Methods for predicting and controlling the withstand voltage of post insulators for anti-icing skirts in converter stations

Abstract

This invention discloses a method for predicting and controlling the withstand voltage of post insulators with interleaved anti-icing skirts in converter stations, relating to the technical field of converter stations. Based on the structure of post insulators with interleaved large skirts within the station, this invention conducts flashover tests under different icing conditions to obtain the flashover voltage of typical interleaved skirt post insulators under different icing gap distances and pollution levels, establishing an ice flashover voltage fitting model. Considering monitoring conditions within the converter station, maintenance personnel or monitoring devices take images of the insulators from multiple angles at regular intervals during periods of severe icing. Combining meteorological conditions and environmental factors, the development of icing is comprehensively predicted. Based on the icing development and pollution levels, the ice flashover voltage fitting model is used to predict the ice flashover voltage value, determine the safe operating voltage for the next time period, and decide whether voltage reduction is necessary and the magnitude of the reduction.

Description

Technical Field

[0001] This invention relates to the field of converter station technology, specifically a method for predicting and controlling the withstand voltage of post insulators for interleaved anti-icing skirts in converter stations. Background Technology

[0002] In recent years, frequent weather events, particularly rain and snow in winter, have easily caused icing on insulators in transmission lines, AC converter stations, and DC converter stations. Under icing conditions, the insulator flashover voltage may be lower than the normal operating voltage, leading to tripping accidents and severely impacting the safe and stable transmission of electricity. When icing is severe and the risk of tripping is high, AC lines need to be shut down. DC transmission differs from AC transmission lines; while voltage regulation can achieve reduced-voltage operation, it requires monitoring the development of icing conditions and predicting the insulator flashover voltage to implement appropriate operational controls in advance.

[0003] The main factors affecting insulator ice flashover are: 1. After the insulator string is covered with ice, the ice layer becomes thicker and may bridge the insulator skirts, turning the discharge path that was originally along the edge of the insulator into a path bridged by the ice layer; 2. During the icing process, the ice water contains pollutants, and the conductivity of the ice surface will also increase when it is about to melt, which will affect the flashover voltage.

[0004] Most current research focuses on icing flashover tests on insulators suspended on transmission lines, including different types of glass insulators, porcelain insulators, and composite insulators, analyzing the impact of their structure on icing conditions and flashover voltage. Existing patents often conduct risk assessments or predict ice flashover in transmission line insulators icing. For example, one patent (CN201610698549.1) assesses icing severity by analyzing the icing state through images of the insulator's umbrella extension. This involves taking photos of the insulator before and after icing using a fixed camera and evaluating the degree of icing by observing changes in the extension of the insulator umbrella. Another patent (CN202211338519.1) defines icing severity as five levels: slight, moderate, heavy, and severe. The corresponding criteria are: no obvious ice layer; ice ridge length reaches 1 / 3 of the gap distance; ice ridge length reaches 1 / 2 of the gap distance; ice ridge length reaches 2 / 3 of the gap distance; and ice ridges are basically bridged except for gaps at the high-voltage end. This method primarily judges insulator performance based on factors such as electric field non-uniformity, icing severity, and ice surface condition, directly defining risk values ​​and then calculating the risk using formulas. A patented method for predicting insulator ice flashover risk based on a deduction rule (patent number CN202011015869.5) also employs image processing to assess icing and other characteristic parameters. It scores parameters such as icing type, icing degree, bridging status, and contaminated surface, and then makes a judgment. Another patented method for assessing the risk of flashover due to icing insulators in ultra-high voltage AC transmission lines (patent number CN201610217229.X) primarily targets icing insulators on transmission lines. Due to the long length of the lines and the difficulty in deploying monitoring devices along the entire length, the method primarily estimates the icing weight based on meteorological data, and then uses the line insulator flashover voltage correction formula to calculate the flashover voltage value under icing conditions.

[0005] The aforementioned existing technologies primarily target line insulators, comprehensively considering factors such as ice thickness, ice weight, and bridging distance when assessing icing. This differs somewhat from intermittent icing umbrellas used for post insulators. Intermittent insulators have larger skirts, and the post insulators are vertical. The smaller umbrellas in the middle of the intermittent icing umbrella skirts experience less icing, and the bridging effect of the large umbrella group is the primary factor affecting flashover voltage. Furthermore, the aforementioned patents do not consider predicting icing development in conjunction with meteorological conditions. Station maintenance personnel can conduct periodic inspections and take photos, or combine these with monitoring devices, making it more convenient and effective than transmission line monitoring. Multiple monitoring results can be combined to dynamically correct and predict icing development. Moreover, existing patents are mostly risk assessments, with relatively coarse parameter classifications. Risk assessments often use scoring mechanisms, lacking accuracy and effectiveness. The correction formulas for line insulator ice flashover in some patents are also difficult to apply to post insulators with intermittent icing umbrellas within converter stations. Summary of the Invention

[0006] To address the aforementioned technical problems, the present invention aims to provide a method for predicting and controlling the withstand voltage of post insulators with intermittent anti-icing umbrella skirts in converter stations. This method considers the flashover characteristics of post insulators with intermittent anti-icing umbrellas in converter stations, selects appropriate characteristic parameters based on experiments, fits and corrects the flashover voltage under different operating conditions, proposes a method for predicting icing development, determines the corresponding safe operating voltage, decides whether voltage reduction is required and the magnitude of voltage reduction, and ensures the safe and stable operation of DC transmission. Specifically, based on the structure of the post insulators with interleaved large umbrella skirts within the station, flashover tests were conducted under different icing conditions to obtain the flashover voltage of typical interleaved umbrella skirt post insulators under different icicle bridging gap distances and pollution levels, and an ice flashover voltage fitting model was established. Considering the monitoring conditions within the converter station, maintenance personnel or monitoring devices took images of the insulators from multiple angles at regular intervals when icing was severe. Combining meteorological conditions and environmental factors, the development of icing was comprehensively predicted. Based on the icing development and pollution conditions, the ice flashover voltage fitting model was used to predict the ice flashover voltage value, determine the safe operating voltage for the next time period, and decide whether voltage reduction is required and the magnitude of the voltage reduction.

[0007] This invention provides a method for predicting and controlling the withstand voltage of post insulators for interleaved anti-icing skirts in converter stations, comprising the following steps:

[0008] Step 1: Using test samples with the same structure as the post insulators with intermittent anti-icing umbrella skirts installed in the DC converter station, the post insulator test samples were coated with dirt according to different pollution levels. Then, icing was simulated in the environmental climate laboratory to obtain post insulator test samples under typical ice bridging conditions.

[0009] Step 2: Using the ice-melting flashover test method, conduct ice flashover tests on post insulator specimens under typical pollution conditions and ice bridging conditions. Perform M tests for each working condition, record and calculate the average flashover voltage;

[0010] Step 3: Combining the pollution level and ice bridging situation, obtain the mapping relationship between the two main influencing factors and the ice flashover voltage, and establish a fitting model for the ice flashover voltage of the post insulator with intermittent anti-icing umbrella skirt.

