A power equipment thermal fault diagnosis method and device based on decision-level fusion and a medium

Abstract

The application discloses a kind of power equipment thermal fault diagnosis method, equipment and medium based on decision level fusion, method includes: obtaining the infrared image of power equipment, using infrared image as pixel level infrared spectrum feature, infrared image is segmented into several superpixels using superpixel segmentation algorithm, each superpixel is regarded as 1 empty spectrum similarity feature;Pixel level infrared spectrum feature is recorded as d i＝1 , the rest superpixel level empty spectrum similarity feature is recorded as d i＝2，3，… ;Using low rank matrix decomposition method, respectively from pixel level infrared spectrum feature d i＝1 And all superpixel level empty spectrum similarity features d j＝2，3，… Extract respective thermal fault area s i＝1 , s i＝2，3，… ;All extracted thermal fault areas are fused, and the thermal fault area of power equipment is obtained.The application improves the thermal fault diagnosis precision of power equipment, to improve the intelligent operation and inspection level of power equipment.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of equipment thermal fault diagnosis, and particularly relates to a power equipment thermal fault diagnosis method based on decision-level fusion. BACKGROUND

[0002] Thermal fault is a common fault type in the operation process of power equipment. If the local temperature continues to rise, it is easy to cause the power equipment to trip and power off. Timely diagnosing the thermal fault of power equipment plays a crucial role in the stable and safe operation of the power system. At present, the infrared thermal imaging technology is a common technical means for diagnosing the thermal fault of power equipment. The infrared spectral information can reflect the temperature field information of the image pixel points. However, due to the interference of environmental factors such as image noise and background shielding, it is difficult to obtain a high-precision thermal fault diagnosis result only by relying on the temperature information of single pixel-level features. Therefore, effectively extracting more discriminative image features has become a research hotspot in the field of thermal fault diagnosis of power equipment in recent years.

[0003] In the prior art, the image segmentation method based on superpixel-level features can fully utilize the local similarity space to improve the fault diagnosis result of the smooth region of the infrared image, and effectively solve the problem of difficult thermal fault recognition in the presence of image noise. However, inaccurate superpixel segmentation is easy to cause missegmentation in the edge, structure and texture dense region. On the other hand, the image processing method based on single pixel-level features can better reflect the temperature information of single pixel points. However, due to the interference of environmental factors such as image noise and background shielding, the thermal fault diagnosis method based on single pixel-level features does not fully utilize the spatial similarity information of the infrared image, and is easy to produce misrecognition results in the smooth region. SUMMARY

[0004] The application provides a power equipment thermal fault diagnosis method based on decision-level fusion, which improves the thermal fault diagnosis precision of power equipment, and thereby improves the intelligent operation and inspection level of power equipment.

[0005] To achieve the above technical purposes, the application adopts the following technical scheme:

[0006] A power equipment thermal fault diagnosis method based on decision-level fusion, comprising:

[0007] An infrared image of the power equipment is acquired, the infrared image is used as a pixel-level infrared spectral feature, a superpixel segmentation algorithm is used to segment the infrared image into a plurality of superpixels, each superpixel is taken as a 1 empty spectral similarity feature, and the pixel-level infrared spectral feature is denoted as d i＝1 , and the remaining superpixel-level empty spectral similarity features are denoted as d i＝2，3，… .

[0008] A low-rank matrix decomposition method is used to respectively extract the pixel-level infrared spectral feature d i＝1 and the superpixel-level empty spectral similarity features d i＝2，3，… .i＝1 and all superpixel-level spatial spectral similarity features d j＝2，3，… Extract their respective thermal fault regions s i＝1 s i＝2，3，… ;

[0009] All extracted thermal fault regions are merged to obtain the thermal fault regions of the power equipment.

[0010] Furthermore, the low-rank matrix decomposition model used to extract the thermal fault region is as follows:

[0011]

[0012] Where i = 1, 2, 3…; s i and l i Representing feature d respectively i The sparse and low-rank terms in the equation correspond to the thermal fault region and the background region, respectively; ||·||0 represents the zero norm, and λ is the regularization parameter.

[0013] Solving the above low-rank matrix factorization model yields all the feature d. i thermal fault region s i .

[0014] Furthermore, the method for solving the low-rank matrix factorization model is as follows:

[0015] First, the non-convex optimization problem shown in equation (1) is relaxed by replacing the rank of the matrix with the matrix kernel norm and replacing the 0 norm with the 1 norm, thus obtaining the convex optimization problem shown in equation (2) below:

[0016]

[0017] Then, the GoDec algorithm is used to solve the optimization problem shown in equation (2):

[0018]

[0019] In the formula, ||·|| F Let represent the Frobenius norm, rank(·) and card(·) represent calculating the rank and sparsity of the matrix, respectively; r and k represent the upper limit thresholds for the rank and sparsity of the matrix.

