In-service wind turbine blade flap deformation online monitoring, early warning and flap deformation amount prediction method

Abstract

This invention belongs to the field of wind turbine blade deformation monitoring technology, and relates to a method for online monitoring, early warning, and prediction of flapping deformation of in-service wind turbine blades. The steps include: synchronously and continuously acquiring static data of the blades via a satellite positioning device; obtaining the static representative points and static reference plane equations of each blade tip; obtaining the dynamic coordinate training sequence of each blade; constructing and calibrating an independent adaptive Kalman filter model for each blade; activating the adaptive Kalman filter model corresponding to each blade to obtain the filtered dynamic coordinate sequence; calculating the flapping deformation of the blade; judging the blade state and executing corresponding control strategies based on preset multi-level early warning thresholds; predicting the blade flapping deformation over a future period and issuing multi-level early warning prompts based on the prediction results. This invention can acquire the flapping deformation of the blades in real time during operation and provide real-time feedback to the wind turbine, allowing for advance adjustment of the wind turbine blade control strategy.

Description

Technical Field

[0001] This invention belongs to the field of wind turbine blade deformation monitoring technology, specifically involving online monitoring, early warning and prediction of flapping deformation of in-service wind turbine blades. Background Technology

[0002] Wind power generation, due to its abundant resources, environmental friendliness, and mature technology, has been widely applied and rapidly developed, becoming the third largest power source after thermal power and hydropower. Wind turbine generators are the core equipment for converting wind energy, with the blades being the key components that capture wind energy and convert it into mechanical energy. Their structural safety and operational status directly affect the overall performance, power generation efficiency, and service life of the turbine.

[0003] As wind power technology develops towards larger scale and higher power, the blade length continues to increase and the structure becomes increasingly complex. The aerodynamic loads, gravity loads, inertial loads and environmental corrosion that the blades bear during operation are becoming increasingly severe, which can easily lead to structural fatigue, deformation or even fracture, seriously affecting the safe and economical operation of the unit.

[0004] Blade flapping deformation, or bending vibration perpendicular to the plane of rotation, is a common dynamic response of blades under complex wind conditions. Excessive flapping deformation can not only lead to decreased aerodynamic performance and reduced power generation efficiency, but may also cause serious consequences such as blade-to-tower collisions, structural fatigue accumulation, and damage to critical components, and even induce catastrophic accidents.

[0005] Currently, the wind power industry mainly relies on periodic inspections, offline testing, and laboratory simulation tests to monitor the condition of in-service blades, lacking real-time, online, and continuous monitoring methods for blade flapping deformation. While traditional methods such as strain gauges and accelerometers can be used to measure local deformation, they are difficult to comprehensively reflect the overall dynamic deformation state of the blade, and are susceptible to environmental interference, complex to install, and have poor durability.

[0006] Furthermore, existing methods primarily focus on post-fault diagnosis and health assessment, failing to provide real-time feedback and proactive control during operation. This hinders timely early warning and adjustment measures in the initial stages of deformation, thus failing to effectively reduce blade operating loads, delay fatigue damage, and extend turbine lifespan. Therefore, researching a technology capable of real-time monitoring of blade flapping deformation and providing early warning and trend prediction, achieving closed-loop linkage with the turbine control system, is of great significance for improving the operational safety, reliability, and economy of wind turbines. Summary of the Invention

[0007] In view of the shortcomings of the prior art, the purpose of this invention is to provide an online monitoring, early warning and prediction method for flapping deformation of in-service wind turbine blades. It can acquire the flapping deformation of wind turbine blades in real time during operation, predict the subsequent deformation trend by combining future wind field prediction data, and provide real-time feedback to the wind turbine to adjust the control strategies such as steering and pitch angle of the wind turbine blades in advance. At the same time, it provides data support for the preventive maintenance of the blades and timely carry out targeted maintenance.

[0008] To achieve the above objectives, this invention provides a method for online monitoring, early warning, and prediction of flapping deformation of in-service wind turbine blades, comprising the following steps:

[0009] S1. Fix satellite positioning devices to the tips of each blade of the wind turbine generator set. While the blades remain stationary, collect static data of the blades synchronously and continuously. Use the main control system of the wind turbine generator set to monitor the data solution type in real time and retain only the data with the solution type of fixed solution as valid static data.

[0010] S2. Using effective static data, obtain the static representative points at the tips of each blade and the reference plane equation when the blade is at rest.

[0011] S3. Under typical operating conditions of different wind speeds and corresponding unit speeds, synchronously and continuously collect dynamic deformation data of the blades and perform preprocessing to form a dynamic coordinate training sequence for each blade.

[0012] S4. Construct and calibrate an independent adaptive Kalman filter model for each blade, train the sequence using the dynamic coordinates corresponding to each blade, and use the optimization algorithm to calibrate the initial value of the noise covariance matrix corresponding to different operating conditions for each blade, and establish a mapping relationship library between unit speed and initial value of noise covariance matrix.

[0013] S5. Based on the current unit speed, retrieve the corresponding initial value of the optimal noise covariance matrix from the mapping relationship library as the initial parameter, start the adaptive Kalman filter model corresponding to each blade, and obtain the filtered dynamic coordinate sequence.

[0014] S6. Substitute the filtered dynamic coordinate sequence into the reference plane equation, calculate the vertical distance of the dynamic deformation data points from the reference plane, and obtain the flapping deformation of the blade.

