Method and System for Feature Extraction of High Voltage Fuse Status Data

Abstract

This invention relates to the field of data processing technology and discloses a method and system for extracting features from the status data of high-voltage fuses. The method includes: real-time acquisition of operating current data, environmental data, and insulator leakage current data of the target object within a monitoring period; by acquiring the operating current data, environmental data, and insulator leakage current data, the medium environment stress factor is obtained through preprocessing and analysis, stress modulation current trajectory is generated, and singular fluctuation entropy and dispersion offset are extracted. The comprehensive degradation potential index and health status indicator are calculated collaboratively. This method can accurately capture scattered, random, and weak features such as slight fluctuations in leakage current and slow rate increases in local temperature, enabling the determination of early insulation degradation and contact deterioration of the target object, improving the accuracy of monitoring, and ensuring the stable operation of the target object in coastal salt spray and rainwater corrosion environments.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, specifically to a method and system for extracting features from the status data of high-voltage fuses. Background Technology

[0002] Currently, in the extraction of status data features of high-voltage fuses, traditional extraction methods such as time domain analysis, frequency domain analysis, and wavelet transform are usually used to process the collected current and temperature data, and the operating status of the high-voltage fuse is determined based on the data characteristics.

[0003] However, the above data processing method still has the following defects: In coastal areas, high-voltage fuses are affected by rainwater and salt spray corrosion for a long time, which leads to the degradation of their insulation layer performance and changes in their internal contact performance. These early weak state anomalies are only manifested as slight fluctuations in leakage current and slow rise in local temperature. The changes in these weak features often do not reach the threshold and are scattered and random, making it difficult for conventional feature extraction methods to capture these weak and scattered state anomalies and to identify the operating status of high-voltage fuses in a timely manner. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a method and system for extracting features from the status data of high-voltage fuses, thus solving the aforementioned problems.

[0005] The above-mentioned technical objective of the present invention is achieved through the following technical solution:

[0006] A method for extracting features from the condition data of high-voltage fuses, including:

[0007] Step S1: Real-time acquisition of operating current data, environmental data, and insulator leakage current data of the target object within the monitoring period;

[0008] Step S2: After preprocessing the environmental data and the insulator leakage current data respectively, the dynamic correlation between the two is analyzed to obtain the medium environment stress factor, which reflects the comprehensive electrochemical stress level of the insulating medium.

[0009] Step S3: Analyze the operating current data based on the medium environment stress factor to generate a stress-modulated current trajectory representing the current distortion caused by early insulation degradation and contact deterioration.

[0010] Step S4: Analyze the stress-modulated current trajectory, extract weak anomaly features, and generate singular fluctuation entropy that reflects the complexity of the trajectory's slight fluctuations and dispersed offset that reflects the trajectory's slight offset.

[0011] Step S5: Perform joint calculation of singular fluctuation entropy, dispersion offset, and medium environment stress factor to generate a health status identifier for determining the operating status of the target object.

[0012] Furthermore, after preprocessing the environmental data and insulator leakage current data respectively, and analyzing their dynamic correlation, a dielectric environmental stress factor reflecting the comprehensive electrochemical stress level of the insulating medium was obtained, including:

[0013] The environmental load index is generated by calculating the temperature and humidity in the environmental data.

[0014] Fluctuation analysis was performed on the leakage current data of the insulators to obtain the pollution response.

[0015] The dynamic correlation between the environmental load index and the fouling responsiveness was analyzed to obtain the electrochemical coupling coefficient, which reflects the closeness between the two.

[0016] Furthermore, after preprocessing the environmental data and insulator leakage current data respectively, and analyzing their dynamic correlation, a dielectric environmental stress factor reflecting the comprehensive electrochemical stress level of the insulating medium was obtained, which also includes:

[0017] The environmental load index is reconstructed based on the electrochemical coupling coefficient to generate a dynamic stress intensity sequence that reflects the real-time electrochemical stress intensity under wet pollution conditions.

[0018] Multi-scale analysis of dynamic stress intensity sequences is performed to generate a medium environment stress factor that reflects the comprehensive electrochemical stress level of the insulating medium under wet and polluted conditions.

[0019] Furthermore, based on the dielectric environment stress factor, the operating current data is analyzed to generate a stress-modulated current trajectory representing the current distortion caused by early insulation degradation and contact deterioration, including:

[0020] The operating current data is modulated based on the medium environment stress factor to generate a stress-modulated current sequence that reflects the dynamic loading effect of environmental stress.

[0021] Local similarity analysis of the stress-modulated current sequence yields a degraded excitation spectrum representing the excitation intensity of internal defects in the target object.

[0022] Furthermore, based on the dielectric environment stress factor, the operating current data is analyzed to generate a stress-modulated current trajectory representing the current distortion caused by early insulation degradation and contact deterioration. This also includes:

[0023] Based on the degradation excitation spectrum, resonance points related to insulation degradation and contact degradation are identified, and degradation resonance feature vectors are generated.

[0024] The degradation resonance eigenvectors are dynamically synthesized in the time domain to generate stress-modulated current trajectories that represent current distortions caused by early insulation degradation and contact deterioration.

[0025] Furthermore, the stress-modulated current trajectory is analyzed to extract weak anomaly features, generating singular fluctuation entropy reflecting the complexity of the trajectory's micro-amplitude fluctuations and dispersion offset reflecting the weak trajectory deviations, including:

[0026] The stress-modulated current trajectory is decomposed into multiple sub-trajectory components at different time scales to generate the trajectory eigenvalue matrix.

[0027] Based on the trajectory eigenvalue matrix, the state and transition probability of each sub-trajectory component are analyzed to generate a fluctuation mode tensor representing the fluctuation complexity.

[0028] Furthermore, the stress-modulated current trajectory is analyzed to extract weak anomaly features, generating singular fluctuation entropy reflecting the complexity of the trajectory's micro-amplitude fluctuations and dispersion offset reflecting the weak trajectory deviations. This also includes:

[0029] The degree of deviation of the stress-modulated current trajectory is analyzed, and a trajectory divergence field representing the overall and local deviation characteristics of the trajectory is generated.

[0030] The fluctuation mode tensor and the trajectory divergence field are calculated separately to generate the singular fluctuation entropy, which reflects the complexity of the trajectory's slight fluctuations, and the dispersion offset, which reflects the trajectory's slight offset.

[0031] Furthermore, singular fluctuation entropy, dispersion offset, and medium environment stress factor are jointly calculated to generate a health status indicator for determining the operating status of the target object, including:

[0032] Based on the medium environment stress factor, evaluate the contribution weights of singular fluctuation entropy and dispersion offset to the current environmental stress level, and generate a feature weight vector.

[0033] By synergistically fusing singular fluctuation entropy and dispersion offset based on the feature weight vector, a comprehensive degradation potential index representing the overall state of the target object is obtained.

[0034] Furthermore, the method involves co-calculating singular fluctuation entropy, dispersion offset, and medium environment stress factor to generate a health status indicator for determining the operating status of the target object. This also includes:

[0035] The comprehensive degradation potential index is analyzed to generate a health status indicator for determining the operating status of the target object.

[0036] Furthermore, a high-voltage fuse status data feature extraction system, applied to the above extraction method, includes:

[0037] The data acquisition unit is used to collect real-time operating current data, environmental data, and insulator leakage current data of the target object during the monitoring period.

