the theoretical calculation method of azimuth angle in the field of traditional wind turbines and provides technical support for the monitoring of the operating status of wind turbine systems.

Description

Technical Field

[0001] The present application relates to the field of wind power technology, and in particular to a method and device for calculating the azimuth angle of a wind turbine generator set based on measured load data. Background Art

[0002] During the operation of the wind turbine, different blades are located in different positions and azimuths at the same time. The change in azimuth will cause differences in blade swing loads and flapping loads at different positions. The size of the azimuth will directly affect the force state of the blades, thereby significantly affecting the overall power generation performance of the wind turbine. Currently, most wind turbine SCADA systems do not provide azimuth data, and the traditional method of obtaining azimuth data mainly relies on the installation of actual measurement sensors. This method not only increases the cost of equipment, but also brings inconvenience to the maintenance and operation of the wind turbine. In addition, there are technical challenges in the installation and maintenance of sensors, especially in harsh environmental conditions, which further limits the widespread application of this method.

[0003] In the prior art, there is no method that can accurately reconstruct the azimuth of a wind turbine without additional sensors. Considering the strong correlation between the azimuth of a wind turbine and some SCADA data and measured load data, if it is possible to install measured load sensors on wind turbine prototypes, combine the wind turbine SCADA data, build an azimuth-load model, and predict the azimuth time series of the prototype, combined with the characteristics that the performance of wind turbines of the same model is roughly the same, this calculation method can be extended to other wind turbines of the same model in the wind farm that are not equipped with sensors, which will greatly reduce the cost and complexity of obtaining azimuth data. Summary of the invention

[0004] In view of the problem that the traditional method of obtaining azimuth data of wind farms mainly relies on the installation of measured sensors, the present application provides a method and device for calculating the azimuth of wind turbines based on measured load data. By installing sensors on wind turbine prototypes, a theoretical model is constructed to predict the azimuth data of wind turbines of the same model in the wind farm that are not equipped with sensors.

[0005] According to a first aspect of the present application, a method for calculating the azimuth angle of a wind turbine generator system based on measured load data is provided, the method comprising:

[0006] The correlation between SCADA system parameters and measured load data and azimuth angle was analyzed based on the Pearson correlation coefficient to determine the strongly correlated parameters;

[0007] According to the strongly correlated parameters, the pre-built ModernTCN neural network model is trained to obtain a trained ModernTCN neural network model;

[0008] According to the trained ModernTCN neural network model, predict the difference between the azimuth angle and the initial position during the operation of the wind turbine;

[0009] According to the difference, the load curve is fitted by Fourier series to obtain a predicted azimuth-load model;

[0010] The unknown azimuth angle of the initial position of the wind turbine is obtained according to the predicted azimuth angle-load model and the measured azimuth angle-load model, thereby determining the azimuth angle time series during the entire wind turbine operation process.

[0011] According to the above aspects and any possible implementation, an implementation is further provided, wherein the analysis of the correlation between the SCADA system parameters and the measured load data and the azimuth based on the Pearson correlation coefficient to determine the strongly correlated parameters includes:

[0012] Acquire all parameter data related to the azimuth angle in the SCADA system and measured load data based on sensors, and pre-process the parameter data and load data;

[0013] Calculating the Pearson correlation coefficient between each data and the azimuth according to the preprocessed parameter data and load data;

[0014] According to the Pearson correlation coefficient between each data and the azimuth, the strength of the correlation between each data and the azimuth is judged, and the strong correlation parameter is determined.

[0015] According to the above aspects and any possible implementation manner, an implementation manner is further provided, wherein judging the correlation strength between each data and the azimuth angle according to the Pearson correlation coefficient between each data and the azimuth angle comprises:

[0016] If the Pearson correlation coefficient is between the first coefficient threshold and 1 or is equal to the first coefficient threshold, then there is a strong positive correlation between the data and the azimuth;

[0017] If the Pearson correlation coefficient is between the first coefficient threshold and the second coefficient threshold, the data is weakly correlated with the azimuth;

[0018] If the Pearson correlation coefficient is between the second coefficient threshold and -1 or is equal to the second coefficient threshold, then there is a strong negative correlation between the data and the azimuth;

[0019] Among them, 1>first coefficient threshold>0>second coefficient threshold>-1.

[0020] According to the aspects described above and any possible implementation manner, an implementation manner is further provided, wherein the strongly correlated parameters include blade flapping load, blade swing load, hub wind speed, and wind rotor rotation speed.