[0011] Step 4: Measure the equivalent salt density of the standby insulators located next to the operating post insulators in the DC converter station, which are in the same environment, to determine the pollution level;

[0012] Step 5: From the start of icing, at intervals of T ps By using maintenance and inspection personnel or monitoring devices, images of ice accumulation on the support insulators from all angles can be captured to obtain information on the bridging gaps caused by ice floes.

[0013] Step 6: Based on the monitoring data of meteorological factors within the converter station, establish an ice growth prediction model based on SVM, and then use interval T... ps Images of insulators covered with ice were used to train and optimize the model until the prediction accuracy of ice growth length was ≥η%, which was then used for dynamic prediction of ice growth in subsequent ice-covered areas.

[0014] Step 7: When an icing disaster occurs at the converter station, the length of ice growth is predicted according to Step 6. Based on the measured degree of pollution and the predicted ice bridging distance, combined with the ice flashover voltage fitting model of the post insulator in Step 3, the ice flashover voltage of the post insulator is predicted. If the ice flashover voltage value is higher than the rated voltage, normal operation continues. If it is lower than the corresponding voltage, the voltage is reduced. Dynamic prediction is performed at regular intervals to determine the operating voltage value.

[0015] Preferably, in step 1, the dirt levels are divided into N types based on a preset level of dirtiness, and dirtiness is simulated according to the corresponding equivalent salt density and equivalent ash density, wherein each dirtiness level takes the median value; soluble substances are simulated using pure NaCl, and insoluble substances are simulated using diatomaceous earth, in a ratio of 1:X. h Mix the materials to form a simulated contaminant, and then manually and evenly apply the contaminant to the cleaned insulator surface using the solid layer method. Place the insulator in a dry and ventilated place. gz Time is pending.

[0016] Preferably, the icing method employs live icing. Tests have shown that live icing at the rated operating phase voltage will cause flashover during the icing process; therefore, the rated voltage k is used. N %Electrified icing. Before spraying icing, the insulators are pretreated by spraying water onto their surface to form a water film, which then forms an ice film to prevent contamination from washing away. Cooling water is then sprayed onto the post insulators from nozzles on both sides, gradually forming an ice layer under the low temperatures of an environmental climate laboratory. Icing is stopped when the ice layer bridges N different gap distances, resulting in N test samples for each level of contamination.

[0017] Preferably, in step 2, considering the presence of a water film during actual ice melting, which makes flashover more likely, the ice flash test considers a more stringent situation, namely, adopting the ice melting flashover test method. After the ice covers the ice to the corresponding ice length, the temperature in the climate environment room is controlled to rise to a certain temperature T1 close to 0 degrees Celsius, and then the temperature is increased at a rate of St°C / h. When there is a water film on the ice surface and water droplets are left on the ice, the conditions for starting the ice flash test are met, and pressurization is prepared to begin.

[0018] Preferably, the following pressurization method is used: During the first pressurization, the approximate range of the ice flash voltage has not yet been obtained; at this time, according to U... t The voltage is increased at a rate of kV / s until a flashover occurs, and the flashover voltage U at this point is recorded. tcSubsequent tests will first rapidly boost the voltage to U. tc k tc %, and then each time U increases tc k jy %, and maintain voltage withstand time T ns In T ns The voltage is then increased again until breakdown occurs, thus obtaining an ice flash voltage that is closer to the actual situation.

[0019] Preferably, in step 3, by combining the pollution level and the ice bridging situation, the mapping relationship between the two main influencing factors and the ice flashover voltage is obtained, and a fitting model for the ice flashover voltage of the post insulator with intermittent anti-icing umbrella skirt is established.

[0020] To ensure the effectiveness of the experiment and the reliability of the data, the pollution levels of the insulators were assigned values ​​of ESDD1, ESDD2, ESDD3, ESDD4, and ESDD5, respectively. The length of the insulator icicles was represented by the bridging ratio between the large umbrella skirts, which were p1, p2, p3, p4, and p5, respectively. The pollution level and icicle bridging ratio of each insulator were combined in pairs, and three repeated breakdown tests were carried out. Abnormal situations were excluded, and the average breakdown voltage was used to fit the model.

[0021] (1)

[0022] In the formula, p is the ice bridging ratio, l is the length of the insulator ice, and d is the spacing between the insulator skirts;

[0023] The insulator flashover voltage decreases with increasing pollution level, while the insulator ice flashover voltage decreases with increasing ice length. A surface-fitting mathematical model is constructed by iterating the function multiple times using the least squares approximation method. This allows for the calculation and prediction of the breakdown voltage corresponding to pollution level and bridging ratio, excluding characteristic parameters. The surface fit is R. 2 The expression for the surface is shown in equation (2):

[0024] (2)

[0025] In the formula, U is the breakdown voltage under different salt density and ice bridging ratios, x is the ice bridging ratio, and y is the insulator salt density.

[0026] Preferably, in step 4, the normally operating post insulators cannot be removed for testing. Several post insulators specifically for testing are placed next to the normally operating insulators under the same natural environment and in the same pollution conditions. The corresponding insulators with an area of ​​S are then tested. jc The filth from the area was collected and dissolved to a volume of V. jcIn distilled water, the conductivity of the solution is measured. Then, pure NaCl is added to the same volume of water. When the conductivity of the solution is the same as the conductivity of the dissolved impurities, the number of milligrams of pure NaCl at this point is recorded. Dividing this number by the area yields the equivalent salt density, expressed in mg / cm³. 2 And determine the level of filth.

[0027] Preferably, in step 5, the specific processing method is as follows:

[0028] S5.1: Images of iced insulators are taken by staff at regular intervals to obtain comprehensive icing conditions of the insulators at different monitoring stages. Before taking the images, the structural characteristics and parameters of the insulator are obtained to calibrate the images. By using the object's length, area, and feature point positions, the relationship between image pixels and the object's actual physical dimensions is established to avoid errors caused by viewing angle and lens distortion, thereby improving the accuracy of image measurement and analysis. When taking multi-angle images of iced insulators, obstruction and shaking should be avoided during shooting. Appropriate focus and aperture should be adjusted, and supplementary lighting should be used when necessary to obtain comprehensive and clear images of the iced insulators. In addition, the shooting positions should be repeated at different times to avoid image processing errors.

[0029] S5.2: When processing images of ice-covered insulators, we focus on extracting image features such as ice length and bridging ratio. We use grayscale conversion and median filtering algorithms to preprocess the images, enhance target features, and reduce background noise, so as to facilitate subsequent calibration and feature extraction. For the color features of ice-covered insulators and the sensitivity of human eye perception, we use a weighted average method for grayscale conversion, as shown in Equation (3):

[0030] (3)

[0031] In the formula, M is the weighted gray value, and R, G, and B are the components of the three colors: red, green, and blue, respectively.

[0032] To separate the insulator image from its background image, the pixels of both are divided according to their gray values, and then image thresholding is achieved based on the difference in gray levels.

[0033] (4)

[0034] In the formula, f(x, y) is the original image, g(x, y) is the segmented image, and t1 and t2 are the grayscale ranges of the segmentation;

[0035] The Canny edge detection algorithm is used to extract the contour and feature edges of insulators. The calculation process includes image size and gradient direction calculation, non-maximum suppression, double threshold filtering, and edge connection.