[0038] The first data processing unit is used to preprocess environmental data and insulator leakage current data respectively, and analyze the dynamic correlation between the two to obtain the medium environment stress factor, which reflects the comprehensive electrochemical stress level of the environment on the insulating medium.

[0039] The second data processing unit is used to analyze the operating current data based on the medium environment stress factor and generate a stress-modulated current trajectory that represents the current distortion caused by early insulation degradation and contact deterioration.

[0040] The data feature extraction unit is used to analyze the stress-modulated current trajectory, extract weak anomaly features, and generate singular fluctuation entropy that reflects the complexity of the trajectory's slight fluctuations and dispersion offset that reflects the trajectory's slight deviation.

[0041] The status judgment unit is used to perform collaborative calculations on singular fluctuation entropy, dispersion offset, and medium environment stress factor to generate a health status identifier for determining the operating status of the target object.

[0042] In summary, the present invention has the following main beneficial effects:

[0043] Beneficial effects of the technical solution

[0044] The data acquisition unit collects real-time operating current data, environmental data, and insulator leakage current data of the target object during the monitoring period. Then, by analyzing the dynamic correlation between environmental data and insulator leakage current data, a medium environmental stress factor is generated, which includes environmental load index, pollution response degree, and electrochemical coupling coefficient. This factor accurately reflects the comprehensive electrochemical stress level of the environment on the insulating medium. Based on the medium environmental stress factor, the operating current data is modulated to generate stress-modulated current sequence, degradation excitation spectrum, and degradation resonance feature vector. The stress-modulated current trajectory is further synthesized, which can capture the current distortion caused by early insulation degradation and contact deterioration of the target object.

[0045] By decomposing the trajectory to generate the trajectory intrinsic matrix, fluctuation mode tensor, and trajectory divergence field, the singular fluctuation entropy reflecting the complexity of the trajectory's slight fluctuations and the dispersion offset reflecting the trajectory's slight deviation are extracted. Finally, the feature weight vector and comprehensive degradation potential index are calculated, and the health status identifier is obtained through iterative analysis. This scheme can accurately capture the dispersed and random weak features such as slight fluctuations in leakage current and slow rise in local temperature, realize the determination of early insulation degradation and contact deterioration of the target object, improve the accuracy of condition monitoring, and ensure the stable operation of the target object in the coastal salt spray and rainwater corrosion environment. Attached Figure Description

[0046] Figure 1 This is a flowchart illustrating the steps of the high-voltage fuse status data feature extraction method of the present invention;

[0047] Figure 2 This is a schematic diagram of the high-voltage fuse status data feature extraction system of the present invention. Detailed Implementation

[0048] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0049] refer to Figure 1 and Figure 2 A method for extracting features from high-voltage fuse status data, including:

[0050] Step S1: Real-time acquisition of operating current data, environmental data, and insulator leakage current data of the target object within the monitoring period. The monitoring period is 1 hour, and the sampling frequency is once every 1 minute. The target object is a high-voltage fuse. The monitoring period is set to 1 hour, the sampling frequency is set to 1 minute / time, and the early insulation degradation characteristic period is 30~60 minutes. 1 hour can completely cover the characteristic period. This monitoring period and sampling frequency can completely capture the slight changes in insulation and contact degradation of the high-voltage fuse, while avoiding data redundancy.

[0051] Environmental data includes temperature and humidity, which are the core environmental parameters affecting the insulation performance of insulators and are key factors that cause salt spray corrosion.

[0052] Step S2: After preprocessing the environmental data and the insulator leakage current data respectively, the dynamic correlation between the two is analyzed to obtain the medium environment stress factor, which reflects the comprehensive electrochemical stress level of the insulating medium.

[0053] Step S3: Analyze the operating current data based on the medium environment stress factor to generate a stress-modulated current trajectory representing the current distortion caused by early insulation degradation and contact deterioration.

[0054] Step S4: Analyze the stress-modulated current trajectory, extract weak anomaly features, and generate singular fluctuation entropy that reflects the complexity of the trajectory's slight fluctuations and dispersed offset that reflects the trajectory's slight offset.

[0055] Step S5: Perform joint calculation of singular fluctuation entropy, dispersion offset, and medium environment stress factor to generate a health status identifier for determining the operating status of the target object.

[0056] In one embodiment, after preprocessing the environmental data and the insulator leakage current data, and analyzing their dynamic correlation, a dielectric environmental stress factor reflecting the comprehensive electrochemical stress level of the insulating medium is obtained, including:

[0057] The environmental load index is generated by calculating the temperature and humidity in the environmental data. Specifically, the median of the temperature series within the monitoring period is used as the characteristic temperature, and the median of the humidity series within the monitoring period is used as the characteristic humidity. The median can eliminate interference from extreme temperature and humidity data, which meets the stable performance requirements of the environmental load. The difference between the current temperature and the characteristic temperature is divided by the range of the temperature series within the monitoring period to obtain the relative temperature offset of each sampling point. The relative humidity offset can be calculated in the same way as the temperature offset.

[0058] Calculate the absolute value of the temperature difference between the current temperature and the previous sampling point, and divide the absolute value by the average absolute value of the temperature change of all adjacent sampling points within the monitoring period to obtain the instantaneous temperature change intensity coefficient for each sampling point. The instantaneous humidity change intensity coefficient can be calculated in the same way as the temperature instantaneous change intensity coefficient. The instantaneous change intensity coefficient is mainly used to reflect the degree of abrupt change in temperature and humidity, because abrupt change is an important factor that accelerates insulation degradation.

[0059] Calculate the cross-correlation coefficient between the temperature series and the humidity series throughout the entire monitoring period. Multiply the relative temperature offset of the sampling point by the relative humidity offset, and then multiply by the absolute value of the cross-correlation coefficient to obtain the coupling term value for each sampling point.

[0060] The instantaneous comprehensive environmental stress value is obtained by summing the relative temperature offset, relative humidity offset, instantaneous temperature change intensity coefficient, instantaneous humidity change intensity coefficient, and coupling term value of each sampling point.

[0061] The environmental load index is obtained by calculating the square root of the sum of squares of all instantaneous comprehensive environmental stress values ​​during the entire monitoring period and dividing the square root by the square root of the total number of sampling points during the monitoring period. The environmental load index is mainly used to reflect the severity of the environment. The square root calculated here can quantify temperature and humidity parameters of different dimensions, and thus the environmental load index can be used to reflect the severity of the environment.

[0062] Fluctuation analysis of insulator leakage current data is performed to obtain pollution response, specifically including: preprocessing the insulator leakage current data collected during the monitoring period and arranging the preprocessed leakage current into a leakage current sequence;

[0063] Calculate the arithmetic mean of all positive data points in the leakage current sequence to obtain the positive half-cycle characteristic current; calculate the arithmetic mean of all negative data points in the leakage current sequence to obtain the negative half-cycle characteristic current; divide the value of each data point in the leakage current sequence by its corresponding half-cycle characteristic current to obtain the normalized leakage current sequence for the entire cycle.

[0064] The range of data from every five consecutive sampling points in the normalized leakage current sequence is calculated, and the ranges are arranged to generate the first subsequence. Five sampling points are selected because rapid fluctuations refer to short-term leakage current jumps of 1 to 5 minutes. These jumps correspond to rapid physical processes such as instantaneous wetting of the insulator surface, instantaneous dissolution of salt particles, and local micro-discharge. Five sampling points can cover the shortest rapid fluctuation period and can fully extract the maximum change amplitude of a single rapid fluctuation without introducing long-term trends.