[0021] According to the above aspects and any possible implementation, an implementation is further provided, wherein the pre-built ModernTCN neural network model is trained according to the strongly correlated parameters, including:

[0022] The blade flapping load data, blade swing load data, hub wind speed data, and wind rotor speed data are transmitted to the input layer of the ModernTCN neural network model, wherein the ModernTCN neural network model includes at least one time convolution block;

[0023] The input layer of the ModernTCN neural network model performs data preprocessing to form time series samples, and extracts feature information of the sequence samples at different time scales through a time convolution block;

[0024] The pre-built ModernTCN neural network model is trained according to the feature information of the sequence samples at different time scales.

[0025] According to the above aspects and any possible implementation, an implementation is further provided, wherein the pre-built ModernTCN neural network model is trained according to the feature information of the sequence samples at different time scales, including:

[0026] According to the characteristic information of the sequence samples at different time scales, a loss function between the predicted azimuth and the actual azimuth is established;

[0027] By minimizing the loss function, the weight parameters of the convolution layer in the temporal convolution block are adjusted to learn the nonlinear relationship between the input parameters and the azimuth change.

[0028] According to the above aspects and any possible implementation, an implementation is further provided, wherein according to the difference, fitting the load curve by Fourier series to obtain a predicted azimuth-load model comprises:

[0029] Determining the Fourier coefficients by the least square method according to the difference between the azimuth angle and the initial position during the operation of the fan;

[0030] Based on the Fourier coefficients and the difference, a predicted azimuth-load model is determined.

[0031] According to the above aspects and any possible implementation, an implementation is further provided, wherein the unknown azimuth of the initial position of the wind turbine is obtained according to the predicted azimuth-load model and the measured azimuth-load model, thereby determining the azimuth time series during the entire operation of the wind turbine, including:

[0032] The predicted azimuth-load model is combined with the measured azimuth-load model to determine a mathematical expression between the predicted azimuth-load model and the measured azimuth-load model;

[0033] The unknown azimuth angle of the initial position of the wind turbine is calculated according to the mathematical expression between the predicted azimuth angle-load model and the measured azimuth angle-load model;

[0034] According to the unknown azimuth angle of the initial position of the wind turbine, the azimuth angle time series during the entire operation process of the wind turbine is determined.

[0035] According to the above aspects and any possible implementation, an implementation is further provided, wherein the mathematical expression between the predicted azimuth-load model and the measured azimuth-load model is:

[0036] ;

[0037] in, is the measured azimuth, is the measured azimuth-load model; To predict the azimuth, To predict the azimuth-load model The azimuth-load model obtained by considering the initial position azimuth and measured azimuth The results of the coincidence on the mathematical parameter model; is the unknown azimuth of the initial position.

[0038] According to a second aspect of the present application, a device for calculating the azimuth angle of a wind turbine generator system based on measured load data is provided. The device comprises:

[0039] A strong correlation parameter determination module is used to analyze the correlation between the SCADA system parameters and the measured load data and the azimuth angle based on the Pearson correlation coefficient to determine the strong correlation parameters;

[0040] A model training module, used to train the pre-built ModernTCN neural network model according to the strongly correlated parameters to obtain a trained ModernTCN neural network model;

[0041] An azimuth difference prediction module is used to predict the difference between the azimuth and its initial position during the operation of the wind turbine according to the trained ModernTCN neural network model;

[0042] An azimuth-load curve simulation module is used to obtain a predicted azimuth-load model by fitting the load curve through Fourier series according to the difference;

[0043] The azimuth time series determination module is used to obtain the unknown azimuth of the initial position of the wind turbine according to the predicted azimuth-load model and the measured azimuth-load model, so as to determine the azimuth time series during the entire operation of the wind turbine.

[0044] According to a third aspect of the present application, an electronic device is provided. The electronic device comprises: a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the program, the method described above is implemented.

[0045] According to a fourth aspect of the present application, a computer-readable storage medium is provided, on which a computer program is stored, and when the program is executed by a processor, the method according to the first aspect of the present application is implemented.

[0046] The wind turbine azimuth angle calculation method based on measured load data provided by the present invention utilizes the characteristics that the same type of wind turbines have roughly the same performance, combines the data of wind turbine prototypes equipped with load sensors and azimuth angle sensors, and realizes accurate prediction of the azimuth angles of other wind turbines of the same type in the wind farm through a neural network model without the need for additional sensors. First, the Pearson correlation coefficient is used to analyze the correlation between the azimuth angle and parameters such as wind speed, flapping load and speed mean in the SCADA system, and the strongly correlated parameters are selected as input to construct a ModernTCN neural network model to predict the azimuth angle difference within a certain period of time. The model is trained and verified by measured data, and the initial azimuth angle is inferred by combining the azimuth angle-load model, thereby obtaining the azimuth angle time series during the entire operation of the wind turbine. It makes up for the shortcomings of the azimuth angle theoretical calculation method in the traditional wind turbine field and provides technical support for the monitoring of the operating status of the wind turbine system.

