A method and system for measuring attitude parameters of a wind turbine

Abstract

The present application relates to the field of fan assembly, specifically relates to a kind of attitude parameter measurement method and system of wind driven generator, comprising: the reference attitude data of hoist in hoisting process, rope inclination data and the target attitude data of hoisted component are collected, inclination data and reference attitude data are fused using first weight, and predicted attitude is obtained, and then target attitude angle and predicted attitude are fused using second weight, and attitude parameter is obtained;Time instability is calculated by historical estimation error distribution, and the first weight or the second weight is adaptively updated according to instability size, and the attitude parameter of hoisted component is continuously measured.The present application can inhibit the accumulation of IMU integral error in long-time hoisting process, and improve the measurement accuracy of the attitude parameter of hoisted component of wind driven generator.

Description

Technical Field

[0001] This invention relates to the field of wind turbine assembly, and specifically to a method and system for measuring the attitude parameters of a wind turbine generator. Background Technology

[0002] In wind turbine installation operations, a single inertial measurement unit (IMU) is typically installed on the component being installed (such as the tower, nacelle, and blades). An inertial navigation algorithm integrates the acceleration and angular velocity data collected by the IMU to obtain the attitude angles (pitch, roll, and yaw) of the component. To suppress errors in the inertial navigation system, current technology employs a Kalman filter algorithm to correct the IMU's calculation results.

[0003] However, a single hoisting operation of a wind turbine typically lasts for several hours. The integral error caused by the gyroscope bias and accelerometer drift of the IMU will continue to accumulate over time. Kalman filtering has limited ability to suppress errors in long-term nonlinear systems, and the attitude measurement accuracy decreases significantly as the hoisting time increases. Summary of the Invention

[0004] To address the problem of significantly reduced attitude measurement accuracy during long-term attitude measurement using a single inertial measurement unit (IMU) in the hoisting process of wind turbines, this invention provides a method and system for measuring the attitude parameters of wind turbines.

[0005] The present invention provides a method and system for measuring the attitude parameters of a wind turbine generator, which adopts the following technical solution: One embodiment of the present invention provides a method for measuring the attitude parameters of a wind turbine, the method comprising the following steps: For the components being hoisted during the wind turbine hoisting process, reference attitude data of the hoisting equipment and the inclination angle data of the rope between each lifting point of the component and each hook lifting point of the hoisting equipment are collected, and target attitude data of the component being hoisted is also collected; the reference attitude data or target attitude data includes attitude angles. The tilt angle data is integrated into the reference attitude data using a first weight to obtain the predicted attitude of the hoisted component. The attitude angles from the target attitude data are integrated into the predicted attitude using a second weight to obtain the attitude parameters of the hoisted component. The difference between the attitude parameters and the attitude angles in the target attitude data is recorded as the estimation error. The instability of the distribution of all estimation errors over a historical time period is obtained. When the instability is greater than or equal to a preset instability threshold, the second weight is updated; when the instability is less than the preset instability threshold, the first weight is updated. The updated first and second weights are used to continue obtaining the attitude parameters of the hoisted component. Preferably, the specific steps of integrating the tilt angle data into the reference attitude data using the first weight to obtain the predicted attitude of the hoisted component are as follows: Construct the objective function The objective function Positively correlated with the first difference, and positively correlated with the second difference The twofold positive correlation is present. in This indicates the position of the hoisted component in the world coordinate system. This represents the rotation matrix from the coordinate system of the hoisted component to the world coordinate system; the first difference refers to the position of the i-th component's hoisting point in the world coordinate system. The position of the i-th hook lifting point in the world coordinate system The difference between the distance and the length of the i-th rope; Depend on , Sure; Determined by the reference attitude data of the spreader; the second difference refers to: unit vector and Differences; This represents the direction vector of the i-th rope in the world coordinate system, which is determined by the inclination angle data of the i-th rope; 1 indicates the first weight. Indicates the modulus length; Minimum This is denoted as the predicted rotation matrix; the attitude angles in this predicted rotation matrix are denoted as the predicted attitude. Preferably, the attitude parameters of the hoisted component... ,in This represents the attitude angle in the target attitude data. Indicates the predicted attitude. This indicates the second weight.

[0006] Preferably, the specific steps for obtaining the instability of the distribution of all prediction errors over a historical time period are as follows: After obtaining the attitude parameters, for a preset historical time period prior to the current time, this historical time period is divided into several sub-segments. All the estimated errors obtained in each sub-segment constitute an error sample set. All the estimated errors in the error sample sets of all sub-segments are normalized to obtain the distribution characteristics of all normalized estimated errors in each error sample set. The distribution characteristics of the error sample sets corresponding to all sub-segments constitute a distribution characteristic sequence. The instability is determined by the correlation of all elements in the distribution characteristic sequence with time.

[0007] Preferably, the specific steps for obtaining the distribution characteristics of all normalized prediction errors in each error sample set are as follows: For any dimension of the normalized prediction error, calculate the mean and standard deviation of all normalized prediction errors in each error sample set for that dimension, and use the vector formed by the mean and standard deviation as the dimension feature of that dimension; the vector formed by concatenating the dimension features of all dimensions is used as the distribution feature of each error sample set.

