A multi-modal online detection method and system for transformer oil

Abstract

The present invention discloses a multimodal online detection method and system for transformer oil, which relate to the technical field of transformer oil online monitoring. The method derives the dielectric constant based on Maxwell's equations and the principle of electric displacement invariance in combination with real-time voltage; adopts a priori smoothing approximation algorithm combined with a segmentation, marking and sharpening method to segment, mark and sharpen real-time visual images; adopts a filter mapping enhancement method to denoise and enhance real-time thermal images to generate real-time enhanced thermal images, and derives temperature based on regional screening and colorimetric mapping; derives acid value based on the grayscale value of a second target visual image in combination with a fluorescence standard curve, obtains the actual height of the oil column in a third target visual image through an improved edge detection and scale expansion method, and calculates the change in oil column height to reflect viscosity; generates an aging index through logistic regression analysis, and dynamically decides whether to adjust the monitoring interval or send a warning prompt based on a warning decision mechanism, thereby realizing online adaptive detection and warning of transformer oil.

Description

Technical Field

[0001] The present invention relates to the technical field of transformer oil online monitoring, and in particular to a transformer oil multi-modal online detection method and system. Background Art

[0002] Power transmission and transformation equipment is a vital component of the power system, and its operating condition is directly related to the safe and economical operation of the entire power grid. As a core component in power transmission and transformation equipment, transformers serve as the hub for power transmission and distribution. The condition of the transformer oil within them has a crucial impact on the transformer's service life and failure risk.

[0003] The existing Chinese patent application with publication number CN103512962A proposes a transformer oil and gas chromatography online detection simulation system and method. The system includes an oil sample configuration unit, an online testing unit and a comparative analysis unit; the oil sample configuration unit completes the configuration of the oil sample; the online testing unit includes a circulating oil circuit, a circulating pump, an air pump, an oil tank and an oil circuit temperature control device, and the online testing unit completes the online testing of the oil sample; the comparative analysis unit includes measurement error analysis, cross-sensitivity test, minimum detection cycle test, measurement repeatability test, data transmission test and data analysis function check.

[0004] Existing technologies for online detection of transformer oil are mostly based on separating gases in the transformer oil and indirectly judging the status of the transformer oil through chromatographic detection. However, parameters such as the acid value and dielectric constant of the transformer oil require complex experimental equipment or manual operation to be carried out. The detection cycle is long and time-consuming and labor-intensive, and it is impossible to monitor the status of the transformer oil in real time and detect potential faults in a timely manner. Summary of the Invention

[0005] In view of the deficiencies in the prior art, the present invention proposes a transformer oil multi-modal online detection method and system.

[0006] The technical solutions adopted to achieve the purpose of the present invention are:

[0007] A multi-modal online detection method for transformer oil includes the following specific steps:

[0008] Get real-time visual images at the current node t Real-time thermal images And real-time voltage U t ;

[0009] Based on Maxwell's equations and the principle of electric displacement invariance combined with real-time voltage U t Derive the dielectric constant ε of the current node t t ;

[0010] Using Qu's priori smoothing approximation algorithm to process real-time visual images Utilize curvature information to build a non-convex high-order model to reduce the step effect, construct an adaptive variational problem and perform smooth approximation through optimization algorithms to generate real-time smooth visual images.

[0011] Segmentation mark sharpening method is used to process real-time smooth visual images Perform segmentation, labeling and sharpening to generate the first target visual image Second target visual image and third target visual image

[0012] Removing real-time thermal images by filter mapping enhancement The noise in the image is converted into an optimized grayscale image And generate real-time enhanced thermal images through local histogram equalization Use area screening and color mapping to obtain the temperature T of the current node t t ;

[0013] Based on the second target visual image The fluorescence intensity is obtained by combining the gray value in the fluorescence standard curve to derive the pH value of the current node t. t ;

[0014] The third target visual image is transformed into Oil column pixel height in Converted into the corresponding actual height of the oil column l t o And calculate the oil column height change Δl at the current node t t o ;

[0015] The dielectric constant ε of the current node t t , temperature T t , acid value pH t and oil column height change Δl t o Input the logistic regressor, perform dimensionality increase, dimensionality reduction and nonlinear mapping in turn to generate the aging index ψ of the current node t t , and dynamically decide whether to adjust the monitoring interval Δt or send an early warning prompt based on the early warning decision mechanism.

[0016] Furthermore, the dielectric constant ε of the current node t t The specific derivation process is as follows:

[0017] The normal charge and normal voltage of the capacitor are Q0 and U0 respectively. According to Maxwell's equations, the normal electric displacement D0 is the product of the air dielectric constant ε0 and the normal electric field strength E0. At the same time, the normal electric displacement D0 is proportional to the normal charge Q0 and inversely proportional to the electrode distance r, from which the normal electric field strength E0 is derived;

[0018] Since the normal field strength E0 of the capacitor is the differential of the normal voltage U0 with respect to the electrode distance r, the first equation of the normal voltage U0 and the air dielectric constant ε0 can be deduced in reverse;

[0019] The real-time charge and voltage of the capacitor at the current node t are Q(t) and U(t), respectively. Based on the principle of electric displacement invariance, the real-time electric displacement D(t) is equal to the normal electric displacement D0. According to Maxwell's equations, the real-time charge Q(t) is equal to the normal charge Q0.

[0020] Since the capacitor has not changed, the second equation generated by replacing the normal voltage U0 and the air dielectric constant ε0 in the first equation with the real-time voltage U(t) and the real-time dielectric constant ε(t) at the current node t still holds.

[0021] Dividing the second equation by the first equation yields the following: the ratio of the real-time voltage U(t) to the normal voltage U0 is always equal to the ratio of the real-time dielectric constant ε(t) to the air dielectric constant ε0. This allows us to derive the real-time dielectric constant ε(t) at the current node t.