[0020] Finally, the optimization problem shown in equation (3) is decomposed into two sub-optimization problems, and solved alternately until convergence. The sparse term at this point is the obtained thermal fault region. The two sub-optimization problems obtained by decomposition are:

[0021]

[0022]

[0023] In the formula, t in the upper right superscript represents the number of iterations.

[0024] Further, the value range of the upper limit threshold r of the matrix rank is [3, 10], and the value range of the upper limit threshold k of the matrix sparsity is [0.1, 0.5].

[0025] Further, the method for fusing all the extracted thermal fault regions is: performing splicing on all the thermal fault regions s i＝2，3，… of the superpixels, and performing average summation with the thermal fault region s i＝1 of the pixel-level infrared spectrum feature to obtain the final thermal fault diagnosis result, which is represented as:

[0026]

[0027] wherein s final is the thermal fault region of the power equipment obtained by fusion, N sup is the number of superpixels; and the value of the thermal fault region s is the gray value of each pixel in the respective region, and the size is [0, 1], wherein the greater the gray value is, the greater the probability of abnormal overheating is, and vice versa, the smaller the gray value is, the greater the probability of background is; and the addition of the regions is the addition of the gray values of the corresponding pixels.

[0028] Further, the method for dividing the infrared image into superpixels by using the superpixel segmentation algorithm is: first, acquiring the first principal component in the infrared image by principal component analysis and forming a base image; then, extracting a preset number of superpixels in the base image by using the entropy rate over-segmentation method, and each superpixel is composed of a plurality of pixels with similar gray values.

[0029] Further, the preset number of superpixels is N sup =m*n / 100, m and n are the length and width dimensions of the infrared image of the power equipment, respectively.

[0030] An electronic device includes a memory and a processor, the memory stores a computer program, and the computer program is executed by the processor to enable the processor to implement the power equipment thermal fault diagnosis method based on decision-level fusion according to any one of the technical solutions.

[0031] A computer readable storage medium stores a computer program, and the computer program is executed by a processor to implement the power equipment thermal fault diagnosis method based on decision-level fusion according to any one of the technical solutions.

[0032] Compared with the prior art, the technical effect of the present application is that the pixel-level infrared spectrum information and the super-pixel-level spatial similarity information are combined through the decision-level fusion technology, thereby improving the thermal fault diagnosis accuracy in the structure-rich area and the smooth area at the same time, obtaining a thermal fault diagnosis result with higher accuracy, solving the diagnosis problem under the condition of image noise and background interference, meeting the real-time and high-precision monitoring requirements of thermal faults, and solving the problem of low efficiency of artificial inspection, improving the intelligent operation and inspection level of power equipment. It has great significance and practical value in power intelligent inspection and operation. BRIEF DESCRIPTION OF DRAWINGS

[0033] Figure 1 is the principle diagram of the method described in the embodiments of the present application;

[0034] Figure 2 is a thermal fault diagnosis result comparison diagram of the embodiments of the present application, wherein (a) is an original input infrared image, (b) is a fault diagnosis result obtained by using only pixel-level features, (c) is a thermal fault diagnosis result obtained by splicing the results corresponding to all super-pixel-level features, and (d) is a fault diagnosis result obtained by the method of the present application. DETAILED DESCRIPTION

[0035] The technical solutions in the embodiments of the present application will be described in detail and completely with reference to the drawings in the embodiments of the present application. As shown in the drawings, Figure 1 the input is an infrared image of a power equipment, and the output is a thermal fault diagnosis result image. The specific implementation details are as follows:

[0036] Step 1, obtaining an infrared image of a power equipment, using the infrared image as a pixel-level infrared spectrum feature, using a super-pixel segmentation algorithm to segment the infrared image into a plurality of super-pixels, taking each super-pixel as a 1 empty spectrum similarity feature; the pixel-level infrared spectrum feature is denoted as d i＝1 , and the remaining super-pixel-level empty spectrum similarity features are denoted as d i＝2，3，… .

[0037] The pixel-level infrared spectrum feature constituted by the infrared image can effectively reflect the spectral image of the infrared image, wherein the pixel value of each pixel point corresponds to the temperature information of the pixel point. The super-pixel-level infrared spectrum feature constituted by the super-pixel can effectively reflect the spatial similarity information of the infrared image.

