A method for evaluating the insulation performance of a post-washing live device and related equipment

Abstract

The embodiment of the application discloses a kind of after washing charged equipment insulation performance evaluation method and related equipment, by obtaining the image data of charged equipment surface after washing, and using spectrum analyzer to collect the spectral characteristic signal of multiple pollutants, can accurately, comprehensively capture the pollutant information of equipment surface.The spectral characteristic signal of multiple pollutants is analyzed, the concentration distribution data of pollutant type can be obtained, to simultaneously identify multiple residues and understand its distribution state.When judging that it contains salt and oil stain mixture, the residual concentration mean value and concentration fluctuation coefficient of salt and oil stain are obtained, and the accurate insulation performance evaluation result can be obtained by combining the residual concentration mean value and concentration fluctuation coefficient of salt and oil stain, avoid the inaccurate insulation performance evaluation caused by ignoring part of pollutants in traditional method, effectively reduce the flashover risk in equipment operation, guarantee the safe and stable operation of charged equipment.

Description

Technical Field

[0001] This invention relates to the field of intelligent maintenance and testing technology for power equipment, and in particular to a method and related equipment for evaluating the insulation performance of live equipment after rinsing. Background Technology

[0002] The safe operation of live equipment is crucial to the stability of the power system, and its surface cleanliness directly affects insulation performance and equipment lifespan. During operation, live equipment can become contaminated due to environmental pollution, dust accumulation, or chemical corrosion. This contamination can lead to electric arcs, flashovers, and even equipment failure. Therefore, regularly flushing live equipment and ensuring its surface cleanliness is a key aspect of ensuring the reliable operation of the power system. Currently, post-flushing inspection and evaluation have become a key research area because they not only concern equipment safety but also directly impact the overall stability of the power system.

[0003] Existing detection methods mainly rely on manual inspection or simple chemical testing, which have significant limitations. Manual inspection is limited by the operator's experience, making it difficult to guarantee consistency and accuracy, especially in complex environments where trace amounts of residual contaminants are difficult to distinguish with the naked eye. Traditional chemical testing equipment can usually only perform single-item analysis of specific contaminants, lacking the ability to comprehensively assess multiple residues and their distribution. For example, after flushing a high-voltage transformer, trace amounts of salt and oil mixture may remain on the surface. Traditional detection methods may only identify the salt while ignoring the oil, leading to inaccurate insulation performance assessments and leaving the equipment at risk of flashover during operation. Summary of the Invention

[0004] In view of this, the present invention provides a method and related equipment for evaluating the insulation performance of energized equipment after rinsing.

[0005] The specific technical solution of the first embodiment of the present invention is as follows: a method for evaluating the insulation performance of energized equipment after rinsing, the method comprising: acquiring image data of the surface of the energized equipment after rinsing; collecting spectral characteristic signals of multiple pollutants in the image data using a spectral analyzer; performing pollutant analysis on the spectral characteristic signals of the multiple pollutants to obtain pollutant types and pollutant concentration distribution data in the image data; determining whether the pollutant types contain a mixture of salt and oil; if the mixture of salt and oil is included, acquiring the first residual concentration average and first concentration fluctuation coefficient of salt in the pollutant concentration distribution data, and acquiring the second residual concentration average and second concentration fluctuation coefficient of oil; and obtaining the insulation performance evaluation result of the surface of the energized equipment based on the first residual concentration average, the first concentration fluctuation coefficient, the second residual concentration average, and the second concentration fluctuation coefficient.

[0006] Preferably, obtaining the insulation performance evaluation result of the surface of the live equipment based on the first average residual concentration, the first concentration fluctuation coefficient, the second average residual concentration, and the second concentration fluctuation coefficient includes: obtaining simulated distribution data of the salt and oil mixture on the surface of the live equipment based on the first average residual concentration, the first concentration fluctuation coefficient, the second average residual concentration, and the second concentration fluctuation coefficient; the simulated distribution data includes electric field strength distribution and surface current density; comparing the simulated distribution data with insulation performance benchmark values ​​to obtain distribution data deviation values; the distribution data deviation values ​​include electric field strength distribution deviation values ​​and surface current density deviation values, and the insulation performance benchmark values ​​include standard electric field strength distribution and standard surface current density when there is no salt and oil mixture; inputting the distribution data deviation values, the first concentration fluctuation coefficient, and the second concentration fluctuation coefficient into a preset insulation analysis model to obtain the insulation risk index of the surface of the live equipment after rinsing; and obtaining a first insulation performance evaluation score of the surface of the live equipment based on the insulation risk index and a first preset insulation performance evaluation rule, wherein the first insulation performance evaluation score is used to characterize the insulation performance evaluation result of the surface of the live equipment.

[0007] Preferably, the first concentration fluctuation coefficient and the second concentration fluctuation coefficient are obtained by the following method: dividing the image data into several cells; based on the pollutant concentration distribution data, obtaining the salt residue concentration value and oil residue concentration value of each cell; according to the salt residue concentration value and oil residue concentration value of all cells, obtaining the salt residue standard deviation, oil residue standard deviation, salt residue mean concentration, and oil residue mean concentration; obtaining a first ratio of the salt residue standard deviation to the salt residue mean concentration, the first ratio being the first concentration fluctuation coefficient; obtaining a second ratio of the oil residue standard deviation to the oil residue mean concentration, the second ratio being the second concentration fluctuation coefficient.

[0008] Preferably, determining whether the pollutant type contains a mixture of salt and oil includes: determining whether the spectral characteristic signal of the pollutant contains a first characteristic peak of salt and a second characteristic peak of oil; if it contains the first characteristic peak and the second characteristic peak, then the pollutant type contains a mixture of salt and oil.

[0009] Preferably, after obtaining the first insulation performance evaluation score of the surface of the energized equipment, the method further includes: determining whether the first insulation performance evaluation score is less than a first preset threshold; if it is less than the first preset threshold, then extracting the chemical distribution features other than salt and oil from the spectral feature signals of the multiple pollutants using a feature extraction algorithm; classifying the chemical distribution features using a support vector machine algorithm to obtain the types of chemical substances other than salt and oil in the spectral feature signals of the multiple pollutants; obtaining the pollutant concentration, distribution density, and distribution uniformity of the multiple pollutants in the pollutant concentration distribution data; the multiple pollutants include the types of chemical substances, the salt, and the oil; and obtaining the second insulation performance evaluation score of the surface of the energized equipment based on the pollutant concentration, the distribution density, and the distribution uniformity.

[0010] Preferably, after obtaining the pollutant concentration, distribution density, and distribution uniformity of multiple pollutants in the pollutant concentration distribution data, the method further includes: obtaining the interference intensity of multiple pollutants according to a preset interference level classification rule, and obtaining the interference frequency of multiple pollutants in the pollutant concentration distribution data; then, obtaining the second insulation performance evaluation score of the surface of the energized equipment based on the pollutant concentration, the distribution density, and the distribution uniformity includes: obtaining the second insulation performance evaluation score of the surface of the energized equipment based on the pollutant concentration, the distribution density, the distribution uniformity, the interference intensity, and the interference frequency of the multiple pollutants.

[0011] Preferably, obtaining the second insulation performance evaluation score of the surface of the electrical equipment based on the pollutant concentration, distribution density, distribution uniformity, interference intensity, and interference frequency of the multiple pollutants includes: integrating the pollutant types, pollutant concentrations, distribution densities, distribution uniformity, interference intensity, and interference frequency to obtain a preliminary feature vector; obtaining an environmental feature vector of the electrical equipment, the environmental feature vector including vectors composed of the ambient humidity, ambient temperature, image data sampling time, and sampling location of the electrical equipment; performing a weighted calculation on the preliminary feature vector and the environmental feature vector to obtain a weighted feature matrix; and obtaining the second insulation performance evaluation score of the surface of the electrical equipment based on the value of the weighted feature matrix and a second preset insulation performance evaluation rule.

[0012] The specific technical solution of the second embodiment of the present invention is as follows: a system for evaluating the insulation performance of energized equipment after rinsing, the system comprising: a feature signal acquisition module, a contaminant analysis module, a judgment module, a numerical acquisition module, and an evaluation module; the feature signal acquisition module is used to acquire image data of the surface of the energized equipment after rinsing, and to collect spectral feature signals of multiple contaminants in the image data through a spectral analyzer; the contaminant analysis module is used to perform contaminant analysis on the spectral feature signals of the multiple contaminants to obtain contaminant types and contaminant concentration distribution data in the image data; the judgment module is used to determine whether the contaminant types include a mixture of salt and oil; the numerical acquisition module is used to, if the mixture of salt and oil is included, acquire the first residual concentration average and the first concentration fluctuation coefficient of salt in the contaminant concentration distribution data, and acquire the second residual concentration average and the second concentration fluctuation coefficient of oil; the evaluation module is used to obtain the insulation performance evaluation result of the surface of the energized equipment based on the first residual concentration average, the first concentration fluctuation coefficient, the second residual concentration average, and the second concentration fluctuation coefficient.

[0013] The specific technical solution of the third embodiment of the present invention is as follows: an insulation performance evaluation device for energized equipment after rinsing, comprising a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of the method as described in any one of the first embodiments of this application.

[0014] The specific technical solution of the fourth embodiment of the present invention is as follows: a computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the processor performs the steps of the method as described in any one of the first embodiments of this application.

[0015] Implementing the embodiments of the present invention will have the following beneficial effects: This invention acquires image data of the surface of the equipment after rinsing and uses a spectrometer to collect spectral characteristic signals of various pollutants. It can accurately and comprehensively capture pollutant information on the surface of the equipment, without being affected by human experience or complex environments, thus ensuring the consistency and accuracy of the detection.

[0016] Analyzing the spectral characteristics of multiple pollutants allows for the acquisition of pollutant types and concentration distribution data, enabling the simultaneous identification of multiple residues and understanding their distribution patterns. When a mixture of salt and oil is identified, the average residual concentration and concentration fluctuation coefficient of both are obtained. The average residual concentration directly reflects the average content of salt and oil on the equipment surface; higher concentrations indicate more severe contamination and a greater negative impact on insulation performance. The concentration fluctuation coefficient reflects the uneven distribution of salt and oil on the equipment surface. A larger fluctuation coefficient indicates higher concentrations of pollutants in certain areas, potentially leading to severe localized contamination. These areas of high localized contamination are often weak points in insulation performance, making them more susceptible to flashover and other faults. Therefore, combining the average residual concentration and concentration fluctuation coefficient of salt and oil provides accurate insulation performance assessment results, avoiding the inaccuracies caused by neglecting some pollutants in traditional methods. This effectively reduces the risk of flashover during equipment operation and ensures the safe and stable operation of live equipment. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A flowchart illustrating the steps of a method for evaluating the insulation performance of live equipment after rinsing; Figure 2 A schematic diagram of the structure of a first embodiment of a system for evaluating the insulation performance of live equipment after rinsing; Figure 3 A schematic diagram of a second embodiment of a system for evaluating the insulation performance of live equipment after rinsing; The modules are as follows: 201. Feature signal acquisition module; 202. Pollutant analysis module; 203. Judgment module; 204. Numerical acquisition module; 205. Evaluation module; 301. Image acquisition module; 302. Image processing module; 303. Pollutant fusion module; 304. Insulation evaluation module; 305. Interference identification module; 306. Comprehensive scoring module; 307. Risk analysis module; 308. Risk prediction module. Detailed Implementation

[0019] 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 embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0020] The terms "first," "second," etc., used in the specification, claims, and drawings of this application are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or modules is not limited to the listed steps or modules, but may optionally include steps or modules not listed, or may optionally include other steps or modules inherent to such processes, methods, products, or apparatus.

