A wind turbine ice monitoring method

Abstract

The present invention relates to the technical field of wind turbine monitoring and discloses a method for monitoring wind turbine ice buildup. The method comprises: performing denoising and segmented enhancement processing on collected wind turbine blade images; constructing a wind turbine blade ice buildup recognition model, optimizing the wind turbine blade ice buildup recognition model using an improved non-convex optimization method to obtain an optimal wind turbine blade ice buildup recognition model; and extracting ice edge images from the enhanced wind turbine blade images using the optimal wind turbine blade ice buildup recognition model, and calculating the ice thickness on the wind turbine blade surfaces. The present invention performs denoising and enhancement processing on images collected under harsh environments, constructs a wind turbine blade ice buildup recognition model based on image morphological features, selects an optimal edge detection operator template to perform edge detection on the images, and then calculates the ice thickness on the wind turbine blade surfaces, thereby achieving real-time monitoring of wind turbines.

Description

Technical Field

[0001] The present invention relates to the technical field of wind turbine monitoring, and in particular to a wind turbine icing monitoring method. Background Art

[0002] Wind power generation has become one of the primary methods of electricity production in my country. It is not only clean and pollution-free, but also effectively reduces production costs and improves resource and energy utilization. However, in some wind power generation areas, winter temperatures are relatively low. Large temperature differences in the external environment can lead to ice accumulation on wind turbine blades, severely impacting the normal operation of the unit and significantly reducing the quality and efficiency of wind power generation. To address this issue, the present invention proposes a wind turbine ice monitoring method to ensure safe operation of wind turbines. Summary of the Invention

[0003] In view of this, the present invention provides a wind turbine ice monitoring method, the purpose of which is: 1) Since there are a large number of image noise signals in the collected wind turbine blade images under severe weather scenes, multi-scale image decomposition is adopted, and the noise signal filtering processing is performed on the decomposition results at different scales by using an adaptive threshold combined with the decomposition scale, wherein the higher the scale, the larger the threshold, and the lower the processing degree of the high-scale decomposition result, thereby avoiding filtering a large amount of contour information at high scales, realizing denoising processing of the collected wind turbine blade images, and performing segmented enhancement processing on the denoised wind turbine blade images. When the original grayscale value of the image pixel is too low or too high, the pixel grayscale value is processed by logarithmic transformation, the range of low grayscale value is expanded and the range of high grayscale value is compressed, that is, the dark part in the image is expanded. , suppress the bright parts in the image and highlight the dark parts in the image collected under severe weather conditions, which is beneficial to the monitoring of wind turbine ice in severe weather conditions, and use gamma transform in other grayscale value intervals to enhance the difference between different grayscale levels, achieve image detail enhancement, and improve the accuracy of ice thickness monitoring; 2) By constructing a wind turbine blade ice recognition model combined with image morphological features, the optimal edge detection operator template is selected based on the image morphological features, and the edge detection is performed using the selected edge detection operator template to obtain the ice edge image of the wind turbine blade, and then the ice thickness on the surface of the wind turbine blade is calculated. In the model optimization process, the approximate terms of the model parameter vector to be optimized are added to adjust and optimize the model parameter vector to avoid the model parameter vector after iterative update deviating too much from the stable result.

[0004] To achieve the above objectives, the present invention provides a method for monitoring ice walking on a wind turbine, comprising the following steps:

[0005] S1: collecting wind turbine blade images and performing denoising on the collected wind turbine blade images to obtain denoised wind turbine blade images, wherein an adaptive threshold is a main implementation method of the denoising process;

[0006] S2: performing segment enhancement processing on the denoised wind turbine blade image to obtain an enhanced wind turbine blade image;

[0007] S3: constructing a wind turbine blade ice recognition model, wherein the model takes the enhanced wind turbine blade image as input and the ice edge image as output;

[0008] S4: optimizing the wind turbine blade ice identification model using an improved non-convex optimization method to obtain an optimal wind turbine blade ice identification model, wherein an approximate alternating direction optimization method is a main implementation method of the non-convex optimization method;

[0009] S5: Use the optimal wind turbine blade ice recognition model to extract the ice edge image from the enhanced wind turbine blade image, and calculate the ice thickness on the wind turbine blade surface based on the ice edge image. If the ice thickness exceeds the preset ice threshold, an early warning is issued.

[0010] As a further improvement method of the present invention:

[0011] Optionally, in step S1, collecting the wind turbine blade image and performing denoising on the collected wind turbine blade image includes:

[0012] A camera is deployed near the wind turbine. The camera is used to collect the image of the wind turbine blade at the current moment in real time. The collected image of the wind turbine blade is denoised to obtain the denoised image of the wind turbine blade. The denoising process is as follows:

[0013] S11: grayscale processing is performed on the collected wind turbine blade image I, wherein the grayscale processing formula is:

[0014] g(i,j)=max{R(i,j),G(i,j),B(i,j)},i∈[1,M],j∈[1,N]

[0015] in:

[0016] R(i,j), G(i,j), B(i,j) represent the color values ​​of pixel I(i,j) in the wind turbine blade image I in the R, G, and B color channels respectively; pixel I(i,j) represents the pixel in the i-th row and j-th column of the wind turbine blade image I;

[0017] g(i,j) represents the grayscale value of pixel I(i,j) in the wind turbine blade image I;

[0018] M represents the number of row pixels of the wind turbine blade image, and N represents the number of column pixels of the wind turbine blade image;

[0019] S12: Construct spatial filter coefficient matrix W:

[0020]

[0021] in:

[0022] w(a1, a2) represents the filter coefficients in the spatial filter coefficient matrix W, a1, a2 = {-1, 0, 1}

