Wind turbine generator blade health state assessment method

Abstract

The invention relates to the technical field of wind power, in particular to a wind turbine generator blade health state assessment method. Comprising the following steps: generating a first-class RGB thermodynamic image of wind speed power operation according to real-time time sequence data of a wind turbine generator SCADA system, identifying and marking abnormal data of the wind turbine generator from the first-class RGB thermodynamic image, and generating a second-class RGB thermodynamic image of blade key fault features according to historical time sequence data of the wind turbine generator SCADA system, and performing deep learning on the second type of RGB thermodynamic images to construct a health state evaluation model for image recognition, inputting the first type of RGB thermodynamic images into the health state evaluation model, and outputting an evaluation result. According to the method, the whole abnormal data recognition process is visualized, the method does not depend on human experience such as reference images, the health state evaluation model is constructed based on deep learning, hidden information in SCADA data is fully mined and analyzed, and accurate evaluation of the health state of the fan blade is achieved.

Description

Technical Field

[0001] The invention relates to the technical field of wind power generation, and in particular to a method for evaluating the health status of blades of a wind turbine generator set. Background Art

[0002] In order to cope with the energy crisis, countries around the world have begun to pay attention to the development of renewable energy. With the vigorous development of green energy in recent years, wind power generation has become the fastest growing renewable energy in the world. Among them, the operation and maintenance of wind turbines has become an issue that must be faced. Large-scale wind farms are usually distributed in remote mountainous areas, uninhabited plains or offshore areas. The climate in these places is complex and changeable, and the working environment of wind turbine components is relatively complex, which is prone to various faults. Blade failure is one of the most serious faults of wind turbines. Blade strain can cause problems such as abnormal gearbox speed and excessive cabin temperature. If the fault is still not found after the blade breaks, the wind turbine will suffer serious damage. Therefore, it is increasingly important to assess the health status of wind turbine blades.

[0003] At present, the fault status detection of wind turbine blades usually adopts the method of manual regular inspection, which will cause the detection results to have a certain lag. Therefore, at present, the various parameters of wind turbine operation are monitored by installing a supervisory control and data acquisition (SCADA) system in the wind farm. When the blade crack develops rapidly, some parameters recorded in the SCADA system can indirectly reflect the fault. The blade health status can be evaluated through data mining, and an early warning can be issued before the blade breaks. There are two evaluation methods: one is to use specific sensor signals such as sound to evaluate the health status of wind turbine blades, but this type of method has poor versatility; another method is to use artificial intelligence technology to evaluate the health status of wind turbine blades based on SCADA data, which involves autoencoders, deep belief networks, etc. However, due to the complex working environment of wind turbines, the data recorded by sensors such as anemometers and thermometers may be biased, and the noise in data transmission will cause the parameters such as power and power generation recorded by the SCADA system to fluctuate. Conventional computing tools cannot process a large amount of high-dimensional SCADA data in a short time, and the abnormal points in the SCADA data will have a great impact on the health status assessment of wind turbine blades. Therefore, it is of great research significance to quickly and accurately identify abnormal data in SCADA data.

[0004] At present, the wind turbine SCADA abnormal data recognition methods are mainly divided into three categories: recognition methods based on wind speed power curve model, recognition methods based on neural network, and recognition methods based on data image conversion and image processing. The recognition method based on wind speed power curve model is the most widely used. This method has a clear physical basis and is excellent in identifying abnormal noise data, but it requires sufficient data volume and has weak recognition ability for power-limited abnormal data: the recognition method based on neural network can identify a variety of abnormal data and process complex distributed data, but the recognition accuracy is related to the proportion of abnormal data and requires sufficient data volume; the image processing-based method is applied to the field of SCADA abnormal data recognition, which can accurately separate the normal operating range and has high efficiency, but it relies on prior knowledge such as reference images, the classification accuracy of abnormal data is low, and the research on this method is less and lacks universality. Summary of the invention

[0005] The technical problem to be solved by the embodiments of the present invention is to provide a method for evaluating the health status of wind turbine blades to solve the problem of inaccurate health status evaluation of wind turbine blades caused by low accuracy in identifying abnormal data of wind turbine SCADA in the prior art.

[0006] The present invention discloses a method for evaluating the health status of blades of a wind turbine generator set, comprising: Acquire real-time time series data of the wind turbine SCADA system, extract and transform the real-time time series data to generate a first-class RGB thermal image of wind speed and power operation; According to the first type of RGB thermal images of wind speed and power operation, identifying and marking abnormal data of the wind turbine generator set from the first type of RGB thermal images through image processing; Acquire historical time series data of the wind turbine SCADA system, and extract and transform the historical time series data to generate a second type of RGB thermal image of key fault features of the blade; According to the second type of RGB thermal images of key fault features of the blade, the image set of the second type of RGB thermal images is divided into a training set and a test set, and a health status assessment model for image recognition is constructed through deep learning of the training set and the test set; The first type of RGB thermal images marked with abnormal data are input into the health status assessment model, and an assessment result of the health status of the wind turbine blades is obtained by output.

