Unmanned aerial vehicle cooperative wind power plant intelligent inspection method and system

Abstract

The invention provides an unmanned aerial vehicle cooperative wind power plant intelligent inspection method and system, and relates to the technical field of unmanned aerial vehicle inspection. The method comprises the following steps: receiving a foundation cloud picture and meteorological characteristic data of a wind power plant in a target area; preprocessing the foundation cloud picture to obtain a foundation cloud picture image, and performing segmentation, feature extraction and recognition processing on the foundation cloud picture image to obtain foundation cloud picture features; extracting features of the meteorological feature data to obtain meteorological features; and inputting the ground-based cloud picture features and the meteorological features into a preset fan equipment operation risk prediction model, calculating to obtain a risk index of the fan equipment, and controlling the unmanned aerial vehicle to perform inspection according to the risk index. The problems that the existing inspection system does not consider the influence of meteorological characteristics on the inspection work, so that the inspection system of a wind power system cannot accurately predict the operation risk of fan equipment, the working difficulty is increased, and the operation requirement of a smart wind power plant is difficult to meet are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of unmanned aerial vehicle inspection, and in particular to a wind farm intelligent inspection method and system coordinated by unmanned aerial vehicles. Background Art

[0002] With the rapid development of my country's wind power industry, the number of wind turbines has increased dramatically. However, the natural environment of wind farms is usually very complex. They may be located in remote mountainous areas or at sea and are exposed to harsh climates for a long time. Wind turbines are hundreds of meters apart, and the turbine cabins are tens of meters high. The use of manual inspections is costly, inefficient, error-prone, and dangerous. How to reduce the operation and maintenance management costs of a large number of wind turbines, promote the intelligence and informatization of wind power systems, and thus increase the power generation revenue of wind farms has become a major problem that needs to be solved in the wind power industry.

[0003] With the introduction of the concept of "smart" wind farms, wind power systems are being comprehensively optimized and upgraded in parts manufacturing, data management, operation and maintenance. Smart wind farms are mainly based on measurement and control technology, communication technology, sensor technology, big data processing technology and various intelligent algorithms to achieve intelligent wind turbine control management, equipment status perception, inspection and maintenance, etc. Through the various sensors inside the wind turbine, the working status data of each device is obtained, and the sensor data is integrated and processed by the edge server inside the wind turbine. It can be known whether the wind turbine is currently in normal working condition, and the possible cause of the failure can be found when the wind turbine fails.

[0004] However, wind power systems have not yet achieved true intelligence. The existing inspection system does not take into account the impact of meteorological characteristics on inspection work, which results in the inability of the wind power system inspection system to accurately predict the operating risks of wind turbine equipment, thereby increasing the difficulty of work and making it difficult to meet the operating requirements of smart wind farms. Summary of the invention

[0005] The problem solved by the present invention is that the existing inspection system does not take into account the impact of meteorological characteristics on the inspection work, which results in the inspection system of the wind power system being unable to accurately predict the operating risks of the wind turbine equipment, thereby increasing the difficulty of the work and making it difficult to meet the operating requirements of the smart wind farm.

[0006] In order to solve the above problems, on the one hand, the present invention provides a wind farm intelligent inspection method coordinated by a drone, the method comprising: Receive ground-based cloud images and meteorological characteristic data of wind farms in the target area; Preprocess the ground-based cloud image to obtain a ground-based cloud image, segment the ground-based cloud image, extract features and perform recognition processing to obtain ground-based cloud features; Extracting features of meteorological feature data to obtain meteorological features; The ground-based cloud map features and meteorological features are input into the preset wind turbine equipment operation risk prediction model, the risk index of the wind turbine equipment is calculated, and the drones are controlled to conduct collaborative inspections based on the risk index.

[0007] Preferably, the process of preprocessing the ground-based cloud image to obtain the ground-based cloud image is as follows: Read the ground cloud image and convert it from RGB to grayscale image; Perform histogram equalization on the grayscale image, use the edge detection algorithm to detect the edge of the grayscale image, and determine whether to apply different Gaussian blur parameters for denoising based on the average value of the edge detection result; Output ground-based cloud image.

