A wind turbine blade active obstacle avoidance method

Abstract

The application provides an active obstacle avoidance method for a wind turbine blade, comprising: installing a laser radar and a visual sensor on a wind turbine generator, and the active obstacle avoidance steps are as follows: S1, performing point cloud sampling and image taking on the laser radar in front of a fan impeller; S2, filtering the point cloud sampling of the laser radar; S3, removing outliers in the point cloud sampling data processed in S2; S4, obtaining the coordinate position and range of the foreign matter in front in a three-dimensional coordinate system, and sending the coordinate position to a control mechanism of a multi-dimensional rotating mechanism of the visual sensor; S5, identifying and classifying the anchor box image by using a cross entropy function; and S6, performing speed control on the impeller according to the coordinate position and range of the foreign matter. The active control of the wind turbine is used to control the speed of the impeller, so that the object is prevented from colliding with the impeller; and the data information of each active obstacle avoidance is collected, the data is trained in an offline end, and the obstacle avoidance action of the wind turbine is optimized.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of wind power technology, in particular to a wind turbine blade active obstacle avoidance method. BACKGROUND

[0002] With the continuous growth of global demand for renewable energy, wind power as a clean and renewable energy form has become increasingly important. The development of wind power technology not only concerns the optimization of energy structure, but also is one of the key ways to achieve carbon neutrality. Wind turbine blades, as the core components of wind turbine generators, their performance directly determines the efficiency and reliability of wind power generation. However, in the actual operation process, wind turbine blades face a variety of complex environmental challenges, one of which is how to effectively avoid collisions with obstacles in front.

[0003] Traditional wind turbine blade obstacle avoidance methods mainly rely on passive protection measures, such as setting up protective nets, increasing blade strength, etc. Although these methods can reduce the risk of blade damage to some extent, they cannot fundamentally solve the problem of collisions between blades and obstacles. Especially in complex and variable wind field environments, such as mountainous areas, urban outskirts, and other complex terrain areas, wind turbine blades are more vulnerable to threats from obstacles such as birds, trees, buildings, etc.

[0004] In recent years, with the rapid development of image recognition technology, machine learning algorithms, and sensor technology, new possibilities have been provided for active obstacle avoidance technology. By obtaining real-time image information in front of the wind turbine impeller, and combining advanced image recognition technology, the rapid and accurate identification of objects in front can be achieved. Further, based on the pre-loaded object database, a weight library can be constructed to determine the avoidance priority of different object types, thereby making more intelligent obstacle avoidance decisions.

[0005] However, existing active obstacle avoidance technology still faces many challenges. On the one hand, how to ensure the efficiency of wind turbine power generation while achieving precise obstacle avoidance of the blade is a problem that needs to be solved. On the other hand, due to the complexity and uncertainty of the wind turbine operating environment, how to continuously optimize the obstacle avoidance strategy based on real-time data to improve the accuracy and reliability of obstacle avoidance is also a focus of current research.

[0006] Therefore, the application provides a wind turbine blade active obstacle avoidance method, which obtains image information in front of the impeller, combines image recognition technology and an object database, and constructs a complete active obstacle avoidance system. The system can not only determine the type and avoidance priority of the object in front of the wind turbine in real time, but also predict the motion trajectories of the wind turbine and the obstacle according to the position information of the impeller of the wind turbine and the position of the obstacle, so as to determine the point to be avoided. The speed of the impeller can be adjusted through active control of the wind turbine, so that the object can be effectively prevented from colliding with the impeller, and the safe operation of the wind turbine is ensured. Meanwhile, the method also collects data information of each active obstacle avoidance, trains and optimizes the data in an offline terminal, and continuously improves the accuracy and efficiency of the obstacle avoidance action of the wind turbine. The proposal of the innovative technology is expected to provide new ideas and solutions for the active obstacle avoidance technology in the field of wind power generation. SUMMARY

[0007] The application provides a wind turbine blade active obstacle avoidance method, which avoids and controls foreign matters according to detection of a wind turbine generator.