[0023] S13: Using the spatial filter coefficient matrix W, perform spatial filtering on the grayscale value of any pixel I(i, j) in the wind turbine blade image. The formula for the spatial filtering is:

[0024]

[0025] in:

[0026] g′(i,j) represents the grayscale value of pixel I(i,j) after spatial filtering;

[0027] S14: With the center pixel of the wind turbine blade image as the origin, the horizontal direction as the X-axis direction, and the vertical direction as the Y-axis direction, the pixel position in the wind turbine blade image is mapped to a coordinate system, where f(x, y) represents the grayscale value of the pixel at the coordinate (x, y), and the grayscale value is the grayscale value of the pixel after spatial filtering;

[0028] S15: Adjust the image decomposition scale b and perform multi-scale decomposition processing on the wind turbine blade image:

[0029]

[0030] in:

[0031] D(b) represents the decomposition result of the wind turbine blade image at scale b, b∈[1,B], [1,B] represents the decomposition scale range of the wind turbine blade image, and B represents the maximum image decomposition scale;

[0032] q represents the imaginary unit, q 2 =-1;

[0033] e represents a natural constant;

[0034] S16: Constructing an adaptive threshold function:

[0035]

[0036] in:

[0037] λ b represents the threshold for processing the decomposition result of the wind turbine blade image at scale b;

[0038] λ represents the baseline threshold;

[0039] S17: Processing the decomposition results of the wind turbine blade images at different scales based on a threshold function to filter out noise signals from the decomposition results of the images at different scales. The processing formula is:

[0040]

[0041] in:

[0042] D′(b) represents the decomposition result of the wind turbine blade image at scale b after processing;

[0043] S18: Reconstruct the image decomposition result after filtering the noise signal to obtain a denoised wind turbine blade image:

[0044]

[0045] in:

[0046] I′ represents the denoised wind turbine blade image, and f′(x, y) represents the grayscale value of the pixel at coordinate (x, y) in I′.

[0047] Optionally, in step S2, performing segmented enhancement processing on the denoised wind turbine blade image includes:

[0048] The denoised wind turbine blade image I′ is subjected to segmented enhancement processing to obtain a segmented enhanced wind turbine blade image I″, wherein the segmented enhancement processing formula is:

[0049]

[0050] in:

[0051] f1,f2 are segmented grayscale values, f1 <f2;

[0052] f′(x,y) represents the grayscale value of the pixel at coordinate (x,y) in I′;

[0053] f″(x,y) represents the grayscale value of the pixel at coordinate (x,y) in I′ after segment enhancement.

[0054] In an embodiment of the present invention, when the original grayscale value is too low or too high, a logarithmic transformation is used to process the pixel grayscale value to expand the range of low grayscale values ​​and compress the range of high grayscale values, that is, to expand the dark part of the image, suppress the bright part of the image, and highlight the dim part of the image collected under severe weather conditions, which is beneficial to the ice monitoring of wind turbines under severe weather conditions. Gamma transformation is used in other grayscale value ranges to enhance the details of the pixels.

[0055] Optionally, the step S3 of constructing a wind turbine blade ice recognition model includes:

[0056] A wind turbine blade ice detection model was constructed. The model took an enhanced wind turbine blade image as input and an ice edge image as output. A morphological analysis-based cell machine was the main structure of the wind turbine blade ice detection model.

[0057] The constructed model includes an input layer, a morphological feature extraction layer, a template selection layer, and an output layer. The input layer is used to receive the enhanced wind turbine blade image. The morphological feature extraction layer is used to extract the morphological features of the enhanced wind turbine blade image, select an edge detection operator template based on the morphological features, and use the selected edge detection operator template to perform edge detection marking on the enhanced wind turbine blade image. The edge marking results are used to form an ice edge image as the output result of the output layer.

[0058] The morphological feature extraction layer includes a morphological processing layer and a convolution layer. The morphological processing layer is used to perform morphological processing on the enhanced wind turbine blade image using a dilation matrix and an erosion matrix. The convolution layer is used to perform a convolution operation on the morphological processing result to obtain a feature map. The size of the convolution kernel in the convolution layer is 3×3 pixels.

[0059] The template selection layer includes a pooling layer and an activation function, which performs pooling on the feature map. The template matching function is used to calculate the correlation between the pooling result and the edge detection operator template. The edge detection operator template with the highest correlation is selected as the template selection result. In this embodiment of the present invention, the edge detection operator template includes a Sobel operator template, a Roberts operator template, a Laplacian operator template, a Canny operator template, and the like.

[0060] Optionally, in step S4, optimizing the wind turbine blade ice identification model using an improved non-convex optimization method includes:

[0061] The wind turbine blade ice identification model is optimized using an improved non-convex optimization method, where the model optimization process is as follows:

[0062] S41: Collect denoised and enhanced wind turbine blade images under different environments, and mark and extract the edges of the collected images to obtain real ice edge images, which constitute the training set data for model training:

[0063] data={(U n ,E n )|n∈[1,H]}

[0064] in:

[0065] U n represents the nth denoised and enhanced wind turbine blade image collected, E n represents the real ice edge image of the nth denoised and enhanced wind turbine blade image, and H represents the total number of denoised and enhanced wind turbine blade images in the training set data;

[0066] S42: Constructing the training objective function of the wind turbine blade ice recognition model:

[0067]

[0068] in:

[0069] L(θ) represents the training objective function of the wind turbine blade ice detection model, and θ represents the model parameter vector to be trained and optimized, including the expansion matrix, erosion matrix, convolutional layer weights, and bias parameters;

[0070] sim(·) represents cosine similarity calculation, Hu(·) represents Hu moment feature extraction;

[0071] Indicates that U n Input into the wind turbine blade ice recognition model based on the model parameter vector θ, and the model outputs the ice edge image;

[0072] S43: Initialize the model parameter vector θ0 and set the current iteration number t of model training. The initial value of t is 0.