[0008] Preferably, the process of segmenting the ground-based cloud image is as follows: The ground-based cloud image is segmented. The specific steps include: Use Otsu's method to perform global thresholding on the ground-based cloud image and invert the result; Use the opening operation to remove noise, then expand the background area through the dilation operation, apply the distance transform to highlight the foreground object, and use the thresholding to extract the foreground area of ​​the ground-based cloud image. The specific process is as follows: The boundary of the foreground area of ​​the ground-based cloud image is shrunk inward by the erosion operation to eliminate small noise points or small foreground objects. The foreground area in the ground-based cloud image is expanded by the dilation operation to replace each pixel with the maximum value of its surrounding area, connect the separated parts in the erosion operation, and maintain the important image structure features of the ground-based cloud image. By calculating the distance from each pixel to the nearest background pixel, the boundaries and features of the foreground area are highlighted, so that the pixel value of the foreground area is adjusted according to the distance from the nearest background pixel, thereby enhancing the edges and features of the foreground area. The distance transformed image is converted into a binary image according to the set threshold, and the foreground area is separated from the background; The foreground area is converted into an 8-bit image, and the unknown area of ​​the ground-based cloud image is determined by subtraction operation; Mark the foreground area, add 1 to all marks, and mark the unknown area as 0; The watershed algorithm is used to segment the image, and the segmentation results are merged to obtain the segmented ground cloud image.

[0009] Preferably, the process of extracting features from the segmented ground cloud image is as follows: Extracting various ground-based cloud image features from an area of ​​the segmented ground-based cloud image, wherein the ground-based cloud image features include regional features, texture features, color features, deep learning features, and multi-scale features; Traverse all regions, extract the ground-based cloud map features of each region, and add the ground-based cloud map features of each region to the feature list; Convert a list of features to a NumPy array.

[0010] Preferably, the process of identifying the ground-based cloud image after feature extraction includes: Define a feature array and a label array, wherein the feature array includes a NumPy array of various features extracted from the ground-based cloud image, and the label array includes a NumPy array of cloud type labels corresponding to each ground-based cloud image; Encode the cloud type label into a number to obtain an encoded digital label array, where the number of label categories is n, where n is a natural number greater than 0; The data set is divided into a training set and a test set, where the test set accounts for 20% of the total data set; Build a sequential model and add a fully connected layer with 256 neurons and ReLU activation function, a dropout layer, and an output layer with n neurons and softmax activation function to obtain a deep learning model; Use the adam optimizer to compile the model, and use the training set data to train the model. Use the test machine for verification during the training process, and use the test set to evaluate the model's loss value and accuracy. Use the trained model to predict the test set and obtain the probability distribution of each sample belonging to each category.

[0011] Preferably, the specific process of using the test set to evaluate the loss value and accuracy of the model is as follows: Classify the test set data according to different cloud types and obtain their respective subsets; Use the trained model to predict the test samples of each cloud type and obtain the prediction results; The loss value is calculated for the entire test set, and the formula is as follows: in, is the total number of samples in the test set, The model is The loss value between the prediction of samples and the true label, is the actual label of the i-th sample, is the probability that the model predicts the positive class; Calculate the accuracy of the entire test set, the formula is as follows: in, is the accuracy of the entire test set, is the number of times the predicted result matches the actual true result, is the total number of samples in the test set; Traverse the subset of each cloud type and calculate the accuracy of the model on each cloud type. The formula is as follows: in, is the accuracy of the model on each cloud type, are different cloud types, Indicates the cloud type The number of samples predicted correctly is Is of cloud type The total number of samples, Is the cloud type A subset of Combining the loss value of the entire test set, the overall accuracy, and the accuracy of each cloud type, we get a comprehensive evaluation index, as follows: in, is the weighted average of the loss value of the entire test set, the overall accuracy, and the accuracy of each cloud type. are all weight parameters.