[0036] (5)

[0037] (6)

[0038] (7)

[0039] In the formula, g x (x, y), g y (x, y) are the partial derivatives of the graph in the x and y directions, respectively, and g 45 (x, y), g 135 (x, y) are the partial derivatives of the image at 45° and 135°, respectively, e x e y Let F(x, y) be the first-order partial derivative of the image in the x and y directions, and let g be the magnitude of the gradient. d (x, y) represents the direction of the gradient;

[0040] S5.3: Using the image processing and edge detection methods described above, the edge features of the un-iced insulator image in the same orientation are obtained. The edges of the iced and un-iced insulator images are compared, and the edge difference is the ice thickness of the iced insulator. Each pixel of the skirt edge in the un-iced insulator image is covered onto the iced insulator image, which serves as the starting point for ice growth. This pixel is the seed pixel. If there is a pixel with the same name below or around the seed pixel in the iced insulator image, it is considered that the ice is growing towards that point, and this pixel is used as the new seed pixel. The above growth process is repeated to continuously find the pixels in the iced insulator image. Combined with the image calibration in step S5.1, the ice length of the iced insulator can be obtained. Then, the proportion of ice bridging the gap between the insulators can be calculated.

[0041] Repeat steps S5.1 and S5.2 above to calculate the length of ice floes and bridging conditions across the entire ice-covered insulator, and take the maximum ice floe length to calculate the gap bridging ratio of the entire insulator.

[0042] Preferably, in step 6, an SVM-based ice growth prediction model is established by combining the monitoring data of meteorological factors within the converter station, and the model is trained and optimized until the prediction accuracy of ice growth length is less than a certain threshold, so as to be used for the dynamic prediction of subsequent ice growth.

[0043] The modeling steps for the ice crystal growth prediction model are as follows:

[0044] S6.1: The icing growth process on the insulator surface is strongly correlated with temperature, humidity, rainfall, wind speed, and wind direction. Furthermore, the icing growth rate is non-linear. Lower temperatures are a necessary condition for the phase change freezing of liquid water or water mist. Within the range of -25 to 0℃, the lower the temperature, the higher the probability of the insulator capturing and adsorbing liquid water, and the faster the icing rate. Humidity characterizes the content of liquid water in the atmosphere; higher humidity results in more water droplets being captured on the insulator surface. In rainfall conditions, larger raindrop diameters increase the probability of colliding with the insulator surface, leading to greater icing. Wind speed... Wind direction directly affects the location and shape of icing, and under the action of convective heat transfer, it will accelerate the freezing rate of liquid water. Meteorological data is obtained from the meteorological monitoring device of the converter station and the local meteorological platform. The meteorological monitoring device includes a main unit, temperature sensor, humidity sensor, wind speed sensor, wind direction sensor, rainfall sensor and air pressure sensor to dynamically monitor the environmental conditions of the converter station and send meteorological data to the management backend in real time via wireless network. The system can also obtain real-time forecast data such as temperature, humidity and rainfall from the meteorological platform to predict the growth of insulator icing.

[0045] S6.2: The SVM model has good performance in dealing with the multidimensional factors affecting insulator icing in complex environments. Considering factors such as ambient temperature, humidity, wind speed, rainfall, and raindrop diameter, an insulator icing growth prediction model based on the SVM model is established.

[0046] The extracted feature parameters affecting ice growth directly impact the model's prediction results. Too many uncorrelated or weakly correlated feature vectors will reduce the SVM prediction accuracy. Therefore, to improve model accuracy, the Pearson correlation coefficient method is used to select feature parameters. The calculation formula is as follows:

[0047] (8)

[0048] In the formula, X i and Y i These represent the i-th values ​​of a certain characteristic parameter X and the ice length Y within a unit of time;

[0049] The features are selected for training the model. The temperature change, humidity change, wind speed change, rainfall change, and raindrop diameter change per unit time are shown in equations (9)-(13).

[0050] (9)

[0051] (10)

[0052] (11)

[0053] (12)

[0054] (13)

[0055] In the formula, ∆t represents the unit time interval, taken as 10 minutes, and T i+1 T i Let R represent the temperature at time i and after time ∆t, respectively. i+1 R i V represents the humidity at time i and after time ∆t, respectively. i+1 V i Let J represent the wind speed at time i and after time ∆t, respectively. i+1 J i Y represents the rainfall at time i and after time ∆t, respectively. i+1 Y i Let represent the diameter of the raindrop at time i and after time ∆t, respectively;

[0056] To eliminate the influence of dimensions, each characteristic parameter is normalized.

[0057] (14)

[0058] In the formula, x i Let x be the feature parameters of a certain input model. min and x max These are the maximum and minimum values ​​of the parameter, x. i ' represents the normalized value;

[0059] S6.3: The SVM model is used to predict the length of insulator icicles per unit time. The obtained data on insulator icicle growth includes insulator structural features, electrical features, environmental features, and the corresponding length of icicle growth per unit time. A large amount of sample data is divided into training set and test set. The training set is input into the initial SVM model to obtain the corresponding penalty coefficient and kernel function. Then, the test set data is input into the model, and the length of icicles in the test set data is compared with the length predicted by the SVM. When the prediction error is greater than k%, the improved GS algorithm is used to optimize the parameters of the support vector machine model until the error meets the expected requirements, and the final optimized SVM model is obtained.

[0060] The final SVM model has a penalty coefficient of C, a kernel function of γ, and a mean absolute percentage error of k1% on the test set.

[0061] Therefore, based on the current temperature, humidity, and wind speed meteorological data, combined with the meteorological data predicted by the meteorological platform, the meteorological changes after a unit of time can be obtained. Then, the meteorological change data can be substituted into the SVM ice prediction model to obtain the ice growth length at the next moment. Furthermore, the ice bridging ratio can be calculated based on the ice length and the insulator gap distance.

[0062] S6.4: Before making actual predictions using an SVM model trained on historical data, the current environment and parameter characteristics can be obtained for targeted model correction. For example, when making a prediction using the model, initial meteorological data can be obtained and substituted into the model. The GS algorithm can be used to optimize the model again to obtain a new SVM prediction model. After the model is optimized in a targeted manner, its prediction results are closer to the actual situation. Then, based on the current temperature, humidity, and wind speed meteorological data, combined with the meteorological data predicted by the meteorological platform, the meteorological changes after a unit of time can be obtained, and then predictions can be made to obtain the ice growth length and gap bridging ratio at the next moment.

[0063] Preferably, in step 7, when an icing disaster occurs at the converter station, the length of ice growth is predicted according to step 6. Based on the measured degree of pollution and the predicted ice bridging distance, combined with the ice flashover voltage fitting model of the post insulator in step 3, the ice flashover voltage of the post insulator is predicted. If the ice flashover voltage value is higher than the rated voltage, normal operation continues. If it is lower than the corresponding voltage, voltage reduction operation is performed. Dynamic prediction is performed at regular intervals to determine the operating voltage value.