[0073] S44: Generate parameter approximation:

[0074]

[0075] in:

[0076] α represents the regulation constant between 0 and 1;

[0077] z t represents the parameter approximation term of the t-th iteration;

[0078] ||·|| represents the L1 norm;

[0079] S45: Iteratively update the model parameter vector:

[0080]

[0081] If ||θ t+1 -θ t ||<∈, then the model parameter vector iteration ends, let θ t+1 is the optimal model parameter vector obtained through training, and an optimal wind turbine blade ice detection model is constructed based on the optimal model parameter vector; otherwise, t=t+1 is set, and the process returns to step S44.

[0082] Optionally, in step S5, performing ice edge image extraction on the enhanced wind turbine blade image using the optimal wind turbine blade ice recognition model includes:

[0083] The optimal wind turbine blade ice recognition model is used to extract the ice edge image from the enhanced wind turbine blade image. The ice edge image extraction process is as follows:

[0084] S51: inputting the enhanced wind turbine blade image I″ into the optimal wind turbine blade ice recognition model;

[0085] S52: Using the expansion matrix and the corrosion matrix in the optimal wind turbine blade ice recognition model, the enhanced wind turbine blade image is sequentially expanded and eroded, and the morphologically processed image I is processed. * Perform convolution operation:

[0086] F=Conv(Q T I * +h)

[0087] in:

[0088] Conv(·) represents the convolution operation of the convolution layer;

[0089] Q represents the weight parameter of the convolution kernel in the convolution layer, and h represents the bias of the convolution kernel;

[0090] F represents the feature map;

[0091] S53: Perform pooling operation and activation function processing on the feature map:

[0092] F′=σ[C(F)]

[0093] in:

[0094] C(·) represents the pooling operation with a step size of 2;

[0095] σ(·) represents the activation function, and the ReLU activation function is selected;

[0096] F′ represents the feature map after pooling operation and activation function processing;

[0097] S54: Calculate the feature map F′ and the edge detection operator template S k Correlation Cor(F′,S k ):

[0098]

[0099] in:

[0100] S k Represents the kth edge detection operator template;

[0101] ||·||2 represents the L2 norm;

[0102] Select the edge detection operator template with the highest correlation as the template selection result;

[0103] S55: Using the selected edge detection operator template, edge detection and marking are performed on the enhanced wind turbine blade image, and the edge marking results are used to form an ice edge image as the output result of the model.

[0104] Optionally, the step S5 of calculating the ice thickness on the surface of the wind turbine blade based on the ice edge image includes:

[0105] The inner center point of the closed edge image in the ice edge image is used as the reference point, the direction of the wind blade is used as the coordinate axis direction, and a one-dimensional coordinate axis is constructed. An arbitrary coordinate point h on the one-dimensional coordinate axis is selected, and a ray is emitted from the coordinate point to the edges on both sides. The emitted ray is perpendicular to the one-dimensional coordinate axis, and the intersection of the ray and the edges on both sides is obtained. The distance between the two intersection points is calculated as the surface thickness v of the wind blade corresponding to the coordinate point h. h , and calculate the ice thickness v corresponding to the coordinate point h h -v, where v represents the blade thickness. The selected coordinate point is within the closed edge image in the ice edge image. If the ice thickness exceeds a preset ice threshold, an alert is issued. In this embodiment of the present invention, the ice edge image contains several closed edge images, each corresponding to the edge image of a wind turbine blade.

[0106] In order to solve the above problem, the present invention provides an electronic device, comprising:

[0107] a memory storing at least one instruction;

[0108] Communication interfaces to enable electronic equipment to communicate; and

[0109] The processor executes the instructions stored in the memory to implement the above-mentioned wind turbine icing monitoring method.

[0110] In order to solve the above problem, the present invention also provides a computer-readable storage medium, in which at least one instruction is stored. The at least one instruction is executed by a processor in an electronic device to implement the above-mentioned wind turbine icing monitoring method.

[0111] Compared with the existing technology, the present invention proposes a wind turbine ice monitoring method, which has the following advantages:

[0112] First, this scheme proposes a denoising and enhancement method for wind turbine blade images. The spatial filter coefficient matrix W is used to perform spatial filtering on the grayscale value of any pixel I(i, j) in the wind turbine blade image. The formula for spatial filtering is:

[0113]

[0114] Where: g′(i,j) represents the grayscale value of pixel I(i,j) after spatial filtering processing; with the center pixel of the wind turbine blade image as the origin, the horizontal direction is the X-axis direction, and the vertical direction is the Y-axis direction, the pixel position in the wind turbine blade image is mapped to the coordinate system, where f(x,y) represents the grayscale value of the pixel at the coordinate (x,y), which is the grayscale value of the pixel after spatial filtering processing; adjust the image decomposition scale b, and perform multi-scale decomposition processing on the wind turbine blade image:

[0115]

[0116] Where: D(b) represents the decomposition result of the wind turbine blade image at scale b, b∈[1,B], [1,B] represents the decomposition scale range of the wind turbine blade image, B represents the maximum image decomposition scale; q represents the imaginary unit, q 2 = -1; e represents a natural constant; construct an adaptive threshold function:

[0117]

[0118] Where: b represents the threshold for processing the decomposition results of the wind turbine blade image at scale b; λ represents the reference threshold; based on the threshold function, the decomposition results of the wind turbine blade image at different scales are processed to filter the noise signals of the image decomposition results at different scales. The processing formula is:

[0119]

[0120] Where: D′(b) represents the decomposition result of the wind turbine blade image at scale b after processing; the image decomposition result after filtering the noise signal is reconstructed to obtain the denoised wind turbine blade image:

[0121]

[0122] Where I′ represents the denoised wind turbine blade image, and f′(x,y) represents the grayscale value of the pixel at coordinate (x,y) in I′. This scheme employs multi-scale image decomposition and uses an adaptive threshold based on the decomposition scale to filter noise signals from the decomposition results at different scales. The higher the scale, the larger the threshold, and the less processing is applied to the high-scale decomposition results, avoiding filtering of large amounts of high-scale contour information. This denoises the captured wind turbine blade images and performs segmented enhancement on the denoised wind turbine blade images. When the original grayscale values ​​of image pixels are too low or too high, a logarithmic transformation is applied to the pixel grayscale values, expanding the range of low grayscale values ​​and compressing the range of high grayscale values. This expands the dark portions of the image and suppresses the bright portions, highlighting the dim portions of images captured in inclement weather. This is beneficial for wind turbine ice monitoring in inclement weather. A gamma transform is also applied to other grayscale value ranges to enhance the differences between different grayscale levels, enhancing image detail and improving the accuracy of ice thickness monitoring.

[0123] At the same time, this scheme proposes a wind turbine blade ice recognition model and optimization strategy. By constructing a wind turbine blade ice recognition model combined with image morphological features, the optimal edge detection operator template is selected based on the image morphological features, and edge detection is performed using the selected edge detection operator template to obtain the ice edge image of the wind turbine blade. Then, the ice thickness on the surface of the wind turbine blade is calculated, and the wind turbine blade ice recognition model is optimized using an improved non-convex optimization method. The model optimization process is as follows: denoised and enhanced wind turbine blade images under different environments are collected, and the edges of the collected images are marked and extracted to obtain real ice edge images, which constitute the training set data for model training:

[0124] data={(U n ,E n )|n∈[1,H]}

[0125] Among them: U n represents the nth denoised and enhanced wind turbine blade image collected, E n represents the real ice edge image of the nth denoised and enhanced wind turbine blade image, and H represents the total number of denoised and enhanced wind turbine blade images in the training set data. The training objective function of the wind turbine blade ice recognition model is constructed as follows:

[0126]

[0127] Where: L(θ) represents the training objective function of the wind turbine blade ice recognition model, θ represents the model parameter vector to be trained and optimized, including the expansion matrix, erosion matrix, convolution layer weights, and bias parameters; sim(·) represents cosine similarity calculation, and Hu(·) represents Hu moment feature extraction; Indicates that U n Input the ice edge image into the wind turbine blade ice recognition model based on the model parameter vector θ; initialize the model parameter vector θ0, set the current iteration number t of the model training, and the initial value of t is 0; generate the parameter approximation term:

[0128]

[0129] Where: α represents the adjustment constant between 0 and 1; z t represents the parameter approximation term of the t-th iteration; ||·|| represents the L1 norm; the model parameter vector is iteratively updated:

[0130]

[0131] If ||θ t+1 -θ t ||<∈, then the model parameter vector iteration ends, let θ t+1 The optimal model parameter vector obtained through training is used to construct the optimal wind turbine blade ice recognition model based on the optimal model parameter vector. Otherwise, let t = t + 1. During the model optimization process, this scheme adds an approximate term of the model parameter vector to be optimized to adjust and optimize the model parameter vector, preventing the iteratively updated model parameter vector from deviating too much from the stable result. BRIEF DESCRIPTION OF THE DRAWINGS

[0132] Figure 1 A schematic flow chart of a method for monitoring ice walking on a wind turbine according to an embodiment of the present invention;

[0133] Figure 2 A schematic structural diagram of an electronic device for implementing a method for monitoring ice walking on a wind turbine according to an embodiment of the present invention.

[0134] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings and in conjunction with the embodiments. DETAILED DESCRIPTION

[0135] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0136] The present invention provides a method for monitoring wind turbine icing. The method can be performed by at least one electronic device, such as a server or terminal, that can be configured to perform the method provided by the present invention. In other words, the method can be performed by software or hardware installed on a terminal or server device, where the software can be a blockchain platform. The server can include, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster.

[0137] Example 1:

[0138] S1: Collecting wind turbine blade images and performing denoising on the collected wind turbine blade images to obtain denoised wind turbine blade images.

[0139] In step S1, the wind turbine blade image is collected and denoising is performed on the collected wind turbine blade image, including:

[0140] A camera is deployed near the wind turbine. The camera is used to collect the image of the wind turbine blade at the current moment in real time. The collected image of the wind turbine blade is denoised to obtain the denoised image of the wind turbine blade. The denoising process is as follows:

[0141] S11: grayscale processing is performed on the collected wind turbine blade image I, wherein the grayscale processing formula is:

[0142] g(i,j)=max{R(i,j),G(i,j),B(i,j)},i∈[1,M],j∈[1,N]

[0143] in:

[0144] R(i,j), G(i,j), B(i,j) represent the color values ​​of pixel I(i,j) in the wind turbine blade image I in the R, G, and B color channels respectively; pixel I(i,j) represents the pixel in the i-th row and j-th column of the wind turbine blade image I;

[0145] g(i,j) represents the grayscale value of pixel I(i,j) in the wind turbine blade image I;

[0146] M represents the number of row pixels of the wind turbine blade image, and N represents the number of column pixels of the wind turbine blade image;

[0147] S12: Construct spatial filter coefficient matrix W:

[0148]

[0149] in:

[0150] w(a1, a2) represents the filter coefficients in the spatial filter coefficient matrix W, a1, a2 = {-1, 0, 1}

[0151] S13: Using the spatial filter coefficient matrix W, perform spatial filtering on the grayscale value of any pixel I(i, j) in the wind turbine blade image. The formula for the spatial filtering is:

[0152]

[0153] in:

[0154] g′(i,j) represents the grayscale value of pixel I(i,j) after spatial filtering;

[0155] S14: With the center pixel of the wind turbine blade image as the origin, the horizontal direction as the X-axis direction, and the vertical direction as the Y-axis direction, the pixel position in the wind turbine blade image is mapped to a coordinate system, where f(x, y) represents the grayscale value of the pixel at the coordinate (x, y), and the grayscale value is the grayscale value of the pixel after spatial filtering;

[0156] S15: Adjust the image decomposition scale b and perform multi-scale decomposition processing on the wind turbine blade image:

[0157]

[0158] in:

[0159] D(b) represents the decomposition result of the wind turbine blade image at scale b, b∈[1,B], [1,B] represents the decomposition scale range of the wind turbine blade image, and B represents the maximum image decomposition scale;

[0160] q represents the imaginary unit, q 2 =-1;

[0161] e represents a natural constant;

[0162] S16: Constructing an adaptive threshold function:

[0163]

[0164] in:

[0165] λ b represents the threshold for processing the decomposition result of the wind turbine blade image at scale b;

[0166] λ represents the baseline threshold;

[0167] S17: Processing the decomposition results of the wind turbine blade images at different scales based on a threshold function to filter out noise signals from the decomposition results of the images at different scales. The processing formula is:

[0168]

[0169] in:

[0170] D′(b) represents the decomposition result of the wind turbine blade image at scale b after processing;

[0171] S18: Reconstruct the image decomposition result after filtering the noise signal to obtain a denoised wind turbine blade image:

[0172]

[0173] in:

[0174] I′ represents the denoised wind turbine blade image, and f′(x, y) represents the grayscale value of the pixel at coordinate (x, y) in I′.

[0175] S2: performing segment enhancement processing on the denoised wind turbine blade image to obtain an enhanced wind turbine blade image.

[0176] The step S2 includes performing segmented enhancement processing on the denoised wind turbine blade image, including:

[0177] The denoised wind turbine blade image I′ is subjected to segmented enhancement processing to obtain a segmented enhanced wind turbine blade image I″, wherein the segmented enhancement processing formula is:

[0178]

[0179] in:

[0180] f1,f2 are segmented grayscale values, f1 <f2;

[0181] f′(x,y) represents the grayscale value of the pixel at coordinate (x,y) in I′;

[0182] f″(x,y) represents the grayscale value of the pixel at coordinate (x,y) in I′ after segment enhancement.

[0183] S3: Construct a wind turbine blade ice detection model, wherein the model takes the enhanced wind turbine blade image as input and the ice edge image as output.

[0184] The step S3 constructs a wind turbine blade ice recognition model, including:

[0185] A wind turbine blade ice detection model was constructed. The model took an enhanced wind turbine blade image as input and an ice edge image as output. A morphological analysis-based cell machine was the main structure of the wind turbine blade ice detection model.

[0186] The constructed model includes an input layer, a morphological feature extraction layer, a template selection layer, and an output layer. The input layer is used to receive the enhanced wind turbine blade image. The morphological feature extraction layer is used to extract the morphological features of the enhanced wind turbine blade image, select an edge detection operator template based on the morphological features, and use the selected edge detection operator template to perform edge detection marking on the enhanced wind turbine blade image. The edge marking results are used to form an ice edge image as the output result of the output layer.

[0187] The morphological feature extraction layer includes a morphological processing layer and a convolution layer. The morphological processing layer is used to perform morphological processing on the enhanced wind turbine blade image using a dilation matrix and an erosion matrix. The convolution layer is used to perform a convolution operation on the morphological processing result to obtain a feature map. The size of the convolution kernel in the convolution layer is 3×3 pixels.

[0188] The template selection layer includes a pooling layer and an activation function, which performs pooling on the feature map and uses the template matching function to calculate the correlation between the pooling result and the edge detection operator template, and selects the edge detection operator template with the highest correlation as the template selection result.

[0189] S4: The wind turbine blade icing identification model is optimized using an improved non-convex optimization method to obtain the optimal wind turbine blade icing identification model.

[0190] In step S4, the wind turbine blade ice identification model is optimized using an improved non-convex optimization method, including:

[0191] The wind turbine blade ice identification model is optimized using an improved non-convex optimization method, where the model optimization process is as follows:

[0192] S41: Collect denoised and enhanced wind turbine blade images under different environments, and mark and extract the edges of the collected images to obtain real ice edge images, which constitute the training set data for model training:

[0193] data={(U n ,E n )|n∈[1,H]}

[0194] in:

[0195] U n represents the nth denoised and enhanced wind turbine blade image collected, E nrepresents the real ice edge image of the nth denoised and enhanced wind turbine blade image, and H represents the total number of denoised and enhanced wind turbine blade images in the training set data;

[0196] S42: Constructing the training objective function of the wind turbine blade ice recognition model:

[0197]

[0198] in:

[0199] L(θ) represents the training objective function of the wind turbine blade ice detection model, and θ represents the model parameter vector to be trained and optimized, including the expansion matrix, erosion matrix, convolutional layer weights, and bias parameters;

[0200] sim(·) represents cosine similarity calculation, Hu(·) represents Hu moment feature extraction;

[0201] Indicates that U n Input into the wind turbine blade ice recognition model based on the model parameter vector θ, and the model outputs the ice edge image;

[0202] S43: Initialize the model parameter vector θ0 and set the current iteration number t of model training. The initial value of t is 0.