[0008] Preferably, the distribution characteristics of the error sample set corresponding to all the sub-segments constitute a distribution characteristic sequence, and the instability is determined by the correlation of all elements in the distribution characteristic sequence over time. The specific steps include the following: The mean of the time contained in the sub-segment is recorded as the reference time of the error sample set corresponding to the sub-segment; the distribution characteristics of all error sample sets are arranged in chronological order according to the reference time to obtain the distribution characteristic sequence; The values ​​of each dimension of all elements in the distribution feature sequence constitute each reference sequence; the reference times of all error sample sets are arranged in chronological order to obtain the reference time series; the Pearson correlation coefficient between each reference sequence and the reference time series is calculated, and the sum of the absolute values ​​of the Pearson correlation coefficients between all reference sequences and the reference time series is denoted as the instability.

[0009] Preferably, updating the second weight means reducing the second weight by a preset percentage.

[0010] Preferably, the specific steps for updating the first weight are as follows: The target attitude data includes attitude angles and position; Minimum This is denoted as the predicted position; The position from the target attitude data is incorporated into the predicted position using a second weight to obtain the position parameters of the hoisted component; the position parameters are expressed as... The attitude parameters are expressed as , the objective function In 1 is considered as the quantity to be optimized and solved. Optimize, the optimized Make the objective function Take the minimum value and Similarly, the optimized version at this time As the first weight after the update.

[0011] Preferably, whenever the second weight is updated, if the next update is of the first weight, the predicted error obtained after updating the first weight is marked as the target error; the specific steps included each time the first weight is updated are as follows: The most recently labeled target errors are denoted as target error samples, and the mean of any dimension among all target error samples is obtained. and standard deviation ,Will As the first interval, if any dimension of the most recently obtained prediction error does not fall within the first interval, the second weight will be rolled back to the value before the most recent update.

[0012] Preferably, another embodiment of the present invention provides an attitude parameter measurement system for a wind turbine, the system including a multi-sensor data acquisition module, the module comprising: an IMU sensor on the lifting device, an IMU sensor on the lifted component, and an inclination sensor on each rope; the module uses the data output by the IMU sensors and calculates reference attitude data and target attitude data based on an inertial navigation algorithm; the module uses the inclination sensors to acquire the inclination data of the ropes; The system also includes an attitude parameter measurement module, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor runs the computer program, it uses reference attitude data, target attitude data, and tilt angle data collected by the multi-sensor data acquisition module to implement all the steps of the above-mentioned wind turbine attitude parameter measurement method. The system also includes a visualization module that displays a 3D model of the hoisted component, with the attitude parameters of the 3D model set to the attitude parameters measured by the attitude parameter measurement module.

[0013] The beneficial effects of the technical solution of the present invention are: This invention obtains the predicted attitude by fusing the lifting reference attitude data and rope tilt angle data with a first weight, and then obtains the final attitude parameters by fusing the predicted attitude with the target attitude data directly measured by the lifted component with a second weight. The first and second weights are adaptively updated based on the temporal instability of the historical prediction error distribution, realizing the dynamic adaptive fusion of multi-sensor data. This can continuously suppress the accumulation of IMU integral error during the long-term lifting of wind turbines and effectively improve the measurement accuracy of the attitude parameters of the lifted component.

[0014] Specifically, the degree of error accumulation is judged by predicting the temporal instability of the error distribution. When the instability is greater than the threshold, the second weight is updated, which can quickly suppress the excessive integration of IMU accumulated error in the target attitude data. When the instability is less than the threshold, the first weight is updated, which can further optimize the fusion ratio of tilt angle data after the error accumulation is controlled, and continuously optimize the attitude measurement accuracy. The above dynamic weight update mechanism can adapt to the changes in error characteristics during hoisting, and further ensure the stability and accuracy of attitude measurement during long-term hoisting. Attached Figure Description

[0015] Figure 1 This is a flowchart illustrating the steps of a method for measuring the attitude parameters of a wind turbine, as provided in an embodiment of the present invention. Detailed Implementation

[0016] To facilitate the description of all embodiments of the present invention, some preparatory work performed before running all embodiments of the present invention will be described first: 1. In the 3D design phase of the wind turbine's components to be hoisted, the positions and number of lifting points on the components are determined in advance. Finite element analysis is then used to simulate and evaluate the component deformation under different lifting point arrangements, thereby selecting the optimal lifting point positions (i.e., obtaining the position coordinates of the lifting points in the component's coordinate system). This ensures uniform stress and initial attitude stability during hoisting. This preparatory work eliminates the need for manual measurement and temporary determination of lifting points on-site, improving hoisting efficiency and providing a precise geometric reference for subsequent attitude calculations.

[0017] 2. Multiple hook lifting points are also set below the lifting device. The position coordinates of each hook lifting point in the lifting device coordinate system are obtained through calibration using the 3D model of the lifting device. For each component being lifted, each hook lifting point corresponds one-to-one with a component lifting point on it. During subsequent lifting, each pair (hook lifting point, component lifting point) is connected by ropes. Different rope lengths are selected for different components being lifted and different lifting point pairings to ensure that the component being lifted has a preset initial posture (e.g., horizontal posture) during lifting. The determination of rope length is also completed during the 3D design process of the component being lifted.

[0018] The coordinate system of the lifting device and the coordinate system of the lifted component both refer to the coordinate systems of the three-dimensional model of the lifting device and the three-dimensional model of the lifted component in the three-dimensional design stage (that is, the coordinate system in CAD).

[0019] In summary, before operating any embodiment of the present invention, the subsequent hoisting process (including the design of component hoisting points, the design of hook hoisting points and the ropes between component hoisting points, etc.) is considered during the 3D design process. Since the above-mentioned preparatory work is routine operation and existing technology in the 3D design stage of wind turbines, and is not the problem to be solved by the present invention, it will not be described in detail here.