[0022] Furthermore, the real-time visual image is processed by using the Qu first a priori smoothing approximation algorithm. Generate real-time smooth visual images The specific steps include:

[0023] Real-time visual images The TV regularization term is obtained by integrating all horizontal curves on the Lipschitz continuous boundary domain Ω The specific formula is as follows:

[0024]

[0025] in, and Real-time visual images Grayscale images and pixels, is the first-order gradient image corresponding to the grayscale image;

[0026] Introducing the curvature curve model, through the grayscale image The curvature regularization term is calculated by fitting the divergence of the unit normal vector in The specific formula is as follows:

[0027]

[0028] Among them, α is the fitting coefficient, Grayscale image The unit second-order gradient map of

[0029] Improve the weight matrix in the adaptive TV model to construct an adaptive variational problem. The adaptive variational problem is as follows:

[0030]

[0031] Among them, λ and η are fixed regularization coefficients, b is the approximation coefficient, and Pixels The unit second-order gradient, weight matrix and first-order gradient at pixel point The weight matrix at Based on Gaussian kernel Get;

[0032] Use optimization algorithm to solve the adaptive variational problem and obtain the optimal approximation coefficient b * and calculates real-time smooth visual images

[0033] Furthermore, the segmentation mark sharpening method is used to process the real-time smooth visual image Generate the first target visual image Second target visual image and third target visual image The specific steps include:

[0034] Mining real-time smooth visual images through feature extraction layers based on inverted residual structures The feature extraction layer based on the inverted residual structure achieves real-time smoothing of visual images through convolution, depth-separable convolution and convolution again. Dimensionality increase, feature mining and dimensionality reduction, and generate feature map f through ReLU and residual connection t v ;

[0035] The coordinate attention mechanism is used to capture long-range dependencies along the horizontal and vertical directions and retain precise location information to generate a direction-aware attention map f. t v,d and position-aware attention map f t v,p And collaboratively enhance the feature map f t v , generate the attention feature map f t v,a ;

[0036] Real-time smoothing of visual images through multi-scale feature fusion layers Divide into several patches, and generate corresponding weight vectors for each patch through secondary extraction and upsampling, and smooth the visual image in real time according to the patch The positions in the corresponding splicing all weight vectors generate a weight graph

[0037] By splicing the prediction layer, the attention feature map f t v,a and weight graph Perform linear modulation and splicing, and then perform convolution and bicubic interpolation to restore the real-time visual markup image.

[0038] Select Single Marker to visually mark the image in real time The pixel values ​​of all pixels that do not carry a single mark are reset to the boundary pixel value with the largest difference from the average pixel value of the pixels carrying a single mark, generating the first target visual image. Second target visual image and third target visual image

[0039] Furthermore, the real-time thermal image is processed by filter mapping enhancement method Generate real-time enhanced thermal imagery And obtain the temperature T of the current node t through area screening and color mapping t , including the following specific steps:

[0040] Use median filtering to filter real-time thermal images Generate real-time optimized thermal images using salt and pepper noise and Gaussian noise

[0041] Get optimized grayscale image And divided into N local grayscale areas, N is the total number of local grayscale areas;

[0042] The probability of occurrence of each gray value in each local gray area is accumulated to obtain the cumulative distribution function;

[0043] Get optimized grayscale image The total number of grayscale values ​​in the grayscale area is deducted by 1 and multiplied by the cumulative distribution function of each local grayscale area and rounded down to generate a real-time enhanced thermal image.

[0044] Thermal imagery will be enhanced in real time Middle and first target visual image The pixels that overlap with the non-boundary pixel values ​​are retained;

[0045] Perform colorimetric mapping based on the color temperature mapping function between the pixel value and temperature in the colorimetric bar to obtain real-time enhanced thermal images The temperature corresponding to all retained pixels in the grid is taken as the average temperature T of the current node t. t .

[0046] Furthermore, the actual height l of the oil column at the current node t is obtained by using the improved edge detection and scale expansion method. t o and oil column height change Δl t o , including the following specific steps:

[0047] Using improved recursive filtering to process the third target visual image To solve the problem that Gaussian filtering is not effective in processing salt and pepper noise and is prone to blurring edges, a third target optimized visual image is generated.

[0048] Using Sobel operator to calculate the third objective to optimize visual image The gradient magnitude and gradient direction of the image are calculated, and bilinear interpolation is used to interpolate the four pixels including the target pixel in the gradient direction to achieve sub-pixel positioning of the edge;

[0049] Using double threshold judgment method to determine the third target to optimize visual image The oil column pixel edge in the image is obtained, and the oil column pixel height of the current node t is obtained by rectangular approximation. and infiltration tube pixel height The actual height of the infiltration tube Pixel height of the infiltration tube The ratio of is used as the scaling scale;

[0050] The oil column pixel height Multiply by the telescopic scale to get the actual height l of the oil column at the current node t t o , and the initial oil column height Make the difference to get the oil column height change Δl at the current node t t o .

[0051] Furthermore, the improved recursive filtering includes the following specific steps:

[0052] Set the initial sliding window width w0 and the maximum sliding window width w max and judgment threshold ξ δ , define the third target visual image Standard deviation of pixels within the sliding window is the local contrast;

[0053] Perform local contrast judgment before filtering the target pixel and calculate the local contrast within the current sliding window And judge whether it is less than the judgment threshold ξ δ ;

[0054] If it is less than the judgment threshold ξ δ , update the pixel value of the target pixel to the median of the pixels in the current sliding window, and move the sliding window to the next pixel;

[0055] If it is greater than or equal to the judgment threshold ξ δ , execute sliding window judgment to determine whether the current sliding window width w is less than the maximum sliding window width w max ;

[0056] If it is less than the maximum sliding window width w max , expand the sliding window width to w+2, and jump to perform local comparison judgment; w is the current sliding window width;

[0057] If it is equal to the maximum sliding window width w max , update the pixel value of the target pixel to the pixel median of the four adjacent pixels;

[0058] Repeat the above steps until the third target visual image All pixels in the image are filtered.