[0021] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0022] Please see Figure 1 This is a flowchart illustrating the steps of a method for evaluating the insulation performance of energized equipment after rinsing, as described in the first embodiment of this application, to improve the accuracy and reliability of the insulation performance evaluation results. The method includes: Step 101: Obtain image data of the surface of the energized equipment after rinsing, and collect spectral characteristic signals of various pollutants in the image data using a spectral analyzer; Step 102: Perform pollutant analysis on the spectral feature signals of the multiple pollutants to obtain pollutant type and pollutant concentration distribution data in the image data; Step 103: Determine whether the contaminant type contains a mixture of salt and oil. Step 104: If the mixture of salt and oil is included, obtain the first residual concentration average and the first concentration fluctuation coefficient of salt in the pollutant concentration distribution data, and obtain the second residual concentration average and the second concentration fluctuation coefficient of oil. Step 105: Obtain the insulation performance evaluation result of the surface of the energized equipment based on the first average residual concentration, the first concentration fluctuation coefficient, the second average residual concentration, and the second concentration fluctuation coefficient.

[0023] Specifically, a high-definition industrial camera mounted on a drone is used to acquire images of the washed surface of the electrical equipment, ensuring clear and unobstructed images to accurately reflect the surface condition. The acquired image data is then imported into a spectrometer, which can identify and collect spectral characteristic signals of various contaminants in the images, including but not limited to common contaminants such as salt and oil. After processing, the spectrometer outputs data on contaminant type and concentration distribution. It is then determined whether the contaminant type includes a mixture of salt and oil. If such a mixture is detected, the following parameters are extracted: the first residual concentration average and its first concentration fluctuation coefficient for salt, and the second residual concentration average and its second concentration fluctuation coefficient for oil. These parameters reflect the average residual degree and uniformity of salt and oil distribution on the equipment surface. Based on these parameters, a pre-set algorithm model is used for comprehensive analysis. This model considers the combined effects of salt and oil on insulation performance. Combining the first residual concentration average, the first concentration fluctuation coefficient, the second residual concentration average, and the second concentration fluctuation coefficient, the insulation performance evaluation result of the electrical equipment surface is calculated. The evaluation result is presented in the form of quantitative indicators, intuitively reflecting the insulation status of the equipment surface.

[0024] This embodiment acquires image data of the surface of energized equipment after rinsing and uses a spectrometer to collect spectral characteristic signals of various contaminants. This allows for accurate and comprehensive capture of contaminant information on the equipment surface, unaffected by human experience or complex environments, ensuring consistent and accurate detection. Convolutional Neural Network (CNN) feature extraction and classification are performed on the acquired images to identify the type and distribution of contaminants. Analyzing the spectral characteristic signals of various contaminants yields data on contaminant type and concentration distribution, enabling simultaneous identification of multiple residues and understanding their distribution. When a mixture of salt and oil is identified, the average residual concentration and concentration fluctuation coefficient of salt and oil are obtained. The average residual concentration directly reflects the average content of salt and oil on the equipment surface; higher concentrations indicate more severe contamination and a greater negative impact on insulation performance. The concentration fluctuation coefficient reflects the uneven distribution of salt and oil on the equipment surface. A large fluctuation coefficient indicates that the concentration of pollutants is high in some areas, which may lead to severe local pollution. These areas of high local pollution are often weak points in insulation performance and are more prone to flashover and other faults. Therefore, combining the average residual concentration of salt and oil with the concentration fluctuation coefficient can obtain accurate insulation performance assessment results, avoiding the inaccurate insulation performance assessment caused by neglecting some pollutants in traditional methods. This effectively reduces the risk of flashover during equipment operation and ensures the safe and stable operation of live equipment.

[0025] In a specific embodiment, obtaining the insulation performance evaluation result of the surface of the live equipment based on the first average residual concentration, the first concentration fluctuation coefficient, the second average residual concentration, and the second concentration fluctuation coefficient includes: obtaining simulated distribution data of the salt and oil mixture on the surface of the live equipment based on the first average residual concentration, the first concentration fluctuation coefficient, the second average residual concentration, and the second concentration fluctuation coefficient; the simulated distribution data includes electric field strength distribution and surface current density; comparing the simulated distribution data with insulation performance benchmark values ​​to obtain distribution data deviation values; the distribution data deviation values ​​include electric field strength distribution deviation values ​​and surface current density deviation values, and the insulation performance benchmark values ​​include standard electric field strength distribution and standard surface current density without salt and oil mixture; inputting the distribution data deviation values, the first concentration fluctuation coefficient, and the second concentration fluctuation coefficient into a preset insulation analysis model to obtain the insulation risk index of the surface of the live equipment after rinsing; and obtaining a first insulation performance evaluation score of the surface of the live equipment based on the insulation risk index and a first preset insulation performance evaluation rule, wherein the first insulation performance evaluation score is used to characterize the insulation performance evaluation result of the surface of the live equipment.

[0026] Specifically, based on the average first residual concentration and first concentration fluctuation coefficient of salt, and the average second residual concentration and second concentration fluctuation coefficient of oil, professional simulation software is used to generate simulated distribution data of the salt and oil mixture on the surface of the equipment, covering electric field intensity distribution and surface current density. The simulated electric field intensity distribution and surface current density are compared with pre-set insulation performance benchmark values, which are the standard electric field intensity distribution and standard surface current density without salt and oil mixture, thus obtaining the electric field intensity distribution deviation value and surface current density deviation value. The distribution data deviation value, the first concentration fluctuation coefficient, and the second concentration fluctuation coefficient are input into a preset insulation analysis model. This model is constructed based on a large amount of experimental data and theoretical analysis, and calculates the insulation risk index of the equipment surface after washing. Based on the insulation risk index and the first preset insulation performance evaluation rules, such as setting different risk index ranges corresponding to different score intervals, the first insulation performance evaluation score of the equipment surface is obtained, thus characterizing the equipment surface insulation performance evaluation result.

[0027] By comprehensively considering multiple parameters to generate simulated distribution data and comparing it in detail with benchmark values, the system can comprehensively and accurately reflect the impact of salt and oil mixtures on insulation performance, thus improving the accuracy of the assessment. It outputs an insulation risk index and assessment score, presenting insulation performance in a quantitative form, allowing maintenance personnel to intuitively understand the insulation status of the equipment. In a specific embodiment, the first concentration fluctuation coefficient and the second concentration fluctuation coefficient are obtained by the following method: dividing the image data into several cells; based on the pollutant concentration distribution data, obtaining the salt residue concentration value and oil residue concentration value of each cell; according to the salt residue concentration value and oil residue concentration value of all cells, obtaining the salt residue standard deviation, oil residue standard deviation, salt residue mean concentration, and oil residue mean concentration; obtaining a first ratio of the salt residue standard deviation to the salt residue mean concentration, the first ratio being the first concentration fluctuation coefficient; obtaining a second ratio of the oil residue standard deviation to the oil residue mean concentration, the second ratio being the second concentration fluctuation coefficient.

[0028] Specifically, to obtain the first and second concentration fluctuation coefficients, the acquired image data is divided into several uniformly sized cells. The number of cells is determined based on the size of the equipment surface and the required detection accuracy, for example, 100 cells. Based on the pollutant concentration distribution data, the residual salt concentration and residual oil concentration values ​​for each cell are obtained and recorded. The standard deviation of the residual salt concentration is calculated using the standard deviation formula based on the recorded residual salt concentration values ​​in all cells. Simultaneously, the average residual salt concentration values ​​in all cells are calculated to obtain the mean residual salt concentration. Similarly, the standard deviation and mean residual oil concentration are calculated based on the residual oil concentration values ​​in all cells. The first ratio of the residual salt standard deviation to the mean residual salt concentration is calculated; this first ratio is the first concentration fluctuation coefficient. The second ratio of the residual oil standard deviation to the mean residual oil concentration is calculated; this second ratio is the second concentration fluctuation coefficient. Based on the first concentration fluctuation coefficient, the second concentration fluctuation coefficient, and other relevant parameters, simulated distribution data is generated according to the aforementioned method, compared with the insulation performance benchmark value, input into the preset insulation analysis model, and finally the insulation performance evaluation result of the surface of the live equipment is obtained.

[0029] By dividing image data into cells and calculating the residual standard deviation and mean concentration of salt and oil respectively, a concentration fluctuation coefficient is obtained. This quantitative method can more accurately reflect the uneven distribution of contaminants on the equipment surface, providing more accurate data support for insulation performance evaluation. By comprehensively considering the concentration fluctuations of the two main contaminants, salt and oil, their impact on insulation performance is fully analyzed, avoiding the limitations of single-factor evaluation and making the evaluation results more reliable.

[0030] In a specific embodiment, determining whether the pollutant type contains a mixture of salt and oil includes: determining whether the spectral characteristic signal of the pollutant contains a first characteristic peak of salt and a second characteristic peak of oil; if it contains the first characteristic peak and the second characteristic peak, then the pollutant type contains a mixture of salt and oil.

[0031] Specifically, a unique first characteristic peak is preset for salt, and a corresponding second characteristic peak is preset for oil. The spectrometer then meticulously compares and judges the acquired spectral characteristic signals based on these preset characteristic peaks. If, during the analysis, the spectral characteristic signal simultaneously contains the first characteristic peak of salt and the second characteristic peak of oil, then it can be determined that the contaminant type includes a mixture of salt and oil.

[0032] By detecting specific characteristic peaks to determine whether a mixture of salt and oil is present, this method, based on the unique spectral characteristics of pollutants, can quickly and accurately identify the type of pollutant, greatly improving detection efficiency and reducing errors that may occur with manual judgment.

[0033] In a specific embodiment, after obtaining the first insulation performance evaluation score of the surface of the energized equipment, the method further includes: determining whether the first insulation performance evaluation score is less than a first preset threshold; if it is less than the first preset threshold, then extracting chemical distribution features other than salt and oil from the spectral feature signals of the multiple pollutants using a feature extraction algorithm; classifying the chemical distribution features using a support vector machine algorithm to obtain the types of chemical substances other than salt and oil in the spectral feature signals of the multiple pollutants; obtaining the pollutant concentration, distribution density, and distribution uniformity of the multiple pollutants in the pollutant concentration distribution data; the multiple pollutants include the types of chemical substances, the salt, and the oil; and obtaining the second insulation performance evaluation score of the surface of the energized equipment based on the pollutant concentration, the distribution density, and the distribution uniformity.

