An insulation defect classification method based on a photon pulse sequence pattern
Patent Information
- Application Number
- CN202611141322.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-07-30
- Publication Date
- 2026-10-30
AI Technical Summary
1.本发明通过构建考虑工频周期关联特性的光子脉冲序列模式图谱，能够有效刻画不同绝缘缺陷在光子发光过程中的相位滞后特性与极性分布差异，克服了传统相位分辨光子计数方法中不同缺陷特征分布相似、难以区分的问题。在此基础上，通过对光子核密度分布进行建模，并提取包括极值特征、统计分布特征及空间分布特征在内的多维统计参数，实现了对光子发光行为的多维度表征，使不同缺陷在特征空间中形成明显可分离的分布区域，从而显著提升绝缘缺陷分类的准确性与可靠性。
Smart Images
Figure CN123009518A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of power equipment condition monitoring and insulation defect diagnosis technology, and in particular to an insulation defect classification method based on photon pulse sequence spectrum. Background Technology
[0002] Gas-insulated transmission lines (GIL) and gas-insulated switchgear (GIS) are important components of ultra-high voltage power grids, and their operational reliability directly affects the safe and stable operation of the power system. In actual operation, insulation defects (such as metal particles, surface scratches, and protruding defects) are prone to partial discharge under the influence of an electric field, which can lead to deterioration of the equipment's insulation performance or even breakdown failure.
[0003] Currently, detection methods for internal defects in gas-insulated equipment mainly include ultra-high frequency detection, ultrasonic detection, and pulsed current detection. However, these methods are susceptible to electromagnetic interference and environmental noise, and lack sufficient sensitivity in identifying weak discharges and early defects. In contrast, photon counting-based detection methods can directly reflect the luminescence behavior of insulation defects under the influence of an electric field, and have advantages such as strong anti-interference ability and high sensitivity. Therefore, they are gradually becoming an important means of insulation defect detection.
[0004] However, in practical applications, different types of insulation defects may exhibit similar statistical characteristics in amplitude distribution and phase properties during different emission stages (such as electroluminescence, gas ionization, and partial discharge). Especially in the maps constructed using the Phase-Resolved Photon Counting (PRPC) method, the distribution patterns of different defect types overlap significantly, making it difficult for traditional methods to effectively distinguish defect types and limiting the application effect of photon detection technology in defect classification.
[0005] Therefore, there is an urgent need to propose an insulation defect classification method that can effectively extract deep features of photon signals at different emission stages and enhance the differences between different defects, so as to improve the accuracy and reliability of defect identification in gas-insulated equipment. Summary of the Invention
[0006] To address the above shortcomings, this invention provides an insulation defect classification method based on photon pulse sequence maps. By constructing a photon pulse sequence pattern map that considers the correlation characteristics of power frequency cycles and extracting multidimensional statistical features, it effectively distinguishes insulation defects under different emission stages. This method can achieve stable classification under four typical insulation defect types and six emission stage conditions, and has the advantages of high accuracy and strong applicability, which helps to improve the identification capability and early warning level of insulation defects in gas-insulated equipment.
[0007] To achieve the above objectives, the present invention adopts the following technical solution: An insulation defect classification method based on photon pulse sequence maps includes the following steps: (1) Photon data acquisition: Construct a defect electrode model, collect photon count data of different defect types in multiple emission stages under AC voltage excitation conditions, and construct phase-resolved photon count distribution data based on voltage phase information; (2) Photon pulse sequence construction: The phase-resolved photon counting data obtained from multiple repeated measurements are spliced together in chronological order to form a continuous photon pulse sequence. Background noise is eliminated by amplitude threshold denoising. At the same time, the sequence is smoothed by the sliding window method to improve the signal-to-noise ratio of the photon signal. (3) Feature mapping construction: Based on the power frequency cycle characteristics, establish the mapping relationship between the current phase photon pulse amplitude and the photon pulse amplitude that lags by one power frequency cycle, and construct a two-dimensional photon pulse feature space to characterize the phase correlation and polarity distribution characteristics in the photon emission process; (4) Photon pulse sequence pattern construction: The probability density model of the sample data in the two-dimensional photon pulse feature space is modeled using the kernel density estimation method, and the obtained probability density is normalized to generate a photon pulse sequence pattern map. (5) Defect classification: Based on the photon pulse sequence pattern spectrum, multidimensional statistical feature parameters characterizing the photon distribution characteristics are extracted, and feature fusion and dimensionality reduction are performed on the feature parameters. Based on the distribution differences of different defects in the feature space, the insulation defect type is classified and identified.