[0064] Compared with the prior art, the beneficial effects of the present invention are:

[0065] 1. Based on the experimental results, when the anti-icing skirts are inserted between the post insulators, the main icing and ice formation are mainly concentrated on the large skirts. The breakdown is mainly related to the pollution conditions and the ice bridging distance. This invention differs from the ice flashover risk prediction method used in existing patents. It no longer relies on a multi-factor scoring mechanism. Based on the condition of the anti-icing skirts inserted between the post insulators, ice flashover tests are conducted. Based on the test results, the influencing factors are reasonably simplified. Considering that only salt can be easily introduced during the icing process after the outside of the insulator is covered with ice, and will affect the discharge during the melting of ice, the test results are combined with only two parameters: equivalent salt density and ice bridging. The ice flashover voltage curve is obtained by fitting. Compared with the multi-factor scoring mechanism, this method is more accurate in determining whether the post insulator is experiencing ice flashover under the operating voltage.

[0066] 2. Existing patents mainly focus on identifying existing icing conditions. However, DC control requires predicting the length of icing over a certain period and whether flashover will occur during normal operation. This necessitates developing control strategies based on the corresponding flashover voltage values ​​to achieve a special DC voltage reduction operation mode. The voltage reduction process takes time, and safety must be ensured during this period. Therefore, this invention, based on existing image recognition and considering the characteristics of converter station inspections, takes into account the long icing development time. It establishes an icing growth model by periodically monitoring icing growth in the early stages of an ice storm, and dynamically corrects the model based on actual conditions. This allows for more accurate prediction of icing growth and further prediction of the flashover characteristics of post insulators. Attached Figure Description

[0067] Figure 1 This is a surface fitting diagram of the insulator breakdown voltage.

[0068] Figure 2 This is a flowchart of image processing for ice-covered insulators.

[0069] Figure 3 This is a flowchart for predicting ice length.

[0070] Figure 4 A flowchart for ice flashover prediction and operation control of post insulators. Detailed Implementation

[0071] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0072] like Figure 4 As shown, this invention provides a method for predicting and controlling the withstand voltage of post insulators for anti-icing skirts in converter stations, comprising the following steps:

[0073] Step 1: Using test samples with the same structure as the post insulators with intermittent anti-icing umbrella skirts installed in the DC converter station, the post insulator test samples were coated with dirt according to different pollution levels. Then, icing was simulated in the environmental climate laboratory to obtain post insulator test samples under typical ice bridging conditions.

[0074] Step 2: Using the ice-melting flashover test method, conduct ice flashover tests on post insulator specimens under typical pollution conditions and ice bridging conditions. Perform 3 tests for each working condition, record and calculate the average flashover voltage;

[0075] Step 3: Combining the pollution level and ice bridging situation, obtain the mapping relationship between the two main influencing factors and the ice flashover voltage, and establish a fitting model for the ice flashover voltage of the post insulator with intermittent anti-icing umbrella skirt.

[0076] Step 4: Measure the equivalent salt density of the standby insulators located next to the operating post insulators in the DC converter station, which are in the same environment, to determine the pollution level;

[0077] Step 5: From the start of icing, at intervals of T ps By using maintenance and inspection personnel or monitoring devices, comprehensive images of the ice-covered support insulators can be captured to obtain information on the bridging gaps caused by ice frost.

[0078] Step 6: Based on the monitoring data of meteorological factors within the converter station, establish an ice growth prediction model based on SVM, and then use interval T... ps Images of insulators covered with ice were used to train and optimize the model until the prediction accuracy of ice growth length was ≥η%, which could then be used for dynamic prediction of subsequent ice growth.

[0079] Step 7: When an icing disaster occurs at the converter station, the ice growth length is predicted according to Step 6. Based on the measured pollution level and the predicted ice bridging distance, combined with the ice flashover voltage fitting model of the post insulators in Step 3, the ice flashover voltage of the post insulators is predicted. If the ice flashover voltage value is higher than the rated voltage, normal operation continues; if it is lower than the corresponding voltage, voltage reduction operation is implemented. Dynamic prediction is performed at regular intervals to determine the operating voltage value.

[0080] In step 1, five levels of filth are classified according to their severity, and filth is simulated using corresponding equivalent salt density and equivalent ash density. Each filth level uses a median value of 0.015 mg / cm³. 2 0.045 mg / cm 2 0.08 mg / cm 2 0.175 mg / cm 2 0.3 mg / cm 2 Soluble substances were simulated using pure NaCl, and insoluble substances were simulated using diatomaceous earth, in a ratio of 1:X. h Mix the materials to form a simulated pollutant, and then manually and evenly apply the pollutant to the cleaned insulator surface using the solid layer method. Place the insulator in a dry and ventilated place for 7 days before use.

[0081] Table 1 Classification of Filth Levels

[0082]

[0083] The icing method employed was live-line icing. Tests showed that live-line icing at the rated operating phase voltage would cause flashover during the icing process; therefore, 80% of the rated voltage was used for live-line icing. Before spray icing, the insulators were pre-treated by spraying a water film onto their surface to form an ice film, preventing surface contaminants from washing away. Then, cooling water was sprayed onto the post insulators from nozzles on both sides, gradually forming an ice layer under the low temperatures of an environmental climate laboratory. Icing was stopped when the ice layer bridged five different gap distances, resulting in five test samples for each level of contamination.

[0084] In step 2, considering the presence of a water film during actual ice melting, which makes flashover more likely, the ice flashover test takes a more stringent approach, employing the ice melting flashover test method. After the ice layer reaches the corresponding ice length, the temperature in the controlled environment room is raised to -2°C, and then the temperature is increased at a rate of 5°C / h. When a water film forms on the ice surface and water droplets remain on the ice, the conditions for starting the ice flashover test are met, and pressurization is prepared to begin.

[0085] Considering the characteristics of ice melting, ice flashover may occur during the water film dripping process. The following pressurization method is adopted: During the first pressurization, the approximate range of the ice flashover voltage is not yet obtained. At this time, the voltage is increased at a rate of 3kV / s until flashover occurs, and the flashover voltage U at this time is recorded. tc Subsequent tests will first rapidly boost the voltage to U. tc 70%, then each time U increases tc The voltage is increased by 2%, and the voltage withstand time is maintained for 5 seconds. After 5 seconds, the voltage is increased again until breakdown occurs. The ice flash voltage obtained in this way is closer to the actual situation.

[0086] In step 3, by combining the pollution level and the ice bridging situation, the mapping relationship between the two main influencing factors and the ice flashover voltage is obtained, and a fitting model for the ice flashover voltage of the post insulator with intermittent anti-icing umbrella skirt is established.

[0087] To ensure the validity of the experiment and the reliability of the data, the pollution level of the insulator was set to 0 mg / cm³. 2 0.045 mg / cm 2 0.08 mg / cm 2 0.175 mg / cm 2 0.3 mg / cm 2 The length of the insulator ice bridging is represented by the bridging ratio between the large umbrella skirts, which are 0%, 25%, 50%, 75%, and 100%. The pollution level and ice bridging ratio of each insulator are paired, and three repeated breakdown tests are carried out to eliminate abnormal situations. The average breakdown voltage of the insulator is used to fit the model, as shown in Table 2.

[0088] (1)

[0089] In the formula, p is the ice bridging ratio, l is the length of the insulator ice, and d is the spacing between the insulator skirts.

