Wind power generation rotor pneumatic unbalance detection method and system based on video analysis

Abstract

The invention relates to the technical field of wind power generators, in particular to a wind power generation rotor pneumatic unbalance detection method and system based on video analysis, and the method comprises the steps: obtaining the operation video data of a wind power generator rotor from multiple angles; performing frame extraction on the operation video data of the wind driven generator rotor, and performing feature extraction on the extracted key frame to obtain a plurality of feature data of the wind driven generator rotor; and constructing a rotor pneumatic unbalance state prediction model based on a random forest algorithm, and after training, inputting a plurality of feature data of the wind driven generator rotor into the rotor pneumatic unbalance state prediction model for calculation to obtain a rotor pneumatic unbalance prediction state. The characteristics of the wind driven generator rotor are extracted by using the video data, the pneumatic unbalance state is accurately predicted through the random forest algorithm, the monitoring real-time performance and accuracy are improved, the fault occurrence rate is reduced, and the operation efficiency and reliability of the wind driven generator are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of wind power generators, in particular to a wind power rotor aerodynamic imbalance detection method and system based on video analysis. BACKGROUND

[0002] As a renewable energy source, wind power is increasingly valued, especially in addressing global climate change and energy transition. The performance of wind power generator rotors directly affects the efficiency and reliability of wind power generation. The aerodynamic imbalance of the rotor not only leads to reduced power generation efficiency, but also may cause mechanical failure, thereby affecting the safe operation of the wind power generator. Although existing monitoring technologies play an important role in the maintenance of wind power generators, they often lack real-time and accuracy. Many traditional methods rely on periodic manual inspection, which cannot timely detect potential problems of the rotor. In addition, the monitoring method based on vibration sensors may be disturbed by noise in complex environments, resulting in reduced accuracy and difficulty in accurately determining the aerodynamic imbalance state. In the process of using data collection, the delay in data transmission and processing may also cause a delayed response to problems. In addition, the current monitoring data analysis of the rotor relies more on experience and knowledge, which is easily affected by subjective judgment, leading to misjudgment or missed judgment, increasing the maintenance cost of the wind power generator and affecting its overall operating efficiency. SUMMARY

[0003] The present application aims to provide a wind power rotor aerodynamic imbalance detection method and system based on video analysis, which extracts features of the wind power generator rotor using video data, accurately predicts the aerodynamic imbalance state through a random forest algorithm, improves monitoring real-time and accuracy, reduces the failure rate, and improves the operating efficiency and reliability of the wind power generator.

[0004] The present application is achieved by the following technical solutions:

[0005] The wind power rotor aerodynamic imbalance detection method based on video analysis includes the following steps:

[0006] Obtaining running video data of the wind power generator rotor from multiple angles;

[0007] Frame extraction is performed on the wind power generator rotor running video data, and feature extraction is performed on the extracted key frames to obtain multiple feature data of the wind power generator rotor;

[0008] A rotor aerodynamic imbalance state prediction model is constructed based on a random forest algorithm, and after training, the multiple feature data of the wind power generator rotor are input into the rotor aerodynamic imbalance state prediction model for calculation to obtain the rotor aerodynamic imbalance prediction state.

[0009] Optionally, the multiple-angle acquisition of the wind turbine rotor operation video data is specifically as follows:

[0010] A camera that meets both the set resolution and the set frame rate is selected.

[0011] A plurality of shooting positions are selected, and corresponding shooting angles are selected for the shooting positions to alternately collect the wind turbine rotor operation video data.

[0012] The camera is configured with a stabilizer and an environment sensor to prevent video shaking and to obtain environment data of the wind turbine rotor.

[0013] Optionally, the method further comprises data processing of the wind turbine rotor operation video data, which is specifically as follows:

[0014] The wind turbine rotor operation video data is format-converted.

[0015] The format-converted wind turbine rotor operation video data is frame-extracted.

[0016] The extracted frames are denoised and feature-extracted to obtain a plurality of feature data of the wind turbine rotor.

[0017] Optionally, the plurality of feature data of the wind turbine rotor includes motion features of the wind turbine rotor, image features of the wind turbine rotor, and vibration features of the wind turbine rotor.