[0034] Specifically, a first preset threshold is set at 70 points (out of 100). If the first insulation performance evaluation score is 65 points, which is less than the first preset threshold, a further testing process is initiated. A feature extraction algorithm is used to extract the distribution features of chemical substances other than salt and oil from the spectral feature signals of various pollutants. This feature extraction algorithm accurately identifies and separates the spectral features of other chemical substances based on the unique waveform and peak characteristics of the spectral signals. A support vector machine (SVM) algorithm is used to classify these extracted chemical substance distribution features. The SVM algorithm constructs a classification hyperplane in a high-dimensional space to accurately distinguish the spectral features of different chemical substances, thereby obtaining the types of chemical substances other than salt and oil in the spectral feature signals of various pollutants, such as detecting the presence of a certain acidic substance. The pollutant concentration, distribution density, and distribution uniformity are obtained from the pollutant concentration distribution data for various pollutants. For salt, oil, and newly detected acidic substances, the average concentration is calculated by analyzing the concentration values ​​of each region in the image data as the pollutant concentration; the distribution density is obtained by counting the number of pollutants per unit area; and the distribution uniformity is determined by statistically analyzing the distribution of pollutants on the equipment surface. Based on the concentration, distribution density, and uniformity of pollutants, a pre-set comprehensive evaluation model is used to comprehensively consider the influence of these factors on insulation performance, and a second insulation performance evaluation score for the surface of the live equipment is calculated, thereby more comprehensively and accurately evaluating the insulation performance of the equipment surface.

[0035] When the initial insulation performance assessment score is unsatisfactory, further testing for chemicals other than salt and oil allows for a more comprehensive understanding of the contaminants on the equipment surface, preventing inaccurate assessments due to overlooking other substances that may affect insulation performance. By acquiring multi-dimensional information such as the concentration, distribution density, and uniformity of various contaminants, and comprehensively calculating the second insulation performance assessment score, the assessment results better reflect the actual insulation condition of the equipment surface, improving the accuracy and reliability of the assessment.

[0036] In a specific embodiment, after obtaining the pollutant concentration, distribution density, and distribution uniformity of multiple pollutants in the pollutant concentration distribution data, the method further includes: obtaining the interference intensity of multiple pollutants according to a preset interference level classification rule, and obtaining the interference frequency of multiple pollutants in the pollutant concentration distribution data; then, obtaining the second insulation performance evaluation score of the surface of the energized equipment based on the pollutant concentration, the distribution density, and the distribution uniformity includes: obtaining the second insulation performance evaluation score of the surface of the energized equipment based on the pollutant concentration, the distribution density, the distribution uniformity, the interference intensity, and the interference frequency of the multiple pollutants.

[0037] Specifically, the interference intensity of various pollutants is obtained according to a preset interference level classification rule. This rule sets different levels based on the degree of influence of pollutants on insulation performance. For example, salt, due to its high conductivity, is set as high interference intensity; oil, which reduces surface insulation performance, is set as medium interference intensity; and metal oxides are set as low interference intensity based on their properties. Simultaneously, the interference frequency of various pollutants in the pollutant concentration distribution data is obtained. By statistically analyzing the number of times pollutants interfere with insulation performance per unit time (e.g., twice per hour), it is found that salt has a high interference frequency, followed by oil, and metal oxides have a low frequency. Based on the pollutant concentration, distribution density, distribution uniformity, interference intensity, and interference frequency of various pollutants, a comprehensive evaluation model is used to calculate the second insulation performance evaluation score of the equipment surface. This model comprehensively considers the influence weight of each factor on insulation performance. For example, pollutant concentration and interference intensity are given higher weights. Through weighted calculation, the second insulation performance evaluation score is 60 points (out of 100), thus providing a more comprehensive and accurate assessment of the insulation performance of the equipment surface.

[0038] In a specific embodiment, obtaining the second insulation performance evaluation score of the surface of the electrical equipment based on the pollutant concentration, distribution density, distribution uniformity, interference intensity, and interference frequency of the various pollutants includes: integrating the pollutant types, pollutant concentrations, distribution densities, distribution uniformity, interference intensity, and interference frequency to obtain a preliminary feature vector; acquiring the environmental feature vector of the electrical equipment, which includes vectors composed of the ambient humidity, ambient temperature, image data sampling time, and sampling location of the electrical equipment; performing a weighted calculation on the preliminary feature vector and the environmental feature vector to obtain a weighted feature matrix; and obtaining the second insulation performance evaluation score of the surface of the electrical equipment based on the value of the weighted feature matrix and a second preset insulation performance evaluation rule.

[0039] Specifically, features are integrated from pollutant types (e.g., coded as salt 01, oil 02, acidic substances 03), pollutant concentrations, distribution densities, distribution uniformity (e.g., quantified values, such as 0.2 for poor uniformity, 0.5 for average, and 0.8 for good uniformity), interference intensity (e.g., 0.9 for high, 0.6 for medium, and 0.3 for low), and interference frequency to obtain a preliminary feature vector. Environmental feature vectors for the energized equipment are then acquired. For example, the average ambient humidity is 80%, represented by 0.8; ambient temperature varies depending on the season and time of day, but the average temperature during the assessment period is 25℃, represented by 0.25; image data was sampled at 10:00 AM, assumed to be represented by 0.1 (quantized over 24 hours); and equipment location, quantified using coordinate codes, is assumed to be represented by 0.5. The preliminary feature vector and the environmental feature vector are then weighted. Weights are assigned based on the importance of each factor to insulation performance, such as a pollutant-related feature weight of 0.7 and an environmental feature weight of 0.3. A weighted feature matrix is ​​obtained through matrix operations. Based on the value of the weighted feature matrix and the second preset insulation performance evaluation rules (such as setting different numerical ranges to correspond to different insulation performance levels and scores), the second insulation performance evaluation score of the surface of the live equipment is 60 points (out of 100), indicating that the insulation performance of the equipment is poor and needs to be maintained in a timely manner.

[0040] In a specific embodiment, please refer to Figure 2 The second embodiment of this application provides a schematic diagram of a system for evaluating the insulation performance of energized equipment after rinsing. The system includes: a feature signal acquisition module 201, a contaminant analysis module 202, a judgment module 203, a numerical acquisition module 204, and an evaluation module 205. The feature signal acquisition module 201 is used to acquire image data of the surface of the energized equipment after rinsing, and to collect spectral feature signals of multiple contaminants in the image data using a spectral analyzer. The contaminant analysis module 202 is used to perform contaminant analysis on the spectral feature signals of the multiple contaminants to obtain contaminant types and contaminant concentration distribution data in the image data. The judgment module 203 is used to determine whether the contaminant types include a mixture of salt and oil. The numerical acquisition module 204 is used to, if the mixture of salt and oil is included, acquire the first residual concentration average and the first concentration fluctuation coefficient of salt in the contaminant concentration distribution data, and acquire the second residual concentration average and the second concentration fluctuation coefficient of oil. The evaluation module 205 is used to obtain the insulation performance evaluation result of the surface of the energized equipment based on the first residual concentration average, the first concentration fluctuation coefficient, the second residual concentration average, and the second concentration fluctuation coefficient.

[0041] In a specific embodiment, please refer to Figure 3A second embodiment of the insulation performance evaluation system for energized equipment after rinsing, according to this application, includes: an image acquisition module 301, which acquires surface image data of the energized equipment after rinsing, collects spectral feature signals of various contaminants using a spectral analyzer, and extracts corresponding insulation performance benchmark values ​​from a preset database; an image processing module 302, which processes the image data using a convolutional neural network algorithm based on the acquired spectral feature signals to obtain preliminary classification results of contaminant types and distribution states; a contaminant fusion module 303, which determines whether the contaminant types in the preliminary classification results include a mixture of salt and oil; if so, it fuses the spectral feature signals with the classification results to determine a quantitative index of residual concentration; and an insulation evaluation module 304, which acquires simulated data of equipment surface insulation recovery based on the determined quantitative index of residual concentration. The system employs a support vector machine algorithm to analyze the deviation between simulated data and benchmark values, obtaining an evaluation score for the impact on insulation performance. The interference identification module 305 extracts additional chemical distribution features from the spectral feature signal if the evaluation score is below a preset threshold, determining the potential interference level of these features on insulation performance. The comprehensive scoring module 306 integrates the pollutant distribution status and interference features using a random forest algorithm based on the determined interference level to determine the overall comprehensive score for cleanliness. The risk analysis module 307 obtains matching records of historical equipment operation data based on the determined comprehensive score and determines the trend of similar pollutant residues' impact on lifespan in the matching records. The risk prediction module 308, based on the determined impact trend, fuses the comprehensive score and trend data to obtain a predicted value for the subsequent operational risk of the energized equipment.

[0042] This system achieves intelligent evaluation of the washing effect of live equipment through multimodal data fusion. An image acquisition module, mounted on the drone, simultaneously acquires images and spectral signals of the equipment surface after washing using a high-resolution camera and a spectrometer. The spectrometer detects the reflection characteristics of different contaminants in specific wavelengths and compares them with standard spectral templates in a database to extract insulation performance benchmark values. The image processing module performs convolutional neural network (CNN) feature extraction and classification on the acquired images to identify the type and distribution of contaminants. When the detection results include combined contamination such as salt and oil, the contaminant fusion module fuses the spectral feature signals with the classification results and calculates a quantitative index of residue concentration using multivariate regression. The insulation assessment module generates simulated data of electric field distribution and surface leakage paths based on the quantitative index of residue concentration, and uses a support vector machine (SVM) algorithm to calculate the deviation between the simulated results and the insulation performance benchmark value, obtaining an insulation performance assessment score. If the score is below a threshold, the system automatically triggers an interference identification module to extract potential interference features such as organic compounds and sulfides from the spectral signals and quantify their impact on insulation performance. The comprehensive scoring module inputs pollutant distribution characteristics, interfering factors, and assessment scores into a random forest model, outputting a comprehensive score for overall cleanliness. Finally, the risk analysis module compares the comprehensive score with a historical operational database to infer the impact trend of similar pollutant residues on equipment lifespan. The risk prediction module, based on the trend model and the current comprehensive score, outputs future operational risk values, providing data support for subsequent maintenance decisions.

[0043] This UAV-based live-line water flushing system significantly improves the intelligence and safety of live-line cleaning by introducing a multimodal fusion mechanism that combines target recognition and spectral analysis. Compared to traditional assessment methods that rely on manual visual inspection, this system can quantitatively analyze the types, distribution, and residual concentrations of contaminants, thus providing accurate insulation recovery assessment results. Through the combined application of CNN, SVM, and random forest algorithms, the system exhibits higher accuracy and robustness in contaminant identification, insulation performance assessment, and risk prediction. Spectral signals can effectively identify chemical residues that are difficult to distinguish with the naked eye, enabling automated judgment of flushing quality. The risk prediction module provides forward-looking decision-making based on historical operating trends, helping to prevent the risk of equipment aging or insulation breakdown. Overall, this invention improves the safety factor of live-line work, reduces the number of power outages for maintenance, and achieves intelligent maintenance.