[0020] The following description, in conjunction with the accompanying drawings, details the specific scheme of the attitude parameter measurement method and system for a wind turbine provided by the present invention.

[0021] Please see Figure 1 The diagram illustrates a flowchart of a method for measuring the attitude parameters of a wind turbine according to an embodiment of the present invention. The method includes the following steps: Step S101: For the components being hoisted during the wind turbine hoisting process, collect the reference attitude data of the hoisting equipment, as well as the inclination angle data of the rope between each component hoisting point and each hook hoisting point on the hoisting equipment, and collect the target attitude data of the components being hoisted.

[0022] In this embodiment, an IMU sensor is installed on the lifting device, and another IMU sensor is installed on the component being lifted. Additionally, an inclination sensor is installed on each rope. Once the component is lifted, each IMU sensor calculates its position and attitude angle in real time using an inertial navigation algorithm. The calculation results from the lifting device's IMU constitute the reference attitude data for the lifting device, while the calculation results from the IMU on the lifted component constitute the target attitude data. The inclination sensors on the ropes measure the inclination angle of each rope relative to the direction of gravity in real time. Specifically, each rope inclination sensor outputs two angle values: the angle α between the rope's projection in the XZ plane and the Z-axis (direction of gravity), and the angle β between the rope's projection in the YZ plane and the Z-axis (direction of gravity).

[0023] The XYZ coordinate system refers to a coordinate system with the upper end of each rope (i.e., each hook lifting point) as the origin and parallel to the lifting device coordinate system.

[0024] The inertial navigation algorithm is a well-known technology and will not be described in detail in this embodiment. It should be noted that all IMU sensors in this embodiment are of the same model, and the sampling frequency of all IMU sensors is 200Hz. Furthermore, this embodiment synchronously collects a set of multi-sensor data (i.e., the reference attitude data of the lifting device, the target attitude data of the lifted component, and the tilt angle data of all ropes) every time interval t. This set of multi-sensor data is then fused to complete one attitude parameter measurement process, resulting in an attitude parameter (including pitch angle, roll angle, and yaw angle). In this embodiment, t is set to 0.5 seconds. In other embodiments, the value of t can be set according to the specific processor computing power. For example, for embodiments with high computing power, the value of t can be reduced, for example, t can be set to 0.1 seconds, to further ensure the real-time performance of the measurement process. Conversely, for embodiments with low computing power, the value of t can be increased, for example, t can be set to 1 second, to avoid the lag situation where the previous measurement process has not yet finished when the next measurement process arrives. This embodiment does not specifically limit the value of t.

[0025] One simple method to determine whether a component being hoisted has been lifted is: Before the hoisting operation begins, a pressure sensor is placed under the component to be hoisted. Before hoisting, the pressure sensor will show a reading greater than 0 due to the reaction force of the weight of the component. When the pressure sensor shows a reading of 0, it is determined that the component to be hoisted is to be lifted.

[0026] Step S102: Use the first weight to integrate the tilt angle data into the reference attitude data to obtain the predicted attitude of the hoisted component.

[0027] Although the aforementioned reference attitude data and tilt angle data are not direct measurements of the attitude of the lifted component, they originate from different components of the lifting device and collectively constitute the geometric and mechanical constraints on the attitude of the lifted component. This embodiment utilizes a first weight to fuse these multi-sensor data to obtain the predicted attitude of the lifted component. This process takes into account the definite kinematic constraint relationship between the attitude of the lifting device, the tilt angle of the rope, and the attitude of the lifted component. Knowing the spatial positions of the rope's connection points and the rope direction (described by the tilt angle), combined with the attitudes of the lifting device and each hook's lifting point, the attitude of the lifted component can be calculated in reverse. The reference attitude data (from the IMU) and the tilt angle data (from the tilt sensor) have different physical meanings and error characteristics. The role of the first weight is to reasonably control the contribution ratio of sensor data with different physical meanings and error characteristics in the fusion process, thereby obtaining a more reliable predicted attitude than data from a single sensor.

[0028] Step S103: Use the second weight to integrate the attitude angle in the target attitude data into the predicted attitude to obtain the attitude parameters of the hoisted component, and record the difference between the attitude parameters and the attitude angle in the target attitude data as the prediction error.

[0029] The predicted attitude is inferred from the indirect constraints of the hoisting device, while the target attitude data is a direct measurement of the hoisted component. Both represent observations of the hoisted component, but they utilize different observation methods. This embodiment fuses the observation results from the two methods to obtain attitude parameters of the hoisted component that more closely approximate its true attitude.

[0030] This embodiment utilizes a second weight to perform a secondary fusion of the predicted attitude and target attitude data to obtain the final output attitude parameters. The second weight controls the relative contribution of the two observation methods in the fusion, ensuring that the attitude parameters take into account the observation results of both methods and are constrained by multiple sensor data. Compared to using only one observation method or sensor, the obtained attitude parameters are more accurate and reliable. The difference between the attitude parameters and the attitude angles in the target attitude data is defined as the prediction error, which serves as a feedback indicator of the fusion effect.

[0031] Step S104: Obtain the instability of the distribution of all estimated errors over time within the historical time period. When the instability is greater than or equal to the preset instability threshold, update the second weight; when the instability is less than the preset instability threshold, update the first weight; use the updated first and second weights to continue obtaining the attitude parameters of the hoisting component.