[0038] Wherein, the method for segmenting the infrared image into a plurality of superpixels by using a superpixel segmentation algorithm is as follows: first, a first principal component in the infrared image is obtained by principal component analysis and a base image is formed; then, an entropy rate over-segmentation method is used to extract a preset number of superpixels in the base image, each superpixel being composed of a plurality of pixels with similar gray values, and each superpixel effectively representing spatial similar information in the infrared image. Each superpixel Y(i) is reshaped into a matrix Y(i) e R B×P , wherein B=3, indicating three channels of the input infrared pseudo-color image, and P is the number of pixels of each superpixel.

[0039] In the embodiment, the preset number of superpixels is N sup =m*n / 100, m and n are respectively the length and width dimensions of the infrared image of the power equipment. Therefore, all superpixel-level spatial-spectral similarity features are denoted as

[0040] Step 2, using a low-rank matrix decomposition method, the respective thermal fault regions s i＝1 , s i＝2，3，… are extracted from the pixel-level infrared spectral features d i＝1 and all superpixel-level spatial-spectral similarity features d j＝2，3，… .

[0041] Based on the temperature difference characteristics of the thermal fault region and the image background region in the infrared image, the low-rank matrix decomposition method can be used to calculate the thermal fault region in the pixel-level infrared spectral features and the superpixel-level spatial-spectral similarity features.

[0042] The proportion of the number of pixels of the background pixels in the infrared image is large, and a few pixels can generate the background information by linear combination, so the background information presents a low-rank characteristic. Since the thermal fault region occupies a small proportion of the number of pixels in the entire input image, it is difficult to be generated by linear combination of a few pixels, and therefore presents a sparse characteristic. In order to diagnose the thermal fault region in the infrared image, the low-rank matrix decomposition method is used to process the input infrared image d, which is decomposed into a low-rank term l and a sparse term s:

[0043] d=l+s

[0044] The purpose of the low-rank matrix decomposition method is to separate the thermal fault region s which presents a sparse distribution from the background component l which presents a low-rank characteristic, so as to realize the initial diagnosis of the thermal fault region s.

[0045] In the embodiment, the matrix decomposition technology is used to model the thermal fault region as follows:

[0046] min l，s rank(l)+λ||s||0, s.t.d=l+s (1)

[0047] Since the above formula is a non-convex optimization problem, it is difficult to determine the optimal solution. In this embodiment, the matrix kernel norm is used to replace the matrix rank, and the 1 norm is used to replace the 0 norm. λ is a regularization parameter that adjusts the weight of the sparse terms. Therefore, the above formula can be rewritten as:

[0048] min l，s ||l|| * +λ||s||1,std=l+s (2)

[0049] λ is a balance factor with a magnitude of m and n are the length and width of the input infrared image. To solve the above formula, this embodiment uses the GoDec algorithm:

[0050] min l，s ||dls|| F ,strank(l)≤r,card(s)≤k (3)

[0051] In the formula, ||·|| F Let represent the Frobenius norm, rank(·) and card(·) represent calculating the rank and sparsity of the matrix, respectively; r and k represent the upper limit thresholds for the rank and sparsity of the matrix. In this embodiment, the upper limit threshold r for the rank of the matrix is ​​set empirically in the range of [3, 10], and the upper limit threshold k for the sparsity of the matrix is ​​set empirically in the range of [0.1, 0.5].

[0052] Finally, the optimization problem shown in equation (3) is decomposed into two sub-optimization problems, and solved alternately until convergence. The sparse term at this point is the obtained thermal fault region. The two sub-optimization problems obtained by decomposition are:

[0053]

[0054]

[0055] In the formula, t in the superscript represents the number of iterations.

[0056] Following the same method described above for extracting thermal fault regions from infrared images (i.e., pixel-level infrared spectral features), thermal fault regions are extracted from each extracted superpixel (i.e., superpixel-level spatial spectral similarity features) to obtain the thermal fault regions in each superpixel.

[0057] Step 3: Merge all extracted thermal fault regions to obtain the thermal fault regions of the power equipment.

[0058] Infrared image d i＝1 The thermal fault region in the middle is denoted as s.i＝1 , all the thermal fault areas in all superpixels (d i＝2，3，… ) are recorded as s i＝2，3，… The method for fusing all the thermal fault areas is: splicing the thermal fault areas s j＝2，3，… of all superpixels, and then performing average summation with the thermal fault areas s j＝1 of the pixel-level infrared spectrum features, so as to sufficiently mine the complementary information of the pixel-level and superpixel-level initial diagnostic results, and obtain the final thermal fault diagnostic result, thereby improving the thermal fault diagnostic performance.

