Anomaly data processing method for a wind farm

Abstract

The application provides an abnormal data processing method for a wind power plant, and relates to a data processing technology, and the method comprises the following steps: in response to the completion of first collection of the first rotating side of a wind driven generator by a UAV, determining a passing time window corresponding to a gap space formed by adjacent rotating blades based on the real-time rotating speed of the corresponding wind driven generator, and determining an estimated passing duration of the UAV passing through the gap space; based on the matching relationship between the passing time window and the estimated passing duration, controlling the UAV to perform passing flight or circling flight of the corresponding wind driven generator, so that the UAV reaches a second position on the second rotating side of the wind driven generator, which has a symmetrical relationship with the first position; controlling the UAV to perform second collection of the second rotating side of the wind driven generator based on the second position, and determining the detection attribute of the corresponding wind driven generator based on the first image and the second image obtained based on the first collection and the second collection. The application at least improves the detection efficiency.

Description

Technical Field

[0001] This invention relates to data processing technology, and more particularly to a method for processing abnormal data in wind farms. Background Technology

[0002] With the rapid development of renewable energy, wind power has become one of the main forms of clean energy. To ensure the safe and stable operation of wind turbine generators, it is necessary to regularly conduct health checks on their huge rotating blades.

[0003] Currently, in routine drone inspection operations, in order to comprehensively acquire the health status of the blades and make comprehensive anomaly judgments, drones usually need to reach both the windward and leeward sides of the wind turbine to perform all-round image acquisition. However, in the actual flight process across the rotor plane, existing drones often make long-distance detours along the safe range of the blade tip when performing all-round image acquisition across the rotor plane, which greatly increases the one-way detour distance and the total flight time, resulting in low on-site inspection efficiency.

[0004] Therefore, there is an urgent need to provide a method for processing abnormal data in wind farms that can improve detection efficiency. Summary of the Invention

[0005] In view of the above problems, the present invention is proposed to provide a method for processing abnormal data in wind farms that overcomes or at least partially solves the above problems.

[0006] According to one aspect of the present invention, a method for processing abnormal data in wind farms is provided, comprising: The drone completes the first data acquisition on the first rotating side of the wind turbine based on the first position, and obtains the real-time rotation speed of the corresponding wind turbine. Based on the real-time rotational speed, the crossing time window corresponding to the gap space formed by the adjacent rotating blades of the wind turbine is determined, and the estimated crossing time of the UAV crossing through the gap space is determined. Based on the matching relationship between the crossing time window and the estimated crossing time, the UAV is controlled to perform a crossing flight or a circling flight around the corresponding wind turbine, so that the UAV reaches a second position located on the second rotating side of the wind turbine and symmetrical to the first position. The drone is controlled to perform a second acquisition on the second rotating side of the wind turbine based on the second position, and the detection attributes of the corresponding wind turbine are determined based on the first acquisition, the first acquisition, and the second acquisition of the first image and the second image.

[0007] Optionally, in the method according to the present invention, obtaining the real-time rotational speed of the rotating blades of the wind turbine includes: Based on the first image obtained from the first acquisition, determine the blade regions in the first image that indicate each rotating blade; The first image is processed into coordinates to determine the coordinate points of each blade that makes up each blade region. The distances of each blade coordinate point relative to the rotation center point of the wind turbine are obtained, and the blade coordinate point with the largest corresponding distance is determined as the marker coordinate point. The response determines that there are at least two positively aligned marker coordinate points with the same pixel value based on the rotation direction of the rotating blade, and identifies the pixel value as the marker pixel value; otherwise, the marker pixel value is determined based on a preset pixel matching strategy. Control the drone to hover, determine the image acquisition frame pre-set in the image acquisition unit of the drone, and determine the time interval when the identifier pixel value appears continuously in the same image position; The number of blades corresponding to all rotating blades of the wind turbine is obtained, and the real-time rotational speed is determined based on the ratio of the rotational distance determined by the number of blades and the distance between the points corresponding to the marked coordinates, and the time interval.

[0008] Optionally, in the method according to the present invention, determining the identifier pixel value based on a preset pixel matching strategy includes: The coordinate points corresponding to different blade regions are determined as primary marker points, and marker connection lines connecting the primary marker points to the rotation center point are generated. The blade coordinate point located at the center of the label connection line is determined as the slave label point, and the pixel values ​​corresponding to the master label point and the slave label point are combined into the label pixel value.

[0009] Optionally, in the method according to the present invention, determining the crossing time window corresponding to the gap space formed by adjacent rotating blades of the wind turbine based on the real-time rotational speed, and determining the estimated crossing time of the UAV crossing through the gap space, includes: Using the rotation center point as the center, generate a rotation profile that passes through each marked coordinate point, and generate a marker connecting line that connects the marked coordinate points to the rotation center point; Based on the first image, a gap dividing line is formed starting from the rotation center point to divide adjacent marker connecting lines and connect the rotation contour; The gap dividing lines are arrayed, and with the rotation center point as the center, arc line segments are generated that connect adjacent blade regions and pass through each array point. Determine the chord length of each arc segment, and in response to any chord length exceeding the device size of the corresponding UAV, determine the array point corresponding to that chord length as a valid point; The crossing time window for each valid point is determined based on the real-time rotation speed and the arc length of each corresponding arc segment, and the estimated crossing time is determined based on the equipment distance of the UAV to the wind turbine.

[0010] Optionally, in the method according to the present invention, determining the estimated travel time based on the equipment distance of the UAV to the corresponding wind turbine includes: A first distance is determined along the gap dividing line corresponding to the rotation center point of each effective point, and a first wind disturbance coefficient is determined based on the first distance for the corresponding tower dimension; A second distance is determined along the gap dividing line corresponding to the rotation profile of each effective point, and a second wind disturbance coefficient is determined based on the second distance for the corresponding tip dimension; The first and second wind disturbance coefficients corresponding to the same effective point are weighted and summed to obtain the comprehensive wind disturbance coefficients. The initial flight time is determined based on the distance and speed of the drone to the corresponding wind turbine. The estimated crossing time for each effective point is obtained by multiplying the initial flight time with each comprehensive wind disturbance coefficient.

[0011] Optionally, in the method according to the present invention, controlling the UAV to perform a crossing flight or a circumnavigation flight around the corresponding wind turbine based on the matching relationship between the crossing time window and the estimated crossing duration includes: The crossing time window corresponding to the same valid point is compared with the estimated crossing time. If the crossing time window is greater than the estimated crossing time, the valid point is determined as the crossing point. The second distance corresponding to each crossing point is summed with the preset redundancy distance. The estimated detour time is determined based on the obtained one-way detour distance. The estimated crossing time corresponding to the same crossing point is compared with the estimated detour time. If the estimated crossing time for any crossing point is less than the estimated detour time, the preparation time for the UAV to fly from the first position to the crossing point is determined based on the device speed. The preparation time and estimated crossing time for the same crossing point are summed, and the UAV is controlled to perform crossing flight to the corresponding wind turbine based on the crossing point with the minimum total flight time; or, If the estimated crossing time for all crossing points is not less than the estimated detour time, control the UAV to perform detour flight around the corresponding wind turbine based on the first position.

[0012] Optionally, in the method according to the invention, determining the preparation time for the UAV to fly from the first position to the crossing point based on the device speed includes: Obtain the acquisition center point of the corresponding image acquisition frame, and generate a center connection line that passes through the acquisition center point and connects the rotation center point and the rotation contour; The response determines the crossing point of any rotating blade on the central connecting line based on the image acquisition frame, and determines the position of the crossing point on the central connecting line as the moving position. The drone is controlled to adjust its movement along the rotation plane of the corresponding wind turbine based on the device speed, so that the acquisition center point coincides with the movement position, and the preparation time corresponding to the adjustment movement is obtained.

[0013] Optionally, in the method according to the present invention, determining the detection attributes of the corresponding wind turbine based on the first image and the second image obtained from the first acquisition and the second acquisition includes: Based on the first image and the second image, the first blade region and the second blade region corresponding to each rotating blade are determined, and the first blade region and the second blade region corresponding to the same rotating blade are grouped into the same blade identifier group. The first and second leaf regions located in the same leaf identification group are extracted, and the first and second feature sets of the corresponding first and second leaf regions are compared to obtain the feature deviation value of each detection point. The first and second feature sets include each detection point that is symmetrically distributed along the rotation plane. If any feature deviation value is greater than a preset deviation threshold, the detection point corresponding to that feature deviation value will be identified as an abnormal point. If the characteristic deviation value of any detection point is less than or equal to the preset deviation threshold, the characteristic data of the detection point will be compared with the retrieved historical data. If the obtained historical deviation value is greater than the preset deviation threshold, the detection point will be identified as an abnormal point. The detection attributes of the corresponding wind turbine are determined based on the number and distribution of all corresponding anomaly points.