[0015] S7. Transmit the flapping deformation amount to the main control system in real time, and judge the blade status and execute the corresponding control strategy based on the preset multi-level early warning thresholds.

[0016] S8. Obtain real-time wind speed data and future wind speed forecast data, construct a piecewise multinomial regression model, predict the blade flapping deformation over a future period of time, and issue multi-level early warning prompts based on the prediction results.

[0017] As a preferred embodiment of the present invention, in S1, the satellite positioning device adopts an RTK positioning module or a PPK positioning module, and the data acquisition frequency of the satellite positioning device is set to 20 times per second. When the wind turbine generator is stopped and the blades remain stationary, the main control system controls each satellite positioning device to synchronously and continuously acquire static data of the blades for 10 minutes. During the acquisition process, the main control system monitors the data solution type in real time, retains only the data with the solution type of fixed solution, and discards the data with the solution type of floating point solution or single point solution.

[0018] As a preferred embodiment of the present invention, in S2, the least squares sphere fitting algorithm is used to process the collected static data. By fitting a spatial sphere model, the spatial distribution characteristics of the leaf tip in a static state are characterized, and the center of the sphere is extracted as the static representative point of the leaf tip. Based on the static representative point, the reference plane equation when the leaf is at rest is determined, and the plane constant is determined.

[0019] As a preferred embodiment of the present invention, in S3, the data acquisition time for each typical working condition is not less than 30 minutes, using... The criteria are used to filter the dynamic coordinate sequences in the dynamic deformation data to obtain the dynamic coordinate training sequences for each blade tip used for model training:

[0020] ;

[0021] In the formula, This represents the dynamic coordinate training sequence of the tip of the k-th leaf at time t; , , These represent the X-axis, Y-axis, and Z-axis coordinates of the tip of the k-th blade at time t, respectively.

[0022] As a preferred embodiment of the present invention, in S4, based on various typical working conditions An independent adaptive Kalman filter model is constructed and calibrated for each blade.

[0023] Define the state vector of the k-th blade at time t. Represented as:

[0024] ;

[0025] In the formula, , , These represent the X-axis, Y-axis, and Z-axis coordinates of the tip of the k-th blade at time t in the global coordinate system, which are the actual position states to be estimated. , , Let X and Y represent the velocities of the tip of the k-th blade at time t in the X, Y, and Z coordinate directions, respectively; the superscript T indicates transpose.

[0026] By employing maximum likelihood estimation or expectation-maximization algorithms, the initial values ​​of the optimal noise covariance matrix corresponding to different operating conditions are determined for the k-th blade, including the optimal process noise covariance matrix. and the optimal observation noise covariance matrix Establish a mapping relationship library for the initial values ​​of the unit speed-noise covariance matrix.

[0027] As a preferred embodiment of the present invention, in S5, after the blades are put into real-time monitoring operation, the main control system synchronously receives the real-time coordinate sequences uploaded by each satellite positioning device. ,in, This represents the actual coordinate sequence of the tip of the k-th blade at time t. , , These represent the actual X-axis, Y-axis, and Z-axis coordinates of the tip of the k-th blade at time t in the global coordinate system;

[0028] Based on the current unit speed, retrieve the corresponding [database name] from the mapping database. , As initial parameters, the adaptive Kalman filter model corresponding to each blade is activated. Using observation input as the input, an iterative process of state prediction and observation update is employed, and the noise covariance parameter is dynamically adjusted based on the real-time information sequence to achieve adaptive fusion of the model and observations. Finally, the optimal estimate of the state vector for each blade is output, and the position component is extracted from the optimal estimate to obtain the filtered dynamic coordinate sequence. .

[0029] As a preferred embodiment of the present invention, in S6, the flapping deformation of the k-th blade at time t... The calculation formula is:

[0030] ;

[0031] In the formula, , , These are the components of the plane normal vector along the X, Y, and Z axes of the global coordinate system, respectively; D is the plane constant.

[0032] As a preferred embodiment of the present invention, in S7, based on the elastic deformation limit of the blade material and historical operating data, three levels of early warning thresholds are preset, namely:

[0033] ;

[0034] In the formula, , , These represent the first-level, second-level, and third-level thresholds for the k-th leaf, respectively. This represents the average peak value of the flapping deformation of the k-th blade under normal operating conditions. Let be the standard deviation of the flapping deformation of the k-th blade;

[0035] The specific steps for judging the blade status and executing the corresponding control strategy are as follows:

[0036] when When this occurs, a Level 1 warning is triggered: the main control system outputs a prompt message to the monitoring center of the wind turbine generator, reminding maintenance personnel to pay attention to the blade status;

[0037] when At this time, a level two warning is triggered: the main control system adjusts the pitch angle to reduce the aerodynamic load on the blades;

[0038] when When this occurs, a Level 3 warning is triggered: the emergency shutdown procedure is immediately initiated, the blades are feathered to 90 degrees, and the mechanical brakes are activated.

[0039] As a preferred embodiment of the present invention, in step S8, a wind speed meter is used to acquire real-time wind speed data. And obtain wind speed forecast data for the next 24 hours provided by the wind field. ,right , Max-min standardization was performed separately to obtain standardized wind speed data. Wind speed forecast data ;

[0040] Based on wind speed, the wind speed is divided into four ranges: low wind speed, medium wind speed, high wind speed, and extreme wind speed.

