Active obstacle avoidance method for wind turbine blades

Abstract

The invention provides an active obstacle avoidance method for a wind turbine blade, which comprises the following steps: installing a laser radar and a visual sensor on a wind turbine generator, and comprises the following active obstacle avoidance steps: S1, carrying out point cloud sampling and image shooting of the laser radar in front of a fan impeller; s2, filtering point cloud sampling of the laser radar; s3, outliers in the point cloud sampling data processed in the step S2 are removed; s4, the coordinate position and range of the front foreign matter in the three-dimensional coordinate system are obtained, and the coordinate position is sent to a control mechanism of a multi-dimensional rotating mechanism of the visual sensor; s5, identifying and classifying the anchoring frame image by using a cross entropy function; and S6, according to the coordinate position and range of the foreign matter, the impeller is subjected to speed regulation control. The speed of the impeller is adjusted through active control of the wind turbine, and the impeller is prevented from being collided by objects; data information of active obstacle avoidance each time is collected, data are trained at the offline end, and the obstacle avoidance action of the wind turbine is optimized.

Description

Technical Field

[0001] The present invention relates to the field of wind power technology, and in particular to an active obstacle avoidance method for wind turbine blades. Background Art

[0002] As the global demand for renewable energy continues to grow, wind power generation, as a clean and renewable form of energy, is becoming increasingly important. The development of wind power generation technology is not only related to the optimization of the energy structure, but also one of the key ways to achieve the goal of carbon neutrality. As the core component of a wind turbine generator set, the performance of wind turbine blades directly determines the efficiency and reliability of wind power generation. However, in actual operation, wind turbine blades face a variety of complex environmental challenges, one of which is how to effectively avoid collisions with obstacles ahead.

[0003] Traditional wind turbine blade obstacle avoidance methods mainly rely on passive protection measures, such as setting up protective nets and increasing blade strength. Although these methods can reduce the risk of blade damage to a certain extent, they cannot fundamentally solve the problem of collision between blades and obstacles. Especially in complex and changeable wind farm environments, such as mountainous areas and urban suburbs, wind turbine blades are more vulnerable to obstacles such as flying birds, trees, and buildings.

[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 acquiring image information in front of the wind turbine impeller in real time and combining it with advanced image recognition technology, it is possible to quickly and accurately identify objects in front. Furthermore, by building a weight library based on a pre-loaded object database, the avoidance priority of different object types can be determined, thereby making more intelligent obstacle avoidance decisions.

[0005] However, the existing active obstacle avoidance technology still faces many challenges. On the one hand, how to achieve accurate obstacle avoidance of blades while ensuring the power generation efficiency of wind turbines is an urgent problem 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 and improve the accuracy and reliability of obstacle avoidance is also the focus of current research.

[0006] Therefore, the present invention proposes a method for active obstacle avoidance of wind turbine blades, which acquires image information in front of the impeller, combines image recognition technology and object database, and constructs a complete active obstacle avoidance system. The system can not only judge the type and avoidance priority of the object in front in real time, but also predict the movement trajectory of the two according to the impeller position information of the wind turbine main engine and the position of the obstacle, so as to determine the point that needs to be avoided. By actively controlling the wind turbine to adjust the speed of the impeller, it is possible to effectively avoid objects from hitting the impeller and ensure the safe operation of the wind turbine. At the same time, the method also collects data information of each active obstacle avoidance, trains and optimizes the data at the offline end, and continuously improves the accuracy and efficiency of the wind turbine's obstacle avoidance action. The introduction of this innovative technology is expected to provide new ideas and solutions for active obstacle avoidance technology in the field of wind power generation. Summary of the invention

[0007] The present invention provides a method for actively avoiding obstacles of wind turbine blades, and performs obstacle avoidance control on foreign objects according to the detection of wind turbine generator sets.

