A wind speed and direction instrument calibration method based on blade load data

Abstract

The application provides a wind speed and direction instrument calibration method based on blade load data, comprising the following steps: S1, obtaining a blade root load; S2, collecting original data and performing pretreatment; S3, establishing a load model of a wind turbine generator based on force analysis; S4, establishing an equation function of the blade root load and the load model of the wind turbine generator; S5, constructing a neural network prediction model; S6, using a genetic algorithm to optimize the weight and threshold of the neural network, taking the error value of the output value and the actual value of the training sample as a fitness function, so as to obtain a prediction model with minimum error; S7, inputting a test set into the minimum error model for prediction, and verifying the result; and S8, obtaining a real-time wind speed value. After model training and verification are completed, the compensation value of the wind speed can be directly obtained through the model and the blade root sensing value, so that other calculations based on the wind speed are more accurate.

Description

Technical Field

[0001] This invention relates to the field of wind power technology, and specifically to a method for calibrating wind speed and direction instruments based on blade load data. Background Technology

[0002] In the wind power industry, as wind turbines become increasingly larger, various sensors need to be installed on them to monitor different environmental conditions and turbine operation. These sensors include temperature, humidity, wind speed, position and angle sensors for each motor, and voltage and current sensors. The anemometer of a wind turbine is installed at the rear of the nacelle. The wind speed at this location is affected by the turbulence caused by the rotation of the impeller at the front of the nacelle, which introduces a certain degree of error. Therefore, when performing various calculations using the anemometer data, there are often some errors, which can lead to the calculated target value being out of tolerance. Summary of the Invention

[0003] This invention provides a method for calibrating anemometers based on blade load data, which can compensate for wind speed detection data to make it closer to the actual wind speed.

[0004] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: A method for calibrating anemometers based on blade load data includes: S1. Install a blade root sensor at the blade root to detect the real-time changes in the load on the blade during torsion and tension, and obtain the blade root load. S2. Collect raw data and perform preprocessing; including setting up wind speed sensors at the wind turbine generator to collect real-time wind speed data, collecting the pitch angle of the wind turbine blades, collecting the azimuth angle information of the blades, collecting temperature and humidity data from the weather station, all raw data with time beacons, using the processed raw data as input values, and the actual wind speed compensation value as output values. S3. Establish a load model for the wind turbine generator based on stress analysis; S4. Convert the blade root load in S1 to the same load as the load model of the wind turbine generator set established in S3 and perform calculations to establish the equation function between the blade root load and the load model of the wind turbine generator set. S5. Divide the processed data in S2 into two sets of data, one set as the training set and the other set as the test set. The training set is used as the input sample of the neural prediction network. The correlation index between the theoretical load calculated by the load model of the wind turbine in S3 and the measured load converted from the blade root load in S4 is used as the output value to construct the neural network prediction model. S6. Use a genetic algorithm to optimize the weights and thresholds of the neural network, and use the error between the output value of the training sample and the actual value as the fitness function to obtain the prediction model with the minimum error. S7. Input the test set into the prediction model with the minimum error for prediction, and verify the results. If the error exceeds the limit, continue to increase the number of samples in the training set and repeat steps S5-S6 until the error does not exceed the limit, and obtain the prediction model for the wind speed compensation value. S8. Input the real-time raw data value into the prediction model of the wind speed compensation value verified in S7 to obtain the real-time wind speed compensation value. Add the wind speed compensation value to the detected wind speed value to obtain the real-time wind speed value.

