A method and system for assessing the risk of power transmission line tripping due to wildfires based on video images.

Abstract

This application provides a method and system for assessing the risk of power transmission line tripping due to wildfires based on video images. The method includes: acquiring original monitoring images of mountainous areas; enhancing and preprocessing the original monitoring images, then inputting them into a deep learning object detection network to output smoke detection boxes and flame detection boxes; determining the flame bridging ratio based on the original monitoring images and flame detection boxes; determining the line tripping risk coefficient based on the smoke detection boxes; and determining the wildfire tripping / blocking risk level based on the flame bridging ratio or the line tripping risk coefficient. The proposed method and system, on the one hand, achieve spatial quantification of physical features based on "visual perception" and break down the disciplinary barriers between visual features and underlying electrical engineering theory, making the tripping assessment results more scientific; on the other hand, it innovatively proposes a multi-dimensional smoke fine-grained classification algorithm, improving the accuracy of wildfire tripping assessment under smoke bridging conditions.

Description

Technical Field

[0001] This application relates to the field of risk detection in mountainous fire areas, and in particular to a method and system for assessing the risk of power transmission line tripping due to wildfires based on video images. Background Technology

[0002] Currently, intelligent inspection systems based on video surveillance are widely deployed in the power grid to monitor wildfires along transmission lines. The closest existing technology typically involves installing visible light or infrared cameras on transmission towers and using target detection algorithms from computer vision to analyze the monitoring footage in real time to identify the presence of "flames" or "smoke" targets in the two-dimensional image. Once a suspected fire is detected, the system will select the target area and send a qualitative alarm message to the dispatch center.

[0003] Meanwhile, in terms of wildfire tripping risk assessment, existing assessment methods mostly rely on satellite remote sensing fire point data, meteorological station macro-environmental data, and vegetation distribution measured manually on-site, combined with traditional electrical design specifications, to conduct offline or semi-online macro-risk simulations.

[0004] Although existing video surveillance and assessment methods can detect fires to some extent, they still have the following significant drawbacks when it comes to protecting power transmission lines from wildfire tripping: (1) The alarm is limited to two-dimensional qualitative alarm and lacks precise quantification of three-dimensional spatial topological relationships. Existing video surveillance algorithms can only output the bounding boxes at the pixel level of two-dimensional images and cannot restore the relative positional relationship between the actual fire location and the power transmission line in three-dimensional physical space. This makes the existing system only able to "detect the fire" but unable to quantitatively answer "how close the fire is to the power line and how much safety margin is left".

[0005] (2) There is a "disconnect" between visual characteristics and electrical breakdown mechanisms, and the assessment lacks physical support. The high-temperature ionized gases and conductive smoke generated by wildfires reduce the insulation withstand strength of air gaps, thus causing line tripping. Existing visual monitoring only stays at the "seeing" level and fails to effectively convert visual parameters such as flame height and relative position extracted from images into physical parameters for calculating insulation breakdown. Due to the lack of an interdisciplinary coupling mechanism that directly substitutes front-end visual parameters into back-end electrical breakdown models, existing risk assessments are often not precise enough and are prone to generating a large number of invalid false alarms when the fire does not yet pose an electrical threat.

[0006] (3) Failure to quantify the differentiated impact of different types of smoke on air insulation. The characteristics of smoke produced by different combustibles have significantly different effects on insulation. "Dense black smoke" containing a large number of conductive carbon particles is extremely damaging to air insulation, while "light gray smoke" which is mainly composed of moisture has a relatively minor impact. Existing target detection technologies and wildfire breakdown models usually treat all smoke in a general way, lacking multi-dimensional in-depth analysis of smoke transmission concentration, color tone, and texture density. They cannot output specific smoke types and corresponding electrical correction coefficients, resulting in extremely inaccurate assessment of wildfire tripping risk under "smoke bridging" conditions. Summary of the Invention

[0007] This application provides a method and system for assessing the risk of power transmission line wildfire tripping based on video images. To solve the above-mentioned technical problems, this application adopts the following technical methods: This application provides a method for assessing the risk of power transmission line tripping due to wildfires based on video images, including: Obtain raw surveillance images of mountainous areas; After enhancement and preprocessing, the original monitoring images of the mountainous area are input into a deep learning object detection network, which outputs smoke detection boxes and flame detection boxes. Based on the original monitoring images of the mountainous area and the flame detection frame, the flame bridging ratio is determined; Based on the smoke detection frame, the line tripping risk coefficient is determined; Based on the flame bridging ratio or the line tripping risk coefficient, the wildfire tripping / blocking risk level is determined.

[0008] Optionally, the enhancement is: A preset image enhancement algorithm is used to enhance the original monitoring image of the mountainous area to obtain an enhanced monitoring image of the mountainous area; the preset image enhancement algorithm includes adaptive histogram equalization or dark channel enhancement algorithm; The preprocessing is as follows: The enhanced monitoring images of the mountainous area are uniformly scaled to the standard inference resolution, and mean shift and variance standardization are performed.

[0009] Optionally, determining the flame bridging ratio based on the original monitoring image of the mountainous area and the flame detection frame includes: Based on the original monitoring images of the mountainous area, the two-dimensional curve equation of the power transmission line in the current image is determined; Based on the two-dimensional curve equation and the flame detection frame, the flame bridging ratio is determined.

