Unmanned aerial vehicle cluster intelligent inspection collaborative operation system for wind power plant

Abstract

The invention relates to the technical field of unmanned aerial vehicle inspection and wind field intelligent management, in particular to an unmanned aerial vehicle cluster intelligent inspection collaborative operation system for a wind power plant, which comprises an unmanned aerial vehicle cluster and a server, the server comprises a digital twin modeling module, a cluster cooperative control module, a data processing module, an analysis detection module and a risk report module, and can also comprise a communication adaptation module. The digital twinborn modeling module constructs a three-dimensional digital twinborn body of the wind field, the cluster cooperative control module distributes inspection subtasks through a weighted optimization function, plans a low-resistance path in combination with a rapid exploration random tree optimization algorithm, and dynamically redistributes tasks when a fault occurs; the data processing module pre-processes the multi-source sensor data and fuses the multi-source sensor data into feature vectors; the analysis detection module outputs a defect detection result through the pre-training model; and the risk report module calculates a risk coefficient and generates an evaluation report. And the communication adaptation module ensures stable data transmission.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of unmanned aerial vehicle inspection and wind farm intelligent management, in particular to an unmanned aerial vehicle cluster intelligent inspection collaborative operation system for wind farms. BACKGROUND

[0002] Wind farms are generally characterized by wide distribution, large number of wind turbines, and complex geographical environment. A wind turbine is composed of multiple key components such as blades, nacelles, and towers. Its operating environment often involves complex weather conditions such as strong winds and large temperature differences. Therefore, there are high requirements for the efficiency, accuracy, and stability of the inspection. The traditional wind farm inspection method mainly relies on manual inspection. The staff needs to climb the wind turbine or observe it closely from the ground. This not only has high labor intensity and low operation efficiency, but also faces safety risks such as high-altitude falling and equipment collision. In addition, the identification of defects is highly dependent on personnel experience, which may lead to missed or false inspections due to subjective judgment bias.

[0003] With the application of unmanned aerial vehicle technology, single unmanned aerial vehicle inspection gradually replaces part of the manual work, but still has obvious limitations: single unmanned aerial vehicle has limited endurance and narrow coverage, making it difficult to meet the rapid inspection needs of large-scale wind farms; real-time changes in wind speed and direction in wind farms easily cause the unmanned aerial vehicle to deviate from the flight path, and the anti-interference ability is weak; single unmanned aerial vehicle can only carry limited sensors, making it difficult to achieve multi-dimensional and all-around defect detection, and the detection accuracy is limited; at the same time, existing unmanned aerial vehicle cluster technology is mostly applied to general scenarios, lacking special adaptation design for wind farm environment, and having the following problems: on the one hand, in the complex environment of densely distributed wind turbines in wind farms and real-time wind interference, the task allocation flexibility is insufficient during cluster collaborative scheduling, which easily leads to repeated inspection or missed inspection, and the cluster obstacle avoidance response delay is high, which has a risk of collision; on the other hand, the cross-model adaptation ability is poor, making it difficult to accurately identify defects such as blade cracks, bolt loosening, and tower corrosion of different types of wind turbines.

[0004] In summary, the existing inspection technology cannot balance the efficiency, accuracy, and stability of wind farm inspection, and there is an urgent need for an intelligent inspection system that adapts to the complex environment of wind farms and has high collaborative ability. SUMMARY

[0005] In view of the deficiencies of the prior art, the present application aims to provide an unmanned aerial vehicle cluster intelligent inspection collaborative operation system for wind farms, which solves the problems of low efficiency, weak anti-interference, and insufficient defect detection accuracy of existing wind farm inspection, and can improve the intelligent level of inspection.