[0008] Preferably, in the above-mentioned insulation defect classification method based on photon pulse sequence spectrum, the defect electrode model is used to simulate typical insulation defects in gas-insulated equipment, including surface scratch defects, metal particle defects, attached metal defects, and tip protrusion defects.
[0009] Preferably, in the above-mentioned method for classifying insulation defects based on photon pulse sequence maps, the acquisition of photon counting data is carried out in a closed gas environment, which is a test chamber filled with insulating gas, and the photon signals generated by the defects under AC voltage excitation are collected by a photon detector.
[0010] Preferably, in the above-mentioned method for classifying insulation defects based on photon pulse sequence spectrum, the method for constructing the phase-resolved photon count distribution data in step (1) is as follows: divide a power frequency cycle into multiple phase intervals, count the number of photon pulses in each phase interval, and construct a three-dimensional distribution matrix with voltage phase as the horizontal axis, measurement number as the vertical axis, and photon pulse number as the amplitude.
[0011] Preferably, in the above-mentioned method for classifying insulation defects based on photon pulse sequence maps, step (2) involves the following process for constructing the photon pulse sequence: (1) Multiple sets of phase-resolved photon counting data obtained under the same working condition are spliced together in the order of measurement to form the original photon sequence of continuous time series; (2) The original photon sequence is filtered based on a preset amplitude threshold to remove background noise signals below the threshold and obtain an effective photon pulse sequence; Where i and j represent the diameter and the number of signals in the diameter, and [] is the empty set; This is the original photon pulse sequence obtained from the initial acquisition. The effective photon pulse sequence after removing background noise.
[0012] (3) The denoised photon pulse sequence is smoothed using a sliding window to improve the signal-to-noise ratio of the photon signal; Where w is the size of the sliding window, and data points outside the window are symmetrically reflected into the window; This is a sequence of photon pulses smoothed by a sliding window.
[0013] Preferably, in the above-mentioned method for classifying insulation defects based on photon pulse sequence maps, the method for constructing the two-dimensional photon pulse feature space is as follows: based on the processed photon pulse sequence, a mapping relationship is established between the amplitude of the photon pulse in the current power frequency cycle and the amplitude of the photon pulse lagging by one power frequency cycle, which is used to characterize the phase lag characteristics in the photon emission process.
[0014] Where N is The total number of signal points in the photon sequence; T is the number of quantum pulses in the AC period; s is and Tags; The smoothed photon pulse sequence obtained by formula (2); The x-coordinate data in the two-dimensional feature space; The data represents the ordinate in the two-dimensional feature space.
[0015] Preferably, in the above-mentioned insulation defect classification method based on photon pulse sequence spectrum, the kernel density estimation adopts a non-parametric probability density estimation method to model the sample data in the two-dimensional photon pulse feature space to obtain the probability density function of the photon pulse distribution.
[0016] Where h is the bandwidth parameter; K(·) is a kernel function; The x-coordinate data in the two-dimensional feature space in formula (3); y is the ordinate data in the two-dimensional feature space in formula (3); M is the number of samples (number of feature points) in the two-dimensional photon pulse feature space. Let be the eigenvalue corresponding to the s-th two-dimensional feature point; Eigenvalues The corresponding probability density.
[0017] Preferably, in the above-mentioned method for classifying insulation defects based on photon pulse sequence patterns, the photon pulse sequence pattern (PPSP) constructs a three-dimensional distribution map with the amplitude of the current phase photon pulse as the horizontal axis, the amplitude of the photon pulse lagging by one power frequency cycle as the vertical axis, and the normalized photon nuclear density as the third dimension, which is used to characterize the differences in luminescence characteristics of different defects.