[0010] Optionally, determining the two-dimensional curve equation of the power transmission line in the current image based on the original monitoring image of the mountainous area includes: The original monitoring images of the mountainous area are scaled and normalized to generate normalized monitoring images of the mountainous area. The normalized monitoring image of the mountainous area is input into a preset semantic segmentation network, and pixel-by-pixel binary classification prediction is performed to output a probability feature map. According to a preset probability threshold, the probability feature map is binarized to obtain the initial binary mask image of the conductor. The initial binary mask image is subjected to connected component labeling analysis, and the geometric feature set of each independent white connected region is calculated; After filtering out the geometric feature set, a clean mask image is obtained; After applying the Zhang-Suen thinning algorithm to the clean mask image, the edge pixels of the connected components of the conductors are iteratively stripped to extract the conductor skeleton line with a single pixel width; the conductor skeleton line is composed of multiple discrete pixel coordinates; Extract the pixels with a value of 1 from the conductor skeleton line and construct a two-dimensional point set of pixel coordinates; Using the principle of random sampling consistency, a parabolic fit is performed on the two-dimensional point set of pixel coordinates, and smoothing is performed using the least squares method to determine the two-dimensional curve equation of the power transmission line in the current image.

[0011] Optionally, determining the flame bridging ratio based on the two-dimensional curve equation and the flame detection frame includes: The flame detection frame is cropped to generate a flame sub-image; Based on the flame sub-image, the ignition point at the flame root and the flame tip point are determined; Substitute the abscissa of the ignition point at the root of the flame into the two-dimensional curve equation to determine the projection point of the power transmission line directly above the ignition point; In the two-dimensional pixel coordinate system of the image, calculate the vertical pixel difference between the projection point of the power transmission line and the ignition point at the root of the flame; Obtain the distance between the conductor and the ground in the current span of the video surveillance device; Based on the vertical pixel drop and the distance of the conductor to the ground, the pixel-physical dynamic mapping scaling factor of the vertical plane where the flame ignition point is located is determined. Calculate the vertical pixel difference between the flame tip and the flame root ignition point; The actual vertical height of the flame is determined based on the pixel-physical dynamic mapping scaling factor and the vertical pixel difference. The shortest pixel geometric distance from the flame tip to the two-dimensional curve equation is calculated using a point-to-parabola distance algorithm. The actual remaining air gap is calculated based on the shortest pixel geometric distance and the pixel-physical dynamic mapping scaling factor. The flame bridging ratio is calculated based on the actual remaining air gap and the distance between the conductor and the ground.

[0012] Optionally, determining the line tripping risk coefficient based on the smoke detection frame includes: The smoke detection frame is cropped to generate a smoke sub-image; Based on the smoke sub-image, the global saturation mean of the saturation S channel and the global brightness mean of the brightness V channel are determined, along with the smoke concentration index, the average entropy of the gray-level co-occurrence matrix, and the contrast. Based on the global saturation mean, global brightness mean, smoke concentration index, average entropy, and contrast, the corresponding smoke type is determined. Based on the corresponding smoke type, a smoke breakdown impact weighting coefficient is determined; the smoke breakdown impact weighting coefficient includes a smoke particle correction coefficient and a dense smoke correction coefficient. Obtain the actual vertical height of the flame; Based on the actual vertical height of the flame, determine the temperature correction factor at the conductor and the smoke temperature correction factor. The line tripping risk coefficient is determined based on the weighting coefficient of the smoke breakdown effect, the temperature correction coefficient at the conductor, and the smoke temperature correction coefficient.

[0013] Optionally, based on the smoke sub-image, the global saturation mean of the saturation S channel and the global brightness mean of the brightness V channel, the smoke concentration index, the average entropy of the gray-level co-occurrence matrix, and the contrast are determined; including: An adaptive thresholding method is used to binarize the smoke sub-image to extract and generate a set of clean smoke pixels. The color space of the pure smoke pixel set is converted from RGB to HSV space, and then the global saturation mean of the saturation S channel and the global brightness mean of the brightness V channel are calculated respectively. Based on the smoke sub-image, calculate the dark channel image corresponding to the smoke sub-image; Based on the dark channel map, the average transmittance of the smoke region is calculated; The smoke concentration index is calculated based on the average transmittance. The smoke sub-image is converted into a single-channel grayscale image, and the grayscale co-occurrence matrix in multiple directions is calculated. Extract and calculate the average entropy and contrast of the gray-level co-occurrence matrix.

[0014] Optionally, determining the temperature correction coefficient at the conductor and the smoke temperature correction coefficient based on the actual vertical height of the flame includes: The actual vertical height of the flame is calculated inversely to determine the inverted fire line intensity; Obtain the distance between the conductor and the ground in the current span of the video surveillance device; Based on the fire intensity and the distance between the conductor and the ground, the temperature rise of the flame at the conductor and the temperature rise in the middle of the smoke are determined. Get the current ambient temperature; Based on the current ambient temperature and the temperature rise, the temperature correction factor at the conductor is calculated. Based on the current ambient temperature and the temperature rise in the middle of the flue gas, a smoke temperature correction coefficient is calculated.