[0065] The first subsequence mainly reflects the rapid fluctuation characteristics of the current. The standard deviation of the data of every 20 consecutive sampling points in the normalized leakage current sequence is calculated, and multiple standard deviations are arranged to form the second subsequence. The reason for selecting 20 consecutive sampling points is that the slow fluctuation of the leakage current in this application is caused by the gradual change of ambient temperature and humidity and the slow accumulation of contamination on the surface of the insulator. The change period is usually 15 to 20 minutes. Under the condition of 1 minute sampling, 20 sampling points correspond to 20 minutes, which is perfectly matched with the time scale of the slow fluctuation of the leakage current. It can stably reflect the dispersion of the current over a continuous period of time, thereby accurately characterizing the slow current change characteristics caused by the slow development of contamination. The second subsequence mainly reflects the slow fluctuation characteristics of the current.

[0066] The arithmetic mean of all values ​​in the first subsequence is calculated as the intensity of rapid fluctuations, and the arithmetic mean of all values ​​in the second subsequence is calculated as the intensity of slow fluctuations.

[0067] The response factor is obtained by dividing the rapid fluctuation intensity by the slow fluctuation intensity and then multiplying it by the ratio of the peak value of the entire leakage current sequence to the characteristic current of the positive half-cycle. The peak value is the maximum absolute value of each data point. The response factor is then multiplied by the environmental load index to obtain the pollution response degree. The pollution response degree is used to reflect the micro-fluctuation characteristics of the leakage current of the insulator and to determine the degree of degradation of the insulation layer by salt spray and rainwater corrosion.

[0068] The dynamic correlation between environmental load index and pollution response is analyzed to obtain the electrochemical coupling coefficient reflecting the closeness between the two. Specifically, this includes: obtaining the environmental load index of the previous eight historical monitoring cycles of the current period and arranging them in chronological order to form the first historical sequence; simultaneously, the pollution response of the previous eight historical monitoring cycles of the current period and arranging them in chronological order to form the second historical sequence; calculating the arithmetic mean of all elements in the first historical sequence to obtain the environmental load benchmark; and calculating the arithmetic mean of all elements in the second historical sequence to obtain the pollution response benchmark. The first eight historical monitoring cycles are selected because the single monitoring cycle in this application is 1 hour, while in the coastal salt spray environment, the impact of environmental load on insulator pollution has a significant lag: after changes in ambient temperature and humidity and salt spray concentration, the pollution response of leakage current will not be immediately synchronized, but will be reflected gradually after a lag of several hours. The lag effect is mainly concentrated in the most recent 8 hours. The 8 monitoring cycles can not only fully cover the effective lag time from environmental stress to pollution response, but also avoid introducing too much historical data to cause interference from outdated information in the correlation calculation.

[0069] Calculate the absolute value of the difference between the current environmental load index and the environmental load benchmark to obtain the environmental load offset; calculate the absolute value of the difference between the current fouling response and the fouling response benchmark to obtain the fouling response offset.

[0070] Multiply the first historical sequence and the second historical sequence element by element to obtain the product sequence; calculate the arithmetic mean of all elements in the product sequence, and then divide it by the product of the environmental load benchmark and the pollution response benchmark to obtain the historical correlation factor;

[0071] The dynamic synergy degree can be obtained by multiplying the current environmental load offset by the current fouling response offset and dividing the product by the baseline synergy value obtained by multiplying the standard deviation of the first historical sequence by the standard deviation of the second historical sequence. Multiplying the dynamic synergy degree by the historical correlation factor yields the electrochemical coupling coefficient, which reflects the closeness between the two.

[0072] In one embodiment, after preprocessing the environmental data and the insulator leakage current data, and analyzing their dynamic correlation, a dielectric environmental stress factor reflecting the comprehensive electrochemical stress level of the insulating medium is obtained, which further includes:

[0073] The environmental load index is reconstructed based on the electrochemical coupling coefficient to generate a dynamic stress intensity sequence reflecting the real-time electrochemical stress intensity under wet and polluted conditions. Specifically, this includes: multiplying the electrochemical coupling coefficient by the environmental load index to obtain the comprehensive environmental impact factor; for each sampling point, calculating the absolute value of the difference between the temperature at that sampling point and the median of the temperature sequence over the entire monitoring period, and then dividing it by the range of the temperature sequence over the monitoring period to obtain the temperature significance ratio at that sampling point; the humidity significance ratio can be calculated using the same method as the temperature significance ratio.

[0074] Add the temperature significance ratio and the humidity significance ratio, and then multiply by the sum of the instantaneous temperature change intensity coefficient and the instantaneous humidity change intensity coefficient corresponding to the sampling point to obtain the dynamic environmental stress value of each sampling point; multiply the dynamic environmental stress value of each sampling point by the comprehensive environmental impact factor to obtain the primary real-time stress intensity.

[0075] The average value of all primary real-time stress intensity values ​​within the monitoring period is calculated. The primary real-time stress intensity of each sampling point is divided by the average value to obtain the dynamic stress intensity. The dynamic stress intensity is arranged by time to form a dynamic stress intensity sequence that reflects the real-time electrochemical stress intensity under wet pollution conditions.

[0076] Multi-scale analysis of the dynamic stress intensity sequence is performed to generate a medium environment stress factor that reflects the comprehensive electrochemical stress level of the insulating medium under wet pollution. Specifically, this includes: calculating the arithmetic mean and median of the dynamic stress intensity sequence, taking the larger value as the first characteristic value and the smaller value as the second characteristic value; dividing the difference between the two characteristic values ​​by the standard deviation of the dynamic stress intensity sequence and rounding the result up to obtain the number of scales M. If M is less than 3, it is set to 3; otherwise, the original value is retained. The scale division method is determined according to the degree of data dispersion, and 3 is the minimum effective analysis scale. According to the multi-scale entropy analysis standard, fewer than 3 scales may not be able to complete effective multi-scale decomposition.

[0077] The dynamic stress intensity sequence is arranged in ascending order of numerical value to form an ascending sequence. Based on the number of scale divisions M, the percentile division points of this ascending sequence are calculated: the 0th, 100th percentile divided by M, 200th percentile divided by M, ..., 100th percentile. Based on these division points, the ascending sequence is precisely divided into M consecutive numerical intervals. The data points in each interval constitute a subsequence. The percentile division method used here is to uniformly divide the dynamic stress intensity sequence according to the stress intensity, ensuring that each subsequence contains data of similar quantity but different intensity levels, thereby achieving uniform analysis under multiple scales.

[0078] For each subsequence, its kurtosis is calculated to represent the sharp or flat shape of stress fluctuations within that interval. Kurtosis can reflect the distribution pattern of stress fluctuations because the kurtosis of fluctuations will be higher under deterioration conditions. The arithmetic mean of the kurtosis of all subsequences is calculated to obtain the basic morphological index.

[0079] Calculate the variance of each subsequence and the variance of the dynamic stress intensity sequence; for each subsequence, calculate the proportion of its variance to the variance of the entire dynamic stress intensity sequence to obtain a set of proportion values; calculate the standard deviation of this set of proportion values, denoted as the weighting adjustment factor.