[0014] Optionally, in the method according to the present invention, determining the detection attributes of the corresponding wind turbine based on the number and distribution of all corresponding anomaly points includes: If the number of all abnormal points is greater than the preset abnormal threshold, the detection attribute of the corresponding wind turbine is determined as an abnormal attribute; otherwise, all abnormal points corresponding to the same rotating blade are aggregated into the same rotating abnormal group. The system identifies the circumscribed circle of all abnormal points located in the same rotational anomaly group. If the area ratio of the rotating blade corresponding to the abnormal circumscribed circle is greater than a preset ratio threshold, the detection attribute of the corresponding wind turbine is determined as an abnormal attribute; otherwise, it is determined as a normal attribute.

[0015] According to another aspect of the present invention, an anomaly data processing system for wind farms is provided, comprising: The rotation speed acquisition module is configured to respond to the first acquisition performed by the UAV on the first rotating side of the wind turbine based on the first position, and acquire the real-time rotation speed of the corresponding wind turbine. The crossing prediction module is configured to determine the crossing time window corresponding to the gap space formed by the adjacent rotating blades of the wind turbine based on the real-time rotation speed, and to determine the estimated crossing time of the UAV crossing through the gap space. The flight execution module is configured to control the UAV to perform cross-flight or around-flight to the corresponding wind turbine based on the matching relationship between the cross-flight time window and the estimated cross-flight duration, so that the UAV can reach a second position located on the second rotating side of the wind turbine and symmetrical to the first position. The detection and determination module is configured to control the UAV to perform a second acquisition on the second rotating side of the wind turbine based on the second position, and to determine the detection attributes of the corresponding wind turbine based on the first acquisition, the first image obtained from the second acquisition, and the second image.

[0016] According to the solution of the present invention, the present invention has the following beneficial effects: First, by acquiring the real-time rotational speed of the wind turbine, the present invention can accurately determine the crossing time window corresponding to the gap space formed by adjacent rotating blades, and calculate the estimated crossing time of the UAV through the gap space. Based on the matching relationship between the two, the present invention can dynamically control the UAV to perform crossing flight or around flight. When conditions permit, the UAV can directly complete the crossing flight across the impeller plane through the blade gap, which greatly shortens the one-way flight distance and the total flight time, and significantly improves the overall operation efficiency of on-site inspection. Secondly, based on the first image captured by the UAV at the first location, the present invention can extract the coordinate points of the identification area of ​​the blade region and determine the time interval of the continuous appearance of the identification pixel value by combining the image acquisition frame, thereby calculating the real-time speed of the wind turbine. There is no need to deploy additional speed acquisition equipment or access the back-end data of the wind turbine unit. This reduces the intermediate links of data transmission and connection from the source, avoids planning deviations caused by transmission delays, and ensures the efficiency and accuracy of the basic detection process. Next, when assessing the feasibility of crossing, the present invention can determine the first wind disturbance coefficient of the effective point corresponding to the tower dimension and the second wind disturbance coefficient of the corresponding tip dimension along the gap dividing line. The comprehensive wind disturbance coefficient is obtained by weighted summation, and then the estimated crossing time is accurately corrected. In addition, the present invention will also compare the estimated total crossing time with the estimated detour time including the preset redundancy distance. Crossing flight is only performed when the crossing time is shorter and within the safety window. Otherwise, detour flight is performed. This avoids blind detour flight in the whole range and saves the complex spatial collision physics simulation calculation, achieving a perfect balance between flight efficiency and operational safety. Finally, after reaching the second position symmetrical to the first position and completing the second acquisition, the present invention summarizes the data from both sides of the same rotating blade (the first blade area and the second blade area) into the same blade identification group for feature extraction. By comparing the detection points that are symmetrically distributed along the rotation plane, the feature deviation value is calculated, and combined with the number and area distribution of abnormal points, the detection attributes of the corresponding wind turbine are output at once. This avoids the inefficiency of processing the front and back sides of the data in the prior art, and eliminates the need for supplementary acquisition and repeated verification due to different benchmarks, thus achieving cost reduction and efficiency improvement in data analysis. Attached Figure Description

[0017] Figure 1 A flowchart of an abnormal data processing method for a wind farm according to an embodiment of the present invention is shown; Figure 2 A schematic diagram of the corresponding rotational profile in this embodiment is shown; Figure 3 A structural block diagram of an anomaly data processing system for a wind farm according to another embodiment of the present invention is shown. Detailed Implementation

[0018] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0019] To address the problems existing in the prior art, the inventors proposed the solution of this invention. One embodiment of this invention provides a method for processing abnormal data in wind farms. This method can be executed in a computing device, wherein the computing device can be understood as a terminal with data processing capabilities, such as a computer.

[0020] Figure 1 A flowchart of an anomaly data processing method for a wind farm according to an embodiment of the present invention is shown, such as... Figure 1As shown, the abnormal data processing method for wind farms proposed in this embodiment begins with step S1, which includes the following: The responding drone completes the first data acquisition on the first rotating side of the wind turbine based on the first position, and obtains the real-time rotation speed of the corresponding wind turbine.

[0021] For example, in this embodiment, the first location for the UAV to conduct wind farm inspection operations can be determined based on the server. It can be explained that the first location is a fixed spatial point where the UAV performs this data collection operation. This point corresponds to the first rotating side of the wind turbine. The first location can be set at a spatial position that maintains a fixed vertical distance from the rotating plane of the wind turbine blades. This location can completely cover the visible range of the first rotating side of all rotating blades of the wind turbine, thereby ensuring the comprehensiveness of subsequent data collection operations. The first location can be set according to needs, and this embodiment does not limit its specific location. Furthermore, the server can control the drone to maintain a stable hovering state within the determined first position in order to perform the first acquisition on the first rotating side of the wind turbine. It should be noted that the first acquisition is the image acquisition of the rotating blades on the first rotating side of the wind turbine by the drone through its own onboard image acquisition unit. During the acquisition process, the drone needs to remain within the first position without unnecessary positional shifts to ensure that the reference of the acquired image is always consistent. After the drone completes the initial data acquisition on the first rotating side of the wind turbine, the server responds to the trigger signal indicating completion of the acquisition. Based on the image data acquired in this initial acquisition, the server calculates the real-time rotational speed of the corresponding wind turbine. It should be noted that the real-time rotational speed of the wind turbine is the real-time rotational speed of its blades. This parameter is the core foundational data for subsequent inspection path planning. Obtaining the real-time rotational speed based on the images acquired on-site eliminates the need for additional access to the wind turbine's backend data or the deployment of dedicated rotational speed acquisition equipment at the wind farm. This reduces intermediate data transmission steps and avoids the adaptation costs associated with data integration between different turbines. Simultaneously, it ensures that the acquired rotational speed data perfectly matches the actual on-site conditions, preventing planning deviations caused by data transmission delays and significantly improving the accuracy and efficiency of subsequent inspection path planning.

[0022] Furthermore, in this embodiment, the aforementioned "obtaining the real-time rotational speed of the wind turbine blades" may further include the following steps: Based on the first image obtained from the first acquisition, determine the blade regions in the first image that indicate each rotating blade; The first image is processed into coordinates to determine the coordinate points of each blade that makes up each blade region. The distances of each blade coordinate point relative to the rotation center point of the wind turbine are obtained, and the blade coordinate point with the largest corresponding distance is determined as the marker coordinate point. The response determines that there are at least two positively aligned marker coordinate points with the same pixel value based on the rotation direction of the rotating blade, and identifies the pixel value as the marker pixel value; otherwise, the marker pixel value is determined based on a preset pixel matching strategy. Control the drone to hover, determine the image acquisition frame pre-set in the image acquisition unit of the drone, and determine the time interval when the identifier pixel value appears continuously in the same image position; The number of blades corresponding to all rotating blades of the wind turbine is obtained, and the real-time rotational speed is determined based on the ratio of the rotational distance determined by the number of blades and the distance between the points corresponding to the marked coordinates, and the time interval.

