Oil paper insulation thermal aging state diagnosis mode construction method under high-frequency pulse voltage

Abstract

The invention discloses an oil paper insulation thermal aging state diagnosis mode construction method under high-frequency pulse voltage. The method comprises the following steps: firstly, building a partial discharge experiment platform under high-frequency pulse, acquiring a rising edge / falling edge partial discharge signal based on an established sliding phase window model, denoising, and obtaining a phase window normalized partial discharge graph according to the denoised discharge signal; then, extracting phase jump, phase concentration ratio and other pass frequency aging characteristic quantities from the phase window normalized partial discharge spectrum; then, constructing a correlation degree matrix between the pass frequency aging characteristic quantity and the frequency, obtaining a principal component which is in weak correlation with the frequency, and completing dimension reduction; and finally, identifying different stages of thermal aging under the high-frequency pulse based on a support vector machine algorithm optimized by a particle swarm algorithm. According to the method for identifying the oil paper insulation thermal aging stage under the high-frequency pulse voltage, the identification efficiency of the high-frequency aging stage of the transformer can be improved, and the accuracy of evaluating the insulation aging state of the high-frequency transformer is further improved.

Description

Technical Field

[0001] The invention belongs to the field of online insulation monitoring and fault diagnosis of power equipment, and relates to a novel integrated optimization identification method for partial discharge characteristics of insulation. Background Art

[0002] During the manufacturing, transportation, installation and operation of power transformers, it is inevitable that various types of defects will occur in the insulation, resulting in partial discharge and affecting the safe operation of electrical equipment. With the gradual high-frequencyization of modern electrical equipment, the application of high-frequency transformers is becoming more and more extensive. However, while high-frequency transformers improve the efficiency of power transmission, they will also accelerate the aging process of their oil-paper insulation system. When the operating frequency is high, if there are insulation aging defects inside the high-frequency transformer, insulation degradation or breakdown accidents may occur during operation. By identifying the different stages of oil-paper insulation aging under high-frequency pulse voltage, the severity of insulation aging can be understood, and on this basis, the maintenance plan of the high-frequency transformer can be determined, laying the foundation for transformer status monitoring and life prediction and evaluation.

[0003] At present, the research hotspots are mainly the partial discharge characteristics and thermal aging stage identification of power frequency transformers, and there are few studies on the identification methods of thermal aging stages of oil-paper insulation of high-frequency transformers. Therefore, with the development of new power systems and the advancement of high frequencies, it is of great significance to study the identification methods of thermal aging stages of oil-paper insulation of high-frequency transformers. Summary of the invention

[0004] The present invention aims to study a method for constructing a diagnostic model of the thermal aging state of oil-paper insulation under high-frequency pulse voltage, and to provide a systematic identification method for identifying thermal aging of oil-paper insulation of transformers under high-frequency pulse voltage. First, a partial discharge experimental platform under high-frequency pulses is built, and the rising / falling edge partial discharge signals are collected and denoised based on the established sliding phase window model. The phase-window normalized partial discharge spectra at different frequencies are obtained according to the denoised rising / falling edge discharge signals; then, the phase jump and phase concentration and other frequency-broadband aging feature quantities are extracted from the phase-window normalized partial discharge spectra; subsequently, a correlation matrix is ​​constructed by calculating the correlation between different feature quantities and frequency, and the principal components with weak correlation with frequency are obtained and dimensionality reduction is completed; finally, the particle swarm algorithm is used to obtain the optimal penalty factor of the support vector machine. C and kernel parameters γ The combination of the support vector machine classifier is then optimized, and the different stages of thermal aging of oil-paper insulation under high-frequency pulses are identified based on the optimized support vector machine classification algorithm. This recognition method further improves the accuracy and efficiency of recognition on the basis of identifying the thermal aging stages of oil-paper insulation in high-frequency transformers.

[0005] Step 1: Build a partial discharge experimental platform under high-frequency pulses and conduct partial discharge experiments on samples with insulation defects.