[0008] In order to solve the above technical problems, the technical solution adopted by the present invention is: A method for actively avoiding obstacles for wind turbine blades includes installing a laser radar and a visual sensor on a wind turbine set to detect the front of the wind turbine impeller. The visual sensor is provided with a multi-dimensional rotating mechanism to aim at foreign objects. The steps of actively avoiding obstacles are as follows: S1. Perform laser radar point cloud sampling and image acquisition in front of the fan impeller; S2, filtering the point cloud sampling of the laser radar; S3, removing outliers from the point cloud sampling data processed in S2; S4. Establish a three-dimensional coordinate system with the laser radar beam emission point as the center, segment and cluster the laser radar point cloud data, establish a rotation model of the wind turbine impeller, collect the blade azimuth and rotation speed in the wind turbine SCADA system, and establish an impeller model of the current blade azimuth and fit it with the laser radar point cloud data after segmentation and clustering, so as to obtain the impeller point cloud data in the laser radar point cloud data, remove the impeller point cloud data from the laser radar point cloud data after segmentation and clustering, and obtain the segmented clustered point cloud data in front of the wind turbine impeller after the impeller is removed, and obtain the coordinate position and range of the foreign object in the front in the three-dimensional coordinate system, and send this coordinate position to the control mechanism of the multi-dimensional rotation mechanism of the visual sensor; S5, the control mechanism of the visual sensor controls the multi-dimensional rotating mechanism to align the foreign object to capture the visual image, inputs the captured image into the target detection network to output the anchor frame, and uses the cross entropy function to identify and classify the anchor frame image; S6. After the target type of the foreign object is determined, the coordinate position and range of the foreign object, combined with the blade azimuth and current speed in the SCADA system, determine whether it will collide with the blade, and control the impeller speed in combination with the obstacle avoidance logic.

[0009] The point cloud sampling used in the above S1 is downsampling, and the filtering used in S2 is voxel filtering.

[0010] The outlier removal of the point cloud sampling data in the above S3 adopts the distance-based K nearest neighbor algorithm.

[0011] The specific process of the distance-based K nearest neighbor algorithm in S3 above is: S3.1. Set the parameter K of the number of selected neighborhood points; S3.2. Calculate the average distance from each point to K neighboring points : Set the point cloud data Coordinates of 3D points , then the point to any three-dimensional point Distance for: ; Then the average distance to the K nearest neighbors is : ; S3.3. Calculate the mean and standard deviation of the average distance, and then calculate the distance threshold T : Mean: ; Standard Deviation: ; Distance Threshold T : ; in is the mean of all average distances, The distance threshold can be changed according to the requirements. is the standard deviation of the mean distance; S3.4, setting points whose average distance exceeds the distance threshold as outliers; S3.5. Filter to remove outliers.

[0012] In the preferred scheme, the establishment of the wind turbine impeller rotation model in the above-mentioned S4 adopts the pre-established three-dimensional information of a single blade, and then combines the parameters of the hub, the three-dimensional coordinate system centered on the laser radar beam emission point, the current blade azimuth and rotation speed, and the coordinate position relationship between the hub center and the beam emission point to calculate and obtain the three-dimensional rotation model of the wind turbine blade and hub.

[0013] In a preferred solution, the pre-establishment of the three-dimensional information of a single blade in the above S4 is performed by replacing the single blade with a cylindrical model to establish a three-dimensional model of the blade.

[0014] In the above-mentioned S6, the obstacle avoidance logic of the impeller control includes: 1) Determine whether the type of foreign object is the type that requires obstacle avoidance; 2) Whether the current propeller speed exceeds the minimum speed required for obstacle avoidance; 3) When conditions 1 and 2 are both met, the current distance between the foreign object and the propeller blade, the speed of the foreign object, and the current rotation speed of the propeller blade are combined to determine whether the foreign object will hit the propeller blade; 4) If the result of judgment 3 is that there will be a collision, calculate the current wind speed, blade pitch angle and rotation speed, and determine which of the following conditions will result in a greater power generation: the speed of the blades will increase or decrease when the pitch angle is adjusted to increase or decrease, and whether there is enough safe distance for the impeller to avoid obstacles when judging the condition for a greater power generation. If so, choose the speed regulation method with a greater power generation; if not, adopt another regulation method.