[0023] For example, in this embodiment, the determination of the real-time rotational speed can be achieved based on the following method steps: First, based on the first image acquired, the server can perform image recognition processing to determine the blade regions in the first image that indicate each rotating blade. It should be noted that the blade region is the complete image region of a single rotating blade of the wind turbine in the first image. For example, if the wind turbine has three rotating blades, three independent blade regions will be identified in the first image. In this embodiment, by dividing the blade region corresponding to each blade, the effective image range for subsequent speed calculation can be determined without performing indiscriminate analysis on the entire image content of the first image, which greatly reduces the computational load of image analysis and improves the overall detection efficiency from the beginning of speed detection. It can be noted that the image recognition processing can be based on existing neural network learning modules or machine learning models. The training process and recognition process of the neural network learning module or machine learning model can be implemented based on existing technologies. Therefore, this embodiment does not make specific limitations in this regard. Next, the server can perform coordinate processing on the first image after recognition to determine the coordinate points of each blade that makes up each blade region, and simultaneously obtain the distance of each blade coordinate point relative to the rotation center point of the wind turbine. Then, the blade coordinate point with the largest corresponding distance is determined as the marker coordinate point. It can be explained that the rotation center point is the corresponding coordinate position of the rotation axis of the wind turbine rotor in the first image, the point distance is the straight-line distance between the blade coordinate point and the rotation center point, and the coordinate point with the largest point distance corresponds to the blade tip position of the rotating blade. Here, by locking the marker coordinate point corresponding to the blade tip, the core marker for speed calculation can be directly determined without having to verify all the coordinate points of the blade region one by one, further reducing the calculation steps and improving the execution efficiency of speed detection. Then, the server can perform arrangement verification of the marker coordinate points corresponding to different blade areas based on the rotation direction of the rotating blade. When it is determined that there are at least two marker coordinate points with the same pixel value in a forward arrangement, the pixel value can be determined as the marker pixel value. When there are no marker coordinate points that meet the requirements, the marker pixel value is determined based on a preset pixel matching strategy. It can be explained that since the forward arrangement is along the rotation direction of the rotating blade, the marker coordinate points of different blades appear in the order of rotation. Therefore, the marker coordinate points with the same pixel value can ensure that the benchmark for subsequent speed detection is completely consistent. There will be no detection deviation due to inconsistent marker features. There is no need to repeatedly adjust the recognition parameters for multiple verifications, which effectively improves the first pass rate and detection efficiency of speed detection. At the same time, it can also ensure the accuracy of the detection results and avoid repeated detection due to detection deviation. Then, the server can control the drone to hover, determine the image acquisition frame corresponding to the image acquisition unit of the drone, synchronously track the appearance position of the identifier pixel value in the image acquisition frame, and determine the time interval of the identifier pixel value appearing continuously in the same image position. It can be explained that the image acquisition frame is the fixed image acquisition range of the drone's image acquisition unit. When the drone is hovering, the image reference of the image acquisition frame remains completely fixed. The time interval of the identifier pixel value appearing continuously in the same image position is the time difference of the blade tip of two adjacent rotating blades passing through the same image position. By locking the time interval through the fixed image acquisition frame, the core time parameter for speed calculation can be directly obtained without the need to deploy additional dedicated speed detection equipment or access the backend data of the wind turbine unit. This reduces the intermediate links in data transmission and docking, and greatly improves the convenience and overall efficiency of speed detection. Finally, the server can obtain the number of blades corresponding to all rotating blades of the wind turbine, and calculate the real-time rotational speed of the wind turbine blades based on the ratio of the rotation distance determined by the number of blades and the distance between the corresponding coordinate points to the previously obtained time interval. Here, the number of blades is the total number of rotating blades configured in the wind turbine. Conventional wind turbines have three blades. The rotation distance is the distance of one complete rotation of the blade. By calculating the ratio of the number of blades, the distance between the coordinate points, and the time interval, the accurate real-time rotational speed can be obtained directly. The entire calculation process only requires three core basic parameters. The calculation logic is simple and efficient, without the need for complex multi-dimensional data fitting, further improving the calculation efficiency of rotational speed detection. At the same time, the accurate real-time rotational speed can also provide reliable basic data for subsequent blade status detection, thereby improving the overall detection efficiency of the entire wind farm inspection process.

[0024] Furthermore, in this embodiment, the aforementioned "determining the identifier pixel value based on a preset pixel matching strategy" may further include the following steps: The coordinate points corresponding to different blade regions are determined as primary marker points, and marker connection lines connecting the primary marker points to the rotation center point are generated. The blade coordinate point located at the center of the label connection line is determined as the slave label point, and the pixel values ​​corresponding to the master label point and the slave label point are combined into the label pixel value.

[0025] For example, in this embodiment, the determination of the identifier pixel value can be specifically implemented based on the following method steps: First, when it is necessary to determine the identifier pixel value based on a preset pixel matching strategy, the server can first determine the identifier coordinate points corresponding to different blade areas as primary identifier points. It can be explained that different blade areas correspond to independent areas of different rotating blades of the wind turbine in the captured image, and the identifier coordinate points correspond to the tip position of a single blade. Using these as primary identifier points, the core blade identifier with the highest recognition in the image can be directly locked, avoiding interference from other irrelevant elements in the image to the identifier recognition. There is no need to repeatedly screen effective recognition targets, thus improving the overall detection efficiency from the initial stage of identifier determination. Next, the server can generate a marker connection line connecting the primary marker point to the rotation center point. The rotation center point is the corresponding coordinate position of the wind turbine rotor rotation axis in the captured image. The generated marker connection line extends radially along the single blade, completely covering the main body of the blade. Through this marker connection line, the effective range of subsequent pixel matching can be accurately defined, eliminating the need for indiscriminate full-screen pixel scanning of the entire blade area, greatly reducing the computational load of pixel matching, and further improving detection efficiency. Then, the server can determine the blade coordinate point located at the center of the marker connection line as the slave marker point. Here, the center of the marker connection line corresponds to the middle area of ​​a single blade. The blade coordinate point at this position and the master marker point at the blade tip position form a two-level marker reference distributed radially along the blade. By adding slave marker points, the recognition deviation of a single marker point due to interference from the screen environment can be avoided. There is no need to repeatedly verify the accuracy of the marker, reducing the invalid links of repeated detection and improving the stability of the detection process and the overall execution efficiency. Then, the server can combine the pixel values ​​corresponding to the primary and secondary markers into a marker pixel value. It can be explained that the primary and secondary markers correspond to the pixel features of different radial positions of the blade. The combined marker pixel value contains feature information from both levels of position, which greatly improves the uniqueness and recognizability of the marker. In the subsequent marker tracking and detection process, the image position of the target marker can be quickly and accurately locked without scanning the entire screen frame by frame to find the target. This greatly shortens the time spent on marker tracking and improves the overall detection efficiency from the core detection stage.

[0026] Step S2 includes the following: Based on the real-time rotational speed, the crossing time window corresponding to the gap space formed by the adjacent rotating blades of the wind turbine is determined, and the estimated crossing time of the UAV crossing through the gap space is determined.

[0027] For example, in this embodiment, after acquiring the real-time rotation speed, the server can first further clarify the real-time rotation pattern of all rotating blades of the wind turbine based on the real-time rotation speed, so as to locate the gap space formed between adjacent rotating blades. Here, it can be explained that the gap space can be understood as a continuous space that can be passed through by the drone formed by two adjacent rotating blades during rotation, and the position of this space will change synchronously with the rotation of the rotating blades. In real-world scenarios, wind turbines are usually equipped with three rotating blades. Therefore, three independent gap spaces will be formed between two adjacent blades. In this embodiment, by completing the location of the gap space, the effective area that the drone can pass through can be locked. Next, after determining the gap space, the server can further determine the crossing time window corresponding to the gap space formed by adjacent rotating blades based on the real-time rotation speed of the wind turbine. Here, the crossing time window can be understood as the continuous time period during which the gap space rotates to a passable position and can ensure that the drone can safely complete the crossing. In addition, the server also needs to combine the spatial range of the gap space with the basic flight status of the drone to determine the estimated crossing time of the drone through the gap space. The estimated crossing time can be understood as the time required for the drone to complete the entire crossing flight from entering the gap space to completely leaving the gap space.

[0028] Furthermore, in this embodiment, the aforementioned "determining the crossing time window corresponding to the gap space formed by adjacent rotating blades of the wind turbine based on the real-time rotational speed, and determining the estimated crossing time of the UAV crossing through the gap space" may also include the following steps: Using the rotation center point as the center, generate a rotation profile that passes through each marked coordinate point, and generate a marker connecting line that connects the marked coordinate points to the rotation center point; Based on the first image, a gap dividing line is formed starting from the rotation center point to divide adjacent marker connecting lines and connect the rotation contour; The gap dividing lines are arrayed, and with the rotation center point as the center, arc line segments are generated that connect adjacent blade regions and pass through each array point. Determine the chord length of each arc segment, and in response to any chord length exceeding the device size of the corresponding UAV, determine the array point corresponding to that chord length as a valid point; The crossing time window for each valid point is determined based on the real-time rotation speed and the arc length of each corresponding arc segment, and the estimated crossing time is determined based on the equipment distance of the UAV to the wind turbine.