[0203] S44: Generate parameter approximation:

[0204]

[0205] in:

[0206] α represents the regulation constant between 0 and 1;

[0207] z t represents the parameter approximation term of the t-th iteration;

[0208] ||·|| represents the L1 norm;

[0209] S45: Iteratively update the model parameter vector:

[0210]

[0211] If ||θ t+1 -θ t ||<∈, then the model parameter vector iteration ends, let θ t+1 is the optimal model parameter vector obtained through training, and an optimal wind turbine blade ice detection model is constructed based on the optimal model parameter vector; otherwise, t=t+1 is set, and the process returns to step S44.

[0212] S5: Use the optimal wind turbine blade ice recognition model to extract the ice edge image from the enhanced wind turbine blade image, and calculate the ice thickness on the wind turbine blade surface based on the ice edge image. If the ice thickness exceeds the preset ice threshold, an early warning is issued.

[0213] In step S5, the optimal wind turbine blade ice recognition model is used to extract ice edge images from the enhanced wind turbine blade images, including:

[0214] The optimal wind turbine blade ice recognition model is used to extract the ice edge image from the enhanced wind turbine blade image. The ice edge image extraction process is as follows:

[0215] S51: inputting the enhanced wind turbine blade image I″ into the optimal wind turbine blade ice recognition model;

[0216] S52: Using the expansion matrix and the corrosion matrix in the optimal wind turbine blade ice recognition model, the enhanced wind turbine blade image is sequentially expanded and eroded, and the morphologically processed image I is processed. * Perform convolution operation:

[0217] F=Conv(Q T I * +h)

[0218] in:

[0219] Conv(·) represents the convolution operation of the convolution layer;

[0220] Q represents the weight parameter of the convolution kernel in the convolution layer, and h represents the bias of the convolution kernel;

[0221] F represents the feature map;

[0222] S53: Perform pooling operation and activation function processing on the feature map:

[0223] F′=σ[C(F)]

[0224] in:

[0225] C(·) represents the pooling operation with a step size of 2;

[0226] σ(·) represents the activation function, and the ReLU activation function is selected;

[0227] F′ represents the feature map after pooling operation and activation function processing;

[0228] S54: Calculate the feature map F′ and the edge detection operator template S k Correlation Cor(F′,S k):

[0229]

[0230] in:

[0231] S k Represents the kth edge detection operator template;

[0232] ||·||2 represents the L2 norm;

[0233] Select the edge detection operator template with the highest correlation as the template selection result;

[0234] S55: Using the selected edge detection operator template, edge detection and marking are performed on the enhanced wind turbine blade image, and the edge marking results are used to form an ice edge image as the output result of the model.

[0235] Calculating the ice thickness on the surface of the wind turbine blade based on the ice edge image in step S5 includes:

[0236] The inner center point of the closed edge image in the ice edge image is used as the reference point, the direction of the wind blade is used as the coordinate axis direction, and a one-dimensional coordinate axis is constructed. An arbitrary coordinate point h on the one-dimensional coordinate axis is selected, and a ray is emitted from the coordinate point to the edges on both sides. The emitted ray is perpendicular to the one-dimensional coordinate axis, and the intersection of the ray and the edges on both sides is obtained. The distance between the two intersection points is calculated as the surface thickness v of the wind blade corresponding to the coordinate point h. h , and calculate the ice thickness v corresponding to the coordinate point h h -v, where v represents the thickness of the wind blade. The selected coordinate point is inside the closed edge image in the ice edge image. If the ice surface thickness exceeds the preset ice surface threshold, an early warning is issued.

[0237] Example 2:

[0238] like Figure 2 FIG. 1 is a schematic diagram of the structure of an electronic device for implementing a method for monitoring ice walking on a wind turbine provided by an embodiment of the present invention.

[0239] The electronic device 1 may include a processor 10 , a memory 11 , a communication interface 13 and a bus, and may further include a computer program stored in the memory 11 and executable on the processor 10 , such as a program 12 .

[0240] The memory 11 includes at least one type of readable storage medium, and the readable storage medium includes a flash memory, a mobile hard disk, a multimedia card, a card-type memory (for example, an SD or DX memory, etc.), a magnetic memory, a disk, an optical disk, etc. In some embodiments, the memory 11 can be an internal storage unit of the electronic device 1, such as a mobile hard disk of the electronic device 1. In other embodiments, the memory 11 can also be an external storage device of the electronic device 1, such as a plug-in mobile hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. equipped on the electronic device 1. Furthermore, the memory 11 can also include both an internal storage unit of the electronic device 1 and an external storage device. The memory 11 can not only be used to store application software and various types of data installed in the electronic device 1, such as the code of the program 12, etc., but can also be used to temporarily store data that has been output or is to be output.

[0241] In some embodiments, the processor 10 may be comprised of an integrated circuit, such as a single packaged integrated circuit or a plurality of packaged integrated circuits with the same or different functions, including one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and a combination of various control chips. The processor 10 is the control core (Control Unit) of the electronic device, connecting the various components of the entire electronic device using various interfaces and circuits. It executes or runs programs or modules stored in the memory 11 (such as the program 12 for implementing wind turbine ice monitoring) and calls data stored in the memory 11 to perform various functions of the electronic device 1 and process data.

[0242] The communication interface 13 may include a wired interface and / or a wireless interface (such as a WI-FI interface, a Bluetooth interface, etc.), which is generally used to establish a communication connection between the electronic device 1 and other electronic devices, and to realize connection and communication between internal components of the electronic devices.

[0243] The bus may be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. The bus may be divided into an address bus, a data bus, a control bus, etc. The bus is configured to enable connection and communication between the memory 11 and at least one processor 10, etc.

[0244] Figure 2 Only the electronic device with components is shown, and it can be understood by those skilled in the art that Figure 2 The structure shown does not constitute a limitation on the electronic device 1 , and may include fewer or more components than shown in the figure, or combine certain components, or arrange the components differently.

