Complete machine load estimation control method for wind turbine generator

Abstract

The invention provides a complete machine load estimation control method for a wind turbine generator. The method comprises the steps that S1, original data are collected and preprocessed; s2, establishing a load model of the wind generating set based on stress analysis; s3, establishing a load prediction model of the wind generating set based on the variable-pitch motor data; s4, constructing a neural network prediction model; s5, using a genetic algorithm to optimize the weight and the threshold of the neural network for training; and S6, adjusting. Through one-time sensor investment and calculation of a large amount of calculation and training work in the early stage, an actual model for reversely deducing a fan load from parameters of a variable-pitch motor and a generator is obtained and is transmitted to a main controller, so that convenient load reduction control can be realized, subsequent repeated investment can be omitted at a wind turbine generator, a large amount of hardware investment is saved, and the cost is reduced. Adjustment is carried out through continuous monitoring and continuous verification modeling in the later period, and rapid load estimation and control of the whole life cycle of the wind turbine generator can be achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of wind power, and in particular to a method for estimating and controlling the load of a wind turbine set. Background Art

[0002] In the wind power industry, with the increasingly significant trend of large-scale wind turbines and the increasing pressure to reduce costs, how to operate wind turbines efficiently and safely has become an important issue that needs to be solved in the industry. During operation, wind turbines are affected by a variety of dynamic loads, including wind loads, mechanical loads, and transient loads caused by turbulence. These loads have a direct impact on the stable operation, service life, and maintenance costs of wind turbines. Therefore, accurately estimating the load of wind turbines and actively implementing load reduction measures based on the estimation results have become key means to improve the competitiveness and economic benefits of wind turbines.

[0003] At present, most traditional wind turbine load estimation methods rely on offline analysis and additional sensor equipment. These methods are not only costly, but also difficult to reflect the dynamic changes of the unit's operating status in real time. Offline analysis is usually based on historical data and cannot capture the real-time load changes in the current unit operation; and the additional installed sensor equipment increases the complexity and maintenance cost of the system. In addition, the current load calculation also requires different load states according to different standards for different turbulent wind conditions, and the load calculation is complex and difficult.

[0004] Currently, many stress monitoring sensors are set at the root of the blade to obtain the changes in the load on the blade during torsion and tension. However, this load often has a large error with the actual load because it ignores the actual transmission conditions of the wind turbine.

[0005] Therefore, a simple and efficient real-time monitoring and regulation method of wind turbine load is needed to reasonably control and regulate the wind turbine. Summary of the invention

[0006] The present invention provides a method for estimating and controlling the load of a wind turbine generator set, which can perform real-time and efficient calculation and load control on the load of the wind turbine generator set.

[0007] In order to solve the above technical problems, the technical solution adopted by the present invention is: A wind turbine load estimation and control method, comprising: S1. Collecting raw data and preprocessing it; including setting a gas density sensor at the wind turbine generator set to collect real-time data on gas density, setting a wind speed sensor to collect real-time data on wind speed, collecting real-time voltage and current of the generator, speed data of the wind wheel, collecting real-time speed, voltage and current of the variable pitch motor, collecting the variable pitch angle of the wind blade, collecting the azimuth information of the blade, and all raw data are accompanied by a time beacon; S2. Establish a load model of the wind turbine generator set based on force analysis; S3, establishing a load prediction model for wind turbines based on variable pitch motor data; S4, divide the preprocessed raw data in S1 into two groups of data, one group as a training set and the other group as a test set, the training set is used as the input sample of the neural prediction network, and the correlation index of the load prediction model in S3 corresponding to the load model in S2 is used as the output value to construct a neural network prediction model; S5, using a genetic algorithm to optimize the weights and thresholds of the neural network, using the error value between the output value of the training sample and the actual value as the fitness function, so as to obtain the minimum error model of the load prediction model based on the pitch motor data in S3, so that the error between it and the load model in S2 is minimized; S6. Input the test set into the minimum error model of the load prediction model based on the variable pitch motor data in S3, compare the predicted load with the load calculated by the load model of the wind turbine generator set based on the force analysis in S2 with the same time beacon, verify the correctness of the minimum error model of the load prediction model based on the variable pitch motor data. If the error exceeds the limit, continue to increase the number of samples in the training set, repeat steps S4-S6 until the error does not exceed the limit, and obtain the trained load prediction model for the wind turbine generator set based on the variable pitch motor data.