[0041] For each wind speed range, utilize Construct a cubic polynomial regression prediction model for the k-th leaf:

[0042] ;

[0043] In the formula, This represents the predicted sequence of flapping deformation of the k-th blade; , , , The model coefficients for the m-th wind speed interval are obtained by fitting the historical wind speed-flailing deformation matching dataset using the weighted least squares method.

[0044] Will By inputting the cubic polynomial regression prediction model of the k-th blade, we obtain the predicted sequence of future flapping deformation of the k-th blade. ,Will By inverse normalization, a prediction sequence for actual swing deformation is obtained. ,based on Based on preset multi-level warning thresholds, multi-level warning prompts are issued.

[0045] As a preferred embodiment of the present invention, in S8, for any specific future time t, based on and , , Multiple levels of early warning alerts were issued, specifically:

[0046] when When this occurs, a Level 1 early warning is triggered: the main control system outputs an early warning message to the monitoring center, informing the operation and maintenance personnel that "the amount of deformation during waving is likely to increase in the future," and reminding them to pay attention to the operating status in advance;

[0047] when At that time, a level-two early warning is triggered: the main control system combines... Actively adjust the pitch angle to intervene in advance and reduce the expected aerodynamic load;

[0048] when At that time, a level three early warning is triggered: the main control system is activated. Initiate the emergency shutdown preparation procedure and immediately notify maintenance personnel to be on site to complete emergency response preparations.

[0049] The beneficial effects of this invention are:

[0050] This invention relies on a full-process data processing scheme that combines high-frequency satellite positioning acquisition at the blade tip, high-precision data filtering using fixed solutions, and dynamic noise reduction using adaptive Kalman filtering. This enables high-precision, real-time online monitoring of wind turbine blade flapping deformation under all operating conditions. It can accurately capture the overall dynamic deformation state of the blade, overcoming the limitations of traditional sensors that rely on localized measurements and are susceptible to environmental interference. Simultaneously, based on the elastic deformation limit of the blade material and historical operating data, a three-level early warning threshold is set. Combined with a graded response strategy linked to the wind turbine main control system, from maintenance prompts and automatic pitch adjustment to 10-second emergency shutdown, real-time feedback and active control of flapping deformation are achieved. This effectively avoids risks such as blade-to-tower collisions and structural fatigue accumulation, significantly improving the safety of blade operation and the timeliness of fault prevention.

[0051] This invention integrates real-time wind speed data from the wind farm with 24-hour wind speed forecasts. It constructs a cubic polynomial regression prediction model based on wind speed intervals, fits the model coefficients using weighted least squares, and updates them monthly to match actual operating conditions. This allows for accurate prediction of future blade flapping deformation trends. Furthermore, it compares predicted values ​​with early warning thresholds to achieve proactive, tiered control, providing data support for the main control system to adjust control strategies such as pitch, yaw, and capacity reduction in advance. It not only achieves closed-loop linkage with the turbine control system, suppressing excessive blade flapping deformation at the source, delaying fatigue damage, and extending the service life of blades and the turbine, but also provides accurate data references for preventative blade maintenance. This drives the transformation of wind power operation and maintenance from post-fault diagnosis to proactive prevention and forward-looking decision-making, significantly improving the economy and intelligence of wind turbine operation. Attached Figure Description

[0052] Figure 1 This is a flowchart illustrating the principle of this invention;

[0053] Figure 2 This is a schematic diagram of the installation of a satellite positioning device on a wind turbine blade. Detailed Implementation

[0054] The embodiments of the present invention will be further described below with reference to the accompanying drawings:

[0055] Example 1: As Figure 1 As shown, the method for online monitoring, early warning, and prediction of flapping deformation of in-service wind turbine blades includes the following steps:

[0056] S1. Fix satellite positioning devices to the tips of each blade of the wind turbine generator set. While the blades remain stationary, collect static data of the blades synchronously and continuously. Use the main control system of the wind turbine generator set to monitor the data solution type in real time and retain only the data with the solution type of fixed solution as valid static data.

[0057] S2. Using effective static data, obtain the static representative points at the tips of each blade and the reference plane equation when the blade is at rest.

[0058] S3. Under typical operating conditions of different wind speeds and corresponding unit speeds, synchronously and continuously collect dynamic deformation data of the blades and perform preprocessing to form a dynamic coordinate training sequence for each blade.

[0059] S4. Construct and calibrate an independent adaptive Kalman filter model for each blade, train the sequence using the dynamic coordinates corresponding to each blade, and use the optimization algorithm to calibrate the initial value of the noise covariance matrix corresponding to different operating conditions for each blade, and establish a mapping relationship library between unit speed and initial value of noise covariance matrix.

[0060] S5. Based on the current unit speed, retrieve the corresponding initial value of the optimal noise covariance matrix from the mapping relationship library as the initial parameter, start the adaptive Kalman filter model corresponding to each blade, and obtain the filtered dynamic coordinate sequence.

[0061] S6. Substitute the filtered dynamic coordinate sequence into the reference plane equation, calculate the vertical distance of the dynamic deformation data points from the reference plane, and obtain the flapping deformation of the blade (referred to as flapping amount).

[0062] S7. Transmit the flapping deformation amount to the main control system in real time, and judge the blade status and execute the corresponding control strategy based on the preset multi-level early warning thresholds.