[0015] In the above S6, when multiple foreign objects are detected, the obstacle avoidance logic of the impeller control includes: 1) Determine whether the type of foreign object is the type that requires obstacle avoidance; 2) Determine whether the current propeller speed exceeds the minimum speed required for obstacle avoidance; 3) When conditions 1 and 2 are both met, the current number of foreign matter types is combined with the current speed to determine whether it will cause uncontrollable risks to the blades. If so, the impeller is regulated to perform a speed reduction operation along the slurry. If not, the distance relationship between multiple foreign matter is determined. When the distance between two foreign matter is less than the set bonding distance, the two foreign matter are considered as a whole. The speed of the impeller when each foreign matter just passes through the two impellers is determined based on the current number of foreign matter and its speed. The adjustment method of the impeller is calculated based on the current speed and the calculated speed of the impeller when the foreign matter just passes through the two impellers.

[0016] The laser radar and visual sensor on the above-mentioned wind turbine are respectively arranged in one group on the tower and one group on the cabin.

[0017] The present invention provides a method for active obstacle avoidance of wind turbine blades. A point cloud model is established in front of the wind turbine tower by using a laser radar on the wind turbine tower. A three-dimensional model of the impeller is established and fitted with the point cloud data. The data of the impeller is removed to obtain the point cloud data in front of the unit. The type of foreign matter is classified by combining with visual detection. Then, speed control is performed according to the current blade pitch angle and rotation speed. Active detection is used to avoid damage to the blades by foreign matter. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The present invention will be further described below in conjunction with the accompanying drawings and embodiments: Figure 1 is a hardware schematic diagram of the method; Figure 2 yes Figure 1 A partial enlarged schematic diagram of .

[0019] Among them: LiDAR and visual sensor (1. DETAILED DESCRIPTION

[0020] In order to make the purpose, technical solutions and advantages of the present invention clearer, the following content will systematically and completely describe the specific technical solutions of the present invention in combination with the drawings provided according to the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0021] Embodiment 1: like Figure 1 As shown in , a method for active obstacle avoidance of wind turbine blades includes installing a laser radar and a visual sensor on a wind turbine set to detect the front of the wind turbine impeller. The visual sensor is provided with a multi-dimensional rotating mechanism for aligning with foreign objects. The steps of active obstacle avoidance are: S1. Perform laser radar point cloud sampling and image acquisition in front of the fan impeller; S2, filtering the point cloud sampling of the laser radar, while reducing the number of three-dimensional points in the cloud data, maintaining the shape characteristics of the point cloud without destroying the characteristics of the point cloud; S3, removing outliers from the point cloud sampling data processed in S2; S4. Establish a three-dimensional coordinate system with the laser radar beam emission point as the center, segment and cluster the laser radar point cloud data, establish a rotation model of the wind turbine impeller, collect the blade azimuth and rotation speed in the wind turbine SCADA system, and establish an impeller model of the current blade azimuth and fit it with the laser radar point cloud data after segmentation and clustering, so as to obtain the impeller point cloud data in the laser radar point cloud data, remove the impeller point cloud data from the laser radar point cloud data after segmentation and clustering, and obtain the segmented clustered point cloud data in front of the wind turbine impeller after the impeller is removed, and obtain the coordinate position and range of the foreign object in the front in the three-dimensional coordinate system, and send this coordinate position to the control mechanism of the multi-dimensional rotation mechanism of the visual sensor; S5, the control mechanism of the visual sensor controls the multi-dimensional rotating mechanism to align the foreign object to capture the visual image, inputs the captured image into the target detection network to output the anchor frame, and uses the cross entropy function to identify and classify the anchor frame image; S6. After the target type of the foreign object is determined, the coordinate position and range of the foreign object, combined with the blade azimuth and current speed in the SCADA system, determine whether it will collide with the blade, and control the impeller speed in combination with the obstacle avoidance logic.