[0005] The specific process of S2 mentioned above is as follows: S2.1. Take the meteorological data and parameters in the raw data with time stamps as samples. The meteorological data includes gas density, wind speed, humidity and temperature; the parameters include pitch angle and blade azimuth angle. S2.2. The correlation degree of meteorological factors is obtained through grey relational analysis, the weight of each meteorological factor is analyzed, and meteorological factors whose impact on wind turbine load and wind speed compensation is within the set value are screened out. The specific steps are as follows: S2.2.1 Suppose there are n sample data forming the following matrix, denoted as indicator data: ; n is the number of samples, and m is the number of indicators. ; S2.2.2 Determine the reference data series. The reference data series is a comparison standard. It can be constructed using the best or worst values ​​of each indicator, or other reference values ​​can be selected according to the evaluation purpose. It is denoted as: ; Calculate the absolute difference between the indicator data and the reference data: ; S2.2.3 Calculate the grey relational coefficient, using the following formula: ; Where ρ is the discrimination coefficient, 0 < ρ < 1. The smaller ρ is, the greater the difference between the correlation coefficients and the stronger the discrimination ability. S2.2.4 Calculate the grey weighted correlation degree, using the following formula: ; In the formula Let be the grey weighted correlation degree between the i-th evaluation object and the ideal object; S2.2.5. Based on the magnitude of the gray weighted correlation degree, the evaluation objects are sorted to establish the correlation order of the evaluation objects; S2.3 Processing of the integrated meteorological dataset: The meteorological factors obtained from grey relational analysis are denoted as the comprehensive meteorological set. Because each meteorological factor corresponds to a correlation degree, the greater the correlation degree, the greater the impact of the meteorological factor on the wind turbine load. Using the influence level of each influencing factor as the input value, a new meteorological set is obtained after assigning values. Finally, it is processed as input using a formula, which is as follows: ; Where sum is the summation term. Indicates the corresponding time. v This indicates the corresponding wind speed compensation value.

[0006] The process of establishing the load model of the wind turbine generator based on force analysis in S3 above is as follows: Calculate the forces on a single blade, the first... i The thrust outside the plane of rotation on each blade can be expressed as: ; in, The air density is calculated from temperature and altitude. Where is the blade radius; This refers to the wind turbine rotation speed; Let be the thrust coefficient of the blade outside the plane of rotation, the magnitude of which is related to the relative wind speed acting on the blade. Pitch angle and rotational speed Related to; subscript The serial numbers are for the three blades; The relative wind speed acting on the blades can be expressed as the actual wind speed. Vibration of the tower calculate: ; in, The tower height; This refers to the displacement of the tower top caused by the bending of the tower. The azimuth angle of the blade. The value is zero when the blade is at its vertical top. The forces acting in the other two directions on the blade are the resultant forces of air force, gravity, and inertia: ; in, , These are the aerodynamic components in the y and z directions, respectively. , These are the gravitational components in the y and z directions, respectively. , These are the inertial force components in the y and z directions, respectively; The components of aerodynamic force are calculated as follows: ; in, , , where represents the thrust coefficients of the blades in two perpendicularly intersecting directions within the rotor's plane of rotation. Along the blade span direction, Perpendicular to the blade's span, its size is related to the relative wind speed acting on that blade. Pitch angle and rotational speed related; Gravity always acts vertically downwards. As the blades rotate, let... Let be the mass of the blade, then the gravitational component can be expressed as: ; It is caused by acceleration. It is caused by centrifugal force: ; in, For the hub radius, The z-axis coordinate of the blade's center of mass; The torque on the blade is defined as follows: The torque that causes the wind turbine to rotate is not considered as a structural load; the torques in the other two directions are mainly due to aerodynamics, and can be expressed as: ; in, , These are the bending moment coefficients of the blades in two perpendicularly intersecting directions within the rotor's plane of rotation. Along the blade span direction, Perpendicular to the blade's span, its size is related to the relative wind speed acting on that blade. Pitch angle and rotational speed related; After obtaining the load on the blade, the load in the rotating blade coordinate system is transformed to the fixed hub coordinate system: ; in, The mass of the wheel hub; set up Center of mass of wheel hub x Axis coordinates Let the component of the hub's center of mass projected onto a fixed hub coordinate system be the load on the blades. ; In addition to increasing the gravity of the motors and the cabin arrive In addition to axle loads, the loads acting on the engine nacelle are the same as those acting on the wheel hubs: ; shaft and The shaft, also known as the tilting moment and deflection moment, can be expressed by the following formula: ; in, The center height of the wheel hub The distance between the top of the tower and the center of the wind turbine. and These are the components projected from the center of mass of the wheel hub onto the gravity of the motor and the engine compartment, respectively.

[0007] The blade root sensor in S1 described above is installed at the blade root bolt of each blade.

