Insulator contamination level assessment method for insulator operation based on infrared image and noise

Abstract

The present invention provides an insulator pollution level assessment method based on infrared images and noise, comprising the following steps: S1. obtaining noise signals and infrared image information of sample insulators when they are running under different ambient temperatures, different humidity and different pollution levels; S2. preprocessing the infrared image and the noise signal respectively; S3. extracting temperature characteristic parameters from the preprocessed image and extracting time domain characteristic parameters from the noise signal after filtering; S4. performing wavelet transform on the noise signal after filtering and extracting wavelet characteristic parameters within a set frequency band; S5. constructing a first probability neural network with the same structure. network and a second probabilistic neural network, and input the temperature characteristic parameters, ambient temperature and ambient humidity as input features into the first probabilistic neural network to train the probabilistic neural network; input the time domain characteristic parameters, wavelet characteristic parameters, ambient temperature and humidity as characteristic parameter matrices into the second probabilistic neural network for training; S6. Real-time acquisition of infrared images and noise information of the running insulators, after processing in steps S2-S4, input them into the first probabilistic neural network and the second probabilistic neural network for processing, and construct a probability distribution model M(A), and screen out the pollution level corresponding to the maximum value of the M(A) value as the final evaluation result.

Description

Technical Field

[0001] The invention relates to an insulator contamination level assessment method, in particular to an insulator operation insulator contamination level assessment method based on infrared images and noise. Background Art

[0002] Insulators are widely used in transmission lines, power generation and substations, and are an important part of the power system. Due to the increasing environmental and atmospheric pollution, insulators are directly exposed to the air during operation, and dust particles are continuously deposited on the surface of the insulators, eventually attaching a layer of dirt. When dirty insulators operate in weather conditions with high air humidity such as haze, condensation, and drizzle, their insulation performance will be greatly reduced, and flashover accidents are prone to occur under normal operating voltage.

[0003] In order to prevent the occurrence of flashover accidents, the insulators need to be decontaminated. Before the decontamination, the pollution level of the insulators needs to be accurately assessed, so as to provide a reference for the formulation of decontamination measures. In the prior art, the assessment of the pollution level of insulators includes manual estimation and online monitoring. Manual estimation is achieved through staff inspections, etc. This method is affected by human subjective factors and has low accuracy. It is also costly and inefficient. Online methods, such as image monitoring methods, are easily affected by environmental interference and their accuracy is also limited. There are also leakage current detection methods, pulse current detection methods, etc. These methods are easily affected by electromagnetic interference and have low accuracy.

[0004] Therefore, in order to solve the above technical problems, it is urgent to propose a new technical means. Summary of the invention

[0005] In view of this, the purpose of the present invention is to provide an insulator pollution level assessment method based on infrared images and noise. The heat and noise generated by the insulator during operation and discharge after pollution accumulation are used as the basis for judging the pollution level of the insulator. Combined with the real-time ambient temperature and humidity, the accuracy of the final judgment result can be effectively improved, avoiding interference from the electromagnetic environment in the traditional method, and it is highly efficient and reliable.

[0006] The present invention provides an insulator pollution level assessment method for insulator operation based on infrared images and noise, comprising the following steps:

[0007] S1. Obtain the noise signal and infrared image information of the sample insulator when it is running under different ambient temperatures, different humidity and different pollution levels;

[0008] S2. Preprocessing the infrared image and the noise signal respectively;

[0009] S3. Extracting temperature characteristic parameters from the preprocessed image and extracting time domain characteristic parameters from the filtered noise signal;

[0010] S4. Performing wavelet transform on the filtered noise signal and extracting wavelet characteristic parameters within a set frequency band;

[0011] S5. Construct a first probabilistic neural network and a second probabilistic neural network with the same structure, and input the temperature characteristic parameters, ambient temperature and ambient humidity into the first probabilistic neural network to train the probabilistic neural network; input the time domain characteristic parameters, wavelet characteristic parameters, ambient temperature and humidity into the second probabilistic neural network for training;

[0012] S6. Infrared images and noise information of operating insulators are collected in real time, and after being processed in steps S2-S4, they are input into the first probabilistic neural network and the second probabilistic neural network for processing, and a probability distribution model is constructed:

[0013] The pollution level corresponding to the maximum value of M(A) is selected as the final evaluation result, where A i is the pollution level corresponding to infrared noise, i = 1, 2, 3, 4, 5; B j is the pollution level corresponding to the noise signal, j = 1, 2, 3, 4, 5; θ represents B j and A i The union of , φ represents the empty set, m1(A i ) and m2(B j ) is calculated in the same way, Among them, the probabilistic neural network consists of an input layer, a pattern layer, a summation layer, and an output layer. The input vector of the input layer is The output vector of the pattern layer is The output vector of the summation layer is

[0014] Further, the infrared image is preprocessed including:

[0015] S2A1. Convert the infrared image to HSV color space and extract the S component information;

[0016] S2A2 performs bilateral filtering on the S component information;

[0017] S2A3 morphological reconstruction opening and closing algorithm is used to reconstruct the S component information after step S22 to obtain a reconstructed image;

[0018] S2A4. Calculate the local maximum of the reconstructed image and use the local maximum as the foreground mark of the reconstructed image;

[0019] The reconstructed image is converted into a binary image, and the binary image is subjected to distance transformation to obtain a vector distance matrix of the binary image, and the vector distance matrix is ​​subjected to watershed transformation to obtain a boundary line between a foreground area and a background area of ​​the reconstructed image and the boundary line is used as a background mark;

[0020] S2A5. Convert the infrared image into RGB color space, and use Sobel operator to calculate the gradient of R component of RGB color space and S component of HSV color space respectively;

[0021] S2A6. Compare the gradient values ​​of the corresponding pixels of the R component and the S component, select the larger gradient value of the two components as the gradient value of the current pixel, and synthesize the RS component gradient map;

[0022] S2A7. Use the imimposemin function to correct the RS component gradient map so that the corrected RS component gradient map has a minimum value at the foreground mark and the background mark, thereby obtaining the corrected RS component gradient map;

[0023] The watershed transform is performed on the modified RS component gradient map to segment the insulator area in the infrared image.

[0024] Furthermore, the temperature characteristic parameters are extracted from the preprocessed image, and the extraction specifically includes:

[0025] S3A1: determining a target region from the segmented insulator region and converting the target region into a binary matrix;

[0026] S3A2. Extract the temperature matrix of the insulator area from the infrared image and the temperature matrix and the binary matrix have the same order;

[0027] S3A3. Perform dot multiplication of the binary matrix and the temperature matrix to obtain the temperature rise matrix of the target area, and extract temperature characteristic parameters from the temperature rise matrix, including: maximum value, minimum value, average value, median value, mode value, root mean square, range, variance, standard deviation and kurtosis.

[0028] Furthermore, the time domain characteristic parameters include the peak value, average value, standard deviation, root mean square, kurtosis, peak factor, pulse factor and margin factor of the noise;

[0029] Wavelet characteristic parameters include frequency band energy ratio and information entropy.

[0030] Further, in step S2, preprocessing the noise signal includes:

[0031] The noise signal is processed by Fourier transform and then filtered by a band-stop filter;

[0032] The filtered noise signal is then subjected to global filtering using a Gaussian weighted moving average algorithm.

[0033] Further, step S5 specifically includes:

[0034] When performing wavelet transform on the noise signal after filtering, the wavelet basis function is set to db6 and the number of decomposition levels is set to 3;

[0035] The noise signal after wavelet transformation is divided into two J sub-bands, and determine the energy value of each sub-band

[0036] Calculate the ratio of the energy of the set frequency band to the total energy

[0037] in: Indicates the energy value of the mth set frequency band;

[0038] Determine the waveform sequence X of the noise signal and calculate the information entropy H(X):

[0039] H(X)=-∑q i log a q i ;q i is the probability of occurrence of the ith sub-band in the waveform sequence, and a is the set calculation coefficient.

[0040] Beneficial effects of the present invention: Through the present invention, the heat and noise generated by the insulator during discharge after pollution accumulation are used as the basis for judging the pollution level of the insulator, combined with the real-time ambient temperature and humidity, so as to effectively improve the accuracy of the final judgment result, avoid interference such as electromagnetic environment in traditional methods, and have high efficiency and reliability. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] The present invention will be further described below in conjunction with the accompanying drawings and embodiments:

[0042] Figure 1 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0043] The present invention is further described in detail below:

[0044] The present invention provides an insulator pollution level assessment method for insulator operation based on infrared images and noise, comprising the following steps:

[0045] S1. Obtain the noise signal and infrared image information of the sample insulator when it is running under different ambient temperatures, different humidity and different pollution levels;

[0046] S2. Preprocessing the infrared image and the noise signal respectively;

[0047] S3. Extracting temperature characteristic parameters from the preprocessed image and extracting time domain characteristic parameters from the filtered noise signal;