[0080] After normalizing the number of scale divisions, basic morphological indicators, weight adjustment factors, and electrochemical coupling coefficients to the 0-1 interval, multiply them together, and then multiply by the coefficient of variation of the dynamic stress intensity sequence. After normalizing the calculation results to the 0-1 interval, the medium environment stress factor reflecting the comprehensive electrochemical stress level of the insulating medium in a wet and polluted environment can be obtained.

[0081] By preprocessing environmental data and insulator leakage current data, the environmental load index, pollution response, electrochemical coupling coefficient, and dielectric environmental stress factor can be accurately calculated. This can effectively capture scattered and random weak characteristics such as slight fluctuations in leakage current and slow rises in local temperature, accurately reflecting the comprehensive electrochemical stress level of the environment on the insulating medium and the degree of insulation layer degradation. This enables timely identification of early weak state anomalies of high-voltage fuses and improves the accuracy of condition monitoring.

[0082] In one embodiment, the operating current data is analyzed based on the dielectric environment stress factor to generate a stress-modulated current trajectory representing the current distortion caused by early insulation degradation and contact deterioration, including:

[0083] The operating current data is modulated according to the medium environment stress factor to generate a stress-modulated current sequence that reflects the dynamic loading effect of environmental stress. Specifically, the operating current in the operating current data is arranged in order to form an operating current sequence, the skewness and kurtosis of all data points in the operating current sequence are calculated, and the absolute value of the skewness and the kurtosis are added together to obtain the shape coefficient.

[0084] The number of slope sign changes for each local segment consisting of three consecutive adjacent data points in the operating current sequence is calculated. The number of slope sign changes is the total number of times the product of the previous pair of differences and the next pair of differences is negative. The total number of changes is divided by the total number of local segments to obtain the local oscillation frequency. Three sampling points are selected because at least three points are needed to determine the trend change of a signal, such as rising, falling, or turning point, in order to determine whether a trend reversal has occurred. The basic modulation factor is obtained by multiplying the medium environment stress factor by the shape coefficient and then dividing by the sum of the local oscillation frequency and 1.

[0085] For each data point in the operating current sequence, calculate the difference between it and the median of the operating current sequence, divide the difference by the average of the absolute values ​​of the differences between all data points in the operating current sequence and the median to obtain the position offset of the data point, and calculate the standard deviation of the operating current corresponding to the five data points formed by the data point and the two data points before and after it to obtain the local standard deviation.

[0086] Calculate the average of all local standard deviations for the entire operating current sequence, and divide the local standard deviation of each data point by this average to obtain the normalized local fluctuation coefficient.

[0087] Multiply the basic modulation factor, position offset, and normalized local fluctuation coefficient, and then add 1 to obtain the composite modulation coefficient of the data point. Multiply the operating current of the data point by the composite modulation coefficient to obtain the stress modulation current of the data point. Connect the stress modulation currents of all data points in sequence to form a stress modulation current sequence that reflects the dynamic loading effect of environmental stress.

[0088] Local similarity analysis is performed on the stress-modulated current sequence to obtain the deteriorated excitation spectrum representing the excitation intensity of internal defects in the target object. Specifically, this includes: calculating the sum of the number of all local maxima and local minima in the stress-modulated current sequence, using this value as the initial length of the reference window, extracting a subsequence of the initial length from the starting point of the stress-modulated current sequence, calculating the ratio of the standard deviation of this subsequence to the standard deviation of the stress-modulated current sequence, and if the ratio is > 0.5, shifting this subsequence forward by one sampling point. This process is repeated until a subsequence with a standard deviation ratio ≤ 0.5 is found, and this subsequence is determined as the feature reference segment.

[0089] Centered on the current sampling point, a continuous subsequence is extracted from the stress-modulated current sequence as a neighborhood window. The length of the neighborhood window is exactly the same as the length of the feature reference segment. The cross-correlation coefficient between the data in the neighborhood window and the feature reference segment is calculated and denoted as the first similarity. Then, the cross-correlation coefficients between the first half of the neighborhood window and the first half of the feature reference segment, and the cross-correlation coefficients between the second half of the neighborhood window and the second half of the feature reference segment are calculated respectively, and the smaller value is taken as the second similarity.

[0090] Subtracting the first similarity from 1 yields the overall mismatch; subtracting the second similarity from 1 yields the structural distortion; the absolute value of kurtosis in the neighborhood window data is calculated, and the overall mismatch, structural distortion, and absolute kurtosis are multiplied together to generate the instantaneous degradation excitation value for that sampling point. Arranging the instantaneous degradation excitation values ​​of all sampling points in chronological order yields the degradation excitation spectrum representing the intensity of internal defect excitation in the target object.

[0091] In one embodiment, the analysis of operating current data based on the dielectric environment stress factor generates a stress-modulated current trajectory representing the current distortion caused by early insulation degradation and contact deterioration, further comprising:

[0092] Based on the deterioration excitation spectrum, resonance points related to insulation degradation and contact degradation are identified, and a deterioration resonance feature vector is generated. Specifically, for a deterioration excitation spectrum of length N, its autocorrelation coefficients at different hysteresis points k are calculated to obtain the autocorrelation coefficients corresponding to each k, where k is from 0 to N-1; these autocorrelation coefficients are arranged in order of k from 0 to N-1, and the resulting sequence is the autocorrelation sequence.

[0093] In the autocorrelation sequence, identify all local maxima except for the zero lag point, and arrange these local maxima in descending order to form a local maxima sequence.

[0094] The absolute value of the skewness of the stress-modulated current sequence is calculated, divided by its shape coefficient, and then multiplied by the dielectric environment stress factor to obtain the insulation degradation correlation index, which reflects the correlation between the resonance point and the degradation process. The absolute value of the skewness of the degradation excitation spectrum is then calculated and added to the insulation degradation correlation index to obtain the comprehensive selection coefficient.

[0095] If the overall selection coefficient is greater than 1, then the first three points in the local maximum sequence are selected as resonance points; otherwise, the first two points are selected as resonance points.

[0096] For each resonance point, its autocorrelation coefficient is multiplied by the variance of the degraded excitation spectrum, and then divided by the number of hysteresis points k, where k must be greater than 0, to obtain the resonance intensity of that resonance point.

[0097] All calculated resonance intensities are sorted in ascending order of their corresponding hysteresis points to form a one-dimensional array. This array is used as the degraded resonance feature vector, which is used to amplify the intensity of weak and anomalous resonance signals.

[0098] The degradation resonance feature vector is dynamically trajectory synthesized in the time domain to generate a stress-modulated current trajectory representing the current distortion caused by early insulation degradation and contact degradation. Specifically, the degradation excitation spectrum is delayed and aligned based on the resonance intensity and the corresponding number of hysteresis points contained in the degradation resonance feature vector. For each resonance point, the degradation excitation spectrum is shifted backward by k sampling points corresponding to the number of hysteresis points. The missing data at the front end of the degraded excitation spectrum after shifting is filled with the value of the first sampling point of the degradation excitation spectrum, thereby obtaining the corresponding delay sequence.