[0059] Furthermore, a dynamic decision is made based on the early warning decision mechanism on whether to adjust the monitoring interval Δt or send an early warning prompt, including the following specific steps:

[0060] Get the aging index ψ of transformer oil at the current node t t , determine the aging index ψ t Is it greater than or equal to the first aging threshold ψ1?

[0061] If it is greater than or equal to the first aging threshold ψ1, an early warning prompt is issued;

[0062] If it is less than the first aging threshold ψ1, further determine the aging index ψ t Is it less than the second aging threshold ψ2?

[0063] If it is less than the second aging threshold ψ2, the monitoring interval Δt is maintained as the initial monitoring interval Δ0 to generate the next node t+Δt;

[0064] If it is greater than or equal to the second aging threshold ψ2 and less than the first aging threshold ψ1, obtain the aging index ψ t The quantization layer number y t , adjust the monitoring interval Δt to the initial monitoring interval Δ0 minus the quantization layer number y tMultiply by the minimum adjustment interval Δ min , generate the next node t+Δt.

[0065] A transformer oil multi-modal online detection system is used to implement a transformer oil multi-modal online detection method, including a system clock module, a data acquisition module, a multi-scale analysis module and a feedback adjustment module;

[0066] The system clock module sets nodes based on the monitoring interval Δt and sends a collection signal to the data collection module at each node;

[0067] The data acquisition module receives the acquisition signal, obtains real-time visual images, real-time thermal images and real-time voltage, and feeds them back to the multi-scale analysis module;

[0068] The multi-scale analysis module derives the dielectric constant based on Maxwell's equations and the principle of electric displacement invariance. It generates the first, second, and third target visual images based on a priori smoothing approximation algorithm combined with a segmentation marker sharpening method. It processes the real-time thermal image using a filter mapping enhancement method and obtains the temperature by combining regional screening and colorimetric mapping. It obtains the acid value based on the second target visual image combined with a fluorescence standard curve. It processes the third target visual image using an improved edge detection and scale expansion method to obtain the oil column height change. The dielectric constant, temperature, acid value, and oil column height change are then sent to the feedback adjustment module.

[0069] The feedback regulation module uses a logistic regression function to analyze the dielectric constant, temperature, acid value, and oil column height to generate an aging index, and dynamically decides whether to adjust the monitoring interval Δt or send an early warning based on the early warning decision mechanism.

[0070] Compared with the prior art, the present invention has the following significant advantages:

[0071] 1. The introduction of infrared temperature measurement, fluorescent indicator acid value measurement, capillary infiltration, and capacitor induction, along with algorithms or principles such as filter mapping enhancement, fluorescence standard curves, improved edge detection and scaling, and Maxwell's equations, has transformed the previously cumbersome and complex manual inspection process into automated visual processing, infrared measurement, and electrical signal analysis. This has enabled regular, automated online inspection of transformer oil status. The updated inspection method also reduces the use of precision and intelligent equipment, thereby reducing costs.

[0072] 2. Design a priori smoothing approximation algorithm to attenuate abnormal high-frequency components in the real-time visual image to reduce the staircase effect, and locally couple to generate a real-time smooth visual image. This assists the segmentation and labeling sharpening method. The segmentation and labeling sharpening method uses image segmentation technology to identify the first, second, and third targets in the real-time smooth visual image and label them to generate a real-time visual label image. Based on labeling sharpening, the first, second, and third target visual images are generated for subsequent accurate detection.

[0073] 3. A joint analysis and adjustment scheme combining a logistic regressor and an early warning decision-making mechanism was designed. The aging index of the transformer oil was estimated by comprehensively considering the dielectric constant, temperature, acid value, and oil column height change. Based on the range of the aging index, the early warning decision-making mechanism chooses to send early warning prompts or adaptively adjust the monitoring interval to detect transformer oil problems early. This achieves effective early warning of transformer oil and improves the operational safety of the transformer. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] Figure 1 This is a flow chart of a transformer oil multi-modal online detection method in the present invention;

[0075] Figure 2 This is a flow chart of the segmentation mark sharpening method in the present invention;

[0076] Figure 3 This is a flow chart of the improved recursive filtering in the present invention;

[0077] Figure 4 This is a schematic diagram of a transformer oil multi-modal online detection system in the present invention. DETAILED DESCRIPTION

[0078] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0079] Example 1

[0080] like Figure 1 As shown, a specific embodiment of the present invention discloses a multi-modal online detection method for transformer oil, comprising the following specific steps:

[0081] At the current node t, real-time visual images are acquired through active collection of multi-source visual detection equipment and passive sensing of variable capacitance sensors. Real-time thermal images And real-time voltage U t , and real-time visual images and real-time thermal images have the same dimensions;

[0082] Combining Maxwell's equations and the principle of electric displacement invariance, combined with the real-time voltage U tThe ratio of the transformer oil dielectric constant ε at the current node t is determined by the ratio of the transformer oil dielectric constant ε to the normal voltage U0. t ;

[0083] Using Qu's priori smoothing approximation algorithm to process real-time visual images Leveraging curvature information to construct a non-convex high-order model for attenuating real-time visual images The abnormal high-frequency components in the image are used to reduce the step effect, transform the local coupling into an adaptive variational problem, and perform smooth approximation through the optimization algorithm to generate a real-time smooth visual image.