[0090] Table 2 Insulator breakdown voltage under different operating conditions

[0091]

[0092] When the insulator is free of pollution, the breakdown voltages of the insulator under different ice bridging ratios are 722.2kV, 675.5kV, 618.3kV, 550.8kV, and 471.1kV, respectively; when the insulator is free of ice, the breakdown voltages under different pollution levels are 722.2kV, 689.3kV, 642.6kV, 580.2kV, and 493.6kV, respectively.

[0093] The flashover voltage of an insulator decreases with increasing pollution level, while the flashover voltage of an insulator decreases with increasing ice length. By iterating the function multiple times using the least squares approximation method, a surface-fitting mathematical model is constructed. This allows for the calculation and prediction of the breakdown voltage corresponding to pollution level and bridging ratio, excluding characteristic parameters. The fitted surface is shown below. Figure 1 As shown, the fit of the surface is 0.997. The expression of the surface is shown in equation (2).

[0094] (2)

[0095] In the formula, U is the breakdown voltage under different salt density and ice bridging ratios, x is the ice bridging ratio, and y is the insulator salt density.

[0096] In step 4, the normally operating post insulators cannot be removed for testing. Several post insulators specifically designed for testing will be placed next to the normally operating insulators under the same environmental conditions and pollution levels. The corresponding insulators will have an area of ​​2000 cm². 2 The contaminants in the area were collected and dissolved in 500 mL of distilled water. The conductivity of the solution was measured. Then, pure NaCl was added to the same volume of water. When the conductivity of the solution equaled the conductivity of the dissolved contaminants, the number of milligrams of pure NaCl at this point was recorded. This number was then divided by the area to obtain the equivalent salinity, expressed in mg / cm³. 2 The level of contamination was determined according to Table 1.

[0097] In step 5, the specific processing method is as follows:

[0098] S5.1: Image monitoring can acquire a large amount of intuitive image information. It is a mature technology, reliable data source, and highly economical. Furthermore, the processing effectively avoids interference from solar halos and corona discharges, making it widely used in monitoring scenarios such as converter stations and transmission lines. Images of iced insulators are taken periodically by staff to obtain comprehensive icing conditions of the insulators at different monitoring stages. The data source is simple and reliable. The image processing flow for iced insulators is as follows: Figure 3 As shown.

[0099] Before shooting, the structural characteristics and parameters of the support insulator were obtained as shown in Table 2, in order to calibrate the captured images. By using the object's length, area, and feature point positions, the relationship between image pixels and the object's actual physical size can be established, avoiding errors caused by viewing angle, lens distortion, etc., thereby improving the accuracy of image measurement and analysis.

[0100] Table 3 Insulator structural parameters

[0101]

[0102] When capturing images of icing on insulators from multiple angles, avoid obstruction and shaking during shooting. Adjust the appropriate focus and aperture, and take supplementary lighting measures if necessary to obtain clear, comprehensive images of the icing on the insulators. Furthermore, repeat the previous shooting positions at different times to avoid image processing errors.

[0103] S5.2: In low-temperature and high-humidity environments, rainwater or water mist accumulates and freezes on the surface of insulators in liquid form. Irregular ice-covered water films hinder the uniform distribution of water, and water droplets, influenced by wind and gravity, freeze as they move downwards, thus promoting the formation and growth of icicles. As the proportion of insulators bridged by ice increases, their leakage distance gradually shortens, their insulation performance gradually decreases, easily causing electric field distortion in the insulator string, reducing flashover voltage, and ultimately leading to ice flashover accidents. Therefore, when processing images of ice-covered insulators, it is crucial to extract image features such as ice length and bridging ratio.

[0104] Image preprocessing employs grayscale conversion and median filtering algorithms to enhance target features and reduce background noise, facilitating subsequent calibration and feature extraction. Compared to the RGB channels of a color image, a grayscale image contains only one grayscale channel, and image noise is converted into fluctuations in grayscale values. This significantly reduces the amount of data while preserving key features such as insulator edges, textures, and shapes, effectively improving computational speed.

[0105] To address the color characteristics of ice-covered insulators and the sensitivity of human visual perception, a weighted average method was used for grayscale processing, as shown in equation (3).

[0106] (3)

[0107] In the formula, M is the weighted gray value, and R, G, and B are the components of the three colors: red, green, and blue, respectively.

[0108] To separate the insulator image from its background image, the pixels of both are divided according to their gray values, and then image thresholding is achieved based on the difference in gray levels.

[0109] (4)

[0110] In the formula, f(x, y) is the original image, g(x, y) is the segmented image, and t1 and t2 are the grayscale ranges of the segmentation.

[0111] The Canny edge detection algorithm is used to extract the contour and feature edges of insulators. The calculation process includes image size and gradient direction calculation, non-maximum suppression, double threshold filtering, and edge connection.

[0112] (5)

[0113] (6)

[0114] (7)

[0115] In the formula, g x (x, y), g y (x, y) are the partial derivatives of the graph in the x and y directions, respectively, and g 45 (x, y), g 135 (x, y) are the partial derivatives of the image at 45° and 135°, respectively, e x e y Let F(x, y) be the first-order partial derivative of the image in the x and y directions, and let g be the magnitude of the gradient. d (x, y) represents the direction of the gradient.

[0116] S5.3: Using the image processing and edge detection methods described above, obtain the edge features of the un-iced insulator image in the same orientation. Compare the edges of the iced and un-iced insulator images; the difference in edges represents the ice thickness of the iced insulator. Overlay each pixel of the skirt edge in the un-iced insulator image onto the iced insulator image, using this as the starting point for ice growth (this pixel is the seed pixel). If a pixel identical to the seed pixel exists below or around the seed pixel in the iced insulator image, it is considered that the ice is growing towards that point, and this pixel is used as the new seed pixel. Repeat the above growth process, continuously searching for pixels in the iced insulator image, and combine this with the image calibration from step S5.1 to obtain the ice length of the iced insulator. Furthermore, the proportion of ice bridging the insulator gap can be calculated.

[0117] Repeat steps S5.1 and S5.2 above to calculate the ice length and bridging situation across the entire icing insulator, and use the maximum ice length to calculate the overall gap bridging ratio of the insulator. Taking a monitoring result as an example, monitoring was conducted every 10 minutes, and the ice lengths were 17.3mm, 25.5mm, 36.1mm, and 51.2mm, respectively. Therefore, the ice bridging ratios were 14.4%, 21.3%, 30.1%, and 42.7%, respectively.

[0118] In step 6, an SVM-based ice growth prediction model is established by combining the monitoring data of meteorological factors within the converter station. The model is then trained and optimized until the prediction accuracy of ice growth length is less than 5%, so that it can be used for dynamic prediction of subsequent ice growth.

[0119] The modeling steps for the ice crystal growth prediction model are as follows:

[0120] S6.1: The icing growth process on the insulator surface is strongly correlated with factors such as temperature, humidity, rainfall, wind speed, and wind direction, and the icing growth rate is non-linear. Lower temperatures are a necessary condition for the phase change freezing of liquid water or water mist. Within the range of -25 to 0℃, the lower the temperature, the higher the probability of the insulator capturing and adsorbing liquid water, and the faster the icing rate. Humidity characterizes the content of liquid water in the atmosphere; the higher the humidity, the more water droplets are captured on the insulator surface. In the event of rainfall, the larger the diameter of the raindrops, the greater the probability of them colliding with the insulator surface, and the greater the degree of icing. Wind speed and wind direction directly affect the orientation and morphology of icing, and under the action of convective heat transfer, they accelerate the freezing rate of liquid water.