[0008] The specific process of S4 mentioned above is as follows: Let the value of the blade root sensor at each blade be... ,in Number the blades. If the blade root sensor is numbered, then the force it projects onto the tilting torque and deflection torque is... and They are respectively: ; in, and They are respectively The forces projected onto the y-axis and z-axis, where m is the number of root bolts for each blade; Then based on the tilting torque and deflection torque of the blade root sensor and They are respectively: ; in, The distance from the center of the wheel hub to the center of the tower; The equation function for the blade root load and the load model of the wind turbine generator is: ; in, and These are the linear coefficients of the equations for the tilting moment and the yaw moment, respectively. and These are the constants in the equations for tilting moment and deflection moment, respectively.

[0009] The specific process of S6 mentioned above is as follows: S6.1 Encoding and Population Creation: Let the range of the population be [a, b], and the number of individuals in the population be... c The parameter is represented by a binary code of length k, which is randomly generated. Let the code of an individual in the population be k. Set a random number r = rand. If r < 0.5, then... =0, if r>0.5, then =1; and so on; S6.2 Decoding: The binary code of the known individual is... The corresponding decoding formula is: ; The decoded values ​​correspond to the weights and thresholds; S6.3, Evaluation of Individual Fitness: The sample data is fed into a BP neural network with weights and thresholds determined by the initial population for training. The fitness is calculated as the inverse of the mean squared error (MSE) of the output and the actual value corresponding to each sample. The fitness formula is as follows: ; In the formula For the first The actual value corresponding to each training sample; For the first The neural network determined by each individual is the first... The predicted output value of each training sample; the smaller the mean squared error, the higher the fitness. Taking the reciprocal of the mean squared error as the fitness increases the probability of selecting superior individuals. S6.4 Selection: Select individuals from the initial population, using fitness to select better individuals; S6.5 Saving Individuals: Sort the selected individuals by fitness, and save the individuals within the set value range in a loop to prevent negative changes in individuals during crossover and mutation; S6.6 Crossover: The crossover operation uses a single-point crossover operator, that is, there is only one crossover point position. Two individuals in the population are randomly selected and a crossover point position is randomly generated to exchange their genetic codes. S6.7 Mutation: The mutation operation uses uniform mutation, which means that individuals in the population will have a small probability of flipping the set value of their genes, from 0 to 1, or from 1 to 0. S6.8 Repeat steps S6.2 to S6.7 until the optimal weights and thresholds are obtained.

[0010] The specific process of S6.4 above is as follows: S6.4.1 Calculate the total fitness of the population : ; S6.4.2. Calculate the selection probability of each individual : ; S6.4.3. Calculate cumulative probability : ; S6.4.4. Generate a number in the range of [0, 1] using rand, and select an individual in ; S6.4.5. Create new individuals to supplement the deleted individuals.

[0011] The specific process of the above S6.6 is as follows: S6.6.1. Set the crossover probability pcross; S6.6.2. Randomly select two individuals in the population, with the formula as follows: ; where represents the position of an individual in the population, ceil is a rounding function, rand is a random number randomly generated between 0 and 1, and c is the number of individuals in the population; S6.6.3. Randomly generate the position of the crossover point, with the formula as follows: ; where , represents the crossover position, and k represents the length of the individual; S6.6.4. Start crossover, randomly generate a number p=rand. If p<pcross, start the crossover; if p>pcross, skip to the next step, that is, directly skip the mutation and crossover steps, and directly proceed to the recombination step.

[0012] The present invention provides a method for calibrating an anemoscope based on blade load data. A load model of a wind turbine unit based on force analysis is established to obtain the wind turbine load formula model under the current wind speed, pitch angle and meteorological environment, and the actual load is obtained by performing torque conversion on the sensor data at the blade root arranged at the blade root. The data with time stamps of the two are trained through a neural network model based on a genetic algorithm to obtain the influence weight of meteorological data on the load, and further obtain the compensation amount of the detection value of the wind speed sensor to obtain the real wind speed. After the model training and verification are completed, the wind speed compensation value can be directly obtained through the model and the blade root sensor value, so that other calculations based on wind speed are more accurate. Brief Description of the Drawings

[0013] The present invention will be further described below with reference to the accompanying drawings and embodiments: Figure 1 This is a schematic diagram of a wind turbine blade root sensor. Figure 2 yes Figure 1 A magnified view of a portion of the image; Figure 3 This is a schematic diagram of the blade root sensor projected onto a fixed hub coordinate system. Detailed Implementation