[0084] Real-time smoothing of visual images using segmentation and labeling sharpening Real-time smooth visual image recognition through image segmentation technology The pixels belonging to the first target, the second target and the third target are assigned corresponding labels to generate a real-time visual labeling image Generate the first target visual image based on the mark sharpening Second target visual image and third target visual image Among them, the first target is transformer oil, the second target is fluorescent indicator, and the third target is capillary infiltration tube;

[0085] Processing real-time thermal images by filter mapping enhancement method Remove real-time thermal images by filtering The noise in the image is converted into an optimized grayscale image By optimizing the grayscale image The local histogram equalization is performed on the cumulative grayscale distribution function of the local area to generate a real-time enhanced thermal image Combined with the first target visual image Perform regional screening and obtain the temperature T of the transformer oil at the current node t through color mapping t ;

[0086] Since the second target visual image The gray value of each pixel in represents the brightness of the pixel, so the second target visual image The average grayscale value of all pixels in the area where the fluorescent indicator is located is used as the fluorescence intensity, and the acid value pH of the transformer oil at the current node t is derived based on the fluorescence standard curve. t , wherein the fluorescent indicator can be made of a pH-sensitive fluorescent dye based on reversible protonation of an amino group. The fluorescence intensity of the fluorescent indicator changes with the acid value of the contact object, and the fluorescence standard curve can be obtained by pre-measuring multiple groups of transformer oils with different acid values ​​using the fluorescent indicator and fitting them. In this embodiment, the fluorescent indicator is a boron dipyrromethene dye;

[0087] Calibration of third target visual image by improved edge detection and scale expansion method Oil column pixel height in and infiltration tube pixel height Combined with the actual height of the infiltration tube Calculate the telescopic scale to obtain the actual height l of the oil column at the current node t t o And calculate the oil column height change Δl at the current node t t o ;

[0088] The dielectric constant ε of the current node t t , temperature T t , acid value pH t and oil column height change Δl t o As the input of the pre-trained logistic regressor, the logistic regressor first increases the dimension of the input through the fully connected layer and the batch normalization layer, then reduces the dimension through two fully connected layers in sequence, and finally realizes the dielectric constant ε of the current node t through the Sigmoid function. t , temperature T t , acid value pH t and oil column height change Δl t o To transformer oil aging index ψ t The nonlinear mapping is performed, and the early warning decision-making mechanism dynamically decides whether to adjust the monitoring interval Δt or send an early warning prompt. The pre-training of the logistic regressor requires collecting the temperature, dielectric constant, acid value and viscosity of a large number of transformer oil samples, and recording the aging index of these transformer oil samples. The aging index ranges from 0 to 1 and can be obtained through various means such as chemical analysis, physical testing or expert evaluation.

[0089] Furthermore, the varactor sensor can sense the capacitance change of the capacitor and the dielectric constant ε of the transformer oil at the current node t. t The specific derivation process is as follows:

[0090] When transformer oil is not passed between the inner and outer electrodes of the capacitor, the medium is air, and the normal charge and normal voltage of the capacitor are Q0 and U0 respectively;

[0091] According to Maxwell's equations, the normal electric displacement D0 of a capacitor when the medium is air is the product of the air dielectric constant ε0 and the normal electric field strength E0, that is, D0 = ε0·E0. At the same time, the normal electric displacement D0 is proportional to the normal charge Q0 and inversely proportional to the electrode distance r. The specific formula is D0 = Q0 / (2π·r), from which the normal electric field strength E0 = Q0 / (2π·ε0·r) can be deduced;

[0092] Since the field strength E of the capacitor is the differential of the voltage U with respect to the electrode distance r, that is, the normal field strength E0, the normal voltage U0 and the electrode distance r satisfy The first equation of the normal voltage U0 and the air dielectric constant ε0 can be deduced in reverse, and the first equation is as follows:

[0093]

[0094] Where r1 and r2 are the radii of the inner and outer electrodes, respectively;

[0095] When transformer oil passes between the inner and outer electrodes of the capacitor, the medium is transformer oil. The real-time charge and real-time voltage of the capacitor at the current node t are Q(t) and U(t), respectively. Based on the principle of electric displacement invariance, the real-time electric displacement D(t) at the current node t is equal to the normal electric displacement D0. According to Maxwell's equations, the real-time charge Q(t) is also equal to the normal charge Q0.

[0096] Since the capacitor's electrode distance r, inner electrode radius r1, and outer electrode radius r2 have not changed, the second equation generated by replacing the normal voltage U0 and air dielectric constant ε0 in the first equation with the real-time voltage U(t) and real-time dielectric constant ε(t) at the current node t still holds true.

[0097] Dividing the second equation by the first equation shows that the ratio of the capacitor's real-time voltage U(t) at the current node t to the normal voltage U0 is always equal to the ratio of the real-time dielectric constant ε(t) at the current node t to the dielectric constant ε0 of air. Therefore, the real-time dielectric constant ε(t) at the current node t = [U(t)·ε0] / U0.

[0098] Furthermore, the real-time visual image is processed by using the Qu first a priori smoothing approximation algorithm. Generate real-time smooth visual images The specific steps include:

[0099] Due to the real-time visual image For a bounded image on a Lipschitz continuous boundary domain Ω, for real-time visual images The TV regularization term is obtained by integrating all horizontal curves of , the specific formula is as follows:

[0100]

[0101] in, and Real-time visual images Grayscale images and pixels, is the first-order gradient image corresponding to the grayscale image;

[0102] Since the TV model is susceptible to the step effect, the curvature curve model is introduced to t v The curvature regularization term is calculated by fitting the divergence of the unit normal vector in The specific formula is as follows:

[0103]

[0104] Among them, α is the fitting coefficient, Grayscale image The unit second-order gradient map of

[0105] To reduce the effects of noise and blur and fully preserve the real-time visual image The edge, structure and other valuable details of the image are obtained by improving the weight matrix in the adaptive TV model to construct an adaptive variation problem to obtain the grayscale image. Smooth near real-time visual images The adaptive variational problem is as follows:

[0106]

[0107] Among them, λ and η are fixed regularization coefficients, b is the approximation coefficient, and Pixels The unit second-order gradient, weight matrix and first-order gradient at pixel point The weight matrix at Based on Gaussian kernel The specific formula is as follows:

[0108]

[0109] Where γ is the proportionality coefficient, and Grayscale images At the pixel Positive gradient on the horizontal axis i and vertical axis j at location, Gaussian kernel The specific calculation formula is as follows:

[0110]

[0111] Where δ is the standard deviation of the Gaussian kernel, and Pixels Pixel values ​​on the horizontal axis i and the vertical axis j;

[0112] Use optimization algorithm to solve the adaptive variational problem and obtain the optimal approximation coefficient b * , optimal fitting coefficient α *, optimal proportional coefficient γ * and the optimal standard deviation δ * , obtain real-time smooth visual images In this embodiment, the optimization algorithm adopts the alternating direction multiplier method.

