Substation noise control method

Abstract

The invention discloses a transformer substation noise control method, which relates to the field of noise monitoring and comprises the following steps of: detecting and feeding back rough noise distribution around a transformer substation through a sound source equivalent model; analyzing an uncertainty source in noise distribution detection, and establishing an uncertainty model to calibrate the accuracy of noise distribution; substituting the noise distribution result into the infrared image, and processing a noise abnormal part by an infrared image segmentation method; accurate noise distribution and fault point reproduction around the transformer substation are completed through a network drawing algorithm, and transformer substation noise prevention and control are achieved. According to the transformer substation noise monitoring method, the noise distribution around the to-be-monitored transformer substation is confirmed through the sound source equivalent model, the transformer substation noise can be monitored in real time, and the operation condition of the transformer substation can be comprehensively reflected.

Description

Technical Field

[0001] The invention relates to the field of noise monitoring, and in particular to a method for preventing and controlling noise in a substation. Background Art

[0002] In the power transmission system, under the same rated power conditions, the higher the transmission voltage, the smaller the corresponding transmission current, and thus the smaller the energy loss of the line. Therefore, high-voltage and ultra-high-voltage transmission lines are the key development direction of current and future power construction. In the high-voltage transmission system, an ultra-high electric field will be generated near the conductor, causing the nearby air to ionize and discharge. The air discharge will produce a "hissing" sound, forming corona noise. At the same time, due to the periodic change of voltage, the charged particles near the conductor move periodically under the action of the electric field, and the movement period is the same as the alternating current period. In my country, the AC power frequency is 50Hz. The charged particles collide with air molecules to form sound waves that radiate outward. This periodic sound pressure change constitutes the body noise. Due to the nonlinear effect, the body noise also includes the harmonic signals of the power frequency. High-voltage transmission transformers generate a lot of heat, which increases the temperature of the transformer and affects the service life of the transformer. Generally, high-voltage transformers are equipped with cooling facilities, mainly cooling fans, which generate noise, called fan noise. In large high-voltage substations, due to the relatively concentrated transformers and high-voltage transmission lines, the noise in the substation will have an adverse impact on the production and life of the surrounding residents. According to the above analysis, the noise in the high-voltage substation mainly includes corona noise, body noise and fan noise. The three types of noise have different noise sources and exist in the substation at the same time. At present, there is no method to measure the three types of noise separately. Therefore, a new method is urgently needed to solve the current situation of substation noise.

[0003] Chinese patent document CN104537227B discloses a "substation noise separation method". A microphone is used to pick up the noise of the high-voltage substation. The noise of the high-voltage substation includes corona noise, body noise, and fan noise. Its signal model is: s(n) = c(n) + b(n) + e(n), w0 = 2πf0 / fs, where s(n) is the high-voltage substation noise, c(n) is the corona noise, b(n) is the body noise, e(n) is the fan noise, w0 is the digital domain frequency, f0 is the body vibration fundamental frequency, f0 = 50Hz, fs is the signal sampling frequency, and P is the harmonic individual number; Bk is the harmonic amplitude, φk is the harmonic phase, and n is the discrete sampling time index value; 2) Multilayer wavelet transform is performed on s(n), and six groups of wavelet coefficients are obtained: si(n), i = 1, 2, ... 6; among them, s1(n) is the low-frequency wavelet coefficient, and the frequency band corresponding to s1(n) is [0, fs / 64], and the frequency ranges corresponding to the other five groups of wavelet coefficients si(n), i = 2, ... 6 are [fs / 28-i, fs / 27-i]; among them, the wavelet transform The wavelet transform filter group used in the transformation is a Daubechies filter group, and the wavelet coefficients are calculated using the Mallat fast algorithm; 3) after s1(n) passes through a passband comb filter, an inverse wavelet transform is performed to obtain an estimated signal of the body noise and the separated body noise, and the passband frequency of the passband comb filter is the fundamental frequency f0 of the body noise and each harmonic frequency; 4) after s1(n) passes through a stopband comb filter, an inverse wavelet transform is performed with another five groups of wavelet coefficients to obtain a mixed signal y(n) containing corona noise and fan noise; 5) s5(n) and s6(n) are inversely transformed with wavelets to obtain a corona detection signal sd(n); 6) sd(n) is used to detect whether there is corona noise, and according to the detection result, the fan noise power spectrum is estimated in the time period without corona noise, and in the time period with corona noise, the power spectrum of y(n) is subtracted from the estimated fan noise power spectrum to obtain the estimated corona noise power spectrum, and the power spectrum calculation is realized by fast Fourier transform. However, the accuracy of noise detection in this patent is insufficient, and there is no subsequent solution. Summary of the invention