[0006] The basic scheme provided by the present application is an unmanned aerial vehicle cluster intelligent inspection collaborative operation system for wind farms, which comprises an unmanned aerial vehicle cluster and a server; the unmanned aerial vehicle cluster is equipped with multiple source sensors; the server comprises a digital twin modeling module, a cluster collaborative control module, a data processing module, an analysis and detection module, and a risk reporting module; The digital twin modeling module is configured to construct a wind farm three-dimensional digital twin according to wind farm geographic information data including fan positions, topography, and a patrol area boundary. The cluster cooperative control module includes a task allocation submodule and a path planning submodule. The task allocation submodule is configured to split a preset patrol task into multiple patrol subtasks, construct an optimization function with patrol coverage, task completion time reciprocal, and unmanned aerial vehicle energy consumption reciprocal as weighted items, calculate an adaptation value of each task allocation scheme by weighted summation according to preset wind farm patrol priorities, and select a task allocation scheme with the largest adaptation value to allocate the patrol subtasks to each unmanned aerial vehicle. During the patrol process, when an unmanned aerial vehicle fails, the patrol subtask completion progress of surrounding unmanned aerial vehicles is calculated, and the unmanned aerial vehicle is selected according to the rule that the lower the patrol subtask completion progress, the higher the priority, and the uncompleted subtask is allocated to the selected unmanned aerial vehicle. The path planning submodule is configured to sample space nodes in the wind farm three-dimensional digital twin to construct a path network using a rapid exploration random tree optimization algorithm, calculate the flight resistance of each path node in combination with real-time wind speed data, and select a path with the shortest length and flight resistance less than a preset flight resistance threshold. During the patrol process, when the distance between adjacent unmanned aerial vehicles is less than a preset safety distance threshold, local path adjustment is triggered through real-time sharing of position information between unmanned aerial vehicle clusters. The data processing module is configured to receive multi-source sensor data and corresponding collection positions uploaded by the unmanned aerial vehicle cluster, obtain multi-source standardized data after calibration, noise reduction, and spatiotemporal alignment preprocessing, and convert the multi-source standardized data into a fusion feature vector through encoding. The multi-source standardized data includes visual data, infrared data, and collection position information. The analysis and detection module is configured to input the fusion feature vector into a pre-trained fan defect detection model, and output a defect detection result including a defect position, a defect type, and a defect size. The defect type includes a blade crack, a cabin bolt loosening, a tower corrosion, a cabin sealing leakage, a hub connection deformation, a wire loosening, and a component temperature anomaly. The risk reporting module is configured to determine a component importance coefficient according to the defect type in the defect detection result, determine a defect severity coefficient according to the defect size in the defect detection result, calculate a risk coefficient by multiplying the defect severity coefficient and the component importance coefficient, and generate a risk assessment report including the defect position, the defect type, and the risk coefficient.

[0007] The principle of the application is that: the digital twin modeling module constructs a three-dimensional digital twin based on wind farm geographic information data, restores the distribution of wind turbines, topography and inspection boundary, provides a virtual mapping scene for task allocation and path planning, ensures high adaptation of simulation scheduling calculation and actual wind farm environment; the task allocation submodule of the cluster collaborative control module splits the overall inspection task into several inspection subtasks, quantitatively evaluates the inspection coverage, task completion time and unmanned aerial vehicle energy consumption of each allocation scheme through a weighted optimization function, calculates the adaptation value, the higher the inspection coverage, the larger the adaptation value, the shorter the task completion time, the larger the reciprocal of the task completion time, the larger the adaptation value, the lower the unmanned aerial vehicle energy consumption, the larger the reciprocal of the unmanned aerial vehicle energy consumption, the larger the adaptation value, so only the task allocation scheme with the largest adaptation value after weighted summation is the optimal scheme, the inspection subtasks are allocated to each unmanned aerial vehicle according to the task allocation scheme with the largest adaptation value, and when the unmanned aerial vehicle fails, an adaptive carrier is selected according to the task completion progress of the surrounding unmanned aerial vehicles to ensure continuous execution of the task; the path planning submodule of the cluster collaborative control module constructs a path network based on the spatial nodes of the digital twin, selects the optimal route with low resistance and short path combining real-time wind speed data, reduces the interference of wind on flight, and at the same time, through real-time sharing of cluster positions, path adjustment is triggered when the distance between adjacent unmanned aerial vehicles is too small to avoid collision risk; the data processing module pre-processes the multi-source sensor data collected by the unmanned aerial vehicle, eliminates noise, time sequence deviation and spatial misplacement, and then fuses multi-dimensional features through coding conversion to form a unified fusion feature vector, providing standardized data for defect detection; the analysis and detection module inputs the fusion feature vector into the pre-trained wind turbine defect detection model, identifies the defect position, defect type and defect size through the deep mapping of the model on the multi-dimensional features, and realizes comprehensive defect detection; the risk report module calculates the risk coefficient through the product of double coefficients based on the defect type and defect size, generates a structured risk assessment report, and provides clear decision basis for operation and maintenance work to form an inspection closed loop.