[0008] The above load estimation control method further includes: S7. The real-time variable pitch motor data is brought into the load prediction model of the wind turbine generator set based on the variable pitch motor data trained in S6 to obtain the real-time load of the wind turbine generator set. The real-time load data and the variable pitch angle of the wind blades and the real-time wind speed data are brought into the main control PLC of the wind turbine generator set, and the main control PLC performs load reduction control based on the load calculation results.

[0009] The process of establishing the load model of the wind turbine generator set based on the force analysis in the above S2 is as follows: Calculate the forces on a single blade, i The thrust out of the rotating plane on each blade can be expressed as: ; in, is the air density; is the blade radius; is the wind wheel speed; is the thrust coefficient of the blade outside the rotation plane, and its magnitude is related to the relative wind speed acting on the blade. , pitch angle and speed Related to; Subscript is the serial number of the 3 leaves; The relative wind speed acting on the blades can be expressed as and the tower vibrates forward and backward calculate: ; in, is the tower height; The displacement of the tower top caused by the bending of the tower; is the azimuth of the blade, It is zero when the blade is at the vertical top; The other two forces on the blade are the combined force of air force, gravity and inertia: ; in, , are the air force components in the y and z directions, , are the gravity components in the y and z directions, respectively, , are the inertial force components in the y and z directions respectively; The air force component is calculated as: ; in, , is the thrust coefficient of the blade in two directions perpendicular to each other in the rotating plane of the wind rotor, Along the blade span direction, Perpendicular to the blade span, its size is related to the relative wind speed acting on the blade , pitch angle and speed related; Gravity always points vertically downward. As the blades rotate, is the mass of the blade, the gravity component can be expressed as: ; It is caused by acceleration. It is caused by centrifugal force: ; in, is the hub radius, is the z-axis coordinate of the blade mass center; The moment on the blade is defined as follows, It is the moment that causes the wind wheel to rotate, so it is not considered as a structural load; the moments in the other two directions are mainly caused by aerodynamics, and they can be expressed as: ; in, , is the bending moment coefficient of the blade in two directions perpendicular to each other in the rotating plane of the wind rotor, Along the blade span direction, Perpendicular to the blade span, its size is related to the relative wind speed acting on the blade , pitch angle and speed related; After obtaining the load on the blade, convert the load in the rotating blade coordinate system to the fixed hub coordinate system: ; in, is the mass of the wheel hub; set up The wheel mass center x Axis coordinates, is the component of the hub mass center projected onto the fixed hub coordinate system, then the bending moment on the hub can be expressed by the blade load as: ; In addition to increasing the weight of the motor and cabin arrive In addition to the shaft loads, the loads acting on the nacelle are the same as those acting on the hub: ; Axis and The axis is also called the tilting moment and the yaw moment and can be expressed as follows: ; in, is the center height of the hub, is the distance between the tower top and the center of the wind rotor, and They are the components of the hub mass center projected onto the motor and nacelle gravity respectively.

[0010] When the pitch control mode is unified pitch control, the process of establishing the load prediction model of the wind turbine generator set based on the pitch motor data in the above S3 is as follows: Torque load of pitch motor for: ; in, is the real-time voltage of the pitch motor, I is the real-time current of the pitch motor, is the real-time speed of the pitch motor; Assume that the energy efficiency of the pitch motor is The overall transmission efficiency from the pitch motor to the blade bearing is , the pitch motor torque is used as input to drive the blade pitch angle, and the torque transmitted to each blade is for: ; The real-time voltage of the generator is , the real-time current is , the real-time speed is , then the real-time moment load of the generator is : ; Assume the energy efficiency of the generator is , the overall transmission efficiency from the generator shaft to the generator is , the generator torque is generated by wind input, so the real-time load at the hub calculated from the generator torque load is: ; In the formula, is the real-time load at the thrust-reverse hub.