[0063] S8. Obtain real-time wind speed data and future wind speed forecast data, construct a piecewise multinomial regression model, predict the blade flapping deformation over a future period of time, and issue multi-level early warning prompts based on the prediction results.

[0064] In S1, the satellite positioning device uses an RTK positioning module or a PPK positioning module. The data acquisition frequency of the satellite positioning device is set to 20 times per second (to ensure that the satellite positioning device can measure the flapping deformation of the blade in real time during the blade rotation process). When the wind turbine generator is stopped and the blade remains stationary, the main control system controls each satellite positioning device to synchronously and continuously acquire static data of the blade for 10 minutes. During the acquisition process, the main control system monitors the data solution type in real time, retains only high-precision data with fixed solution type, and discards low-precision data with floating-point solution type and single-point solution type to ensure the accuracy of the reference plane fitting.

[0065] For a wind turbine generator with three blades, the installation diagram of the satellite positioning device is as follows: Figure 2 As shown, satellite positioning devices 1 to 3 are installed at the tips of the three blades respectively.

[0066] In S2, the least squares sphere fitting algorithm is used to process the collected static data. A spatial sphere model is fitted to characterize the spatial distribution characteristics of the leaf tip in a static state, and the center of the sphere is extracted as the static representative point of the leaf tip. The specific process is as follows:

[0067] For a wind turbine generator with three blades, let the effective static coordinate point set of the tip of the k-th blade be:

[0068] ;

[0069] In the formula, , , Let x, y, and z be the X, Y, and Z coordinates of the e-th valid data point in the k-th leaf, respectively, in the global coordinate system, where k = 1, 2, and 3; and e be the index of the valid data point in the k-th leaf. The number of valid data points for the k-th leaf;

[0070] The fitted sphere model is as follows:

[0071] ;

[0072] In the formula, These are universal spatial coordinate variables in the global coordinate system. R corresponds to the X-axis variable, S corresponds to the Y-axis variable, and T1 corresponds to the Z-axis variable. They are used to describe the spatial position of any point on the sphere. Let be the coordinates of the center of the fitted sphere corresponding to the k-th leaf, which is the static representative point of the leaf tip. , , These are the coordinates of the center of the fitted sphere corresponding to the k-th blade in the global coordinate system along the X, Y, and Z axes, respectively, which represent the high-precision static representative point of the blade tip. The radius of the fitted sphere corresponding to the kth blade reflects the spatial dispersion of the static measurement point cloud of that blade.

[0073] The optimal sphere center coordinates are obtained by minimizing the sum of the squares of the differences between the geometric distance from each valid data point to the sphere center and the fitted radius. and radius The objective function is:

[0074] ;

[0075] Solve the objective function to obtain the optimal sphere center coordinates. These are recorded as the static representative points of the three leaf tips:

[0076] ;

[0077] ;

[0078] ;

[0079] In the formula, , , These are the static representative points at the tips of the first, second, and third leaflets, respectively. The optimal center coordinates for the first leaf are given, and the same applies to the others;

[0080] by Using the base point, construct two spatial vectors. , ,calculate and The vector product is used to verify the noncollinearity of the three points, and the result of the vector product is used as the plane normal vector. ,in , , These are the components of the plane normal vector along the X, Y, and Z axes of the global coordinate system, respectively.

[0081] Determine the equation of the reference plane when the blade is at rest:

[0082] ;

[0083] Determine the plane constant D:

[0084] .

[0085] In S3, the data acquisition time for each typical operating condition is no less than 30 minutes (to ensure that the acquired data can fully characterize the flapping deformation features of the blades under that operating condition). The criteria are used to filter the dynamic coordinate sequences in the dynamic deformation data to obtain the dynamic coordinate training sequences for each blade tip used for model training:

[0086] ;

[0087] In the formula, This represents the dynamic coordinate training sequence of the tip of the k-th leaf at time t; , , These represent the X-axis, Y-axis, and Z-axis coordinates of the tip of the k-th blade at time t, respectively.

[0088] In S4, based on various typical working conditions An independent adaptive Kalman filter model is constructed and calibrated for each blade. The core of the adaptive Kalman filter model is to determine the optimal initial values ​​of its process noise covariance matrix Q and observation noise covariance matrix R under different operating conditions.

[0089] Define the state vector of the k-th blade at time t. Represented as:

[0090] ;

[0091] In the formula, , , These represent the X-axis, Y-axis, and Z-axis coordinates of the tip of the k-th blade at time t in the global coordinate system, which are the actual position states to be estimated. , , These represent the velocities of the tip of the k-th blade at time t in the X, Y, and Z coordinate directions, respectively. By establishing the state transition relationship between these velocities and the position, the model's ability to predict motion trends is enhanced; the superscript T indicates transpose.

[0092] By employing maximum likelihood estimation or expectation-maximization algorithms, the initial values ​​of the optimal noise covariance matrix corresponding to different operating conditions are determined for the k-th blade, including the optimal process noise covariance matrix. and the optimal observation noise covariance matrix Establish a mapping relationship library for the initial values ​​of the unit speed-noise covariance matrix.