[0113] like Figure 2 As shown, further, the segmentation mark sharpening method is used to process the real-time smooth visual image Generate the first target visual image Second target visual image and third target visual image The specific steps include:

[0114] Mining real-time smooth visual images through feature extraction layers based on inverted residual structures The corresponding feature map f t v The traditional feature extraction layer is often implemented based on a five-step strategy of dimensionality reduction, convolution, dimensionality increase, ReLU and residual connection. The feature extraction layer based on the inverted residual structure first adds real-time smoothing of the visual image through 1×1 convolution. The number of channels is increased to achieve dimensionality increase, and then feature mining is performed through 3×3 depth-separable convolution, where depth-separable convolution can reduce the number of parameters and improve the operation speed. Finally, the number of channels is reduced again through 1×1 convolution, and then the feature map f is generated through ReLU and residual connection. t v ;

[0115] The coordinate attention mechanism imitates the human visual system to adjust the feature map f t v Different attention levels at different positions, different from the traditional attention mechanism that only focuses on the feature information of the encoding channel, the coordinate attention mechanism captures long-range dependencies along the horizontal direction while retaining precise position information along the vertical direction, generating direction-aware attention maps f t v,d and position-aware attention map f t v,p and used together to enhance the feature map f t v , generate attention feature map

[0116] Real-time smooth visual image enhancement via multi-scale feature fusion layers The context aggregation capability of feature information in the multi-scale feature fusion layer will smooth the visual image in real time. The image is divided into patches of the same size. For each patch, 3×3 convolution, batch normalization and Sigmoid are performed, and then 1×1 convolution, batch normalization and Sigmoid are performed again for secondary extraction. To avoid dimensionality reduction in the convolution process, bicubic interpolation is used for upsampling to generate the weight vector corresponding to each patch, and the visual image is smoothed in real time based on the patch. The position in corresponds to the weight vector of all patches, generating a weight map

[0117] By splicing the prediction layer, the attention feature map f t v,a and weight graph They are linearly modulated to convert to the same dimension and concatenated, and then undergo a 3×3 convolution for dimensionality reduction and bicubic interpolation to restore to the same smooth visual image as the real-time one. Same dimension, generating real-time visually labeled images Real-time visual tagging of images The pixels belonging to the first target, the second target, and the third target are assigned labels 1, 2, and 3 respectively;

[0118] Select a single marker from the three markers in turn to visually mark the image in real time The pixel values ​​of all pixels that do not carry a single marker are reset to the boundary pixel value with the largest difference from the average pixel value of the pixels carrying a single marker to achieve real-time visual marking of the image. The boundary pixel values ​​include the upper limit pixel value 255 and the lower limit pixel value 0. Taking mark 1 as an example, assuming that the average pixel value of all pixels carrying mark 1 is 197, the pixel values ​​of all other pixels not carrying mark 1 should be reset to the lower limit pixel value 0 to generate the first target visual image. Second target visual image and third target visual image

[0119] Furthermore, the real-time thermal image is processed by filter mapping enhancement method Generate real-time enhanced thermal imagery And obtain the temperature T of the current node t through area screening and color mapping t , including the following specific steps:

[0120] Use median filtering to filter real-time thermal images Salt and pepper noise and Gaussian noise in order to reduce noise for real-time thermal images The impact of clarity, generating real-time optimized thermal images

[0121] Get real-time optimized thermal images Corresponding optimized grayscale image And divided into N local grayscale areas, N is the total number of local grayscale areas;

[0122] The probability of each gray value in each local gray area is accumulated to obtain the cumulative distribution function. The cumulative distribution function of the nth local gray area is used as the cumulative distribution function. For example, n=1,2,…,N, the specific formula is as follows:

[0123]

[0124] Among them, u max Indicates the maximum gray value, p n (u) represents the probability that the gray value in the nth local gray area is u;

[0125] Get optimized grayscale image The total number of grayscale values ​​in the image is deducted by 1, and then multiplied by the cumulative distribution function of each local grayscale area and rounded down to the integer. This is used as the enhanced thermal pixel corresponding to the local grayscale area to achieve local histogram equalization and generate a real-time enhanced thermal image. Real-time optimization of thermal images Compared with real-time enhanced thermal imagery It has stronger infrared contrast, which is beneficial for identifying the temperature of transformer oil;

[0126] Since the first target visual image Has been sharpened and can be used as a real-time enhanced thermal image The area filter box will enhance the thermal image in real time Middle and first target visual image The pixels that overlap with the non-boundary pixel values ​​are retained;

[0127] Perform colorimetric mapping based on the color temperature mapping function between the pixel value and temperature in the colorimetric bar to obtain real-time enhanced thermal images The temperature corresponding to all retained pixels in the grid is taken as the average temperature T of the current node t. t , the color temperature mapping function is a linear function, and the specific formula is as follows:

[0128] T=β1·x+β2;

[0129] Where x and T are the pixel value and the corresponding temperature in the colorimetric bar, respectively; β1 and β2 are the slope and intercept of the color temperature mapping function, respectively. The specific calculation formula is as follows:

[0130]

[0131] β2=T max-β1·x max ;

[0132] Among them, x max and T max are the maximum pixel value and the corresponding maximum temperature in the color bar, respectively, min and T min They are the minimum pixel value and the corresponding minimum temperature in the color bar, respectively. Based on the color temperature mapping function, it is only necessary to substitute the real-time enhanced thermal image. The pixel value of the retained pixel can be used to obtain the temperature corresponding to the retained pixel.