[0011] When the pitch control mode is independent pitch control, the process of establishing the load prediction model of the wind turbine generator set based on the pitch motor data in the above S3 is: Torque load of pitch motor for: ; in, is the real-time voltage of the pitch motor, is the real-time current of the pitch motor, is the real-time speed of the pitch motor; Assume that the energy efficiency of the pitch motor is The overall transmission efficiency from the pitch motor to the blade bearing is , the pitch motor torque is used as input to drive the blade pitch angle, and the torque transmitted to each blade is for: ; The real-time voltage of the generator is , the real-time current is , the real-time speed is , then the real-time torque load of the generator is : ; Assume the energy efficiency of the generator is , the overall transmission efficiency from the generator shaft to the generator is , the generator torque is generated by wind input, so the real-time load at the hub calculated from the generator torque load is: ; In the formula, is the real-time load at the thrust-reverse hub.

[0012] The specific steps of the above S5 are: S5.1. Encode and create a population; suppose the range of the population is [a, b], and the number of individuals in the population is c , the parameter is represented by a binary code of length k, the binary code is randomly created, and the code of a population individual is , set the random number a=rand, if a<0.5, then =0, if a>0.5, then =1; and so on; S5.2. Decoding: The binary code of a known individual is , then the corresponding decoding formula is: ; The decoded values ​​correspond to weights and thresholds; S5.3, Detection and evaluation of individual fitness: The test set data is brought into the BP neural network with weights and thresholds determined by the initial population for training. The objective function of the training is: ; ; in, and are the linear coefficients for inverting the actual blade load and hub load from the load calculated by the motor, and They are the calculation constants for inverting the actual blade load and hub load from the calculated load of the motor; The inverse of the mean square error (MSE) between the output result and the actual value corresponding to each sample is taken as the fitness. The fitness formula is as follows: ; In the formula is the actual value corresponding to each sample, is the output value obtained by the neural network. The smaller the mean square error value, the higher the fitness. Considering the selection probability in the selection step, taking the inverse of the mean square error as the fitness, the selection probability of excellent individuals becomes higher; S5.4. Select individuals from the initial population and select better individuals based on fitness; S5.5, save individuals; sort the selected individuals by fitness, select individuals within the set value range and save them in a loop to avoid negative changes in individuals during crossover and mutation; S5.6, crossover; the crossover operation uses a single-point crossover operator, that is, there is only one crossover point position, and two individuals in the population are randomly selected and a crossover point position is randomly generated to exchange gene codes; S5.7, mutation; the mutation operation uses uniform mutation, that is, the individuals in the population realize the small probability of the set value of the gene flipping, from 0 to 1, from 1 to 0; S5.8. Repeat steps S5.1 to S5.7 until the optimal weight and threshold are obtained.

[0013] The specific steps of S5.4 above are: S5.4.1. Calculate the total fitness of the group : ; S5.4.2. Calculate the selection probability of each individual : ; S5.4.3. Calculation of cumulative probability : ; S5.4.4, use rand to generate numbers between [0,1], Select individuals from S5.4.5. Create new individuals to complete the deleted individuals.

[0014] The specific steps of S5.6 above are: S5.6.1, set the crossover probability pcross; S5.6.2. Randomly select two individuals from the population. The formula is as follows: ; in Indicates the position of an individual in a population, ceil is a rounding function, rand is a random number between 0 and 1, and c is the number of populations; S5.6.3. Randomly generate intersection positions using the following formula: ; in , It represents the position of the crossover, and k represents the length of the individual; S5.6.4. 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 enter the recombination step directly.