[0014] To make the objectives, technical solutions, and advantages of this invention clearer, the following will describe the specific technical solutions of this invention systematically and completely in conjunction with the accompanying drawings provided by this invention. Obviously, the described embodiments are only some embodiments of this invention, not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0015] Example 1: like Figure 1-3 As shown, a method for calibrating anemometers based on blade load data includes: S1. Install a blade root sensor at the blade root to detect the real-time changes in the load on the blade during torsion and tension, and obtain the blade root load. During initial testing, blade root sensors can be installed at all bolt locations. The data correlation is determined based on the force and position of the sensor at each bolt. In the actual wind field, based on test data, sensors are selected to be installed at a subset of bolt locations, typically three. Then, the data for the other bolt locations is calculated using the data correlation. Figure 1 and Figure 2 The data at each bolt shown was calculated from data from three blade root sensors; S2. Collect raw data and perform preprocessing; including setting up wind speed sensors at the wind turbine generator to collect real-time wind speed data, collecting the pitch angle of the wind turbine blades, collecting the azimuth information of the blades, collecting temperature and humidity data from the weather station, all raw data with time beacons, using the processed data as input values ​​and the actual wind speed compensation value as output values. S3. Establish a load model for the wind turbine generator based on stress analysis; S4. Convert the blade root load in S1 to the same load as the load model of the wind turbine generator set established in S3 and perform calculations to establish the equation function between the blade root load and the load model of the wind turbine generator set. S5. Divide the processed data in S2 into two sets of data, one set as the training set and the other set as the test set. The training set is used as the input sample of the neural prediction network. The correlation index between the theoretical load calculated by the load model of the wind turbine in S3 and the measured load converted from the blade root load in S4 is used as the output value to construct the neural network prediction model. S6. Use a genetic algorithm to optimize the weights and thresholds of the neural network, and use the error between the output value of the training sample and the actual value as the fitness function to obtain the prediction model with the minimum error. S7. Input the test set into the prediction model with the minimum error for prediction, and verify the results. If the error exceeds the limit, continue to increase the number of samples in the training set and repeat steps S5-S6 until the error does not exceed the limit, and obtain the prediction model for the wind speed compensation value. S8. Input the real-time data value into the prediction model of the wind speed compensation value verified in S7 to obtain the real-time wind speed compensation value. Add the wind speed compensation value to the detected wind speed value to obtain the real-time wind speed value.

[0016] The specific process of S2 mentioned above is as follows: S2.1. Take the meteorological data and parameters in the raw data with time stamps as samples. The meteorological data includes gas density, wind speed, humidity and temperature; the parameters include pitch angle and blade azimuth angle. S2.2. The correlation degree of meteorological factors is obtained through grey relational analysis, the weight of each meteorological factor is analyzed, and meteorological factors whose impact on wind turbine load and wind speed compensation is within the set value are screened out. The specific steps are as follows: S2.2.1 Suppose there are n sample data forming the following matrix, denoted as indicator data: ; n is the number of samples, and m is the number of indicators. ; S2.2.2 Determine the reference data series. The reference data series is a comparison standard. It can be constructed using the best or worst values ​​of each indicator, or other reference values ​​can be selected according to the evaluation purpose. It is denoted as: ; Calculate the absolute difference between the indicator data and the reference data: ; S2.2.3 Calculate the grey relational coefficient, using the following formula: ; Where ρ is the discrimination coefficient, 0 < ρ < 1. The smaller ρ is, the greater the difference between the correlation coefficients and the stronger the discrimination ability. S2.2.4 Calculate the grey weighted correlation degree, using the following formula: ; In the formula Let be the gray weighted correlation degree between the i-th evaluation object and the ideal object; S2.2.5. Based on the magnitude of the gray weighted correlation degree, each evaluation object is sorted to establish the correlation order of the evaluation objects; S2.3 Processing of the integrated meteorological dataset: The meteorological factors obtained from grey relational analysis are denoted as the comprehensive meteorological set. Because each meteorological factor corresponds to a correlation degree, the greater the correlation degree, the greater the impact of the meteorological factor on the wind turbine load. Using the influence level of each influencing factor as the input value, a new meteorological set is obtained after assigning values. Finally, it is processed as input using a formula, which is as follows: ; Where sum is the summation term. Indicates the corresponding time. v This indicates the corresponding wind speed compensation value.