[0018] Optionally, the motion features of the wind turbine rotor are extracted as follows:

[0019] Adjacent frames are extracted from the obtained wind turbine rotor operation video data, and the pixel changes between the adjacent frames are obtained by an optical flow method.

[0020] The optical flow between the adjacent frames is calculated to obtain an optical flow vector of each pixel, and the optical flow vector includes an optical flow amplitude and a direction.

[0021] The calculated optical flow vector is extracted to obtain motion trajectory features and average speed features, specifically: the motion trajectory features are obtained by tracking and connecting the optical flow vector; and the average speed features are obtained by averaging all optical flow amplitudes.

[0022] The motion trajectory features and the average speed features are integrated to represent the motion features of the wind turbine rotor.

[0023] Optionally, the image features of the wind turbine rotor are extracted as follows:

[0024] An edge feature of the wind turbine rotor is obtained by applying a Canny edge detection algorithm to process a key frame image of the wind turbine rotor.

[0025] A shape feature of the wind turbine rotor is extracted by calculating a histogram of oriented gradients of the edge feature.

[0026] The edge feature and the shape feature of the wind turbine rotor image are combined to represent the image feature of the wind turbine rotor.

[0027] Optionally, the vibration feature of the wind turbine rotor is extracted by:

[0028] Acceleration data of the wind turbine rotor is obtained based on a vibration sensor.

[0029] An RMS value is calculated based on the acceleration data of the wind turbine rotor.

[0030] The acceleration data of the wind turbine rotor is subjected to Fourier transform to convert the original time-domain signal into a frequency-domain signal, and a main frequency component is extracted to identify the vibration feature of the wind turbine rotor.

[0031] Optionally, the rotor aerodynamic imbalance state prediction model is constructed based on a random forest algorithm, and the construction process is as follows:

[0032] The motion feature, the image feature, and the vibration feature of the wind turbine rotor are integrated to form a data set, and a target variable is defined as an aerodynamic imbalance state.

[0033] The data set is divided into a training set and a test set, a rotor aerodynamic imbalance state prediction model is constructed based on a random forest algorithm, the rotor aerodynamic imbalance state prediction model is trained by using the training set, the trained rotor aerodynamic imbalance state prediction model is verified by using the test set, and an accuracy of the rotor aerodynamic imbalance state prediction model is calculated.

[0034] The rotor aerodynamic imbalance state prediction model is subjected to hyperparameter tuning based on the accuracy until the training of the rotor aerodynamic imbalance state prediction model is completed.

[0035] Optionally, the accuracy is calculated according to the following formula:

[0036]

[0037] wherein A represents the accuracy, N represents the number of samples, i represents an index, y i represents a true class of the sample, and y

[0038] The wind turbine rotor aerodynamic imbalance detection system based on video analysis comprises:

[0039] The collection unit acquires the operation video data of the wind turbine rotor at multiple angles.

[0040] The extraction unit extracts frames from the wind turbine rotor operation video data and extracts features from the extracted key frames to obtain multiple feature data of the wind turbine rotor.

[0041] The prediction unit constructs a rotor aerodynamic imbalance state prediction model based on a random forest algorithm, and after training, inputs the multiple feature data of the wind turbine rotor into the rotor aerodynamic imbalance state prediction model for calculation to obtain a rotor aerodynamic imbalance prediction state.

[0042] The technical scheme of the present application has at least the following advantages and beneficial effects:

[0043] The present application acquires the operation video data of the wind turbine rotor at multiple angles, performs frame extraction and feature analysis, thereby realizing real-time prediction of the rotor aerodynamic imbalance state, not only reducing the cost of relying on traditional sensors and avoiding the defect that they are easily disturbed, but also ensuring the accuracy and stability of the data; moreover, the rotor aerodynamic imbalance state prediction model constructed based on the random forest algorithm can efficiently learn from massive feature data to improve the accuracy and robustness of the prediction. In addition, real-time monitoring and prediction not only improve the operation efficiency of the wind turbine, but also help to discover potential problems in advance, reduce maintenance cost and downtime risk, and ultimately improve the overall power generation benefit. BRIEF DESCRIPTION OF DRAWINGS

[0044] Figure 1 The flowchart of the wind turbine rotor aerodynamic imbalance detection method based on video analysis provided by the present application is shown.