[0008] The application has the advantages that: the cluster collaborative operation replaces single unmanned aerial vehicle or manual inspection, optimally allocates inspection subtasks through an optimization function, avoids repeated inspection and missed inspection, and improves the inspection efficiency of large-scale wind farms; at the same time, the subtasks are quickly reallocated when a fault occurs to ensure that the inspection process does not interrupt and the operation efficiency is ensured; the path planning selects low-resistance routes combining the three-dimensional digital twin of the wind farm and real-time wind speed data, reduces the influence of wind on flight, and improves the flight stability under complex weather conditions; through the collection of visual, infrared and position information by multi-source sensors, after preprocessing and feature fusion, the defect detection model is used to realize multi-dimensional defect identification, covering various defect types such as blade cracks and temperature abnormalities, reducing missed detection and false detection; through double-coefficient risk assessment associated with defect type and size, the defect risk coefficient is quantified, a clear risk assessment report is generated, the traditional subjective judgment relying on experience is replaced, data support is provided for operation and maintenance work, and the intelligent level of wind farm management is improved.

[0009] Further, the path local adjustment logic of the path planning submodule includes: setting priorities according to the inspection sub-task completion progress of the unmanned aerial vehicle, the higher the inspection sub-task completion progress, the higher the priority; when triggering obstacle avoidance, the unmanned aerial vehicle with low priority is adjusted to the side, and the offset distance is not less than a preset safety redundancy value until the distance between adjacent unmanned aerial vehicles is restored to above the safety distance threshold.

[0010] Through the priority rule and execution mode design of local path adjustment, secondary conflicts caused by unordered adjustment of multiple unmanned aerial vehicles during obstacle avoidance are avoided; the priority setting of high progress task priority ensures that the core inspection task is not affected by obstacle avoidance, and the overall inspection progress is ensured; the preset safety redundancy value further improves the safety of obstacle avoidance, reduces the risk of cluster collision, and enhances the stability and reliability of cluster cooperative flight.

[0011] Further, the fan defect detection model of the analysis and detection module is a lightweight YOLOv8 model, in the pre-training stage, a fan component labeled sample library covering all defect types is constructed and divided into a training set and a test set, the model is trained through the training set and the model performance is verified by using the test set, and a pre-trained fan defect detection model is obtained.

[0012] The lightweight YOLOv8 model is adopted to balance the detection accuracy and operation efficiency, adapt to the real-time data processing demand of the server, and avoid detection delay caused by complex models; through the training and test set verification of the special sample library covering all defect types, the recognition ability of the model to the specific defects of the wind field is ensured, and the accuracy and reliability of the defect detection are improved.

[0013] Further, it further includes a communication adaptation module, the communication adaptation module is used for allocating independent communication time slots for each unmanned aerial vehicle to avoid signal interference in the cluster; when the communication of a certain unmanned aerial vehicle is interrupted, the surrounding unmanned aerial vehicle with normal communication is automatically searched as a relay node to upload data through a multi-node relay transmission mode.

[0014] Through independent communication time slot allocation, signal interference caused by simultaneous data transmission of multiple unmanned aerial vehicles in the cluster is eliminated, and the stability and integrity of data transmission are ensured; the relay transmission mechanism solves the problem of communication interruption caused by terrain obstruction or signal attenuation, ensures that the data collected by the unmanned aerial vehicle can be uploaded to the server in time, avoids data loss, and ensures the continuity of the inspection process. BRIEF DESCRIPTION OF DRAWINGS

[0015] Figure 1 It is the system module diagram of the wind farm unmanned aerial vehicle cluster intelligent inspection cooperative operation system embodiment of the application. DETAILED DESCRIPTION