[0018] Preferably, in the above-mentioned method for classifying insulation defects based on photon pulse sequence maps, the extraction of multidimensional statistical feature parameters in step (5) includes the following: (1) Extract extreme value features from the normalized photon nuclear density distribution, including the maximum value Pmax and the minimum value Pmin; (2) Extract the overall distribution characteristics, including total photon density Stot, average photon density Savr, and standard deviation Vstd; (3) Extract spatial distribution features, including the centroid coordinates Xcen and Ycen of the photon nucleus density distribution; (4) Extract distribution morphology features, including skewness Ske and kurtosis Kur.
[0019] Preferably, in the above-mentioned method for classifying insulation defects based on photon pulse sequence maps, step (5), the feature fusion and dimensionality reduction process uses a nonlinear dimensionality reduction method to map multidimensional statistical features, including the following: (1) Construct the multidimensional statistical feature parameters into a high-dimensional feature vector; (2) The high-dimensional feature vector is reduced in dimensionality using the t-SNE algorithm to obtain a two-dimensional or three-dimensional feature representation; (3) Based on the feature distribution after dimensionality reduction, cluster analysis is performed to determine the cluster centers and classification boundaries of different defect types in the feature space.
[0020] The above-mentioned insulation defect classification method based on photon pulse sequence maps is applicable to the identification of insulation defects in different luminescence stages, including electroluminescence stage, gas ionization stage, micro-discharge stage, and partial discharge stage.
[0021] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages: 1. This invention constructs a photon pulse sequence pattern map that considers the correlation characteristics of power frequency cycles, effectively characterizing the phase lag characteristics and polarity distribution differences of different insulation defects during photon emission. This overcomes the problem of similar and indistinguishable feature distributions of different defects in traditional phase-resolved photon counting methods. Based on this, by modeling the photon nuclear density distribution and extracting multidimensional statistical parameters including extreme value features, statistical distribution features, and spatial distribution features, a multidimensional characterization of photon emission behavior is achieved. This allows different defects to form clearly separable distribution regions in the feature space, thereby significantly improving the accuracy and reliability of insulation defect classification.
[0022] 2. This invention enhances the expressive power of photon signal temporal correlation characteristics by introducing a cross-power frequency cycle photon pulse amplitude mapping relationship. This allows different defects to maintain stable characteristic differences at different emission stages, and effective differentiation is achieved after feature fusion and dimensionality reduction, demonstrating the method's good stage adaptability and robustness. Furthermore, this invention achieves detection based on photon signals, without relying on complex electromagnetic interference suppression measures, exhibiting higher sensitivity in weak discharge and early defect identification. It is suitable for online monitoring and early warning of gas-insulated equipment, possessing significant engineering application value. Attached Figure Description
[0023] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below.
[0024] Figure 1 This is a flowchart of an insulation defect classification method based on photon pulse sequence maps according to Embodiment 1 of the present invention; Figure 2 This is a schematic diagram of the defect electrode structure of an insulation defect classification method based on photon pulse sequence spectrum according to Embodiment 1 of the present invention. In (a), the upper figure is a physical image of a gas-insulated device with surface scratch defects, and the lower figure is a schematic diagram of the defect electrode structure of the surface scratch defects; in (b), the upper figure is a physical image of a gas-insulated device with metal particle defects, and the lower figure is a schematic diagram of the defect electrode structure of the metal particle defects; in (c), the upper figure is a physical image of a gas-insulated device with attached metal, and the lower figure is a schematic diagram of the defect electrode structure of the attached metal defects; in (d), the upper figure is a physical image of a gas-insulated device with a pointed protrusion defect, and the lower figure is a schematic diagram of the defect electrode structure of the pointed protrusion defect.
[0025] Figure 3 This is a schematic diagram of a photon counting measurement system and test circuit for an insulation defect classification method based on photon pulse sequence maps, according to Embodiment 1 of the present invention.
[0026] Figure 4 This is a schematic diagram illustrating the process of constructing a continuous photon pulse sequence based on PRPC results in an insulation defect classification method based on photon pulse sequence maps, as described in Embodiments 1-2 of the present invention.