[0007] Optionally, the method for extracting and converting the first type of RGB thermal image of wind speed and power operation according to the real-time time series data includes: Extracting wind speed / power data based on the real-time time series data obtained from the wind turbine SCADA system; Performing Fourier transform on the wind speed / power data and extracting frequency characteristics of the wind speed / power data; Based on the thermal map generation method, the frequency characteristics of the wind speed / power data are mapped one-to-one to each pixel in the RGB color space, and the RGB color values ​​of all pixels are combined to generate the first type of RGB thermal image.

[0008] Optionally, the method of mapping the frequency characteristics of the wind speed / power data to each pixel in the RGB color space in a one-to-one correspondence includes: According to the frequency characteristics of the extracted wind speed / power data, the data are divided into wind speed frequency characteristics and power frequency characteristics; Mapping the plurality of wind speed frequency features one by one to the color depth of the red channel in the RGB color space, and the higher the wind speed frequency feature, the deeper the red color of the mapped pixel point; The multiple power frequency features are mapped one-to-one to the color depth of the green channel in the RGB color space, and the higher the power frequency feature is, the deeper the green of the mapped pixel point is.

[0009] Optionally, the method for identifying and marking abnormal data of a wind turbine from the first type of RGB thermal images by image processing includes: Defining a shutdown condition associated with the wind speed / power data, identifying all pixel points satisfying the shutdown condition from the first type of RGB thermal image, and marking them as shutdown abnormality data; Convert the first type of RGB thermal image into an HSV color space, define a range representing noise anomaly, identify all pixels within the range from the HSV color space, and mark them as noise anomaly data; The first type of RGB thermal image is converted into a grayscale image, horizontally distributed straight lines are extracted from the grayscale image based on edge detection and Hough transform, all pixel points horizontally distributed between each group of straight lines are identified from the grayscale image, and marked as power restriction abnormal data.

[0010] Optionally, the method of defining the shutdown condition associated with the wind speed / power data, identifying all pixels satisfying the shutdown condition from the first type of RGB thermal image, and marking them as shutdown abnormality data includes: According to the frequency characteristics of the wind speed / power data, a frequency characteristic that the wind speed is greater than the cut-in wind speed and the power is less than or equal to 0 is defined as a shutdown condition; Each pixel point in the first type of RGB thermal image is traversed, and the pixel points corresponding to the wind speed / power data that meet the shutdown conditions are marked by distinguishing colors.

[0011] Optionally, the method of converting the first type of RGB thermal image into an HSV color space, defining a range representing noise anomaly, identifying all pixels within the range from the HSV color space, and marking them as noise anomaly data includes: According to the frequency characteristics of the wind speed / power data, the frequency characteristics below the preset threshold are mapped one-to-one to the color depth of the blue channel in the RGB color space; The first type of RGB thermal image is converted into the HSV color space, and the range representing the noise anomaly is defined in the HSV color space. The formula for the range value is:

[0012] In the formula, To represent the color range of noise anomalies, is the maximum hue, is the minimum hue, is the maximum saturation, is the minimum saturation, is the maximum brightness, is the minimum brightness; All blue pixels in the HSV color space are traversed, and all pixels within the range are marked as noise abnormal data by distinguishing colors.

[0013] Optionally, the method of extracting horizontally distributed straight lines from the grayscale image based on edge detection and Hough transform, identifying all horizontally distributed pixel points between each group of straight lines from the grayscale image, and marking them as power-limiting abnormal data includes: Preprocessing the grayscale image, including sequentially performing morphological processing and filtering processing; According to the preprocessed grayscale image, edge detection is performed using a Kirsch operator; According to the grayscale image after edge detection, Hough transform is used to detect straight lines in the grayscale image, and horizontally distributed straight lines are extracted from the result of Hough transform; According to the extracted multiple groups of horizontally distributed straight lines, the interpolation edge polynomial is used to determine the starting point of each straight line, and the horizontally distributed pixel points between each group of straight lines are marked as power-limiting abnormal data by distinguishing colors.

[0014] Optionally, the method for extracting and converting the second type of RGB thermal image of the key fault characteristics of the blade according to the historical time series data includes: According to the historical time series data, a random forest method is used to analyze the fault parameters associated with the causes of blade fault states in the historical time series data to obtain the proportion of blade faults caused by each fault parameter; Selecting the plurality of fault parameters with the highest proportion as key parameters, and using each of the key parameters as an independent channel to segment the time series data, so as to generate a plurality of data samples corresponding to different fault types; Performing kernel density estimation on multiple segments of the data samples in each channel, and converting the estimation result of each channel into a grayscale heat map, using a grayscale matrix to describe the grayscale values ​​of all pixels in each channel; The grayscale matrices of multiple channels are concatenated into an RGB matrix to generate the second type of RGB thermal image.