[0012] Preferably, the process of calculating the risk index of the wind turbine equipment is as follows: The foundation cloud map characteristics and meteorological characteristics are input into the preset wind turbine equipment operation risk prediction model. The formula of the wind turbine equipment operation risk prediction model is as follows: Among them, R is the risk index, is the activation function, ensuring that the output range is between [0, 1], is a weight matrix used to map input features to output risk index, is the bias term, Input features include ground-based cloud map features and meteorological features; The risk index R is compared with the preset risk threshold Compare and make drone inspection decisions: R> , indicating that the equipment is at high risk and requires urgent inspection; R≤ , indicating that the equipment risk is low and no inspection is required for the time being.

[0013] Preferably, it also includes: When the risk index exceeds the preset risk threshold, multiple drones will conduct inspections in coordination; Based on the geographical location of the wind farm, the mission is assigned to the drone closest to the target wind turbine; If multiple wind turbines are at risk, wind turbines with high risk indexes shall be inspected first; The inspection tasks of drones are scheduled and optimized. The specific task allocation formula is as follows: in, is assigned to The inspection mission of the drone, It is The first drone The distance between the fans, It is the biggest limitation of the drone’s flight range. It is The risk index of a wind turbine.

[0014] In another aspect, a UAV-coordinated wind farm intelligent inspection system is provided, which is used to execute the UAV-coordinated wind farm intelligent inspection method, and the system comprises: A receiving module, used for receiving ground-based cloud images and meteorological characteristic data of wind farms in the target area; A preprocessing module is used to preprocess the ground-based cloud image to obtain a ground-based cloud image, and to perform segmentation, feature extraction and recognition processing on the ground-based cloud image to obtain ground-based cloud features; A feature extraction module is used to extract the features of meteorological feature data to obtain meteorological features; The calculation module is used to input the ground-based cloud map characteristics and meteorological characteristics into the preset wind turbine equipment operation risk prediction model, calculate the risk index of the wind turbine equipment, and control the drone to conduct inspections based on the risk index.

[0015] In another aspect, a computing device for a wind farm intelligent inspection method coordinated by a drone is provided, comprising: One or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the one or more programs include instructions for executing any of the methods described.

[0016] Beneficial effects: The present invention provides a drone-coordinated intelligent inspection method and system for wind farms, which improves the accuracy of wind turbine equipment operation risk prediction by increasing the processing of ground cloud maps and meteorological characteristics of wind farms in target areas, and solves the problem that the existing inspection system does not take into account the impact of meteorological characteristics on the inspection work, which results in the inspection system of the wind power system being unable to accurately predict the operation risk of wind turbine equipment, thereby increasing the difficulty of the work and making it difficult to meet the operation requirements of the smart wind farm. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a flow chart of a method for intelligent inspection of wind farms in collaboration with drones; Figure 2 This is a block diagram of a wind farm intelligent inspection system coordinated by drones. DETAILED DESCRIPTION

[0018] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0019] An embodiment of the present invention provides a method for intelligent inspection of a wind farm in cooperation with a drone, the method comprising: Receive ground-based cloud images and meteorological characteristic data of wind farms in the target area; Preprocess the ground-based cloud image to obtain a ground-based cloud image, segment the ground-based cloud image, extract features and perform recognition processing to obtain ground-based cloud features; Extracting features of meteorological feature data to obtain meteorological features; The ground-based cloud map features and meteorological features are input into the preset wind turbine equipment operation risk prediction model, the risk index of the wind turbine equipment is calculated, and the drones are controlled to conduct collaborative inspections based on the risk index.