[0008] To solve the above technical problems, the technical scheme adopted by the application is:

[0009] A wind turbine blade active obstacle avoidance method, comprising installing a laser radar and a visual sensor on a wind turbine generator for detecting in front of the impeller of the wind turbine, and the visual sensor is provided with a multi-dimensional rotating mechanism for aligning foreign matters, and the active obstacle avoidance steps are as follows:

[0010] S1, performing point cloud sampling and image taking of the laser radar in front of the impeller of the wind turbine;

[0011] S2, filtering the point cloud sampling of the laser radar;

[0012] S3, removing outliers in the point cloud sampling data processed in S2;

[0013] S4, establishing a three-dimensional coordinate system with the beam emission point of the laser radar as the center, performing segmentation and clustering on the laser radar point cloud data, establishing a rotating model of the impeller of the wind turbine generator, collecting the blade azimuth angle and the rotating speed in the SCADA system of the wind turbine generator, establishing the current blade azimuth impeller model, and fitting the segmented and clustered laser radar point cloud data, so as to obtain the impeller point cloud data in the laser radar point cloud data, remove the impeller point cloud data from the segmented and clustered laser radar point cloud data to obtain the segmented and clustered point cloud data in front of the impeller of the wind turbine after removing the impeller, and obtain the coordinate position and range of the foreign matter in front in the three-dimensional coordinate system, and send the coordinate position to the control mechanism of the multi-dimensional rotating mechanism of the visual sensor;

[0014] S5, the control mechanism of the visual sensor controls the alignment of the multi-dimensional rotating mechanism to the foreign matter direction for visual image taking, inputs the visual taken image into the target detection network to output an anchor box, and uses a cross entropy function to recognize and classify the anchor box image;

[0015] S6, after the target type of the foreign matter is determined, whether the foreign matter will collide with the blade is judged according to the coordinate position and range of the foreign matter, the blade azimuth angle and the current rotating speed in the SCADA system, and the speed control of the impeller is performed in combination with the obstacle avoidance logic.

[0016] The point cloud sampling in S1 is downsampling, and the filtering in S2 is voxel filtering.

[0017] The outlier removal of the point cloud sampling data in S3 adopts a distance-based K-nearest neighbor algorithm.

[0018] The specific process of the distance-based K-nearest neighbor algorithm in S3 is as follows:

[0019] S3.1, setting a neighborhood point number parameter K;

[0020] S3.2, calculating the average distance of each point to K nearest neighbor points :

[0021] Suppose the coordinates of the i-th three-dimensional point in the point cloud data are xi, yi and zi, then the distance d i between the i-th three-dimensional point and any three-dimensional point is:

[0022] ;

[0023] The average distance d i of the i-th three-dimensional point to K nearest neighbor points is:

[0024] ;

[0025] S3.3, calculating the mean and standard deviation of the average distance, and then calculating the distance threshold T :

[0026] Mean:

[0027] ;

[0028] Standard deviation:

[0029] ;

[0030] Distance threshold: T :

[0031] ; ​​​​​

[0032] wherein is the mean of all average distances, is the set standard deviation multiple, which can change the size of the distance threshold according to the needs, is the standard deviation of the average distance;

[0033] S3.4, set the points with average distance exceeding the distance threshold as outliers;

[0034] S3.5, filter out the outliers.

[0035] In the preferred scheme, the wind turbine impeller rotation model in S4 is established by pre-establishing a single blade three-dimensional information, combining the parameters of the hub, the three-dimensional coordinate system with the laser radar beam emission point as the center, the current blade azimuth angle and the rotational speed, and the coordinate position relationship between the hub center and the beam emission point to obtain the three-dimensional rotation model of the wind turbine blade and the hub.

[0036] In the preferred scheme, the pre-established single blade three-dimensional information in S4 is established by replacing the single blade with a cylindrical model to establish the three-dimensional model of the blade.