[0045] Figure 2 The extraction process diagram of the wind turbine rotor motion feature provided by the present application is shown.

[0046] Figure 3 The principle diagram of the wind turbine rotor aerodynamic imbalance detection system based on video analysis provided by the present application is shown. DETAILED DESCRIPTION

[0047] To make the purpose, technical scheme and advantages of the present application clearer, the technical scheme in the present application will be described clearly and completely below in combination with the drawings in the present application. Obviously, what is described is part of the present application, not the whole. The components of the present application described and shown in the drawings here can be arranged and designed in various different configurations.

[0048] As Figure 1As shown, the present application provides one embodiment: a wind turbine rotor aerodynamic imbalance detection method based on video analysis, the steps of which include:

[0049] Multi-angle acquisition of wind turbine rotor operation video data;

[0050] Frame extraction is performed on the wind turbine rotor operation video data, and feature extraction is performed on the extracted key frames to obtain multiple feature data of the wind turbine rotor;

[0051] A rotor aerodynamic imbalance state prediction model is constructed based on a random forest algorithm, and after training, the multiple feature data of the wind turbine rotor are input into the rotor aerodynamic imbalance state prediction model for calculation to obtain the rotor aerodynamic imbalance prediction state.

[0052] In this embodiment, in order to improve the accuracy and efficiency of the operation state monitoring and aerodynamic imbalance prediction of the wind turbine rotor, multiple cameras are arranged to acquire high-definition video data of the wind turbine rotor operation from different angles, mainly through the rich visual information of the video data, to comprehensively capture the dynamic characteristics of the rotor. Frame extraction is performed on the acquired operation video data, key frames are selected to reduce the computational burden, and important information in the rotor motion is focused on. The extracted key frames are then subjected to feature extraction, and multiple feature data of the wind turbine rotor are extracted, including motion features, image features, and vibration features, etc. On this basis, the present embodiment adopts a random forest algorithm to construct a rotor aerodynamic imbalance state prediction model, and the random forest integrates multiple decision trees to improve the prediction performance and model robustness. In the model training stage, the extracted feature data of the wind turbine rotor are input into the prediction model, so that the complex relationship between the multiple feature data of the wind turbine rotor and the aerodynamic imbalance state can be learned. After sufficient training, the rotor aerodynamic imbalance state prediction model can quickly and accurately calculate the input feature data, thereby outputting the aerodynamic imbalance prediction state of the wind turbine rotor.

[0053] In the above specific implementation, the multi-angle acquisition of wind turbine rotor operation video data is specifically:

[0054] Select a camera that simultaneously meets the set resolution and set frame rate;

[0055] Select multiple shooting positions and select corresponding shooting angles for the shooting positions to alternately collect wind turbine rotor operation video data;

[0056] Among them, the camera is configured with a stabilizer and an environment sensor to prevent video shaking and acquire environment data of the wind turbine rotor.

[0057] In implementation, the embodiment selects a high-performance camera that can simultaneously achieve the set resolution and the set frame rate to ensure that the captured video data has sufficient clarity and smoothness. The set resolution should effectively present the details of the wind turbine rotor, and the set frame rate should meet the real-time monitoring requirements for the dynamic movement of the rotor. Multiple shooting positions are selected to comprehensively cover the operation of the wind turbine rotor. Each shooting position will select a corresponding shooting angle according to the operation characteristics of the rotor, and alternately collect the operation video of the wind turbine rotor. By obtaining data from multiple angles, a more accurate model can be established to capture subtle changes in the rotor movement process and possible abnormal states. Further, in order to optimize the video quality, the selected camera also needs to be equipped with a stabilizer to prevent video shaking caused by wind influence or other environmental factors. The stabilizer can ensure the stability of the video recording process, making it easier to extract clear feature information during subsequent analysis. The camera is also equipped with environmental sensors to obtain real-time data of the environment in which the wind turbine rotor is located, including wind speed, temperature, humidity, etc. The combination of environmental data and video data provides multi-dimensional information support for subsequent analysis.