[0044] The image acquisition module acquires surface image data of the equipment after rinsing, collects spectral feature signals of various pollutants using a spectrometer, and extracts corresponding insulation performance benchmark values ​​from a preset database. This includes acquiring surface image data of the equipment after rinsing using a camera, and using an image preprocessing algorithm to denoise and enhance the surface image data to obtain second image data; acquiring spectral feature signals of various pollutants from the second image data using a spectrometer, and using signal processing technology to filter and normalize the spectral feature signals to obtain a first feature signal; if the intensity of the first feature signal exceeds a preset threshold, querying the preset database for pollutant types matching the first feature signal, and using a data matching algorithm to determine the pollutant type identifier; based on the pollutant type identifier, extracting the corresponding insulation performance benchmark value from the preset database, and using a performance evaluation algorithm to compare the second image data and the benchmark value to determine the insulation performance level.

[0045] In one possible implementation, the image acquisition module of the target recognition-based UAV-based electrified water washing system captures images of the surface of the electrified equipment after washing using an industrial-grade camera mounted on the UAV. The camera's resolution is preferably 1920×1080 pixels, with a frame rate of 30 frames per second. After capturing the image, the system first executes an image preprocessing algorithm to process the obtained raw image data. The image preprocessing algorithm consists of three consecutive steps: The first step is noise removal, which uses mean filtering or median filtering algorithms to calculate the mean or median of surrounding pixels in the entire image pixel by pixel using a 3×3 convolution window, and replaces the original gray value of the pixel to eliminate random noise caused by air disturbance or drone jitter; the second step is image enhancement, which uses a linear contrast stretching method to remap the gray value range of the image from the original 0 to 255 to the full dynamic range of 0 to 255 to increase the local contrast of the image; the third step is edge thinning, which calculates the gradient intensity of the gray value difference between adjacent pixels in the image, sets weak gradient values ​​below 5 to 0, and retains gradient values ​​above 5, thereby obtaining the second image data with enhanced edge information.

[0046] During the spectral acquisition phase, the spectral analyzer works synchronously with the camera, emitting a spectral probe beam to each corresponding pixel region in the second image data and receiving the reflected spectral signal. The spectral analyzer's detection wavelength range is set to 400 to 1000 nanometers, with a 1-nanometer interval between each sampling point. The system arranges the light intensity values ​​at each wavelength sequentially to form a spectral curve, which is recorded as the raw spectral data. Subsequently, a signal processing flow is executed. The first step is bandpass filtering, setting the passband range to 410 to 990 nanometers to filter out ambient stray light interference. The second step is baseline correction, calculating the average light intensity at both ends of the band and subtracting it as background value. The third step is normalization processing, dividing the light intensity value at each wavelength by the maximum light intensity value in the spectral curve to normalize all light intensity values ​​to a standardized range of 0 to 1, obtaining standardized spectral feature data, which is the first feature signal.

[0047] The peak intensity of the first characteristic signal is then calculated. If this peak intensity is greater than a preset threshold, a contaminant signal is identified. The threshold is determined as follows: before rinsing the equipment, multiple spectral measurements are performed on the same type of equipment before and after cleaning. The average light intensity difference in the main wavelength band of the spectral curve is calculated, and 80% of this difference is taken as the threshold. For example, for common oil contaminants, the main spectral reflectance peak appears in the approximately 720 nm wavelength band. Through multiple experiments, the light intensity difference before and after cleaning is found to be approximately 0.5, so the system sets the threshold to 0.4. When the real-time detected peak intensity exceeds 0.4, the system triggers the contaminant identification procedure.

[0048] The pollutant identification process is executed by a data matching algorithm. The algorithm first calculates the squared difference between the light intensity of the detected first feature signal at each wavelength and the standard pollutant spectral template in the database. It then sums the squared differences at all wavelengths and takes the average to obtain the average error value. The system compares this average error value with the error values ​​of all pollutant templates in the database, selecting the pollutant type with the smallest error as the matching result and outputting the pollutant type identifier. For example, when the average error between the detection curve and the "salt contamination template" is 0.03, while the average error with the "oil contamination template" is 0.08, the system outputs the pollutant type identifier as "salt contamination."

[0049] After determining the type of contaminant, the system extracts the corresponding insulation performance benchmark value from a preset database based on the contaminant type identifier. This benchmark value includes three parameters: surface resistivity, leakage current threshold, and dielectric loss factor. Surface resistivity is the resistance value per unit area of ​​the equipment in a clean state, measured in ohms-meters; the leakage current threshold is the maximum allowable leakage current, measured in microamps; and the dielectric loss factor is a dimensionless value characterizing the degree of energy loss in the insulation material. Taking a typical high-voltage insulator as an example, the surface resistivity benchmark value is 1 × 10¹² ohms-meters, the leakage current threshold is 5 microamps, and the dielectric loss factor is 0.005. The system compares these benchmark values ​​with the corresponding regional features in the second image data. The comparison process includes: segmenting the area containing contaminants in the second image data, calculating the difference between the average brightness value of this area and the brightness of the standard clean area; if the brightness difference exceeds 20%, it is determined that surface residue exists in this area; simultaneously, the residue thickness is calculated based on the peak width changes in the spectral signal, with increased peak width indicating a thicker surface adhesion layer. The system converts the brightness difference and residual thickness into the equivalent surface resistivity change ratio, calculates the insulation performance reduction ratio using an empirical formula, and outputs the insulation performance level result accordingly.

[0050] Insulation performance is classified into four levels: Level 1 indicates complete performance recovery; Level 2 indicates slight residue but minimal impact on insulation performance; Level 3 indicates significant residue with insulation performance degradation exceeding 20%; and Level 4 indicates severe residue with insulation performance degradation exceeding 50%. The system automatically outputs the appropriate level based on the calculated insulation performance degradation percentage. For example, if the calculation shows an insulation performance degradation percentage of 25%, the output level will be 3.

[0051] The image processing module, based on the acquired spectral feature signals, uses a convolutional neural network algorithm to process image data and obtain preliminary classification results of pollutant types and distribution states. This includes acquiring the original spectral feature signals, performing frequency domain transformation on the signals using a fast Fourier transform to obtain a first frequency domain feature dataset; performing dimensionality reduction processing on the first frequency domain feature dataset using principal component analysis to extract key feature vectors, obtaining a second feature dataset; if the feature vector dimension of the second feature dataset meets a preset threshold, then using a convolutional neural network to train the feature vectors for classification, obtaining a preliminary classification model; and based on the preliminary classification model, inputting the second feature dataset for inference to determine the pollutant type and its distribution state, obtaining the final classification result.

[0052] In this embodiment, the system first acquires the raw spectral characteristic signal from a spectral analyzer. This signal is a continuous data sequence showing the variation of reflected light intensity with wavelength in the 400-1000 nm wavelength range on the surface of the equipment after rinsing. Each sampling point corresponds to a light intensity value, with a sampling interval of 1 nm, for a total of 601 sampling points. The system collects each spectral signal at fixed time intervals and stores it as a time-domain signal dataset. To analyze the frequency domain characteristics of the signal, the system performs a Fast Fourier Transform (FFT) calculation. This process consists of three steps: First, the 601 time-domain sampling points from each spectral signal sample are sequentially input into the transform operation module; second, a weighted sum of the sine and cosine components is performed on each sampling point to calculate the amplitude at each frequency component; third, the obtained amplitudes are rearranged in frequency order to form a frequency domain characteristic sequence. The transform result is the first frequency domain characteristic dataset, containing amplitude information for 601 frequency components. Through this transform, the low-frequency part reflects the overall pollution trend, while the high-frequency part reflects the details of pollution distribution.

[0053] Then, the system performs dimensionality reduction using principal component analysis on the first frequency domain feature dataset. The dimensionality reduction process includes five specific steps: First, calculate the average value of each frequency component in all spectral samples; second, calculate the covariance matrix between each frequency component, where covariance reflects the correlation between different frequency components; third, perform eigenvalue decomposition on the covariance matrix to obtain several eigenvalues ​​and corresponding eigenvectors; fourth, arrange all eigenvalues ​​from largest to smallest, calculate the cumulative variance contribution rate, and determine the number of principal components to retain when the cumulative variance contribution rate is greater than or equal to 95%; fifth, select the first few corresponding eigenvectors and project the original frequency domain feature data onto this eigenvector space to obtain the second feature dataset. This dataset is the dimensionality-reduced feature representation, and its dimension is generally 10% to 20% of the original dimension. In this embodiment, if the original dimension is 601, the dimension of the second feature dataset is approximately 60.

[0054] The system then determines whether the feature vector dimension of the second feature dataset meets a preset threshold. This threshold is used to ensure the validity and stability of the input features. The threshold is determined as follows: during the system training phase, feature data with different sample sizes are statistically analyzed. When the dimension is less than 5% of the total number of samples, the classification accuracy drops significantly. Therefore, the threshold is set to 5% of the total number of samples. For example, when the total number of training samples is 1000, the threshold is set to 50 dimensions. When the feature dimension of the second feature dataset is greater than or equal to 50 dimensions, the system considers the feature information sufficient and can proceed to the convolutional neural network classification stage; if it is below the threshold, the system returns to re-acquire the spectral signal.

[0055] If the threshold condition is met, the system invokes a convolutional neural network for classification training. The convolutional neural network consists of an input layer, three convolutional layers, two pooling layers, one fully connected layer, and one output layer. The specific implementation of each layer is as follows: The first convolutional layer has 32 convolutional kernels, a kernel size of 3×3, and a stride of 1. It performs one-dimensional convolution on the input second feature dataset to extract local features. The second convolutional layer has 64 convolutional kernels, a kernel size of 3×3, and a stride of 1. It convolves the output of the previous layer to extract higher-order features. The third convolutional layer has 128 convolutional kernels, a kernel size of 3×3, and a stride of 1. It further extracts deep semantic features. A pooling layer is set between every two convolutional layers, using max pooling with a pooling window size of 2 and a stride of 2. This is used to reduce data dimensionality and retain the main features. The fully connected layer contains 128 neurons and maps the convolutional features to the classification space. The number of output nodes in the output layer is set according to the number of pollutant types in the training samples. For example, if the pollutant types include salt, oil, and dust, then the number of output layer nodes is 3. The system uses cross-entropy as the loss function and updates the weights using the gradient descent algorithm. The training iterations are 100, and the learning rate is set to 0.001.

[0056] After the convolutional neural network completes training, the system obtains a preliminary classification model. The input to the classification model is the feature vector of the second feature dataset, and the output is the probability distribution of pollutant types. The system uses the pollutant category corresponding to the maximum output probability of each sample as the preliminary classification result for that sample. Simultaneously, the system maps the corresponding image region back to the original image based on the response position of the convolutional layer, calculates the location coordinates and area proportion of the pollutant distribution region, and thus obtains the distribution status of the pollutants.