[0015] Optionally, determining the line tripping risk coefficient based on the smoke breakdown impact weighting coefficient, the conductor temperature correction coefficient, and the smoke temperature correction coefficient includes: Based on the actual vertical height of the flame, determine the air gap bridging type of the line affected by flame intrusion; Based on the smoke breakdown influence weighting coefficient, the conductor temperature correction coefficient, and the smoke temperature correction coefficient, the conductor-to-ground wildfire resistance strength and conductor-to-conductor wildfire resistance strength corresponding to the air gap bridging type are determined. Based on the conductor-to-ground wildfire resistance strength and the conductor-to-conductor wildfire resistance strength, the number of line tripping risks is determined.

[0016] Secondly, this application also provides a computer system, comprising: Memory is used to store instructions that can be executed by the processor; A processor for executing the instructions to implement the method as described in the first aspect.

[0017] This application has the following beneficial effects: The method proposed in this application, on the one hand, realizes the spatial quantification of physical features based on "visual perception" and breaks down the disciplinary barriers between visual features and underlying electrical engineering theory, making the trip assessment results more scientific; on the other hand, it innovatively proposes a multi-dimensional smoke fine classification algorithm, which improves the accuracy of wildfire trip assessment under smoke bridging conditions. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating a method for assessing the risk of power transmission line tripping due to wildfires based on video images, provided in an embodiment of this application. Detailed Implementation

[0019] To facilitate understanding by those skilled in the art, the present application will be further described below in conjunction with embodiments and accompanying drawings. The content mentioned in the embodiments is not intended to limit the present application.

[0020] To solve the above technical problems, such as Figure 1As shown, this application proposes a method for assessing the risk of power transmission line tripping due to wildfires based on video images, including: Step S101: Obtain original surveillance images of the mountainous area; In this embodiment, the original monitoring images of the mountainous area are acquired through a combination of periodic snapshots and triggered acquisition. Periodic snapshots refer to the automatic acquisition of high-definition panoramic or preset position images by the transmission tower monitoring device at preset time intervals (e.g., 5-10 minutes) in normal mode. Triggered acquisition refers to the immediate triggering of a high-frequency snapshot mode (e.g., 10 seconds / frame) when auxiliary sensors mounted on the tower (e.g., infrared thermometers, smoke concentration sensors, or ambient thermometers) detect abnormal value fluctuations, thereby ensuring the real-time capture of fire information. In addition to the above acquisition methods, corresponding image acquisition methods can also be set according to the specific scenario or the probability of fire occurrence in the corresponding season; this application does not impose specific limitations here.

[0021] Step S102: After enhancement and preprocessing, the original monitoring image of the mountainous area is input into a deep learning target detection network, which outputs smoke detection boxes and flame detection boxes. Due to the complex field environment, raw surveillance images from mountainous areas need to be standardized before being input into deep learning object detection networks. First, considering the characteristics of foggy and dusty mountainous areas, preset image enhancement algorithms, such as adaptive histogram equalization (AHE) or dark channel enhancement algorithms, are applied to enhance the original monitoring images of the mountainous areas, resulting in enhanced monitoring images. This improves image contrast and ensures the recognizability of smoke features at a distance. It should be noted that the preset image enhancement algorithms are not limited to the two mentioned here. Due to the differences between daytime and nighttime smoke monitoring images, different image enhancement algorithms can be set according to the corresponding time periods, thereby improving the efficiency and accuracy of smoke image recognition.

[0022] The enhanced monitoring images of the aforementioned mountainous areas were uniformly scaled to the standard inference resolution, and mean shift and variance standardization were performed to eliminate calculation bias caused by different light intensities.

[0023] The standardized image is then input into a deep learning object detection network (such as the YOLO series or Faster R-CNN). If a "flame" or "smoke" target is detected in the image, a smoke detection box and a flame detection box are output, proceeding to the next step, S103. If no "flame" or "smoke" target is detected, polling monitoring continues. This ensures that when flames or smoke are detected, subsequent processing and early warning can be performed quickly; otherwise, only monitoring is performed, saving computational resources.

[0024] Step S103: Determine the flame bridging ratio based on the original monitoring image of the mountainous area and the flame detection frame; To address the complex mountainous terrain, a hybrid algorithm was used to identify the location of the guide wires. The specific calculation steps are as follows: The original monitoring images of the mountainous area were scaled up to the standard resolution required by the model (e.g., 512). The image is 512 pixels, and pixel values ​​are normalized to generate a normalized image for mountain monitoring.

[0025] The normalized image from the mountain area surveillance is input into a pre-defined semantic segmentation network (e.g., a U-Net model based on a lightweight backbone network) for pixel-by-pixel binary classification prediction. The output is a probability feature map of the same size as the original image, where each pixel value represents the probability that it belongs to a "transmission line". Then, the probability feature map is binarized according to a pre-defined probability threshold (e.g., 0.5) to obtain the initial binary mask image of the transmission line.

[0026] Against a forest background, straight tree branches or pole edges can easily be misidentified as wires, generating noise. Therefore, a connected component labeling algorithm is used to analyze the initial binary mask image, calculating the geometric feature set of each independent white connected region, including the region area (total number of pixels) and the aspect ratio of the bounding rectangle.