[0004] The present invention mainly solves the technical problem that the original noise monitoring method cannot accurately locate the noise point and process the substation noise, and provides a substation noise prevention and control method. The sound source equivalent model is used to feed back the approximate noise distribution around the substation in real time, and the uncertainty model is used to analyze the source of uncertainty of the noise in the distribution and calibrate it. The calibrated noise part is substituted into the infrared image for cutting and filtering out abnormal noise results. Finally, the closing algorithm is used to accurately feed back the noise distribution around the substation to be tested and its noise fault points to facilitate subsequent fixed-point prevention and control.

[0005] The above technical problem of the present invention is mainly solved by the following technical solution: The steps of the present invention include: S1: The approximate noise distribution around the substation is fed back through the sound source equivalent model detection; S2: Analyze the sources of uncertainty in noise distribution detection and establish an uncertainty model to calibrate the accuracy of noise distribution; S3: Substitute the noise distribution result into the infrared image and process the abnormal noise by infrared image segmentation method; S4: Use the network closing algorithm to complete the accurate noise distribution around the substation and reproduce the fault point, realizing the noise prevention and control of the substation.

[0006] Preferably, in step S1, the active sound source, reflected sound source and background sound source are subjected to equivalent processing after reaching a steady state. In the equivalent processing, the sound pressure level and initial sound power of the equivalent sound source at a certain point around the substation are first determined, and then the background sound source is substituted into the sound pressure level limit formula to obtain the noise distribution condition around the substation.

[0007] Preferably, in step S2, affected by the noise detection method, various types of uncertainties appear in the noise distribution feedback process to interfere with the accuracy of noise detection, and the uncertainty is divided into Class A uncertainty and Class B uncertainty. The Class A uncertainty includes the uncertainty impact caused by the measurement method during the measurement process, and the Class B uncertainty includes the uncertainty impact caused by different substation equipment during the measurement process.

[0008] Preferably, in step S2, a Class A assessment model is established for Class A uncertainty, and the Class A assessment model determines the uncertainty of a single noise measurement and the measurement under the Class A standard. The above measurement is repeated n times to complete the improvement of the Class A assessment model. A Class B assessment model is established for Class B uncertainty. The Class B assessment model confirms the half-width of the noise measurement interval based on historical processing data, assumes the probability distribution of the noise measured value, completes the uncertainty of the Class B standard, and repeats the above measurement n times to complete the improvement of the Class B assessment model.

[0009] Preferably, in step S2, for substation noise that can be directly measured and the measured components are unrelated, the Class A standard and the Class B standard are combined and expanded to generate an extended standard uncertainty assessment model, which integrates and sums to obtain uncertainty and multiplies it by the inclusion factor to obtain the final assessment result.

[0010] Preferably, in step S4, a compression search is performed from the boundary of the variable space through a closing algorithm, with the current global optimal solution as a reference to the noise distribution state around the substation. When searching for reproduction, the initialization grid of the variable space is used as the starting point, and a number of random nodes are selected on the boundary surface surrounded by hyperplanes at the upper and lower limits of each dimension of the initialization grid. If the total optimization space is D-dimensional, each hyperplane at the boundary takes n random nodes, and the grid contains a total of P=2nD nodes. Each initialized node always has a dimension at the upper or lower limit. After substituting each node on the grid into the objective function for calculation, the noise distribution around the substation and the reproduction of the fault point are achieved.