[0058] Further, it also includes data processing of the wind turbine rotor operation video data, which specifically comprises:

[0059] Converting the format of the wind turbine rotor operation video data;

[0060] Extracting frames from the wind turbine rotor operation video data after format conversion;

[0061] Denoising and feature extraction are performed on the extracted frames to obtain multiple feature data of the wind turbine rotor.

[0062] During the implementation of the embodiment, the obtained wind turbine rotor operation video data is converted in format to ensure that the data can be supported by subsequent processing software. Format conversion involves converting video data from one file format (such as AVI, MOV) to another format (such as MP4) that is more suitable for analysis, to improve the compatibility and efficiency of data processing, and to reduce the file size on the basis of ensuring video quality, thereby speeding up the subsequent processing speed. After completing the format conversion, key frames are identified and intercepted by analyzing the video stream to extract representative images in each frame. The frequency and interval of frame extraction are set according to specific requirements, usually selecting several frames per second. The extracted frame data is usually affected by various noises, such as environmental noise and motion blur. In order to improve the accuracy of subsequent feature extraction, various image denoising techniques are used when denoising the extracted frames, such as mean filtering and median filtering. Through denoising processing, the image quality can be significantly improved to provide a clearer and more accurate image basis for feature extraction.

[0063] As Figure 2 shown in the specific application of the present embodiment, the plurality of feature data of the wind turbine rotor includes: motion features of the wind turbine rotor, image features of the wind turbine rotor, and vibration features of the wind turbine rotor.

[0064] The motion features of the wind turbine rotor, the specific extraction process is:

[0065] Extract adjacent frames from the obtained wind turbine rotor running video data, and obtain the pixel changes between adjacent frames by the optical flow method;

[0066] Calculate the optical flow between adjacent frames to obtain the optical flow vector of each pixel, and the optical flow vector includes the optical flow amplitude and direction;

[0067] Extract the calculated optical flow vector to obtain the motion trajectory feature and the average speed feature, specifically: the motion trajectory feature is obtained by tracking and connecting the optical flow vector; the average speed feature is obtained by averaging all optical flow amplitudes;

[0068] Integrate the motion trajectory feature and the average speed feature to represent the motion features of the wind turbine rotor.

[0069] Specifically, the present embodiment obtains the pixel changes between adjacent frames by the optical flow method. The optical flow method can estimate the movement of objects in images by comparing the changes of pixels in consecutive frames, so it is very effective when analyzing the motion state of the rotor. Specifically, set adjacent frames as I1 and I2, and calculate the optical flow vector v(x, y) between adjacent frames by the optical flow method, which contains the displacement information of each pixel over time. The optical flow vector can be expressed as: Where u(x, y) is the horizontal component of the optical flow, and v(x, y) is the vertical component of the optical flow. The optical flow amplitude |v(x, y)| and the direction θ(x, y) are calculated by the following formulas:

[0070]

[0071] After obtaining the optical flow amplitude and direction of each pixel, the next step is to extract the motion trajectory feature and the average speed feature. The motion trajectory feature is visualized by tracking the changes of the optical flow vector and connecting these change points to visualize the motion trajectory of the rotor. By analyzing the changes of the optical flow vector over time, the motion trajectory of the rotor can be plotted, thereby intuitively reflecting its dynamic characteristics. The average speed feature is obtained by averaging all optical flow amplitudes, that is: Where M is the total number of pixels involved in the calculation, and V avgThe average speed feature represents the average speed characteristics, can quickly reflect the overall rotor motion speed, and provides important information about the rotor running state. Finally, the motion trajectory feature and the average speed feature are integrated to form a comprehensive motion feature description for representing the motion state of the wind turbine rotor.

[0072] The image feature of the wind turbine rotor is extracted as follows:

[0073] The Canny edge detection algorithm is applied to process the key frame image of the wind turbine rotor to obtain the edge feature of the wind turbine rotor.

[0074] The direction gradient histogram of the edge feature is calculated to extract the shape feature of the wind turbine rotor.

[0075] The edge feature and the shape feature of the wind turbine rotor image are combined to represent the image feature of the wind turbine rotor.