[0059]

[0060] wherein s final is the thermal fault area of the power equipment obtained by fusion, N sup is the number of superpixels, N sup =m*n / 100, m and n are respectively the length and width dimensions of the infrared image of the power equipment. The value of the thermal fault area s

[0061] Figure 2 The comparative diagram of the thermal fault diagnostic result of the present application is given, wherein (a) is the original input infrared image, (b) is the fault diagnostic result obtained by using only the pixel-level features, (c) is the thermal fault diagnostic result obtained by splicing the corresponding results of all superpixel-level features, and (d) is the fault diagnostic result obtained by the method of the present application.

[0062] The above embodiments are the preferred embodiments of the present application, and those skilled in the art can also make various transformations or improvements on the basis of the above embodiments, and these transformations or improvements shall all belong to the scope of protection of the present application without departing from the general concept of the present application.

Claims

1. A method for diagnosing thermal faults in power equipment based on decision-level fusion, characterized in that, include: Infrared images of power equipment are acquired, and the infrared images are used as pixel-level infrared spectral features. The infrared images are divided into several superpixels using a superpixel segmentation algorithm. Each superpixel is composed of several pixels with similar gray values, and each superpixel is used as a spatial spectral similarity feature. Let the pixel-level infrared spectral features be denoted as d. i＝1 The remaining superpixel-level spatial spectral similarity features are denoted as d. i＝2,3,… ; Using the low-rank matrix factorization method, pixel-level infrared spectral features d are analyzed. i＝1 And all superpixel-level spatial spectral similarity features dj ＝2,3,... Extract their respective thermal fault regions s i＝1 s i＝2,3,... ; All extracted thermal fault regions are merged to obtain the thermal fault regions of the power equipment. All superpixel thermal fault areas s i＝2,3, ...and then stitch them together with the pixel-level infrared spectral features of the thermal fault region s i＝1 The final thermal fault diagnosis result is obtained by averaging and summing, as follows: Among them, s final This refers to the thermal fault region of the power equipment obtained through fusion, N sup Number of superpixels; thermal fault area The value refers to the grayscale value of each pixel in its respective region, with a size of [0,1]. The larger the grayscale value, the greater the probability of it being abnormally overheated. Conversely, the smaller the grayscale value, the greater the probability of it being the background. The sum of the regions is the sum of the grayscale values ​​of the corresponding pixels.

2. The method according to claim 1, characterized in that, The low-rank matrix decomposition model used to extract the thermal fault region is as follows: Where i = 1, 2, 3…; s i and l i Representing feature d respectively i The sparse and low-rank terms in the equation correspond to the thermal fault region and the background region, respectively; ||·|0 represents the zero norm, and λ is the regularization parameter. Solving the above low-rank matrix factorization model yields all the feature d. i thermal fault region s i .

3. The method according to claim 2, characterized in that, The method for solving the low-rank matrix factorization model is as follows: First, the non-convex optimization problem shown in equation (1) is relaxed by replacing the rank of the matrix with the matrix kernel norm and replacing the 0 norm with the 1 norm, thus obtaining the convex optimization problem shown in equation (2) below: Then, the GoDec algorithm is used to solve the optimization problem shown in equation (2): In the formula, ‖·‖ F Let represent the Frobenius norm, rank(·) and card(·) represent calculating the rank and sparsity of the matrix, respectively; r and k represent the upper limit thresholds for the rank and sparsity of the matrix. Finally, the optimization problem shown in equation (3) is decomposed into two sub-optimization problems, and solved alternately until convergence. The sparse term at this point is the obtained thermal fault region. The two sub-optimization problems obtained by decomposition are: In the formula, t in the superscript represents the number of iterations.

4. The method according to claim 3, characterized in that, The upper limit threshold r of the matrix rank is in the range of [3, 10], and the upper limit threshold k of the matrix sparsity is in the range of [0.1, 0.5].

5. The method according to claim 1, characterized in that, The method for segmenting infrared images into superpixels using a superpixel segmentation algorithm is as follows: First, the first principal component in the infrared image is obtained through principal component analysis and a base image is formed; then, the entropy rate oversegmentation method is used to extract a preset number of superpixels from the base image.

6. The method according to claim 5, characterized in that, The preset number of superpixels is N. sup =m*n / 100, where m and n are the length and width dimensions of the infrared image of the power equipment, respectively.

7. An electronic device, comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the computer program is executed by the processor, the processor causes the processor to implement the method as described in any one of claims 1 to 6.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1 to 6.

Images