[0029] For example, in this embodiment, the crossing time window and the estimated crossing duration can be implemented based on the following method steps: First, using the already determined rotation center point of the wind turbine as the center, the server can generate a rotation profile passing through each marked coordinate point, and simultaneously generate a marker connecting line connecting the marked coordinate points to the rotation center point. It can be explained that the rotation center point is the corresponding coordinate position of the rotation axis of the wind turbine rotor in the first image, and the marked coordinate point corresponds to the tip position of a single rotating blade. The generated rotation profile is the complete circular trajectory formed by the rotation of all blade tips, and the marker connecting line is the radial line connecting the tip of a single blade to the rotation center point, completely corresponding to the radial extension range of a single blade. By generating the rotation profile and the marker connecting line, the overall rotation range of the blade and the radial boundary of a single blade can be accurately defined directly in the first image without the need for indiscriminate range recognition of the entire content of the first image, which greatly reduces the effective analysis range of subsequent gap detection and improves the overall detection efficiency from the beginning of the detection process. Next, based on the first image, the server can form a gap dividing line that divides the adjacent marker connecting lines and connects the rotation contour, starting from the rotation center point. Here, the adjacent marker connecting lines correspond to the radial boundaries of two adjacent rotating blades. The gap dividing line is located exactly in the middle of the two adjacent marker connecting lines, with one end connecting to the rotation center point and the other end connecting to the rotation contour. It completely corresponds to the center baseline of the gap between two adjacent blades. It can be explained that by generating the gap dividing line, the core analysis baseline of the gap between adjacent blades can be directly locked without the need for full-range scanning analysis of the entire gap area, which greatly reduces the calculation range of subsequent point detection and further improves the detection efficiency of gap-related parameters. Then, the server can perform array division on the generated gap dividing line, and then generate arc segments connecting adjacent blade regions and passing through each array point, with the rotation center point as the center. Array division is to split the complete line segment of the gap dividing line from the rotation center point to the rotation contour into multiple continuous array points. For example, the gap dividing line can be divided into fifty continuous array points at equal intervals. Each array point corresponds to a fixed radial position on the gap dividing line. The generated arc segments are centered on the rotation center point, with the distance from the rotation center point to the corresponding array point as the radius, and the two ends are connected to two adjacent blade regions respectively, completely corresponding to the arc range of the gap between adjacent blades at the corresponding radial position. Here, by decomposing the entire gap dividing line into multiple independent array points through array division, and then generating arc segments for each point, segmented and accurate detection of the entire radial range of the gap can be achieved. There is no need to perform complex calculations on the entire gap line as a whole. The calculation amount of a single segment after splitting is greatly reduced, which effectively improves the calculation efficiency of detection. At the same time, it can also avoid the omission of local details in the overall detection, and there is no need for subsequent repeated detection to complete the data. Next, the server can determine the chord length of each arc segment. When any chord length is detected to be greater than the corresponding drone's device size, the array point corresponding to that chord length is directly determined as a valid point. It can be explained that the chord length is the straight-line distance between the two endpoints of the arc segment, corresponding to the straight-line gap width between two adjacent blades at the radial position of the array point. The drone's device size is the maximum lateral width of the drone's fuselage. For example, if the chord length of a certain arc segment is two meters and the drone's device size is one meter, the chord length is greater than the drone's device size, and the array point can be determined as a valid point. Here, by directly comparing the chord length with the drone's device size, the screening of passable points can be completed quickly without complex spatial simulation and collision verification. A single comparison can complete the determination of valid points, greatly improving the detection efficiency of valid passable points. At the same time, it can also directly eliminate invalid points that cannot pass, avoiding unnecessary parameter calculations for invalid points in the future, and further reducing the overall detection calculation workload. Finally, based on the real-time rotational speed of the wind turbine and the arc length of each corresponding arc segment, the crossing time window for each valid point is determined. Simultaneously, based on the device distance between the UAV and the wind turbine, the estimated crossing time is determined. It can be noted that the arc length is the total length of the arc of the arc segment. Combined with the real-time rotational speed, the gap corresponding to the valid point can be accurately calculated. The complete passable time period after passing through the fixed reference position is the crossing time window. The device distance between the UAV and the wind turbine is the straight-line flight distance the UAV needs to complete a single crossing flight. Based on this distance, the estimated crossing time required for the UAV to complete the crossing can be directly calculated. Here, core parameter calculations are only performed on the selected valid points, eliminating the need for full calculations on all array points. This significantly reduces the computational load of core parameters, improves the detection efficiency of core parameters, and provides a reliable basis for subsequent operational plans, avoiding repeated detections and plan adjustments due to parameter deviations. This improves the overall detection and operational efficiency throughout the entire process.

[0030] Figure 2 A schematic diagram of the rotated profile in this embodiment is shown, as follows: Figure 2 As shown, the outer dashed lines form the rotational profile connecting the blade tips. A gap dividing line is generated outward from the center of rotation between two adjacent blades, and multiple array points are determined on this dividing line. Then, multiple concentric arc segments are drawn with the center point as the center, passing through these points and connecting the edges of the blades on both sides. These are labeled as L1, L2, L3, and L4, respectively. It can be seen that the arc segments distributed from the inside out intuitively reflect the array division process of the gap space. They are mainly used to calculate the gap width at different radial positions, that is, the chord length of the line segments, so as to assess whether the UAV can safely pass through the corresponding points.

[0031] Furthermore, in this embodiment, the aforementioned "determining the estimated travel time based on the equipment distance to the corresponding wind turbine of the UAV" may also include the following steps: A first distance is determined along the gap dividing line corresponding to the rotation center point of each effective point, and a first wind disturbance coefficient is determined based on the first distance for the corresponding tower dimension; A second distance is determined along the gap dividing line corresponding to the rotation profile of each effective point, and a second wind disturbance coefficient is determined based on the second distance for the corresponding tip dimension; The first and second wind disturbance coefficients corresponding to the same effective point are weighted and summed to obtain the comprehensive wind disturbance coefficients. The initial flight time is determined based on the distance and speed of the drone to the corresponding wind turbine. The estimated crossing time for each effective point is obtained by multiplying the initial flight time with each comprehensive wind disturbance coefficient.

[0032] For example, in this embodiment, determining the estimated traversal time based on the device distance can be achieved through the following method steps: First, the server can determine the first distance from the rotation center point corresponding to each effective point along the gap dividing line, and determine the first wind disturbance coefficient of the corresponding tower dimension based on the first distance. It can be explained that the gap dividing line is the center reference line of the gap between adjacent rotating blades, the effective point is a fixed point that has been screened and can be safely passed by the drone, the rotation center point is the rotation axis of the wind turbine rotor, the first distance is the straight-line length from the effective point to the rotation center point, and the first wind disturbance coefficient of the tower dimension is a correction parameter that reflects the impact of airflow disturbance around the tower on the drone's flight. Here, for example, if the first distance of an effective point from the rotation center point along the gap dividing line is 20 meters, the point is farther from the tower, the tower has less obstruction of the surrounding airflow, and the corresponding first wind disturbance coefficient is closer to the reference value of 1. However, if the first distance of the effective point is 5 meters, the point is closer to the tower, the airflow disturbance brought by the tower is more obvious, and the corresponding first wind disturbance coefficient will be adjusted accordingly. By determining the first distance along a fixed gap dividing line, the fixed benchmark for calculating the wind disturbance coefficient can be directly locked, eliminating the need for full-range wind field scanning and data acquisition across the entire blade rotation plane. This significantly reduces the detection range and computational load of wind disturbance parameters, improving overall detection efficiency from the very beginning of parameter acquisition. Next, the server can determine the second distance of the rotation profile corresponding to each effective point along the gap dividing line, and determine the second wind disturbance coefficient of the corresponding tip dimension based on the second distance. It can be explained that the rotation profile is the complete circular trajectory formed by the rotation of the blade tip, the second distance is the straight-line length from the effective point to the rotation profile, and the second wind disturbance coefficient of the tip dimension is a correction parameter reflecting the impact of the rotating airflow at the blade tip on the UAV's flight. For example, if the second distance of a certain effective point from the rotation profile along the gap dividing line is ten meters, this point is farther from the blade tip, resulting in less tip airflow disturbance caused by the blade rotation, and the corresponding second wind disturbance coefficient is closer to the baseline value. If the second distance of the effective point is three meters, which is closer to the tip of the blade, the airflow disturbance caused by the rotation is more intense, and the corresponding second wind disturbance coefficient will be adjusted accordingly. Here, the baseline and calculation logic that are completely consistent with the first wind disturbance coefficient are adopted. There is no need to switch the detection rules and calculation model. The baseline data of the gap dividing line can be directly reused to complete the determination of the second distance, avoiding repeated baseline positioning operations, further reducing invalid calculations in the detection process, and improving the detection efficiency of wind disturbance parameters. At the same time, the dual-dimensional wind disturbance coefficient detection can avoid the parameter deviation caused by single-dimensional detection, eliminating the need for subsequent repeated verification and correction of parameters, and improving the first pass rate of the detection results. Then, the server can perform a weighted summation of the first and second wind disturbance coefficients corresponding to the same valid point to obtain the comprehensive wind disturbance coefficient. It should be noted that the weighted summation calculation combines two independent wind disturbance coefficients into a single comprehensive correction parameter based on the wind disturbance influence weights of the tower and tip dimensions. The weights can be pre-set according to the typical wind farm environment. For example, if the first wind disturbance coefficient of a certain valid point is 1.05, the second wind disturbance coefficient is 1.1, the weight of the tower dimension is 0.4, and the weight of the tip dimension is 0.6, the comprehensive wind disturbance coefficient obtained after weighted summation is 1.08. Here, the weighted summation calculation combines two independent wind disturbance parameters into a single comprehensive correction parameter, eliminating the need to process two sets of wind disturbance data separately in subsequent time-based calculations. This significantly simplifies the subsequent calculation process, reduces the detection latency caused by parallel processing of multiple parameters, and further improves the overall detection efficiency. Next, the initial flight time is determined based on the distance and speed of the drone to the wind turbine. The distance is the total straight-line distance the drone needs to cover in a single crossing flight, the speed is the drone's preset stable crossing speed, and the initial flight time is the baseline time required for the drone to complete the crossing flight under ideal conditions without considering any airflow disturbances. For example, if the distance between the drone and the wind turbine is ten meters and the drone's speed is five meters per second, the calculated initial flight time is two seconds. Here, the initial flight time can be calculated using only two basic parameters: distance and speed. There is no need to establish a complex drone flight dynamics model; the calculation logic is simple and direct, allowing for rapid acquisition of the baseline time value, significantly reducing the time spent on baseline parameter detection and improving detection efficiency. Finally, the server can calculate the estimated crossing time for each valid point based on the product of the initial flight time and each comprehensive wind disturbance coefficient. It should be noted that the estimated crossing time is the actual time required for the UAV to complete the crossing flight after considering the influence of on-site wind disturbance. This is a core parameter for determining the subsequent operational plan. For example, if the initial flight time for a valid point is two seconds and the comprehensive wind disturbance coefficient is 1.08, the estimated crossing time after the product calculation is 2.16 seconds. Wind disturbance correction for the initial time can be completed with only one product calculation, eliminating the need for multiple iterative simulations and complex environmental fitting calculations. This allows for rapid and accurate output of the estimated crossing time, simplifying the detection logic throughout and reducing unnecessary computational steps. Furthermore, since the final time calculation is only performed for valid points, without indiscriminate calculations for all points, the computational load is further reduced, significantly improving overall detection efficiency.