[0093] In S5, after the blades are put into real-time monitoring operation, the main control system synchronously receives the real-time coordinate sequences uploaded by each satellite positioning device. ,in, This represents the actual coordinate sequence of the tip of the k-th blade at time t. , , These represent the actual X-axis, Y-axis, and Z-axis coordinates of the tip of the k-th blade at time t in the global coordinate system;

[0094] Based on the current unit speed, retrieve the corresponding [database name] from the mapping database. , As initial parameters, the adaptive Kalman filter model corresponding to each blade is activated. Using observational input as the basis, an iterative process of state prediction and observation updates is employed, with the noise covariance parameter dynamically adjusted based on the real-time information sequence to achieve adaptive fusion of the model and observations. Ultimately, the optimal estimate of the state vector for each blade is output. The position components are extracted from the optimal estimate to obtain a filtered, high-precision, all-condition-adaptive dynamic coordinate sequence. .

[0095] The core process of the adaptive Kalman filter model (filter) includes two iterative steps: state prediction and observation update. For the k-th leaf, at each sampling time t, the adaptive Kalman filter model... As observation vector The input, the state prediction step is based on the optimal state estimate of the previous time t-1. And the state transition matrix F, which is used to make a priori predictions of the state at the current time t, to obtain the predicted state value. and its corresponding prediction error covariance matrix :

[0096] ;

[0097] ;

[0098] In the formula, Let be the error covariance matrix of the state estimation at the previous time step; The process noise covariance matrix at the current moment represents the uncertainty of the model; the state transition matrix F can be set according to the blade motion model (e.g., at the sampling interval). If we assume the velocity is constant, then F includes (This term is used to predict the position at the next moment from the position and velocity).

[0099] Then, the observation update step begins. First, based on... With the current observation noise covariance matrix Calculate the Kalman gain matrix :

[0100] ;

[0101] In the formula, H is the observation matrix, which is used to map the state vector to the observation space.

[0102] This determines the trust weights between model predictions and real-time observations in later updates. Then, using... and The prior predicted state is corrected to obtain the optimal posterior estimate of the state at the current time t. and its updated error covariance matrix :

[0103] ;

[0104] ;

[0105] In the formula, Let I be the prior prediction of the current state based on information from the previous time step; I is the identity matrix.

[0106] The observation residuals (i.e., the innovation sequence) reflect the deviation between actual observations and model predictions. Used to balance the weights of prediction and observation in state correction; The correction factor represents the degree to which prediction uncertainty is reduced after fusing observational information;

[0107] The above prediction and update process is executed cyclically within each sampling period (0.05S), thereby achieving continuous and real-time optimal estimation of the blade tip motion state.

[0108] To achieve optimal filtering under all operating conditions, the adaptive mechanism is manifested as follows:

[0109] In each iteration, the observation residuals (new information sequence) are calculated. The residual reflects the deviation between the current observed values ​​and the model predictions, and is a key indicator for evaluating the accuracy of model predictions and the level of observational noise. Based on the statistical properties of this residual (such as its covariance)... Real-time fine-tuning process and The specific value, when A significant increase indicates a relative increase in observation noise or an increase in model prediction bias. In this case, the algorithm will reduce its focus on the current observations. The weighting of the model predictions leads to greater trust in its predictions. Conversely, when the residual covariance decreases, the observation weights are increased to quickly track changes in the actual state. Through this dynamic adjustment, the filter can automatically balance prediction and observation under different wind speeds and rotational speeds, always maintaining optimal estimation performance.

[0110] The iterative update frequency of the adaptive Kalman filter is kept consistent with the data acquisition frequency of the satellite positioning device, both being 20Hz, to ensure the real-time performance of the filtering process.

[0111] In S6, the flapping deformation of the k-th blade at time t. The calculation formula is:

[0112] ;

[0113] In the formula, , , These are the components of the plane normal vector along the X, Y, and Z axes of the global coordinate system, respectively; D is the plane constant.

[0114] In S7, based on the elastic deformation limit of the blade material and historical operating data, three levels of early warning thresholds are preset, namely:

[0115] ;

[0116] In the formula, , , These represent the first-level, second-level, and third-level thresholds for the k-th leaf, respectively. The mean peak value of the flapping deformation of the kth blade under normal operating conditions (e.g., when the wind speed is in the low or medium wind speed range, the unit is operating without faults, and the satellite positioning data solution type is fixed). Let be the standard deviation of the flapping deformation of the k-th blade;

[0117] The specific steps for judging the blade status and executing the corresponding control strategy are as follows:

[0118] when When this occurs, a Level 1 warning is triggered: the main control system outputs a prompt message to the monitoring center of the wind turbine generator, reminding maintenance personnel to pay attention to the blade status;

[0119] when At this time, a level two warning is triggered: the main control system adjusts the pitch angle to reduce the aerodynamic load on the blades;

[0120] when When this occurs, a Level 3 warning is triggered: the emergency shutdown procedure is immediately initiated, the blades are feathered to 90 degrees and the mechanical brakes are activated to ensure that the unit is completely shut down within 10 seconds.

[0121] In S8, a wind speed meter is used to obtain real-time wind speed data. And obtain wind speed forecast data for the next 24 hours provided by the wind field. ,right , Max-min standardization was performed separately to obtain standardized wind speed data. Wind speed forecast data ;

[0122] flapping deformation of the k-th blade Max-min normalization is performed to obtain the normalized swing deformation. Used to fit model coefficients and Inverse standardization;

[0123] Based on wind speed, four wind speed ranges are defined: low wind speed range (greater than 0 m / s and less than 4 m / s), medium wind speed range (greater than or equal to 4 m / s and less than or equal to 10 m / s), high wind speed range (greater than 10 m / s and less than or equal to 25 m / s), and extreme wind speed range (greater than 25 m / s).