[0017] By collecting a large amount of meteorological data and wind turbine parameters with time beacons as samples for training, and obtaining the correlation weight of each meteorological factor in determining the load through correlation analysis, we can prepare undefined calculation factors in advance for the load values ​​converted by blade root sensors to be used in the calculation later. Other parameters in the load model established by the force analysis can be directly obtained from sensors or motors. For example, the pitch angle can be read from the servo encoder of the independent pitch motor or from the circular grating installed on the pitch structure. The gas density can be obtained from the meteorological station. The blade radius, hub radius and other dimensional parameters, as well as the inference coefficient formula, can be provided by the manufacturer.

[0018] After processing the data as described above, the final uncertain factor, wind speed compensation value, is obtained as the output value after subsequent model verification.

[0019] The process of establishing the load model of the wind turbine generator based on force analysis in S3 above is as follows: Calculate the forces on a single blade, the first... i The thrust outside the plane of rotation on each blade can be expressed as: ; in, The air density is calculated from temperature and altitude. Where is the blade radius; This refers to the wind turbine rotation speed; Let be the thrust coefficient of the blade outside the plane of rotation, the magnitude of which is related to the relative wind speed acting on the blade. Pitch angle and rotational speed Its characteristics are specified by the manufacturer before it leaves the factory; subscript The serial numbers are for the three blades; The relative wind speed acting on the blades can be expressed as the actual wind speed. Vibration of the tower calculate: ; in, The tower height; This refers to the displacement of the tower top caused by the bending of the tower. The azimuth angle of the blade. With zero wind speed at the top of the blades (3 / 4 of the blade radius), and assuming the effective wind speed is at the point of application, the tower is considered a cantilever beam with a rotation angle to deflection ratio of 3 / 2. h ; The forces acting in the other two directions on the blade are the resultant forces of air force, gravity, and inertia: ; in, , These are the aerodynamic components in the y and z directions, respectively. , These are the gravitational components in the y and z directions, respectively. , These are the inertial force components in the y and z directions, respectively; The components of aerodynamic force are calculated as follows: ; in, , , where represents the thrust coefficients of the blades in two perpendicularly intersecting directions within the rotor's plane of rotation. Along the blade span direction, Perpendicular to the blade's span, its size is related to the relative wind speed acting on that blade. Pitch angle and rotational speed The relevant characteristics are specified by the manufacturer before the product leaves the factory; Due to air force The force is relatively small and therefore negligible; gravity always acts vertically downwards, and as the blades rotate, let... Let be the mass of the blade, then the gravitational component can be expressed as: ; The forces of inertia differ significantly along the y-axis and z-axis because It is mainly caused by acceleration. It is caused by centrifugal force: ; in, For the hub radius, The z-axis coordinate of the blade's center of mass; The torque on the blade is defined as follows: The torque that causes the wind turbine to rotate is not considered as a structural load; the torques in the other two directions are mainly due to aerodynamics, and can be expressed as: ; in, , These are the bending moment coefficients of the blades in two perpendicularly intersecting directions within the rotor's plane of rotation.

[0020] Along the blade span direction, Perpendicular to the blade's span, its size is related to the relative wind speed acting on that blade. Pitch angle and rotational speed The relevant characteristics are specified by the manufacturer before the product leaves the factory; After obtaining the load on the blade, the load in the rotating blade coordinate system is transformed to the fixed hub coordinate system: ; in, The mass of the wheel hub; set up Center of mass of wheel hub x Axis coordinates Let the component of the hub's center of mass projected onto a fixed hub coordinate system be the load on the blades. ; In addition to increasing the gravity of the motors and cabin arrive In addition to axle loads, the loads acting on the engine nacelle are the same as those acting on the wheel hubs: ; shaft and The shaft, also known as the tilting moment and deflection moment, can be expressed by the following formula: ; in, The center height of the wheel hub The distance between the top of the tower and the center of the wind turbine. and These are the components projected from the center of mass of the wheel hub onto the gravity of the motor and the engine compartment, respectively.

[0021] The blade root sensor in S1 described above is installed at the blade root bolt of each blade.

[0022] The specific process of S4 mentioned above is as follows: Let the value of the blade root sensor at each blade be... ,in Number the blades. If the blade root sensor is numbered, then the force it projects onto the tilting torque and deflection torque is... and They are respectively: ; in, and They are respectively The forces projected onto the y-axis and z-axis, where m is the number of root bolts for each blade; Then based on the tilting torque and deflection torque of the blade root sensor and They are respectively: ; in, The distance from the center of the wheel hub to the center of the tower; The equation function for the blade root load and the load model of the wind turbine generator is: ; in, and These are the linear coefficients of the equations for the tilting moment and the yaw moment, respectively. and These are the constants in the equations for tilting moment and deflection moment, respectively.