[0099] Calculate the cross-correlation coefficient between each delayed sequence and the degraded excitation spectrum to obtain the temporal coupling coefficient; multiply each delayed sequence by its corresponding resonance intensity, and then multiply by the temporal coupling coefficient to obtain a set of resonance modulation sequences;

[0100] Summing the values ​​of all resonant modulation sequences at the same time point generates a resonant background sequence; calculating the arithmetic mean of all values ​​in the resonant background sequence, subtracting this mean from each value in the resonant background sequence, yields an intermediate sequence; multiplying each value in the intermediate sequence by (the standard deviation of the stress modulation current sequence divided by the standard deviation of the intermediate sequence), thereby generating a normalized resonant background sequence with the same standard deviation as the stress modulation current sequence.

[0101] Then, the absolute value of the cross-correlation coefficient between the stress-modulated current sequence and the normalized resonant background sequence is calculated to obtain the dynamic fusion weight.

[0102] For each sampling point, the value of the stress-modulated current sequence at that sampling point is multiplied by the value of the normalized resonant background sequence at that sampling point by the dynamic fusion weight and then added to obtain the trajectory value of that sampling point. By connecting the trajectory values ​​of all sampling points in chronological order, the stress-modulated current trajectory representing the current distortion caused by early insulation degradation and contact deterioration is obtained.

[0103] By modulating the operating current data using the medium environment stress factor, a stress-modulated current sequence is generated. This sequence is then combined with the degradation excitation spectrum and degradation resonance feature vector to synthesize the stress-modulated current trajectory. This amplifies the signal of weak abnormal resonance, accurately identifies current distortion caused by early insulation decay and contact degradation, effectively captures scattered and random weak features such as slight fluctuations in leakage current and slow rise in local temperature, accurately reflects insulation layer degradation and changes in internal contact performance, and enables timely identification of early weak abnormal conditions of high-voltage fuses, ensuring their stable operation in coastal salt spray and rainwater corrosion environments.

[0104] In one embodiment, the stress-modulated current trajectory is analyzed to extract weak anomaly features, generating singular fluctuation entropy reflecting the complexity of the trajectory's micro-amplitude fluctuations and dispersion offset reflecting the weak trajectory deviation, including:

[0105] The stress-modulated current trajectory is decomposed into multiple sub-trajectory components at different time scales to generate the trajectory eigenvalue matrix. Specifically, this includes: identifying all local maxima and local minima in the stress-modulated current trajectory, calculating the time interval between every two adjacent extreme points, and then calculating the arithmetic mean of all these time intervals to obtain the base time scale.

[0106] Calculate the absolute value of the skewness and its kurtosis of the stress-modulated current trajectory, add these two values ​​together and round up to obtain an integer, which is the reference number; if the reference number is greater than 5, then set it to five; otherwise, keep the value of the reference number.

[0107] Construct the same number of analytical scales as the reference scale. The time window length corresponding to the first scale is the base time scale; the time window length corresponding to the second scale is equal to the base time scale multiplied by 2; the time window length corresponding to each subsequent scale is the base time scale multiplied by the number of scales it corresponds to, for example, the third scale is multiplied by 3.

[0108] For each scale, perform the following operations: Using the time window length corresponding to the scale, divide the stress-modulated current trajectory into non-overlapping segments starting from the beginning to form data segments. Discard any data segments whose last segment is less than a complete window. Calculate the arithmetic mean of all data points within each data segment and arrange these means in chronological order to form a low-frequency profile sub-trajectory component. Simultaneously, calculate the standard deviation of all data points within each data segment and arrange these standard deviations in the same order to form a high-frequency fluctuation sub-trajectory component. Use linear interpolation to restore the lengths of these two sub-trajectory components to be the same as the stress-modulated current trajectory.

[0109] The low-frequency profile sub-trajectory components and high-frequency fluctuation sub-trajectory components at all scales are combined to form a preliminary sub-trajectory component set; the cross-correlation coefficient of the stress-modulated current trajectory and each component in the preliminary sub-trajectory component set is calculated; the arithmetic mean of the absolute values ​​of all these cross-correlation coefficients is calculated, and the components whose absolute values ​​of the cross-correlation coefficients with the stress-modulated current trajectory are greater than this arithmetic mean are retained to form the sub-trajectory component set.

[0110] All sub-trajectory components in the sub-trajectory component set are arranged in ascending order of their corresponding analysis scale window lengths. Low-frequency contour sub-trajectory components belonging to the same scale are arranged before high-frequency fluctuation sub-trajectory components. Each sub-trajectory component is treated as a row vector, and all row vectors are stacked vertically from top to bottom to form a two-dimensional matrix. This matrix is ​​the trajectory intrinsic matrix, which is mainly used to remove noise components and prevent noise from masking weak features.

[0111] Based on the trajectory intrinsic matrix, the state and transition probability of each sub-trajectory component are analyzed to generate a fluctuation mode tensor representing the fluctuation complexity. Specifically, this includes: merging the values ​​of all sub-trajectory components in the trajectory intrinsic matrix to obtain a merged dataset; calculating the 25th, 50th, and 75th percentiles of the merged dataset; assigning values ​​less than the 25th percentile to state 1, values ​​between the 25th and 75th percentiles to state 2, and values ​​greater than the 75th percentile to state 3.

[0112] Calculate the standard deviation of each sub-trajectory component and use half of the standard deviation as its threshold;

[0113] Starting from the first data point of each sub-trajectory component, each subsequent data point is checked in turn: if the absolute value of the difference between its value and the previous data point is greater than the threshold of the sub-trajectory component, then the value falls into whichever interval it falls into and is assigned to the corresponding state. For example, if it is less than the 25th percentile, it is assigned to state 1; otherwise, the state of the previous point remains unchanged, and a state sequence of the same length as the sub-trajectory component is generated.

[0114] Traverse the state sequence. For each state except the last one, combine it with the next state to form a state transition pair, such as transitioning from state 1 to state 2. Count the number of times each specific transition (e.g., from state 2 to state 3) occurs in all state transition pairs. Record the counts of all state transition pairs in a 3x3 transition counting matrix, where the row index represents the state before the transition (i.e., the previous state) and the column index represents the state after the transition (i.e., the next state).

[0115] Divide all elements in each row of the transition count matrix by the sum of the elements in that row to convert the absolute count into transition probabilities, thus obtaining the transition probability matrix. Then, arrange the transition probability matrices corresponding to all sub-trajectory components from top to bottom according to their respective scales to form a three-dimensional fluctuation mode tensor representing the complexity of fluctuations.

[0116] In one embodiment, the stress-modulated current trajectory is analyzed to extract weak anomaly features, generating singular fluctuation entropy reflecting the complexity of the trajectory's micro-amplitude fluctuations and dispersion offset reflecting the weak trajectory deviation. The analysis also includes:

[0117] The degree of deviation of the stress-modulated current trajectory is analyzed, and a trajectory divergence field representing the overall and local deviation characteristics of the trajectory is generated. Specifically, this includes: calculating the number of times the sign of the difference between the values ​​of all adjacent data points in the stress-modulated current trajectory changes, that is, the total number of times the difference between the next point and the previous point changes from positive to negative or from negative to positive; calculating the arithmetic square root of the total number of times; and rounding the result up. The integer obtained is the standard length of the local analysis segment.

[0118] For each sampling point on the stress-modulated current trajectory, take several points before and after it to form a local segment, such that the total number of points in the local segment is equal to the standard length. However, for sampling points at the beginning or end of the stress-modulated current trajectory, since the number of adjacent points on one side is insufficient, the total length of the local segment is no longer required to reach the standard length. Instead, all available adjacent points on that side are included together with the sampling point and the adjacent points on the other side to form a local segment with a length less than the standard length.