[0124] For each wind speed range, utilize Construct a cubic polynomial regression prediction model for the k-th leaf:

[0125] ;

[0126] In the formula, This represents the predicted sequence of flapping deformation of the k-th blade; , , , The model coefficients for the m-th wind speed interval are obtained by fitting the historical wind speed-flailing deformation matching dataset using the weighted least squares method.

[0127] Will By inputting the cubic polynomial regression prediction model of the k-th blade, we obtain the predicted sequence of future flapping deformation of the k-th blade. ,Will By inverse normalization, a prediction sequence for actual swing deformation is obtained. ,based on Based on preset multi-level warning thresholds, multi-level warning prompts are issued.

[0128] Using max-min standardized historical wind speed data and the corresponding historical waving deformation, a historical wind speed-waving deformation matching dataset is constructed for training a cubic polynomial regression prediction model and fitting the model coefficients.

[0129] The historical wind speed-flaring deformation matching dataset was obtained by fitting it using weighted least squares method. , , , At that time, wind speed stability weights are introduced. To enhance the fitting accuracy in regions with smaller wind speed fluctuations:

[0130] ;

[0131] In the formula, This is the historical average wind speed. Let be the standard deviation of wind speed 10 minutes before and after time t;

[0132] The fitting objective function is:

[0133] ;

[0134] The coefficient vector can be obtained by solving the matrix:

[0135] ;

[0136] In the formula, V is the design matrix, and each row represents a design matrix. ; Let be the vector of coefficients to be determined; This is a diagonal weight matrix, with diagonal elements as follows: ;

[0137] By inverse normalization, a prediction sequence for actual swing deformation is obtained. Specifically:

[0138] ;

[0139] In the formula, This represents the average historical flapping deformation of the k-th blade.

[0140] For any specific future time t (in the actual swing deformation prediction sequence, a specific future time t corresponds to an actual swing deformation prediction value), based on and , , Multiple levels of early warning alerts were issued, specifically:

[0141] when When this occurs, a Level 1 early warning is triggered: the main control system outputs an early warning message to the monitoring center, informing the operation and maintenance personnel that "the amount of deformation during waving is likely to increase in the future," and reminding them to pay attention to the operating status in advance;

[0142] when At that time, a level-two early warning is triggered: the main control system combines... Actively adjust the pitch angle to intervene in advance and reduce the expected aerodynamic load; for example, if No pitch adjustment is needed in low wind speed ranges; moderate pitch adjustment is needed in medium wind speed ranges (e.g., adjusting to 45 degrees to reduce the angle of attack); significant pitch adjustment is needed in high wind speed ranges (e.g., moving closer to 89 degrees to reduce the tip velocity); and pitch adjustment should be initiated directly in extreme wind speed ranges for emergency adjustment.

[0143] when At that time, a level three early warning is triggered: the main control system is activated. Initiate the emergency shutdown preparation procedure and immediately notify maintenance personnel to be on site to complete emergency response preparations; for example, if In low wind speed ranges, the system will not be activated immediately, only monitoring will be strengthened. In medium wind speed ranges, if the wind speed is predicted to continue to rise, the emergency shutdown preparation procedure will be activated. In high wind speed ranges, the emergency shutdown preparation procedure will be activated immediately. In extreme wind speed ranges, the emergency shutdown preparation procedure will be activated directly, and the blades will be feathered to 90° and the mechanical brakes will be activated in preparation. At the same time, maintenance personnel will be notified to be on site to complete the emergency response preparation.

[0144] To ensure the accuracy and environmental adaptability of the predictions, the prediction model is refitted every month using the latest collected operating data, and the model coefficients are updated to ensure that the predicted swing deformation always matches the actual operating conditions of the unit.

[0145] The algorithm calculations involved in the method of this embodiment can be implemented by the main control system executing the corresponding program.

[0146] Example 2: Based on Example 1, when constructing the cubic polynomial regression prediction model for the k-th blade, wind speed change rate, ambient temperature, and ambient humidity are further considered, specifically:

[0147] We acquire predicted data on wind speed change rate, ambient temperature, and ambient humidity for the next 24 hours. We then perform max-min standardization on the predicted data for wind speed change rate, temperature, and humidity to obtain standardized wind speed change rate data. Temperature data Humidity data ;

[0148] Set the critical threshold for wind speed change rate. Construct a dynamic adaptive wind speed interval boundary based on the wind speed change rate: when At the same time, the span of the corresponding wind speed range will be reduced by 50%, so as to achieve fine-grained range division under the condition of sudden wind speed change.

[0149] To construct a multi-factor coupled correction cubic polynomial regression prediction model, wind speed change rate correction term, temperature correction term, and humidity correction term are added to the cubic polynomial regression prediction model for the m-th wind speed interval.

[0150] ;

[0151] In the formula, , , These are the weighting coefficients for the wind speed change rate correction term, temperature correction term, and humidity correction term, respectively.

[0152] The extended matching dataset is a four-dimensional matching dataset of wind speed-wind speed change rate-temperature and humidity-flapping deformation (this dataset can also be used for model training). The model coefficients of the corrected model are obtained by fitting the four-dimensional matching dataset using the weighted least squares method.