[0033] Step S3 includes the following: Based on the matching relationship between the crossing time window and the estimated crossing time, the UAV is controlled to perform a crossing flight or a circling flight around the corresponding wind turbine, so that the UAV reaches a second position located on the second rotating side of the wind turbine and symmetrical to the first position.

[0034] For example, in this embodiment, the server can determine the matching relationship based on the determined crossing time window and the estimated crossing time. Here, the matching relationship refers to the corresponding adaptation relationship between the passable time of the crossing time window and the estimated crossing time required for the drone to complete the crossing. In one scenario, when the matching relationship is determined to meet the safe passage requirements, the drone can be controlled to perform a cross-flight over the corresponding wind turbine. That is, the drone is controlled to fly across the impeller plane through the gap space formed by adjacent rotating blades. This flight path does not require long-distance detours along the blade tip, and the path length is greatly shortened. The corresponding path planning only needs to lock the passage section of the gap space, without the need for full-range path fitting of the overall blade profile, further reducing the complexity of path planning and improving planning efficiency and on-site operation efficiency. In another scenario, when the matching relationship is determined to be inconsistent with the safe passage requirements, the server needs to control the drone to perform a fly-around flight around the corresponding wind turbine. The fly-around flight refers to the drone completing a fly-around flight across the impeller plane along the safe range of the blade tip. In addition, after the drone completes its flight or circumnavigation, it needs to reach a second position on the second rotating side of the wind turbine. This second position is symmetrical to the first position where the drone initially collected data. That is, the second and first positions maintain a completely symmetrical spatial relationship along the rotor's plane of rotation. For example, if the first position is five meters vertically from the rotor's plane of rotation, located on the windward side of the rotor, then the corresponding second position is also five meters vertically from the rotor's plane of rotation, located on the leeward side of the rotor. The spatial coordinates of the two points are perfectly mirrored along the rotor's plane of rotation. This ensures that the baseline for subsequent data collection on both sides is completely consistent, avoiding reshoots and additional path planning due to point deviations, and improving the overall inspection efficiency throughout the entire process.

[0035] Furthermore, in this embodiment, the aforementioned "controlling the UAV to perform cross-flight or around-flight to the corresponding wind turbine based on the matching relationship between the cross-flight time window and the estimated cross-flight duration" may further include the following steps: The crossing time window corresponding to the same valid point is compared with the estimated crossing time. If the crossing time window is greater than the estimated crossing time, the valid point is determined as the crossing point. The second distance corresponding to each crossing point is summed with the preset redundancy distance. The estimated detour time is determined based on the obtained one-way detour distance. The estimated crossing time corresponding to the same crossing point is compared with the estimated detour time. If the estimated crossing time for any crossing point is less than the estimated detour time, the preparation time for the UAV to fly from the first position to the crossing point is determined based on the device speed. The preparation time and estimated crossing time for the same crossing point are summed, and the UAV is controlled to perform crossing flight to the corresponding wind turbine based on the crossing point with the minimum total flight time; or, If the estimated crossing time for all crossing points is not less than the estimated detour time, control the UAV to perform detour flight around the corresponding wind turbine based on the first position.

[0036] For example, in this embodiment, determining the flight mode of the drone based on the matching relationship can be achieved through the following method steps: First, for each identified valid location, the server can directly compare the crossing time window corresponding to the same valid location with the estimated crossing time. It can be explained that a valid location is a fixed location that has been pre-screened and is safe for drones to pass through. The crossing time window is the continuous time interval between adjacent blades corresponding to that valid location that allows the drone to safely cross. The estimated crossing time is the total time actually required for the drone to complete the crossing flight at that location. For example, if the crossing time window for a valid location is three seconds and the estimated crossing time is 2.2 seconds, and the crossing time window is longer than the estimated crossing time, then that valid location can be identified as a crossing location. Here, by first comparing the time of a single location, the initial screening of crossing locations can be quickly completed, directly eliminating invalid locations that do not meet the safety crossing requirements. This eliminates the need for subsequent full-process calculations for all valid locations, significantly reducing the subsequent computational load and improving overall detection efficiency from the very beginning of the matching and judgment process. Next, for all the selected crossing points, the server can sum the second distance and the preset redundancy distance for each crossing point, determine the estimated flight time based on the obtained one-way flight distance, and compare the estimated crossing time and the estimated flight time for the same crossing point. It can be noted that the second distance is the straight-line length from the crossing point along the gap dividing line to the rotation profile; the preset redundancy distance is the safety protection distance reserved during the drone's flight; the one-way flight distance is the total one-way flight length the drone needs to travel from its current position along the blade tip to complete the flight; and the estimated flight time is the time required for the drone to complete the flight. The total time required for a detour is calculated. For example, if the second distance corresponding to a certain crossing point is eight meters and the preset redundancy distance is two meters, the sum of the two results in a one-way detour distance of ten meters. Combined with the speed of the drone, the corresponding estimated detour time can be quickly calculated. Here, a baseline value for the detour time is obtained through a unified calculation logic, eliminating the need to simulate and plan the detour path for each point separately. This greatly simplifies the calculation process for the detour time. At the same time, by comparing the crossing and detour times of the same point, the optimal passage method for that point can be quickly identified, avoiding invalid path planning calculations and further improving the efficiency of detection and planning. Then, when the estimated crossing time for any crossing point is detected to be less than the estimated detour time, the server can determine the preparation time for the drone to fly from the first position to that crossing point based on the drone's device speed. It can be explained that the first position is a fixed spatial point where the drone performs its initial data collection, and the preparation time is the total time required for the drone to fly smoothly from the first position to the target crossing point. For example, if the straight-line flight distance from the first position to a certain crossing point is 15 meters, and the drone's device speed is 5 meters per second, the preparation time for that point can be calculated to be 3 seconds. Here, the preparation time is calculated only for points with better crossing efficiency, eliminating the need for full preparation path calculations for all crossing points. This significantly reduces unnecessary calculation steps, improves the detection efficiency of core parameters, and provides a reliable benchmark for subsequent total time calculations, avoiding duplicate detections due to parameter deviations. Next, the server can sum the preparation time and the estimated crossing time for the same crossing point to obtain the total flight time for that crossing point. Then, from all the qualifying crossing points, it selects the crossing point with the shortest total flight time and controls the drone to perform the crossing flight to the corresponding wind turbine based on this optimal crossing point. It can be explained that the total flight time is the total time taken for the drone to fly from the first position to the target crossing point and complete the entire crossing flight. The crossing point with the shortest total flight time is the most efficient operation throughout the entire process. For example, if the total flight time for crossing point A is five seconds and the total flight time for crossing point B is four seconds, then point B is directly selected as the final execution point. Here, by summing the total time and filtering by the minimum value, the most efficient crossing point can be locked in one step. There is no need to perform full-process path simulation and efficiency comparison for all points, which greatly shortens the time spent on selecting the optimal path and improves the overall detection and operation efficiency. At the same time, the flight action is directly executed based on the selection results without additional manual intervention and secondary verification, further improving the execution efficiency of the entire process. Finally, when the estimated crossing time for all crossing points is not less than the estimated detour time, the server can control the drone to perform a detour flight around the corresponding wind turbine based on the first position. This means that the estimated crossing time for all crossing points is not less than the estimated detour time, indicating that the efficiency of crossing flight is lower than that of detour flight under the current working conditions. Detour flight is the optimal operation method under the current scenario. Here, through the comparison of all points in advance, the optimal flight mode can be directly determined without having to re-plan the entire detour path when the crossing mode is not feasible. This avoids the repetitive rework of the planning process, greatly improves the efficiency of path detection and determination, and ensures that feasible operation solutions can be obtained quickly under all working conditions, without any detection stagnation due to the lack of available solutions.