[0245] For example, although not shown, the electronic device 1 may further include a power source (such as a battery) for powering the various components. Preferably, the power source may be logically connected to the at least one processor 10 via a power management device, thereby implementing functions such as charging management, discharging management, and power consumption management through the power management device. The power source may further include any components such as one or more DC or AC power sources, a recharging device, a power failure detection circuit, a power converter or inverter, a power status indicator, etc. The electronic device 1 may further include various sensors, Bluetooth modules, Wi-Fi modules, etc., which will not be described in detail here.

[0246] Optionally, the electronic device 1 may further include a user interface, which may be a display or an input unit (such as a keyboard). Optionally, the user interface may also be a standard wired interface or a wireless interface. Optionally, in some embodiments, the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touch device. The display may also be appropriately referred to as a display screen or a display unit, which is used to display information processed in the electronic device 1 and to display a visual user interface.

[0247] It should be understood that the embodiment is for illustration only and the scope of the patent application is not limited to this structure.

[0248] The program 12 stored in the memory 11 of the electronic device 1 is a combination of multiple instructions. When executed in the processor 10, it can achieve the following:

[0249] Collecting wind turbine blade images and performing denoising on the collected wind turbine blade images to obtain denoised wind turbine blade images;

[0250] Performing segment enhancement processing on the denoised wind turbine blade image to obtain an enhanced wind turbine blade image;

[0251] Construct a wind turbine blade ice detection model;

[0252] The wind turbine blade icing identification model is optimized using an improved non-convex optimization method, and the optimal wind turbine blade icing identification model is obtained.

[0253] The optimal wind turbine blade ice recognition model is used to extract the ice edge image from the enhanced wind turbine blade image, and the ice thickness on the wind turbine blade surface is calculated based on the ice edge image. If the ice thickness exceeds the preset ice threshold, an early warning is issued.

[0254] Specifically, the specific implementation method of the processor 10 for the above instructions can refer to Figures 1 to 2 The description of the relevant steps in the corresponding embodiments will not be repeated here.

[0255] It should be noted that the serial numbers of the above-mentioned embodiments of the present invention are for descriptive purposes only and do not represent the advantages or disadvantages of the embodiments. In addition, the terms "including", "comprising" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, device, article or method comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, device, article or method. In the absence of further restrictions, an element defined by the sentence "including a ..." does not exclude the presence of other identical elements in the process, device, article or method comprising the element.

[0256] Through the description of the above embodiments, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus the necessary general hardware platform, and of course can also be implemented by hardware, but in many cases the former is a better embodiment. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes a number of instructions for enabling a terminal device (which can be a mobile phone, computer, server, or network device, etc.) to execute the methods described in each embodiment of the present invention.

[0257] The above are only preferred embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention description and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for monitoring ice on a wind turbine, characterized in that: The method comprises: S1: collecting wind turbine blade images and performing denoising on the collected wind turbine blade images to obtain denoised wind turbine blade images; S2: performing segmented enhancement processing on the denoised wind turbine blade image to obtain an enhanced wind turbine blade image; S3: constructing a wind turbine blade ice recognition model, wherein the model takes the enhanced wind turbine blade image as input and the ice edge image as output; S4: optimizing the wind turbine blade ice recognition model using an improved non-convex optimization method to obtain an optimal wind turbine blade ice recognition model; The model optimization process is as follows: S41: Collect denoised and enhanced wind turbine blade images under different environments, and mark and extract the edges of the collected images to obtain real ice edge images, which constitute the training set data for model training: ； in: represents the nth denoised and enhanced wind turbine blade image collected, represents the real ice edge image of the nth denoised and enhanced wind turbine blade image, and H represents the total number of denoised and enhanced wind turbine blade images in the training set data; S42: Constructing the training objective function of the wind turbine blade ice recognition model: ； in: represents the training objective function of the wind turbine blade ice recognition model, Represents the model parameter vector to be trained and optimized, including the expansion matrix, corrosion matrix, convolutional layer weights, and bias parameters; Indicates cosine similarity calculation, Represents Hu moment feature extraction; Indicates that Input to the model parameter vector In the wind turbine blade ice recognition model, the model outputs the ice edge image; S43: Initialize model parameter vector , set the current iteration number t of model training, the initial value of t is 0; S44: Generate parameter approximation: ； in: Indicates the adjustment constant between 0 and 1; represents the parameter approximation term of the t-th iteration; represents the L1 norm; S45: Iteratively update the model parameter vector: ； like , then the model parameter vector iteration ends, let is the optimal model parameter vector obtained through training, and the optimal wind turbine blade ice recognition model is constructed based on the optimal model parameter vector. Otherwise, , return to step S44; S5: Use the optimal wind turbine blade ice recognition model to extract the ice edge image from the enhanced wind turbine blade image, and calculate the ice thickness on the wind turbine blade surface based on the ice edge image. If the ice thickness exceeds the preset ice threshold, an early warning is issued.