[0037] In S6, the obstacle avoidance logic of the impeller control includes:

[0038] 1) judging whether the type of the foreign matter belongs to the type that needs to be avoided;

[0039] 2) judging whether the current blade rotational speed exceeds the minimum rotational speed that needs to be considered for obstacle avoidance;

[0040] 3) when conditions 1 and 2 are both met, judging whether the foreign matter will hit the blade by combining the distance between the current foreign matter and the blade and the speed of the foreign matter with the current blade rotational speed;

[0041] 4) if the result of 3 is that it will hit, calculating the current wind speed, blade pitch angle, and rotational speed to determine which kind of power generation power is greater when the rotational speed of the blade is increased or decreased corresponding to the adjustment of the pitch angle, and judging whether the obstacle avoidance of the impeller leaves enough safety distance when the condition of the greater power generation power is met, if yes, selecting the speed regulation mode of the greater power generation power, and if not, taking another adjustment mode.

[0042] In S6, when multiple foreign matters are detected, the obstacle avoidance logic of the impeller control includes:

[0043] 1) judging whether the type of the foreign matter belongs to the type that needs to be avoided;

[0044] 2) judging whether the current blade rotational speed exceeds the minimum rotational speed that needs to be considered for obstacle avoidance;

[0045] 3) When conditions 1 and 2 are met, the number of current foreign objects is combined with the current rotating speed to determine whether it will cause uncontrollable risk to the blades, if so, the impeller is controlled to operate along the stream and slow down, if not, the distance relationship between multiple foreign objects is determined, when the distance between two foreign objects is less than the set bonding distance, the two foreign objects are considered as a whole, the rotating speed of the impeller when each foreign object just passes between the two impellers is determined according to the number of current foreign objects and the speed thereof, and the adjustment mode of the impeller is calculated according to the current rotating speed and the rotating speed of the impeller when each foreign object just passes between the two impellers.

[0046] The laser radar and the visual sensor on the wind turbine generator set are each arranged on a tower and a nacelle.

[0047] The application provides a wind turbine blade active obstacle avoidance method, a point cloud model is established in front of the wind turbine generator set through a laser radar on a tower of the wind turbine generator set, a three-dimensional model of an impeller is established and fitted with point cloud data, data of the impeller is removed to obtain point cloud data in front of the wind turbine generator set, the type of foreign objects is classified in combination with visual detection, and then speed control is performed according to the current pitch angle and rotating speed of the blades, so that the blades are prevented from being damaged by foreign objects through active detection. BRIEF DESCRIPTION OF DRAWINGS

[0048] The application will be further described below in combination with the drawings and embodiments:

[0049] Figure 1 It is a hardware schematic diagram of the method;

[0050] Figure 2 It is Figure 1 a partial enlarged schematic diagram.

[0051] The laser radar and the visual sensor (1. DETAILED DESCRIPTION

[0052] To make the purpose, technical scheme and advantages of the application clearer, the following content will combine the drawings provided by the application to make a systematic and complete description of the specific technical scheme of the application. Obviously, the described embodiments are part of the embodiments of the application, rather than all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor belong to the protection scope of the application.

[0053] Embodiment 1:

[0054] As shown in the Figure 1 application, an active obstacle avoidance method for a wind turbine blade includes installing a laser radar and a visual sensor on a wind turbine generator set to detect in front of the blades of the wind turbine, the visual sensor is provided with a multi-dimensional rotating mechanism for aiming at foreign objects, and the steps of active obstacle avoidance are as follows:

[0055] S1, laser radar point cloud sampling and image taking are performed on the front of the fan impeller;

[0056] S2, the point cloud sampling of the laser radar is filtered to reduce the number of three-dimensional points of the cloud data while maintaining the shape characteristics of the point cloud, without destroying the characteristics of the point cloud;

[0057] S3, the outliers in the point cloud sampling data processed in S2 are removed;

[0058] S4, a three-dimensional coordinate system is established with the beam emission point of the laser radar as the center, the laser radar point cloud data is segmented and clustered, a rotating model of the wind turbine impeller is established, the blade azimuth angle and the rotating speed in the SCADA system of the wind turbine are collected, and the current blade azimuth impeller model is established and fitted with the segmented and clustered laser radar point cloud data, so that the impeller point cloud data in the laser radar point cloud data can be obtained. The segmented and clustered laser radar point cloud data is obtained by removing the impeller point cloud data, and the coordinate position and range of the foreign matter in front of the three-dimensional coordinate system are obtained. The coordinate position is sent to the control mechanism of the multi-dimensional rotating mechanism of the visual sensor;

[0059] S5, the control mechanism of the visual sensor controls the multi-dimensional rotating mechanism to take a visual image in the direction of the foreign matter, inputs the visual image into a target detection network to output an anchor box, and uses a cross-entropy function to identify and classify the anchor box image;

[0060] S6, after the target type of the foreign matter is determined, whether the foreign matter will collide with the blade is judged according to the coordinate position and range of the foreign matter, the blade azimuth angle in the SCADA system and the current rotating speed, and the speed control of the impeller is performed according to the obstacle avoidance logic.