[0011] Preferably, in step S3, when the relevant parameters of the imaging system are not adjusted accurately, the noise will significantly weaken the visual effect of the image. After substituting the noise distribution result into the infrared image, various forms of noise can be intuitively fed back. Some noise distributions conform to a certain probability statistical model, while some noise distributions and the number of noise points are random. By segmenting the infrared image to find the abnormal noise, the interference of noise on the real information of the image can be reduced.

[0012] Preferably, the OTSU threshold segmentation method is used for segmentation in step S3 to clarify the total number of pixels of the segmented infrared image, the gray level and the number of pixels with gray value. The target image can be divided into two parts by a single threshold. The probability of the two regions is equal to the maximum value of the variance of the two regions, which is the optimal threshold. The infrared image is divided into M layers by OTSU migration to multi-threshold segmentation, and the color image is grayed. Multi-threshold OTSU segmentation is performed to remove redundant error parameters in each layer.

[0013] Preferably, in step S3, infrared images are prone to generate noise during the process of generation and transmission, and opening and closing reconstruction is performed using the structural elements on the infrared image. The opening and closing reconstruction takes the marked infrared image as the starting point of the transformation, transforms two images and a structural element, and uses morphological reconstruction to filter out different types of noise and retain image details. The image is iteratively opened and closed reconstructed and filtered through the structural elements to generate uniform grayscale features and an image that is easy to segment.

[0014] Preferably, when removing noise in step S3, the noisy image is decomposed into n layers of thresholds using wavelet transform. The standard deviation of each layer of wavelet coefficients obtained by wavelet transform of the image will change with the change of the system. If the number of decomposition layers is different, the wavelet threshold is different. When the number of decomposition layers increases, the corresponding wavelet threshold decreases.

[0015] The beneficial effects of the present invention are as follows: in the present invention, the noise distribution around the substation to be tested is confirmed by using a sound source equivalent model. Using the substation noise monitoring method provided by the present invention, the substation noise can be monitored at all times, and the operating status of the substation can be fully reflected. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 The present invention provides a flow chart of a method for monitoring noise in a substation.

[0017] Figure 2 The present invention provides a flowchart for calculating uncertainty of a substation. DETAILED DESCRIPTION

[0018] The technical solution of the present invention is further specifically described below through embodiments and in conjunction with the accompanying drawings.

[0019] Embodiment: The substation noise monitoring method of this embodiment is as follows: Figure 1 and Figure 2 As shown in the figure, the substation and perimeter noise are mainly generated by the transformers and reactors inside the substation and noise sources such as fans. The noise is transmitted outward through the propagation medium. When the sound encounters the surface and corners of the building, part of the energy will be absorbed, and the other part will continue to propagate in the form of reflection, diffraction, etc., and reach a steady state after a certain period of time. All equivalent sound sources include three types: active sound sources, reflected sound sources, and background sound sources. Among them, non-active sound sources mainly include various noise sources, such as transformers, reactors, cooling fans, etc. When performing equivalent sound source processing, if the sound power of a certain point in space is H, and the sound power of an equivalent sound source reaching this point after propagating through space is Hi, then the sound pressure level at this point is the initial sound power of the equivalent sound source, which is defined as Hi2. There is a certain attenuation after the sound source propagates in space. After a point sound source propagates in space to a landing point at a distance R, the power will decay to 4πR of the initial sound source. 2 times. The ideal line sound source will attenuate to 2πR times. However, there are many factors in the actual environment that restrict the ideal propagation form of the sound source. Therefore, when the actual sound source equivalent model is used to feedback the noise distribution, the sound power attenuation of the sound source is defined according to the actual transmission result, and the background sound source is defined as the sound power that does not change. The sound pressure level formula at point P in space is obtained. According to the sound pressure formula, the actual sound pressure level of a total of k points measured in the edge area of ​​the substation within a period of time is defined as Nk, so that the actual sound pressure of each point is infinitely close to the theoretical sound pressure. Combined with the sound source propagation attenuation, the sound source equivalent model is constructed, and the noise distribution state around the substation is restored through the sound source equivalent model. When the objective function in the equivalent model is infinitely close to 0, it means that the sound pressure level calculated by the model matches the actual situation, that is, the noise distribution around the substation is restored.