[0027] Because power transmission lines have an "extremely long and thin" topological feature in the image, an area threshold is set. and aspect ratio threshold Automatically remove areas smaller than [the specified area]. And the aspect ratio is less than The connected components (i.e., filtering out blocky or short line-shaped false detection noise points) are separated, and the long strip-shaped connected components that truly belong to the wires are retained to obtain a clean mask image.

[0028] Furthermore, geometric features such as the centroid location and principal axis direction angle of the connected region can be combined to further distinguish conductors from similar elongated tree branches. For example, the principal axis direction of a conductor region is usually consistent with the direction of the transmission line, while the direction of a tree branch is random. The accuracy of identification can be improved by checking the consistency of the direction angle.

[0029] Since the wire mask obtained from coarse positioning has a certain pixel width (i.e., the wires are relatively thick), it is necessary to extract the center line of the wires to facilitate subsequent mathematical modeling. After applying the Zhang-Suen thinning algorithm to the clean mask image, edge pixels of the connected components of the conductors are iteratively stripped to extract the conductor skeleton line with a single pixel width. At this time, the extracted skeleton consists of a series of discrete pixel coordinates ( )composition.

[0030] Due to gravity, the power transmission line between the two towers forms a catenary (i.e., sag), which is approximately a parabola rather than an absolute straight line in the two-dimensional projection of the image.

[0031] Extract the pixels with a value of 1 from the conductor skeleton line and construct a two-dimensional point set S with pixel coordinates. Using the Random Sample Consensus (RANSAC) principle, a parabolic fit is performed on the two-dimensional point set of pixel coordinates: (1) Randomly select a small number of data points (e.g., 3 points) from the two-dimensional point set S of pixel coordinates, and calculate a set of initial model parameters (a, b, c) based on the fitting of the quadratic polynomial equation: (1) (2) Substitute the remaining points in the two-dimensional point set S of pixel coordinates into the initial model and calculate the geometric distance from each point to the parabola. If the distance is less than the set tolerance error... If so, then that point is considered an "in-place point".

[0032] (3) Iterate the above random sampling process multiple times and retain the set of model parameters that contains the most "inside points". ; Finally, after removing outliers that deviate from the actual conductor, the least squares method is used to perform a final smoothing fit on all local points, outputting the accurate two-dimensional curve equation of the transmission conductor in the current image: (2) To address the complex backgrounds of mountainous areas, this study innovatively combines a lightweight semantic segmentation network (U-Net) with connected component feature-based anti-interference filtering to extract the conductor skeleton. By utilizing random sampling to account for gravity-induced sag, a parabolic fit is performed, highly reproducing the actual conductor morphology. Specifically, multiple pixels are randomly selected from the pure conductor skeleton as fitting sample points. A gravity-induced sag model is introduced, and the least squares method is used to fit a parabolic equation that conforms to the characteristics of an overhead catenary conductor. This accurately recreates the actual trajectory and curvature of the conductor in image space, providing reliable geometric parameters for subsequent sag calculation, wind deflection assessment, and tripping risk evaluation.

[0033] By utilizing the inherent spatial topology of power transmission lines, the pixel coordinates of a two-dimensional image are transformed into physical spatial parameters. The specific steps are as follows: The flame detection bounding box is cropped to generate a flame sub-image. The connected components of the flame target pixels in the flame sub-image are traversed to obtain the center pixel of the lowest edge of the flame outline as the ignition point at the root of the flame. (This point approximately represents the ground), and the highest pixel of the flame outline is used as the flame tip. .

[0034] The x-coordinate of the ignition point at the base of the flame Substituting into equation (2), the ordinate is calculated. The coordinates of this point This refers to the projection point of the power transmission line directly above the point of ignition.

[0035] Since field surveillance cameras typically use telephoto lenses, perspective distortion is relatively small in the area where the fire occurs (distant area), and a linear geometric similarity model can be used to approximate the local spatial calibration.

[0036] First, calculate the projection points of the transmission line in the two-dimensional pixel coordinate system of the image. ignition point at the base of the flame Vertical pixel drop (i.e., pixel distance from the wire to the ground): (3) From the power grid GIS ledger system, retrieve the conductor-to-ground distance (i.e., the vertical physical distance from the lowest point of the conductor to the ground when it is naturally drooping without wind) for the current video surveillance device span, and record it as... (Unit: meters). Then, based on the vertical pixel drop and the distance of the wire to the ground, the "pixel-physical" dynamic mapping scaling factor K of the vertical plane where the fire point is located is calculated: (4) Calculate the vertical pixel difference between the flame tip and the ignition point at the flame root: (5) The actual vertical height of the flame is calculated based on the pixel-physical dynamic mapping scaling factor and the vertical pixel difference. : (6) Because the high-temperature ionized gas at the flame tip will drastically reduce air insulation, it is necessary to calculate the shortest physical distance between the flame tip and the conductor, rather than simply the vertical margin.

[0037] First, calculate the shortest pixel distance. Using a point-to-parabola distance algorithm, calculate the flame tip. to conductor curve The shortest pixel geometric distance, denoted as .

[0038] The actual remaining air gap (shortest physical distance) is calculated based on the shortest pixel geometric distance and the pixel-physical dynamic mapping scaling factor. : (7) To simplify the electrical model calculations, in the smoke bridging model, it is assumed that the flame core path is approximately vertical, at which point the following conditions are met. .