[0061] The laser radar samples the front of the impeller, and the laser radar outputs the coordinate position and range of the foreign matter. A large amount of calculation and model calculation is required for the target recognition and analysis of this three-dimensional information, which is difficult and has a large amount of calculation. Therefore, the image information with shape and color in the visual sensor is used to identify and classify the foreign matter, which can greatly reduce the difficulty and amount of calculation, improve the accuracy of identification, and determine the current and future position information of the blade by establishing a blade model and according to the blade azimuth angle and the rotating position of the blade. Whether a collision accident will occur is judged by combining the current rotating speed, the movement speed and direction of the foreign matter, and different coping strategies are adopted according to the type of the foreign matter, and the rotating speed of the blade is regulated as needed.

[0062] The point cloud sampling in S1 above is downsampling, and the filtering in S2 is voxel filtering.

[0063] Point cloud downsampling can reduce data complexity and improve processing efficiency by reducing the number of points. Voxel-based filtering downsampling creates a voxel grid, loads the overall point cloud data into the voxel grid, and then collects all three-dimensional points in the grid to calculate the centroid, which is used as a sampling point to approximate other points in the grid to reduce the amount of data.

[0064] The outlier removal of the point cloud sampling data in S3 above adopts a distance-based K-nearest neighbor algorithm.

[0065] The specific process of the distance-based K-nearest neighbor algorithm in S3 above is as follows:

[0066] S3.1, set the number of selected neighborhood points parameter K;

[0067] S3.2, calculate the average distance of each point to K nearest neighbors :

[0068] Let the coordinates of the i-th three-dimensional point in the point cloud data be xi, yi, and zi, then the distance di between the i-th three-dimensional point and any three-dimensional point is:

[0069] ;

[0070] The average distance of the i-th three-dimensional point to K nearest neighbors is:

[0071] ;

[0072] S3.3, calculate the mean and standard deviation of the average distance, and then calculate the distance threshold T :

[0073] Mean:

[0074] ;

[0075] Standard deviation:

[0076] ;

[0077] Distance threshold: T :

[0078] ;

[0079] wherein the mean of all average distances is , the set standard deviation multiple is , which can change the size of the distance threshold according to the demand, and the standard deviation of the average distance is

[0080] ​​​​​S3.4, setting the points with average distance exceeding the distance threshold as outliers;

[0081] S3.5, filtering to remove outliers.

[0082] In the preferred scheme, the wind turbine impeller rotation model in S4 is established by pre-establishing a single-blade three-dimensional information, combining the parameters of the hub, the three-dimensional coordinate system with the laser radar beam emission point as the center, the current blade azimuth angle and rotational speed, and the coordinate position relationship between the hub center and the beam emission point to obtain the three-dimensional rotation model of the wind turbine blade and the hub.

[0083] In the preferred scheme, the pre-established single-blade three-dimensional information in S4 is established by replacing the single-blade with a cylindrical model to establish the three-dimensional model of the blade.

[0084] In S6, the obstacle avoidance logic of the impeller control includes:

[0085] 1) determining whether the type of foreign matter belongs to a type that needs to be avoided;

[0086] 2) determining whether the current blade rotational speed exceeds the minimum rotational speed that needs to be considered for obstacle avoidance;

[0087] 3) when conditions 1 and 2 are both met, determining whether the foreign matter will collide with the blade by combining the distance between the current foreign matter and the blade, the speed of the foreign matter, and the current blade rotational speed;

[0088] 4) if the result of 3 is that it will collide, calculating the current wind speed, blade pitch angle, and rotational speed to determine which type of power generation power is greater when the pitch angle is adjusted to increase or decrease, and determining whether the obstacle avoidance of the impeller leaves enough safety distance when the condition of greater power generation power is met, if so, selecting the speed regulation mode with greater power generation power, if not, taking another adjustment mode.