[0022] The laser radar samples the area in front of the impeller. Since the laser radar outputs the point cloud information of the coordinate position and range of the foreign object, the recognition and analysis of the target of this three-dimensional information requires a lot of calculations and model calculations, which is computationally intensive and difficult. Therefore, using the image information with shape and color in the visual sensor to identify and classify foreign objects can greatly reduce the difficulty and computational complexity of identification, improve the accuracy of identification, and determine the current and future position information of the blade based on the azimuth and rotational position of the blade by establishing a blade model. By combining the current rotation speed and the speed and direction of the foreign object, it is determined whether a collision accident will occur, and then different response strategies are adopted according to the type of foreign object, and the blade speed is adjusted if necessary.

[0023] The point cloud sampling used in the above S1 is downsampling, and the filtering used in S2 is voxel filtering.

[0024] Point cloud downsampling can improve processing efficiency by reducing the number of points and reducing data complexity. Voxel filtering downsampling creates a voxel grid, puts the entire point cloud data into the voxel grid, and then collects all the 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.

[0025] The outlier removal of the point cloud sampling data in the above S3 adopts the distance-based K nearest neighbor algorithm.

[0026] The specific process of the distance-based K nearest neighbor algorithm in S3 above is: S3.1. Set the parameter K of the number of selected neighborhood points; S3.2. Calculate the average distance from each point to K neighboring points : Set the point cloud data Coordinates of 3D points , then the point to any three-dimensional point Distance for: ; Then the average distance to the K nearest neighbors is : ; S3.3. Calculate the mean and standard deviation of the average distance, and then calculate the distance threshold T : Mean: ; Standard Deviation: ; Distance Threshold T : ; in is the mean of all average distances, The distance threshold can be changed according to the requirements. is the standard deviation of the mean distance; S3.4, setting points whose average distance exceeds the distance threshold as outliers; S3.5. Filter to remove outliers.

[0027] In the preferred scheme, the establishment of the wind turbine impeller rotation model in the above-mentioned S4 adopts the pre-established three-dimensional information of a single blade, and then combines the parameters of the hub, the three-dimensional coordinate system centered on the laser radar beam emission point, the current blade azimuth and rotation speed, and the coordinate position relationship between the hub center and the beam emission point to calculate and obtain the three-dimensional rotation model of the wind turbine blade and hub.

[0028] In a preferred solution, the pre-establishment of the three-dimensional information of a single blade in the above S4 is performed by replacing the single blade with a cylindrical model to establish a three-dimensional model of the blade.

[0029] In the above-mentioned S6, the obstacle avoidance logic of the impeller control includes: 1) Determine whether the type of foreign object is the type that requires obstacle avoidance; 2) Whether the current propeller speed exceeds the minimum speed required for obstacle avoidance; 3) When conditions 1 and 2 are both met, the current distance between the foreign object and the propeller blade, the speed of the foreign object, and the current rotation speed of the propeller blade are combined to determine whether the foreign object will hit the propeller blade; 4) If the result of judgment 3 is that there will be a collision, calculate the current wind speed, blade pitch angle and rotation speed, and determine which of the following conditions will result in a greater power generation: the speed of the blades will increase or decrease when the pitch angle is adjusted to increase or decrease, and whether there is enough safe distance for the impeller to avoid obstacles when judging the condition for a greater power generation. If so, choose the speed regulation method with a greater power generation; if not, adopt another regulation method.