[0119] Calculate the standard deviation of the numerical differences between all adjacent data points within a local segment to obtain the local variation dispersion, which reflects the degree of dispersion of changes within the local segment; connect the first and last data points of the local segment to form a virtual chord, calculate the sum of the absolute values ​​of the vertical distances from all data points within the local segment to this chord, and then divide by the length of the projection of this chord in the horizontal direction to obtain the local curvature, which reflects the degree of curvature of the local segment relative to the line connecting its first and last ends.

[0120] Multiply the local variation dispersion by the local curvature and then divide by the sum of the two to obtain the intrinsic offset coefficient of the sampling point; recalculate the intrinsic offset coefficient with a window of twice the standard length in the same way as the intrinsic offset coefficient to obtain the secondary intrinsic offset coefficient; calculate the absolute value of the difference between the secondary intrinsic offset coefficient and the intrinsic offset coefficient, and use the reciprocal of the absolute value as the multi-scale consistency weight of the sampling point.

[0121] For each sampling point, its intrinsic offset coefficient is multiplied by the multi-scale consistency weight to obtain the overall divergence value of that point. The overall divergence values ​​of all sampling points are arranged in chronological order as the first row sequence, and the local variation dispersion of all sampling points is arranged in chronological order as the second row sequence. The two row sequences are combined to form a two-dimensional matrix, which is the trajectory divergence field representing the overall and local offset characteristics of the trajectory.

[0122] The fluctuation mode tensor and trajectory divergence field are calculated separately to generate singular fluctuation entropy, which reflects the complexity of small trajectory fluctuations, and dispersion offset, which reflects the slight trajectory offset. Specifically, this includes: calculating the transition probability matrix corresponding to each sub-trajectory component in the fluctuation mode tensor, calculating the arithmetic mean of all its off-diagonal elements to obtain the average transition activity of the sub-trajectory component; calculating the standard deviation of the average transition activity of all sub-trajectory components, and then dividing it by the range of the average transition activity of all sub-trajectory components to obtain the mode divergence ratio.

[0123] For the transition probability matrix of each sub-trajectory component, find the maximum value in each row, calculate the standard deviation of these maximum values ​​in each row, and calculate the arithmetic mean of the standard deviations of all sub-trajectory components to obtain the structure-specific benchmark.

[0124] Multiplying the mode divergence ratio by the structure-specific benchmark, then by the total number of sub-trajectory components, and finally by the absolute value of the kurtosis of the stress-modulated current trajectory sequence, we can obtain the singular fluctuation entropy, which reflects the complexity of the trajectory's micro-amplitude fluctuations.

[0125] Separate the comprehensive divergence value sequence of the first row and the local variation dispersion sequence of the second row from the two-dimensional trajectory divergence field, and calculate the cross-correlation coefficient between the two sequences over the entire monitoring period.

[0126] Arrange all values ​​in the comprehensive divergence value sequence in ascending order, calculate the difference between its 75th percentile and 25th percentile, and then divide it by the range of all values ​​in the comprehensive divergence value sequence to obtain the divergence concentration.

[0127] In the sequence of comprehensive divergence values, identify all peak points that are greater than the arithmetic mean of the sequence of comprehensive divergence values, calculate the standard deviation of the time interval between every two adjacent peak points, and divide the standard deviation by the arithmetic mean of these time intervals to obtain the peak divergence coefficient.

[0128] Multiplying the absolute value of the cross-correlation coefficient, the divergence concentration, and the peak divergence coefficient, and then multiplying by the electrochemical coupling coefficient, yields the dispersion offset, which reflects the slight trajectory shift. The dispersion offset is derived from the statistical characteristics of the offset data and directly reflects the slight trajectory shift caused by degradation.

[0129] By decomposing the stress-modulated current trajectory to generate the trajectory intrinsic matrix and eliminating noise, and combining the fluctuation mode tensor and trajectory divergence field to extract singular fluctuation entropy and dispersion offset, we can accurately capture the weak, dispersed, and random characteristics such as the slight fluctuation of leakage current and the slow rate of local temperature rise. This clearly reflects the complexity of the trajectory's slight fluctuations and the slight offset, accurately identifies the current distortion caused by early insulation degradation and contact deterioration, and can promptly detect abnormal conditions of high-voltage fuses, ensuring the accuracy of data analysis and the timeliness of monitoring.

[0130] In one embodiment, the singular fluctuation entropy, dispersion offset, and medium environment stress factor are jointly calculated to generate a health status identifier for determining the operating status of the target object, including:

[0131] Based on the medium environment stress factor, the contribution weights of singular fluctuation entropy and dispersion offset to the current environmental stress level are evaluated, and a feature weight vector is generated. Specifically, the singular fluctuation entropy is multiplied by the dispersion offset, and then divided by the sum of the singular fluctuation entropy and the dispersion offset to obtain the balance factor.

[0132] Multiply the medium environment stress factor by the singular fluctuation entropy and the dispersion offset respectively to obtain two product values; calculate the square root of the singular fluctuation entropy and the square root of the dispersion offset respectively; divide the first product value by the square root of the singular fluctuation entropy to obtain the absolute change intensity of the singular fluctuation entropy; divide the second product value by the square root of the dispersion offset to obtain the absolute change intensity of the dispersion offset.

[0133] Calculate the absolute value of the difference between the medium environment stress factor and 1, multiply the absolute value of the difference by the balance factor, and then add it to 1 to obtain the environment modulation coefficient; multiply the absolute change intensity of the singular fluctuation entropy by the environment modulation coefficient to obtain the modulation intensity of the singular fluctuation entropy; then divide the absolute change intensity of the dispersion offset by the environment modulation coefficient to obtain the modulation intensity of the dispersion offset.

[0134] Calculate the arithmetic sum of the two modulation intensities, divide the modulation intensity of the singular fluctuation entropy by the arithmetic sum to obtain the contribution weight of the singular fluctuation entropy; divide the modulation intensity of the dispersion offset by the arithmetic sum to obtain the contribution weight of the dispersion offset; combine the two contribution weights in sequence into a two-dimensional vector, i.e., the feature weight vector.

[0135] Based on the feature weight vector, singular fluctuation entropy and dispersion offset are synergistically fused to obtain a comprehensive degradation potential index representing the overall state of the target object. Specifically, singular fluctuation entropy and dispersion offset are used as two elements to form a two-dimensional column vector, i.e., the initial feature vector.

[0136] Calculate the square root of the medium environment stress factor, then calculate the ratio of singular fluctuation entropy to dispersion offset, and use the product of the ratio and the square root of the medium environment stress factor as the first transformation value; calculate the reciprocal of the medium environment stress factor, then calculate the ratio of dispersion offset to singular fluctuation entropy, and use the product of the ratio and the reciprocal of the environment stress factor as the second transformation value.

[0137] Construct a dynamic two-dimensional transformation matrix. The main diagonal elements of the two-dimensional transformation matrix are the contribution weights of the singular fluctuation entropy and the contribution weights of the dispersion offset, respectively. The off-diagonal elements of the two-dimensional transformation matrix are the first transformation value and the second transformation value, respectively.