2. The wind turbine icing monitoring method according to claim 1, wherein: In step S1, the wind turbine blade image is collected and denoising is performed on the collected wind turbine blade image, including: A camera is deployed near the wind turbine. The camera is used to collect the image of the wind turbine blade at the current moment in real time. The collected image of the wind turbine blade is denoised to obtain the denoised image of the wind turbine blade. The denoising process is as follows: S11: Collected wind turbine blade images Perform grayscale processing, where the grayscale processing formula is: ； in: Represents an image of a wind turbine blade Medium pixels Color values ​​in R, G, B color channels respectively; pixel Represents an image of a wind turbine blade The pixel at row i and column j in ; Represents an image of a wind turbine blade Medium pixels Gray value of M represents the number of row pixels of the wind turbine blade image; N represents the number of column pixels in the wind turbine blade image; S12: Construct spatial filter coefficient matrix W: ； in: represents the filter coefficients in the spatial filter coefficient matrix W, S13: Use the spatial filter coefficient matrix W to filter any pixel in the wind turbine blade image. The grayscale value is spatially filtered, and the formula for spatial filtering is: ； in: Represents the pixel after spatial filtering Gray value of S14: With the center pixel of the wind turbine blade image as the origin, the horizontal direction as the X-axis direction, and the vertical direction as the Y-axis direction, the pixel position in the wind turbine blade image is mapped to a coordinate system, where Indicates that the coordinate The grayscale value of the pixel is the grayscale value of the pixel after spatial filtering; S15: Adjust the image decomposition scale b and perform multi-scale decomposition processing on the wind turbine blade image: ； in: represents the decomposition result of the wind turbine blade image at scale b, , represents the decomposition scale range of the wind turbine blade image, and B represents the maximum image decomposition scale; q represents the imaginary unit, ; e represents a natural constant; S16: Constructing an adaptive threshold function: ； in: represents the threshold for processing the decomposition result of the wind turbine blade image at scale b; represents the baseline threshold; S17: Process the decomposition results of the wind turbine blade images at different scales based on a threshold function to filter out noise signals from the decomposition results of the images at different scales. The processing formula is: ； in: represents the decomposition result of the wind turbine blade image at scale b after processing; S18: Reconstruct the image decomposition result after filtering the noise signal to obtain a denoised wind turbine blade image: ； in: represents the denoised wind turbine blade image, express The coordinates The grayscale value of the pixel.

3. The wind turbine icing monitoring method according to claim 2, wherein: The step S2 includes performing segmented enhancement processing on the denoised wind turbine blade image, including: The denoised wind turbine blade image Perform segmented enhancement processing to obtain the segmented enhanced image of the wind turbine blade , where the segment enhancement processing formula is: ； in: is the segment gray value, ; express The coordinates The grayscale value of the pixel; express The coordinates The gray value of the pixel after segmented enhancement.

4. The wind turbine icing monitoring method according to claim 1, wherein: The step S3 constructs a wind turbine blade ice recognition model, including: A wind turbine blade ice detection model was constructed. The model took an enhanced wind turbine blade image as input and an ice edge image as output. A morphological analysis-based cell machine was the main structure of the wind turbine blade ice detection model. The constructed model includes an input layer, a morphological feature extraction layer, a template selection layer, and an output layer. The input layer is used to receive the enhanced wind turbine blade image. The morphological feature extraction layer is used to extract the morphological features of the enhanced wind turbine blade image, select an edge detection operator template based on the morphological features, and use the selected edge detection operator template to perform edge detection marking on the enhanced wind turbine blade image. The edge marking results are used to form an ice edge image as the output result of the output layer. The morphological feature extraction layer includes a morphological processing layer and a convolution layer. The morphological processing layer is used to perform morphological processing on the enhanced wind turbine blade image using a dilation matrix and an erosion matrix. The convolution layer is used to perform a convolution operation on the morphological processing result to obtain a feature map. The size of the convolution kernel in the convolution layer is Pixels; The template selection layer includes a pooling layer and an activation function, which performs pooling on the feature map and uses the template matching function to calculate the correlation between the pooling result and the edge detection operator template, and selects the edge detection operator template with the highest correlation as the template selection result.

5. The wind turbine icing monitoring method according to claim 1, wherein: In step S5, the optimal wind turbine blade ice recognition model is used to extract ice edge images from the enhanced wind turbine blade images, including: The optimal wind turbine blade ice recognition model is used to extract the ice edge image from the enhanced wind turbine blade image. The ice edge image extraction process is as follows: S51: Enhanced wind turbine blade image Input into the optimal wind turbine blade ice identification model; S52: Using the expansion matrix and corrosion matrix in the optimal wind turbine blade ice recognition model, the enhanced wind turbine blade image is sequentially expanded and eroded, and the morphologically processed image is processed. Perform convolution operation: ； in: Represents the convolution operation of the convolution layer; Q represents the weight parameter of the convolution kernel in the convolution layer, and h represents the bias of the convolution kernel; F represents the feature map; S53: Perform pooling operation and activation function processing on the feature map: ； in: Represents a pooling operation with a step size of 2; Represents the activation function, select the ReLU activation function; Represents the feature map after pooling operation and activation function processing; S54: Calculate the feature map and edge detection operator template Correlation : ； in: Represents the kth edge detection operator template; represents the L2 norm; Select the edge detection operator template with the highest correlation as the template selection result; S55: Using the selected edge detection operator template, edge detection and marking are performed on the enhanced wind turbine blade image, and the edge marking results are used to form an ice edge image as the output result of the model.

6. The wind turbine ice monitoring method according to claim 5, characterized in that: Calculating the ice thickness on the surface of the wind turbine blade based on the ice edge image in step S5 includes: The inner center point of the closed edge image in the ice edge image is used as the reference point, the direction of the wind blade is used as the coordinate axis direction, and a one-dimensional coordinate axis is constructed. An arbitrary coordinate point h on the one-dimensional coordinate axis is selected, and a ray is emitted from the coordinate point to the edges on both sides. The emitted ray is perpendicular to the one-dimensional coordinate axis, and the intersection of the ray and the edges on both sides is obtained. The distance between the two intersection points is calculated as the surface thickness of the wind blade corresponding to the coordinate point h. , and calculate the ice thickness corresponding to the coordinate point h ,in Indicates the thickness of the wind blade. The selected coordinate point is inside the closed edge image in the ice edge image. If the ice surface thickness exceeds the preset ice surface threshold, an early warning is issued.

Images