[0030] In the above S6, when multiple foreign objects are detected, the obstacle avoidance logic of the impeller control includes: 1) Determine whether the type of foreign object is the type that requires obstacle avoidance; 2) Determine whether the current propeller speed exceeds the minimum speed required for obstacle avoidance; 3) When conditions 1 and 2 are both met, the current number of foreign matter types is combined with the current speed to determine whether it will cause uncontrollable risks to the blades. If so, the impeller is regulated to perform a speed reduction operation along the slurry. If not, the distance relationship between multiple foreign matter is determined. When the distance between two foreign matter is less than the set bonding distance, the two foreign matter are considered as a whole. The speed of the impeller when each foreign matter just passes through the two impellers is determined based on the current number of foreign matter and its speed. The adjustment method of the impeller is calculated based on the current speed and the calculated speed of the impeller when the foreign matter just passes through the two impellers.

[0031] like Figure 2 As shown in , the laser radar and visual sensor on the above-mentioned wind turbine set are respectively arranged in one group on the tower and one group on the cabin.

[0032] The target detection network in S5 above is the Retina Net network, and the cross entropy function uses the Focal Loss function for recognition and classification: Binary cross entropy loss function: ; in , is the manually annotated category label, Yes The predicted probability of set up: ; but: = ; The ordinary cross entropy loss function also has a large loss for easy-to-classify samples and is not suitable for datasets with unbalanced sample categories. Therefore, to solve this problem, a weighting factor is introduced before the loss function. [0,1], the balanced cross entropy is constructed as follows: = ; However, when the sample categories are extremely unbalanced during training, easy-to-classify negative samples occupy most of the loss function, resulting in unstable convergence of the error function. Therefore, Focal Loss adds an adjustable factor to the above formula to reduce the weight of easy-to-classify negative samples, mainly focusing on difficult-to-classify samples. Its definition is as follows: = ; in, is an adjustable focus parameter.

[0033] After the above classification, the classification and identification results of foreign objects are obtained. When the foreign objects are birds, branches and other objects, they will cause a relatively large impact on the impeller, which must be avoided. When the foreign objects are leaves, plastic bags, balloons, etc., the impact will be different depending on the number of foreign objects and the speed of the impeller, so it is necessary to consider the adjustment and control differently.

Claims

1. A method for active obstacle avoidance of wind turbine blades, characterized in that: It includes installing a laser radar and a visual sensor on the wind turbine to detect the front of the wind turbine impeller. The visual sensor is equipped with a multi-dimensional rotating mechanism to aim at foreign objects. The steps of active obstacle avoidance are as follows: S1. Perform laser radar point cloud sampling and image acquisition in front of the fan impeller; S2, filtering the point cloud sampling of the laser radar; S3, removing outliers from the point cloud sampling data processed in S2; S4. Establish a three-dimensional coordinate system with the laser radar beam emission point as the center, segment and cluster the laser radar point cloud data, establish a rotation model of the wind turbine impeller, collect the blade azimuth and rotation speed in the wind turbine SCADA system, and establish an impeller model of the current blade azimuth and fit it with the laser radar point cloud data after segmentation and clustering, so as to obtain the impeller point cloud data in the laser radar point cloud data, remove the impeller point cloud data from the laser radar point cloud data after segmentation and clustering, and obtain the segmented clustered point cloud data in front of the wind turbine impeller after the impeller is removed, and obtain the coordinate position and range of the foreign object in the front in the three-dimensional coordinate system, and send this coordinate position to the control mechanism of the multi-dimensional rotation mechanism of the visual sensor; S5, the control mechanism of the visual sensor controls the multi-dimensional rotating mechanism to align the foreign object to capture the visual image, inputs the captured image into the target detection network to output the anchor frame, and uses the cross entropy function to identify and classify the anchor frame image; S6. After the target type of the foreign object is determined, the coordinate position and range of the foreign object are combined with the blade azimuth and current speed in the SCADA system to determine whether it will collide with the blade, and the impeller speed is controlled in combination with the obstacle avoidance logic.

2. The active obstacle avoidance method for wind turbine blades according to claim 1, characterized in that: The point cloud sampling used in S1 is downsampling, and the filtering used in S2 is voxel filtering.