[0020] In the process of feedback noise distribution, there are many uncertain factors that affect the feedback results. Therefore, after the noise peripheral distribution is restored, it is necessary to analyze the sources of uncertainty in the noise distribution detection and establish an uncertainty model to calibrate the accuracy of the noise distribution. In this embodiment, the uncertainty is divided into Class A uncertainty and Class B uncertainty, where Class A uncertainty includes the uncertainty caused by the measurement method during the measurement process, and Class B uncertainty includes the uncertainty caused by different substation equipment during the measurement process. Class A uncertainty is mainly the uncertainty caused by the measurement method, which consists of two parts: the uncertainty of a single measurement and the uncertainty of repeated measurements. Class B uncertainty is mainly the uncertainty caused by the performance of the instrument during the noise measurement process, which mainly consists of three parts: the frequency weighting of the sound level meter, the linear deviation of the sound level meter, and the uncertainty of the calibration sound source. For Class A uncertainty, within the measurement time T specified by the noise equivalent continuous A sound level model, the noise at the same position of the substation under normal operating conditions is independently repeated n times under the same acoustic environment conditions to obtain n measurement values, and the average energy value measured is the equivalent continuous A sound level model of the substation boundary noise. The total number of samples within the measurement time T is calculated according to the formula to obtain the evaluation model of the standard uncertainty of noise. For the Class A evaluation model of standard uncertainty, the Class A evaluation method is used to evaluate the measurement uncertainty component of the value measured under the specified measurement conditions using the statistical analysis method. The calculation method of the evaluation model of the Class A uncertainty of the noise standard around the substation is as follows: the experimental standard deviation of a single measured value is carried out under repeated measurement conditions in n measurements, and the accuracy of the standard deviation is improved by accumulating the number of measurements.

[0021] For Class B uncertainty, it is necessary to establish a Class B assessment model for standard uncertainty. The Class B assessment method determines the possible value range of the measured value based on relevant information or experience. Assuming the probability distribution of the measured value, determine the assessment model of the standard Class B uncertainty based on the probability distribution and the required probability. When building the model, determine the half-width of the possible value range of the substation noise measurement, and obtain the confidence factor based on probability theory. When the confidence factor is a multiplication factor of the expanded uncertainty, it is called the inclusion factor. Evaluate the Class B uncertainty of field measurement by confidence factor. The specific test steps are as follows: Use a sound level meter to obtain the maximum allowable error or possible value range. When the upper and lower limits of the possible value range of the measured value are estimated using relevant information or experience, and the value outside the range is almost zero, if the probability of the measured value falling at any value in the range is the same, it can be assumed to be uniformly distributed, and the inclusion factor is 3. The Class B assessment model of standard uncertainty of substation boundary noise gives the maximum allowable measurement uncertainty of the sound level meter's metrological performance measurement results. The random variation of the measured value is approximately normally distributed, and the coverage factor is 1.960, and finally the Class B uncertainty is obtained.