[0023] An equation model is established by creating a unit load model based on force analysis and the load model projected by the actual measured blade root sensor. The model is then trained using a neural network based on a genetic algorithm to obtain the linear coefficients and constants of the equation. After verification, real-time data can be input, and the wind speed compensation value can be obtained as the output.

[0024] The specific process of S6 mentioned above is as follows: S6.1 Encoding and Population Creation: Let the range of the population be [a, b], and the number of individuals in the population be... c The parameter is represented by a binary code of length k, which is randomly generated. Let the code of an individual in the population be k. Set a random number r = rand. If r < 0.5, then... =0, if r>0.5, then =1; and so on; S6.2 Decoding: The binary code of the known individual is... The corresponding decoding formula is: ; The decoded values ​​correspond to the weights and thresholds; S6.3, Evaluation of Individual Fitness: The sample data is fed into a BP neural network with weights and thresholds determined by the initial population for training. The fitness is calculated as the inverse of the mean squared error (MSE) of the output and the actual value corresponding to each sample. The fitness formula is as follows: ; In the formula For the first The actual value corresponding to each training sample; For the first The neural network determined by each individual is the first... The predicted output value of each training sample; the smaller the mean squared error, the higher the fitness. Taking the reciprocal of the mean squared error as the fitness increases the probability of selecting superior individuals. S6.4 Selection: Select individuals from the initial population, using fitness to select better individuals; S6.5 Saving Individuals: Sort the selected individuals by fitness, and save the individuals within the set value range in a loop to prevent negative changes in individuals during crossover and mutation; S6.6 Crossover: The crossover operation uses a single-point crossover operator, that is, there is only one crossover point position. Two individuals in the population are randomly selected and a crossover point position is randomly generated to exchange their genetic codes. S6.7 Mutation: The mutation operation uses uniform mutation, which means that individuals in the population will have a small probability of flipping the set value of their genes, from 0 to 1, or from 1 to 0. S6.8 Repeat steps S6.2 to S6.7 until the optimal weights and thresholds are obtained.

[0025] The specific process of S6.4 above is as follows: S6.4.1 Calculate the total fitness of the population : ; S6.4.2 Calculate the selection probability for each individual. : ; S6.4.3 Calculate the cumulative probability : ; S6.4.4. Generate numbers between [0,1] using rand. Select individuals from the list; S6.4.5. Create new individuals to supplement the deleted individuals.

[0026] The specific process of the above S6.6 is as follows: S6.6.1. Set the crossover probability pcross; S6.6.2. Randomly select two individuals in the population, with the formula as follows: ; where represents the position of the individual in the population, ceil is the rounding function, rand is a randomly generated random number between 0 and 1, and c is the number of populations; S6.6.3. Randomly generate the position of the crossover point, with the formula as follows: ; where , represents the crossover position, and k represents the length of the individual; S6.6.4. Start crossover, randomly generate a number p=rand. If p<pcross, start crossover; if p>pcross, skip to the next step, that is, directly skip the mutation and crossover steps and proceed directly to the recombination step.

Claims

1. A method for calibrating anemometers based on blade load data, characterized in that, include: S1. Install a blade root sensor at the blade root to detect the real-time changes in the load on the blade during torsion and tension, and obtain the blade root load. S2. Collect raw data and preprocess it; This includes installing wind speed sensors at wind turbine generators to collect real-time wind speed data, collecting pitch angle of wind turbine blades, collecting azimuth information of blades, collecting temperature and humidity data from weather stations, and using time beacons for all raw data. The processed raw data is used as input values, and the actual wind speed compensation value is used as output values. S3. Establish a load model for the wind turbine generator based on stress analysis; S4. Convert the blade root load in S1 to the same load as the load model of the wind turbine generator set established in S3 and perform calculations to establish the equation function between the blade root load and the load model of the wind turbine generator set. S5. Divide the raw data processed in S2 into two sets of data, one set as the training set and the other set as the test set. The training set is used as the input sample of the neural prediction network. The correlation index between the theoretical load calculated by the load model of the wind turbine in S3 and the measured load converted from the blade root load in S4 is used as the output value to construct the neural network prediction model. S6. Use a genetic algorithm to optimize the weights and thresholds of the neural network, and use the error between the output value of the training sample and the actual value as the fitness function to obtain the prediction model with the minimum error. S7. Input the test set into the prediction model with the minimum error for prediction, and verify the results. If the error exceeds the limit, continue to increase the number of samples in the training set and repeat steps S5-S6 until the error does not exceed the limit, and obtain the prediction model for the wind speed compensation value. S8. Input the real-time raw data value into the prediction model of the wind speed compensation value verified in S7 to obtain the real-time wind speed compensation value. Add the wind speed compensation value to the detected wind speed value to obtain the real-time wind speed value.