[0006] Step 2: Based on the established sliding phase window model, the rising / falling edge high-frequency partial discharge signals are collected and denoised, thereby obtaining the phase window normalized partial discharge spectra at different frequencies; Step 3: Extract the universal frequency characteristic quantities of aging such as phase jump and phase concentration from the phase window normalized partial discharge spectrum; Step 4: construct a correlation matrix by calculating the correlation between different through-band aging characteristic quantities and frequency, obtain the principal components that are weakly correlated with the frequency and complete the dimensionality reduction; Step 5: Use the particle swarm algorithm to optimize the support vector machine classification algorithm, and use the optimized support vector machine classification method to identify the different stages of thermal aging of oil-paper insulation under high-frequency pulses.

[0007] Optionally, a high-frequency pulse partial discharge experimental platform is built in step 1, and the high-frequency pulse partial discharge experimental platform consists of a high-frequency pulse voltage power supply, a protection resistor, a coupling capacitor, a high-voltage differential probe, a defect model, a high-frequency current transformer and a digital oscilloscope. Among them, the input voltage and frequency of the high-frequency pulse voltage power supply are single-phase 220V±10%, and the output frequency is 1kHz to 20kHz (adjustable range). The bandwidth of the digital oscilloscope is 400M, and the maximum sampling frequency is 20G / s; Optionally, the phase window normalized partial discharge spectra at different frequencies in step 2 can be obtained by establishing a sliding phase window model, including: First, the square wave signal is detected by the voltage divider probe, and then based on the rising edge / falling edge interval of the high-frequency pulse square wave ( t 1, t 2) Extract the local discharge signal within this interval. The phase range of the collected local discharge signal is (360 * t 1* f， 360 * t 2* f ); then determine the number of phase windows of the rising edge / falling edge with 1kHz as the base frequency, and then fix the number of phase windows of each rising edge / falling edge. When the frequency changes, adjust the phase window size by changing the frequency so that the number of phase windows of the rising edge / falling edge of different frequencies is consistent, thereby obtaining partial discharge spectra with normalized phase windows at different frequencies.

[0008] Optionally, the phase jump in step 3 is a characteristic quantity describing the jump of the partial discharge signal during the phase change process, and its expression is: (6); Where is the phase angle after the phase jump, is the phase angle before the phase jump; Phase concentration is a characteristic quantity that describes the concentration of phase distribution of partial discharge signal, and its expression is: (7); in C is the phase concentration; N is the sample size; P ( φ i ) is the probability density of each phase angle; Since the high-frequency changes are large in amplitude and fast in speed, aging characteristic quantities such as phase jump and phase concentration are used to reflect the changes in the phase window normalized partial discharge spectrum.

[0009] Optionally, step 4 includes firstly performing a correlation analysis on the relationship between the characteristic quantity extracted from the phase window normalized partial discharge spectrum and the frequency, and calculating the correlation coefficient between each characteristic quantity and the frequency according to formula (3); (8); in: ρ is the resolution coefficient, x 0(k) is the array corresponding to the feature quantity, x i (k) is the frequency correspondence array. The correlation degree is calculated according to formula (4): (9); The grey correlation between each aging characteristic and the frequency is obtained, and a correlation matrix is ​​constructed. Based on the constructed correlation matrix, the principal component with weak correlation with the frequency is obtained, and the dimensionality reduction processing of the through-band aging characteristic is completed.

[0010] Optionally, in step 5, a particle swarm algorithm is used to optimize the support vector machine classification algorithm, and the fitness of all sample particles is evaluated according to formula (5): (10); in: e i For the i Cross-validation misclassification rate; h e is the number of correctly classified samples; h w is the number of misclassified samples; f accuracy The value is determined by the comprehensive mean classification error rate; calculate the fitness of all particle swarms. If the conditions are not met, it can be updated according to formula (5). Update the particle swarm and repeat the steps. When the set number of iterations is met, the algorithm ends and the optimal ( C , γ )combination.

[0011] Optionally, the particle swarm algorithm iteratively calculates the optimal ( C , γ ) combination to optimize the performance of the support vector machine classifier. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 A flow chart of a method for constructing a diagnostic model for the thermal aging state of oil-paper insulation under high-frequency pulse voltage provided by the present invention; Figure 2 A schematic diagram of an insulation defect model provided by the present invention; Figure 3 A schematic diagram of a test circuit system provided by the present invention. DETAILED DESCRIPTION

[0013] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0014] Figure 1 The flowchart of the method for constructing the diagnosis mode of the thermal aging state of oil-paper insulation under high-frequency pulse voltage provided by the present invention is as follows: Figure 1 As shown, the steps of the method include: Step 1: Build a partial discharge experimental platform under high-frequency pulses and conduct partial discharge experiments on samples with insulation defects.