[0037] Furthermore, in this embodiment, the aforementioned "determining the preparation time for the drone to fly from the first position to the crossing point based on the device speed" may further include the following steps: Obtain the acquisition center point of the corresponding image acquisition frame, and generate a center connection line that passes through the acquisition center point and connects the rotation center point and the rotation contour; The response determines the crossing point of any rotating blade on the central connecting line based on the image acquisition frame, and determines the position of the crossing point on the central connecting line as the moving position. The drone is controlled to adjust its movement along the rotation plane of the corresponding wind turbine based on the device speed, so that the acquisition center point coincides with the movement position, and the preparation time corresponding to the adjustment movement is obtained.

[0038] For example, in this embodiment, the determination of the preparation time for flying from the first position to the crossing point can be specifically based on the following method steps: First, the server can obtain the acquisition center point corresponding to the image acquisition frame of the UAV, and generate a center connection line that passes through the acquisition center point and connects the rotation center point and the rotation contour. It can be explained that the image acquisition frame is the fixed image acquisition range of the image acquisition unit on the UAV, the acquisition center point is the center position of the image acquisition frame, the rotation center point is the rotation axis of the wind turbine rotor, and the rotation contour is the complete circular trajectory formed by the rotation of the blade tip. The generated center connection line connects the rotation center point at one end and the rotation contour at the other end, passing through the acquisition center point of the image acquisition frame throughout. It is a fixed baseline for subsequent point detection and UAV movement. Here, by generating a fixed and uniform center connection line, the baseline path of UAV movement can be directly locked without performing indiscriminate path scanning and baseline positioning on the entire rotation plane, which greatly reduces the amount of calculation for subsequent point detection and movement path planning, and improves the overall detection efficiency from the beginning of the pre-detection period. Next, the server can continuously track the real-time position of the crossing point of the corresponding rotating blade based on the image content captured in real time by the image acquisition frame. In response, it determines that the crossing point of any rotating blade appears on the central connecting line based on the image acquisition frame, and determines the position of the point on the central connecting line corresponding to the crossing point as the moving position. It can be explained that the crossing point is the target point of the UAV's crossing flight, and the moving position is the fixed spatial point that the UAV needs to fly to. For example, when the crossing point rotates with the rotating blade to a position that completely coincides with the central connecting line, the fixed spatial point on the central connecting line corresponding to this coincidence position is the moving position that needs to be determined. Here, the moving position is determined only when the crossing point appears on the central connecting line, which can ensure that the reference of the moving position is completely consistent with the reference of the acquisition center point. There is no need to repeatedly calibrate the spatial coordinates of the point, nor is there a need to continuously plan the path of the full rotation trajectory of the crossing point, which greatly reduces invalid point tracking calculations and further improves the efficiency of moving position detection and determination. Then, the server can control the drone to adjust its movement along the rotation plane of the corresponding wind turbine based on the device speed, so that the acquisition center point coincides with the determined moving position. It can be explained that the device speed is the drone's preset stable flight speed, the rotation plane is the fixed plane formed by the rotation of the wind turbine blades, and the adjustment movement is the drone's smooth position adjustment flight from the first position to the moving position. For example, if the straight-line distance between the drone's first position and the determined moving position is twelve meters, and the drone's device speed is four meters per second, the drone will adjust its position along the rotation plane at this stable speed until the acquisition center point of the image acquisition frame completely coincides with the moving position. Here, adjusting the movement along a fixed rotation plane at a uniform device speed can ensure the stability of the movement process and the accuracy of the duration calculation. There is no need to frequently adjust the flight speed and flight direction during the movement, avoiding the detection delay caused by invalid attitude adjustments. At the same time, the requirement for the acquisition center point to coincide with the moving position can ensure that the image reference of subsequent acquisition operations is completely consistent, without the need for subsequent additional image calibration processing, further improving the overall detection efficiency. Finally, as the drone completes all its adjustments and movements, the server can simultaneously acquire the preparation time for this adjustment and movement. This preparation time is the total time consumed by the drone from its initial position to complete all adjustments and movements and stably reach its new position. It is a core parameter for determining subsequent flight plans. Here, acquiring the preparation time simultaneously with the drone's position adjustment eliminates the need for separate time simulation calculations or manual statistics of movement time. It directly provides a precise time value that perfectly matches the actual on-site conditions, avoiding discrepancies between simulation calculations and actual flight. It also eliminates the need for subsequent repeated verification of the time parameter, significantly improving the efficiency and accuracy of preparation time detection. Furthermore, it provides reliable basic data for determining subsequent flight plans, improving overall detection and operational efficiency throughout the entire process.

[0039] Step S4 includes the following: The drone is controlled to perform a second acquisition on the second rotating side of the wind turbine based on the second position, and the detection attributes of the corresponding wind turbine are determined based on the first acquisition, the first acquisition, and the second acquisition of the first image and the second image.

[0040] For example, in this embodiment, after the UAV completes its flight and reaches the second position, the server can control the UAV to perform a second acquisition on the second rotating side of the wind turbine based on the locked second position. Similar to the first acquisition, the second acquisition is a continuous image acquisition operation performed by the UAV on the rotating blades of the second rotating side through its onboard image acquisition unit. During the acquisition process, the UAV always stays within the second position without any unnecessary positional shifts, and all acquired parameters are completely consistent with the first acquisition, thereby ensuring that the acquired image data corresponds to the data acquired in the first acquisition. In addition, after the UAV completes the second data acquisition operation on the second rotating side of the wind turbine, it can simultaneously acquire the second image obtained in this second acquisition, as well as the first image obtained in the first acquisition at the first position. Based on the first and second images obtained in the first and second acquisitions, corresponding data comparison and comprehensive analysis are carried out to determine the detection attributes of the corresponding wind turbine. It can be noted that the detection attributes are a comprehensive judgment result of the overall operation and blade health status of the wind turbine. The overall status is determined in one go by the complete image data from both sides, without the need to process the data of one side separately in batches, or to plan additional verification flight paths to supplement the missing status data. The overall status of the wind turbine can be determined in one analysis, reducing the subsequent additional operation planning steps and further improving the detection efficiency and execution efficiency of the entire inspection process.

[0041] Furthermore, in this embodiment, the aforementioned "determining the detection attributes of the corresponding wind turbine based on the first image and the second image obtained from the first acquisition and the second acquisition" may further include the following steps: Based on the first image and the second image, the first blade region and the second blade region corresponding to each rotating blade are determined, and the first blade region and the second blade region corresponding to the same rotating blade are grouped into the same blade identifier group. The first and second leaf regions located in the same leaf identification group are extracted, and the first and second feature sets of the corresponding first and second leaf regions are compared to obtain the feature deviation value of each detection point. The first and second feature sets include each detection point that is symmetrically distributed along the rotation plane. If any feature deviation value is greater than a preset deviation threshold, the detection point corresponding to that feature deviation value will be identified as an abnormal point. If the characteristic deviation value of any detection point is less than or equal to the preset deviation threshold, the characteristic data of the detection point will be compared with the retrieved historical data. If the obtained historical deviation value is greater than the preset deviation threshold, the detection point will be identified as an abnormal point. The detection attributes of the corresponding wind turbine are determined based on the number and distribution of all corresponding anomaly points.