[0022] Substitute the noise distribution result after the uncertainty model calibration into the infrared image, and extract the edge features of the image through edge detection. When extracting, search for the area where the grayscale in the image changes significantly to extract the image edge. This extraction method is widely used in image segmentation, fusion and computer vision as a basic image preprocessing step. The steps can be divided into three steps: filtering, enhancement and detection. Edge detection methods can be divided into three categories: one is various differential operators; the second is the global extraction method based on the energy minimization criterion, which realizes edge extraction from the global optimum; the third is the algorithm based on new theoretical technology, such as various edge detection algorithms based on wavelet transform, mathematical morphology and combinatorial optimization theory. Noise in infrared images exists in various forms, and the source of noise is mainly in the process of image imaging and image signal reception. For example, the disorder of particles in the internal electronic device load, overload or local short circuit of electronic components will cause noise to be generated during the imaging process. If there are design flaws in the infrared thermal imaging system, the generated image will be polluted by noise; when the relevant parameters of the imaging system are not adjusted accurately, noise will also appear. Noise will significantly weaken the visual effect of the image. Some noise distributions conform to a certain probability statistical model, while some noise distributions and the number of noise points are random. Since noise will interfere with the real information of the image, in order to improve the measurement accuracy of infrared detection, the noise in the infrared image must be filtered out. Select a gray value of a point to be detected as b, and the average gray value of the image as a. If b>a, the point to be detected with gray value b will be displayed as a bright spot in the image; if a>b, the point to be detected will be displayed as a dark spot. Usually, a and b are set as saturation values, that is, the maximum or minimum value of gray in the image. Therefore, the noise of the infrared image is a comprehensive noise composed of internal and external noise. In order to improve the image quality and reduce the impact of noise on subsequent segmentation, it is necessary to perform effective preprocessing. However, in the process of filtering and smoothing, it is inevitable that the image will be blurred and high-frequency detail information will be lost. The better the filtering effect, the higher the attenuation of the detail information, so it is necessary to balance the mutual degree of the two in the preprocessing process.

[0023] Image denoising quality evaluation is mainly divided into subjective evaluation and objective index evaluation. Subjective evaluation is to observe the clarity and color saturation of the image manually and evaluate the quality of the output image. Subjective evaluation is divided into individual evaluation and comparative evaluation. Comparative evaluation is to evaluate with reference images, while individual evaluation is to evaluate the image quality without reference images based on personal senses. For the same denoised image, the image quality determined by different judges is different. Generally, the credibility of subjective evaluation of the image can be improved by increasing the number of evaluators. The subjective evaluation method is not very persuasive due to the lack of unified indicators, so it is necessary to find a quantitative objective indicator to evaluate the quality of the denoised image. Objective evaluation is a model formula formulated based on human visual effects, and the corresponding parameters are formulated according to the image features. According to the formula, an indicator for evaluating the quality of the image is calculated. In order to better illustrate the effect of the denoising algorithm in this paper on the denoising of the motor infrared image, the commonly used mean square error, peak signal-to-noise ratio and structural similarity indicators are selected. The mean square error is one of the most commonly used indicators for analyzing the quality of denoised images. The smaller its value, the better the image denoising effect. The peak signal-to-noise ratio is an important indicator for evaluating the denoising effect of infrared images. It is used to indicate the degree of degradation before and after image denoising. The larger its value, the higher the denoising quality. The structural similarity constitutes its evaluation index by adjusting the adjustment coefficient of instability. The larger the value of structural similarity, the higher the quality of the denoised image. The existing traditional denoising methods mainly include mean filtering, median filtering and wavelet thresholding. Among them, mean filtering is a typical linear filtering algorithm. The principle is to replace the pixel value of the original target pixel with the obtained average value, which is realized by the neighborhood averaging method. In this method, the larger the filter window is, the better the noise suppression effect is. However, if the selected window size is too large, the image will become blurred. The median filter is a nonlinear filtering method that removes noise by statistical sorting. The basic step is to replace the gray value of the original target pixel with the median value of the obtained pixel gray value, which can make the surrounding pixel values ​​closer to the true value and achieve the purpose of removing noise. Therefore, both mean filtering and median filtering can be used to filter out noise and smooth the image. However, there is a clear difference between the two: mean filtering uses linear denoising, while median filtering uses nonlinear denoising. The defect of mean filtering is that the details of the image are not well protected. Although it can remove Gaussian noise, it cannot filter out salt and pepper noise. Median filtering can filter out salt and pepper noise and can effectively protect the sharp edges of the image, but its processing performance for Gaussian noise is not ideal.