[0039] The flame bridging ratio is defined as the degree of safety threat posed by flame intrusion into the conductor to the ground. Considering the most severe operating conditions, the shortest possible spatial gap is used. The projection is conservatively calculated to be equivalent to the vertical height. That is, the flame bridging ratio is calculated based on the actual remaining air gap and the distance of the conductor to the ground. The calculation formula is: (8) This calculation eliminates the need for expensive lidar ranging equipment. Relying solely on existing long-range monitoring images, it achieves precise quantification of the shortest physical air gap between the flame tip and the conductor surface, while automatically calculating the flame bridging ratio. This solution effectively overcomes the shortcomings of traditional ranging methods, which rely on dedicated sensors, are costly, and complex to install. It significantly improves the perception dimension and measurement accuracy of flame spatial distribution characteristics, providing a reliable basis for the quantitative assessment of fire hazards in power transmission lines.

[0040] Step S104: Based on the smoke detection frame, determine the line tripping risk coefficient; The aforementioned smoke detection bounding box is cropped to generate a smoke sub-image. To eliminate background interference such as sky and trees, an adaptive thresholding method (or typical algorithms such as Otsu's method) is used to perform binarization masking on the smoke sub-image to extract a clean set of smoke pixels. Then, the color space of the clean smoke pixel set is converted from RGB to HSV, and the global saturation mean of the S channel is calculated. Global brightness mean of the brightness V channel .

[0041] Considering the high correlation between smoke concentration and image transmittance, i.e., the degree of background occlusion, this application utilizes dark channel prior theory to calculate smoke concentration characteristics: First, calculate the dark channel map of the smoke sub-image. It is defined as the minimum pixel value among the three RGB channels within a local window. The formula is: (9) in, One channel of a color image. In pixels A local window centered on the user.

[0042] Then, based on the atmospheric physics scattering model and the dark channel diagram, the average transmittance of the smoke region is calculated. : (10) in, This is the defogging coefficient. This represents the global atmospheric light value. Transmittance. The lower the value, the denser the smoke.

[0043] Define the smoke concentration index: (11) Right now: (12) The higher the value, the higher the smoke concentration.

[0044] Dense smoke typically appears dense and smooth internally, while light smoke often shows significant edge diffusion and a flocculent internal texture (high contrast). The smoke sub-image is converted into a single-channel grayscale image. Gray-level co-occurrence matrices are calculated for multiple directions (e.g., 0°, 45°, 90°, 135°). Finally, the average entropy (Entrop) and contrast (Contrast) of the matrices are extracted and calculated as texture feature vectors.

[0045] This application is based on the above extraction. , , Five texture features—Entropy, Contrast, and others—were used to construct a multi-dimensional cross-validation threshold determination rule. This rule, by jointly analyzing the relationship between each feature parameter and a preset threshold range, avoids the limitations of single-feature threshold judgment and directly classifies smoke states into four typical types: Saturation threshold Used for verifying the authenticity of smoke.

[0046] Brightness threshold Used to distinguish the color tone of smoke.

[0047] Transmission concentration threshold : Distinguish the physical thickness of smoke.

[0048] Entropy threshold With contrast threshold Used for texture density verification.

[0049] Condition 1 (Dense Black Smoke): If the following conditions are met ( and ), and simultaneously satisfy ( And Entropy And Contrast If the smoke is thick and black, it is classified as "thick black smoke". Condition 2 (Black Light Smoke): If satisfied ( and ), and simultaneously satisfy ( Or Entropy Or Contrast If the value is ), then it is determined to be "black light smoke"; Condition 3 (dense gray-white smoke): If the following conditions are met ( and ), and simultaneously satisfy ( And Entropy And Contrast If the smoke is grayish-white, it is determined to be "thick, grayish-white smoke"; Condition 4 (Grayish-white light smoke): If satisfied ( and ), and simultaneously satisfy ( Or Entropy Or Contrast If the value is 0, it is determined to be "grayish-white light smoke".

[0050] Anomaly exclusion branch (non-smoke interference): If This indicates that the target color is too bright and does not conform to the physical characteristics of smoke. Therefore, it is directly eliminated as a "false target" and the subsequent risk calculation is skipped.

[0051] This multi-dimensional cross-validation strategy effectively improves the accuracy and environmental adaptability of smoke classification, providing reliable technical support for early smoke identification and graded warning in scenarios such as power transmission channels and forest fire prevention.

[0052] Based on the aforementioned multi-dimensional cross-validation results classifying smoke into four types, this application further assigns a corresponding smoke breakdown influence weighting coefficient to each type of smoke according to the density differences of conductive carbon particles contained in different types of smoke and their degree of attenuation to air insulation breakdown voltage. This smoke breakdown influence weighting coefficient includes a "smoke particle correction coefficient". "and dense smoke correction factor" ": For "thick black smoke": It has an extremely high carbon content, severely damaging insulation and affecting output. =2, ; Regarding "black light smoke": it has a certain carbon content and output... =1.5, ; For "dense gray-white smoke": it mainly consists of moisture and ash, leading to a certain decrease in insulation and output coefficient. =1.2, ; For "grayish-white light smoke": almost no conductive carbon particles, minimal impact on insulation, output reference coefficient. =1.0, .