[0133] Furthermore, the actual height l of the oil column at the current node t is obtained by using the improved edge detection and scale expansion method. t o and oil column height change Δl t o , including the following specific steps:

[0134] Improved recursive filtering is used to replace traditional Gaussian filtering to process the third target visual image To solve the problem that Gaussian filtering is not effective in processing salt and pepper noise and is prone to blurring edges, a third target optimized visual image is generated.

[0135] Since the third goal is to optimize the visual image The middle edge usually shows a sharp change in grayscale intensity. The Sobel operator is used to calculate the third objective to optimize the visual image. The gradient magnitude and gradient direction, where the gradient magnitude indicates the degree of change in grayscale intensity, and the gradient direction indicates the direction of change;

[0136] Since non-maximum suppression only compares the gradient amplitude of the target pixel with the gradient amplitude of the two adjacent pixels in the gradient direction, it is easy to cause random errors and false edge problems when the edge and non-pixel points do not overlap. Bilinear interpolation is used to replace the traditional non-maximum suppression process. By interpolating the four pixels including the target pixel in the gradient direction, the edge position in the grid area surrounded by the four pixels is accurately obtained, achieving sub-pixel edge positioning;

[0137] Since the third target visual image before the improved recursive filtering It is generated after sharpening, and the existing double threshold judgment method can be directly used to determine the third target to optimize the visual image The pixel edge of the oil column in the image;

[0138] Use a rectangular approximation box to select the oil column pixels and capillary infiltration tube pixels to obtain the oil column pixel height at the current node t and infiltration tube pixel height The actual height of the infiltration tube Pixel height of the infiltration tube The ratio of is used as the scaling scale;

[0139] The oil column pixel height Multiply by the telescopic scale to get the actual height l of the oil column at the current node t t o , the initial oil column height corresponding to the pre-measured new transformer oil Make the difference to get the oil column height change at the current node t Since the viscosity change of transformer oil will affect the height of the oil column in the capillary impregnation tube, the height change of the oil column Δl t o It can indirectly indicate the change in transformer oil viscosity from new transformer oil to the current node t.

[0140] like Figure 3 As shown, further, the improved recursive filtering introduces local contrast on the basis of traditional recursive median filtering, and adjusts the third target visual image by local contrast. The size of the filter window in different areas in the image is adjusted to better fit the third target visual image. Based on the local features in , the improved recursive filtering includes the following specific steps:

[0141] Set the initial sliding window width to w0, w0 is an odd number, and the maximum sliding window width is set to w max , the judgment threshold is set to ξ δ , define the third target visual image Standard deviation of pixels within the sliding window is the local contrast;

[0142] Visual image of the third target Before filtering any target pixel in the image, perform local contrast judgment and calculate the local contrast within the current sliding window. And judge whether it is less than the judgment threshold ξ δ ;

[0143] If the local contrast within the current sliding window Less than the judgment threshold ξ δ , update the pixel value of the target pixel to the median of the pixels in the current sliding window, and move the sliding window to the next pixel;

[0144] If the local contrast within the current sliding window Greater than or equal to the judgment threshold ξ δ , execute sliding window judgment to determine whether the current sliding window width w is less than the maximum sliding window width w max ;

[0145] If the current sliding window width w is less than the maximum sliding window width w max , expand the sliding window width to w+2, and jump to perform local comparison judgment;

[0146] If the current sliding window width w is equal to the maximum sliding window width w max , update the pixel value of the target pixel to the pixel median of the four adjacent pixels;

[0147] Repeat the above steps until the third target visual image All pixels in the image are filtered.

[0148] Furthermore, a dynamic decision is made based on the early warning decision mechanism on whether to adjust the monitoring interval Δt or send an early warning prompt, including the following specific steps:

[0149] Get the aging index ψ of transformer oil at the current node t t , determine the aging index ψ t Is it greater than or equal to a first aging threshold ψ1? In this embodiment, the first aging threshold ψ1 is set to 0.8?

[0150] If the aging index ψ t If the value is greater than or equal to the first aging threshold ψ1, the transformer oil is judged to be severely aged and an early warning is issued to the staff;

[0151] If the aging index ψ t is less than the first aging threshold ψ1, further determine the aging index ψ t whether it is less than a second aging threshold ψ2. In this embodiment, the second aging threshold ψ2 is set to 0.4;

[0152] If the aging index ψ t If the value is less than the second aging threshold ψ2, the transformer oil is determined to be relatively new, and the monitoring interval Δt is maintained at the initial monitoring interval Δ0 to generate the next node t+Δt;

[0153] If the aging index ψ t Greater than or equal to the second aging threshold ψ2 and less than the first aging threshold ψ1, obtain the aging index ψ t The quantization layer number y t , adjust the monitoring interval Δt to the initial monitoring interval Δ0 minus the quantization layer number y t Multiply by the minimum adjustment interval Δ min , generate the next node t+Δt, where the quantization layer number y t The second aging threshold ψ2 is the starting value, the first aging threshold ψ1 is the ending value, and 0.1 is the quantization step size. When the quantization layer number is y t When, it represents the aging index ψ tLocated in the range [ψ2+0.1·(y t -1),ψ2+0.1·y t ), the minimum adjustment interval Δ in this embodiment min It is 0.2 times of the initial monitoring interval Δ0.

[0154] Example 2

[0155] like Figure 4 As shown, a transformer oil multi-modal online detection system is used to implement a transformer oil multi-modal online detection method, including a system clock module, a data acquisition module, a multi-scale analysis module and a feedback adjustment module;

[0156] The system clock module sets nodes based on the monitoring interval Δt and sends a collection signal to the data collection module at each node;

[0157] The data acquisition module receives the acquisition signal, lowers the fluorescent indicator, opens the oil extraction channel, and allows the transformer oil to pass through the fluorescent indicator, capillary infiltration tube, and capacitor in sequence. It then uses multi-source visual inspection equipment and a variable capacitance sensor to obtain real-time visual images, real-time thermal images, and real-time voltage, and feeds these back to the multi-scale analysis module. The fluorescent indicator is then recycled. During non-collection periods, the fluorescent indicator is placed in a recycling container containing a substance that helps the fluorescent indicator quickly recover its normal fluorescence intensity, thereby enabling its reuse. The multi-source visual inspection equipment consists of an infrared thermal imager and a high-definition camera.