[0032] This embodiment further obtains the distribution of all prediction errors within a historical time period and evaluates the instability of this distribution over time. Instability reflects whether the prediction error shows a trend of continuous growth (or continuous change). When the instability is large, that is, when the instability is greater than or equal to the preset instability threshold, it indicates that the prediction error is continuously accumulating, indicating that the accumulated error in the target attitude data is being over-integrated. At this time, this embodiment updates the second weight (without updating the first weight) to suppress the propagation of accumulated error by changing the fusion weight of the target attitude data.

[0033] When the instability is small, i.e., less than the preset instability threshold, it indicates that the distribution of the estimated error is relatively stable, meaning that the main source of the attitude parameter error has been effectively controlled. At this time, the integration degree of the target attitude data is appropriate. To further improve accuracy, this embodiment chooses to update the first weight because: the tilt sensor is based on the static tilt angle of the gravity accelerometer (calculated only by the gravity component). When the cumulative error of the target attitude data is effectively suppressed, the error of the tilt sensor data (which does not accumulate over time but has fixed measurement noise) becomes another factor that can be optimized to limit accuracy. By adjusting the first weight, the integration method of the tilt data can be optimized, further improving the accuracy of the attitude parameters.

[0034] In summary, this embodiment fuses physically constrained sensor data from different sensors using a first weight and a second weight to obtain the attitude parameters of the hoisted component. These attitude parameters are then displayed on a screen in the hoisting control room, allowing hoisting personnel to monitor the component's attitude. Furthermore, based on dynamic feedback of historical prediction errors, the first and second weights are updated, achieving dynamic fusion of different sensors. This dynamic feedback mechanism and fusion process enable this embodiment to adaptively adjust the fusion strategy according to actual error characteristics during long-term hoisting operations, continuously suppressing measurement errors in the attitude parameters and improving their accuracy.

[0035] In a preferred implementation of this embodiment, the predicted attitude of the hoisted component is obtained by incorporating the tilt angle data into the reference attitude data using a first weight, and the method includes: This embodiment describes the hoisting process of any hoisted component as an example. The hoisted component has N component lifting points, and the hoisting device has N hook lifting points. That is, in this embodiment, N ropes are used in the hoisting process; in this embodiment, N=5.

[0036] In this embodiment, the mass of the hoisted component is much greater than that of the hoisting device, and multiple ropes constrain the hoisted component to the hoisting device. Therefore, the operation of the hoisted component and the operation of the hoisting device are the same in a short period of time, and there is a physical geometric constraint.

[0037] The position coordinates of the i-th hook lifting point in the lifting coordinate system are represented as follows: The position coordinates of the i-th component lifting point in the coordinate system of the component being lifted are represented as follows: The effective length of the i-th rope (the rope between the i-th hook lifting point and the i-th component lifting point) is (i.e. arrive (straight-line distance).

[0038] in , and All parameters are static and were calibrated during the 3D design phase before hoisting operations. Real-time measurement is not required during the hoisting process in all embodiments of this invention. This refers to the length of the rope under tension when the component is being lifted (i.e., under lifting conditions). arrive The straight-line distance (the length of which is measured in the laboratory through a lifting test).

[0039] The position of the i-th hook lifting point in the world coordinate system is: : ,in This indicates the position of the spreader in the world coordinate system, which is also the position in the reference attitude data of the spreader obtained in step S101. The rotation matrix is ​​the one from the hook coordinate system to the world coordinate system. The attitude angles in this rotation matrix (i.e., the vector composed of the three dimensions of pitch angle, roll angle, and yaw angle) refer to the attitude angles in the reference attitude data of the spreader obtained in step S101. The world coordinate system does not need to be specified separately; it is sufficient to use the coordinate system of the position in the reference attitude data of the spreader obtained in step S101 as the world coordinate system (i.e., there is no need to give an explicit definition of the world coordinate system, which does not affect the implementation of this embodiment).

[0040] The position of the i-th component's lifting point in the world coordinate system is: : ,in Let be the position of the hoisted component in the world coordinate system, and be the quantity to be solved. The rotation matrix from the coordinate system of the hoisted component to the world coordinate system is also the quantity to be solved.

[0041] Since the lengths of the ropes at the i-th hook lifting point and the i-th component lifting point are... Therefore, the constraints are satisfied. ,in This represents the magnitude of the vector, that is... and The distance.

[0042] Wherein, the direction vector of the i-th rope in the lifting coordinate system is ,in .in and The tilt angle data is output by the rope tilt angle sensor. It is obtained based on well-known geometric reasoning, and the specific reasoning process will not be described in detail in this embodiment.

[0043] This direction vector Represented in the world coordinate system as Therefore, this embodiment also has the following constraints: .

[0044] Furthermore, this embodiment constructs a configuration based on the above constraints regarding... objective function :

[0045] in 1 indicates the first weight, used to control the contribution ratio of tilt angle data in the fusion process. In this embodiment... The initial value is set to 1.0, where Used to remove dimensions.

[0046] This indicates the position of the i-th component's lifting point in the world coordinate system. The position of the i-th hook lifting point in the world coordinate system The distance (i.e.) The difference between the length of the first rope and the length of the second rope is denoted as the first difference.

[0047] Represents a unit vector With unit vector The difference between the two unit vectors is denoted as the second difference, which is equal to the square of the Euclidean distance between them.

[0048] This embodiment uses the Gauss-Newton method to solve for... Minimum The possible values ​​of , where Minimum Let be the prediction rotation matrix, and let the attitude angles in the prediction rotation matrix be denoted as the prediction attitude, where Minimum The predicted position is denoted as . The Gauss-Newton method is a commonly used existing technique, and will not be described in detail here.