[0015] The present invention provides a method for estimating and controlling the overall load of a wind turbine. By using different coordinate systems, calculation models for the blade load and hub load of the wind turbine are deduced. Based on the real-time characteristic parameters of the pitch motor and the generator, the inverse values of the blade load and hub load based on the motor parameters are calculated. By setting different sensors at the wind turbine, the actual value of the load is calculated. Combining with a neural network model based on the genetic algorithm, the correlation index between the two models is calculated and verified, thereby obtaining an actual model for inferring the wind turbine load from the parameters of the pitch motor and the generator. According to the real-time parameters of the motor parameters, the wind turbine load is calculated and transmitted to the main controller to achieve convenient load reduction control. Through the one-time sensor investment and calculation of a large amount of previous calculation and training work, subsequent repeated investments can be omitted at the wind turbine, saving a large amount of hardware investment. Later, through continuous monitoring and continued verification and adjustment of the modeling, fast load estimation and control throughout the life cycle of the wind turbine can be achieved. Brief Description of the Drawings

[0016] The following further describes the present invention in conjunction with the drawings and embodiments: Figure 1 It is a schematic diagram based on the rotating blade coordinate system; Figure 2 It is a schematic diagram based on the fixed hub coordinate system. Detailed Embodiments

[0017] To make the objectives, technical solutions, and advantages of the present invention clearer, the following content will systematically and completely describe the specific technical solutions of the present invention in conjunction with the drawings provided according to the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0018] Embodiment 1: A method for estimating and controlling the overall load of a wind turbine includes: S1. Collect and preprocess the original data; including setting a gas density sensor at the wind power generation unit to collect real-time gas density data, setting a wind speed sensor to collect real-time wind speed data, collecting the real-time voltage and current of the generator, the rotational speed data of the wind wheel, collecting the real-time rotational speed, voltage, and current of the pitch motor, collecting the pitch angle of the wind turbine blade, and collecting the azimuth angle information of the blade. All the original data has a time beacon; S2. Establish a load model of the wind turbine generator set based on force analysis; S3, establishing a load prediction model for wind turbines based on variable pitch motor data; S4, divide the preprocessed raw data in S1 into two groups of data, one group as a training set and the other group as a test set, the training set is used as the input sample of the neural prediction network, and the correlation index of the load prediction model in S3 corresponding to the load model in S2 is used as the output value to construct a neural network prediction model; Through the data collection in S1 and the load model in S2, the actual blade load and hub load with time stamp under different wind conditions and blade angle control can be calculated. The reverse thrust load model based on motor parameters is established through S3. The connection between the two models can be established by combining the neural network based on genetic algorithm, and the correlation coefficient between the two models can be calculated through training, so that the actual load can be directly inferred from the motor parameters, which is convenient for control. S5, using a genetic algorithm to optimize the weights and thresholds of the neural network, using the error value between the output value of the training sample and the actual value as the fitness function, so as to obtain the minimum error model of the load prediction model based on the pitch motor data in S3, so that the error between it and the load model in S2 is minimized; S6. Input the test set into the minimum error model of the load prediction model based on the variable pitch motor data in S3, compare the predicted load with the load calculated by the load model of the wind turbine generator set based on the force analysis in S2 with the same time beacon, verify the correctness of the minimum error model of the load prediction model based on the variable pitch motor data. If the error exceeds the limit, continue to increase the number of samples in the training set, repeat steps S4-S6 until the error does not exceed the limit, and obtain the trained load prediction model for the wind turbine generator set based on the variable pitch motor data.

[0019] The above load estimation control method further includes: S7. The real-time variable pitch motor data is brought into the load prediction model of the wind turbine generator set based on the variable pitch motor data trained in S6 to obtain the real-time load of the wind turbine generator set. The real-time load data and the variable pitch angle of the wind blades and the real-time wind speed data are brought into the main control PLC of the wind turbine generator set, and the main control PLC performs load reduction control based on the load calculation results.