[0027] Figure 5 The diagrams shown are schematic diagrams of the construction results of the photon pulse sequence pattern map (PPSP) of the insulation defect classification method based on photon pulse sequence map according to Embodiments 1-2 of the present invention; wherein, (a) is a schematic diagram of the construction results of surface scratch defects; (b) is a schematic diagram of the construction results of metal particle defects; (c) is a schematic diagram of the construction results of attached metal defects; and (d) is a schematic diagram of the construction results of tip protrusion defects.
[0028] Figure 6 The following are schematic diagrams of photon nucleus density distribution characteristic curves corresponding to different insulation defects in an insulation defect classification method based on photon pulse sequence spectrum according to Embodiments 1-2 of the present invention: (a) is a schematic diagram of photon nucleus density distribution characteristic curve for surface scratch defects; (b) is a schematic diagram of photon nucleus density distribution characteristic curve for metal particle defects; (c) is a schematic diagram of photon nucleus density distribution characteristic curve for attached metal defects; and (d) is a schematic diagram of photon nucleus density distribution characteristic curve for tip protrusion defects.
[0029] Figure 7 This is a schematic diagram of the defect classification results after dimensionality reduction processing based on multidimensional statistical features, which is an insulation defect classification method based on photon pulse sequence spectrum according to Embodiments 1-2 of the present invention. Detailed Implementation
[0030] 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. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.
[0031] Example 1: like Figure 1 As shown, an insulation defect classification method based on photon pulse sequence maps is presented. The diagnostic process is as follows: Figure 1 As shown, it includes the following steps: (1) First, a defect electrode model is constructed to simulate typical internal defects of gas-insulated equipment, such as Figure 2As shown, defects include surface scratches, metal particles, attached metal, and protruding tips. Then, photon counting statistics are performed, such as... Figure 3 As shown, under AC voltage excitation, the defect electrode is placed in a sealed test chamber filled with insulating gas, and the photon pulse signals generated by the defect are collected by a photon detector to obtain photon count data. The photon detector uses a high-sensitivity photon counting module, and its parameters are shown in Table 1. Table 1. Photon Count Data Acquired by Photon Detector: (2) In order to study the influence of different luminescence stages on the characteristics of insulation defects, tests were conducted under different voltage conditions. The voltage settings for the luminescence stages corresponding to different defects are shown in Table 2.
[0032] Table 2. Voltage table for different defect types during the emission stage. (3) Based on the acquired photon counting data, the photon pulses are statistically analyzed according to the voltage phase. For example... Figure 4 As shown, a power frequency cycle is divided into multiple phase intervals, and the number of photon pulses in each phase interval is counted to construct a phase-resolved photon counting (PRPC) spectrum. Multiple measurements are repeated, and the results are superimposed or arranged sequentially to improve the stability and repeatability of the data.
[0033] (4) The PRPC data obtained from multiple measurements under the same working condition are spliced together in chronological order to form a continuous photon pulse sequence. The original photon sequence is filtered based on a preset amplitude threshold to remove background noise signals; at the same time, the sequence is smoothed using a sliding window method to improve the signal-to-noise ratio of the photon signal and enhance the continuity of the sequence.
[0034] (5) Based on the processed photon pulse sequence, a mapping relationship is established between the amplitude of the photon pulse in the current power frequency cycle and the amplitude of the photon pulse that lags by one power frequency cycle, thus constructing a two-dimensional photon pulse feature space. Through this mapping relationship, the phase lag characteristics and polarity distribution law in the photon emission process can be characterized, thereby enhancing the differences between different defects.
[0035] (6) The kernel density estimation method is used to model the probability density of sample data in the two-dimensional photon pulse feature space, and the obtained density results are normalized to obtain the photon kernel density distribution. Based on this, the amplitude of the current period photon is used as the horizontal axis, the amplitude of the lag period photon is used as the vertical axis, and the normalized kernel density is used as the third dimension, such as... Figure 5 As shown, a photon pulse sequence pattern map (PPSP) is constructed to characterize the photon emission distribution features corresponding to different defects.