[0049] The above , as well as The acquisition process involves rotation and translation transformations of the three-dimensional coordinate system, which is a well-known process and will not be described in detail in this embodiment. , as well as The specific principles and details of the formula.

[0050] In a preferred implementation of this embodiment, the attitude parameters of the hoisted component are obtained by incorporating the attitude angles in the target attitude data into the predicted attitude using a second weight. The method includes: Attitude parameters of the hoisted component ,in This represents the attitude angle in the target attitude data obtained in step S101. Indicates the predicted attitude. This indicates the second weight.

[0051] In a preferred implementation of this embodiment, the method for obtaining the instability of the distribution of all prediction errors over a historical time period as a function of time includes: As can be seen from the above process, after collecting each set of multi-sensor data (i.e., the reference attitude data of the lifting device, the target attitude data of the lifted component, and the inclination angle data of all ropes), an attitude parameter of the lifted component is obtained according to the above method, and a prediction error is further obtained. Here, the attitude parameter is temporarily regarded as the true attitude of the lifted component, and the prediction error represents the error of the attitude measurement result of the lifted component. In this embodiment, it is used to describe the feedback index of the fusion effect of a set of multi-sensor data.

[0052] Each time attitude parameters are obtained, a preset historical time period prior to the current time is divided into multiple sub-segments. All estimated errors obtained within each sub-segment constitute an error sample set. It should be noted that, since the attitude parameters or attitude angles in this embodiment include three dimensions—pitch angle, roll angle, and yaw angle—each estimated error in the error sample set is also a vector composed of these three dimensions.

[0053] Normalize all predicted errors in the error sample set of all sub-segments: In this embodiment, all predicted errors are linearly normalized to values ​​in the same dimension. Specifically, if all values ​​in the same dimension are the same, the normalization result is set to 0. Note that in subsequent processes, when using sequences with all element values ​​of 0 to participate in the calculation of the Pearson correlation coefficient, the Pearson correlation coefficient can be directly set to 0.

[0054] Obtain the distribution characteristics of all normalized prediction errors in each error sample set. The distribution characteristics of the error sample sets corresponding to all sub-segments constitute a distribution characteristic sequence, and the instability is determined by the correlation of all elements in the distribution characteristic sequence over time.

[0055] Preferably, the historical time period refers to 2 to 5 minutes before the current time (inclusive); as an example, the historical time period is set to 3 minutes before the current time. If there are less than 3 minutes before the current time, the first weight and the second weight are not updated (i.e., step S104 is not executed). When the historical time period is 3 minutes before the current time, the historical time period is divided into 10 equal sub-segments. In other embodiments, the duration of the historical time period and the number of sub-segments can be set to other values. This embodiment does not impose specific limitations, but it is necessary to ensure that the error sample set of each sub-segment has at least 11 estimated errors and at least 5 sub-segments, so as to avoid inaccurate acquisition of distribution features due to too few estimated errors in the error sample set, and to avoid unclear relationship between distribution features and time due to too few sub-segments.

[0056] Preferably, the method for obtaining the distribution characteristics of all normalized prediction errors in each error sample set includes: For any dimension of the normalized prediction error, calculate the mean and standard deviation of all normalized prediction errors in that dimension in each error sample set. In this embodiment, the vector formed by the mean and standard deviation is used as the dimension feature of that dimension. The vector formed by concatenating the dimension features of all dimensions (i.e., the three dimensions of the normalized prediction error) is used as the distribution feature corresponding to each error sample set (each distribution feature is represented as a 6-dimensional vector).

[0057] Preferably, the instability is determined by the correlation of all elements in the distribution feature sequence over time, including the following methods: The mean of the times contained in a sub-segment is denoted as the reference time of the error sample set corresponding to that sub-segment. The distribution characteristics of all error sample sets are arranged in chronological order according to the reference times to obtain a distribution characteristic sequence. That is, each element in the distribution characteristic sequence is a distribution characteristic (with 6 dimensions). The values ​​of all elements in the distribution characteristic sequence in the same dimension constitute a reference sequence. Since the distribution characteristics have 6 dimensions, 6 reference sequences are obtained.

[0058] Arrange the reference times of all error sample sets in chronological order to obtain a reference time series. Calculate the Pearson correlation coefficient between each reference series and the reference time series. The sum of the absolute values ​​of the Pearson correlation coefficients of all reference series and reference time series is denoted as the instability. The larger the sum of the absolute values ​​of the Pearson correlation coefficients, the more significant the change in the distribution characteristics over time, indicating a significant change in the distribution of prediction errors over historical periods. In this case, the greater the instability, the more it indicates that the prediction errors are accumulating. Conversely, the smaller the sum of the absolute values ​​of the Pearson correlation coefficients, the less significant the change in the distribution characteristics over time. In this case, the smaller the instability, the more stable the distribution of prediction errors.

[0059] In a preferred implementation of this embodiment, the second weight is updated when the instability is greater than or equal to a preset instability threshold; the first weight is updated when the instability is less than the preset instability threshold. The method includes: When the instability is greater than or equal to the preset instability threshold, it indicates that the prediction error is accumulating continuously, which means that the accumulated error in the target attitude data is being over-integrated. At this time, the second weight is reduced; note that the first weight remains unchanged.

[0060] As an example, this embodiment reduces the second weight by 10%, where the initial value of the second weight is set to 0.65. When the second weight becomes too small, for example, less than 0.15, the second weight is no longer reduced.