[0020] Furthermore, the process of preprocessing the ground-based cloud image to obtain the ground-based cloud image is as follows: Read the ground-based cloud image and convert it from RGB to grayscale image to reduce processing complexity and data volume; Perform histogram equalization on the grayscale image, use the edge detection algorithm to detect the edge of the grayscale image, and determine whether to apply different Gaussian blur parameters for denoising based on the average value of the edge detection result, so as to further optimize the image quality; Output the ground-based cloud image. Among them, the edge detection algorithm detects the edge of the grayscale image, and determines whether to apply different Gaussian blur parameters for denoising based on the average value of the edge detection result. The specific steps are as follows:

[0021] Use common edge detection algorithms (such as Sobel and Canny) to perform edge detection on the grayscale image to obtain an edge intensity image; For the obtained edge intensity image, the average value of all its pixel values ​​is calculated; this average value reflects the average intensity of the edge in the entire image; According to the calculated average edge strength, a threshold or a set of thresholds is set; according to these thresholds, it is determined whether to apply different Gaussian blur parameters for denoising; If the average edge strength is higher than the set threshold, it means that there may be more edge information in the image, and a smaller Gaussian blur kernel size can be selected at this time; conversely, if the average edge strength is lower than the threshold, a larger Gaussian blur kernel size can be selected; the Gaussian blur kernel size can be selected based on the best effect obtained through experience or experiments.

[0022] Furthermore, the process of segmenting the ground-based cloud image is as follows: The ground-based cloud image is segmented. The specific steps include: Use Otsu's method to perform global thresholding on the ground-based cloud image and invert the result; Use the opening operation to remove noise, then expand the background area through the dilation operation, apply the distance transform to highlight the foreground object, and use the thresholding to extract the foreground area of ​​the ground-based cloud image. The specific process is as follows: The boundary of the foreground area of ​​the ground-based cloud image is shrunk inward by the erosion operation to eliminate small noise points or small foreground objects. The foreground area in the ground-based cloud image is expanded by the dilation operation to replace each pixel with the maximum value of its surrounding area, connect the separated parts in the erosion operation, and maintain the important image structure features of the ground-based cloud image. By calculating the distance from each pixel to the nearest background pixel, the boundaries and features of the foreground area are highlighted, so that the pixel value of the foreground area is adjusted according to the distance from the nearest background pixel, thereby enhancing the edges and features of the foreground area. The distance transformed image is converted into a binary image according to the set threshold, and the foreground area is separated from the background; The foreground area is converted into an 8-bit image, and the unknown area of ​​the ground-based cloud image is determined by subtraction operation; Mark the foreground area, add 1 to all marks, and mark the unknown area as 0; The watershed algorithm is used to segment the image, and the segmentation results are merged to obtain the segmented ground cloud image.

[0023] Furthermore, the process of feature extraction of the segmented ground cloud image is as follows: Extracting various ground-based cloud image features from an area of ​​the segmented ground-based cloud image, wherein the ground-based cloud image features include regional features, texture features, color features, deep learning features, and multi-scale features; Traverse all regions, extract the ground-based cloud map features of each region, and add the ground-based cloud map features of each region to the feature list; Convert a list of features to a NumPy array.

[0024] Furthermore, the process of identifying the ground-based cloud image after feature extraction includes: Define a feature array and a label array, wherein the feature array includes a NumPy array of various features extracted from the ground-based cloud image, and the label array includes a NumPy array of cloud type labels corresponding to each ground-based cloud image; Encode the cloud type label into a number to obtain an encoded digital label array, where the number of label categories is n, where n is a natural number greater than 0; The data set is divided into a training set and a test set, where the test set accounts for 20% of the total data set; Build a sequential model and add a fully connected layer with 256 neurons and ReLU activation function, a dropout layer, and an output layer with n neurons and softmax activation function to obtain a deep learning model; Use the adam optimizer to compile the model, and use the training set data to train the model. Use the test machine for verification during the training process, and use the test set to evaluate the model's loss value and accuracy. Use the trained model to predict the test set and obtain the probability distribution of each sample belonging to each category, thereby verifying the effectiveness and accuracy of the model.

[0025] Furthermore, the specific process of using the test set to evaluate the loss value and accuracy of the model is as follows: Classify the test set data according to different cloud types and obtain their respective subsets; Use the trained model to predict the test samples of each cloud type and obtain the prediction results; The loss value is calculated for the entire test set, and the formula is as follows: in, is the total number of samples in the test set, The model is The loss value between the prediction of samples and the true label, is the actual label of the i-th sample, is the probability that the model predicts the positive class; Calculate the accuracy of the entire test set, the formula is as follows: in, is the accuracy of the entire test set, is the number of times the predicted result matches the actual true result, is the total number of examples in the test set; that is, the proportion of examples for which the model made correct predictions.