[0053] This paper innovatively proposes a multi-dimensional refined smoke classification algorithm, improving the assessment accuracy under smoke bridging conditions. Addressing the weakness of traditional algorithms that treat smoke indiscriminately, this application designs a multi-dimensional judgment rule integrating HSV color space conversion, dark channel prior theory, and gray-level co-occurrence matrix. This rule can accurately extract and calculate saturation, brightness, concentration index, average entropy, and contrast, characterizing smoke properties, precisely classifying smoke into four types. More importantly, this application assigns differentiated "smoke particle correction coefficients" and "dense smoke correction coefficients" based on the varying degrees to which the carbon content of different smoke types weakens the air insulation breakdown voltage. This fills the gap in the existing assessment system regarding the quantification of smoke insulation damage and effectively solves the problem of underreporting caused by smoldering.

[0054] The extracted vertical height of the flames is incorporated into the physical model of air gap breakdown. By inversely calculating the fire line intensity and spatial temperature distribution, the risk of power line tripping during a wildfire is quantitatively assessed. The specific steps are as follows: First, traditional fire intensity estimates often rely on macroscopic meteorological data. This application directly utilizes the actual vertical height of the flame measured from video images. Perform precise back-calculation to determine the inverted fire line intensity I: (13) Then, based on the inversely calculated fire intensity, the temperature rise of the flame at the conductor is calculated. : (14) When bridging smoke, it is necessary to calculate the temperature in the middle of the flue gas, where the temperature rise is... : (15) Combining the current ambient temperature T (°C) and temperature rise The temperature correction factor at the conductor was calculated. ; (16) Combining the current ambient temperature T and the temperature rise in the middle of the flue gas The smoke temperature correction factor was calculated, and the temperature correction factor at the conductor was calculated separately. and smoke temperature correction factor : (17) Calculate the height of the flame continuum ,Pick Based on the actual vertical height of the flame With conductor height Based on the spatial geometric relationships, determine the air gap bridging type of the line affected by flame intrusion: (1) Fully bridged flame: When At that time, the continuous flame directly contacts the wire; (2) Flame half-bridge: when At that time, discontinuous flames directly contact the conductor; (3) Smoke bridging: when At that time, only smoke enveloped the wires, with no open flame in contact.

[0055] Based on the weighting coefficient for smoke breakdown, the temperature correction coefficient at the conductor, and the smoke temperature correction coefficient, the conductor-to-ground wildfire withstand strength and conductor-to-conductor wildfire withstand strength corresponding to the air gap bridging type are determined; the specific steps are as follows: Calculate the fire resistance of conductors under different bridging types. (Unit: kV): (18) In the formula: This is an altitude correction factor; when the altitude is less than 1000m, (19) When the altitude is greater than 1000m: (20) In the formula: Altitude, in meters (m).

[0056] This is the vegetation density correction factor; (twenty one) In the formula: W represents the combustible material load, measured in tons per square kilometer (t / km²). The correction coefficients for vegetation types are: 1.13 for Yunnan pine, 1.0 for Chinese fir, 0.96 for eucalyptus, 0.93 for shrubs, and 0.88 for cogongrass. This is the correction factor for smoke particles; This is the correction factor for dense smoke.

[0057] The average breakdown electric field strength in the continuous zone of a standard flame is expressed in kilovolts per meter (kV / m), and is taken as 60 kV / m. The average breakdown electric field strength in the discontinuous zone of a standard flame is expressed in kilovolts per meter (kV / m), and is taken as 173.7 kV / m.

[0058] The average breakdown electric field strength of a standard air gap is expressed in kilovolts per meter (kV / m). The size, take (twenty two) Wire - Wire Wildfire Resistance (Unit: kV) (twenty three) In the formula: The distance between conductors is expressed in meters (m). This is the smoke height correction factor, which refers to the dilution effect of smoke near the conductor due to diffusion as the conductor moves further away from the flame. Take 1 if the height is less than 5m, otherwise take 1. (twenty four) The value is tentatively set at 0.02 (this needs to be adjusted based on the specific case).

[0059] Based on the conductor-to-ground wildfire withstand strength and the conductor-to-conductor wildfire withstand strength, the line tripping risk coefficient is determined. This risk coefficient can reflect the comprehensive threat level of wildfires to the safe operation of transmission lines in real time, providing a quantitative basis for early warning decision-making and differentiated emergency response. The specific calculation steps are as follows: AC lines The tripping coefficient for AC power lines due to wildfires is calculated using the tripping risk coefficient between the conductor and ground. and phase-to-phase trip risk coefficient The larger of the two: (25) (26) In the formula: This refers to the rated operating voltage of the line, expressed in kilovolts (kV). This refers to the withstand voltage of the wildfire gap between the conductor and ground, expressed in kilovolts (kV). The wildfire gap withstand voltage between conductors is expressed in kilovolts (kV). The maximum allowable deviation coefficient of the system is 1.15 for lines below 330kV and 1.1 for lines of 330kV and above.

[0060] DC lines: The wildfire blocking factor B for DC lines is taken as the blocking risk factor B between the conductor and ground. G1 Phase-to-phase interlocking risk coefficient BP The maximum value between the two: (27) (28) In the formula, C dc This is the voltage polarity correction factor, with 1.19 for positive polarity and 1 for negative polarity.