[0061] When the instability is less than the preset instability threshold, it indicates that the distribution of the estimated error is stable, meaning that the main source of the attitude parameter error has been effectively controlled. To further improve accuracy, this embodiment selects to update the first weight, specifically including: Since both the attitude angle of the target attitude data and the predicted attitude are attitude estimates of the hoisted component, but both have errors with different distributions, when the attitude angle of the target attitude data is integrated into the predicted attitude to obtain attitude parameters, and the distribution of the predicted error is stable (i.e., the instability is less than a preset instability threshold), it indicates that the fused attitude parameters are more reliable than the predicted attitude obtained by a single measurement method. It also indicates that the error in the tilt angle data is another optimizable factor limiting accuracy. Therefore, this embodiment first uses a second weight to integrate the position in the target attitude data into the predicted position to obtain the position parameters of the hoisted component. The position parameters are expressed as... The attitude parameters are expressed as The above objective function In 1 is considered as the quantity to be optimized. Optimize, the optimized Make the objective function Take the minimum value and Similarly, the optimized version at this time This serves as the first weight after the update. That is, the objective function described above. middle After replacing with the updated first weight, When minimized The value of is equal to the more accurate and reliable one mentioned above. This indicates that the updated first weight can reasonably integrate the tilt angle data into the reference attitude data, ensuring that the obtained attitude parameters are more accurate.

[0062] Specifically, the position in the target attitude data is incorporated into the predicted position using a second weight to obtain the position parameters of the hoisted component; this process is similar to the process of obtaining attitude parameters, and specifically includes: the position parameters of the hoisted component. ,in This indicates the position in the target attitude data obtained in step S101. Indicates the predicted location.

[0063] In summary, the above process achieves the update of the first weight or the second weight. When the attitude parameters are measured again (that is, when steps S102 and S103 are executed again), the updated first weight or the second weight is used. Then, the instability is obtained, and the first weight or the second weight is updated again based on the magnitude of the instability (that is, step S104 is executed again based on the updated first weight or the second weight). Thus, while the attitude parameters of the hoisted component are continuously measured, the first weight and the second weight are dynamically updated, thereby improving the accuracy of the attitude parameters.

[0064] It should be noted that the objective function of the above process is... In the process, the collected reference attitude data also contains accumulated errors. Therefore, the predicted attitude indirectly includes accumulated errors. That is, when the instability is greater than or equal to the preset instability threshold, the reason for the continuous accumulation of the prediction error is not only the integration of accumulated errors in the target attitude data, but also the integration of accumulated errors indirectly included in the predicted attitude. However, the process of obtaining the predicted attitude is achieved through two processes: the integration of tilt angle data and the optimization of the objective function. Compared with the target attitude data that directly contains accumulated errors, the accumulated error of the predicted attitude is smaller. That is, the continuous accumulation of the prediction error is mainly affected by the accumulated error of the target attitude data. Therefore, when the instability is large (that is, greater than or equal to the preset instability threshold), the second weight is updated first, rather than the first weight. When the instability obtained after updating the second weight becomes smaller (that is, less than the preset instability threshold), it indicates that by modifying the integration degree of the target attitude data, the accumulated error of the target attitude data is weakened. Only on this basis can the first weight be updated.

[0065] It should be further noted that the purpose of dynamically updating the first and second weights in this embodiment is not to completely remove all errors (such as the cumulative error between the reference attitude data and the target attitude data), but to minimize the introduction of errors by optimizing the fusion strategy of multi-sensor data, thereby further improving the accuracy of attitude parameter measurement results.

[0066] To further improve the accuracy of attitude parameter measurement results, some other embodiments also include the following steps: Whenever the second weight is updated, if the first weight is updated instead of the second weight during the next update (i.e., when updating the first or second weight), the predicted error obtained after updating the first weight is marked as the target error.

[0067] The above process indicates that after the second weight is updated, the first weight is then updated. This demonstrates that by progressively changing the fusion strategy for multiple sensor data, the objective function is optimized while mitigating the error accumulation effect. This optimizes the fusion strategy of multiple sensor data, resulting in more reliable attitude parameters. Based on this, the predicted error (i.e., target error) can relatively accurately describe the error distribution of the target attitude data.

[0068] After each update of the first weight, the n most recently labeled target errors are obtained and denoted as target error samples. Each target error sample is also multi-dimensional (i.e., including pitch angle, roll angle, and yaw angle). The mean of all target error samples in any dimension is obtained. and standard deviation ,Will As the first interval of this dimension, if any value of any dimension in the most recently obtained prediction error is not in the first interval of this dimension, the value of the second weight is restored, that is, the second weight is rolled back to the value before the most recent update, and at most rolled back to its initial value and then no longer rolled back.

[0069] Preferably, n is greater than or equal to 5. This embodiment ensures a sufficient number of target error samples for calculating the mean. and standard deviation And ensure that the target error sample includes the nearest target error. Here, we will use n=9 as an example for explanation. When there are fewer than 9 target errors, the first interval is no longer obtained. At this time, if the second weight is not too small, for example, not less than 0.15, then no rollback is performed; otherwise, the rollback continues to prevent the second weight from approaching 0 further.

[0070] When a situation arises where the target orientation data is not within the first interval of the corresponding dimension, it indicates that after multiple updates to the second weight in the aforementioned process, the second weight may have become too small. This leads to the target orientation data being ignored, essentially evaluating the orientation parameters of the hoisted component solely based on the predicted orientation (equivalent to losing the constraint or feedback effect of an IMU sensor), which is detrimental to improving the accuracy of the orientation parameters. In this case, this embodiment restores the value of the second weight to ensure long-term, continuous fusion of all sensor data. Through the fusion or constraint effect between different sensors, the accuracy of the parameters is improved.