[0026] Traverse the subset of each cloud type and calculate the accuracy of the model on each cloud type. The formula is as follows: in, is the accuracy of the model on each cloud type, are different cloud types, Indicates the cloud type The number of samples predicted correctly is Is of cloud type The total number of samples, Is the cloud type to understand the model's ability to identify different cloud types; Combining the loss value of the entire test set, the overall accuracy, and the accuracy of each cloud type, we get a comprehensive evaluation index, as follows: in, is the weighted average of the loss value of the entire test set, the overall accuracy, and the accuracy of each cloud type. They are all weight parameters, and the values ​​of weight parameters are between (0,1); they can be weighted average accuracy or other forms of comprehensive evaluation, taking into account the performance of the model on the overall data set and different subsets. By reasonably setting the weight parameters, the performance of the model can be more comprehensively evaluated, providing an important reference for further model optimization and improvement; according to the calculated comprehensive evaluation indicators, the parameters in the model can be reasonably adjusted.

[0027] Furthermore, the process of calculating the risk index of wind turbine equipment is as follows: The foundation cloud map characteristics and meteorological characteristics are input into the preset wind turbine equipment operation risk prediction model. The formula of the wind turbine equipment operation risk prediction model is as follows: Among them, R is the risk index, is the activation function, ensuring that the output range is between [0, 1], is a weight matrix used to map input features to output risk index, is the bias term, Input features include ground-based cloud map features and meteorological features; The risk index R is compared with the preset risk threshold Compare and make drone inspection decisions: R> , indicating that the equipment is at high risk and requires urgent inspection; R≤ , indicating that the equipment risk is low and no inspection is required for the time being.

[0028] Preferably, it also includes: When the risk index exceeds the preset risk threshold, multiple drones will conduct inspections in coordination; Based on the geographical location of the wind farm, the mission is assigned to the drone closest to the target wind turbine; If multiple wind turbines are at risk, wind turbines with high risk indexes shall be inspected first; The inspection tasks of drones are scheduled and optimized. The specific task allocation formula is as follows: in, is assigned to The inspection mission of the drone, It is The first drone The distance between the fans, It is the biggest limitation of the drone’s flight range. It is The risk index of a wind turbine.

[0029] Another embodiment of the present invention provides a UAV-coordinated wind farm intelligent inspection system, which is used to execute the UAV-coordinated wind farm intelligent inspection method, and the system includes: A receiving module, used for receiving ground-based cloud images and meteorological characteristic data of wind farms in the target area; A preprocessing module is used to preprocess the ground-based cloud image to obtain a ground-based cloud image, and to perform segmentation, feature extraction and recognition processing on the ground-based cloud image to obtain ground-based cloud features; A feature extraction module is used to extract the features of meteorological feature data to obtain meteorological features; The calculation module is used to input the ground-based cloud map characteristics and meteorological characteristics into the preset wind turbine equipment operation risk prediction model, calculate the risk index of the wind turbine equipment, and control the drone to conduct inspections based on the risk index.

[0030] In order to ensure the safety and reliability of the device, the following improvements can be made based on the virtual device: Energy storage optimization: According to the prediction results, the charging and discharging strategies of the photovoltaic energy storage system are intelligently adjusted to optimize energy utilization efficiency; grid demand and electricity price fluctuations are taken into account to maximize economic benefits.

[0031] Data security, ensuring the security of data transmission and storage, preventing data leakage and tampering; implementing encryption measures and identity authentication mechanisms to protect the system from cyber attacks; System redundancy: design redundant mechanisms to ensure that the system can still operate normally in the event of sensor failure or data loss; perform regular system maintenance and updates to ensure long-term stable operation.

[0032] Through comprehensive consideration of the above multiple angles, it is possible to design an efficient and reliable UAV-coordinated intelligent inspection method and device for wind farms, provide accurate wind turbine equipment risk prediction for wind farms, and optimize energy management and utilization.