[0061] This application departs from extensive macro-meteorological data estimation, instead directly utilizing video-measured flame height for precise back-calculation to obtain fire line intensity and spatial temperature distribution. These visually derived parameters are then deeply coupled into a physical model of wildfire withstand strength for conductor-to-ground and conductor-to-conductor connections. This interdisciplinary innovation allows the system not only to "see the fire" but also to "calculate the electrical conductivity," outputting wildfire tripping / blocking risk coefficients that incorporate AC / DC voltage level differences. This significantly reduces false alarms caused by experience-based judgments in traditional systems, providing a solid electrophysical foundation for the assessment results.

[0062] Step S105: Determine the wildfire tripping / blocking risk level based on the flame bridging ratio or the line tripping risk coefficient.

[0063] Based on the line tripping risk coefficient / blocking risk coefficient, a dual cross-verification method using both geometric and electrical methods is employed, and the wildfire tripping / blocking risk level is determined according to Table 1. The system combines the tripping risk coefficient and the blocking risk coefficient... The risk level is compared with the predefined threshold ranges in Table 1. If either the geometric or electrical verification meets the threshold condition corresponding to the current risk level, the risk level of the wildfire tripping / lockdown is directly determined. This "OR" logic avoids missed detections due to abnormal data from a single sensor or environmental interference, while also reducing the stringency of simultaneously meeting dual conditions, ensuring timely and reliable triggering of corresponding warning and protection actions in complex wildfire scenarios. Table 1 provides an example of the risk coefficient ranges and critical conditions for geometric and electrical criteria corresponding to different risk levels. Specific values ​​can be calibrated based on line voltage levels, terrain, and climate conditions.

[0064] Table 1 Criteria for Determining the Risk Level of Wildfire Tripping / Lockdown

[0065] In summary, the method proposed in this application, on the one hand, is based on the spatial quantification of physical features of "visual perception" and breaks down the disciplinary barriers between visual features and underlying electrical engineering theory, making the tripping assessment results more scientific; on the other hand, it innovatively proposes a multi-dimensional smoke fine classification algorithm, which improves the assessment accuracy under smoke bridging conditions.

[0066] In some embodiments, this application also provides a computer system including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0067] This application also provides a computer-readable storage medium for storing a computer program. This computer-readable storage medium can be applied to a computer device, and the computer program causes the computer device to execute the corresponding processes in the methods described above in the embodiments of this application; for brevity, further details are omitted here.

[0068] The above embodiments are preferred implementations of this application. In addition, this application can be implemented in other ways. Any obvious substitutions without departing from the concept of this technical solution are within the protection scope of this application.

[0069] To facilitate understanding by those skilled in the art of the improvements made by this application compared to the prior art, some of the accompanying drawings and descriptions have been simplified, and for clarity, some other elements have been omitted from this application. Those skilled in the art should realize that these omitted elements may also constitute the content of this application.

Claims

1. A method for assessing the risk of power transmission line tripping due to wildfires based on video images, characterized in that, include: Obtain raw surveillance images of mountainous areas; After enhancement and preprocessing, the original monitoring images of the mountainous area are input into a deep learning object detection network, which outputs smoke detection boxes and flame detection boxes. Based on the original monitoring images of the mountainous area and the flame detection frame, the flame bridging ratio is determined; Based on the smoke detection frame, the line tripping risk coefficient is determined; Based on the flame bridging ratio or the line tripping risk coefficient, the wildfire tripping / blocking risk level is determined.

2. The method according to claim 1, characterized in that, The enhancement is as follows: A preset image enhancement algorithm is used to enhance the original monitoring image of the mountainous area to obtain an enhanced monitoring image of the mountainous area; the preset image enhancement algorithm includes adaptive histogram equalization or dark channel enhancement algorithm; The preprocessing is as follows: The enhanced monitoring images of the mountainous area are uniformly scaled to the standard inference resolution, and mean shift and variance standardization are performed.

3. The method according to claim 2, characterized in that, The step of determining the flame bridging ratio based on the original monitoring image of the mountainous area and the flame detection frame includes: Based on the original monitoring images of the mountainous area, the two-dimensional curve equation of the power transmission line in the current image is determined; Based on the two-dimensional curve equation and the flame detection frame, the flame bridging ratio is determined.

4. The method according to claim 3, characterized in that, The step of determining the two-dimensional curve equation of the power transmission line in the current image based on the original monitoring image of the mountainous area includes: The original monitoring images of the mountainous area are scaled and normalized to generate normalized monitoring images of the mountainous area. The normalized monitoring image of the mountainous area is input into a preset semantic segmentation network, and pixel-by-pixel binary classification prediction is performed to output a probability feature map. According to a preset probability threshold, the probability feature map is binarized to obtain the initial binary mask image of the conductor. The initial binary mask image is subjected to connected component labeling analysis, and the geometric feature set of each independent white connected region is calculated; After filtering out the geometric feature set, a clean mask image is obtained; After applying the Zhang-Suen thinning algorithm to the clean mask image, the edge pixels of the connected components of the conductors are iteratively stripped to extract the conductor skeleton line with a single pixel width; the conductor skeleton line is composed of multiple discrete pixel coordinates; Extract the pixels with a value of 1 from the conductor skeleton line and construct a two-dimensional point set of pixel coordinates; Using the principle of random sampling consistency, a parabolic fit is performed on the two-dimensional point set of pixel coordinates, and smoothing is performed using the least squares method to determine the two-dimensional curve equation of the power transmission line in the current image.