[0057] During the inference phase, the system inputs the second feature dataset to be analyzed into a pre-trained convolutional neural network preliminary classification model, performing one forward propagation process. Specifically, this includes: first, passing the input feature vector through convolution, pooling, activation, and fully connected operations sequentially; second, calculating the predicted probability of each pollutant category at the output layer; and third, selecting the category with the highest predicted probability as the final judgment result. Simultaneously, the system generates a pollutant distribution heatmap, using different grayscale values ​​to indicate the spatial distribution of pollutants on the surface of electrical equipment, with higher grayscale values ​​indicating greater pollution concentration. Finally, the system outputs the pollutant type and its distribution status as the preliminary classification result.

[0058] The pollutant fusion module determines whether the pollutant type in the preliminary classification result includes a mixture of salt and oil. If so, it fuses the spectral feature signals with the classification result to determine the quantitative index of residual concentration. This includes acquiring spectral signals from a spectral sensor, performing data preprocessing on the spectral signals, extracting spectral features through Fourier transform to obtain preprocessed spectral feature data; if the spectral feature data contains characteristic peaks of salt and oil, the pollutant type is determined using a pre-established classification model to obtain a preliminary classification result; a support vector machine algorithm is used to fuse the spectral feature data with the preliminary classification result to determine the residual concentration of salt and oil, obtaining concentration distribution data; and using the concentration distribution data, the mean and standard deviation of the residual concentration of salt and oil are calculated to generate a quantitative index, resulting in the final quantitative result of residual concentration.

[0059] In this embodiment, the system first acquires spectral signals in real time from a spectral sensor. The sampling band of the spectral sensor is from 400 to 1000 nanometers, with a sampling interval of 1 nanometer, resulting in 601 light intensity sampling points per acquisition. The acquired spectral signals first enter the data preprocessing stage, performing three consecutive steps: The first step is baseline correction. The system calculates the average light intensity of the two bands at both ends of the spectral signal (400 to 420 nanometers and 980 to 1000 nanometers), takes the average value as the background intensity value, and subtracts this background intensity from the entire spectral curve to eliminate ambient light interference; the second step is noise removal. The system uses a moving average filtering method, smoothing the light intensity sequence with a sliding window of width 5, replacing the light intensity value of each sampling point with the average of the two sampling points before and after it, plus the value of itself, for a total of 5 values; the third step is normalization. The system divides all light intensity values ​​of the spectral curve by the maximum light intensity value of the curve, normalizing the light intensity range to between 0 and 1, obtaining the preprocessed spectral signal.

[0060] Subsequently, the system performs a Fourier transform on the preprocessed spectral signal to extract frequency domain features. This transform converts the wavelength domain signal into a frequency domain signal. Specifically, the process involves inputting 601 sampling points into the Fourier transform module, calculating the weighted sum of the sine and cosine components for each sampling point, and obtaining the amplitude information corresponding to the 601 frequency components. The system then arranges these frequency components in ascending order of frequency to form spectral feature data. Low-frequency components reflect the overall absorption trend of pollutants, while high-frequency components reflect the local superposition characteristics of mixed pollutants.

[0061] The system identifies characteristic peaks corresponding to salt and oil contamination in the spectral feature data. Salt contamination peaks typically appear in the approximately 720 nm to 760 nm wavelength range, exhibiting high reflected light intensity; oil contamination peaks appear in the approximately 880 nm to 930 nm wavelength range, with a greater absorption depth. The system determines the presence of mixed contamination based on the intensity variations within these two peak ranges. If significant peaks (both peak intensities exceeding 0.4) are present simultaneously in the spectral feature data, it is determined to be mixed salt and oil contamination. The peak threshold of 0.4 was obtained through multiple experimental calibrations: spectral curves were measured on clean surfaces, single-contaminated surfaces, and mixed-contaminated surfaces, and the minimum average peak intensity under mixed contamination conditions was taken as the threshold.

[0062] When mixed pollution is detected, the system invokes a pre-established classification model to determine the pollutant type. This classification model is trained on a large number of known pollution samples. The model input consists of the amplitude of the main frequency components of the spectral feature data, and the output is the probability of the pollutant category. The system infers for each input sample and outputs three probabilities: the probability of salinity, the probability of oil, and the probability of a mixed category. The category with the highest probability is selected as the preliminary classification result. If the probability of a mixed category is greater than 0.5, the presence of a mixture of salt and oil is confirmed.

[0063] After confirming mixed pollution, the system initiates a Support Vector Machine (SVM) algorithm to fuse spectral feature data with preliminary classification results to determine the residual concentrations of salt and oil. The input to the SVM algorithm includes the peak intensity of the corresponding band in the spectral feature data and the classification result label. The system uses a radial basis function (RBF) kernel as the SVM kernel function, determining the optimal classification surface by minimizing the combined cost of classification error and model complexity. The specific execution process is as follows: First, the system establishes a training set based on experimental samples, where each sample contains two input parameters: the peak intensity of the salt band and the peak intensity of the oil band; the output is the known concentration ratio of salt and oil in that sample. Second, the system uses the training set to train the SVM, determining the kernel function parameter γ and the penalty coefficient C. The parameter γ controls the model's sensitivity to local features, and its optimal value is selected from 0.01 to 1 using cross-validation. The penalty coefficient C balances error and generalization ability, and its value is set to 10. In the third step, after training, the system inputs real-time detected spectral feature data, and the support vector machine outputs predicted values ​​for salt and oil concentrations. The output is expressed as a percentage, ranging from 0 to 100%. For example, the output might be a salt concentration of 30% and an oil concentration of 20%.

[0064] The system generates concentration distribution data based on the concentration prediction results output by the support vector machine. To obtain more comprehensive information on surface residue, the system divides the spectral detection area into several grid cells, each cell corresponding to a set of concentration prediction values. The system calculates the average and standard deviation of salt and oil concentrations in all cells. The average value represents the overall residue level, and the standard deviation represents the uniformity of residue distribution. The standard deviation is calculated by averaging the squared differences between the concentration value of each cell and the average concentration value, and then taking the square root. If the standard deviation is less than 0.1, it indicates that the pollution distribution is uniform; if the standard deviation is greater than 0.3, it indicates that the pollution distribution is uneven and there is severe local residue.

[0065] Finally, the system generates a quantitative index of residual concentration based on the average concentration and standard deviation. This index includes two values: the mean residual concentration and the concentration fluctuation coefficient. The concentration fluctuation coefficient is calculated as the ratio of the standard deviation to the mean. The system outputs these two values ​​as the final quantitative result of the residual concentration. For example, when the mean salt residue is 25%, the standard deviation is 0.05, and the fluctuation coefficient is 0.2, the system outputs "Salt residue 25%, uniformly distributed"; when the mean oil residue is 40%, the standard deviation is 0.2, and the fluctuation coefficient is 0.5, the system outputs "Oil residue 40%, unevenly distributed". This result can serve as input data for subsequent insulation performance evaluation, providing a quantitative basis for judging the quality of equipment cleaning.

[0066] The insulation assessment module acquires simulated data on the insulation recovery of the equipment surface using a defined residual concentration quantification index. It then uses a support vector machine algorithm to analyze the deviation between the simulated data and a benchmark value to obtain an assessment score for the impact on insulation performance. This includes acquiring residual concentration data from the equipment surface, using sensor scanning technology to perform multi-point sampling of the surface state to obtain a quantification index of residual concentration; generating simulated insulation recovery data using a data simulation tool based on the quantification index; obtaining a preliminary deviation value if the simulated data deviates from a preset benchmark value; analyzing the deviation value using a support vector machine algorithm to determine the deviation analysis result; and calculating the insulation performance assessment score using a scoring algorithm based on the deviation analysis result. If the assessment score reaches a preset performance assessment threshold, the final insulation performance impact assessment score is obtained.

[0067] In this embodiment, the system first acquires residual concentration data from the device surface. To achieve high-precision measurement, the system employs sensor scanning technology to deploy sampling points on the device surface in an equidistant grid pattern. The grid spacing is determined based on the device size; when the device surface area is greater than 1 square meter, the sampling grid spacing is set to 10 centimeters; when the area is less than 1 square meter, the spacing is set to 5 centimeters. The coordinates of each sampling point are recorded in real time by the UAV's pose sensor, with the position error controlled within ±2 centimeters. At each sampling point, the system uses a spectral sensor to measure the residual salt and oil concentration, collecting the concentration data for each point and storing it in a data table. Based on the concentration data from all sampling points, the system calculates the average and standard deviation to form a quantitative index of residual concentration. This quantitative index includes two items: average residual concentration and concentration fluctuation coefficient, where the concentration fluctuation coefficient is the ratio of the standard deviation to the average, and the unit is dimensionless.

[0068] Next, the system uses data simulation tools to generate simulated data for insulation recovery based on the obtained quantitative indicators. This simulation process is achieved by inputting residual concentration distribution parameters into a three-dimensional electric field model. First, the surface material parameters of the equipment are set in the electric field simulation model, including dielectric constant, surface resistivity, and humidity coefficient. The dielectric constant is determined by the characteristics of the equipment material, such as 6.5 for ceramic insulators and 3.2 for composite insulation materials; the surface resistivity is determined based on a baseline value under clean conditions, such as 1 × 10¹² ohm-meters. The system multiplies the average residual concentration by a pollution coefficient (a scaling factor between 0.01 and 0.05, obtained experimentally based on the type of pollutant) to obtain the percentage decrease in surface resistivity. For example, when the average salt residue is 20% and the pollution coefficient is 0.03, the percentage decrease in surface resistivity is 0.6%. The simulation tool randomly distributes this resistivity variation region on the model surface and adjusts the variation amplitude according to the concentration fluctuation coefficient, thereby generating simulated data for insulation recovery under contaminated conditions. The simulated data includes parameters such as electric field intensity distribution, surface current density, and the location of leakage risk points.

[0069] The system then compares the simulated data with preset benchmark values ​​to calculate the deviation. The benchmark values ​​are the standard electric field intensity distribution and surface current density distribution on the equipment surface under clean conditions. The standard electric field intensity is 3 kV per millimeter in the central region of the surface, and the surface current density does not exceed 0.02 microamps per square centimeter. The system calculates the difference between the simulated data and the benchmark data point by point, and takes the average of the deviations at all sampling points as the preliminary deviation value. If the deviation value is greater than 5%, it is considered that there is a significant residual effect on the equipment surface.

[0070] To address this deviation, the system employs a Support Vector Machine (SVM) algorithm to analyze the distribution of deviations between the simulated data and the baseline values, determining the deviation analysis results. The SVM model inputs the electric field strength deviation, surface current density deviation, and residual concentration fluctuation coefficient for each sampling point, and outputs the insulation performance degradation level. During model training, a training sample set is established using historical experimental data. The training samples include deviation data under known pollution conditions and their corresponding insulation performance degradation levels. The system uses a radial basis function kernel and determines the kernel parameter γ and penalty parameter C through cross-validation. Parameter γ is tested within the range of 0.01 to 1, and the final value of 0.1, which yields the highest classification accuracy, is selected. The penalty parameter C is set to 10 to balance error penalty with model generalization ability. After training, the system inputs real-time simulated data and outputs the deviation analysis results. The output results include the insulation performance degradation level and the corresponding risk index, with the risk index ranging from 0 to 1; a higher value indicates a more significant performance degradation.