[0020] The process of establishing the load model of the wind turbine generator set based on the force analysis in the above S2 is as follows: Calculate the forces on a single blade, i The thrust out of the rotating plane on each blade can be expressed as: ; in, is the air density; is the blade radius; is the wind wheel speed; is the thrust coefficient of the blade outside the rotation plane, and its magnitude is related to the relative wind speed acting on the blade. , pitch angle and speed The characteristics are given by the manufacturer before leaving the factory; is the serial number of the 3 leaves; The relative wind speed acting on the blades can be expressed as and the tower vibrates forward and backward calculate: ; in, is the tower height; The displacement of the tower top caused by the bending of the tower; is the azimuth of the blade, When the blade is at the top of the vertical, it is zero. Assuming that the wind speed at 3 / 4 of the blade radius is the effective wind speed position, the tower is regarded as a cantilever beam, and the ratio of its rotation angle to deflection is 3 / 2. h ; The other two forces on the blade are the combined force of air force, gravity and inertia: ; in, , are the air force components in the y and z directions, , are the gravity components in the y and z directions, respectively, , are the inertial force components in the y and z directions respectively; The air force component is calculated as: ; in, , is the thrust coefficient of the blade in two directions perpendicular to each other in the rotating plane of the wind rotor, Along the blade span direction, Perpendicular to the blade span, its size is related to the relative wind speed acting on the blade , pitch angle and speed The characteristics are given by the manufacturer before leaving the factory; Due to air force is small, so it can be ignored; gravity is always vertically downward, as the blades rotate, is the mass of the blade, the gravity component can be expressed as: ; The inertial force is quite different on the y-axis and z-axis because Mainly caused by acceleration, It is caused by centrifugal force: ; in, is the hub radius, is the z-axis coordinate of the blade mass center; The moment on the blade is defined as follows, It is the moment that causes the wind wheel to rotate, so it is not considered as a structural load; the moments in the other two directions are mainly caused by aerodynamics, and they can be expressed as: ; in, , is the bending moment coefficient of the blade in two directions perpendicular to each other in the rotating plane of the wind rotor, Along the blade span direction, Perpendicular to the blade span, its size is related to the relative wind speed acting on the blade , pitch angle and speed The characteristics are given by the manufacturer before leaving the factory; After obtaining the load on the blade, convert the load in the rotating blade coordinate system to the fixed hub coordinate system: ; in, is the mass of the wheel hub; set up The wheel mass center x Axis coordinates, is the component of the hub mass center projected onto the fixed hub coordinate system, then the bending moment on the hub can be expressed by the blade load as: ; In addition to increasing the weight of the motor and cabin arrive In addition to the shaft loads, the loads acting on the nacelle are the same as those acting on the hub: ; Axis and The axis is also called the tilting moment and the yaw moment and can be expressed as follows: ; in, is the center height of the hub, is the distance between the tower top and the center of the wind rotor, and They are the components of the hub mass center projected onto the motor and nacelle gravity respectively.

[0021] When the pitch control mode is unified pitch control, the process of establishing the load prediction model of the wind turbine generator set based on the pitch motor data in the above S3 is: Torque load of pitch motor for: ; in, is the real-time voltage of the pitch motor, I is the real-time current of the pitch motor, is the real-time speed of the pitch motor; Assume that the energy efficiency of the pitch motor is The overall transmission efficiency from the pitch motor to the blade bearing is , the pitch motor torque is used as input to drive the blade pitch angle, and the torque transmitted to each blade is for: ; The real-time voltage of the generator is , the real-time current is , the real-time speed is , then the real-time moment load of the generator is : ; Assume the energy efficiency of the generator is , the overall transmission efficiency from the generator shaft to the generator is , the generator torque is generated by wind input, so the real-time load at the hub calculated from the generator torque load is: ; In the formula, is the real-time load at the thrust-reverse hub.

[0022] When the pitch control mode is independent pitch control, the process of establishing the load prediction model of the wind turbine generator set based on the pitch motor data in the above S3 is as follows: Torque load of pitch motor for: ; in, is the real-time voltage of the pitch motor, is the real-time current of the pitch motor, is the real-time speed of the pitch motor; Assume that the energy efficiency of the pitch motor is The overall transmission efficiency from the pitch motor to the blade bearing is , the pitch motor torque is used as input to drive the blade pitch angle, and the torque transmitted to each blade is for: ; The real-time voltage of the generator is , the real-time current is , the real-time speed is , then the real-time moment load of the generator is : ; Assume the energy efficiency of the generator is , the overall transmission efficiency from the generator shaft to the generator is , the generator torque is generated by wind input, so the real-time load at the hub calculated from the generator torque load is: ; In the formula, is the real-time load at the thrust-reverse hub.