[0016] The following will be further described in detail through specific embodiments: The embodiment is substantially as shown in the accompanying Figure 1 The unmanned aerial vehicle cluster intelligent inspection collaborative operation system for a wind farm includes an unmanned aerial vehicle cluster and a server; the unmanned aerial vehicle cluster is loaded with multi-source sensors; the server includes a digital twin modeling module, a cluster collaborative control module, a data processing module, an analysis and detection module, and a risk reporting module; The digital twin modeling module is configured to construct a wind farm three-dimensional digital twin according to wind farm geographic information data including wind turbine positions, topography, and inspection area boundaries; The cluster collaborative control module includes a task allocation sub-module and a path planning sub-module; The task allocation sub-module is configured to split a preset inspection task into multiple inspection sub-tasks, construct an optimization function with inspection coverage, task completion time reciprocal, and unmanned aerial vehicle energy consumption reciprocal as weighted items, set weight coefficients of the weighted items in the optimization function according to wind farm inspection priorities, calculate an adaptation value of each task allocation scheme by weighted summation, select a task allocation scheme with the largest adaptation value to allocate the inspection sub-tasks to each unmanned aerial vehicle, and in the inspection process, when an unmanned aerial vehicle fails, calculate the inspection sub-task completion progress of surrounding unmanned aerial vehicles, select unmanned aerial vehicles according to the rule that the lower the inspection sub-task completion progress, the higher the priority, and allocate the uncompleted sub-tasks to the selected unmanned aerial vehicles; The path planning sub-module is configured to use a rapid exploration random tree optimization algorithm to sample space nodes in the wind farm three-dimensional digital twin to construct a path network, calculate flight resistance of each path node in combination with real-time wind speed data, and select a path with flight resistance less than a preset flight resistance threshold and shortest length; in the inspection process, when the distance between adjacent unmanned aerial vehicles is less than a preset safety distance threshold, trigger local adjustment of the path through real-time sharing of position information between the unmanned aerial vehicle cluster; The data processing module is configured to receive multi-source sensor data and corresponding collection positions uploaded by the unmanned aerial vehicle cluster, obtain multi-source standardized data after calibration, noise reduction, and spatio-temporal alignment preprocessing, encode and convert the multi-source standardized data into a fusion feature vector, and the multi-source standardized data includes visual data, infrared data, and collection position information; The analysis and detection module is configured to input the fusion feature vector into a pre-trained wind turbine defect detection model, and output a defect detection result including a defect position, a defect type, and a defect size, and the defect type includes a blade crack, a cabin bolt loosening, a tower corrosion, a cabin sealing leakage, a hub connection deformation, a wire loosening, and a component temperature anomaly; The risk report module is configured to determine a component importance coefficient according to a defect type in the defect detection result, determine a defect severity coefficient according to a defect size in the defect detection result, calculate a risk coefficient by multiplying the defect severity coefficient and the component importance coefficient, and generate a risk assessment report containing a defect position, a defect type, and the risk coefficient.

[0017] Further, the path local adjustment logic of the path planning submodule includes setting a priority according to the inspection sub-task completion progress of the unmanned aerial vehicle, and the higher the inspection sub-task completion progress, the higher the priority; when obstacle avoidance is triggered, the unmanned aerial vehicle with low priority is adjusted to the side, and the offset distance is not less than a preset safety redundancy value until the distance between adjacent unmanned aerial vehicles is restored to above a safety distance threshold.

[0018] Further, the fan defect detection model of the analysis and detection module is a lightweight YOLOv8 model, in the pre-training stage, a fan component labeled sample library covering all defect types is constructed and divided into a training set and a test set, the model is trained through the training set and the model performance is verified by using the test set, and a pre-trained fan defect detection model is obtained.

[0019] Further, the communication adaptation module is further included, and the communication adaptation module is configured to allocate independent communication time slots for each unmanned aerial vehicle to avoid signal interference in the cluster; when communication of a certain unmanned aerial vehicle with the server is interrupted, a surrounding unmanned aerial vehicle with normal communication is automatically searched as a relay node to upload data through a multi-node relay transmission mode.

[0020] In the embodiment, the geographic information data collection specifically includes: obtaining the longitude and latitude range of the wind field and the terrain elevation data through the GIS system, obtaining the specific coordinates of the fan, the tower height, the blade length and other parameters through GPS positioning, and obtaining the surface texture data of the wind field by aerial photography of the unmanned aerial vehicle; The three-dimensional modeling process includes: importing the terrain elevation data to generate a terrain model by using a Unity 3D engine, reconstructing a three-dimensional model of the fan (including blade, cabin, tower and other detailed structures) based on the fan parameters and aerial photography data, binding the fan position information and the inspection area boundary, generating a three-dimensional digital twin of the actual wind field, and supporting spatial node query and path planning calling.