[0153] The rate of change of wind speed can be calculated differentially from continuous wind speed data obtained by a wind farm velocimeter, while ambient temperature and humidity can be synchronously collected / predicted using existing temperature and humidity sensors in the wind farm. The introduction of the rate of change of wind speed solves the problem of low prediction accuracy caused by fixed interval division under sudden wind speed changes. The refined dynamic interval can accurately capture abrupt changes in flapping deformation caused by gusts and wind shear. The temperature and humidity correction term compensates for the model deviation caused by changes in blade material properties (elastic modulus, structural damping) with ambient temperature and humidity. Changes in blade temperature will change the elastic deformation limit of the material, and humidity will affect the blade structural damping. Multi-factor correction makes the prediction model more in line with actual physical laws.

[0154] Example 3: This example illustrates one configuration method for a satellite positioning device, specifically as follows:

[0155] The satellite positioning device is installed on the blade tip via a magnetic-mechanical composite adaptive fixing device, which includes:

[0156] Permanent magnet adsorption base, shape memory alloy buffer layer, piezoelectric ceramic preload monitoring module and electromagnetic tripping protection mechanism;

[0157] The permanent magnet adsorption base uses a neodymium iron boron permanent magnet array, which is distributed in a fan shape on a flexible bonding surface that is adapted to the curved surface of the blade tip. The permanent magnet surface is covered with an anti-corrosion coating. The bottom of the base is provided with a contoured groove that matches the curvature of the blade tip surface, and a shape memory alloy buffer layer is embedded in the groove.

[0158] The shape memory alloy buffer layer is a mesh structure woven from NiTi-based shape memory alloy wires. Its phase transformation temperature range is set from -20℃ to 60℃. It exhibits a superelastic state at room temperature and can adaptively compensate for the slight radial deformation of the blade tip caused by centrifugal force during rotation, maintaining a constant contact pressure between the satellite positioning device and the blade tip surface.

[0159] The piezoelectric ceramic preload monitoring module is embedded between the permanent magnet adsorption base and the shape memory alloy buffer layer. It monitors the preload changes of the magnetic fixing device in real time. When the preload is detected to be lower than the safety threshold, it sends an early warning signal to the main control system.

[0160] The electromagnetic tripping protection mechanism includes an electromagnetic coil and a mechanical locking pin. The electromagnetic coil is linked to the safety chain system of the wind turbine generator set. When the main control system triggers a level 3 warning or receives an emergency stop command, the electromagnetic coil is energized to generate a reverse magnetic field, which counteracts the attraction force of the permanent magnet adsorption base. At the same time, the mechanical locking pin pops out, forcibly separating the satellite positioning device from the blade tip, preventing the satellite positioning device from being damaged by inertial impact during emergency braking of the unit.

Claims

1. A method for online monitoring, early warning and predicting the amount of waving deformation of an in-service wind turbine blade, characterized in that, Includes the following steps: S1. Fix satellite positioning devices to the tips of each blade of the wind turbine generator set. While the blades remain stationary, collect static data of the blades synchronously and continuously. Use the main control system of the wind turbine generator set to monitor the data solution type in real time and retain only the data with the solution type of fixed solution as valid static data. S2. Using effective static data, obtain the static representative points of each blade tip and the reference plane equation when the blade is at rest. Specifically, the collected static data is processed by using the least squares sphere fitting algorithm. A spatial sphere model is fitted to characterize the spatial distribution characteristics of the blade tip in a static state. The center of the sphere is extracted as the static representative point of the blade tip. Based on the static representative point, the reference plane equation when the blade is at rest is determined, and the plane constant is determined. S3. Under typical operating conditions with different wind speeds and corresponding unit speeds, synchronously and continuously collect dynamic deformation data of the blades and preprocess it to form a dynamic coordinate training sequence for each blade, specifically: The collection time of each typical working condition is not less than 30 minutes, and the data is collected by using The dynamic coordinate sequence in the dynamic deformation data is filtered according to the criterion, and the dynamic coordinate training sequence of the blade tip of each blade for model training is obtained. ； In the formula, represents the dynamic coordinate training sequence of the kth blade tip at time t; 、 、 respectively represent the X-axis, Y-axis, and Z-axis coordinate values of the kth blade tip at time t. S4. Construct and calibrate an independent adaptive Kalman filter model for each blade. Use the dynamic coordinates of each blade to train the sequence, and optimize the algorithm to determine the optimal initial value of the noise covariance matrix for each blade corresponding to different operating conditions. Establish a mapping relationship library between unit speed and the initial value of the noise covariance matrix, specifically: Based on various typical working conditions An independent adaptive Kalman filter model is constructed and calibrated for each blade. Define the state vector of the k-th blade at time t. Represented as: ； In the formula, , , These represent the X-axis, Y-axis, and Z-axis coordinates of the tip of the k-th blade at time t in the global coordinate system, which are the actual position states to be estimated. , , Let X and Y represent the velocities of the tip of the k-th blade at time t in the X, Y, and Z coordinate directions, respectively; the superscript T indicates transpose. By employing maximum likelihood estimation or expectation-maximization algorithms, the initial values ​​of the optimal noise covariance matrix corresponding to different operating conditions are determined for the k-th blade, including the optimal process noise covariance matrix. and the optimal observation noise covariance matrix Establish a mapping relationship library for the initial values ​​of the unit speed-noise covariance matrix; S5. Based on the current unit speed, retrieve the corresponding initial value of the optimal noise covariance matrix from the mapping relationship library as the initial parameter, start the adaptive Kalman filter model corresponding to each blade, and obtain the filtered dynamic coordinate sequence. S6. Substitute the filtered dynamic coordinate sequence into the reference plane equation, calculate the vertical distance of the dynamic deformation data points from the reference plane, and obtain the flapping deformation of the blade. S7. Transmit the flapping deformation amount to the main control system in real time, and judge the blade status and execute the corresponding control strategy based on the preset multi-level early warning thresholds. S8. Obtain real-time wind speed data and future wind speed forecast data, construct a piecewise multinomial regression model, predict the blade flapping deformation over a future period of time, and issue multi-level early warning prompts based on the prediction results.