[0023] The specific steps of the above S5 are: S5.1. Encode and create a population; suppose the range of the population is [a, b], and the number of individuals in the population is c , the parameter is represented by a binary code of length k, the binary code is randomly created, and the code of a population individual is , set the random number a=rand, if a<0.5, then =0, if a>0.5, then =1; and so on; S5.2. Decoding: The binary code of a known individual is , then the corresponding decoding formula is: ; The decoded values ​​correspond to weights and thresholds; S5.3, Detection and evaluation of individual fitness: The test set data is brought into the BP neural network with weights and thresholds determined by the initial population for training. The objective function of the training is: ; ; in, and are the linear coefficients for inverting the actual blade load and hub load from the load calculated by the motor, and They are the calculation constants for inverting the actual blade load and hub load from the calculated load of the motor; The inverse of the mean square error (MSE) between the output result and the actual value corresponding to each sample is taken as the fitness. The fitness formula is as follows: ; In the formula is the actual value corresponding to each sample, is the output value obtained by the neural network. The smaller the mean square error value, the higher the fitness. Considering the selection probability in the selection step, taking the inverse of the mean square error as the fitness, the selection probability of excellent individuals becomes higher; S5.4. Select individuals from the initial population and select better individuals based on fitness; S5.5, save individuals; sort the selected individuals by fitness, select individuals within the set value range and save them in a loop to avoid negative changes in individuals during crossover and mutation; S5.6, crossover; the crossover operation uses a single-point crossover operator, that is, there is only one crossover point position, and two individuals in the population are randomly selected and a crossover point position is randomly generated to exchange gene codes; S5.7, mutation; the mutation operation uses uniform mutation, that is, the individuals in the population realize the small probability of the set value of the gene flipping, from 0 to 1, from 1 to 0; S5.8. Repeat steps S5.1 to S5.7 until the optimal weight and threshold are obtained.

[0024] The specific steps of S5.4 above are: S5.4.1. Calculate the total fitness of the group : ; S5.4.2. Calculate the selection probability of each individual : ; S5.4.3. Calculation of cumulative probability : ; S5.4.4, use rand to generate numbers between [0,1], Select individuals from S5.4.5. Create new individuals to complete the deleted individuals.

[0025] The specific steps of S5.6 above are: S5.6.1, set the crossover probability pcross; S5.6.2. Randomly select two individuals from the population. The formula is as follows: ; in represents the position of an individual in the population, ceil is the rounding function, rand is a random number generated randomly between 0 and 1, and c is the number of populations; S5.6.3. Randomly generate the position of the crossover point, and the formula is as follows: ; where , represents the position of the crossover, and k represents the length of the individual; S5.6.4. Start the 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 crossover step and directly enter the recombination step.

Claims

1. A method for estimating and controlling the load of a wind turbine generator set, characterized in that: include: S1, collect raw data and preprocess them; It includes setting a gas density sensor at the wind turbine generator set to collect real-time data on gas density, setting a wind speed sensor to collect real-time data on wind speed, collecting real-time voltage and current of the generator, speed data of the wind wheel, collecting real-time speed, voltage and current of the variable pitch motor, collecting the variable pitch angle of the wind blade, collecting the azimuth angle information of the blade, and all the original data are accompanied by a time beacon; S2. Establish a load model of the wind turbine generator set based on force analysis; S3, establishing a load prediction model for wind turbines based on variable pitch motor data; S4, divide the preprocessed raw data in S1 into two groups of data, one group as a training set and the other group as a test set, the training set is used as the input sample of the neural prediction network, and the correlation index of the load prediction model in S3 corresponding to the load model in S2 is used as the output value to construct a neural network prediction model; S5, using a genetic algorithm to optimize the weights and thresholds of the neural network, using the error value between the output value of the training sample and the actual value as the fitness function, so as to obtain the minimum error model of the load prediction model based on the pitch motor data in S3, so that the error between it and the load model in S2 is minimized; S6. Input the test set into the minimum error model of the load prediction model based on the variable pitch motor data in S3, compare the predicted load with the load calculated by the load model of the wind turbine generator set based on the force analysis in S2 with the same time beacon, verify the correctness of the minimum error model of the load prediction model based on the variable pitch motor data. If the error exceeds the limit, continue to increase the number of samples in the training set, repeat steps S4-S6 until the error does not exceed the limit, and obtain the trained load prediction model for the wind turbine generator set based on the variable pitch motor data.