[0048] S4. Performing wavelet transform on the filtered noise signal and extracting wavelet characteristic parameters within a set frequency band;

[0049] S5. Construct a first probabilistic neural network and a second probabilistic neural network with the same structure, and input the temperature characteristic parameters, ambient temperature and ambient humidity into the first probabilistic neural network to train the probabilistic neural network; input the time domain characteristic parameters, wavelet characteristic parameters, ambient temperature and humidity into the second probabilistic neural network for training;

[0050] S6. Infrared images and noise information of operating insulators are collected in real time, and after being processed in steps S2-S4, they are input into the first probabilistic neural network and the second probabilistic neural network for processing, and a probability distribution model is constructed:

[0051] The pollution level corresponding to the maximum value of M(A) is selected as the final evaluation result, where A i is the pollution level corresponding to infrared noise, i = 1, 2, 3, 4, 5; B i is the pollution level corresponding to the noise signal, j = 1, 2, 3, 4, 5; θ represents B j and A i The union of , φ represents the empty set, m1(A i ) and m2(B j ) is calculated in the same way, Among them, the probabilistic neural network consists of an input layer, a pattern layer, a summation layer, and an output layer. The input vector of the input layer is The output vector of the pattern layer is The output vector of the summation layer is

[0052] Through the above method, the heat and noise generated by the insulator during discharge after pollution accumulation are used as the basis for judging the pollution level of the insulator. Combined with the real-time ambient temperature and humidity, the accuracy of the final judgment result can be effectively improved, avoiding interference such as electromagnetic environment in traditional methods, and it is highly efficient and reliable.

[0053] In the above, the sample insulator is set in the laboratory. A type of insulator is selected as the sample insulator, such as a double umbrella porcelain insulator, and then different ambient temperatures, ambient humidity and different pollution levels are simulated in the laboratory, where:

[0054] The ambient temperature is generally set to 5℃-25℃. This temperature range can be divided into multiple temperature intervals. Corresponding experiments are carried out in each temperature interval. Generally, two intervals of 5℃-15℃ and 15℃-25℃ can meet the needs of sample data.

[0055] Humidity refers to relative humidity, which is generally 60% to 100%; it is generally divided into three intervals to meet the needs of sample data, namely 65%-75%, 75%-85%, and 85%-95%;

[0056] There are five levels of pollution, namely, level 0, level Ⅰ, level Ⅱ, level Ⅲ, and level Ⅳ. 0.03mg / cm2, 0.05mg / cm2, 0.08mg / cm2, 0.15mg / cm2, and 0.25mg / cm2 are selected to represent the equivalent salt density of level 0, level Ⅰ, level Ⅱ, level Ⅲ, and level Ⅳ, respectively. The ratio of ash density to salt density is 6:1. Sodium chloride, diatomaceous earth, and silicon dioxide are used to simulate pollutants.

[0057] Under the above conditions, a DC voltage is applied to the insulator to obtain corresponding noise and infrared images under different conditions. The probabilistic neural network can use existing technology.

[0058] In this embodiment, preprocessing the infrared image specifically includes:

[0059] S2A1. Convert the infrared image to HSV color space and extract the S component information;

[0060] S2A2. Perform bilateral filtering on the S component information to filter out noise interference, where:

[0061] The calculation formula of bilateral filtering is as follows:

[0062]

[0063] Among them: u(x), u BF (x) are the image matrices before and after bilateral filtering, w d (x, y) is the filter coefficient determined by the geometric space distance, w r (x, y) is the filter coefficient determined by the grayscale difference of the pixel, C d，r It is the normalization coefficient, which refers to the number of pixels in the calculation area each time.

[0064] S2A3. The S component information after step S22 is reconstructed by using a morphological reconstruction opening and closing algorithm to obtain a reconstructed image; wherein, through this step, the pseudo local extrema caused by fine texture and noise interference can be effectively removed, and the complete contour information of the insulator is retained, so as to provide a guarantee for the accuracy of subsequent processing, wherein the morphological reconstruction opening and closing algorithm is a prior art and will not be described in detail here;

[0065] S2A4. Calculate the local maximum of the reconstructed image and use the local maximum as the foreground mark of the reconstructed image;

[0066] The reconstructed image is converted into a binary image, and the binary image is transformed to obtain the vector distance matrix of the binary image. The vector distance matrix is ​​transformed to obtain the boundary between the foreground area and the background area of ​​the reconstructed image and the boundary is used as the background mark; wherein, the binary image is transformed by Euclidean distance, D is the distance between each pixel and the nearest non-zero pixel, and the calculation formula is as follows:

[0067]

[0068] Where: p(x1, y1), q(x2, y2) are the coordinates of the foreground and background pixel values ​​respectively. The watershed transform is the existing watershed algorithm, which will not be described in detail here;

[0069] S2A5. Convert the infrared image into RGB color space, and use the Sobel operator to calculate the gradient of the R component of the RGB color space and the S component of the HSV color space respectively; wherein the Sobel operator and the process of calculating the gradient thereof are prior art and will not be described in detail here;

[0070] S2A6. Compare the gradient values ​​of the corresponding pixels of the R component and the S component, select the larger gradient value of the two components as the gradient value of the current pixel, and synthesize the RS component gradient map; for example: the gradient of the R component and the gradient of the S component are both descriptions of the pixel points, such as an image with 3*3 pixels, that is Then m 11 If the gradient value of the R component of the pixel is greater than the gradient value of the S component, then the gradient value of this pixel is the gradient value of the R component. 12 If the gradient value of the R component of the pixel is less than the gradient value of the S component, then the gradient value of this pixel is the gradient value of the S component, and so on, thereby merging the R component gradient map and the S component gradient map into an RS component gradient map.

[0071] S2A7. The imimposemin function is used to correct the RS component gradient map so that the corrected RS component gradient map has minimum values ​​at the foreground mark and the background mark, thereby obtaining the corrected RS component gradient map; through the above method, the interference of pseudo minimum values ​​on image segmentation can be reduced, and the accuracy of the final result can be improved.

[0072] The watershed transform is performed on the modified RS component gradient map to segment the insulator area in the infrared image.

[0073] In this embodiment, extracting temperature characteristic parameters from the preprocessed image specifically includes:

[0074] S3A1: determining a target region from the segmented insulator region and converting the target region into a binary matrix;

[0075] S3A2. Extract the temperature matrix of the insulator area from the infrared image and the temperature matrix and the binary matrix have the same order;

[0076] S3A3. Perform dot multiplication of the binary matrix and the temperature matrix to obtain the temperature rise matrix of the target area, and extract temperature characteristic parameters from the temperature rise matrix, including: maximum value, minimum value, average value, median value, mode value, root mean square, range, variance, standard deviation and kurtosis.

[0077] In this embodiment, the time domain characteristic parameters include the peak value, average value, standard deviation, root mean square, kurtosis, peak factor, pulse factor and margin factor of the noise;

[0078] Wavelet characteristic parameters include frequency band energy ratio and information entropy.

[0079] In step S2, preprocessing the noise signal includes:

[0080] The noise signal is processed by Fourier transform, and then filtered by a band-stop filter; wherein: the process of Fourier transform is the prior art;

[0081] The filtered noise signal is then subjected to global filtering using a Gaussian weighted moving average algorithm, wherein the Gaussian weighted moving average algorithm is an existing technology, and its calculation formula is:

[0082]

[0083] Where: L is the number of sliding points, h n is the weighted average factor, y k is the filtering result.

[0084] In this embodiment, step S5 specifically includes:

[0085] When performing wavelet transform on the noise signal after filtering, the wavelet basis function is set to db6 and the number of decomposition levels is set to 3;

[0086] The noise signal after wavelet transformation is divided into two J sub-bands, and determine the energy value of each sub-band

[0087] Calculate the ratio of the energy of the set frequency band to the total energy

[0088] in: Indicates the energy value of the mth set frequency band;

[0089] Determine the waveform sequence X of the noise signal and calculate the information entropy H(X):

[0090] H(X)=-∑q i log a q i ;q i is the probability of occurrence of the ith sub-band in the waveform sequence, and a is the set calculation coefficient.

[0091] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solution of the present invention, which should be included in the scope of the claims of the present invention.