[0033] Another embodiment of the present invention provides a computing device for a wind farm intelligent inspection method coordinated by a drone, including: One or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the one or more programs include instructions for executing any of the methods described.

[0034] Although the present invention is disclosed as above, the protection scope of the present invention is not limited thereto. Those skilled in the art may make various changes and modifications without departing from the spirit and scope of the present invention, and these changes and modifications will fall within the protection scope of the present invention.

Claims

1. A wind farm intelligent inspection method coordinated by drones, characterized in that: The method comprises: Receive ground-based cloud images and meteorological characteristic data of wind farms in the target area; Preprocess the ground-based cloud image to obtain a ground-based cloud image, segment the ground-based cloud image, extract features and perform recognition processing to obtain ground-based cloud features; Extracting features of meteorological feature data to obtain meteorological features; The ground-based cloud map features and meteorological features are input into the preset wind turbine equipment operation risk prediction model, the risk index of the wind turbine equipment is calculated, and the drones are controlled to conduct collaborative inspections based on the risk index.

2. The UAV-coordinated wind farm intelligent inspection method according to claim 1 is characterized in that: The process of preprocessing the ground-based cloud image to obtain the ground-based cloud image is as follows: Read the ground cloud image and convert it from RGB to grayscale image; Perform histogram equalization on the grayscale image, use the edge detection algorithm to detect the edge of the grayscale image, and determine whether to apply different Gaussian blur parameters for denoising based on the average value of the edge detection result; Output ground-based cloud image.

3. The UAV-coordinated wind farm intelligent inspection method according to claim 1 is characterized in that: The process of segmenting the ground-based cloud image is as follows: The ground-based cloud image is segmented. The specific steps include: Use Otsu's method to perform global thresholding on the ground-based cloud image and invert the result; Use the opening operation to remove noise, then expand the background area through the dilation operation, apply the distance transform to highlight the foreground object, and use the thresholding to extract the foreground area of ​​the ground-based cloud image. The specific process is as follows: The boundary of the foreground area of ​​the ground-based cloud image is shrunk inward by the erosion operation to eliminate small noise points or small foreground objects. The foreground area in the ground-based cloud image is expanded by the dilation operation to replace each pixel with the maximum value of its surrounding area, connect the separated parts in the erosion operation, and maintain the important image structure features of the ground-based cloud image. By calculating the distance from each pixel to the nearest background pixel, the boundaries and features of the foreground area are highlighted, so that the pixel value of the foreground area is adjusted according to the distance from the nearest background pixel, thereby enhancing the edges and features of the foreground area. The distance transformed image is converted into a binary image according to the set threshold, and the foreground area is separated from the background; The foreground area is converted into an 8-bit image, and the unknown area of ​​the ground-based cloud image is determined by subtraction operation; Mark the foreground area, add 1 to all marks, and mark the unknown area as 0; The watershed algorithm is used to segment the image, and the segmentation results are merged to obtain the segmented ground cloud image.

4. The UAV-coordinated wind farm intelligent inspection method according to claim 1 is characterized in that: The process of feature extraction of the segmented ground cloud image is as follows: Extracting various ground-based cloud image features from an area of ​​the segmented ground-based cloud image, wherein the ground-based cloud image features include regional features, texture features, color features, deep learning features, and multi-scale features; Traverse all regions, extract the ground-based cloud map features of each region, and add the ground-based cloud map features of each region to the feature list; Convert a list of features to a NumPy array.

5. The UAV-coordinated wind farm intelligent inspection method according to claim 1 is characterized in that: The process of identifying the ground-based cloud image after feature extraction includes: Define a feature array and a label array, wherein the feature array includes a NumPy array of various features extracted from the ground-based cloud image, and the label array includes a NumPy array of cloud type labels corresponding to each ground-based cloud image; Encode the cloud type label into a number to obtain an encoded digital label array, where the number of label categories is n, where n is a natural number greater than 0; The data set is divided into a training set and a test set, where the test set accounts for 20% of the total data set; Build a sequential model and add a fully connected layer with 256 neurons and ReLU activation function, a dropout layer, and an output layer with n neurons and softmax activation function to obtain a deep learning model; Use the adam optimizer to compile the model, and use the training set data to train the model. Use the test machine for verification during the training process, and use the test set to evaluate the model's loss value and accuracy. Use the trained model to predict the test set and obtain the probability distribution of each sample belonging to each category.