[0042] For example, in this embodiment, the determination of the detection attribute can be specifically implemented based on the following method steps: First, based on the first image obtained from the first acquisition and the second image obtained from the second acquisition, the server can determine the first blade region and the second blade region corresponding to each rotating blade, respectively. Then, the first blade region and the second blade region corresponding to the same rotating blade are grouped into the same blade identification group. It can be explained that the first blade region comes from the acquisition screen of the first rotating side, and the second blade region comes from the acquisition screen of the second rotating side. The same blade identification group corresponds to the front and back sides of the same rotating blade. For example, the area in the first rotating side image and the area in the second rotating side image of the first rotating blade of a wind turbine will be grouped into the same blade identification group. Here, by grouping the blades in a unified manner, subsequent detection can be carried out directly on a single blade basis, avoiding repeated comparisons caused by chaotic image areas, and improving detection efficiency from the data processing stage. Next, the server can extract the first and second leaf regions located in the same leaf identification group, and then compare the first feature set of the first leaf region and the second feature set of the second leaf region to obtain the feature deviation value for each detection point. It can be noted that the first and second feature sets include each detection point that is symmetrically distributed along the rotation plane. The symmetrical distribution means that for each detection point on the first rotation side, there is a corresponding detection point on the second rotation side. The feature deviation value obtained by comparison can directly reflect the state difference of the corresponding positions on the front and back sides of the same leaf. Here, the method of direct comparison of symmetrical points is adopted, which eliminates the need for indiscriminate feature scanning of the entire screen, reduces the amount of calculation for feature extraction and comparison, speeds up the initial judgment of anomalies, and effectively improves the detection efficiency. Then, in response to any feature deviation value exceeding the preset deviation threshold, the server can identify the detection point corresponding to the feature deviation value as an abnormal point. It can be explained that the preset deviation threshold is the maximum feature difference value allowed when the blade is in a healthy state. When the feature deviation value of the detection point exceeds this value, it can be directly identified as an abnormal point. For example, if the feature deviation value of a certain symmetrical detection point is greater than the set preset deviation threshold, the point will be quickly marked as an abnormal point. Here, the determination is made directly through a single threshold, without the need for multiple complex verifications, which can quickly complete the marking of abnormal points, shorten the time for single point determination, and improve the overall detection speed. Then, if the feature deviation value of any detection point is less than or equal to the preset deviation threshold, the server can compare the feature data of the corresponding detection point with the retrieved historical data. If the obtained historical deviation value is greater than the preset deviation threshold, the detection point is identified as an abnormal point. It can be explained that the historical data is the feature record of the point in its previous normal state, and the historical deviation value is the difference between the current feature and the historical feature. Even if there is no obvious abnormality when comparing the two sides, a gradual abnormality can still be found when comparing with the historical data. This dual judgment method can avoid missed detections, and at the same time, there is no need to add an extra collection link. The verification can be completed with the existing data alone, without increasing the detection time, maintaining efficient detection while improving detection accuracy. Finally, based on the number and distribution of all corresponding anomaly points, the detection attributes of the corresponding wind turbine are determined. It can be noted that the number of points reflects the density of anomalies, while the distribution reflects the location range of the anomalies on the blades. Here, by comprehensively judging the two pieces of information, the overall detection result of the wind turbine can be obtained at once, without the need for multiple item-by-item analysis and judgment. This simplifies the process of generating the final conclusion, greatly improves the efficiency of the detection result output, and makes the entire wind farm detection process faster and smoother.

[0043] Furthermore, in this embodiment, the aforementioned "determining the detection attributes of the corresponding wind turbine based on the number and distribution of all corresponding abnormal points" may further include the following steps: If the number of all abnormal points is greater than the preset abnormal threshold, the detection attribute of the corresponding wind turbine is determined as an abnormal attribute; otherwise, all abnormal points corresponding to the same rotating blade are aggregated into the same rotating abnormal group. The system identifies the circumscribed circle of all abnormal points located in the same rotational anomaly group. If the area ratio of the rotating blade corresponding to the abnormal circumscribed circle is greater than a preset ratio threshold, the detection attribute of the corresponding wind turbine is determined as an abnormal attribute; otherwise, it is determined as a normal attribute.

[0044] For example, in this embodiment, determining the detection attributes based on the number and distribution of abnormal points can be achieved through the following method steps: First, if the number of all abnormal points in response exceeds the preset abnormal threshold, the server can determine the detection attribute of the corresponding wind turbine as an abnormal attribute. It can be explained that the abnormal points are the abnormal locations marked by the previous detection, and the preset abnormal threshold is the critical value for the number of points to determine the overall abnormality of the wind turbine. When the total number of all abnormal points exceeds this critical value, the overall abnormality can be quickly determined without the need for additional regional analysis, simplifying the overall determination process and effectively improving the output efficiency of detection results. Next, when the total number of all abnormal points is less than or equal to the preset abnormal threshold, the server can aggregate all abnormal points on the same rotating blade into the same rotating abnormal group. It can be explained that the same rotating abnormal group corresponds to all marked abnormal points on a single blade. Classifying abnormal points by single blade allows subsequent analysis to focus on the single blade range, avoids cross-interference of multi-blade data, reduces the complexity of data processing, and further improves the efficiency of detection and analysis. Then, the server can determine the outer circle of all abnormal points located in the same rotational anomaly group. It can be explained that the outer circle of the anomaly is the smallest circular area that can completely enclose all abnormal points in the rotational anomaly group. The range of anomaly concentration can be quickly delineated geometrically without the need to perform separate position statistics for each abnormal point, which greatly reduces the calculation time of the anomaly range and makes the determination of the anomaly area more intuitive and efficient. Next, the server can calculate the area ratio of the abnormal circumscribed circle relative to the rotating blade and compare this ratio with a preset ratio threshold. It can be explained that the area ratio is the ratio of the area of ​​the abnormal circumscribed circle to the total area of ​​the rotating blade, and the preset ratio threshold is the area threshold for determining the degree of blade abnormality. For example, if the ratio of the area of ​​the abnormal circumscribed circle to the area of ​​the blade exceeds the preset ratio threshold, it can be determined that the abnormal influence range of the blade has reached the warning standard. Here, through the simple calculation and comparison of the area ratio, the quantitative determination of the abnormal range can be completed quickly, avoiding complex regional analysis and improving the detection speed. Finally, when the area ratio corresponding to the abnormal circumcircle is greater than the preset ratio threshold, the server can determine the detection attribute of the corresponding wind turbine as an abnormal attribute. When the area ratio corresponding to the abnormal circumcircle is less than or equal to the preset ratio threshold, the detection attribute of the corresponding wind turbine is simultaneously determined as a normal attribute. Here, through different two-level judgment rules, the final detection result can be output quickly, with no redundant calculations or repeated verifications throughout the process. While ensuring the accuracy of detection, the execution efficiency of the overall detection process is maximized.

[0045] Another embodiment of the present invention provides an anomaly data processing system for wind farms. Figure 3 Its corresponding system block diagram includes: The rotation speed acquisition module is configured to respond to the first acquisition performed by the UAV on the first rotating side of the wind turbine based on the first position, and acquire the real-time rotation speed of the corresponding wind turbine. The crossing prediction module is configured to determine the crossing time window corresponding to the gap space formed by the adjacent rotating blades of the wind turbine based on the real-time rotation speed, and to determine the estimated crossing time of the UAV crossing through the gap space. The flight execution module is configured to control the UAV to perform cross-flight or around-flight to the corresponding wind turbine based on the matching relationship between the cross-flight time window and the estimated cross-flight duration, so that the UAV can reach a second position located on the second rotating side of the wind turbine and symmetrical to the first position. The detection and determination module is configured to control the UAV to perform a second acquisition on the second rotating side of the wind turbine based on the second position, and to determine the detection attributes of the corresponding wind turbine based on the first acquisition, the first image obtained from the second acquisition, and the second image.

[0046] In the specification provided herein, the algorithms and displays are not inherently related to any particular computer, virtual system, or other device. Various general-purpose systems can also be used with the examples of this invention. The required structure for constructing such systems is apparent from the above description. Furthermore, this invention is not directed to any particular programming language. It should be understood that the contents of the invention described herein can be implemented using various programming languages, and the above description of specific languages ​​is for the purpose of disclosing preferred embodiments of the invention.

[0047] Numerous specific details are set forth in the specification provided herein. However, it will be understood that embodiments of the invention may be practiced without these specific details. In some instances, well-known methods, structures, and techniques have not been shown in detail so as not to obscure the understanding of this specification.

[0048] Similarly, it should be understood that, in order to streamline this disclosure and aid in understanding one or more of the various aspects of the invention, in the above description of exemplary embodiments of the invention, various features of the invention are sometimes grouped together in a single embodiment, figure, or description thereof.

[0049] Those skilled in the art will understand that modules, units, or components of the devices disclosed in the examples herein can be arranged in the devices described in this embodiment, or alternatively, can be located in one or more devices different from the devices in this example. The modules in the foregoing examples can be combined into a single module or, in addition, can be divided into multiple sub-modules.

[0050] Those skilled in the art will understand that the modules in the device of the embodiment can be adaptively changed and placed in one or more devices different from that embodiment. Modules, units, or components in the embodiment can be combined into a single module, unit, or component, and further, they can be divided into multiple sub-modules, sub-units, or sub-components.

[0051] Furthermore, those skilled in the art will understand that although some embodiments described herein include certain features included in other embodiments but not others, combinations of features from different embodiments are meant to be within the scope of the invention and form different embodiments.

[0052] Furthermore, some of the embodiments described herein are methods or combinations of method elements that can be implemented by a processor of a computer system or by other means of performing the functions. Therefore, a processor having the necessary instructions for implementing the methods or method elements forms means for implementing the methods or method elements. Furthermore, the elements described herein in the apparatus embodiments are examples of means for implementing the functions performed by elements for the purposes of carrying out the invention.