[0158] The multi-scale analysis module derives the dielectric constant by combining Maxwell's equations and the principle of electric displacement invariance with real-time voltage. It processes the real-time visual image based on the Qu priori smoothing approximation algorithm and uses the segmentation marker sharpening method to generate the first target visual image, the second target visual image, and the third target visual image. It processes the real-time thermal image using the filter mapping enhancement method and obtains the temperature by combining regional screening and colorimetric mapping. It derives the fluorescence intensity based on the second target visual image and obtains the acid value by combining the fluorescence standard curve. It processes the third target visual image using the improved edge detection and scale expansion method to obtain the oil column height change. The dielectric constant, temperature, acid value, and oil column height change are sent to the feedback adjustment module.

[0159] The feedback regulation module uses a logistic regression function to analyze the dielectric constant, temperature, acid value, and oil column height to generate an aging index, and dynamically decides whether to adjust the monitoring interval Δt or send an early warning based on the early warning decision mechanism.

[0160] The present invention discloses a multimodal online detection method and system for transformer oil. The method derives the dielectric constant based on Maxwell's equations and the principle of electric displacement invariance in combination with real-time voltage; adopts a Qu priori smoothing approximation algorithm combined with a segmentation, labeling and sharpening method to segment, label and sharpen real-time visual images; adopts a filter mapping enhancement method to denoise and enhance real-time thermal images to generate real-time enhanced thermal images, and derives temperature based on regional screening and colorimetric mapping; derives acid value based on the grayscale value of a second target visual image in combination with a fluorescence standard curve, obtains the actual height of the oil column in a third target visual image through an improved edge detection and scale expansion method, and calculates the change in oil column height to reflect viscosity; generates an aging index through logistic regression analysis, and dynamically decides whether to adjust the monitoring interval or send a warning prompt based on a warning decision mechanism, thereby realizing online adaptive detection and early warning of transformer oil.

[0161] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiment. All technical solutions based on the concept of the present invention are within the scope of protection of the present invention. It should be noted that for those skilled in the art, various improvements and modifications that do not depart from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A transformer oil multi-modal online detection method, characterized in that: The specific steps include: Acquire real-time visual images, real-time thermal images, and real-time voltage at the current node, and derive the dielectric constant of the current node based on Maxwell's equations and the principle of electric displacement invariance; The non-convex high-order model and adaptive variational problem are constructed by using the Qu-based priori smoothing approximation algorithm, and the real-time smooth visual image is obtained by solving the problem through the optimization algorithm. The first target visual image, the second target visual image and the third target visual image are generated by the segmentation mark sharpening method. The filter mapping enhancement method is used to remove noise from the real-time thermal image and convert it into an optimized grayscale image. The real-time enhanced thermal image is generated through local histogram equalization. The temperature of the current node is obtained by combining regional screening and color mapping. Based on the grayscale value of the second target visual image and combined with the fluorescence standard curve, the acid value of the current node is deduced. The oil column pixel height in the third target visual image is converted into the actual oil column height through the improved edge detection and scale expansion method, and the oil column height change of the current node is calculated. The dielectric constant, temperature, acid value, and oil column height of the current node are analyzed through a logistic regression model and mapped to generate an aging index for the current node. Based on the early warning decision-making mechanism, a dynamic decision is made on whether to adjust the monitoring interval or send an early warning prompt.

2. A transformer oil multi-modal online detection method according to claim 1, characterized in that: The acquisition of a real-time smooth visual image comprises the following specific steps: The TV regularization term is obtained by integrating all horizontal curves of the real-time visual image on the Lipschitz continuous boundary domain; The curvature curve model is introduced, and the curvature regularization term is calculated by fitting the divergence of the unit normal vector in the grayscale image; The weight matrix in the adaptive TV model is improved based on the Gaussian kernel, and the regularization coefficient is introduced to construct the adaptive variational problem; An optimization algorithm is used to solve the adaptive variational problem, obtain the optimal approximation coefficients and calculate the real-time smooth visual image.

3. A transformer oil multi-modal online detection method according to claim 1, characterized in that: The method of using the segmentation mark sharpening method to generate the first target visual image, the second target visual image and the third target visual image comprises the following specific steps: The feature extraction layer based on the inverted residual structure mines real-time smooth visual images. It achieves dimensionality increase, feature mining and dimensionality reduction of real-time smooth visual images through convolution, depthwise separable convolution and re-convolution respectively, and generates feature maps through ReLU and residual connections. The coordinate attention mechanism is used to capture long-range dependencies along the horizontal and vertical directions and retain position information, generate direction-aware attention maps and position-aware attention maps, and jointly enhance feature maps to generate attention feature maps; The multi-scale feature fusion layer divides the real-time smooth visual image into multiple patches. Each patch generates a corresponding weight vector through secondary extraction and upsampling, and then the weight map is generated by splicing the patches according to their positions in the real-time smooth visual image. The attention feature map and weight map are linearly modulated and spliced ​​through the splicing prediction layer, and then restored through convolution and bicubic interpolation to generate a real-time visual markup image; A single marker is selected, and the pixel values ​​of all pixels in the real-time visual marker image that do not carry the single marker are reset to the boundary pixel values ​​with the largest difference from the average pixel values ​​of the pixels carrying the single marker, thereby generating a first target visual image, a second target visual image, and a third target visual image.