[0121] Meteorological data is acquired from the converter station's meteorological monitoring device and the local meteorological platform. The meteorological monitoring device mainly consists of a main unit, temperature sensor, humidity sensor, wind speed sensor, wind direction sensor, rainfall sensor, and air pressure sensor. It dynamically monitors the environmental conditions of the converter station and transmits meteorological data to the management backend in real time via a wireless network. Real-time forecast data on temperature, humidity, and rainfall from the meteorological platform is also acquired to predict the growth of icing on the insulators.

[0122] S6.2: The SVM model demonstrates good performance in handling the multidimensional factors affecting insulator icing under complex environments. Considering factors such as ambient temperature, humidity, wind speed, rainfall, and raindrop diameter, an SVM-based model for predicting insulator icing growth is established.

[0123] The extracted characteristic parameters affecting ice growth are shown in Table 4.

[0124] Table 4 Feature Parameters

[0125]

[0126] Feature parameters directly impact the model's prediction results; too many irrelevant or weakly correlated feature vectors will reduce the prediction accuracy of SVM. Therefore, to improve model accuracy, the Pearson correlation coefficient method is used to select feature parameters, and the calculation formula is as follows:

[0127] (8)

[0128] In the formula, X i and Y i Let X and Y represent the i-th values ​​of a certain characteristic parameter X and the length Y of the ice floe, respectively, within a unit of time.

[0129] Table 5. Pearson correlation coefficients of feature parameters

[0130]

[0131] The feature parameters used to train the model are shown in Table 6 after feature selection.

[0132] Table 6 Feature parameters of the SVM model input

[0133]

[0134] Among them, the changes in temperature per unit time, humidity per unit time, wind speed per unit time, rainfall per unit time, and raindrop diameter per unit time are shown in equations (9)-(13).

[0135] (9)

[0136] (10)

[0137] (11)

[0138] (12)

[0139] (13)

[0140] In the formula, ∆t represents the unit time interval, taken as 10 minutes, and T i+1 T i Let R represent the temperature at time i and after time ∆t, respectively. i+1 R i V represents the humidity at time i and after time ∆t, respectively. i+1 V i Let J represent the wind speed at time i and after time ∆t, respectively. i+1 J i Y represents the rainfall at time i and after time ∆t, respectively. i+1 Y iLet represent the diameter of the raindrop at time i and after time ∆t, respectively.

[0141] To eliminate the influence of dimensions, each characteristic parameter is normalized.

[0142] (14)

[0143] In the formula, x i Let x be the feature parameters of a certain input model. min and x max These are the maximum and minimum values ​​of the parameter, x. i ' represents the normalized value.

[0144] S6.3: The process of predicting the length of insulator icicles per unit time using an SVM model is as follows: Figure 2 As shown. The acquired data on insulator icing growth includes insulator structural features, electrical features, environmental features, and the corresponding length of icing growth per unit time. A large amount of sample data is divided into training and testing sets. The training set is input into the initial SVM model to obtain the corresponding penalty coefficient and kernel function. Then, the test set data is input into the model, and the icing length in the test set data is compared with the length predicted by the SVM. When the prediction error is greater than 5%, a modified GS algorithm is used to optimize the parameters of the support vector machine model. This process continues until the error meets the expected requirements, at which point the final optimized SVM model is obtained.

[0145] Taking the icicle growth on the topmost skirt of an insulator as an example. At the initial monitoring time, the temperature was -12℃, humidity was 92%, wind speed was 3.3m / s, rainfall was 5mm, raindrop diameter was approximately 2.2mm, and icicle length was 17.3mm. The voltage on the insulator was 500kV, and leakage current was negligible. Before and after the same time point, meteorological factors affecting the insulator environment were monitored five times, and the average values ​​are shown in Table 5. The icicle length at each time point was also recorded. A series of data were input into an SVM model to train and optimize the model. The final SVM model had a penalty coefficient C of 56.42 and a kernel function γ of 0.0183. The mean absolute percentage error (MAE) of the test set was 3.1%.

[0146] Table 7 shows the ice formation process on insulators during a certain monitoring session.

[0147]

[0148] S6.4: Before using an SVM model trained on historical data for actual prediction, the current environmental and parameter characteristics can be obtained for targeted model correction. For example, in a prediction using the model, the initial conditions were: ice length 12.2 mm, ambient temperature -6℃, humidity 93%, wind speed 2.7 m / s, rainfall 7 mm, and raindrop diameter 2.4 mm. Substituting five sets of data obtained simultaneously into the model, the GS algorithm was used to optimize the model again, resulting in a new SVM model with a penalty coefficient C of 54.27 and a kernel function γ of 0.0181. Future data obtained from the meteorological platform showed an ambient temperature of -5℃, humidity 92%, wind speed 2.9 m / s, rainfall 6 mm, and raindrop diameter 2.2 mm. Using the unoptimized model, the predicted ice length was 15.4 mm; using the specifically optimized model, the predicted ice length was 15.7 mm.

[0149] Ten minutes later, the icing condition of the insulator was photographed, and the length of the icicles was calculated to be 16.1 mm after image processing. It can be seen that the prediction error of the unoptimized model was 4.3%, which met the requirements; however, the prediction error of the optimized model was 2.5%, and the prediction result was closer to the actual situation.

[0150] Therefore, by substituting current meteorological data into the SVM ice crystal model for targeted optimization, a model with higher prediction accuracy can be obtained. Then, based on current meteorological data such as temperature, humidity, and wind speed, combined with meteorological data predicted by the meteorological platform, the meteorological changes after a unit of time can be obtained. Further prediction can then be made to obtain the ice crystal growth length and gap bridging ratio at the next moment.

[0151] In step 7, when an icing disaster occurs at the 500kV DC converter station, the ice growth length is predicted according to step 6. Based on the measured pollution level and the predicted ice bridging distance, combined with the ice flashover voltage fitting model of the post insulators in step 3, the ice flashover voltage of the post insulators is predicted. If the ice flashover voltage value is higher than the rated voltage, normal operation continues; if it is lower than the corresponding voltage, voltage reduction operation is implemented. Dynamic prediction is performed at regular intervals to determine the operating voltage value.

[0152] For example, when the pollution level of the insulator is measured to be 0.08 mg / cm², during monitoring, when the ice bridging ratio reaches 50%, the breakdown voltage is 542.1 kV, which is within the safe range. However, when predicting the ice length for the next stage, with the ice bridging ratio reaching 75%, the breakdown voltage obtained through the fitting model is 476.3 kV, exceeding the safe range. Therefore, to ensure the safe and stable operation of the power grid, the converter station needs to step down the voltage to 350 kV at this point.