[0015] Optionally, the method for constructing a health status assessment model for image recognition through deep learning of a training set and a test set includes: The second type of RGB thermal image in the training set is decomposed into three independent channels of R, G, and B through the sign enhancement algorithm, and converted into grayscale images respectively. Randomly combine the grayscale images of the three channels R, G, and B at different time points to generate image data samples; Select AlexNet as the network architecture to build a basic model, input the image data samples in the training set into the basic model in batches for training, and obtain a health status assessment model by iteratively optimizing model parameters; The second type of RGB thermal images in the test set are input into the trained health status assessment model for model evaluation.

[0016] Optionally, the method of selecting AlexNet as the network architecture to build a basic model includes: Create a neural network structure corresponding to the AlexNet architecture; Replace all ReLU activation functions in the neural network with Leaky ReLU and configure the preset parameters; The Adam optimizer is set in the neural network, and the initial learning rate, Momentum value parameters and other parameters are specified to obtain the basic model based on the improved AlexNet architecture.

[0017] Compared with the prior art, the method for evaluating the health status of wind turbine blades provided by the embodiment of the present invention has the following beneficial effects: The real-time time series data obtained from the wind turbine SCADA system is converted into the first type of RGB thermal images of wind speed and power operation, and the abnormal data of the wind turbine is identified and marked from the first type of RGB thermal images based on image processing technology, so as to convert the data classification problem into an image processing problem, fully explore the hidden features of the SCADA system, and visualize the entire abnormal data identification process, which does not rely on human experience such as reference images, has good generalization and good accuracy, thus laying the foundation for the health status assessment of wind turbine blades. Then, the second type of RGB thermal images of the key fault features of the blades are generated according to the historical time series data of the wind turbine SCADA system, and a health status assessment model for image recognition is constructed based on deep learning, so as to fully explore and analyze the hidden information in the SCADA data by fusing and enhancing the feature information, and the first type of RGB thermal images marked with abnormal data are used as the input data of the health status assessment model to achieve accurate assessment of the health status of the wind turbine blades. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments, in which: Figure 1 A schematic block diagram of the steps of a method for evaluating the health status of wind turbine blades provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0019] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. Now, in conjunction with the accompanying drawings, the preferred embodiments of the present invention are described in detail.

[0020] The present invention discloses a method for evaluating the health status of blades of a wind turbine generator set, comprising: S1. Obtain the real-time time series data of the wind turbine SCADA system, extract and transform the first type of RGB thermal image of wind speed and power operation according to the real-time time series data; S2. According to the first type of RGB thermal images of wind speed and power operation, identifying and marking abnormal data of the wind turbine generator set from the first type of RGB thermal images through image processing; S3, obtaining the historical time series data of the wind turbine SCADA system, extracting and converting the historical time series data to generate the second type of RGB thermal image of the key fault characteristics of the blade; S4. According to the second type of RGB thermal images of the key fault characteristics of the blade, the image set of the second type of RGB thermal images is divided into a training set and a test set, and a health status assessment model for image recognition is constructed through deep learning of the training set and the test set; S5. Input the first type of RGB thermal image marked with abnormal data into the health status assessment model, and output the assessment result of the health status of the wind turbine blades.

[0021] Through the implementation of the above-mentioned wind turbine blade health status assessment method embodiment, the real-time time series data obtained from the wind turbine SCADA system is converted into a first-class RGB thermal image of wind speed and power operation, and the abnormal data of the wind turbine is identified and marked from the first-class RGB thermal image based on image processing technology, so as to convert the data classification problem into an image processing problem, fully explore the hidden features of the SCADA system, and visualize the entire abnormal data identification process, which does not rely on human experience such as reference images, has good generalization and good accuracy, and thus lays a foundation for the health status assessment of wind turbine blades. Then, according to the historical time series data of the wind turbine SCADA system, a second-class RGB thermal image of the key fault features of the blade is generated, and a health status assessment model for image recognition is constructed based on deep learning, so as to fully explore and analyze the hidden information in the SCADA data by fusing and enhancing the feature information, and the first-class RGB thermal image marked with abnormal data is used as the input data of the health status assessment model to achieve accurate assessment of the health status of the wind turbine blade.