2. According to the method for estimating and controlling the load of a wind turbine generator set as claimed in claim 1, it is characterized in that: The load estimation control method further includes: S7. The real-time variable pitch motor data is brought into the load prediction model of the wind turbine generator set based on the variable pitch motor data trained in S6 to obtain the real-time load of the wind turbine generator set. The real-time load data and the variable pitch angle of the wind blades and the real-time wind speed data are brought into the main control PLC of the wind turbine generator set, and the main control PLC performs load reduction control based on the load calculation results.

3. According to the method for estimating and controlling the load of a wind turbine generator set as claimed in claim 2, it is characterized in that: The process of establishing the load model of the wind turbine generator set based on the force analysis in S2 is as follows: Calculate the forces on a single blade, i The thrust out of the rotating plane on each blade can be expressed as: ； in, is the air density; is the blade radius; is the wind wheel speed; is the thrust coefficient of the blade outside the rotation plane, and its magnitude is related to the relative wind speed acting on the blade. , pitch angle and speed Related to; Subscript is the serial number of the 3 leaves; The relative wind speed acting on the blades can be expressed as and the tower vibrates forward and backward calculate: ； in, is the tower height; The displacement of the tower top caused by the bending of the tower; is the azimuth of the blade, It is zero when the blade is at the vertical top; The other two forces on the blade are the combined force of air force, gravity and inertia: ； in, , are the air force components in the y and z directions, , are the gravity components in the y and z directions, , are the inertial force components in the y and z directions respectively; The air force component is calculated as: ； in, , is the thrust coefficient of the blade in two directions perpendicular to each other in the rotating plane of the wind rotor, Along the blade span direction, Perpendicular to the blade span, its size is related to the relative wind speed acting on the blade , pitch angle and speed related; Gravity always points vertically downward. As the blades rotate, is the mass of the blade, the gravity component can be expressed as: ； It is caused by acceleration. It is caused by centrifugal force: ； in, is the wheel hub radius, is the z-axis coordinate of the blade mass center; The moment on the blade is defined as follows, It is the moment that causes the wind wheel to rotate, so it is not considered as a structural load; the moments in the other two directions are mainly caused by aerodynamics, and they can be expressed as: ； in, , is the bending moment coefficient of the blade in two directions perpendicular to each other in the rotating plane of the wind rotor, Along the blade span direction, Perpendicular to the blade span, its size is related to the relative wind speed acting on the blade , pitch angle and speed related; After obtaining the load on the blade, convert the load in the rotating blade coordinate system to the fixed hub coordinate system: ; in, is the mass of the wheel hub; set up The wheel mass center x Axis coordinates, is the component of the hub mass center projected onto the fixed hub coordinate system, then the bending moment on the hub can be expressed by the blade load as: ； In addition to increasing the weight of the motor and cabin arrive In addition to the shaft loads, the loads acting on the nacelle are the same as those acting on the hub: ; Axis and The axis is also called the tilting moment and the yaw moment and can be expressed as follows: ； in, is the center height of the hub, is the distance between the tower top and the center of the wind rotor, and They are the components of the hub mass center projected onto the motor and nacelle gravity respectively.

4. A method for estimating and controlling the load of a wind turbine generator system according to claim 3, characterized in that: The process of establishing the load prediction model of the wind turbine generator set based on the variable pitch motor data in S3 is as follows: When the pitch control mode is unified pitch control, the torque load of the pitch motor is for: ； in, is the real-time voltage of the pitch motor, I is the real-time current of the pitch motor, is the real-time speed of the pitch motor; Assume that the energy efficiency of the pitch motor is The overall transmission efficiency from the pitch motor to the blade bearing is , the pitch motor torque is used as input to drive the blade pitch angle, and the torque transmitted to each blade is for: ； The real-time voltage of the generator is , the real-time current is , the real-time speed is , then the real-time moment load of the generator is : ； Assume the energy efficiency of the generator is , the overall transmission efficiency from the generator shaft to the generator is , the generator torque is generated by wind input, so the real-time load at the hub calculated from the generator torque load is: ； In the formula, is the real-time load at the thrust-reverse hub.