[0076] Specifically, the Canny edge detection algorithm is applied to process the key frame image of the wind turbine rotor to extract the edge feature of the rotor. The steps are as follows: the key frame image is converted to a grayscale image to reduce data complexity while retaining important shape information. The grayscale image is denoted as I g (x, y), and a Gaussian filter is used to smooth the grayscale image to reduce noise effects, with the formula: where σ is the standard deviation of the Gaussian kernel. The gradient of the image is calculated by the Sobel operator to identify the edges. The horizontal and vertical components of the gradient are defined as: The gradient magnitude and direction are calculated by the following formula: A non-maximum suppression and double threshold strategy is applied to identify and connect the edge points, so as to finally obtain the edge feature of the wind turbine rotor, denoted as E(x, y). With the help of the edge feature, the direction gradient histogram is calculated to extract the shape feature of the wind turbine rotor. For each pixel point (x, y), its direction gradient can be obtained by the aforementioned formula. The HOG feature is formed by dividing the image into N x N small blocks and counting the gradient direction distribution in each small block: where K is the number of small blocks, and v k is the optical flow vector in the corresponding small block. After combining the Canny edge detection and the direction gradient histogram, the edge feature E(x, y) and the shape feature HOG of the wind turbine rotor are combined to generate a more comprehensive image feature representation, which significantly improves the representation ability of the shape and edge of the wind turbine rotor.

[0077] The vibration feature of the wind turbine rotor is extracted as follows:

[0078] Obtaining acceleration data of the wind turbine rotor based on a vibration sensor;

[0079] Solving an RMS value according to the wind turbine rotor acceleration data;

[0080] Performing Fourier transform on the wind turbine rotor acceleration data, converting the original time domain signal into a frequency domain signal, and extracting a main frequency component to identify the vibration characteristics of the wind turbine rotor.

[0081] Specifically, the vibration sensor in the embodiment is usually installed at a key position of the rotor, and can effectively capture the acceleration changes caused by rotor rotation, aerodynamic load, mechanical friction and the like. In the obtained acceleration data, the root mean square (RMS) value of the acceleration is calculated to quantify the intensity of the rotor vibration. The calculation formula of the RMS value is: wherein ai is the acceleration value at i time, and H is the total sampling point number of the acceleration data. The RMS value can effectively reflect the overall vibration level of the rotor, and provide a preliminary basis for judging the running state thereof. Fourier transform is performed on the wind turbine rotor acceleration data, converting the original time domain signal into a frequency domain signal. The formula of the Fourier transform is: wherein X(f) represents the representation in the frequency domain, and a(t) is the acceleration signal in the time domain. Through the Fourier transform, a complex time signal can be decomposed into the superposition of each frequency component, and the frequency information contained in the signal can be revealed. After completing the Fourier transform, the main frequency component is calculated and extracted from the output frequency domain signal. The main frequency component refers to the frequency component with the highest amplitude in the frequency domain, which is usually related to a specific vibration mode of the wind turbine rotor, and can effectively judge the vibration characteristics of the rotor.

[0082] In the embodiment, the rotor aerodynamic imbalance state prediction model is constructed based on the random forest algorithm, specifically:

[0083] The motion characteristics, image characteristics and vibration characteristics of the wind turbine rotor are integrated to form a data set, and the target variable is defined as the aerodynamic imbalance state;

[0084] The data set is divided into a training set and a test set, a rotor aerodynamic imbalance state prediction model is constructed based on the random forest algorithm, the rotor aerodynamic imbalance state prediction model is trained through the training set, the trained rotor aerodynamic imbalance state prediction model is verified through the test set, and the accuracy of the rotor aerodynamic imbalance state prediction model is calculated;

[0085] Based on the accuracy, the hyperparameters of the rotor aerodynamic imbalance state prediction model are optimized until the training of the rotor aerodynamic imbalance state prediction model is completed.