[0053] As used herein, unless otherwise specified, the use of ordinal numbers such as “first,” “second,” “third,” etc., to describe ordinary objects merely indicates different instances of similar objects and is not intended to imply that the objects being described must have a given order in time, space, ordering, or any other manner.

[0054] Although the invention has been described with respect to a limited number of embodiments, those skilled in the art will understand from the foregoing description that other embodiments are conceivable within the scope of the invention described herein. Furthermore, it should be noted that the language used in this specification has been chosen primarily for readability and edibility purposes, and not for the purpose of explaining or limiting the subject matter of the invention.

Claims

1. A method for processing abnormal data in wind farms, characterized in that, include: The drone completes the first data acquisition on the first rotating side of the wind turbine based on the first position, and obtains the real-time rotation speed of the corresponding wind turbine. Based on the real-time rotational speed, the crossing time window corresponding to the gap space formed by the adjacent rotating blades of the wind turbine is determined, and the estimated crossing time of the UAV crossing through the gap space is determined. Based on the matching relationship between the crossing time window and the estimated crossing time, the UAV is controlled to perform a crossing flight or a circling flight around the corresponding wind turbine, so that the UAV reaches a second position located on the second rotating side of the wind turbine and symmetrical to the first position. The drone is controlled to perform a second acquisition on the second rotating side of the wind turbine based on the second position, and the detection attributes of the corresponding wind turbine are determined based on the first acquisition, the first acquisition, and the second acquisition of the first image and the second image.

2. The method according to claim 1, characterized in that, The step of obtaining the real-time rotational speed of the wind turbine blades includes: Based on the first image obtained from the first acquisition, determine the blade regions in the first image that indicate each rotating blade; The first image is processed into coordinates to determine the coordinate points of each blade that makes up each blade region. The distances of each blade coordinate point relative to the rotation center point of the wind turbine are obtained, and the blade coordinate point with the largest corresponding distance is determined as the marker coordinate point. The response determines that there are at least two positively aligned marker coordinate points with the same pixel value based on the rotation direction of the rotating blade, and identifies the pixel value as the marker pixel value; otherwise, the marker pixel value is determined based on a preset pixel matching strategy. Control the drone to hover, determine the image acquisition frame pre-set in the image acquisition unit of the drone, and determine the time interval when the identifier pixel value appears continuously in the same image position; The number of blades corresponding to all rotating blades of the wind turbine is obtained, and the real-time rotational speed is determined based on the ratio of the rotational distance determined by the number of blades and the distance between the points corresponding to the marked coordinates, and the time interval.

3. The method according to claim 2, characterized in that, The step of determining the identifier pixel value based on a preset pixel matching strategy includes: The coordinate points corresponding to different blade regions are determined as primary marker points, and marker connection lines connecting the primary marker points to the rotation center point are generated. The blade coordinate point located at the center of the label connection line is determined as the slave label point, and the pixel values ​​corresponding to the master label point and the slave label point are combined into the label pixel value.

4. The method according to claim 2, characterized in that, The process of determining the crossing time window corresponding to the gap space formed by adjacent rotating blades of the wind turbine based on real-time rotational speed, and determining the estimated crossing time of the UAV through the gap space, includes: Using the rotation center point as the center, generate a rotation profile that passes through each marked coordinate point, and generate a marker connecting line that connects the marked coordinate points to the rotation center point; Based on the first image, a gap dividing line is formed starting from the rotation center point to divide adjacent marker connecting lines and connect the rotation contour; The gap dividing lines are arrayed, and with the rotation center point as the center, arc line segments are generated that connect adjacent blade regions and pass through each array point. Determine the chord length of each arc segment, and in response to any chord length exceeding the device size of the corresponding UAV, determine the array point corresponding to that chord length as a valid point; The crossing time window for each valid point is determined based on the real-time rotation speed and the arc length of each corresponding arc segment, and the estimated crossing time is determined based on the equipment distance of the UAV to the wind turbine.

5. The method according to claim 4, characterized in that, The method for determining the estimated travel time based on the equipment distance to the corresponding wind turbine of the drone includes: A first distance is determined along the gap dividing line corresponding to the rotation center point of each effective point, and a first wind disturbance coefficient is determined based on the first distance for the corresponding tower dimension; A second distance is determined along the gap dividing line corresponding to the rotation profile of each effective point, and a second wind disturbance coefficient is determined based on the second distance for the corresponding tip dimension; The first and second wind disturbance coefficients corresponding to the same effective point are weighted and summed to obtain the comprehensive wind disturbance coefficients. The initial flight time is determined based on the distance and speed of the drone to the corresponding wind turbine. The estimated crossing time for each effective point is obtained by multiplying the initial flight time with each comprehensive wind disturbance coefficient.

6. The method according to claim 5, characterized in that, The step of controlling the UAV to perform a flight or circumnavigation around the corresponding wind turbine based on the matching relationship between the flight time window and the estimated flight duration includes: The crossing time window corresponding to the same valid point is compared with the estimated crossing time. If the crossing time window is greater than the estimated crossing time, the valid point is determined as the crossing point. The second distance corresponding to each crossing point is summed with the preset redundancy distance. The estimated detour time is determined based on the obtained one-way detour distance. The estimated crossing time corresponding to the same crossing point is compared with the estimated detour time. If the estimated crossing time for any crossing point is less than the estimated detour time, the preparation time for the UAV to fly from the first position to the crossing point is determined based on the device speed. The preparation time and estimated crossing time for the same crossing point are summed, and the UAV is controlled to perform crossing flight to the corresponding wind turbine based on the crossing point with the minimum total flight time; or, If the estimated crossing time for all crossing points is not less than the estimated detour time, control the UAV to perform detour flight around the corresponding wind turbine based on the first position.

7. The method according to claim 6, characterized in that, The process of determining the preparation time for the UAV to fly from the first position to the crossing point based on the device speed includes: Obtain the acquisition center point of the corresponding image acquisition frame, and generate a center connection line that passes through the acquisition center point and connects the rotation center point and the rotation contour; The response determines the crossing point of any rotating blade on the central connecting line based on the image acquisition frame, and determines the position of the crossing point on the central connecting line as the moving position. The drone is controlled to adjust its movement along the rotation plane of the corresponding wind turbine based on the device speed, so that the acquisition center point coincides with the movement position, and the preparation time corresponding to the adjustment movement is obtained.

8. The method according to claim 1, characterized in that, The determination of the detection attributes of the corresponding wind turbine based on the first image and the second image obtained from the first acquisition and the second acquisition includes: Based on the first image and the second image, the first blade region and the second blade region corresponding to each rotating blade are determined, and the first blade region and the second blade region corresponding to the same rotating blade are grouped into the same blade identifier group. The first and second leaf regions located in the same leaf identification group are extracted, and the first and second feature sets of the corresponding first and second leaf regions are compared to obtain the feature deviation value of each detection point. The first and second feature sets include each detection point that is symmetrically distributed along the rotation plane. If any feature deviation value is greater than a preset deviation threshold, the detection point corresponding to that feature deviation value will be identified as an abnormal point. If the characteristic deviation value of any detection point is less than or equal to the preset deviation threshold, the characteristic data of the detection point will be compared with the retrieved historical data. If the obtained historical deviation value is greater than the preset deviation threshold, the detection point will be identified as an abnormal point. The detection attributes of the corresponding wind turbine are determined based on the number and distribution of all corresponding anomaly points.

9. The method according to claim 8, characterized in that, The determination of the detection attributes of the corresponding wind turbine based on the number and distribution of all corresponding anomaly points includes: If the number of all abnormal points is greater than the preset abnormal threshold, the detection attribute of the corresponding wind turbine is determined as an abnormal attribute; otherwise, all abnormal points corresponding to the same rotating blade are aggregated into the same rotating abnormal group. The system identifies the circumscribed circle of all abnormal points located in the same rotational anomaly group. If the area ratio of the rotating blade corresponding to the abnormal circumscribed circle is greater than a preset ratio threshold, the detection attribute of the corresponding wind turbine is determined as an abnormal attribute; otherwise, it is determined as a normal attribute.

10. An anomaly data processing system for wind farms, characterized in that, include: The rotation speed acquisition module is configured to respond to the first acquisition performed by the UAV on the first rotating side of the wind turbine based on the first position, and acquire the real-time rotation speed of the corresponding wind turbine. The crossing prediction module is configured to determine the crossing time window corresponding to the gap space formed by the adjacent rotating blades of the wind turbine based on the real-time rotation speed, and to determine the estimated crossing time of the UAV crossing through the gap space. The flight execution module is configured to control the UAV to perform cross-flight or around-flight to the corresponding wind turbine based on the matching relationship between the cross-flight time window and the estimated cross-flight duration, so that the UAV can reach a second position located on the second rotating side of the wind turbine and symmetrical to the first position. The detection and determination module is configured to control the UAV to perform a second acquisition on the second rotating side of the wind turbine based on the second position, and to determine the detection attributes of the corresponding wind turbine based on the first acquisition, the first image obtained from the second acquisition, and the second image.

Images