[0047] It should be understood that the contents described in the Summary of the Invention are not intended to limit the key or important features of the embodiments of the present application, nor are they intended to limit the scope of the present application. Other features of the present application will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] The above and other features, advantages and aspects of the embodiments of the present application will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. The accompanying drawings are used to better understand the present solution and do not constitute a limitation of the present application. In the accompanying drawings, the same or similar reference numerals represent the same or similar elements, among which:

[0049] Figure 1 A flow chart of a method for calculating the azimuth angle of a wind turbine generator system based on measured load data according to an embodiment of the present application is shown;

[0050] Figure 2 shows a Pearson correlation coefficient diagram according to an embodiment of the present application;

[0051] Figure 3 A schematic diagram of gravity during the rotation of blades according to an embodiment of the present application is shown;

[0052] Figure 4 shows an azimuth angle-shimmy load relationship diagram according to an embodiment of the present application;

[0053] Figure 5 shows an azimuth angle-flapping load relationship diagram according to an embodiment of the present application;

[0054] Figure 6 shows an azimuth angle-rotation speed relationship diagram according to an embodiment of the present application;

[0055] Figure 7 shows an azimuth angle-load prediction effect diagram according to an embodiment of the present application;

[0056] Figure 8 A block diagram of a device for calculating the azimuth angle of a wind turbine generator system based on measured load data according to an embodiment of the present application is shown. DETAILED DESCRIPTION

[0057] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0058] In addition, the term "and / or" in this article is only a description of the association relationship between the associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship.

[0059] The present application provides a method for calculating the azimuth angle of a wind turbine based on measured load data. First, the Pearson correlation coefficient is used to analyze the correlation between the wind speed, the average speed, the measured load data based on the sensor and the azimuth angle in the SCADA system, and the strongly correlated parameters are selected as inputs. Secondly, a ModernTCN neural network model is constructed, and the azimuth angle difference within a certain period of time is predicted according to the model. Finally, the model is trained and verified by the measured data, and the initial azimuth angle is inferred by combining the azimuth angle-load model, and then the azimuth angle time series during the whole operation of the wind turbine is obtained. This method not only reduces the system cost, but also improves the integrity and accuracy of the wind turbine operation data.

[0060] Figure 1A flow chart of a method 100 for calculating the azimuth angle of a wind turbine generator system based on measured load data according to an embodiment of the present application is shown. Figure 1 As shown:

[0061] S110, analyzing the correlation between the SCADA system parameters and the measured load data and the azimuth angle based on the Pearson correlation coefficient, and determining a strongly correlated parameter.

[0062] In some embodiments, the Pearson Correlation Coefficient is a statistic that measures the linear relationship between two variables. Its calculation is based on covariance (the degree to which two variables deviate from their respective means at the same time) and standard deviation (the degree of dispersion of the variables). By standardizing the covariance, a correlation coefficient with a fixed range can be obtained, which facilitates the comparison of the strength of the linear relationship between different data sets. Its value range is between -1 and 1. The calculation formula of the Pearson Correlation Coefficient is as follows:

[0063] ;

[0064] in, is the Pearson correlation coefficient, x i and y i are the observed values ​​of the two variables, and They are x and y The average value of .

[0065] In some embodiments, in order to determine the parameters in the SCADA system data that are strongly correlated with the azimuth angle, first, all parameter data related to the azimuth angle in the SCADA system (such as wind speed, average rotation speed, etc.) can be obtained, and the measured load data can be obtained based on the load sensor, and the parameter data and load data can be preprocessed; then, based on the preprocessed parameter data and load data, the Pearson correlation coefficient between each data and the azimuth angle is calculated; finally, based on the Pearson correlation coefficient between each data and the azimuth angle, the correlation strength between each data and the azimuth angle is judged to determine the strongly correlated parameters.

[0066] Specifically, if the Pearson correlation coefficient is between the first coefficient threshold and 1, then there is a strong positive correlation between the data and the azimuth;

[0067] If the Pearson correlation coefficient is between the first coefficient threshold and the second coefficient threshold, the data is weakly correlated with the azimuth;

[0068] If the Pearson correlation coefficient is between the second coefficient threshold and -1, there is a strong negative correlation between the data and the azimuth;

[0069] Among them, 1>first coefficient threshold>0>second coefficient threshold>-1.