[0036] (7) Based on the PPSP spectrum, perform statistical analysis on the photon nucleus density distribution, such as... Figure 6 As shown, multidimensional feature parameters are extracted, including the maximum photon density. P max , minimum photon density P min Total photon density S tot Average photon density S avr Standard deviation V std centroid in the horizontal axis direction X cen 、 centroid in the ordinate direction Y cen 、 Skewness S ke and kurtosis K ur Feature parameters, etc. In this embodiment, statistical analysis was performed on the feature parameters extracted from different defects, and their average values are shown in Table 3.
[0037] Table 3. Statistical table of feature parameters extracted from different defects: (8) such as Figure 7 As shown, the above feature parameters are constructed into a high-dimensional feature vector, and a nonlinear dimensionality reduction method is used for feature fusion processing to achieve classification and identification of different insulation defect types.
[0038] Example 2: Based on Example 1, in order to verify the applicability of the method of the present invention under different emission stages, photon signal acquisition and feature extraction were repeatedly performed on each defect under multiple emission stages based on the different voltage conditions shown in Table 2.
[0039] For each emission stage, a photon pulse sequence pattern map was constructed according to the method in Example 1, and the corresponding statistical feature parameters were extracted. The results show that the feature parameters exhibit certain trends under different emission stages, but significant differences remain between different defects.
[0040] Furthermore, the feature parameters extracted at each stage are uniformly reduced in dimension. In the reduced dimension space, different defect types form relatively independent distribution regions at each luminescence stage, indicating that the method of the present invention has good stage adaptability.
[0041] As can be seen from the statistical results shown in Table 2 of Example 1, although the photon emission characteristics evolve with voltage changes, the overall distribution of characteristic parameters corresponding to different defects remains stable, thus ensuring the reliability of the classification results.
[0042] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.
Claims
1. A method for classifying insulation defects based on photon pulse sequence maps, characterized in that, Includes the following steps: (1) Photon data acquisition: Construct a defect electrode model, collect photon count data of different defect types in multiple emission stages under AC voltage excitation conditions, and construct phase-resolved photon count distribution data based on voltage phase information; (2) Photon pulse sequence construction: The phase-resolved photon counting data obtained from multiple repeated measurements are spliced together in chronological order to form a continuous photon pulse sequence. Background noise is eliminated by amplitude threshold denoising. At the same time, the sequence is smoothed by the sliding window method to improve the signal-to-noise ratio of the photon signal. (3) Feature mapping construction: Based on the power frequency cycle characteristics, establish the mapping relationship between the current phase photon pulse amplitude and the photon pulse amplitude that lags by one power frequency cycle, and construct a two-dimensional photon pulse feature space to characterize the phase correlation and polarity distribution characteristics in the photon emission process; (4) Photon pulse sequence pattern construction: The probability density model of the sample data in the two-dimensional photon pulse feature space is modeled using the kernel density estimation method, and the obtained probability density is normalized to generate a photon pulse sequence pattern map. (5) Defect classification: Based on the photon pulse sequence pattern spectrum, multidimensional statistical feature parameters characterizing the photon distribution characteristics are extracted, and feature fusion and dimensionality reduction are performed on the feature parameters. Based on the distribution differences of different defects in the feature space, the insulation defect type is classified and identified.
2. The method for classifying epoxy insulation defects in gas-insulated transmission lines based on photon pulse sequence modes according to claim 1, characterized in that: The defect electrode model is used to simulate typical insulation defects in gas-insulated transmission lines, including surface scratch defects, metal particle defects, attached metal defects, and pointed protrusion defects.
3. The method for classifying epoxy insulation defects in gas-insulated transmission lines based on photon pulse sequence modes according to claim 1, characterized in that: The acquisition of photon counting data is carried out in a closed gas environment, which is a test chamber filled with insulating gas, and the photon signals generated by the defects under AC voltage excitation are collected by a photon detector.
4. The insulation defect classification method based on photon pulse sequence spectrum according to claim 1, characterized in that: In step (1), the method for constructing phase-resolved photon counting distribution data is as follows: a power frequency cycle is divided into a preset number of phase intervals, and the number of photon pulses in each phase interval is counted to construct a three-dimensional distribution matrix with voltage phase as the horizontal axis, the number of measurements as the vertical axis, and the number of photon pulses as the amplitude; by superimposing or arranging the results of multiple repeated measurements, a phase-resolved photon counting pattern with phase distribution characteristics is formed.