5. The method according to claim 4, characterized in that, The determination of the flame bridging ratio based on the two-dimensional curve equation and the flame detection frame includes: The flame detection frame is cropped to generate a flame sub-image; Based on the flame sub-image, the ignition point at the flame root and the flame tip point are determined; Substitute the abscissa of the ignition point at the root of the flame into the two-dimensional curve equation to determine the projection point of the power transmission line directly above the ignition point; In the two-dimensional pixel coordinate system of the image, calculate the vertical pixel difference between the projection point of the power transmission line and the ignition point at the root of the flame; Obtain the distance between the conductor and the ground in the current span of the video surveillance device; Based on the vertical pixel drop and the distance of the conductor to the ground, the pixel-physical dynamic mapping scaling factor of the vertical plane where the flame ignition point is located is determined. Calculate the vertical pixel difference between the flame tip and the flame root ignition point; The actual vertical height of the flame is determined based on the pixel-physical dynamic mapping scaling factor and the vertical pixel difference. The shortest pixel geometric distance from the flame tip to the two-dimensional curve equation is calculated using a point-to-parabola distance algorithm. The actual remaining air gap is calculated based on the shortest pixel geometric distance and the pixel-physical dynamic mapping scaling factor. The flame bridging ratio is calculated based on the actual remaining air gap and the distance between the conductor and the ground.

6. The method according to claim 1, characterized in that, The determination of the line tripping risk coefficient based on the smoke detection frame includes: The smoke detection frame is cropped to generate a smoke sub-image; Based on the smoke sub-image, the global saturation mean of the saturation S channel and the global brightness mean of the brightness V channel are determined, along with the smoke concentration index, the average entropy of the gray-level co-occurrence matrix, and the contrast. Based on the global saturation mean, global brightness mean, smoke concentration index, average entropy, and contrast, the corresponding smoke type is determined. Based on the corresponding smoke type, a smoke breakdown impact weighting coefficient is determined; the smoke breakdown impact weighting coefficient includes a smoke particle correction coefficient and a dense smoke correction coefficient. Obtain the actual vertical height of the flame; Based on the actual vertical height of the flame, determine the temperature correction factor at the conductor and the smoke temperature correction factor. The line tripping risk coefficient is determined based on the weighting coefficient of the smoke breakdown effect, the temperature correction coefficient at the conductor, and the smoke temperature correction coefficient.

7. The method according to claim 6, characterized in that, Based on the smoke sub-image, determine the global saturation mean of the S-channel and the global brightness mean of the V-channel, the smoke density index, the average entropy of the gray-level co-occurrence matrix, and the contrast; including: An adaptive thresholding method is used to binarize the smoke sub-image to extract and generate a set of clean smoke pixels. The color space of the pure smoke pixel set is converted from RGB to HSV space, and then the global saturation mean of the saturation S channel and the global brightness mean of the brightness V channel are calculated respectively. Based on the smoke sub-image, calculate the dark channel image corresponding to the smoke sub-image; Based on the dark channel map, the average transmittance of the smoke region is calculated; The smoke concentration index is calculated based on the average transmittance. The smoke sub-image is converted into a single-channel grayscale image, and the grayscale co-occurrence matrix in multiple directions is calculated. Extract and calculate the average entropy and contrast of the gray-level co-occurrence matrix.

8. The method according to claim 6, characterized in that, The determination of the temperature correction coefficient at the conductor and the smoke temperature correction coefficient based on the actual vertical height of the flame includes: The actual vertical height of the flame is calculated inversely to determine the inverted fire line intensity; Obtain the distance between the conductor and the ground in the current span of the video surveillance device; Based on the fire intensity and the distance between the conductor and the ground, the temperature rise of the flame at the conductor and the temperature rise in the middle of the smoke are determined. Get the current ambient temperature; Based on the current ambient temperature and the temperature rise, the temperature correction factor at the conductor is calculated. Based on the current ambient temperature and the temperature rise in the middle of the flue gas, a smoke temperature correction coefficient is calculated.

9. The method according to claim 6, characterized in that, The determination of the line tripping risk coefficient based on the smoke breakdown impact weighting coefficient, the conductor temperature correction coefficient, and the smoke temperature correction coefficient includes: Based on the actual vertical height of the flame, determine the air gap bridging type of the line affected by flame intrusion; Based on the smoke breakdown influence weighting coefficient, the conductor temperature correction coefficient, and the smoke temperature correction coefficient, the conductor-to-ground wildfire resistance strength and conductor-to-conductor wildfire resistance strength corresponding to the air gap bridging type are determined. Based on the conductor-to-ground wildfire resistance strength and the conductor-to-conductor wildfire resistance strength, the number of line tripping risks is determined.

10. A computer system, characterized in that, include: Memory is used to store instructions that can be executed by the processor; A processor for executing the instructions to implement the method as described in any one of claims 1 to 9.

Images