[0086] In the specific implementation of the present embodiment, the motion characteristics, image characteristics and vibration characteristics of the wind turbine rotor are integrated to form a comprehensive data set. The data set includes various representative characteristics, which can fully reflect the running state and aerodynamic characteristics of the rotor. In order to realize data integration, it is necessary to first standardize and normalize each characteristic to ensure that the characteristics are in the same dimension and avoid the adverse effects of different scale characteristics on model training. At the same time, the target variable is defined as the aerodynamic imbalance state, which indicates whether the rotor is in an aerodynamic imbalance state. After completing the integration of the data set, the data set is divided into a training set and a test set. Based on the integrated data set, a rotor aerodynamic imbalance state prediction model is constructed based on the random forest algorithm. Random forest is an ensemble learning method that constructs multiple decision trees and votes on their prediction results to obtain the final result. The training set is used to train the model, which optimizes the prediction ability of the aerodynamic imbalance state by statistically analyzing the results of each tree. After training, the trained rotor aerodynamic imbalance state prediction model is verified using the test set. The verification process includes inputting the test set into the model and comparing the accuracy between the model prediction result and the true value. The accuracy of the model is calculated by the following formula: Where A is the accuracy, N is the number of samples, i is the index, y i is the true class of the sample, is the predicted class of the sample. Based on the results of the first training and verification, the rotor aerodynamic imbalance state prediction model is optimized using the accuracy. Hyperparameters are parameters that need to be manually set before model training, such as the number of decision trees, the depth of the tree, the number of features used by each tree, etc. By using cross-validation techniques, the hyperparameters are repeatedly adjusted to achieve optimal performance on the training set, thereby improving the generalization ability on the test set. After continuous iteration and optimization, the training of the rotor aerodynamic imbalance state prediction model is completed, which can provide strong data support for the operation and maintenance of wind turbines.

[0087] As an application, after the training of the rotor aerodynamic imbalance state prediction model is completed, the real-time video data of the wind turbine rotor is processed by the present embodiment to obtain the corresponding feature data. The model will generate a prediction result based on the input feature vector, and the result is a classification label of the aerodynamic imbalance state, 0 indicating a normal state and 1 indicating an imbalance state, indicating the possibility of the rotor being in an imbalance state. If the model warns that the rotor is in an imbalance state, the operation and maintenance team can promptly carry out inspection and maintenance based on this information to ensure the safe and stable operation of the equipment.

[0088] As Figure 3 shown, the wind turbine rotor aerodynamic imbalance detection system based on video analysis includes:

[0089] The collection unit acquires the operation video data of the wind turbine rotor at multiple angles;

[0090] The extraction unit extracts frames from the wind turbine rotor operation video data and extracts features from the extracted key frames to obtain multiple feature data of the wind turbine rotor;

[0091] The prediction unit constructs a rotor aerodynamic imbalance state prediction model based on a random forest algorithm, and after training, inputs the multiple feature data of the wind turbine rotor into the rotor aerodynamic imbalance state prediction model for calculation to obtain a rotor aerodynamic imbalance prediction state.

[0092] The above is only the preferred embodiment of the present application, and is not used to limit the present application. For those skilled in the art, the present application can have various modifications and changes. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for detecting aerodynamic imbalance of a wind turbine rotor based on video analysis, characterized in that: The steps of the method include: Acquire the operation video data of wind turbine rotors from multiple angles; Performing frame extraction on the video data of the wind turbine rotor operation, and performing feature extraction on the extracted key frames to obtain multiple feature data of the wind turbine rotor; A rotor aerodynamic imbalance state prediction model is constructed based on the random forest algorithm. After training, multiple characteristic data of the wind turbine rotor are input into the rotor aerodynamic imbalance state prediction model for calculation to obtain the rotor aerodynamic imbalance prediction state.

2. The method for detecting aerodynamic imbalance of a wind turbine rotor based on video analysis according to claim 1, characterized in that: The multi-angle acquisition of the wind turbine rotor operation video data is specifically as follows: Select a camera that can achieve both the set resolution and the set frame rate; Selecting multiple shooting positions and selecting corresponding shooting angles for the shooting positions to alternately collect operation video data of the wind turbine rotor; The camera is equipped with a stabilizer and an environmental sensor to prevent video shaking and obtain environmental data of the wind turbine rotor.