Claims

1. A method for evaluating the pollution level of insulators in operation based on infrared images and noise, characterized in that: The following steps are involved: S1. Obtain the noise signal and infrared image information of the sample insulator when it is running under different ambient temperatures, different humidity and different pollution levels; S2. Preprocessing the infrared image and the noise signal respectively; S3. Extracting temperature characteristic parameters from the preprocessed image and extracting time domain characteristic parameters from the filtered noise signal; S4. Performing wavelet transform on the filtered noise signal and extracting wavelet characteristic parameters within a set frequency band; S5. Construct a first probabilistic neural network and a second probabilistic neural network with the same structure, and input the temperature characteristic parameters, ambient temperature and ambient humidity into the first probabilistic neural network to train the probabilistic neural network; input the time domain characteristic parameters, wavelet characteristic parameters, ambient temperature and humidity into the second probabilistic neural network for training; S6. Infrared images and noise information of operating insulators are collected in real time, and after being processed in steps S2-S4, they are input into the first probabilistic neural network and the second probabilistic neural network for processing, and a probability distribution model is constructed: The pollution level corresponding to the maximum value of M(A) is selected as the final evaluation result, where A i is the pollution level corresponding to infrared noise, i = 1, 2, 3, 4, 5; B i is the pollution level corresponding to the noise signal, j = 1, 2, 3, 4, 5; m1(A i ) and m2(B j ) is calculated in the same way, Among them, the probabilistic neural network consists of an input layer, a pattern layer, a summation layer, and an output layer. The input vector of the input layer is The output vector of the pattern layer is p=1,2,3,…,q; the output vector of the summation layer is 2. The method for evaluating the pollution level of insulators in operation based on infrared images and noise according to claim 1 is characterized by: Preprocessing of infrared images includes: S2A1. Convert the infrared image to HSV color space and extract the S component information; S2A2 performs bilateral filtering on the S component information; S2A3 morphological reconstruction opening and closing algorithm is used to reconstruct the S component information after step S22 to obtain a reconstructed image; S2A4. Calculate the local maximum of the reconstructed image and use the local maximum as the foreground mark of the reconstructed image; The reconstructed image is converted into a binary image, and the binary image is subjected to distance transformation to obtain a vector distance matrix of the binary image, and the vector distance matrix is ​​subjected to watershed transformation to obtain a boundary line between a foreground area and a background area of ​​the reconstructed image and the boundary line is used as a background mark; S2A5. Convert the infrared image into RGB color space, and use Sobel operator to calculate the gradient of R component of RGB color space and S component of HSV color space respectively; S2A6. Compare the gradient values ​​of the corresponding pixels of the R component and the S component, select the larger gradient value of the two components as the gradient value of the current pixel, and synthesize the RS component gradient map; S2A7. Use the imimposemin function to correct the RS component gradient map so that the corrected RS component gradient map has a minimum value at the foreground mark and the background mark, thereby obtaining the corrected RS component gradient map; The watershed transform is performed on the modified RS component gradient map to segment the insulator area in the infrared image.

3. The method for evaluating the pollution level of insulators in operation based on infrared images and noise according to claim 2 is characterized in that: The temperature feature parameters extracted from the preprocessed image specifically include: S3A1: determining a target region from the segmented insulator region and converting the target region into a binary matrix; S3A2. Extract the temperature matrix of the insulator area from the infrared image and the temperature matrix and the binary matrix have the same order; S3A3. Perform dot multiplication of the binary matrix and the temperature matrix to obtain the temperature rise matrix of the target area, and extract temperature characteristic parameters from the temperature rise matrix, including: maximum value, minimum value, average value, median value, mode value, root mean square, range, variance, standard deviation and kurtosis.

4. The method for evaluating the pollution level of insulators in operation based on infrared images and noise according to claim 1 is characterized in that: The time domain characteristic parameters include the peak value, average value, standard deviation, root mean square, kurtosis, peak factor, pulse factor and margin factor of the noise; Wavelet characteristic parameters include frequency band energy ratio and information entropy.

5. The method for evaluating the pollution level of insulators in operation based on infrared images and noise according to claim 4 is characterized in that: In step S2, preprocessing the noise signal includes: The noise signal is processed by Fourier transform and then filtered by a band-stop filter; The filtered noise signal is then subjected to global filtering using a Gaussian weighted moving average algorithm.

6. The method for evaluating the pollution level of insulators in operation based on infrared images and noise according to claim 5 is characterized by: Step S5 specifically includes: When performing wavelet transform on the noise signal after filtering, the wavelet basis function is set to db6 and the number of decomposition levels is set to 3; The noise signal after wavelet transformation is divided into two J sub-bands, and determine the energy value of each sub-band Calculate the ratio of the energy of the set frequency band to the total energy in: Indicates the energy value of the mth set frequency band; Determine the waveform sequence X of the noise signal and calculate the information entropy H(X): H(X)=-∑q i log a q i ;q i is the probability of occurrence of the ith sub-band in the waveform sequence, and a is the set calculation coefficient.

Images