[0024] The existing wavelet denoising methods are mainly divided into wavelet transform modulus maximum, adjacent scale wavelet coefficient correlation denoising and wavelet threshold denoising. Among them, the wavelet threshold denoising method with simple and easy-to-understand denoising concept and excellent denoising effect is widely used. The basic processing flow of wavelet threshold denoising is to transform the image noise signal in the space and time domain into the wavelet domain, then process the coefficients of each layer after decomposition, and finally obtain the denoised image through wavelet reconstruction. In wavelet threshold denoising, since the image and noise have different characteristics after wavelet transform, a suitable threshold needs to be set: when the amplitude of the wavelet transform coefficient of the image signal is greater than or equal to the threshold, it is considered that the coefficient is caused by the image signal itself and is retained; when the amplitude of the wavelet coefficient of the image signal is less than the threshold, it is considered to be caused by noise and is removed. By comparing different thresholds, the wavelet threshold denoising method is divided into wavelet hard threshold, soft threshold denoising and improved wavelet threshold denoising. The wavelet hard threshold denoising method optimizes the wavelet threshold coefficient and is the basis of the wavelet hard threshold denoising method. The basic idea of ​​the hard threshold function is to set the initial wavelet coefficients whose absolute values ​​are less than the threshold to zero, and the rest are not processed. When the hard threshold function is in place, the wavelet coefficients will not be compressed, there is no deviation problem, and the edge information of the image can be well preserved. The basic processing flow of the wavelet soft threshold denoising method is: the initial wavelet coefficients whose absolute values ​​are less than the threshold are still assigned to zero, but when the absolute value of the initial wavelet coefficients is greater than or equal to the threshold, it is subtracted from the threshold. When performing multi-layer wavelet threshold denoising, the image information will be concentrated in the finite wavelet transform coefficients, and the noise energy will be distributed in the entire wavelet decomposition coefficients. The traditional threshold is single and the same at each decomposition level, but in the process of wavelet decomposition of the image, the noise energy will decrease with the increase of the number of decomposition layers, and the energy of the useful image signal will become larger and larger, so the single threshold will lose the original image information in actual processing, and the effect is not good. Therefore, the noisy image wavelet is decomposed into n layers of thresholds, and the standard deviation of each layer of wavelet coefficients obtained by the image wavelet transform will change with the change of the system. If the number of decomposition layers is different, the wavelet threshold is different. When the number of decomposition layers increases, the corresponding wavelet threshold decreases, which avoids excessive denoising of the image and retains more detailed information of the original image.