[0071] Finally, the system calculates the insulation performance evaluation score based on the deviation analysis results using a scoring algorithm. The scoring algorithm is based on a risk index, mapping it to a score range of 0 to 100. The system uses a linear scoring function: when the risk index is 0, the evaluation score is 100; when the risk index is 1, the evaluation score is 0. If the risk index falls between these two values, the system calculates the corresponding score proportionally. For example, when the risk index is 0.3, the evaluation score is 70. The system sets a performance evaluation threshold to determine the insulation performance recovery status. The threshold is determined by statistically analyzing the lowest evaluation score achieved after cleaning in multiple sets of live equipment experiments. Experiments show that when the evaluation score is below 60, the probability of leakage risk during equipment operation increases significantly; therefore, the threshold is set to 60. When the evaluation score is higher than or equal to 60, the system outputs "Insulation performance restored to normal"; when the evaluation score is lower than 60, it outputs "Insulation performance insufficient recovery". The final insulation performance impact evaluation score is the calculated result.

[0072] The interference identification module, if the evaluation score is lower than a preset threshold, extracts additional chemical substance distribution features from the spectral feature signal to determine the potential interference level of these features on insulation performance. This includes obtaining initial spectral data from the spectral feature signal, performing denoising and baseline correction on the initial spectral data using signal preprocessing methods to obtain standardized first spectral data; if the evaluation score of the first spectral data is lower than the preset threshold, extracting chemical substance distribution features from the first spectral data using a feature extraction algorithm to obtain a first chemical feature set; classifying the chemical substance distribution features using a support vector machine algorithm based on the first chemical feature set to determine the interference level of the chemical substance distribution on insulation performance, obtaining an interference level classification result; comparing the interference level classification result with a preset insulation performance standard to verify the potential impact of the chemical substance distribution features on insulation performance, obtaining a final verification result.

[0073] In this embodiment, the system first acquires initial spectral data from the spectral feature signal. The spectral signal is acquired in real time by a spectral sensor, with a sampling wavelength range of 400 to 1000 nanometers, a sampling interval of 1 nanometer, and 601 sampling points per sample. The system performs signal preprocessing on the acquired spectral data to ensure the accuracy and comparability of the signal. The preprocessing method includes two steps: denoising and baseline correction. In the denoising step, the system uses a moving average method with a sliding window of width 5 to smooth the light intensity data. Each light intensity value is obtained by averaging its own value and the light intensity values ​​of its two adjacent points, thereby eliminating random noise caused by air disturbances, illuminance variations, etc. In the baseline correction step, the system selects the average light intensity at both ends of the spectrum (400 to 420 nanometers and 980 to 1000 nanometers) as the baseline level and subtracts this average value from the entire spectral curve to eliminate the baseline drift caused by the device response curve. After these two steps, the system obtains standardized first spectral data. The normalized spectral signal intensity values ​​are limited to the range of 0 to 1, where 0 represents the lowest reflection intensity and 1 represents the highest reflection intensity.

[0074] Next, the system determines whether interference identification is needed based on the evaluation score output by the insulation assessment module. The evaluation score threshold is set as follows: in multiple sets of experiments, the average score when insulation performance deteriorates but no breakdown occurs is calculated, and 95% of this value is taken as the threshold. Taking a typical experiment as an example, if the equipment can still operate safely when the score is between 60 and 65, the system sets the threshold to 60. When the current evaluation score is below 60, the system enters the chemical substance feature extraction stage.

[0075] During the feature extraction stage, the system calls a feature extraction algorithm to analyze the standardized first spectral data. This algorithm scans absorption and reflection peaks across the entire wavelength range to extract key parameters reflecting the distribution of chemical substances. The extraction steps include: First, the system calculates the first derivative of the spectral curve to detect peak positions. Whenever the intensity curve shows a change in derivative from positive to negative, this position is recorded as a characteristic peak. Second, the system calculates the peak height and peak width of each characteristic peak. Peak height reflects absorption or reflection intensity, and peak width reflects the concentration distribution range of the chemical component. Third, the system categorizes peak positions according to the characteristic absorption wavelengths of different chemical substances. For example, the characteristic absorption peaks of organosilicon pollutants appear in the approximately 450 nm to 480 nm wavelength range, while the characteristic absorption peaks of sulfide pollutants appear in the approximately 870 nm to 910 nm wavelength range. The system combines the peak height, peak width, and position parameters within the corresponding wavelength range to form a first chemical feature set. Each feature set sample includes values ​​such as wavelength position, peak height, peak width, and normalized intensity ratio, and the data is stored in a structured manner for subsequent analysis.

[0076] Subsequently, the system employs a Support Vector Machine (SVM) algorithm to classify the first chemical feature set to determine the degree of interference of chemical substance distribution characteristics on insulation performance. The SVM input is the parameter vector of the chemical feature set (including peak height, peak width, and band position, etc.), and the output is the interference level classification result. During model training, a training set is built based on historical data samples. The output label of each sample corresponds to a known level of insulation interference, categorized into three levels: "low interference," "medium interference," and "high interference." The system selects a radial basis function (RBF) kernel as the SVM kernel function to balance nonlinear feature mapping capability and computational efficiency. The parameters γ and penalty coefficient C are determined through cross-validation. γ ranges from 0.01 to 1, and the value with the highest classification accuracy is selected during validation. C ranges from 1 to 100, and the optimal value with the smallest error and model stability is selected experimentally, typically set to 10. After training, the model can automatically output the interference level of chemical substance distribution based on spectral characteristics.

[0077] The system inputs the real-time collected chemical feature set into a pre-trained support vector machine model, which outputs the probability distribution of the interference level for each sample. The system uses the level with the highest probability as the final classification result. For example, if the probability of "high interference" is 0.7, "medium interference" is 0.2, and "low interference" is 0.1, then the interference level corresponding to this chemical feature set is determined to be "high interference". To enhance the reliability of the results, the system averages the probability values ​​of the classification results from multiple sampling points to calculate the overall interference level.

[0078] Finally, the system compares the interference level classification results with preset insulation performance standards to verify the potential impact of chemical substance distribution characteristics on insulation performance. The preset insulation performance standards include an electric field uniformity index and a lower limit for surface resistivity. The electric field uniformity index refers to the relative deviation of the electric field strength at different locations on the surface of the energized equipment, with an allowable upper limit of 10%; the lower limit for surface resistivity is 1 × 10¹⁰ ohm-meters.

[0079] The pollution distribution characteristic parameters corresponding to the interference classification results are input into the verification module, which compares them with insulation performance records under the corresponding conditions in the standard database. If the electric field deviation corresponding to the detected "high interference" chemical distribution area exceeds 10% or the surface resistivity is lower than the above lower limit, the system outputs "significant insulation interference exists"; if the interference level is "moderate interference" and the deviation is between 5% and 10%, the system outputs "slight insulation interference exists"; if the interference level is "low interference" and the deviation is less than 5%, the system outputs "interference is negligible". The system finally outputs this result as the verification result of potential interference to insulation performance and records it in the database for subsequent risk analysis module use.

[0080] The comprehensive scoring module, based on the determined level of interference, uses a random forest algorithm to integrate pollutant distribution status and interference characteristics to determine the overall cleaning quality score. This includes acquiring pollutant distribution status data and interference characteristic data, and using the random forest algorithm to integrate the pollutant concentration, pollutant type, distribution density, distribution uniformity, interference intensity, and interference frequency to obtain a preliminary feature vector. Based on the preliminary feature vector, combined with environmental humidity, environmental temperature, sampling time, and sampling location, a weighted average method is used to calculate the weight of each feature, resulting in a weighted feature matrix. If the pollutant concentration in the weighted feature matrix is ​​higher than a preset threshold, a clustering analysis method is used to group the distribution density and distribution uniformity to determine cleaning priority groups. Based on the cleaning priority groups and interference frequency, a linear regression method is used to predict the cleaning efficiency to obtain the comprehensive score.

[0081] In this implementation, the system first obtains pollutant distribution status data and interference characteristic data from the preceding modules. The pollutant distribution status data includes four indicators: pollutant concentration, pollutant type, distribution density, and distribution uniformity. The interference characteristic data includes two indicators: interference intensity and interference frequency. Pollutant concentration represents the average mass ratio of surface residues, expressed as a percentage. Pollutant type is the classification result, with values ​​such as "salt," "oil," or "mixed pollution." Distribution density is the coverage ratio of polluted areas per unit area. Distribution uniformity is the ratio of standard deviation to average concentration. Interference intensity is the numerical result of insulation interference level output by the support vector machine, ranging from 0 to 1. Interference frequency is the ratio of the number of interference events occurring within the detection period to the total number of samples. The system uniformly organizes these six indicators into a standardized input dataset. All values ​​are normalized to a range of 0 to 1 to ensure consistent comparability of indicators with different dimensions.

[0082] Next, the system uses the Random Forest algorithm to integrate the six features mentioned above, generating a preliminary feature vector. The implementation process of the Random Forest algorithm includes the following four steps: First, the system establishes a training set based on historical sample data. Each sample contains six input features and a corresponding cleanliness level (e.g., "Excellent," "Good," "Medium," "Poor"). Second, during the training phase, the system randomly extracts a subset of samples and a subset of features to construct multiple decision trees, each trained independently. The information gain criterion is used to divide nodes during training to maximize feature discriminative power. The number of trees is set to 100 to ensure the stability of the results. Third, in each decision tree, starting from the root node, the nodes are divided level by level according to feature thresholds. For example, if the pollutant concentration is greater than 0.3 and the distribution density is greater than 0.4, the branch is judged as "Medium Cleanliness"; if the concentration is less than 0.1 and the interference intensity is less than 0.2, the branch is judged as "High Cleanliness". Fourth, after all decision trees have been trained, the system performs a vote on the output of each tree, and outputs the cleanliness level that appears most frequently as the overall output of the random forest. Simultaneously, the system performs a weighted average of the feature weights of all decision tree nodes to generate a preliminary feature vector. This preliminary feature vector is a numerical sequence of length 6, representing the relative contribution of each input feature to the overall cleanliness evaluation.

[0083] Subsequently, the system performs weighted adjustments based on the initial feature vector and environmental parameters. These environmental parameters include ambient humidity, ambient temperature, sampling time, and sampling location. Ambient humidity is measured as a percentage from a humidity sensor mounted on the UAV; ambient temperature is measured in degrees Celsius from a temperature sensor; sampling time is recorded as a timestamp accurate to the minute; and sampling location is provided by the Global Positioning System (GPS), with latitude and longitude accuracy within ±5 meters. The system employs a weighted average method to adjust the six features. The calculation process is as follows: First, the system determines the influence coefficient of each environmental parameter based on experience or historical statistics. The influence coefficient for humidity is set to 0.3, for temperature to 0.2, for time to 0.2, and for location to 0.3. Second, the system multiplies these influence coefficients by the standardized value of the corresponding feature, and then averages all the results to obtain the environmentally adjusted weighted feature matrix. This matrix reflects the intensity of cleanliness quality characteristics under the combined effects of pollutant characteristics and environmental conditions.