[0153] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A method for predicting and controlling the withstand voltage of post insulators for anti-icing skirts in converter stations, characterized in that, Includes the following steps: Step 1: Using test samples with the same structure as the post insulators with intermittent anti-icing umbrella skirts installed in the DC converter station, the post insulator test samples were coated with dirt according to different pollution levels. Then, icing was simulated in the environmental climate laboratory to obtain post insulator test samples under typical ice bridging conditions. Step 2: Using the ice-melting flashover test method, conduct ice flashover tests on post insulator specimens under typical pollution conditions and ice bridging conditions. Perform M tests for each working condition, record and calculate the average flashover voltage; Step 3: Combining the pollution level and ice bridging situation, obtain the mapping relationship between the two main influencing factors and the ice flashover voltage, and establish a fitting model for the ice flashover voltage of the post insulator with intermittent anti-icing umbrella skirt. Step 4: Measure the equivalent salt density of the standby insulators located next to the operating post insulators in the DC converter station, which are in the same environment, to determine the pollution level; Step 5: From the start of icing, TPS will take all-around images of the icing on the support insulators at regular intervals through maintenance personnel or monitoring devices to obtain information on the icing bridging gaps. Step 6: Combine the monitoring data of meteorological factors in the converter station to establish an SVM-based ice growth prediction model. Train and optimize the model using insulator icing images at intervals of Tps until the prediction accuracy of ice growth length is ≥η%, which will be used for subsequent dynamic prediction of ice growth. Step 7: When an icing disaster occurs at the converter station, the length of ice growth is predicted according to Step 6. Based on the measured degree of pollution and the predicted ice bridging distance, combined with the ice flashover voltage fitting model of the post insulator in Step 3, the ice flashover voltage of the post insulator is predicted. If the ice flashover voltage value is higher than the rated voltage, normal operation continues. If it is lower than the corresponding voltage, the voltage is reduced. Dynamic prediction is performed at regular intervals to determine the operating voltage value.

2. The method for predicting and controlling the withstand voltage of the post insulator for the anti-icing umbrella skirt in a converter station according to claim 1, characterized in that, In step 1, N pollution levels are defined according to a preset pollution level. Pollution is simulated using equivalent salt density and equivalent ash density, with each pollution level taking the median value. Soluble substances are simulated using pure NaCl, and insoluble substances are simulated using diatomaceous earth. They are mixed in a ratio of 1:Xh to form simulated pollution substances. The pollution is then manually and evenly applied to the cleaned insulator surface using the solid layer method and placed in a dry and ventilated place for Tgz time before use.

3. The method for predicting and controlling the withstand voltage of the post insulator for the anti-icing umbrella skirt in a converter station according to claim 2, characterized in that, The icing method employs live icing at a rated voltage of kN%. Before spraying icing, the insulator is pretreated by spraying water onto its surface to form a water film, which then forms an ice film to prevent contamination from escaping. Cooling water is then sprayed onto the post insulator from nozzles on both sides, gradually forming an ice layer under the low-temperature conditions of an environmental climate laboratory. Icing is stopped when the ice layer bridges N different gap distances, resulting in N test samples for each degree of contamination.

4. The method for predicting and controlling the withstand voltage of the post insulator for the anti-icing umbrella skirt in a converter station according to claim 3, characterized in that, In step 2, considering the presence of a water film during actual ice melting, which makes flashover more likely, the ice flash test considers a more stringent situation. That is, the ice melting flashover test method is adopted. After the ice covers the ice to the corresponding ice length, the temperature in the climate environment room is controlled to rise to a certain temperature T1 close to 0 degrees Celsius. Then, the temperature is increased at a rate of St°C / h. When there is a water film on the ice surface and water droplets are left on the ice, the conditions for starting the ice flash test are met, and pressurization is prepared to begin.

5. The method for predicting and controlling the withstand voltage of the post insulator for the anti-icing umbrella skirt in a converter station according to claim 4, characterized in that, The following pressurization method was adopted: During the first pressurization, the voltage was increased at a rate of UtkV / s until flashover occurred, and the flashover voltage Utc was recorded at this time. In subsequent tests, the voltage was first rapidly increased to ktc% of Utc, and then increased by kjy% of Utc each time, while maintaining the voltage withstand time Tns. After Tns, the voltage was increased again until breakdown occurred. The ice flashover voltage obtained in this way is closer to the actual situation.

6. The method for predicting and controlling the withstand voltage of the post insulator for the anti-icing umbrella skirt in a converter station according to claim 5, characterized in that, In step 3, by combining the pollution level and the ice bridging situation, the mapping relationship between the two main influencing factors and the ice flashover voltage is obtained, and a fitting model of the ice flashover voltage of the post insulator with intermittent anti-icing umbrella skirt is established. To ensure the effectiveness of the experiment and the reliability of the data, the pollution levels of the insulators were assigned values ​​of ESDD1, ESDD2, ESDD3, ESDD4, and ESDD5, respectively. The length of the insulator icicles was represented by the bridging ratio between the large umbrella skirts, which were p1, p2, p3, p4, and p5, respectively. The pollution level and icicle bridging ratio of each insulator were combined in pairs, and three repeated breakdown tests were carried out. Abnormal situations were excluded, and the average breakdown voltage was used to fit the model. （1）； In the formula, p is the ice bridging ratio, l is the length of the insulator ice, and d is the spacing between the insulator skirts; The flashover voltage of an insulator decreases with increasing pollution level, and the flashover voltage of an insulator decreases with increasing ice length. By iterating the function multiple times using the least squares approximation method, a surface fitting mathematical model is constructed. This model can then be used to calculate and predict the breakdown voltage corresponding to the pollution level and bridging ratio, excluding the characteristic parameters. The fit degree of the surface is R², and the expression of the surface is shown in equation (2). （2）； In the formula, U is the breakdown voltage under different salt density and ice bridging ratios, x is the ice bridging ratio, and y is the insulator salt density.

7. The method for predicting and controlling the withstand voltage of the post insulator for the anti-icing umbrella skirt in a converter station according to claim 6, characterized in that, In step 4, the normally operating post insulators cannot be removed for testing. Several post insulators specifically for testing are placed next to the normally operating insulators under the same natural environment and with the same pollution conditions. The pollution in the area of ​​Sjc on the corresponding insulator is collected and dissolved in a volume of Vjc of distilled water. The conductivity of the solution is measured. Then, pure NaCl is added to the same volume of water. When the conductivity of the solution is the same as the conductivity of the dissolved pollution, the number of milligrams of pure NaCl at this time is recorded. The equivalent salt density is obtained by dividing by the area, with the unit being mg / cm2, and the pollution level is determined.