5. A wind turbine load estimation and control method according to claim 3, characterized in that: The process of establishing the load prediction model of the wind turbine generator set based on the variable pitch motor data in S3 is as follows: When the pitch control mode is independent pitch control, the torque load of the pitch motor for: ； in, is the real-time voltage of the pitch motor, is the real-time current of the pitch motor, is the real-time speed of the pitch motor; Assume that the energy efficiency of the pitch motor is The overall transmission efficiency from the pitch motor to the blade bearing is , the pitch motor torque is used as input to drive the blade pitch angle, and the torque transmitted to each blade is for: ； The real-time voltage of the generator is , the real-time current is , the real-time speed is , then the real-time moment load of the generator is : ； Assume the energy efficiency of the generator is , the overall transmission efficiency from the generator shaft to the generator is , the generator torque is generated by wind input, so the real-time load at the hub calculated from the generator torque load is: ； In the formula, is the real-time load at the thrust-reverse hub.

6. A method for estimating and controlling the load of a wind turbine generator system according to any one of claims 4 and 5, characterized in that: The specific steps of S5 are: S5.

1. Encode and create a population; suppose the range of the population is [a, b], and the number of individuals in the population is c , the parameter is represented by a binary code of length k, the binary code is randomly created, and the code of a population individual is , set the random number a=rand, if a<0.5, then =0, if a>0.5, then =1; and so on; S5.

2. Decoding: The binary code of a known individual is , then the corresponding decoding formula is: ； The decoded values ​​correspond to weights and thresholds; S5.3, Detection and evaluation of individual fitness: The test set data is brought into the BP neural network with weights and thresholds determined by the initial population for training. The objective function of the training is: ； ； in, and are the linear coefficients for inverting the actual blade load and hub load from the load calculated by the motor, and They are the calculation constants for inverting the actual blade load and hub load from the calculated load of the motor; The inverse of the mean square error (MSE) between the output result and the actual value corresponding to each sample is taken as the fitness. The fitness formula is as follows: ； In the formula is the actual value corresponding to each sample, is the output value obtained by the neural network; the smaller the mean square error value, the higher the fitness. Considering the selection probability in the selection step, taking the inverse of the mean square error as the fitness, the selection probability of excellent individuals becomes higher; S5.

4. Select individuals from the initial population and choose more excellent individuals based on fitness; S5.

5. Save the individuals; sort the selected individuals according to fitness, and select the individuals within the set value range for saving in the loop to prevent negative changes in the individuals during crossover and mutation; S5.

6. Crossover; the crossover operation uses the single-point crossover operator, that is, there is only one crossover point position. Randomly select two individuals from the population and randomly generate a crossover point position to exchange the gene codes; S5.

7. Mutation; the mutation operation uses uniform mutation, that is, to make the individuals in the population flip the gene with a small probability set value, changing from 0 to 1 and from 1 to 0; S5.

8. Repeat the steps from S5.1 to S5.7 until the optimal weights and thresholds are obtained.

7. A method for estimating and controlling the load of a wind turbine generator system according to claim 6, characterized in that: The specific steps of S5.4 are as follows: S5.4.

1. Calculate the total fitness of the group : ； S5.4.

2. Calculate the selection probability of each individual : ； S5.4.

3. Calculation of cumulative probability : ； S5.4.4, use rand to generate numbers between [0,1], Select individuals from S5.4.

5. Create new individuals to complement the deleted individuals.

8. A method for estimating and controlling the load of a wind turbine generator system according to claim 7, characterized in that: The specific steps of S5.6 are as follows: S5.6.

1. Set the crossover probability pcross; S5.6.

2. Randomly select two individuals from the population, and the formula is as follows: ； in Indicates the position of an individual in a population, ceil is a rounding function, rand is a random number between 0 and 1, and c is the number of populations; S5.6.

3. Randomly generate the crossover point position, and the formula is as follows: ； in , It represents the position of the crossover, and k represents the length of the individual; S5.6.

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 directly enter the recombination step.

Images