[0084] When the contaminant concentration in the weighted feature matrix exceeds a preset threshold, the system enters the cluster analysis phase to determine cleaning priority grouping. The contaminant concentration threshold was determined through multiple experiments: insulation recovery tests were conducted on samples with different concentrations, and the rate of change of surface leakage current after cleaning was calculated. A change rate exceeding 10% was considered insufficient cleaning. Experimental results showed that when the contaminant concentration was greater than 0.25, the leakage current change rate exceeded 10%, therefore the system set the threshold to 0.25. The cluster analysis employed a hierarchical clustering algorithm, clustering based on two characteristics: distribution density and distribution uniformity. The algorithm execution steps are as follows: First, calculate the Euclidean distance between each pair of samples. The smaller the distance, the more similar the distribution characteristics. Second, the system merges sample groups according to the shortest distance based on the distance matrix, forming several clusters. Third, the system sorts the clustering results according to pollutant concentration and interference intensity. Clusters with high pollutant concentration and high interference intensity are determined as "priority cleaning groups," clusters with medium concentration and medium interference are "secondary cleaning groups," and clusters with low concentration and low interference are "cleaning completed groups." Fourth, the system marks each group with a priority level of 1, 2, and 3, where 1 represents priority cleaning and 3 represents thorough cleaning.

[0085] After grouping, the system uses linear regression to predict cleaning efficiency and calculate a comprehensive score based on cleaning priority grouping and interference frequency. The specific calculation process is as follows: First, the system uses the average contaminant concentration, interference frequency, and priority level of each cluster as input variables, and the manually verified percentage of insulation recovery after cleaning as the target variable. Second, the system fits a linear relationship using the minimum square error principle to obtain the prediction model parameters. Third, real-time data is input into the model to obtain the predicted cleaning efficiency value. Cleaning efficiency is expressed as a percentage, representing the proportion of insulation performance recovery on the equipment surface after cleaning. The system then maps the predicted cleaning efficiency to a comprehensive score. The mapping rules are: cleaning efficiency above 90% corresponds to a score of 90 to 100; cleaning efficiency of 70% to 90% corresponds to a score of 70 to 89; cleaning efficiency of 50% to 70% corresponds to a score of 50 to 69; and cleaning efficiency below 50% corresponds to a score of 0 to 49. When the overall score is greater than or equal to 80 points, the system outputs "Excellent cleaning effect"; when the score is between 60 and 79 points, it outputs "Good cleaning effect"; when the score is less than 60 points, it outputs "Insufficient cleaning effect". Different cleaning effects will result in different insulation properties of the equipment.

[0086] The risk analysis module, through a determined comprehensive score, obtains matching records of historical equipment operation data and determines the impact trend of similar pollutant residues on lifespan in the matching records. This includes obtaining equipment operation data and environmental parameters from a historical database, extracting time-series formatted operation status and pollutant residue data using a data acquisition module to obtain an initial dataset after cleaning; calculating a comprehensive score based on the initial dataset, using a weighted average method to score the operation status and pollutant residue data, with weights determined by pre-established pollutant impact factors, to obtain a comprehensive score result; querying matching records in the historical database using the comprehensive score result, if the comprehensive score is greater than a preset threshold, extracting the corresponding equipment operation data and pollutant residue data to obtain a set of matching records; and using a linear regression algorithm to analyze the relationship between pollutant residues and equipment lifespan for the time-series data in the set of matching records, calculating the slope to determine the impact trend, and obtaining the lifespan impact trend result.

[0087] In this implementation, the system first retrieves the operating data and environmental parameters of the energized equipment from a historical database. The historical database serves as a long-term storage platform, recording information such as the equipment's operating voltage, current, surface temperature, humidity, operating time, and residual contaminant concentration. To ensure data consistency, the system invokes the data acquisition module to convert the raw records into a time-series format. The time resolution is set to one hour, with each hour serving as a sampling interval. The system extracts the corresponding operating status parameters for each time point, including operating voltage, load current, and surface temperature, while simultaneously extracting the residual contaminant concentration and ambient humidity at the same time, forming a complete set of data records. The system obtains the initial dataset after cleaning through hourly sampling. This dataset includes fields such as a time-series index, operating parameters, residual contaminant values, and cleaning time identifiers, ensuring data continuity and traceability.

[0088] Next, the system calculates a comprehensive score based on the initial dataset. The score calculation is performed using a weighted average method of operating status parameters and contaminant residue parameters. Operating status parameters include voltage stability, load balancing, and temperature stability; contaminant residue parameters include salt residue concentration and oil residue concentration. The system assigns a weight to each parameter, determined by a pre-established contaminant impact factor. The contaminant impact factor is obtained by statistically analyzing the contribution of each contaminant to insulation performance degradation in historical samples. For example, in the past 100 operating samples, the average proportion of insulation life reduction due to salt residue was 30%, and the average proportion due to oil residue was 20%. Therefore, the weight of salt is set to 0.3, and the weight of oil residue is set to 0.2. The weights of the three operating status parameters are set to 0.5, and within each parameter, they are allocated according to their relative impact as follows: voltage stability 0.2, load balancing 0.15, and temperature stability 0.15. Each parameter is standardized, unifying the numerical range to between 0 and 1. Then, the products of their respective weights are summed and averaged to obtain the comprehensive score. The scoring results range from 0 to 100 points, with higher scores indicating more ideal equipment operation and contaminant conditions. The system stores this comprehensive scoring result for matching analysis.

[0089] After obtaining the comprehensive score result, the system performs a matching record search in the historical database. Search criteria include three items: equipment model, operating environment category, and pollutant type. If records with the same or similar characteristics as the current equipment exist in the database, the system calculates the comprehensive score value for these historical samples. If the current comprehensive score is greater than a preset threshold, the equipment is considered to be in good condition, and the system extracts the corresponding equipment operation data and pollutant residue data from the database to form a matching record set. The comprehensive score threshold is determined through statistical analysis of a large amount of sample data. Specifically, the average operating life of the equipment is analyzed under different comprehensive scores. When the lifespan decay rate is less than 10%, the corresponding score range is considered a stable state. Statistical results show that the corresponding score is 80 points, therefore the threshold is set to 80. If the comprehensive score is greater than or equal to 80 points, the system determines that the equipment is in a stable state and extracts the corresponding matching records; if the score is lower than 80 points, the system only retains a portion of the samples for comparative analysis.

[0090] After acquiring the set of matching records, the system performs lifetime trend analysis on the time series data. The analysis process includes the following steps: First, the system arranges each matching record in chronological order and extracts two core variables: operating time and residual pollutant concentration. Operating time is expressed in hours, and residual pollutant concentration is expressed as a percentage. Second, the system smooths each set of data to eliminate the interference of occasional fluctuations. The smoothing process uses a three-point moving average method, replacing the pollutant concentration at each time point with the average of the concentrations at the previous, current, and next times. Third, the system uses a linear regression algorithm to analyze the relationship between residual pollutants and equipment lifespan. To ensure interpretability, the system uses operating time as the independent variable and the rate of change of residual pollutant concentration or the rate of insulation performance degradation as the dependent variable, fitting the data trend for each sample. The linear regression process determines the best-fit line using the principle of minimum squared error, resulting in a straight line representing the trend of pollutant influence. Fourth, the system calculates the slope value based on the regression results and determines the direction of the trend. If the slope is negative and its absolute value is greater than 0.1, it indicates that the pollutant residue concentration increases over time, leading to a significant decrease in lifetime, and the system outputs a trend result of "significant negative impact." If the absolute value of the slope is between 0.05 and 0.1, it indicates a slow decline in lifetime, and the system outputs "slight negative impact." If the slope is close to 0 or positive, it indicates that the pollutant residue has no significant impact on lifetime, and the system outputs "no obvious trend." In the fifth step, the system generates a lifetime impact trend report based on the trend results, which includes the pollutant type, residue level, operating environment, regression slope value, and impact level. The report also records the corresponding sample size and time range for subsequent data fusion in the risk prediction module.

[0091] The risk prediction module, based on the assessed impact trend, integrates comprehensive scores and trend data to obtain predicted values ​​for the subsequent operational risks of the live equipment. This includes acquiring real-time and historical data of the live equipment, querying operating status, equipment parameters, environmental factors, and load levels from a pre-established database, and using data cleaning methods to remove outliers to obtain a standardized dataset. Based on the standardized dataset, the module integrates comprehensive scores and trend data. If the operating status is abnormal and the load level exceeds a preset threshold, a weighted average algorithm is used to calculate the failure probability, resulting in a preliminary risk assessment value. For the preliminary risk assessment value, combined with environmental factors and maintenance records, if environmental factors exceed safe limits, the failure probability weights are adjusted, and a risk level is determined through a mapping table to obtain a risk level classification result. Based on the risk level classification result and the comparative trend of historical and real-time data, a random forest method is used to predict the subsequent operational risks of the live equipment, resulting in the final risk prediction value.

[0092] In this embodiment, the system first acquires real-time and historical data of the energized equipment. Real-time data includes the equipment's operating voltage, current, surface temperature, humidity, wind speed, load power, and current operating time. Historical data includes the equipment's lifespan, fault records, maintenance intervals, and pollutant residue levels, as well as data for the same model of equipment. The system retrieves this data from a pre-established database, while simultaneously acquiring relevant environmental parameters (including air humidity, ambient temperature, wind speed, and air pressure) and the equipment's operating load level. To ensure data consistency and validity, the system performs data cleaning to remove outliers. The cleaning process includes three steps: First, range verification. The system uses the operating parameter ranges specified in the equipment's technical manual, such as a voltage fluctuation range of ±10% of the rated value and a temperature range of 0 to 80 degrees Celsius, to determine and remove records exceeding these ranges as outliers. Second, logic verification. The system checks whether the voltage, current, and power relationships at the same time point conform to power calculation logic; for example, the power should not exceed a 10% error range of the product of voltage and current. Third, time continuity verification. The system checks whether the data time interval is continuous. If the sampling interval is greater than 2 hours or more than 3 consecutive records are missing, the average of the preceding and following records is filled in. After cleaning, the system normalizes all valid data, converting the numerical range of each indicator to a uniform range of 0 to 1, resulting in a standardized dataset.