[0022] Furthermore, the method for extracting and converting the first type of RGB thermal image of wind speed and power operation according to the real-time time series data includes: Extract wind speed / power data based on real-time time series data obtained from the wind turbine SCADA system; Perform Fourier transform on wind speed / power data and extract frequency characteristics of wind speed / power data; Based on the thermal map generation method, the frequency characteristics of wind speed / power data are mapped one-to-one to each pixel in the RGB color space, and the RGB color values ​​of all pixels are combined to generate the first type of RGB thermal image.

[0023] Through the implementation of the above-mentioned wind turbine blade health status assessment method embodiment, the extracted wind speed / power data is used to obtain the key data of the wind turbine operating status for analysis, providing a basis for subsequent analysis. By performing Fourier transform on the wind speed / power data, the main frequency components contained in these data, such as vibration mode, fault frequency, etc., can be analyzed to identify the periodic or non-periodic mode in the operation of the wind turbine. Heat map is a commonly used visualization tool that uses color to represent the intensity or frequency of data. By corresponding the extracted frequency features to the color values ​​in the RGB color space, each color block represents a specific frequency, so that the color values ​​of all pixels are combined together to obtain a visual image, which intuitively shows the frequency feature distribution of the wind speed / power data, and can quickly identify the frequency distribution, which is convenient for analyzing and understanding the patterns in the data.

[0024] Furthermore, the method of mapping the frequency characteristics of the wind speed / power data to each pixel in the RGB color space in a one-to-one correspondence includes: According to the frequency characteristics of the extracted wind speed / power data, it is divided into wind speed frequency characteristics and power frequency characteristics; Map multiple wind speed frequency features one by one to the color depth of the red channel in the RGB color space, and the higher the wind speed frequency feature, the deeper the red of the mapped pixel point; The multiple power frequency features are mapped one-to-one to the color depth of the green channel in the RGB color space, and the higher the power frequency feature, the deeper the green of the mapped pixel.

[0025] Through the implementation of the above-mentioned embodiment of the wind turbine blade health status assessment method, in the first type of RGB thermal image, the thermal value of the pixel represents the frequency of the data in the overall sample, and the color changes in a band from blue to red, corresponding to the frequency of data points from low to high. Thus, complex time series data is converted into an easy-to-understand thermal image, which is convenient for providing data support for the maintenance and optimization of wind turbines by identifying abnormal frequencies or patterns, and is helpful for subsequent data analysis.

[0026] Furthermore, the method of identifying and marking abnormal data of the wind turbine from the first type of RGB thermal image by image processing includes: Define the shutdown conditions associated with the wind speed / power data, identify all pixels that meet the shutdown conditions from the first type of RGB thermal image, and mark them as abnormal shutdown data; The first type of RGB thermal image is converted into the HSV color space, and the range representing the noise anomaly is defined. All pixels within the range are identified from the HSV color space and marked as noise anomaly data; The first type of RGB thermal image is converted into a grayscale image. Horizontally distributed straight lines are extracted from the grayscale image based on edge detection and Hough transform. All horizontally distributed pixel points between each group of straight lines are identified from the grayscale image and marked as power restriction abnormal data.

[0027] Through the implementation of the above-mentioned wind turbine blade health status assessment method embodiment, by associating wind speed / power data with shutdown conditions, a series of thresholds or rules can be set to determine when the wind turbine is considered to be in a shutdown state, providing a clear judgment standard for the identification of abnormal data, and helping to quickly identify the shutdown event of the wind turbine. The image is also converted from the RGB color space to the HSV color space, that is, hue, saturation and brightness, so that the color information can be better separated, so that it is easier to define and identify pixels within the color range, thereby improving the detection accuracy of noise anomalies. In addition, the color image is converted into a grayscale image to simplify subsequent image processing steps, such as edge detection and line extraction. Through edge detection and Hough transform, straight line features can be identified from the grayscale image, which may be related to the power-limiting state of the wind turbine, so that the horizontally distributed pixel points between the identified straight lines represent the power-limiting events related to the wind turbine. Then, in the first type of RGB thermal image, abnormal data such as shutdown anomaly, noise anomaly and power restriction anomaly can be quickly identified through the above-defined conditions and marked to fully explore the hidden features of the SCADA system. The constructed first type of RGB thermal image transforms the data classification problem into an image processing problem, and can visualize the entire abnormal data recognition process. At the same time, it does not rely on human experience such as reference images, and has good generalization and good accuracy.

[0028] Furthermore, a method for defining a shutdown condition associated with wind speed / power data, identifying all pixel points satisfying the shutdown condition from the first type of RGB thermal image, and marking them as shutdown abnormal data includes: According to the frequency characteristics of wind speed / power data, the frequency characteristics of wind speed greater than the cut-in wind speed and power less than or equal to 0 are defined as shutdown conditions; Traverse each pixel in the first type of RGB thermal image, and mark the pixel corresponding to the wind speed / power data that meets the shutdown conditions by distinguishing colors.