2. The method for online monitoring, early warning, and prediction of flapping deformation of in-service wind turbine blades according to claim 1, characterized in that, In S1, the satellite positioning device uses an RTK positioning module or a PPK positioning module. The data acquisition frequency of the satellite positioning device is set to 20 times per second. When the wind turbine generator is stopped and the blades remain stationary, the main control system controls each satellite positioning device to synchronously and continuously acquire static data of the blades for 10 minutes. During the acquisition process, the main control system monitors the data solution type in real time, retains only the data with the solution type of fixed solution, and discards the data with the solution type of floating point solution or single point solution.

3. The method for online monitoring, early warning, and prediction of flapping deformation of in-service wind turbine blades according to claim 1, characterized in that, In S5, after the blades are put into real-time monitoring operation, the main control system synchronously receives the real-time coordinate sequences uploaded by each satellite positioning device. ,in, This represents the actual coordinate sequence of the tip of the k-th blade at time t. , , These represent the actual X-axis, Y-axis, and Z-axis coordinates of the tip of the k-th blade at time t in the global coordinate system; Based on the current unit speed, retrieve the corresponding [database name] from the mapping database. , As initial parameters, the adaptive Kalman filter model corresponding to each blade is activated. Using observation input as the input, an iterative process of state prediction and observation update is employed, and the noise covariance parameter is dynamically adjusted based on the real-time information sequence to achieve adaptive fusion of the model and observations. Finally, the optimal estimate of the state vector for each blade is output, and the position component is extracted from the optimal estimate to obtain the filtered dynamic coordinate sequence. .

4. The method for online monitoring, early warning, and prediction of flapping deformation of in-service wind turbine blades according to claim 3, characterized in that, In S6, the flapping deformation of the k-th blade at time t. The calculation formula is: ； In the formula, , , These are the components of the plane normal vector along the X, Y, and Z axes of the global coordinate system, respectively; D is the plane constant.

5. The method for online monitoring, early warning, and prediction of flapping deformation of in-service wind turbine blades according to claim 1, characterized in that, In S7, based on the elastic deformation limit of the blade material and historical operating data, three levels of early warning thresholds are preset, namely: ； In the formula, , , These represent the first-level, second-level, and third-level thresholds for the k-th leaf, respectively. This represents the average peak value of the flapping deformation of the k-th blade under normal operating conditions. Let be the standard deviation of the flapping deformation of the k-th blade; The specific steps for judging the blade status and executing the corresponding control strategy are as follows: when When this occurs, a Level 1 warning is triggered: the main control system outputs a prompt message to the monitoring center of the wind turbine generator, reminding maintenance personnel to pay attention to the blade status; when At this time, a level two warning is triggered: the main control system adjusts the pitch angle to reduce the aerodynamic load on the blades; when When this occurs, a Level 3 warning is triggered: the emergency shutdown procedure is immediately initiated, the blades are feathered to 90 degrees, and the mechanical brakes are activated.

6. The method for online monitoring, early warning, and prediction of flapping deformation of in-service wind turbine blades according to claim 5, characterized in that, In step S8, a wind speed meter is used to acquire real-time wind speed data. And obtain wind speed forecast data for the next 24 hours provided by the wind field. ,right , Max-min standardization was performed separately to obtain standardized wind speed data. Wind speed forecast data ; Based on wind speed, the wind speed is divided into four ranges: low wind speed, medium wind speed, high wind speed, and extreme wind speed. For each wind speed range, utilize Construct a cubic polynomial regression prediction model for the k-th leaf: ； In the formula, This represents the predicted sequence of flapping deformation for the k-th blade. , , , The model coefficients for the m-th wind speed interval are obtained by fitting the historical wind speed-flailing deformation matching dataset using the weighted least squares method. Will By inputting the cubic polynomial regression prediction model of the k-th blade, we obtain the predicted sequence of future flapping deformation of the k-th blade. ,Will By inverse normalization, a prediction sequence for actual swing deformation is obtained. ,based on Based on preset multi-level warning thresholds, multi-level warning prompts are issued.

7. The method for online monitoring, early warning, and prediction of flapping deformation of in-service wind turbine blades according to claim 6, characterized in that, In S8, for any specific future time t, based on and , , Multiple levels of early warning alerts were issued, specifically: when When this occurs, a Level 1 early warning is triggered: the main control system outputs an early warning message to the monitoring center, informing maintenance personnel that "the amount of deformation during waving is likely to increase in the future," reminding them to pay attention to the operating status in advance; when At that time, a level-two early warning is triggered: the main control system combines... Actively adjust the pitch angle to intervene in advance and reduce the expected aerodynamic load; when At that time, a level three early warning is triggered: the main control system is activated. Initiate the emergency shutdown preparation procedure and immediately notify maintenance personnel to be on site to complete emergency response preparations.

Images