[0138] Multiply the initial feature vector by the two-dimensional transformation matrix to obtain a new two-dimensional vector after collaborative fusion; calculate the square root of the product of the two elements of the new two-dimensional vector, then calculate the sum of the two elements of the vector, and divide the square root value by the sum of the elements to obtain the preliminary aggregation coefficient.

[0139] Multiplying the initial aggregation coefficient by the medium environment stress factor, and then by the absolute value of the difference between the two elements of the new two-dimensional vector, yields the comprehensive degradation potential index representing the overall state of the target object.

[0140] In one embodiment, the method of jointly calculating the singular fluctuation entropy, dispersion offset, and medium environment stress factor to generate a health status indicator for determining the operating status of the target object further includes:

[0141] The comprehensive degradation potential index is analyzed to generate a health status indicator for judging the operating status of the target object. Specifically, the comprehensive degradation potential index is used as the initial iteration value and three iterations are performed. The rule for each iteration is: multiply the current iteration value by the medium environment stress factor, and then calculate the arctangent function of the product value. The resulting radian value is used as the input value for the next iteration. The radian values ​​obtained from the three iterations are recorded and arranged in the iteration order to form a feature sequence.

[0142] Calculate the absolute value of the difference between the first and third values ​​in the feature sequence to obtain the stride length; calculate the standard deviation of the differences between adjacent values ​​in the feature sequence to obtain the gait divergence.

[0143] Dividing stride length by gait divergence yields a preliminary ratio. Dividing this preliminary ratio by the absolute value of the comprehensive deterioration potential index yields a decision characteristic value representing the convergence and anomalous changes in the target object's behavior.

[0144] If the decision feature value is less than or equal to 1, the health status identifier is set to 1, indicating that the target object is in a normal state; otherwise, the health status identifier is set to 0, indicating that the target object is in an abnormal state.

[0145] By combining the fluctuation mode tensor and the trajectory divergence field to extract singular fluctuation entropy and dispersion offset, and then co-calculating the three to generate a feature weight vector and a comprehensive degradation potential index, a health status identifier is obtained through iterative analysis. This can accurately capture weak, dispersed, and random features such as slight fluctuations in leakage current and slow increases in local temperature, enabling the determination of early insulation degradation and contact deterioration, and accurately identifying the operating status of high-voltage fuses.

[0146] In one embodiment, the high-voltage fuse status data feature extraction system is applied to the above-described extraction method, including:

[0147] The data acquisition unit is used to collect real-time operating current data, environmental data, and insulator leakage current data of the target object during the monitoring period.

[0148] The first data processing unit is used to preprocess environmental data and insulator leakage current data respectively, and analyze the dynamic correlation between the two to obtain the medium environment stress factor, which reflects the comprehensive electrochemical stress level of the environment on the insulating medium.

[0149] The second data processing unit is used to analyze the operating current data based on the medium environment stress factor and generate a stress-modulated current trajectory that represents the current distortion caused by early insulation degradation and contact deterioration.

[0150] The data feature extraction unit is used to analyze the stress-modulated current trajectory, extract weak anomaly features, and generate singular fluctuation entropy that reflects the complexity of the trajectory's slight fluctuations and dispersion offset that reflects the trajectory's slight deviation.

[0151] The status judgment unit is used to perform collaborative calculations on singular fluctuation entropy, dispersion offset, and medium environment stress factor to generate a health status identifier for determining the operating status of the target object.

[0152] In one aspect of this embodiment, in order to verify that the method of the present invention can accurately capture early weak, scattered, and random abnormal features and solve the problem that traditional methods cannot identify early insulation degradation and contact deterioration features such as slight fluctuations in leakage current and slow rise in local temperature under coastal salt spray and rainwater corrosion environments, a comparative experiment is used for verification.

[0153] I. Experimental Conditions and Subjects

[0154] The target of implementation is a 10kV outdoor high-voltage fuse, simulating a coastal salt spray environment with a temperature of 20℃~30℃, a relative humidity of 75%~98%, and a salt spray concentration of 0.05mg / cm³.

[0155] The monitoring cycle is 1 hour, and the sampling frequency is once every 1 minute;

[0156] II. Comparison Methods

[0157] The present invention group uses steps S1 to S5 of this application to determine the operating status of the high-voltage fuse;

[0158] Control group 1: The operating status of high-voltage fuses was determined using the traditional threshold method;

[0159] Control group 2: The operating status of high-voltage fuses was determined using the traditional frequency domain FFT analysis method;

[0160] Among them, 30 high-voltage fuses with the same health status were selected and continuously operated in a simulated coastal salt spray environment to accelerate the early slight degradation; the status was monitored simultaneously using the method of this invention, the traditional time-domain threshold method, and the traditional frequency-domain FFT method.

[0161] Data was recorded every 2 hours and continued until the fuse showed significant deterioration, resulting in 150 valid samples.

[0162] III. Evaluation Indicators

[0163] The weak feature recognition rate is the proportion of the total number of tests used to capture weak anomalies such as slight fluctuations in leakage current and slow rise in local temperature.

[0164] The false alarm rate reflects the proportion of tests that were judged as abnormal but showed no deterioration out of the total number of tests.

[0165] Early warning lead time is the length of time that is given in advance compared to the occurrence of obvious failures;

[0166] Table 1 below shows a comparison of the effectiveness of different methods in identifying weak anomalies.

[0167] ;

[0168] Table 2 below mainly shows the comparison of recognition performance in typical weak scenes.

[0169] ;

[0170] IV. Comparison with Implementation Conclusions

[0171] The method of this invention has a weak feature recognition rate of 92.7%, which is much higher than that of traditional time-domain and frequency-domain methods. It can stably capture early weak anomalies such as slight fluctuations in leakage current and slow temperature rise.

[0172] The method of this invention is designed to identify weak degradation features that are scattered, random, and irregular, and has a strong identification capability, thus overcoming the limitation of traditional methods that can only identify regular large-amplitude signals.

[0173] The method of this invention can achieve an average early warning of 18.5 hours, providing sufficient time for safe operation and maintenance of high-voltage fuses in coastal salt spray environments;

[0174] Experimental data fully demonstrate that this invention, through the collaborative calculation of medium environment stress factor, stress modulation current trajectory, singular fluctuation entropy, and dispersion offset, can accurately capture early weak, scattered, and random abnormal features, and the technical effect is real and verifiable.