2. The anemometer calibration method based on blade load data according to claim 1, characterized in that, The specific process of S2 is as follows: S2.

1. Take the meteorological data and parameters in the raw data with time stamps as samples. The meteorological data includes gas density, wind speed, humidity and temperature; the parameters include pitch angle and blade azimuth angle. S2.

2. The correlation degree of meteorological factors is obtained through grey relational analysis, the weight of each meteorological factor is analyzed, and meteorological factors whose impact on wind turbine load and wind speed compensation is within the set value are screened out. The specific steps are as follows: S2.2.1 Suppose there are n sample data forming the following matrix, denoted as indicator data: ； n is the number of samples, and m is the number of indicators. ; S2.2.2 Determine the reference sequence, denoted as: ； Calculate the absolute difference between the indicator data and the reference data: ； S2.2.3 Calculate the grey relational coefficient, using the following formula: ； Where ρ is the discrimination coefficient, 0 < ρ < 1. The smaller ρ is, the greater the difference between the correlation coefficients and the stronger the discrimination ability. S2.2.4 Calculate the grey weighted correlation degree, using the following formula: ； In the formula Let be the grey weighted correlation degree between the i-th evaluation object and the ideal object; S2.2.

5. Based on the magnitude of the gray weighted correlation degree, the evaluation objects are sorted to establish the correlation order of the evaluation objects; S2.3 Processing of the integrated meteorological dataset: The meteorological factors obtained from grey relational analysis are denoted as the comprehensive meteorological set. Because each meteorological factor corresponds to a correlation degree, the greater the correlation degree, the greater the impact of the meteorological factor on the wind turbine load. Using the influence level of each influencing factor as the input value, a new meteorological set is obtained after assigning values. Finally, it is processed as input using a formula, which is as follows: ； Where sum is the summation term. Indicates the corresponding time. v This indicates the corresponding wind speed compensation value.

3. The anemometer calibration method based on blade load data according to claim 2, characterized in that, The process of establishing the load model of the wind turbine generator based on force analysis in S3 is as follows: Calculate the forces on a single blade, the first... i The thrust outside the plane of rotation on each blade can be expressed as: ； in, The air density is calculated from temperature and altitude. Where is the blade radius; This refers to the wind turbine rotation speed; Let be the thrust coefficient of the blade outside the plane of rotation, the magnitude of which is related to the relative wind speed acting on the blade. Pitch angle and rotational speed Related to; subscript The serial numbers are for the three blades; The relative wind speed acting on the blades can be expressed as the actual wind speed. Vibration of the tower calculate: ； in, The tower height; This refers to the displacement of the tower top caused by the bending of the tower. The azimuth angle of the blade. The value is zero when the blade is at its vertical top. The forces acting in the other two directions on the blade are the resultant forces of air force, gravity, and inertia: ； in, , These are the aerodynamic components in the y and z directions, respectively. , These are the gravitational components in the y and z directions, respectively. , These are the inertial force components in the y and z directions, respectively; The components of aerodynamic force are calculated as follows: ； in, , , where represents the thrust coefficients of the blades in two perpendicularly intersecting directions within the rotor's plane of rotation. Along the blade span direction, Perpendicular to the blade's span, its size is related to the relative wind speed acting on that blade. Pitch angle and rotational speed related; Gravity always acts vertically downwards. As the blades rotate, let... Let be the mass of the blade, then the gravitational component can be expressed as: ； It is caused by acceleration. It is caused by centrifugal force: ； in, For the hub radius, The z-axis coordinate of the blade's center of mass; The torque on the blade is defined as follows: The torque that causes the wind turbine to rotate is not considered as a structural load; the torques in the other two directions are mainly due to aerodynamics, and can be expressed as: ； in, , These are the bending moment coefficients of the blades in two perpendicularly intersecting directions within the rotor's plane of rotation. Along the blade span direction, Perpendicular to the blade's span, its size is related to the relative wind speed acting on that blade. Pitch angle and rotational speed related; After obtaining the load on the blade, the load in the rotating blade coordinate system is transformed to the fixed hub coordinate system: ; in, The mass of the wheel hub; set up Center of mass of wheel hub x Axis coordinates Let the component of the hub's center of mass projected onto a fixed hub coordinate system be the load on the blades. ； In addition to increasing the gravity of the motors and cabin arrive In addition to axle loads, the loads acting on the engine nacelle are the same as those acting on the wheel hubs: ; shaft and The shaft, also known as the tilting moment and deflection moment, can be expressed by the following formula: ； in, The center height of the wheel hub The distance between the top of the tower and the center of the wind turbine. and These are the components projected from the center of mass of the wheel hub onto the gravity of the motor and the engine compartment, respectively.