[0021] The cluster cooperative control module comprises: a task allocation submodule: taking a preset inspection task of "full coverage inspection of 50 wind turbine components in the whole wind field" as an example, the task is split into 100 inspection subtasks (2 core component inspection subtasks for each wind turbine); the optimization function is set as: the weight of inspection coverage rate is 0.4, the weight of task completion time inverse is 0.3, and the weight of unmanned aerial vehicle energy consumption inverse is 0.3; when a certain unmanned aerial vehicle triggers a fault alarm, the inspection subtask completion progress of the unmanned aerial vehicles within a range of 3 km around the unmanned aerial vehicle is calculated (for example, unmanned aerial vehicle A completes 60%, and unmanned aerial vehicle B completes 30%), unmanned aerial vehicle B is selected as an adaptive carrier, and the uncompleted subtask of the fault unmanned aerial vehicle is allocated; The path planning submodule: the space node sampling interval of the rapid exploration random tree optimization algorithm is 5 m, and the initial path network is constructed; according to the fact that the flight resistance is proportional to the square of the speed, the flight resistance of each node is obtained by multiplying the air resistance coefficient and the real-time wind speed, the air resistance coefficient is obtained by simulating experimental data statistics, and is taken as 0.05, and the preset flight resistance threshold is 1.25 (corresponding to a wind speed of 5 m / s); the safety distance threshold is preset as 5 m according to the size of the unmanned aerial vehicle body, and the path adjustment is triggered when the distance between adjacent unmanned aerial vehicles is less than 5 m.

[0022] The data processing module: the preprocessing procedure: the NTP protocol is used to synchronize the time stamps of multiple source sensor data, the Gaussian filter is used to remove the noise of the visual data, the median filter is used to smooth the infrared data, the pixel coordinates collected by the sensor are mapped to the world coordinates of the digital twin through a coordinate conversion matrix, and spatial alignment is realized; code conversion: a CNN feature extraction network is used to code the visual data, the infrared data and the collection position information, the extracted feature vectors are spliced into a fusion feature vector, and the data format is uniformly adapted to the input of the defect detection model.

[0023] The analysis and detection module: in the pre-training stage: the sample library is constructed as follows: 10,000 wind turbine component labeled images covering 7 types of defects are collected, and are divided into a training set (7,000 images) and a test set (3,000 images) according to a ratio of 7:3, the labeled information includes defect position, defect type and defect size; based on a lightweight YOLOv8 model, the initial learning rate is set to 0.01, the training rounds are 100, the model parameters are optimized through the training set, the model performance is verified by using the test set (the training is stopped when the test set accuracy is greater than 90%), the model input is the fusion feature vector, and the output is the defect position, the defect type and the defect size.

[0024] Risk report module: the defect severity coefficient is divided into three levels according to the defect size, the defect severity coefficient is 1.0 when the defect size is less than or equal to 5cm, the defect severity coefficient is 2.0 when the defect size is greater than 5cm and less than or equal to 10cm, and the defect severity coefficient is 3.0 when the defect size is greater than 10cm; the component importance coefficient is divided into three levels according to the influence degree: the core component is 3.0, the important component is 2.0, and the ordinary component is 1.0, and the core component, the important component and the ordinary component are pre-classified according to their influence on the operation of the fan, such as: the blade and the hub are classified as core components.

[0025] Communication adaptation module: the server allocates independent communication time slots (10ms per time slot) for the unmanned aerial vehicle in the cluster, and polls the transmission data according to the unmanned aerial vehicle number, and the transmission bandwidth is stable at more than 100Mbps; relay transmission: when the packet loss rate of communication between the unmanned aerial vehicle and the server is greater than or equal to 50% for 3 seconds in a row, it is determined that the communication is interrupted, and the surrounding 1km range is searched for an unmanned aerial vehicle with normal communication, and the relay node is selected according to the signal strength, supporting up to 3-hop transmission.

[0026] The system workflow is as follows: Preparation phase: the digital twin modeling module updates the wind field geographic information data, and constructs a three-dimensional digital twin body; the task allocation sub-module splits the inspection task, and generates an optimal allocation scheme through an optimization function; the path planning sub-module plans the initial flight path of each unmanned aerial vehicle; inspection execution: the unmanned aerial vehicle cluster takes off according to the allocation scheme and the initial flight path, and the multi-source sensor synchronously collects data, which is uploaded to the server after being preprocessed and encoded into a fusion feature vector; the server shares the unmanned aerial vehicle position in real time, triggers obstacle avoidance adjustment, and quickly reallocates sub-tasks in case of failure; defect detection and decision: the analysis detection module calls a pre-trained model to process the fusion feature vector, and outputs a defect detection result; the risk report module calculates a risk coefficient, generates and pushes a risk assessment report; inspection completion: the unmanned aerial vehicle returns autonomously after completing the task, the server integrates the inspection data to form a summary report for subsequent review and maintenance by the operation and maintenance personnel.