[0093] Next, the system integrates comprehensive scores and trend data based on a standardized dataset. The comprehensive score, derived from the comprehensive scoring module, represents the current cleaning quality and insulation recovery level, ranging from 0 to 100. The trend data, derived from the risk analysis module, represents the impact of contaminant residues on equipment lifespan, ranging from -1 to 1, where negative values ​​indicate a declining lifespan and positive values ​​indicate an extending lifespan. The system first standardizes the comprehensive score to a range of 0 to 1 (e.g., a score of 100 corresponds to 1, and a score of 0 corresponds to 0), and then performs a weighted fusion with the trend data. The weight ratios are determined experimentally, with a comprehensive score weight of 0.6 and a trend data weight of 0.4, to ensure the model maintains a balance between long-term trends and the current state. The system calculates the fusion result and determines whether the operating status is abnormal. The criteria for determining abnormal operating status are current fluctuations exceeding ±15% or voltage deviations exceeding ±10%; if either condition is met, it is considered abnormal. When an abnormal operating status is detected and the load level is higher than a preset threshold, the system calculates a preliminary risk assessment value. The load threshold is determined through long-term operational sample statistics. When the load exceeds 80% of the rated power, the equipment temperature rise and failure rate increase significantly; therefore, the threshold is set to 0.8. If the current load level is higher than 0.8 and the operating status is abnormal, the system calculates the failure probability using a weighted average algorithm. The execution process of the weighted average algorithm is as follows: First, the four indicators—comprehensive score, trend data, load level, and abnormal status—are multiplied by their respective weight coefficients. The weight coefficients are determined based on the historical contribution of the features to the failure rate, with the comprehensive score having a weight of 0.3, the trend data a weight of 0.2, the load level a weight of 0.3, and the abnormal status a weight of 0.2. Second, the weighted results are summed and averaged to obtain a preliminary risk assessment value. The assessment value ranges from 0 to 1, with values ​​closer to 1 indicating a higher failure probability. For example, if the comprehensive score is 0.6, the trend data is -0.4, the load level is 0.9, and the abnormal status is 1, the calculated result is approximately 0.7, corresponding to a medium-to-high risk.

[0094] Subsequently, the system modifies the initial risk assessment value based on environmental factors and maintenance records. When the ambient humidity exceeds 90%, the temperature exceeds 80 degrees Celsius, or the wind speed exceeds 15 meters per second, the system considers the environmental factors to be outside the safe range. In this case, the failure probability is corrected by adjusting the weights. The correction method is as follows: if any environmental factor exceeds the limit, the environmental risk weight is set to 0.2 and added to the calculation, so that the total weight is normalized back to 1. For example, in a high humidity environment, environmental risk will increase the final risk assessment value by approximately 10%. In addition, the system adjusts the weights based on maintenance records. If equipment has not been maintained for more than 30% of the recommended maintenance cycle, the maintenance delay weight is set to 0.1, increasing the corresponding failure probability. The corrected risk assessment value is converted into a risk level through a preset mapping table. The mapping table is established based on historical fault data statistics, and the risk level is divided into four levels: Risk Level 1 (Low Risk): Risk value 0 to 0.3; Risk Level 2 (Medium Risk): Risk value 0.31 to 0.6; Risk Level 3 (High Risk): Risk value 0.61 to 0.8; Risk Level 4 (Very High Risk): Risk value 0.81 to 1. The system directly outputs the corresponding level based on the evaluation value. For example, if the evaluation value is 0.75, then "High Risk" is output.

[0095] Finally, the system uses a random forest method to predict the subsequent operational risk value of the equipment by combining the risk level classification results with the comparison trend of historical and real-time data. The random forest model uses historical samples as the training set, and each sample includes the aforementioned standardized features (operating status, environmental factors, load level, comprehensive score, trend data, risk level, etc.) and the corresponding actual failure results. The model training process uses 100 decision trees, with each tree having a depth of no more than 10 layers to prevent overfitting. After training, the system inputs real-time data and obtains the risk prediction output through the voting results of all decision trees. Each tree outputs a "fault" or "normal" judgment, and the system calculates the proportion of "fault" outputs to the total number of trees as the final risk prediction value. For example, if 65 out of 100 trees judge it as "fault," the final risk prediction value is 0.65, indicating that there is a 65% potential failure risk in the future operation cycle of the equipment. The system outputs this result in relation to the aforementioned risk level; for example, a risk prediction value of 0.65 corresponds to a "high risk" level. The prediction results are stored in the database and fed back to the monitoring terminal, providing a basis for UAV operation scheduling and equipment maintenance planning.

[0096] In a specific embodiment, the third embodiment of this application provides an insulation performance evaluation device for live equipment after rinsing, including a memory and a processor. The memory stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of the method as described in any of the first embodiments of this application.

[0097] In a specific embodiment, the fourth embodiment of this application provides a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the steps of the method as described in any one of the first embodiments of this application.

[0098] The above embodiments merely illustrate several implementation methods of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

[0099] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments for application in other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A method for evaluating the insulation performance of live equipment after rinsing, characterized in that, The method includes: Image data of the surface of the energized equipment after rinsing is acquired, and spectral characteristic signals of various pollutants in the image data are collected by a spectral analyzer. Pollutant analysis is performed on the spectral characteristic signals of the various pollutants to obtain pollutant types and pollutant concentration distribution data in the image data; Determine whether the contaminant type contains a mixture of salt and oil; If the mixture of salt and oil is included, the average first residual concentration and the first concentration fluctuation coefficient of salt in the pollutant concentration distribution data are obtained, and the average second residual concentration and the second concentration fluctuation coefficient of oil are obtained. The insulation performance evaluation results of the surface of the energized equipment are obtained based on the first average residual concentration, the first concentration fluctuation coefficient, the second average residual concentration, and the second concentration fluctuation coefficient.

2. The method for evaluating the insulation performance of live equipment after rinsing as described in claim 1, characterized in that, The step of obtaining the insulation performance evaluation result of the surface of the live equipment based on the first average residual concentration, the first concentration fluctuation coefficient, the second average residual concentration, and the second concentration fluctuation coefficient includes: Simulated distribution data of the salt and oil mixture on the surface of the electrical equipment are obtained based on the first average residual concentration, the first concentration fluctuation coefficient, the second average residual concentration, and the second concentration fluctuation coefficient; the simulated distribution data includes electric field intensity distribution and surface current density; The simulated distribution data is compared with the insulation performance benchmark value to obtain the distribution data deviation value; the distribution data deviation value includes the electric field intensity distribution deviation value and the surface current density deviation value; the insulation performance benchmark value includes the standard electric field intensity distribution and the standard surface current density when there is no salt and oil mixture. The distribution data deviation value, the first concentration fluctuation coefficient, and the second concentration fluctuation coefficient are input into a preset insulation analysis model to obtain the insulation risk index of the surface of the washed live equipment. Based on the insulation risk index and the first preset insulation performance evaluation rule, a first insulation performance evaluation score is obtained for the surface of the live equipment. The first insulation performance evaluation score is used to characterize the insulation performance evaluation result of the surface of the live equipment.

3. The method for evaluating the insulation performance of live equipment after rinsing as described in claim 1, characterized in that, The first concentration fluctuation coefficient and the second concentration fluctuation coefficient were obtained using the following method: The image data is divided into several cells; Based on the pollutant concentration distribution data, the salt residue concentration value and oil residue concentration value of each cell are obtained; Based on the salt and oil residue concentration values ​​of all cells, obtain the standard deviation of salt residue, the standard deviation of oil residue, the mean of salt residue concentration, and the mean of oil residue concentration. Obtain a first ratio of the standard deviation of salt residue to the mean of salt residue concentration, where the first ratio is the first concentration fluctuation coefficient; Obtain a second ratio of the standard deviation of the oil residue to the mean concentration of the oil residue, whereby the second ratio is the second concentration fluctuation coefficient.

4. The method for evaluating the insulation performance of energized equipment after rinsing as described in claim 1, characterized in that, The determination of whether the contaminant type contains a mixture of salt and oil includes: Determine whether the spectral characteristic signal of the pollutant contains a first characteristic peak of salt and a second characteristic peak of oil; If the first characteristic peak and the second characteristic peak are included, then the contaminant type includes a mixture of salt and oil.

5. The method for evaluating the insulation performance of live equipment after rinsing as described in claim 2, characterized in that, After obtaining the first insulation performance evaluation score of the surface of the energized equipment, the method further includes: Determine whether the first insulation performance evaluation score is less than a first preset threshold; If the value is less than the first preset threshold, the distribution characteristics of chemical substances other than salt and oil are extracted from the spectral characteristic signals of the multiple pollutants using a feature extraction algorithm. The distribution characteristics of the chemical substances are classified using the support vector machine algorithm to obtain the types of chemical substances other than salt and oil in the spectral feature signals of the various pollutants; The concentration, distribution density, and distribution uniformity of various pollutants in the pollutant concentration distribution data are obtained respectively; the various pollutants include the chemical substances, the salt content, and the oil pollution; A second insulation performance evaluation score for the surface of the energized equipment is obtained based on the pollutant concentration, the distribution density, and the distribution uniformity.

6. The method for evaluating the insulation performance of energized equipment after rinsing as described in claim 5, characterized in that, After obtaining the pollutant concentration, distribution density, and distribution uniformity of various pollutants in the pollutant concentration distribution data, the method further includes: The interference intensity of multiple pollutants is obtained according to the preset interference level classification rules, and the interference frequency of multiple pollutants in the pollutant concentration distribution data is obtained. The step of obtaining the second insulation performance evaluation score of the surface of the energized equipment based on the pollutant concentration, the distribution density, and the distribution uniformity includes: A second insulation performance evaluation score for the surface of the energized equipment is obtained based on the pollutant concentration, distribution density, distribution uniformity, interference intensity, and interference frequency of the various pollutants.

7. The method for evaluating the insulation performance of live equipment after rinsing as described in claim 6, characterized in that, The method of obtaining a second insulation performance evaluation score for the surface of the energized equipment based on the pollutant concentration, distribution density, distribution uniformity, interference intensity, and interference frequency of the multiple pollutants includes: The pollutant type, pollutant concentration, distribution density, distribution uniformity, interference intensity, and interference frequency are integrated to obtain a preliminary feature vector; Obtain the environmental feature vector of the powered equipment, wherein the environmental feature vector includes vectors composed of the ambient humidity, ambient temperature, sampling time of image data, and sampling location of the powered equipment; The preliminary feature vector and the environmental feature vector are weighted to obtain a weighted feature matrix; Based on the value of the weighted feature matrix and the second preset insulation performance evaluation rule, the second insulation performance evaluation score of the surface of the live equipment is obtained.

8. A system for evaluating the insulation performance of live equipment after rinsing, characterized in that, The system includes: a feature signal acquisition module, a pollutant analysis module, a judgment module, a numerical acquisition module, and an evaluation module; The feature signal acquisition module is used to acquire image data of the surface of the electrical equipment after rinsing, and to collect spectral feature signals of various pollutants in the image data through a spectral analyzer. The pollutant analysis module is used to perform pollutant analysis on the spectral feature signals of the multiple pollutants to obtain pollutant type and pollutant concentration distribution data in the image data; The judgment module is used to determine whether the pollutant type contains a mixture of salt and oil. The numerical acquisition module is used to acquire, if the mixture of salt and oil is included, the first residual concentration average and the first concentration fluctuation coefficient of salt in the pollutant concentration distribution data, and the second residual concentration average and the second concentration fluctuation coefficient of oil. The evaluation module is used to obtain the insulation performance evaluation result of the surface of the electrical equipment based on the first residual concentration average, the first concentration fluctuation coefficient, the second residual concentration average, and the second concentration fluctuation coefficient.

9. A device for evaluating the insulation performance of live equipment after rinsing, comprising a memory and a processor, characterized in that, The memory stores a computer program that, when executed by the processor, causes the processor to perform the steps of the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it causes the processor to perform the steps of the method as described in any one of claims 1 to 7.

Images