[0070] It should be noted that the first coefficient threshold and the second coefficient threshold can be selected according to actual needs. In this embodiment, the first coefficient threshold can be 0.05, and the second coefficient threshold can be -0.05.

[0071] like Figure 2 The following is a Pearson correlation coefficient analysis diagram. The target parameters for analysis include the flapping of blades 1, 2, and 3, the shimmy load, hub wind speed, rotor speed, generator power, ambient temperature, cabin temperature, etc. Figure 2 It can be seen that parameters such as blade flapping load, blade swing load, hub wind speed, and wind rotor speed have a significant correlation with the wind rotor azimuth. Therefore, blade flapping load, blade swing load, hub wind speed, and wind rotor speed can be determined as strongly correlated parameters of the azimuth.

[0072] Specifically, the target parameter analysis process is as follows:

[0073] (1) Blade flapping load and swing load: During the rotation of the blade, the gravity and aerodynamic load act together on the blade root, generating bending moment. The gravity diagram during the rotation of the blade is shown in the figure below: Figure 3 As shown in Figure 2, the direction and magnitude of the resultant force of these loads will change accordingly as the azimuth angle of the wind rotor changes. The flapping load and the swing load are the vector sum of the aerodynamic load and the gravity load, and their changes directly reflect the dynamic response of the blade at different azimuth angles.

[0074] Figure 4 , Figure 5 They are the azimuth-swing load relationship diagram and the azimuth-flapping relationship diagram, which are drawn based on measured data using the polar coordinate system. The polar coordinate system can intuitively reflect the variation of the swing load of blade 1 at different azimuths. The radial coordinate represents the amplitude of the swing load, which is negative in the center and gradually increases outward. The angular coordinate represents the azimuth of the wind wheel. A cycle of 360 degrees conforms to the azimuth value range. The polar coordinate system can well reflect the relationship between the load and the azimuth. The distribution of scattered points shows the periodic fluctuation of the load, revealing the dynamic response characteristics of the blade during rotation. The fluctuation amplitude in the figure reflects the maximum and minimum values ​​of the blade swing load.

[0075] For periodic data such as azimuth-load data, Fourier fitting is particularly suitable for processing. Fourier fitting uses the sum of some sine terms and cosine terms to approximate the observed data set to achieve the effect of smoothing discrete data. It is a commonly used curve fitting method.

[0076] Fourier series fit basic model:

[0077] ;

[0078] ;

[0079] ;

[0080] in, and are the Fourier coefficients, is the nth order harmonic, , T is the period.

[0081] The matrix form can be expressed as AX=Y, where

[0082] ;

[0083] ;

[0084] Among them, m n / 2, can be solved by the least squares method, and the least squares solution is:

[0085] ;

[0086] The coefficients of the Fourier fitting equation can be solved from the above formula.

[0087] The following methods are used to determine the parameters of Fourier fitting effect:

[0088] Coefficient of determination

[0089] ;

[0090] and variance

[0091] ;

[0092] Root mean square error

[0093] ;

[0094] The coefficient of determination (COD), also known as the coefficient of determination, is a relative indicator used to measure the goodness of fit between the fitted curve and the observed data. It reflects the proportion of the variation of the dependent variable explained by the independent variable. The value range of the coefficient of determination is between 0 and 1. The closer its value is to 1, the stronger the model's ability to explain the dependent variable, that is, the better the model fits the data. A larger coefficient of determination R2 indicates that the fitted model can explain most of the uncertainty of the dependent variable, and the goodness of fit of the regression equation is higher. In addition, the squared error (SSE) represents the total sum of squares of the errors between the original data and the fitted data, while the root mean square error (RMSE) is called the standard deviation of the regression model. Smaller SSE and RMSE indicate that the model selection is reasonable and the fitting effect is good, which also means that the model has stronger predictive ability.

[0095] (2) Hub wind speed: The turbulent characteristics of wind speed will cause the blades to be subjected to uncertain aerodynamic loads, which in turn affect the blades' flapping and shimmying loads. Since flapping and shimmying are the vector sum of aerodynamic loads and blade gravity loads, the relationship between wind speed and aerodynamic loads needs to be considered when analyzing the relationship between azimuth angle and load. In addition, the wind speed determines the rotor speed, which in turn has a direct impact on the speed of azimuth angle changes.