[0029] Through the implementation of the above-mentioned wind turbine blade health status assessment method embodiment, the shutdown condition is based on the frequency characteristics of the wind speed / power data. By using clearly defined shutdown conditions, the actual shutdown state of the wind turbine can be accurately identified, rather than misjudgment caused by other factors. By traversing each pixel point in the first type of RGB thermal image and combining the wind speed / power data to determine which pixels correspond to the shutdown state, the shutdown state can be visualized to help operators or automatic monitoring systems quickly identify the shutdown status of the wind turbine. And at the pixel points corresponding to the wind speed / power data that meets the shutdown conditions, they are marked by changing the color, for example, from red to green, to provide an intuitive way to distinguish normal data from abnormal data, so that the abnormal area is more prominent in the image. Therefore, through the marked abnormal data, it is easier to analyze and process the relevant wind turbine performance data.

[0030] Further, the first type of RGB thermal image is converted into the HSV color space, and a range representing noise anomaly is defined. All pixel points within the range are identified from the HSV color space and marked as noise anomaly data. The method includes: According to the frequency characteristics of the wind speed / power data, the frequency characteristics below the preset threshold are mapped one-to-one to the color depth of the blue channel in the RGB color space; The first type of RGB thermal image is converted into the HSV color space, and the range representing the noise anomaly is defined in the HSV color space. The formula for the range value is:

[0031] In the formula, in the formula, To represent the color range of noise anomalies, is the maximum hue, is the minimum hue, is the maximum saturation, is the minimum saturation, is the maximum brightness, is the minimum brightness; Traverse all blue pixels in the HSV color space and mark all pixels within the range as noise anomaly data by distinguishing colors.

[0032] Through the implementation of the above-mentioned wind turbine blade health status assessment method embodiment, the frequency features below the preset threshold are mapped to the blue channel in the RGB color space to allow the low-frequency features to be associated with specific colors, which is convenient for subsequent identification in the thermal map. The first type of RGB thermal image is converted into the HSV color space so that in the HSV color space, the hue and saturation channels can be adjusted independently of the brightness, which helps to separate and identify specific color features. By defining a color range representing noise anomalies in the HSV color space, it is helpful to accurately identify and mark abnormal pixels that may be caused by noise or interference. Then traverse all blue pixels in the HSV color space and mark the pixels within the noise anomaly range, which can effectively identify and isolate noise anomaly data, improve the accuracy and reliability of data analysis, thereby reducing interference with the interpretation of actual data, and facilitate users to quickly identify noise anomalies, so as to diagnose problems more accurately.

[0033] Furthermore, the method of extracting horizontally distributed straight lines from the grayscale image based on edge detection and Hough transform, identifying all horizontally distributed pixel points between each group of straight lines from the grayscale image, and marking them as power-limiting abnormal data includes: Preprocessing the grayscale image, including morphological processing and filtering processing in sequence; According to the preprocessed grayscale image, the Kirsch operator is used for edge detection; According to the grayscale image after edge detection, Hough transform is used to detect the straight lines in the grayscale image, and the horizontally distributed straight lines are extracted from the results of Hough transform; According to the extracted multiple groups of horizontally distributed straight lines, the interpolation edge polynomial is used to determine the starting point of each straight line, and the horizontally distributed pixel points between each group of straight lines are marked as power-limiting abnormal data by distinguishing colors.

[0034] Through the implementation of the above-mentioned wind turbine blade health status assessment method embodiment, the grayscale image is subjected to morphological processing and filtering to remove noise and improve image quality, which can improve the accuracy of subsequent edge detection and Hough transform and reduce the interference of noise on line detection. The Kirsch operator is used for edge detection, which can provide direction-sensitive edge detection and help extract important features in the image. Hough transform is then applied to detect the straight lines in the image, and horizontally distributed straight lines are extracted from the results. By interpolating to determine the starting point, the straight line can be marked more accurately and the integrity of the straight line features can be ensured. According to the extracted horizontal straight lines, the horizontally distributed pixel points between each group of straight lines are marked as power-limiting abnormal data to visualize the abnormal area in the image, which helps to quickly identify and respond to power-limiting situations. Thus, by quickly identifying power-limiting abnormalities, the system can respond more quickly and take corresponding measures, and help maintenance personnel quickly locate problem areas and optimize maintenance work.