[0089] In S6, when multiple foreign matters are detected, the obstacle avoidance logic of the impeller control includes:

[0090] 1) determining whether the type of foreign matter belongs to a type that needs to be avoided;

[0091] 2) determining whether the current blade rotational speed exceeds the minimum rotational speed that needs to be considered for obstacle avoidance;

[0092] 3) When conditions 1 and 2 are both met, the number of current foreign matter types is combined with the current rotating speed to determine whether it will cause uncontrollable risk to the paddle, if so, the impeller is adjusted to operate along the stream and slow down; if not, the distance relationship between multiple foreign matters is determined, when the distance between two foreign matters is less than the set bonding distance, the two foreign matters are considered as a whole, the rotating speed of the impeller when each foreign matter just passes between the two impellers is determined according to the number of current foreign matters and their speed, and the adjustment mode of the impeller is calculated according to the current rotating speed and the calculated rotating speed of the impeller when each foreign matter just passes between the two impellers.

[0093] As shown in Figure 2 , the laser radar and visual sensor on the wind turbine described above are each provided with a set on the tower and the nacelle.

[0094] In the above S5, the target detection network is Retina Net network, and the cross-entropy function uses Focal Loss function for identification and classification:

[0095] Binary classification cross-entropy loss function:

[0096] ;

[0097] Where , is the class label marked by artificial labeling, is the predicted probability of ;

[0098] Let:

[0099] ;

[0100] Then:

[0101] = ;

[0102] The ordinary cross-entropy loss function also has a large loss for easy-to-classify samples, and is not suitable for sample category unbalanced data sets, so in order to solve this problem, a weighting factor [0,1] is introduced in front of the loss function, and the balanced cross-entropy is constructed as follows:

[0103] = ;

[0104] But when the sample categories are extremely unbalanced during the training process, the easy-to-classify negative samples occupy most of the loss function, resulting in unstable error function convergence, so Focal Loss increases an adjustable factor to reduce the weight of easy-to-classify negative samples based on the above formula, mainly focusing on difficult-to-classify samples, which is defined as follows:

[0105] = ;

[0106] wherein, is an adjustable focusing parameter.

[0107] After the above classification, the classification recognition result of the foreign matter is obtained. When the foreign matter is an object such as a bird or a branch, a relatively large impact impact on the impeller will occur, which is a situation that must be avoided. When the foreign matter is a leaf, a plastic bag, or a balloon, the influence caused by the foreign matter is different according to the number of foreign matters and the rotating speed of the impeller, and therefore, the adjustment and control thereof need to be considered differently.

Claims

1. A wind turbine blade active obstacle avoidance method, characterized in that, The method comprises the following steps: S1, sampling point cloud and taking image of the front of the wind turbine impeller by laser radar; S2, filtering the point cloud; S3, removing outliers in the point cloud data processed in S2; S4, establishing a three-dimensional coordinate system with the light beam emitting point of the laser radar as the center, segmenting and clustering the laser radar point cloud data, establishing a rotating model of the wind turbine impeller, collecting the blade azimuth angle and the rotating speed in the SCADA system of the wind turbine, establishing the current blade azimuth impeller model, fitting the segmented and clustered laser radar point cloud data, obtaining the impeller point cloud data in the laser radar point cloud data, removing the impeller point cloud data from the segmented and clustered laser radar point cloud data to obtain the segmented and clustered point cloud data of the front of the wind turbine impeller without the impeller, and obtaining the coordinate position and range of the foreign matter in the three-dimensional coordinate system; S5, controlling the multi-dimensional rotating mechanism of the visual sensor to take a visual image in the direction of the foreign matter, inputting the visual image into a target detection network to output an anchor box, and identifying and classifying the anchor box image by using a cross-entropy function; and S6, determining the target type of the foreign matter, judging whether the foreign matter will collide with the blade according to the coordinate position and range of the foreign matter, the blade azimuth angle and the current rotating speed in the SCADA system, and controlling the rotating speed of the impeller according to the obstacle avoidance logic. The point cloud sampling in S1 is downsampling, and the filtering in S2 is voxel filtering. The outliers in the point cloud data in S3 are removed by using a distance-based K-nearest neighbor algorithm. The distance-based K-nearest neighbor algorithm in S3 comprises the following steps: S3.1, setting a parameter K for selecting the number of neighborhood points; Mean: Standard deviation:

2. A wind turbine blade active obstacle avoidance method as claimed in claim 1, wherein, S3.4, setting a point as an outlier if the average distance of the point exceeds a distance threshold; 3. A wind turbine blade active obstacle avoidance method as claimed in claim 2, wherein, S3.5, filtering and removing the outliers.

4. A method of active obstacle avoidance for a wind turbine blade according to claim 3, characterised in that, The rotating model of the wind turbine impeller in S4 is established by pre-establishing three-dimensional information of a single blade, combining the parameters of the hub, the three-dimensional coordinate system with the light beam emitting point of the laser radar as the center, the current blade azimuth angle and rotating speed, and the coordinate position relationship between the hub center and the light beam emitting point, and calculating a three-dimensional rotating model of the wind turbine blade and the hub. The three-dimensional information of the single blade in S4 is established by replacing the single blade with a cylindrical model to establish a three-dimensional model of the blade. S3.2, compute the average distance of each point to the K nearest neighbors : Suppose the point cloud data is in the first Coordinates of a 3D point Then, from that point to any three-dimensional point... distance for: ； the average distance to the K nearest neighbors : ； S3.3, calculate the mean and standard deviation of the average distances, and then calculate the distance threshold T : The obstacle avoidance logic for the impeller control in S6 comprises the following steps: ； 1) judging whether the type of the foreign matter belongs to a type that needs to be avoided; ； distance threshold T : ； wherein is the mean of all average distances, is the set number of standard deviations, which can vary the size of the distance threshold according to the needs, is the standard deviation of the average distances; 2) judging whether the current rotating speed of the blade exceeds a minimum rotating speed that needs to be avoided; 3) when conditions 1 and 2 are both met, judging whether the foreign matter will collide with the blade according to the distance between the foreign matter and the blade, the speed of the foreign matter, and the current rotating speed of the blade.

5. A method of active obstacle avoidance for a wind turbine blade as claimed in claim 4, wherein, ​ 6. A wind turbine blade active obstacle avoidance method according to claim 5, wherein, ​ 7. A wind turbine blade active obstacle avoidance method as claimed in claim 5, wherein, ​ ​ ​ ​ 4) If the result of 3 is that the blades will collide with the foreign matter, the current wind speed, pitch angle and rotational speed are calculated to determine whether the rotational speed of the blades will be increased or decreased, which one will result in a larger power generation, and whether the impeller has enough safety distance when the power generation is larger. If yes, the speed adjustment mode with larger power generation is selected, and if not, another adjustment mode is adopted.

8. A wind turbine blade active obstacle avoidance method as claimed in claim 5, wherein, The obstacle avoidance logic of the impeller when multiple foreign matters are detected in S6 includes: 1) determining whether the type of the foreign matter belongs to the type that needs to be avoided; 2) determining whether the current rotational speed of the blades exceeds the minimum speed that needs to be considered for obstacle avoidance; 3) when conditions 1 and 2 are both met, the number of the current foreign matter type is combined with the current rotational speed to determine whether it will cause uncontrollable risk to the blades, if yes, the impeller is controlled to operate at a speed lower than the current speed; if not, the distance relationship between multiple foreign matters is determined, when the distance between two foreign matters is less than the set adhesion distance, the two foreign matters are considered as a whole, the rotational speed of the impeller when each foreign matter just passes between the two impellers is determined according to the number and speed of the current foreign matter, and the adjustment mode of the impeller is calculated according to the current rotational speed and the calculated rotational speed of the impeller when each foreign matter just passes between the two impellers.

9. A wind turbine blade active obstacle avoidance method as claimed in claim 1, wherein, The laser radar and visual sensor on the wind turbine generator set are arranged on the tower and the nacelle respectively.

Images