[0096] The blade element momentum theory is widely used in the design and performance calculation of wind turbines. It assumes that the force acting on the blade element can be calculated by the two-dimensional airfoil characteristics of the angle of attack measured by the incident combined velocity on the element section, that is, the force acting on the blade element is only related to the momentum change of the gas through the ring swept by the blade element. Therefore, it is assumed that there is no radial interaction between the airflows passing through adjacent rings. Therefore, the velocity component in the span direction is ignored, and the three-dimensional effect is also ignored. The velocity component at a certain radial position of the blade is expressed by the wind speed. Knowing the angle of attack and the lift and drag coefficients as well as the axial and tangential induction factors on each blade element, the force acting on the blade can be calculated.

[0097] From the momentum theory, we know that the thrust acting on the dr micro segment is and torque They are:

[0098] ;

[0099] ;

[0100] According to the blade element theory, the thrust acting on the dr micro segment is and torque They are:

[0101] ;

[0102] ;

[0103] in is the radius of the blade element from the hub center; is the air density; is the incoming wind speed; is the axial induction factor; Tangential induction factor; is the blade angular velocity; is the number of leaves; is the string length; is the normal force coefficient; is the tangential force coefficient; is the relative air velocity, It can be calculated according to the following expression:

[0104] ;

[0105] in w is the impeller speed.

[0106] According to the strip theory, the thrust and torque obtained from the momentum theory and blade element theory are equal, that is: .

[0107] The axial induction factor can be obtained by sorting and tangential induction factor The expression is as follows:

[0108] ;

[0109] ;

[0110] in: is the inflow angle; σ is the solidity, .

[0111] (3) Wind rotor speed: There is an obvious causal relationship between wind rotor speed and azimuth. The speed determines the speed at which the wind rotor responds to changes in wind direction, that is, the rate of change of the azimuth. By analyzing the difference between the average wind rotor speed and the azimuth at two time points, the dynamic relationship between the speed and azimuth can be captured more accurately.

[0112] For the azimuth data, the sliding difference method is used, and a fixed window size of 30 data points is selected, which is recorded as , subtract the i+30th azimuth data from the i-th azimuth data, that is

[0113] ;

[0114] Where i is an index that increases from 1 to the end of the sample. This process continues sliding the window until all data points are processed, and finally a series of azimuth data sequences are generated after difference calculation.

[0115] For the speed data, the same window size is used for arithmetic averaging. The speed data is defined as , sum and average the data points from i to i+29, that is,

[0116] ;

[0117] Then the window is slid sequentially, and the process is repeated for each group of 30 data points to finally generate a smooth speed data series.

[0118] Through sliding processing, short-term fluctuations in the data can be effectively smoothed to obtain more stable azimuth and rotation speed characteristics for subsequent modeling and analysis. Figure 7 As shown in the figure, it is a comparison of the trend of the front and rear azimuth angle difference and the average speed of some measured data.

[0119] S120: training the pre-built ModernTCN neural network model according to the strongly correlated parameters to obtain a trained ModernTCN neural network model.

[0120] In some embodiments, ModernTCN is a deep learning model that is improved on the traditional Temporal Convolutional Network (TCN). The model captures local and global patterns in time series data through temporal convolution operations, showing superior parallel computing capabilities and lower memory overhead compared to traditional recurrent neural networks (RNN) or long short-term memory networks (LSTM), so it has significant advantages in processing long-term dependent sequence data. ModernTCN can further improve the model's expressiveness and performance through a series of innovative structural optimizations.

[0121] Generally speaking, the network architecture of ModernTCN includes the following main components:

[0122] (1) Input layer: This layer receives multidimensional time series data as the input of the model and performs preprocessing on the data, such as normalization and feature extraction, to improve the generalization ability and prediction accuracy of the model. The design of the input layer is usually adjusted according to the specific task and data type.

[0123] (2) Temporal convolutional blocks: The core of ModernTCN consists of multiple stacked temporal convolutional blocks. Each convolutional block contains several temporal convolutional layers, which perform sliding convolution operations along the time dimension, so as to effectively capture local patterns of different scales in the time series. By introducing dilated convolutions, ModernTCN overcomes the problem of insufficient receptive field in traditional convolutional neural networks and can cover a larger time range while maintaining a small number of parameters, thereby fully capturing global dependencies in long time series. In addition, each convolutional block also contains residual connections and normalization operations to ensure that the gradient can be effectively propagated, further improving the stability and convergence speed of the network.