[0175] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for extracting features from the status data of high-voltage fuses, characterized in that, include: Step S1: Real-time acquisition of operating current data, environmental data, and insulator leakage current data of the target object within the monitoring period; Step S2: After preprocessing the environmental data and the insulator leakage current data respectively, the dynamic correlation between the two is analyzed to obtain the medium environment stress factor, which reflects the comprehensive electrochemical stress level of the insulating medium. Step S3: Analyze the operating current data based on the medium environment stress factor to generate a stress-modulated current trajectory representing the current distortion caused by early insulation degradation and contact deterioration. Step S4: Analyze the stress-modulated current trajectory, extract weak anomaly features, and generate singular fluctuation entropy that reflects the complexity of the trajectory's micro-amplitude fluctuations and dispersed offset that reflects the trajectory's weak offset. This includes: decomposing the stress-modulated current trajectory into multiple sub-trajectory components at different time scales and generating the trajectory eigenvalue matrix. Based on the trajectory eigenvalue matrix, the state and transition probability of each sub-trajectory component are analyzed to generate a fluctuation mode tensor representing the fluctuation complexity. The degree of deviation of the stress-modulated current trajectory is analyzed, and a trajectory divergence field representing the overall and local deviation characteristics of the trajectory is generated. The fluctuation mode tensor and trajectory divergence field are calculated separately to generate singular fluctuation entropy, which reflects the complexity of small trajectory fluctuations, and dispersion offset, which reflects the slight trajectory offset. Specifically, this includes: calculating the transition probability matrix corresponding to each sub-trajectory component in the fluctuation mode tensor, calculating the arithmetic mean of all its off-diagonal elements to obtain the average transition activity of the sub-trajectory component; calculating the standard deviation of the average transition activity of all sub-trajectory components, and then dividing it by the range of the average transition activity of all sub-trajectory components to obtain the mode divergence ratio. For the transition probability matrix of each sub-trajectory component, find the maximum value in each row, calculate the standard deviation of these maximum values ​​in each row, and calculate the arithmetic mean of the standard deviations of all sub-trajectory components to obtain the structure-specific benchmark. Multiplying the mode divergence ratio by the structure-specific benchmark, then by the total number of sub-trajectory components, and finally by the absolute value of the kurtosis of the stress-modulated current trajectory sequence, we can obtain the singular fluctuation entropy, which reflects the complexity of the trajectory's micro-amplitude fluctuations. Step S5: Perform joint calculation of singular fluctuation entropy, dispersion offset, and medium environment stress factor to generate a health status identifier for determining the operating status of the target object.

2. The method for extracting features from high-voltage fuse status data according to claim 1, characterized in that, After preprocessing the environmental data and insulator leakage current data respectively, and analyzing their dynamic correlation, the dielectric environmental stress factor, reflecting the comprehensive electrochemical stress level of the insulating medium, was obtained, including: The environmental load index is generated by calculating the temperature and humidity in the environmental data. Fluctuation analysis was performed on the leakage current data of the insulators to obtain the pollution response. The dynamic correlation between the environmental load index and the fouling responsiveness was analyzed to obtain the electrochemical coupling coefficient, which reflects the closeness between the two.

3. The method for extracting features from high-voltage fuse status data according to claim 2, characterized in that, After preprocessing environmental data and insulator leakage current data respectively, and analyzing their dynamic correlation, a dielectric environmental stress factor reflecting the comprehensive electrochemical stress level of the insulating medium is obtained, which also includes: The environmental load index is reconstructed based on the electrochemical coupling coefficient to generate a dynamic stress intensity sequence that reflects the real-time electrochemical stress intensity under wet pollution conditions. Multi-scale analysis of dynamic stress intensity sequences is performed to generate a medium environment stress factor that reflects the comprehensive electrochemical stress level of the insulating medium under wet and polluted conditions.

4. The method for extracting features from high-voltage fuse status data according to claim 3, characterized in that, Based on the analysis of the operating current data using the dielectric environment stress factor, a stress-modulated current trajectory representing the current distortion caused by early insulation degradation and contact deterioration is generated, including: The operating current data is modulated based on the medium environment stress factor to generate a stress-modulated current sequence that reflects the dynamic loading effect of environmental stress. Local similarity analysis of the stress-modulated current sequence yields a degraded excitation spectrum representing the excitation intensity of internal defects in the target object.

5. The method for extracting features from high-voltage fuse status data according to claim 4, characterized in that, Based on the analysis of the operating current data using the dielectric environment stress factor, a stress-modulated current trajectory representing the current distortion caused by early insulation degradation and contact deterioration is generated, which also includes: Based on the degradation excitation spectrum, resonance points related to insulation degradation and contact degradation are identified, and degradation resonance feature vectors are generated. The degradation resonance eigenvectors are dynamically synthesized in the time domain to generate stress-modulated current trajectories that represent current distortions caused by early insulation degradation and contact deterioration.

6. The method for extracting features from high-voltage fuse status data according to claim 3, characterized in that, The singular fluctuation entropy, dispersion offset, and medium environment stress factor are jointly calculated to generate a health status indicator for determining the operating status of the target object, including: Based on the medium environment stress factor, evaluate the contribution weights of singular fluctuation entropy and dispersion offset to the current environmental stress level, and generate a feature weight vector. By synergistically fusing singular fluctuation entropy and dispersion offset based on the feature weight vector, a comprehensive degradation potential index representing the overall state of the target object is obtained.

7. The method for extracting features from high-voltage fuse status data according to claim 6, characterized in that, The system performs collaborative calculations of singular fluctuation entropy, dispersion offset, and medium environment stress factor to generate a health status indicator for determining the operational status of the target object. This also includes: The comprehensive degradation potential index is analyzed to generate a health status indicator for determining the operating status of the target object.

8. A high-voltage fuse status data feature extraction system, applied in the extraction method according to any one of claims 1-7, characterized in that, include: The data acquisition unit is used to collect real-time operating current data, environmental data, and insulator leakage current data of the target object during the monitoring period. The first data processing unit is used to preprocess environmental data and insulator leakage current data respectively, and analyze the dynamic correlation between the two to obtain the medium environment stress factor, which reflects the comprehensive electrochemical stress level of the environment on the insulating medium. The second data processing unit is used to analyze the operating current data based on the medium environment stress factor and generate a stress-modulated current trajectory that represents the current distortion caused by early insulation degradation and contact deterioration. The data feature extraction unit is used to analyze the stress-modulated current trajectory, extract weak anomaly features from it, and generate singular fluctuation entropy that reflects the complexity of the trajectory's micro-amplitude fluctuations and dispersed offset that reflects the trajectory's weak offset. This includes: decomposing the stress-modulated current trajectory into multiple sub-trajectory components at different time scales and generating the trajectory intrinsic matrix. Based on the trajectory eigenvalue matrix, the state and transition probability of each sub-trajectory component are analyzed to generate a fluctuation mode tensor representing the fluctuation complexity. The degree of deviation of the stress-modulated current trajectory is analyzed, and a trajectory divergence field representing the overall and local deviation characteristics of the trajectory is generated. The fluctuation mode tensor and trajectory divergence field are calculated separately to generate singular fluctuation entropy, which reflects the complexity of small trajectory fluctuations, and dispersion offset, which reflects the slight trajectory offset. Specifically, this includes: calculating the transition probability matrix corresponding to each sub-trajectory component in the fluctuation mode tensor, calculating the arithmetic mean of all its off-diagonal elements to obtain the average transition activity of the sub-trajectory component; calculating the standard deviation of the average transition activity of all sub-trajectory components, and then dividing it by the range of the average transition activity of all sub-trajectory components to obtain the mode divergence ratio. For the transition probability matrix of each sub-trajectory component, find the maximum value in each row, calculate the standard deviation of these maximum values ​​in each row, and calculate the arithmetic mean of the standard deviations of all sub-trajectory components to obtain the structure-specific benchmark. Multiplying the mode divergence ratio by the structure-specific benchmark, then by the total number of sub-trajectory components, and finally by the absolute value of the kurtosis of the stress-modulated current trajectory sequence, we can obtain the singular fluctuation entropy, which reflects the complexity of the trajectory's micro-amplitude fluctuations. The status judgment unit is used to perform collaborative calculations on singular fluctuation entropy, dispersion offset, and medium environment stress factor to generate a health status identifier for determining the operating status of the target object.

Images