[0025] After image preprocessing is completed, it needs to be further processed with the help of image segmentation. Image segmentation refers to dividing the image into several connected regions with internal consistency and different characteristics by using some feature information of the image and complying with certain similarity criteria. That is, the pixels inside the region have high similarity in grayscale, color, texture and other features, while there are obvious differences between pixels in different regions. When image segmentation is described in a set manner, an image to be segmented is set as a total set R, and each segmented region is a non-empty subset. According to the segmentation conditions, different regions are classified into different non-empty subsets to complete the image segmentation. At present, the commonly used image segmentation is divided into threshold segmentation methods, which mainly include maximum inter-class variance threshold segmentation (OTSU), iterative threshold segmentation, minimum error threshold segmentation, etc. The key to the OTSU threshold segmentation method is to find a suitable threshold to divide the image into two parts: the target and the background. If there is an infrared image with a total number of pixels N, a grayscale L, and a number of pixels with a grayscale value of Ni, then the target image can be divided into two parts, C0 and C1, by a single threshold. The probability of the two regions is w0. When the variance between the two classes is the maximum, t at this time is the optimal threshold. In the target infrared image, the grayscale of the electrical equipment in the substation is different from that of the background. If only a single threshold method is used for processing, it will be difficult to obtain the contours of each part. Therefore, OTSU is migrated to multi-threshold segmentation. When an infrared image is divided into M layers, when the grayscale image contains less information and the grayscale difference is small, the image background and the target are integrated after single threshold processing and cannot be recognized, resulting in over-segmentation. Further, the pre-processed target image is not gray-processed, and multi-threshold OTSU segmentation is performed directly; the edge detection algorithm uses a differential operator to convolve with the image to be segmented to obtain the edge contour of the image. The cluster segmentation method groups the pixels or features in the image according to certain rules and similarities to form different regions or objects for cluster segmentation. According to the actual segmentation requirements, the above segmentation method is selected to remove the abnormal noise obtained from the infrared image.

[0026] After filtering out the abnormal noise, the accurate noise distribution around the substation and the reproduction of the fault point are completed with the help of the closing algorithm. The traditional linear optimization algorithm can no longer solve the problem, so it is necessary to introduce a nonlinear optimization algorithm. The closing algorithm is an optimization method that starts from the boundary of the variable space and uses the current global optimal solution as a reference for compression search. Under the same computing scale, compared with various conventional group intelligence optimization algorithms such as improved genetic algorithms and particle swarm algorithms, it can obtain optimization results more quickly and accurately, and has good convergence. The basic process of the noise parameter identification method based on the closing algorithm is: initialize the mesh (group). The mesh is composed of a number of random nodes (feasible solutions) selected on the boundary surface surrounded by hyperplanes at the upper and lower limits of each dimension. If the total optimization space is D-dimensional, each hyperplane at the boundary takes n random nodes, then the mesh contains a total of P = 2nD nodes, that is, in this way, each initialized node always has a dimension at the upper or lower limit. Substitute each node on the network into the objective function for calculation, and obtain the current global optimal solution to iteratively update each node; cross-bounds judgment, if some dimensions of a node exceed their own upper and lower limits after update, limit them; return to the process until k outputs the optimization result. Since the current optimal node of each round is still itself after calculation, the previous "optimal node" will only be updated when a better node appears. In this way, the non-inferiority of the group can be guaranteed, that is, the optimal feasible solution obtained in each round can be guaranteed to be better than all previous feasible solutions. By updating each optimal node to improve the noise distribution of different spatial points around the substation, the relevant personnel can quickly grasp the abnormal noise points in the substation with the help of the reproduction of the fault point, and quickly and effectively propose to reduce the abnormal noise.

Claims

1. A method for preventing and controlling noise in a substation, characterized in that the steps include: S1: The approximate noise distribution around the substation is fed back through the sound source equivalent model detection; S2: Analyze the sources of uncertainty in noise distribution detection and establish an uncertainty model to calibrate the accuracy of noise distribution; S3: Substitute the noise distribution result into the infrared image and process the abnormal noise by infrared image segmentation method; S4: Use the network closing algorithm to complete the accurate noise distribution around the substation and reproduce the fault point, realizing the noise prevention and control of the substation.

2. A method for preventing and controlling noise in a substation according to claim 1, characterized in that: In step S1, the active sound source, reflected sound source and background sound source are subjected to equivalent processing after reaching a steady state. In the equivalent processing, the sound pressure level and initial sound power of the equivalent sound source at a certain point around the substation are first determined, and then the background sound source is substituted into the sound pressure level limit formula to obtain the noise distribution condition around the substation.