4. A transformer oil multi-modal online detection method according to claim 1, characterized in that: Generating a real-time enhanced thermal image and obtaining the temperature of the current node by combining region screening and color mapping includes the following specific steps: Use median filtering to remove noise from real-time thermal images and generate real-time optimized thermal images; Obtain an optimized grayscale image and divide it into multiple local grayscale regions, and accumulate the probability of occurrence of each grayscale value in each local grayscale region to obtain a cumulative distribution function; Obtain the total number of grayscale values ​​in the optimized grayscale image, subtract 1 from the total number of grayscale values, multiply it by the cumulative distribution function of each local grayscale area, and round it down to generate a real-time enhanced thermal image; retaining the pixel points in the real-time enhanced thermal image that overlap with the pixel points of the non-boundary pixel values ​​in the first target visual image; Colorimetric mapping is performed based on the color temperature mapping function between the pixel value and temperature in the colorimetric bar, and the temperatures corresponding to all retained pixels in the real-time enhanced thermal image are obtained, and the average value is used as the temperature of the current node.

5. A transformer oil multi-modal online detection method according to claim 1, characterized in that: The calculation of the oil column height change at the current node includes the following specific steps: Using improved recursive filtering to process the third target visual image to generate a third target optimized visual image; The Sobel operator is used to calculate the gradient magnitude and gradient direction of the third objective optimized visual image, and bilinear interpolation is used to interpolate the four pixels including the target pixel in the gradient direction; A double-threshold method is used to determine the oil column pixel edge in the third objective optimization visual image. The oil column pixel height and the infiltration tube pixel height of the current node are obtained through rectangular approximation. The ratio of the actual infiltration tube height to the infiltration tube pixel height is used as the scaling scale. The actual height of the oil column at the current node is obtained by multiplying the pixel height of the oil column by the telescopic scale, and the difference between the actual height and the initial height of the oil column is used to obtain the changed height of the oil column at the current node.

6. A transformer oil multi-modal online detection method according to claim 5, characterized in that: The improved recursive filtering comprises the following specific steps: Set the initial sliding window width, maximum sliding window width and judgment threshold, and define the pixel standard deviation of the third target visual image within the sliding window as the local contrast; Before filtering the target pixel, perform local contrast judgment, calculate the local contrast within the current sliding window and determine whether it is less than the judgment threshold; If it is less than the judgment threshold, the pixel value of the target pixel is updated to the median of the pixels in the current sliding window, and the sliding window is moved to the next pixel; If it is greater than or equal to the judgment threshold, the sliding window judgment is performed to determine whether the current sliding window width is less than the maximum sliding window width; If it is less than the maximum sliding window width, the sliding window width is expanded and the local comparison judgment is performed; If it is equal to the maximum sliding window width, update the pixel value of the target pixel to the median of the four adjacent pixels; Repeat the above steps until all pixels in the third target visual image are filtered.

7. A transformer oil multi-modal online detection method according to claim 1, characterized in that: The specific derivation process of the dielectric constant of the current node is as follows: Based on Maxwell's equations, the normal electric displacement is the product of the air dielectric constant and the normal electric field strength. It is proportional to the normal charge and inversely proportional to the distance between the electrodes. The normal electric field strength is derived. Since the normal field strength is the differential of the normal voltage with respect to the electrode distance, the first equation of the normal voltage and the dielectric constant of air is derived; Based on the principle of electric displacement invariance combined with Maxwell's equations, the real-time electric displacement is equal to the normal electric displacement, and the real-time charge is equal to the normal charge; The real-time voltage and the real-time dielectric constant of the current node are replaced with the normal voltage and the air dielectric constant in the first equation to generate the second equation; Dividing the second equation by the first equation yields that the ratio of the real-time voltage to the normal voltage is always equal to the ratio of the real-time dielectric constant to the dielectric constant of air, and the real-time dielectric constant of the current node can be derived.

8. A transformer oil multi-modal online detection method according to claim 1, characterized in that: The dynamic decision-making based on the early warning decision mechanism on whether to adjust the monitoring interval or send an early warning prompt includes the following specific steps: Obtaining an aging index of the transformer oil at the current node, and determining whether the aging index is greater than or equal to a first aging threshold; If it is greater than or equal to the first aging threshold, an early warning prompt is issued; If it is less than the first aging threshold, determining whether the aging index is less than the second aging threshold; If it is less than the second aging threshold, the monitoring interval is maintained as the initial monitoring interval and the next node is generated; If it is greater than or equal to the second aging threshold and less than the first aging threshold, obtain the quantization layer number where the aging index is located, adjust the monitoring interval to the initial monitoring interval minus the quantization layer number multiplied by the minimum adjustment interval, and generate the next node.

9. A transformer oil multi-modal online detection system, characterized in that: It includes system clock module, data acquisition module, multi-scale analysis module and feedback regulation module; The system clock module sets nodes based on monitoring intervals and sends acquisition signals to each node; The data acquisition module receives the acquisition signal, obtains real-time visual images, real-time thermal images and real-time voltage, and feeds them back to the multi-scale analysis module; The multi-scale analysis module derives the dielectric constant based on Maxwell's equations and the principle of electric displacement invariance. It generates the first, second, and third target visual images using a Qu priori smoothing approximation algorithm combined with a segmentation marker sharpening method. It processes real-time thermal images using a filter mapping enhancement method and combines regional screening and colorimetric mapping to obtain temperature. It derives the acid value based on the second target visual image combined with a fluorescence standard curve. It processes the third target visual image using an improved edge detection and scaling method to obtain the oil column height. The feedback regulation module uses a logistic regression function to analyze the dielectric constant, temperature, acid value, and oil column height to generate an aging index, and dynamically decides whether to adjust the monitoring interval or send an early warning based on the early warning decision mechanism.

10. A transformer oil multi-modal online detection system according to claim 9, characterized in that: The data acquisition module receives the acquisition signal, lowers the fluorescent indicator, opens the oil extraction channel to allow the transformer oil to pass through the fluorescent indicator, the capillary infiltration tube and the capacitor in sequence, obtains real-time visual images, real-time thermal images and real-time voltage through the multi-source visual detection equipment and the variable capacitance sensor, and recovers the fluorescent indicator.

Images