[0071] The core purpose of this step is to suppress the problem of the second weight becoming too small after multiple updates, and to ensure that this embodiment always relies on the mutual fusion and mutual constraint of multiple sensor data to measure attitude parameters, thus avoiding the problem of a significant decrease in attitude measurement accuracy when a single sensor is used for long-term measurement.

[0072] As an example, for Optimize, the optimized Make the objective function Take the minimum value and Similarly, the optimized version at this time The methods included as the updated first weight are: The above objective function It can be viewed as a variable about the independent variable. Continuous functions This process is equivalent to: solving for a numerical value. This makes the continuous function When the minimum value is reached (at which point the function G is a multivariate quadratic function), TB and RB are equal to TB0 and RB0, respectively. This is a conventional mathematical problem, and it is obvious that the following results are obtained regarding... The equation:

[0073] Seeking the middle Representing continuous functions exist The partial derivatives with respect to TB and RB are used. This embodiment utilizes Brent's method to... Minimize, minimize That is the optimized version .

[0074] In summary, for Optimization is a conventional continuous function optimization problem and is a well-known technique. This embodiment will not provide specific reasoning or description.

[0075] In a preferred implementation of this embodiment, the method for setting the instability threshold is as follows: In the laboratory, set up a track (e.g., a circular track with a radius of 2 meters and an angle of 45° with the horizontal plane), and set up several sampling points on the track (e.g., one sampling point every 0.1 meters).

[0076] An object (e.g., a metal block of arbitrary mass or volume driven by a linear motor) is made to move along a circular track at a random speed (e.g., a random speed of -10 to 10 m / s) for 0.5 hours. The object is equipped with the same type of IMU as described in step S101. A photoelectric module is installed at each sampling point to detect whether the object has passed through each sampling point. When the object passes through any sampling point, the observation attitude angle of the object is calculated using an inertial navigation algorithm based on the data collected by the IMU.

[0077] In addition, the true attitude angles (including pitch, roll, and yaw angles) of the object at each sampling point are manually measured. For example, one sampling point is taken as a reference point, and its true attitude angles are set to [0, 0, 0]. The true attitude angles of other sampling points are determined by measuring the relative attitudes of other sampling points with respect to the reference point (that is, measuring the angles of rotation of other sampling points relative to the reference point in the three dimensions of pitch, roll, and yaw). Furthermore, the object starts moving from the reference point (that is, its observed attitude angles at the reference point are also [0, 0, 0]).

[0078] The difference between the true attitude angle and the observed attitude angle at each sampling point is recorded as a test error.

[0079] The pre-defined historical time period before the object's motion ends is divided into multiple sub-segments. All test errors obtained within each sub-segment constitute a test error sample set. All test errors in the test error sample sets of all sub-segments are then normalized. Obtain the distribution characteristics of all normalized test errors in each test error sample set. The distribution characteristics of the test error sample sets corresponding to all sub-segments constitute a test distribution characteristic sequence. The correlation of all elements in the test distribution characteristic sequence over time is denoted as test instability.

[0080] The process of testing instability and obtaining instability described above are similar, and will not be described in detail in this embodiment.

[0081] Since the instability in this embodiment is based on the attitude parameters obtained by fusion, the instability in this embodiment is smaller than the test instability obtained under a single sensor. Therefore, the maximum value of the instability threshold does not exceed the test instability. In this embodiment, the instability threshold is preferably set to 30-60% of the test instability. Specifically, in this embodiment, 45% of the test instability is used as the instability threshold.

[0082] Another embodiment of the present invention provides an attitude parameter measurement system for a wind turbine. This system includes a multi-sensor data acquisition module containing multiple sensors, including: an IMU sensor on the lifting device, an IMU sensor on the lifted component, and an inclination sensor on each rope. The module utilizes data output from the IMU sensors on the lifting device (or the lifted component) and calculates reference attitude data (or target attitude data) based on an inertial navigation algorithm; the module also uses the inclination sensors to acquire the inclination data of the ropes.

[0083] The system also includes an attitude parameter measurement module, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor runs the computer program, it uses a multi-sensor data acquisition module to acquire reference attitude data, target attitude data, and tilt angle data to implement all the steps of the above embodiments.

[0084] The system also includes a visualization module that displays a 3D model of the hoisted component, with the attitude parameters of the 3D model set to the attitude parameters measured by the attitude parameter measurement module.

[0085] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for measuring the attitude parameters of a wind turbine generator, characterized in that, The method includes the following steps: For the components being hoisted during the wind turbine hoisting process, reference attitude data of the hoisting equipment and the inclination angle data of the rope between each lifting point of the component and each hook lifting point of the hoisting equipment are collected, and target attitude data of the component being hoisted is also collected; the reference attitude data or target attitude data includes attitude angles. The tilt angle data is incorporated into the reference attitude data using a first weight to obtain the predicted attitude of the hoisted component. The attitude angles in the target attitude data are incorporated into the predicted attitude using a second weight to obtain the attitude parameters of the hoisted component. The difference between the attitude parameters and the attitude angles in the target attitude data is recorded as the prediction error. The instability of the distribution of all prediction errors over a historical period is obtained. When the instability is greater than or equal to a preset instability threshold, the second weight is updated; when the instability is less than the preset instability threshold, the first weight is updated. The updated first and second weights are used to continue obtaining the attitude parameters of the hoisted component.