6. The UAV-coordinated wind farm intelligent inspection method according to claim 1 is characterized in that: The specific process of using the test set to evaluate the loss value and accuracy of the model is as follows: Classify the test set data according to different cloud types and obtain their respective subsets; Use the trained model to predict the test samples of each cloud type and obtain the prediction results; The loss value is calculated for the entire test set, and the formula is as follows: in, is the total number of samples in the test set, The model is The loss value between the prediction of samples and the true label, is the actual label of the i-th sample, is the probability that the model predicts the positive class; Calculate the accuracy of the entire test set, the formula is as follows: in, is the accuracy of the entire test set, is the number of times the predicted result matches the actual true result, is the total number of samples in the test set; Traverse the subset of each cloud type and calculate the accuracy of the model on each cloud type. The formula is as follows: in, is the accuracy of the model on each cloud type, are different cloud types, Indicates the cloud type The number of samples predicted correctly is Is of cloud type The total number of samples, Is the cloud type A subset of Combining the loss value of the entire test set, the overall accuracy, and the accuracy of each cloud type, we get a comprehensive evaluation index, as follows: in, is the weighted average of the loss value of the entire test set, the overall accuracy, and the accuracy of each cloud type. are all weight parameters.

7. The UAV-coordinated wind farm intelligent inspection method according to claim 1 is characterized in that: The process of calculating the risk index of wind turbine equipment is as follows: The foundation cloud map characteristics and meteorological characteristics are input into the preset wind turbine equipment operation risk prediction model. The formula of the wind turbine equipment operation risk prediction model is as follows: Among them, R is the risk index, is the activation function, ensuring that the output range is between [0, 1], is a weight matrix used to map input features to output risk index, is the bias term, Input features include ground-based cloud map features and meteorological features; The risk index R is compared with the preset risk threshold Compare and make drone inspection decisions: R> , indicating that the equipment is at high risk and requires urgent inspection; R≤ , indicating that the equipment risk is low and no inspection is required for the time being.

8. The UAV-coordinated wind farm intelligent inspection method according to claim 7 is characterized in that: Also includes: When the risk index exceeds the preset risk threshold, multiple drones will conduct inspections in coordination; Based on the geographical location of the wind farm, the mission is assigned to the drone closest to the target wind turbine; If multiple wind turbines are at risk, wind turbines with high risk indexes shall be inspected first; The inspection tasks of drones are scheduled and optimized. The specific task allocation formula is as follows: in, is assigned to The inspection mission of the drone, It is The first drone The distance between the fans, It is the biggest limitation of the drone’s flight range. It is The risk index of a wind turbine.

9. A UAV-coordinated wind farm intelligent inspection system, used to execute the UAV-coordinated wind farm intelligent inspection method according to any one of claims 1 to 8, characterized in that: The system comprises: A receiving module, used for receiving ground-based cloud images and meteorological characteristic data of wind farms in the target area; A preprocessing module is used to preprocess the ground-based cloud image to obtain a ground-based cloud image, and to perform segmentation, feature extraction and recognition processing on the ground-based cloud image to obtain ground-based cloud features; A feature extraction module is used to extract the features of meteorological feature data to obtain meteorological features; The calculation module is used to input the ground-based cloud map characteristics and meteorological characteristics into the preset wind turbine equipment operation risk prediction model, calculate the risk index of the wind turbine equipment, and control the drone to conduct inspections based on the risk index.

10. A computing device for a wind farm intelligent inspection method coordinated by drones, characterized in that: include: One or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the one or more programs include instructions for executing any of the methods according to claims 1-8.

Images