3. The method for detecting aerodynamic imbalance of a wind turbine rotor based on video analysis according to claim 2, characterized in that: It also includes data processing of the wind turbine rotor operation video data, which is specifically: Convert the format of wind turbine rotor operation video data; Perform frame extraction on the format-converted wind turbine rotor operation video data; The extracted frames are subjected to denoising and feature extraction to obtain multiple feature data of the wind turbine rotor.

4. The method for detecting aerodynamic imbalance of a wind turbine rotor based on video analysis according to claim 3, characterized in that: The plurality of characteristic data of the wind turbine rotor include: motion characteristics of the wind turbine rotor, image characteristics of the wind turbine rotor, and vibration characteristics of the wind turbine rotor.

5. The method for detecting aerodynamic imbalance of a wind turbine rotor based on video analysis according to claim 4, characterized in that: The specific extraction process of the motion characteristics of the wind turbine rotor is as follows: Adjacent frames are extracted from the acquired video data of wind turbine rotor operation, and the pixel changes between adjacent frames are obtained by optical flow method; Calculate the optical flow between adjacent frames to obtain the optical flow vector of each pixel, which includes the optical flow amplitude and direction; The calculated optical flow vectors are extracted to obtain the motion trajectory characteristics and average speed characteristics of the rotor. Specifically, the motion trajectory characteristics are obtained by tracking and connecting the optical flow vectors; average The speed feature is obtained by averaging all optical flow amplitudes; The motion trajectory characteristics and average speed characteristics are integrated to represent the motion characteristics of the wind turbine rotor.

6. The method for detecting aerodynamic imbalance of a wind turbine rotor based on video analysis according to claim 4, characterized in that: The specific extraction process of the image features of the wind turbine rotor is as follows: Apply the Canny edge detection algorithm to process the key frame image of the wind turbine rotor to obtain the edge features of the wind turbine rotor; Calculate the directional gradient histogram of edge features to extract the shape features of the wind turbine rotor; The edge features and shape features of the wind turbine rotor image are combined to characterize the image features of the wind turbine rotor.

7. The method for detecting aerodynamic imbalance of a wind turbine rotor based on video analysis according to claim 4, characterized in that: The specific extraction process of the vibration characteristics of the wind turbine rotor is as follows: Acquire acceleration data of the wind turbine rotor based on a vibration sensor; Determine the RMS value based on the wind turbine rotor acceleration data; Fourier transform is performed on the wind turbine rotor acceleration data to convert the original time domain signal into a frequency domain signal, and the main frequency component is extracted to identify the vibration characteristics of the wind turbine rotor.

8. The method for detecting aerodynamic imbalance of a wind turbine rotor based on video analysis according to any one of claims 4 to 7, characterized in that: The rotor aerodynamic imbalance state prediction model is constructed based on the random forest algorithm, which is specifically as follows: Integrating the motion characteristics, image characteristics, and vibration characteristics of the wind turbine rotor to form a data set, and defining the target variable as an aerodynamic imbalance state; The data set is divided into a training set and a test set. A rotor aerodynamic imbalance state prediction model is constructed based on the random forest algorithm. The rotor aerodynamic imbalance state prediction model is trained using the training set and verified using the test set. The accuracy of the rotor aerodynamic imbalance state prediction model is also calculated. Hyperparameter tuning is performed on the rotor aerodynamic imbalance state prediction model based on the accuracy until the training of the rotor aerodynamic imbalance state prediction model is completed.

9. The method for detecting aerodynamic imbalance of a wind turbine rotor based on video analysis according to claim 8, characterized in that: The accuracy rate is calculated as follows: Among them, A is the accuracy, N is the number of samples, i is the index, y i is the true category of the sample, is the predicted category of the sample.

10. A wind turbine rotor aerodynamic imbalance detection system based on video analysis, characterized in that: include: The acquisition unit acquires the operating video data of the wind turbine rotor from multiple angles; An extraction unit extracts frames from the video data of the wind turbine rotor operation and extracts features from the extracted key frames to obtain a plurality of feature data of the wind turbine rotor; The prediction unit constructs a rotor aerodynamic imbalance state prediction model based on the random forest algorithm. After training, multiple characteristic data of the wind turbine rotor are input into the rotor aerodynamic imbalance state prediction model for calculation to obtain the rotor aerodynamic imbalance prediction state.

Images