5. The insulation defect classification method based on photon pulse sequence spectrum according to claim 1, characterized in that: In step (2), the construction of the photon pulse sequence Includes the following processes: (1) Multiple sets of phase-resolved photon counting data acquired under the same working condition are spliced together in the order of measurement to form the original photon sequence of continuous time series; (2) Based on the preset amplitude threshold θd The original photon sequence is filtered to remove background noise signals below a threshold, resulting in an effective photon pulse sequence. Φ The processing method is as follows: Where i and j represent the diameter and the number of signals in the diameter, and [] is the empty set; This is the original photon pulse sequence obtained from the initial acquisition. The effective photon pulse sequence after removing background noise; (3) The denoised photon pulse sequence is smoothed using a sliding window to improve the signal-to-noise ratio. The processing method is as follows: Where w is the size of the sliding window; data points outside the window are symmetrically reflected into the window; This is a sequence of photon pulses smoothed by a sliding window.
6. The insulation defect classification method based on photon pulse sequence spectrum according to claim 5, characterized in that: In step (3), the method for constructing the two-dimensional photon pulse feature space is as follows: based on the processed photon pulse sequence The mapping relationship between the current power frequency cycle and the photon pulse amplitude lagging by one power frequency cycle is constructed, and its expression is: Where N is The total number of signal points in the photon sequence; T is the number of quantum pulses in the AC period; s is and Tags; The photon pulse sequence obtained by formula (2) after smoothing through a sliding window; The x-coordinate data in the two-dimensional feature space; The data represents the ordinate in the two-dimensional feature space.
7. The insulation defect classification method based on photon pulse sequence spectrum according to claim 6, characterized in that: In step (4), the kernel density estimation uses a non-parametric probability density estimation method to model the sample data in the two-dimensional photon pulse feature space. The calculation process includes: Where h is the bandwidth parameter, and K(·) is a kernel function; The x-coordinate data in the two-dimensional feature space of formula (3); The vertical coordinate data in the two-dimensional feature space of formula (3) is M; M is the number of feature points in the two-dimensional photon pulse feature space. Let be the eigenvalue corresponding to the s-th two-dimensional feature point; Eigenvalues The corresponding probability density.
8. The insulation defect classification method based on photon pulse sequence spectrum according to claim 1, characterized in that: The photon pulse sequence pattern map is constructed with the amplitude of the current phase photon pulse on the horizontal axis, the amplitude of the photon pulse lagging by one power frequency cycle on the vertical axis, and the corresponding normalized photon nuclear density as the third dimension, to construct a three-dimensional distribution map, which is used to characterize the phase lag characteristics and polarity distribution features in the defect luminescence process.
9. The insulation defect classification method based on photon pulse sequence spectrum according to claim 1, characterized in that: In step (5), the extraction of multidimensional statistical feature parameters includes the following: (1) Extract extreme value features from the normalized photon nuclear density distribution, including the maximum value Pmax and the minimum value Pmin; (2) Extract the overall distribution characteristics, including total photon density Stot, average photon density Savr, and standard deviation Vstd; (3) Extract spatial distribution features, including the centroid coordinates Xcen and Ycen of the photon nucleus density distribution; (4) Extract distribution morphology features, including skewness Ske and kurtosis Kur.
10. The insulation defect classification method based on photon pulse sequence spectrum according to claim 1, characterized in that: In step (5), the feature fusion and dimensionality reduction process uses a nonlinear dimensionality reduction method to map multidimensional statistical features, including the following: (1) Construct the multidimensional statistical feature parameters into a high-dimensional feature vector; (2) The high-dimensional feature vector is reduced in dimensionality using the t-SNE algorithm to obtain a two-dimensional or three-dimensional feature representation; (3) Based on the feature distribution after dimensionality reduction, cluster analysis is performed to determine the cluster centers and classification boundaries of different defect types in the feature space.