[0124] (3) Output layer: The output layer is responsible for converting the output of the last temporal convolutional block into the required prediction results. Depending on the requirements of different tasks, the output layer can use different activation functions and loss functions, such as linear activation functions in regression tasks or Softmax activation functions in classification tasks. The design of the output layer is usually optimized in combination with specific application scenarios and objective functions to maximize the prediction performance and robustness of the model.

[0125] ModernTCN achieves effective feature extraction and pattern capture in the modeling of time series data, and can show stronger modeling capabilities and higher computing efficiency in long time series data. In complex time series prediction tasks, it shows stronger parallelization capabilities, lower memory resource consumption and better generalization performance than traditional RNN or LSTM architectures.

[0126] In the wind turbine azimuth angle reconstruction method based on the ModernTCN neural network of this embodiment, first, strongly correlated parameters (flapping load, swing load, wind speed, rotation speed) are used as input data, and the model input layer preprocesses these strongly correlated parameters to form time series samples in order to capture the dynamic change characteristics existing in the system.

[0127] Specifically, the ModernTCN model relies on dilated convolution operations to extract local and global features from the input sequence through a layer-by-layer temporal convolutional network. The input layer of the model accepts four parameter values ​​(flapping load, sway load, wind speed, and rotation speed) in each time step and passes these values ​​to the stacked temporal convolution blocks. Each convolution block processes the data with a sliding window to extract feature information of the input sequence at different time scales.

[0128] In the training phase, based on the feature information of the input sequence at different time scales, a loss function is established between the azimuth prediction value and the actual observation value. By minimizing the loss function, the weight parameters in the convolution layer are gradually adjusted. In this process, the measured wind speed, load and rotation speed data are used for repeated training, so that the model can learn the nonlinear relationship between the parameters and the azimuth change.

[0129] During the testing and validation phase, the predictions of the model were tested using an independent measured data set. The accuracy and stability of the model were evaluated by comparing the azimuth changes predicted by the model with the actual measured azimuths. The performance of the model was quantified using metrics such as the root mean square error (RMSE) and the coefficient of determination (R²) to ensure that the model was able to effectively reconstruct the azimuth sequence of the wind turbine under different operating conditions.

[0130] S130, predicting the difference between the azimuth angle and the initial position during the operation of the wind turbine according to the trained ModernTCN neural network model.

[0131] In some embodiments, since the azimuth angle of the fan changes continuously, the trained ModernTCN neural network model ultimately outputs not the absolute azimuth angle value, but the change in azimuth angle, that is, the difference between the azimuth angle and its initial position during the operation of the fan.

[0132] S140, fitting the load curve by Fourier series according to the difference to obtain a predicted azimuth-load model.

[0133] In some embodiments, the Fourier fitting method is particularly suitable for processing periodic data such as azimuth-load data. Fourier fitting uses the sum of some sine terms and cosine terms to approximate the observed data set to achieve the effect of smoothing discrete data. It is a commonly used curve fitting method.

[0134] Specifically, the Fourier coefficients can be determined by the least square method according to the difference between the azimuth angle and its initial position during the operation of the wind turbine; and the predicted azimuth angle-load model can be determined according to the Fourier coefficients and the difference.

[0135] S150, obtaining the unknown azimuth of the initial position of the wind turbine according to the predicted azimuth-load model and the measured azimuth-load model, thereby determining the azimuth time series during the entire operation of the wind turbine.

[0136] In some embodiments, in order to reconstruct the azimuth time series during the entire wind turbine operation process, the predicted azimuth-load model can be combined with the measured azimuth-load model to determine the mathematical expression between the predicted azimuth-load model and the measured azimuth-load model; secondly, based on the mathematical expression between the predicted azimuth-load model and the measured azimuth-load model, the unknown azimuth of the initial position of the wind turbine is calculated; finally, based on the unknown azimuth of the initial position of the wind turbine, the azimuth time series during the entire wind turbine operation process is determined.

[0137] Among them, the mathematical expression between the predicted azimuth-load model and the measured azimuth-load model is:

[0138]

[0139] in, is the measured azimuth, is the measured azimuth-load model; To predict the azimuth, To predict the azimuth-load model The azimuth-load model obtained by considering the initial position azimuth and measured azimuth The results of the coincidence on the mathematical parameter model; is the unknown azimuth of the initial position.

[0140] Specific as Figure 7 As shown in the figure, the predicted azimuth-load model is combined with the measured azimuth-load model. It can be seen that the actual azimuth-load curve and the predicted azimuth-load curve only have an initial azimuth offset. The unknown azimuth of the initial position of the wind turbine can be calculated and simulated through the mathematical expression between the predicted azimuth-load model and the measured azimuth-load model, thereby successfully reconstructing the accurate azimuth time series of the wind turbine blades.