[0035] Furthermore, the method for extracting and converting the second type of RGB thermal image of the key fault characteristics of the blade according to the historical time series data includes: According to the historical time series data, the random forest method is used to analyze the fault parameters associated with the causes of blade fault status in the historical time series data, and the proportion of blade failure caused by each fault parameter is obtained; Select multiple fault parameters with the highest proportion as key parameters, use each key parameter as an independent channel to segment the time series data, and generate multiple data samples corresponding to different fault types; Perform kernel density estimation on multiple data samples in each channel, and convert the estimation results of each channel into a grayscale heat map, using a grayscale matrix to describe the grayscale values ​​of all pixels in each channel; The grayscale matrices of multiple channels are concatenated into an RGB matrix to generate the second type of RGB thermal image.

[0036] Through the implementation of the above-mentioned wind turbine blade health status assessment method embodiment, the random forest method can be used to effectively process high-dimensional data and identify fault parameters that have an important impact on the blade fault state. The fault parameter with the highest proportion is selected as the key parameter, so that the system focuses on the most important fault parameter to simplify the data analysis process and improve the pertinence of fault diagnosis. By performing kernel density estimation on multiple data samples in each key parameter channel, a smooth estimate of the data distribution can be provided, and the result is converted into a grayscale thermal map to intuitively display the probability distribution of the fault parameter. The grayscale matrices of multiple channels are then spliced ​​into an RGB matrix so that the distribution of different fault parameters can be compared and analyzed in a unified visual framework. Thus, the second type of RGB thermal image is used to provide an intuitive way to display the distribution of different fault parameters, so as to help establish a more accurate fault prediction model and reduce future downtime and maintenance costs.

[0037] Furthermore, through deep learning of training sets and test sets, a method for constructing a health status assessment model for image recognition includes: The second type of RGB thermal image in the training set is decomposed into three independent channels of R, G, and B through the sign enhancement algorithm, and converted into grayscale images respectively. Randomly combine the grayscale images of the three channels R, G, and B at different time points to generate image data samples; Select AlexNet as the network architecture to build the basic model, input the image data samples in the training set into the basic model in batches for training, and obtain the health status assessment model by iteratively optimizing the model parameters; The second type of RGB thermal images in the test set are input into the trained health status assessment model for model evaluation.

[0038] Through the implementation of the above-mentioned wind turbine blade health status assessment method embodiment, the second type of RGB thermal image is decomposed into three independent channels of R, G, and B using a symbol enhancement algorithm, which can help reduce image noise and improve the interpretability of features, and convert it into a grayscale image to simplify subsequent processing and model input. By randomly combining the grayscale images of the three channels of R, G, and B at different time points, data diversity can be increased, which helps to improve the generalization ability of the model, so that it can make accurate predictions when facing unseen situations. AlexNet is selected as the network architecture to build a basic model. Based on its powerful convolutional neural network architecture, it is suitable for image recognition tasks, and its pre-trained characteristics help to improve the recognition accuracy of the model. Finally, by using the second type of RGB thermal images in the test set to evaluate the trained health status assessment model, the performance of the health status assessment model on actual data can be verified, ensuring that the health status assessment model has good generalization ability and reliability. Thus, the constructed health status assessment model fully mines and analyzes the hidden information in the SCADA data, fuses and enhances the feature information, and realizes an accurate assessment of the health status of the wind turbine blades.

[0039] Furthermore, the method of selecting AlexNet as the network architecture to build the basic model includes: Create a neural network structure corresponding to the AlexNet architecture; Replace all ReLU activation functions in the neural network with Leaky ReLU and configure the preset parameters; Set the Adam optimizer in the neural network and specify the initial learning rate, Momentum value parameters and other parameters to obtain a basic model based on the improved AlexNet architecture.

[0040] Through the implementation of the above-mentioned wind turbine blade health status assessment method embodiment, a neural network structure corresponding to the AlexNet architecture is created, and its multi-layer convolution and pooling layer design is retained. The use of the AlexNet architecture has been proven to be very effective in image recognition tasks, so retaining its core design can ensure that the health status assessment model has a strong feature extraction capability. The ReLU activation function is replaced with Leaky ReLU, and the preset parameters are configured so that Leaky ReLU can solve the gradient disappearance problem in the ReLU function, and make the neural network more stable during the training process, which helps to improve the learning efficiency and final performance of the model. Since Adam is an optimization algorithm for adaptive learning rate, it combines momentum and adaptive learning rate, sets the Adam optimizer, and specifies the initial learning rate, Momentum value parameters and other related parameters. The health status assessment model can converge quickly and provide better optimization effects. Thus, the improved AlexNet architecture can be applied to different visual tasks.