[0027] The above is only an embodiment of the present application, and the common knowledge of specific structures and characteristics in the scheme is not described in detail, and the ordinary skilled person in the art knows all the ordinary technical knowledge in the technical field of the present application before the application date or the priority date, can know all the prior art in the field, and has the ability to apply conventional experimental means before that date, and the ordinary skilled person in the art can improve and implement the present scheme under the guidance of the present application, and some typical known structures or known methods should not be an obstacle for the ordinary skilled person in the art to implement the present application. It should be pointed out that for those skilled in the art, without departing from the structure of the present application, a number of modifications and improvements can be made, which should also be considered as the protection scope of the present application, which will not affect the effect and practicality of the patent. The protection scope of the present application should be subject to the content of its claims, and the specific implementation mode and the like in the specification can be used to explain the content of the claims.

Claims

1. A collaborative intelligent inspection system for wind farm drone swarms, characterized in that: It includes a drone swarm and a server; the drone swarm is equipped with multi-source sensors; the server includes a digital twin modeling module, a swarm collaborative control module, a data processing module, an analysis and detection module, and a risk reporting module; The digital twin modeling module is used to construct a three-dimensional digital twin of the wind farm based on wind farm geographic information data, including wind turbine location, topography, and inspection area boundaries. The cluster collaborative control module includes a task allocation submodule and a path planning submodule; The task allocation submodule is used to split the preset inspection task into multiple inspection subtasks, construct an optimization function with inspection coverage, reciprocal of task completion time, and reciprocal of UAV energy consumption as weighting terms. The weight coefficients of each weighting term in the optimization function are preset according to the wind farm inspection priority. The adaptation value of each task allocation scheme is calculated by weighted summation, and the task allocation scheme with the largest adaptation value is selected to allocate the inspection subtask to each UAV. During the inspection, when a drone malfunctions, the progress of the inspection sub-tasks of surrounding drones is calculated. Drones are selected according to the rule that the lower the progress of the inspection sub-tasks, the higher the priority. Unfinished sub-tasks are assigned to the selected drones. The path planning submodule is used to construct a path network by sampling spatial nodes in the three-dimensional digital twin of the wind field using a fast exploratory random tree optimization algorithm, and to calculate the flight resistance of each path node by combining real-time wind speed data, and to select the path with the shortest length and flight resistance less than a preset flight resistance threshold. During the inspection, by sharing location information in real time among the drone clusters, when the distance between adjacent drones is less than a preset safe distance threshold, a local path adjustment is triggered. The data processing module is used to receive multi-source sensor data and corresponding acquisition locations uploaded by the UAV cluster, and obtain multi-source standardized data after calibration, noise reduction and spatiotemporal alignment preprocessing. The multi-source standardized data is encoded and converted into a fusion feature vector. The multi-source standardized data includes visual data, infrared data and acquisition location information. The analysis and detection module is used to input the fused feature vector into the pre-trained wind turbine defect detection model and output defect detection results including defect location, defect type and defect size. The defect types include blade cracks, loose nacelle bolts, tower corrosion, nacelle seal leakage, hub connection deformation, wire loosening and component temperature abnormality. The risk reporting module is used to determine the component importance coefficient based on the defect type in the defect detection results, determine the defect severity coefficient based on the defect size in the defect detection results, calculate the risk coefficient by multiplying the defect severity coefficient and the component importance coefficient, and generate a risk assessment report containing the defect location, defect type and risk coefficient.

2. The intelligent inspection and collaborative operation system for wind farm drone swarms according to claim 1, characterized in that: The path planning submodule's path local adjustment logic includes: setting priorities according to the completion progress of the UAV's inspection sub-tasks, with higher priority for higher completion progress; when obstacle avoidance is triggered, the low-priority UAV adjusts its course to the side, with the offset distance not less than the preset safety redundancy value, until the distance between adjacent UAVs is restored to above the safety distance threshold.

3. The intelligent inspection and collaborative operation system for wind farm drone swarms according to claim 2, characterized in that: The wind turbine defect detection model of the analysis and detection module is a lightweight YOLOv8 model. In the pre-training stage, a wind turbine component labeled sample library covering all defect types is constructed and divided into training set and test set. The model is trained using the training set and the model performance is verified using the test set to obtain the pre-trained wind turbine defect detection model.

4. The intelligent inspection and collaborative operation system for wind farm drone swarms according to claim 3, characterized in that: It also includes a communication adaptation module, which is used to allocate independent communication time slots to each drone to avoid signal interference within the cluster; when a drone loses communication with the server, it automatically searches for nearby drones with normal communication as relay nodes and uploads data through a multi-node relay transmission method.

Images