4. The anemometer calibration method based on blade load data according to claim 3, characterized in that, The blade root sensor in S1 is installed at the blade root bolt of each blade.

5. The anemometer calibration method based on blade load data according to claim 4, characterized in that, The specific process of S4 is as follows: Let the value of the blade root sensor at each blade be... ,in Number the blades. If the blade root sensor is numbered, then the force it projects onto the tilting torque and deflection torque is... and They are respectively: ; in, and They are respectively The forces projected onto the y-axis and z-axis, where m is the number of root bolts for each blade; Then based on the tilting torque and deflection torque of the blade root sensor and They are respectively: ； in, The distance from the center of the wheel hub to the center of the tower; The equation function for the blade root load and the load model of the wind turbine generator is: ; in, and These are the linear coefficients of the equations for the tilting moment and the yaw moment, respectively. and These are the constants in the equations for tilting moment and deflection moment, respectively.

6. The anemometer calibration method based on blade load data according to claim 5, characterized in that, The specific process of S6 is as follows: S6.1 Encoding and Population Creation: Let the range of the population be [a, b], and the number of individuals in the population be... c The parameter is represented by a binary code of length k, which is randomly generated. Let the code of an individual in the population be k. Set a random number r = rand. If r < 0.5, then... =0, if r>0.5, then =1; and so on; S6.2 Decoding: The binary code of the known individual is... The corresponding decoding formula is: ； The decoded values ​​correspond to the weights and thresholds; S6.3, Evaluation of Individual Fitness: The sample data is fed into a BP neural network with weights and thresholds determined by the initial population for training. The fitness is calculated as the inverse of the mean squared error (MSE) of the output and the actual value corresponding to each sample. The fitness formula is as follows: In the formula For the first The actual value corresponding to each training sample; For the first The neural network determined by each individual is the first... The predicted output value of each training sample; the smaller the mean squared error, the higher the fitness. Taking the reciprocal of the mean squared error as the fitness increases the probability of selecting superior individuals. S6.4 Selection: Select individuals from the initial population, using fitness to select better individuals; S6.5 Saving Individuals: Sort the selected individuals by fitness, and save the individuals within the set value range in a loop to prevent negative changes in individuals during crossover and mutation; S6.6 Crossover: The crossover operation uses a single-point crossover operator, that is, there is only one crossover point position. Two individuals in the population are randomly selected and a crossover point position is randomly generated to exchange their genetic codes. S6.7 Mutation: The mutation operation uses uniform mutation, which means that individuals in the population will have a small probability of flipping the set value of their genes, from 0 to 1, or from 1 to 0. S6.8 Repeat steps S6.2 to S6.7 until the optimal weights and thresholds are obtained.

7. The anemometer calibration method based on blade load data according to claim 6, characterized in that, The specific process of S6.4 is as follows: S6.4.1 Calculate the total fitness of the population : ； S6.4.2 Calculate the selection probability for each individual. : ； S6.4.3 Calculate the cumulative probability : ； S6.4.

4. Generate numbers between [0,1] using rand. Select individuals from the list; S6.4.

5. Create new individuals to replace the deleted individuals.

Images