[0041] The wind turbine blade health status assessment method of the present invention includes abnormal data recognition based on image processing and establishment of a health status assessment model based on deep learning. Based on abnormal data recognition based on image processing, color space conversion, Kirsch operator, Hough transform and other technologies are used to fully explore the hidden features of the SCADA system, and the first type of RGB thermal image is constructed to convert the data classification problem into an image processing problem. The entire abnormal data recognition process can be visualized, and it does not rely on human experience such as reference images, and has good generalization and good accuracy. Based on the establishment of a health status assessment model based on deep learning, a health status assessment model is constructed using random forest, improved AlexNet and symbol enhancement technologies to fully explore and analyze the hidden information in the SCADA data, and the feature information is integrated and enhanced, thereby achieving an accurate assessment of the health status of the wind turbine blades.

[0042] Based on the above-mentioned wind turbine blade health status assessment method, the present invention also discloses an assessment device, including: The first type of thermal image generation module is used to obtain the real-time time series data of the wind turbine SCADA system, extract and convert the real-time time series data to generate the first type of RGB thermal image of wind speed and power operation; an abnormal data identification module, used for identifying and marking abnormal data of the wind turbine generator set from the first type of RGB thermal image according to the first type of RGB thermal image of wind speed and power operation through image processing; The second type of thermal image generation module is used to obtain the historical time series data of the wind turbine SCADA system, extract and transform the historical time series data to generate the second type of RGB thermal image of the key fault characteristics of the blade; A model building module is used to divide the image set of the second-category RGB thermal images into a training set and a test set according to the second-category RGB thermal images of key fault characteristics of the blade, and to build a health status assessment model for image recognition through deep learning of the training set and the test set; The model evaluation module is used to input the first type of RGB thermal image marked with abnormal data into the health status evaluation model, and output the evaluation result of the health status of the wind turbine blade.

[0043] Based on the above-mentioned method for assessing the health status of wind turbine blades, the present invention further discloses a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the above-mentioned method for real-time monitoring of carbon emissions from coal-fired units are implemented.

[0044] Based on the above-mentioned method for assessing the health status of wind turbine blades, the present invention also discloses a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that when the processor executes the computer program, the steps of the above-mentioned method for real-time monitoring of carbon emissions from coal-fired units are implemented.

[0045] The present invention is described in terms of flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to specific embodiments. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0046] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0047] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0048] It should be understood that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. For those skilled in the art, the technical solutions described in the above embodiments can be modified, or some of the technical features therein can be replaced by equivalents; and all these modifications and replacements should fall within the scope of protection of the claims attached to the present invention.

Claims

1. A method for evaluating the health status of wind turbine blades, characterized in that: The wind turbine blade health status assessment method comprises: Acquire real-time time series data of the wind turbine SCADA system, extract and transform the real-time time series data to generate a first-class RGB thermal image of wind speed and power operation; According to the first type of RGB thermal images of wind speed and power operation, identifying and marking abnormal data of the wind turbine generator set from the first type of RGB thermal images through image processing; Acquire historical time series data of the wind turbine SCADA system, and extract and transform the historical time series data to generate a second type of RGB thermal image of key fault features of the blade; According to the second type of RGB thermal images of key fault features of the blade, the image set of the second type of RGB thermal images is divided into a training set and a test set, and a health status assessment model for image recognition is constructed through deep learning of the training set and the test set; The first type of RGB thermal images marked with abnormal data are input into the health status assessment model, and an assessment result of the health status of the wind turbine blades is obtained by output.

2. The method for evaluating the health status of wind turbine blades according to claim 1, characterized in that: The method for extracting and converting the first type of RGB thermal image of wind speed and power operation according to the real-time time series data includes: Extracting wind speed / power data based on the real-time time series data obtained from the wind turbine SCADA system; Performing Fourier transform on the wind speed / power data and extracting frequency characteristics of the wind speed / power data; Based on the thermal map generation method, the frequency characteristics of the wind speed / power data are mapped one-to-one to each pixel in the RGB color space, and the RGB color values ​​of all pixels are combined to generate the first type of RGB thermal image.

3. The method for evaluating the health status of wind turbine blades according to claim 2, characterized in that: The method of mapping the frequency characteristics of the wind speed / power data to each pixel point in the RGB color space in a one-to-one correspondence includes: According to the frequency characteristics of the extracted wind speed / power data, the data are divided into wind speed frequency characteristics and power frequency characteristics; Mapping the plurality of wind speed frequency features one by one to the color depth of the red channel in the RGB color space, and the higher the wind speed frequency feature, the deeper the red color of the mapped pixel point; The multiple power frequency features are mapped one-to-one to the color depth of the green channel in the RGB color space, and the higher the power frequency feature is, the deeper the green of the mapped pixel point is.