[0141] According to the embodiments of the present application, the following technical effects are achieved:

[0142] (1) This application utilizes the fact that wind turbines of the same model have roughly the same performance characteristics, and combines the data of wind turbine prototypes equipped with load sensors and azimuth sensors to accurately predict the azimuth angles of other wind turbines of the same model in the wind farm through a neural network model without the need for additional sensors.

[0143] (2) This application first uses the Pearson correlation coefficient to analyze the correlation between the wind speed, flapping load, speed difference and azimuth in the SCADA system, and selects the strongly correlated parameters as input. Secondly, a ModernTCN neural network model is constructed to predict the azimuth difference within a certain period of time. Finally, the model is trained and verified through measured data, and the initial azimuth is inferred by combining the azimuth-load model, and then the azimuth time series during the entire operation of the wind turbine is obtained. This method not only reduces the system cost, but also improves the integrity and accuracy of the wind turbine operation data.

[0144] It should be noted that, for the aforementioned method embodiments, for the sake of simplicity, they are all expressed as a series of action combinations, but those skilled in the art should be aware that the present application is not limited by the described order of actions, because according to the present application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all optional embodiments, and the actions and modules involved are not necessarily required by the present application.

[0145] The above is an introduction to the method embodiment. The following is a further explanation of the scheme described in this application through an apparatus embodiment.

[0146] Figure 8 FIG. 8 is a block diagram of a wind turbine azimuth angle calculation device 800 based on measured load data according to an embodiment of the present application. Figure 8 As shown, the device 800 includes:

[0147] A strongly correlated parameter determination module 810 is used to analyze the correlation between the SCADA system parameters and the measured load data and the azimuth based on the Pearson correlation coefficient to determine the strongly correlated parameters;

[0148] A model training module 820 is used to train the pre-built ModernTCN neural network model according to the strongly correlated parameters to obtain a trained ModernTCN neural network model;

[0149] An azimuth difference prediction module 830 is used to predict the difference between the azimuth and its initial position during the operation of the wind turbine according to the trained ModernTCN neural network model;

[0150] An azimuth-load curve simulation module 840 is used to obtain the unknown azimuth of the initial position of the wind turbine by fitting the load curve through Fourier series according to the difference, so as to obtain a predicted azimuth-load model;

[0151] The azimuth angle time series determination module 850 is used to determine the azimuth angle time series during the entire wind turbine operation process according to the predicted azimuth angle-load model and the measured azimuth angle-load model.

[0152] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the described module can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0153] In the technical solution of this application, the acquisition, storage and application of user personal information involved are in compliance with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0154] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps recorded in this application can be executed in parallel, sequentially or in different orders, as long as the expected results of the technical solution disclosed in this application can be achieved, and this document is not limited here.

[0155] The above specific implementations do not constitute a limitation on the protection scope of this application. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principles of this application should be included in the protection scope of this application.

Claims

1. A method for calculating the azimuth angle of a wind turbine generator system based on measured load data, characterized in that: include: The correlation between SCADA system parameters and measured load data and azimuth angle is analyzed based on the Pearson correlation coefficient, and strong correlation parameters are determined, including: Acquire all parameter data related to the azimuth angle in the SCADA system and measured load data based on sensors, and pre-process the parameter data and load data; According to the preprocessed parameter data and load data, the Pearson correlation coefficient between each data and the azimuth angle is calculated; According to the Pearson correlation coefficient between each data and the azimuth angle, the correlation strength between each data and the azimuth angle is judged, and a strongly correlated parameter is determined; wherein the strongly correlated parameter includes blade flapping load, blade swing load, hub wind speed, and wind rotor speed; According to the strongly correlated parameters, the pre-built ModernTCN neural network model is trained to obtain a trained ModernTCN neural network model; According to the trained ModernTCN neural network model, predict the difference between the azimuth angle and the initial position during the operation of the wind turbine; According to the difference, the load curve is fitted by Fourier series to obtain a predicted azimuth-load model; The unknown azimuth angle of the initial position of the wind turbine is obtained according to the predicted azimuth angle-load model and the measured azimuth angle-load model, thereby determining the azimuth angle time series during the entire wind turbine operation process.