2. The method for measuring the attitude parameters of a wind turbine according to claim 1, characterized in that, The specific steps for incorporating the tilt angle data into the reference attitude data using a first weight to obtain the predicted attitude of the hoisted component are as follows: Construct the objective function The objective function Positively correlated with the first difference, and positively correlated with the second difference The twofold positive correlation is present. in This indicates the position of the hoisted component in the world coordinate system. This represents the rotation matrix from the coordinate system of the hoisted component to the world coordinate system; the first difference refers to the position of the i-th component's hoisting point in the world coordinate system. The position of the i-th hook lifting point in the world coordinate system The difference between the distance and the length of the i-th rope; Depend on , Sure; Determined by the reference attitude data of the spreader; the second difference refers to: unit vector and Differences; This represents the direction vector of the i-th rope in the world coordinate system, which is determined by the inclination angle data of the i-th rope; 1 indicates the first weight. Represents the magnitude of the vector; Minimum Let denot be the predicted rotation matrix; let the attitude angles in the predicted rotation matrix be denoted as the predicted attitude.

3. The method for measuring the attitude parameters of a wind turbine according to claim 1, characterized in that, The attitude parameters of the hoisted component ,in This represents the attitude angle in the target attitude data. Indicates the predicted attitude. This indicates the second weight.

4. The method for measuring the attitude parameters of a wind turbine according to claim 1, characterized in that, The specific steps involved in obtaining the instability of the distribution of all prediction errors over a historical time period are as follows: After obtaining the attitude parameters, for a preset historical time period prior to the current time, this historical time period is divided into several sub-segments. All the estimated errors obtained in each sub-segment constitute an error sample set. All the estimated errors in the error sample sets of all sub-segments are normalized to obtain the distribution characteristics of all normalized estimated errors in each error sample set. The distribution characteristics of the error sample sets corresponding to all sub-segments constitute a distribution characteristic sequence. The instability is determined by the correlation of all elements in the distribution characteristic sequence with time.

5. The method for measuring the attitude parameters of a wind turbine according to claim 4, characterized in that, The specific steps for obtaining the distribution characteristics of all normalized prediction errors in each error sample set are as follows: For any dimension of the normalized prediction error, calculate the mean and standard deviation of all normalized prediction errors in each error sample set for that dimension, and use the vector formed by the mean and standard deviation as the dimension feature of that dimension; the vector formed by concatenating the dimension features of all dimensions is used as the distribution feature of each error sample set.

6. The method for measuring the attitude parameters of a wind turbine according to claim 5, characterized in that, The distribution characteristics of the error sample sets corresponding to all the sub-segments constitute a distribution characteristic sequence. The instability is determined by the correlation of all elements in the distribution characteristic sequence over time, including the following specific steps: The mean of the time contained in the sub-segment is recorded as the reference time of the error sample set corresponding to the sub-segment; the distribution characteristics of all error sample sets are arranged in chronological order according to the reference time to obtain the distribution characteristic sequence; The values ​​of each dimension of all elements in the distribution feature sequence constitute each reference sequence; the reference times of all error sample sets are arranged in chronological order to obtain the reference time series; the Pearson correlation coefficient between each reference sequence and the reference time series is calculated, and the sum of the absolute values ​​of the Pearson correlation coefficients between all reference sequences and the reference time series is denoted as the instability.

7. The method for measuring the attitude parameters of a wind turbine according to claim 1, characterized in that, Updating the second weight means reducing the second weight by a preset percentage.

8. The method for measuring the attitude parameters of a wind turbine according to claim 2, characterized in that, The specific steps involved in updating the first weight are as follows: The target attitude data includes attitude angles and position; Minimum This is denoted as the predicted position; The position from the target attitude data is incorporated into the predicted position using a second weight to obtain the position parameters of the hoisted component; the position parameters are expressed as... The attitude parameters are expressed as , the objective function In 1 is considered as the quantity to be solved. Optimize, the optimized Make the objective function Take the minimum value and Similarly, the optimized version at this time As the first weight after the update.

9. The method for measuring the attitude parameters of a wind turbine according to claim 8, characterized in that, Whenever the second weight is updated, if the next update is of the first weight, the predicted error obtained after updating the first weight is marked as the target error. The specific steps involved after each update of the first weight are as follows: The most recently labeled target errors are denoted as target error samples, and the mean of any dimension among all target error samples is obtained. and standard deviation ,Will As the first interval, if any dimension of the most recently obtained prediction error does not fall within the first interval, the second weight will be rolled back to the value before the most recent update.

10. A wind turbine attitude parameter measurement system, characterized in that, The system includes a multi-sensor data acquisition module, which contains: IMU sensors on the lifting device, IMU sensors on the lifted component, and tilt sensors on each rope; the module uses the data output by the IMU sensors and calculates the reference attitude data and target attitude data based on the inertial navigation algorithm; the module uses the tilt sensors to acquire the tilt angle data of the ropes; The system also includes an attitude parameter measurement module, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor runs the computer program, it uses reference attitude data, target attitude data, and tilt angle data collected by the multi-sensor data acquisition module to implement all the steps of the attitude parameter measurement method for a wind turbine according to any one of claims 1 to 9. The system also includes a visualization module that displays a 3D model of the hoisted component, with the attitude parameters of the 3D model set to the attitude parameters measured by the attitude parameter measurement module.

Images