4. The method for evaluating the health status of wind turbine blades according to claim 2 or 3, characterized in that: The method for identifying and marking abnormal data of a wind turbine from the first type of RGB thermal images by image processing comprises: Defining a shutdown condition associated with the wind speed / power data, identifying all pixel points satisfying the shutdown condition from the first type of RGB thermal image, and marking them as shutdown abnormality data; The first type of RGB thermal image is converted into HSV color space, and the noise anomaly is defined range, from the HSV color space to identify the All pixels within the range are marked as noise anomaly data; The first type of RGB thermal image is converted into a grayscale image, horizontally distributed straight lines are extracted from the grayscale image based on edge detection and Hough transform, all pixel points horizontally distributed between each group of straight lines are identified from the grayscale image, and marked as power restriction abnormal data.

5. The method for evaluating the health status of wind turbine blades according to claim 4, characterized in that: The method of defining the shutdown condition associated with the wind speed / power data, identifying all pixel points satisfying the shutdown condition from the first type of RGB thermal image, and marking them as abnormal shutdown data includes: According to the frequency characteristics of the wind speed / power data, a frequency characteristic that the wind speed is greater than the cut-in wind speed and the power is less than or equal to 0 is defined as a shutdown condition; Each pixel point in the first type of RGB thermal image is traversed, and the pixel points corresponding to the wind speed / power data that meet the shutdown conditions are marked by distinguishing colors.

6. The method for evaluating the health status of wind turbine blades according to claim 4, characterized in that: The first type of RGB thermal image is converted into HSV color space, and a noise anomaly is defined. range, from the HSV color space to identify the Methods for identifying all pixels within a certain range and marking them as noise anomaly data include: According to the frequency characteristics of the wind speed / power data, the frequency characteristics below the preset threshold are mapped one-to-one to the color depth of the blue channel in the RGB color space; The first type of RGB thermal image is converted into HSV color space, and the noise anomaly is defined in HSV color space. scope, The formula for determining the range is: In the formula, To represent the color range of noise anomalies, is the maximum hue, is the minimum hue, is the maximum saturation, is the minimum saturation, is the maximum brightness, is the minimum brightness; Traverse all blue pixels in the HSV color space and find All pixels within the range are marked as noise anomaly data by distinguishing colors.

7. The method for evaluating the health status of wind turbine blades according to claim 4, characterized in that: The method of extracting horizontally distributed straight lines from the grayscale image based on edge detection and Hough transform, identifying all horizontally distributed pixel points between each group of straight lines from the grayscale image, and marking them as power-limiting abnormal data includes: Preprocessing the grayscale image, including sequentially performing morphological processing and filtering processing; According to the preprocessed grayscale image, edge detection is performed using a Kirsch operator; According to the grayscale image after edge detection, Hough transform is used to detect straight lines in the grayscale image, and horizontally distributed straight lines are extracted from the result of Hough transform; According to the extracted multiple groups of horizontally distributed straight lines, the interpolation edge polynomial is used to determine the starting point of each straight line, and the horizontally distributed pixel points between each group of straight lines are marked as power-limiting abnormal data by distinguishing colors.

8. The method for evaluating the health status of wind turbine blades according to claim 1, characterized in that: The method for extracting and converting the second type of RGB thermal image of the key fault characteristics of the blade according to the historical time series data includes: According to the historical time series data, a random forest method is used to analyze the fault parameters associated with the causes of blade fault states in the historical time series data to obtain the proportion of blade faults caused by each fault parameter; Selecting the plurality of fault parameters with the highest proportion as key parameters, and using each of the key parameters as an independent channel to segment the time series data, so as to generate a plurality of data samples corresponding to different fault types; Performing kernel density estimation on multiple segments of the data samples in each channel, and converting the estimation result of each channel into a grayscale heat map, using a grayscale matrix to describe the grayscale values ​​of all pixels in each channel; The grayscale matrices of multiple channels are concatenated into an RGB matrix to generate the second type of RGB thermal image.

9. The method for evaluating the health status of wind turbine blades according to claim 1, characterized in that: The method for constructing a health status assessment model for image recognition through deep learning of a training set and a test set includes: The second type of RGB thermal image in the training set is decomposed into three independent channels of R, G, and B through the sign enhancement algorithm, and converted into grayscale images respectively. Randomly combine the grayscale images of the three channels R, G, and B at different time points to generate image data samples; Select AlexNet as the network architecture to build a basic model, input the image data samples in the training set into the basic model in batches for training, and obtain a health status assessment model by iteratively optimizing model parameters; The second type of RGB thermal images in the test set are input into the trained health status assessment model for model evaluation.

10. The method for evaluating the health status of wind turbine blades according to claim 9, characterized in that: The method of selecting AlexNet as the network architecture to build a basic model includes: Create a neural network structure corresponding to the AlexNet architecture; Replace all ReLU activation functions in the neural network with Leaky ReLU and configure the preset parameters; The Adam optimizer is set in the neural network, and the initial learning rate, Momentum value parameters and other parameters are specified to obtain the basic model based on the improved AlexNet architecture.

Images