2. A method for calculating the azimuth angle of a wind turbine generator system based on measured load data according to claim 1, characterized in that: The step of judging the correlation strength between each data and the azimuth angle according to the Pearson correlation coefficient between each data and the azimuth angle includes: If the Pearson correlation coefficient is between the first coefficient threshold and 1 or is equal to the first coefficient threshold, then there is a strong positive correlation between the data and the azimuth; If the Pearson correlation coefficient is between the first coefficient threshold and the second coefficient threshold, the data is weakly correlated with the azimuth; If the Pearson correlation coefficient is between the second coefficient threshold and -1 or is equal to the second coefficient threshold, then there is a strong negative correlation between the data and the azimuth; Among them, 1>first coefficient threshold>0>second coefficient threshold>-1.

3. The method for calculating the azimuth angle of a wind turbine generator system based on measured load data according to claim 1, characterized in that: The step of training the pre-built ModernTCN neural network model according to the strongly correlated parameters includes: The blade flapping load data, blade swing load data, hub wind speed data, and wind rotor speed data are transmitted to the input layer of the ModernTCN neural network model, wherein the ModernTCN neural network model includes at least one time convolution block; The input layer of the ModernTCN neural network model performs data preprocessing to form time series samples, and extracts feature information of the sequence samples at different time scales through a time convolution block; The pre-built ModernTCN neural network model is trained according to the feature information of the sequence samples at different time scales.

4. A method for calculating the azimuth angle of a wind turbine generator system based on measured load data according to claim 3, characterized in that: The step of training the pre-built ModernTCN neural network model according to the feature information of the sequence samples at different time scales includes: According to the characteristic information of the sequence samples at different time scales, a loss function between the predicted azimuth and the actual azimuth is established; By minimizing the loss function, the weight parameters of the convolution layer in the temporal convolution block are adjusted to learn the nonlinear relationship between the input parameters and the azimuth change.

5. The method for calculating the azimuth angle of a wind turbine generator system based on measured load data according to claim 1, characterized in that: The method of fitting the load curve by Fourier series according to the difference to obtain a predicted azimuth-load model comprises: Determining the Fourier coefficients by the least square method according to the difference between the azimuth angle and the initial position during the operation of the fan; Based on the Fourier coefficients and the difference, a predicted azimuth-load model is determined.

6. The method according to claim 1, characterized in that The method of obtaining the unknown azimuth angle of the initial position of the wind turbine according to the predicted azimuth angle-load model and the measured azimuth angle-load model, thereby determining the azimuth angle time series during the entire operation of the wind turbine, includes: The predicted azimuth-load model is combined with the measured azimuth-load model to determine a mathematical expression between the predicted azimuth-load model and the measured azimuth-load model; The unknown azimuth angle of the initial position of the wind turbine is calculated according to the mathematical expression between the predicted azimuth angle-load model and the measured azimuth angle-load model; According to the unknown azimuth angle of the initial position of the wind turbine, the azimuth angle time series during the entire operation process of the wind turbine is determined.

7. A method for calculating the azimuth angle of a wind turbine generator system based on measured load data according to claim 6, characterized in that: The mathematical expression between the predicted azimuth-load model and the measured azimuth-load model is: ； in, is the measured azimuth, is the measured azimuth-load model; To predict the azimuth, To predict the azimuth-load model The azimuth-load model obtained by considering the initial position azimuth and measured azimuth The results of the coincidence on the mathematical parameter model; is the unknown azimuth of the initial position.

8. A device for calculating the azimuth angle of a wind turbine generator system based on measured load data, comprising: The strongly correlated parameter determination module is used to analyze the correlation between the SCADA system parameters and the measured load data and the azimuth based on the Pearson correlation coefficient, and determine the strongly correlated parameters, including: Acquire all parameter data related to the azimuth angle in the SCADA system and measured load data based on sensors, and pre-process the parameter data and load data; According to the preprocessed parameter data and load data, the Pearson correlation coefficient between each data and the azimuth angle is calculated; According to the Pearson correlation coefficient between each data and the azimuth angle, the correlation strength between each data and the azimuth angle is judged, and a strongly correlated parameter is determined; wherein the strongly correlated parameter includes blade flapping load, blade swing load, hub wind speed, and wind rotor speed; A model training module, used to train the pre-built ModernTCN neural network model according to the strongly correlated parameters to obtain a trained ModernTCN neural network model; An azimuth difference prediction module is used to predict the difference between the azimuth and its initial position during the operation of the wind turbine according to the trained ModernTCN neural network model; An azimuth-load curve simulation module is used to obtain a predicted azimuth-load model by fitting the load curve through Fourier series according to the difference; The azimuth time series determination module is used to obtain the unknown azimuth of the initial position of the wind turbine according to the predicted azimuth-load model and the measured azimuth-load model, so as to determine the azimuth time series during the entire operation of the wind turbine.

Images