3. The active obstacle avoidance method for wind turbine blades according to claim 2, characterized in that: The outlier removal of the point cloud sampling data in S3 adopts the distance-based K nearest neighbor algorithm.

4. The active obstacle avoidance method for wind turbine blades according to claim 3, characterized in that: The specific process of the distance-based K nearest neighbor algorithm in S3 is as follows: S3.

1. Set the parameter K of the number of selected neighborhood points; S3.

2. Calculate the average distance from each point to K neighboring points : Set the point cloud data Coordinates of 3D points , then the point to any three-dimensional point Distance for: ； Then the average distance to the K nearest neighbors is : ； S3.

3. Calculate the mean and standard deviation of the average distance, and then calculate the distance threshold T : Mean: ； Standard Deviation: ； Distance Threshold T : ； in is the mean of all average distances, The distance threshold can be changed according to the requirements. is the standard deviation of the mean distance; S3.4, setting points whose average distance exceeds the distance threshold as outliers; S3.

5. Filter to remove outliers.

5. The active obstacle avoidance method for wind turbine blades according to claim 4, characterized in that: The establishment of the wind turbine impeller rotation model in S4 adopts the pre-established three-dimensional information of a single blade, and then combines the parameters of the hub, the three-dimensional coordinate system centered on the laser radar beam emission point, the current blade azimuth and rotation speed, and the coordinate position relationship between the hub center and the beam emission point to calculate and obtain the three-dimensional rotation model of the wind turbine blade and hub.

6. The active obstacle avoidance method for wind turbine blades according to claim 5, characterized in that: The pre-establishment of the three-dimensional information of a single blade in S4 is carried out by replacing the single blade with a cylindrical model to establish a three-dimensional model of the blade.

7. The active obstacle avoidance method for wind turbine blades according to claim 5, characterized in that: In the above-mentioned S6, the obstacle avoidance logic of the impeller control includes: 1) Determine whether the type of foreign object is the type that requires obstacle avoidance; 2) Whether the current propeller speed exceeds the minimum speed required for obstacle avoidance; 3) When conditions 1 and 2 are both met, the current distance between the foreign object and the propeller blade, the speed of the foreign object, and the current rotation speed of the propeller blade are combined to determine whether the foreign object will hit the propeller blade; 4) If the result of judgment 3 is that there will be a collision, calculate the current wind speed, blade pitch angle and rotation speed, and determine which of the following conditions will result in a greater power generation: the speed of the blades will increase or decrease when the pitch angle is adjusted to increase or decrease, and whether there is enough safe distance for the impeller to avoid obstacles when judging the condition for a greater power generation. If so, choose the speed regulation method with a greater power generation; if not, adopt another regulation method.

8. The method for active obstacle avoidance of wind turbine blades according to claim 5, characterized in that: In the above S6, when multiple foreign objects are detected, the obstacle avoidance logic of the impeller control includes: 1) Determine whether the type of foreign object is the type that requires obstacle avoidance; 2) Determine whether the current propeller speed exceeds the minimum speed required for obstacle avoidance; 3) When conditions 1 and 2 are both met, the current number of foreign matter types is combined with the current speed to determine whether it will cause uncontrollable risks to the blades. If so, the impeller is regulated to perform a speed reduction operation along the slurry. If not, the distance relationship between multiple foreign matter is determined. When the distance between two foreign matter is less than the set bonding distance, the two foreign matter are considered as a whole. The speed of the impeller when each foreign matter just passes through the two impellers is determined based on the current number of foreign matter and its speed. The adjustment method of the impeller is calculated based on the current speed and the calculated speed of the impeller when the foreign matter just passes through the two impellers.

9. The active obstacle avoidance method for wind turbine blades according to claim 1, characterized in that: The laser radar and visual sensor on the wind turbine are arranged in one group on the tower and one group on the cabin.

Images