8. The method for predicting and controlling the withstand voltage of the post insulator for the anti-icing umbrella skirt in a converter station according to claim 7, characterized in that, In step 5, the specific processing method is as follows: S5.1: Images of iced insulators are taken by staff at regular intervals to obtain comprehensive icing conditions of the insulators at different monitoring stages. Before taking the images, the structural characteristics and parameters of the insulator are obtained to calibrate the images. By using the object's length, area, and feature point positions, the relationship between image pixels and the object's actual physical dimensions is established to avoid errors caused by viewing angle and lens distortion, thereby improving the accuracy of image measurement and analysis. When taking multi-angle images of iced insulators, obstruction and shaking should be avoided during shooting. Appropriate focus and aperture should be adjusted to obtain comprehensive and clear images of the iced insulators. In addition, the shooting positions should be repeated at different times to avoid image processing errors. S5.2: When processing images of ice-covered insulators, the focus is on extracting image features of ice length and bridging ratio. Grayscale and median filtering algorithms are used to preprocess the images to enhance target features and reduce background noise, so as to facilitate subsequent calibration and feature extraction. For the color features of ice-covered insulators and the sensitivity of human eye perception, a weighted average method is used for grayscale processing, as shown in Equation (3): （3）； In the formula, M is the weighted gray value, and R, G, and B are the components of the three colors: red, green, and blue, respectively. To separate the insulator image from its background image, the pixels of both are divided according to their gray values, and then image thresholding is achieved based on the difference in gray levels. （4）； In the formula, f(x, y) is the original image, g(x, y) is the segmented image, and t1 and t2 are the grayscale ranges of the segmentation; The Canny edge detection algorithm is used to extract the contour and feature edges of insulators. The calculation process includes image size and gradient direction calculation, non-maximum suppression, double threshold filtering, and edge connection. （5）； （6）； （7）； In the formula, gx(x,y) and gy(x,y) are the partial derivative functions of the image in the x and y directions, respectively; g45(x,y) and g135(x,y) are the partial derivative functions of the image in the 45° and 135° directions, respectively; ex and ey are the first-order partial derivatives of the image in the x and y directions, respectively; F(x,y) is the magnitude of the gradient; and gd(x,y) is the direction of the gradient. S5.3: Using the image processing and edge detection methods described above, the edge features of the un-iced insulator image in the same orientation are obtained. The edges of the iced and un-iced insulator images are compared, and the edge difference is the ice thickness of the iced insulator. Each pixel of the skirt edge in the un-iced insulator image is covered onto the iced insulator image, which serves as the starting point for ice growth. This pixel is the seed pixel. If there is a pixel with the same name below or around the seed pixel in the iced insulator image, it is considered that the ice is growing towards that point, and this pixel is used as the new seed pixel. The above growth process is repeated to continuously find the pixels in the iced insulator image. Combined with the image calibration in step S5.1, the ice length of the iced insulator is obtained. Then, the ratio of ice bridging the gap between the insulators is calculated. Repeat steps S5.1 and S5.2 above to calculate the length of ice floes and bridging conditions across the entire ice-covered insulator, and take the maximum ice floe length to calculate the gap bridging ratio of the entire insulator.

9. A method for predicting and controlling the withstand voltage of post insulators for anti-icing skirts in converter stations according to claim 8, characterized in that, In step 6, an SVM-based ice growth prediction model is established by combining the monitoring data of meteorological factors in the converter station. The model is then trained and optimized until the prediction accuracy of ice growth length is less than a certain threshold, so that it can be used for dynamic prediction of subsequent ice growth. The modeling steps for the ice crystal growth prediction model are as follows: S6.1: The icing growth process on the insulator surface is strongly correlated with temperature, humidity, rainfall, wind speed, and wind direction. Furthermore, the icing growth rate is non-linear. Within the range of -25 to 0℃, the lower the temperature, the higher the probability of the insulator capturing and adsorbing liquid water, resulting in a faster icing rate. Humidity characterizes the content of liquid water in the atmosphere; higher humidity leads to more water droplets being captured on the insulator surface. In rainfall conditions, larger raindrop diameters increase the probability of colliding with the insulator surface, thus increasing the degree of icing. Wind speed and direction directly affect the orientation and shape of the icing, and convective heat transfer accelerates the freezing rate of liquid water. Meteorological data is acquired from the converter station's meteorological monitoring device and the local meteorological platform. The meteorological monitoring device includes a main unit, temperature sensor, humidity sensor, wind speed sensor, wind direction sensor, rainfall sensor, and air pressure sensor. It dynamically monitors the environmental conditions of the converter station and transmits meteorological data to the management backend in real time via a wireless network. Real-time temperature, humidity, and rainfall forecast data from the meteorological platform are also obtained to predict the icing growth status of the insulators. S6.2: The SVM model has good performance in dealing with the multidimensional factors affecting insulator icing in complex environments. Considering factors such as ambient temperature, humidity, wind speed, rainfall, and raindrop diameter, an insulator icing growth prediction model based on the SVM model is established. The extracted feature parameters that influence ice growth directly affect the model's prediction results. Too many uncorrelated or weakly correlated feature vectors will reduce the SVM prediction accuracy. Therefore, to improve the model's accuracy, the Pearson correlation coefficient method is used to select feature parameters. The calculation formula is as follows: （8）； In the formula, Xi and Yi represent the i-th values ​​of a certain characteristic parameter X and the ice length Y within a unit time. The features are selected for training the model. The temperature change, humidity change, wind speed change, rainfall change, and raindrop diameter change per unit time are shown in equations (9)-(13). （9）； （10）； （11）； （12）； （13）； In the formula, ∆t represents the unit time interval, which is taken as 10 minutes; Ti+1 and Ti represent the temperature at time i and after time ∆t, respectively; Ri+1 and Ri represent the humidity at time i and after time ∆t, respectively; Vi+1 and Vi represent the wind speed at time i and after time ∆t, respectively; Ji+1 and Ji represent the rainfall at time i and after time ∆t, respectively; and Yi+1 and Yi represent the raindrop diameter at time i and after time ∆t, respectively. To eliminate the influence of dimensions, each characteristic parameter is normalized. （14）； In the formula, xi is a feature parameter of a certain input model, xmin and xmax are the maximum and minimum values ​​of the parameter, respectively, and xi' is the normalized value; S6.3: The SVM model is used to predict the length of insulator icicles per unit time. The obtained data on insulator icicle growth includes insulator structural features, electrical features, environmental features, and the corresponding length of icicle growth per unit time. A large amount of sample data is divided into training set and test set. The training set is input into the initial SVM model to obtain the corresponding penalty coefficient and kernel function. Then, the test set data is input into the model, and the length of icicles in the test set data is compared with the length predicted by the SVM. When the prediction error is greater than k%, the improved GS algorithm is used to optimize the parameters of the support vector machine model until the error meets the expected requirements, and the final optimized SVM model is obtained. The final SVM model has a penalty coefficient of C, a kernel function of γ, and a mean absolute percentage error of k1% on the test set. Based on the current temperature, humidity, and wind speed meteorological data, combined with the meteorological data predicted by the meteorological platform, the meteorological changes after a unit of time are obtained. Then, the meteorological change data is substituted into the SVM ice prediction model to obtain the ice growth length at the next moment. Then, the ice bridging ratio is calculated based on the ice length and the insulator gap distance. S6.4: Before making actual predictions using the SVM model trained with historical data, the current environmental and parameter characteristics are obtained for targeted model correction. First, the initial meteorological data is obtained and substituted into the model. The GS algorithm is used to optimize the model again to obtain a new SVM prediction model. Then, based on the current temperature, humidity, and wind speed meteorological data, combined with the meteorological data predicted by the meteorological platform, the meteorological changes after a unit of time are obtained, and then predictions are made to obtain the ice growth length and gap bridging ratio at the next moment.

Images