3. A method for preventing and controlling noise in a substation according to any one of claims 1 or 2, characterized in that: In step S2, affected by the noise detection method, various uncertainties appear in the noise distribution feedback process to interfere with the accuracy of noise detection. The uncertainty is divided into Class A uncertainty and Class B uncertainty. The Class A uncertainty includes the uncertainty impact caused by the measurement method during the measurement process, and the Class B uncertainty includes the uncertainty impact caused by different substation equipment during the measurement process.

4. A method for preventing and controlling noise in a substation according to claim 3, characterized in that: In the step S2, a Class A assessment model is established for Class A uncertainty. The Class A assessment model determines the uncertainty of a single noise measurement and the measurement under the Class A standard. The above measurement is repeated n times to complete the improvement of the Class A assessment model. A Class B assessment model is established for Class B uncertainty. The Class B assessment model confirms the half-width of the noise measurement interval based on historical processing data, assumes the probability distribution of the noise measured value, completes the uncertainty of the Class B standard, and repeats the above measurement n times to complete the improvement of the Class B assessment model.

5. A method for preventing and controlling noise in a substation according to any one of claims 3 or 4, characterized in that: In step S2, for substation noise that can be directly measured and the measured components are unrelated, the Class A standard and the Class B standard are combined and expanded to generate an extended standard uncertainty assessment model, which integrates and sums to obtain uncertainty and multiplies it by the inclusion factor to obtain the final assessment result.

6. A method for preventing and controlling noise in a substation according to claim 5, characterized in that: In step S4, starting from the boundary of the variable space, a compression search is performed with reference to the current global optimal solution through a closing algorithm to determine the noise distribution state around the substation. When searching for recurrence, the initialization grid of the variable space is used as the starting point, and a number of random nodes are selected on the boundary surface surrounded by hyperplanes at the upper and lower limits of each dimension of the initialization grid. If the total optimization space is D-dimensional, each hyperplane at the boundary takes n random nodes, and the grid contains a total of P=2nD nodes. Each initialization node always has a dimension at the upper or lower limit. Substituting each node on the grid into the objective function for calculation, the noise distribution around the substation and the reproduction of the fault point are achieved.

7. A method for preventing and controlling noise in a substation according to claim 4, characterized in that: In step S3, when the relevant parameters of the imaging system are not adjusted accurately, the noise will significantly weaken the visual effect of the image. After substituting the noise distribution result into the infrared image, various forms of noise can be intuitively fed back. Some noise distributions conform to a certain probability statistical model, while some noise distributions and the number of noise points are random. By segmenting the infrared image where the noise is abnormal, the interference of noise on the real information of the image can be reduced.

8. A method for preventing and controlling noise in a substation according to claim 7, characterized in that: In step S3, the OTSU threshold segmentation method is used for segmentation to clarify the total number of pixels of the segmented infrared image, the gray level and the number of pixels with gray value. The target image can be divided into two parts through a single threshold. The probability of the two regions is equal to the maximum value of the variance of the two regions, which is the optimal threshold. The infrared image is divided into M layers through OTSU migration to multi-threshold segmentation, and the color image is grayed. Multi-threshold OTSU segmentation is performed to remove redundant error parameters in each layer.

9. A method for preventing and controlling noise in a substation according to claim 1, characterized in that: In step S3, infrared images are prone to noise during generation and transmission, and the structural elements on the infrared images are used for opening and closing reconstruction. The opening and closing reconstruction takes the marked infrared image as the starting point of the transformation, transforms two images and a structural element, and filters out different types of noise and retains image details through morphological reconstruction. The image is iteratively opened and closed reconstructed and filtered through the structural elements to generate uniform grayscale features and an image that is easy to segment.

10. A method for preventing and controlling noise in a substation according to claim 8, characterized in that: When removing noise in step S3, the noisy image is decomposed into n layers of thresholds using wavelet transform. The standard deviation of each layer of wavelet coefficients obtained by wavelet transform will change with the change of the system. If the number of decomposition layers is different, the wavelet threshold is different. When the number of decomposition layers increases, the corresponding wavelet threshold decreases.

Images