[0015] Build a high-frequency pulse partial discharge experimental platform, such as Figure 3 As shown in the figure, the high-frequency pulse partial discharge experimental platform consists of a high-frequency pulse voltage power supply, a protective resistor, a coupling capacitor, a high-voltage differential probe, a defect model, a high-frequency current transformer and a digital oscilloscope. The input voltage and frequency of the high-frequency pulse voltage power supply are single-phase 220V±10%, and the output frequency is 1kHz to 20kHz (adjustable range). The bandwidth of the digital oscilloscope is 400M, and the maximum sampling frequency is 20G / s; Step 2: Use the established sliding phase window model to obtain the phase window normalized partial discharge spectrum at different frequencies: First, detect the square wave signal through the voltage divider probe, and then based on the rising edge / falling edge interval of the high-frequency pulse square wave ( t 1, t 2) Extract the local discharge signal within this interval. The phase range of the collected local discharge signal is (360 * t 1*f， 360 * t 2* f ). Then, the number of phase windows of the rising edge / falling edge is determined with 1kHz as the base frequency, and the number of phase windows of each rising edge / falling edge is fixed. When the frequency changes, the phase window size is adjusted by changing the frequency, so that the number of phase windows of the rising edge / falling edge of different frequencies is consistent, thereby obtaining the normalized partial discharge spectrum of the phase window at different frequencies.

[0016] Step 3: Extract the common frequency characteristic quantities of aging such as phase jump and phase concentration from the phase window normalized partial discharge spectrum: Phase jump is a characteristic quantity that describes the jump of partial discharge signal during phase change, and its expression is: (11); Where is the phase angle after the phase jump, is the phase angle before the phase jump; Phase concentration is a characteristic quantity that describes the concentration of phase distribution of partial discharge signal, and its expression is: (12); in C is the phase concentration; N is the sample size; P ( φ i ) is the probability density of each phase angle; Step 4: By calculating the correlation between different through-band aging characteristic quantities and frequency, a correlation matrix is ​​constructed to obtain the principal component with weak correlation with frequency and complete dimensionality reduction: First, the relationship between the characteristic quantity extracted from the phase window normalized partial discharge spectrum and the frequency is analyzed for correlation, and the correlation coefficient between each characteristic quantity and frequency is calculated according to formula (8); (13); in: ρ is the resolution coefficient, x 0(k) is the array corresponding to the feature quantity, x i (k) is the frequency correspondence array. The correlation degree is calculated according to formula (9): (14); The grey correlation between each aging characteristic and the frequency is obtained, and a correlation matrix is ​​constructed; based on the constructed correlation matrix, the principal component with a weak correlation with the frequency is obtained, and the dimension reduction processing of the through-frequency aging characteristic is completed; Step 5: Use the particle swarm algorithm to optimize the support vector machine classification algorithm, and identify the different stages of thermal aging of oil-paper insulation under high-frequency pulses based on the optimized support vector machine classification method.

[0017] The method implementation case is as follows: First, the insulation classic defect model is established, such as Figure 2 As shown, in the partial discharge test system, such as Figure 3 As shown in the figure, the partial discharge signal at the rising edge / falling edge of each type of defect is collected; during the test, a high-frequency pulse voltage is applied to each test sample, and 10 samples are tested for each defect. Secondly, based on the established sliding phase window model, the high-frequency partial discharge signals at the rising edge / falling edge of the samples at different frequencies are collected and denoised, and the phase window normalized partial discharge spectrum at different frequencies is obtained through the denoised rising edge / falling edge discharge signals. Then, the through-frequency aging feature quantities such as phase jump and phase concentration are extracted from the phase window normalized partial discharge spectrum. Subsequently, the correlation matrix is ​​constructed by calculating the correlation between different through-frequency aging feature quantities and frequency, and the principal components with weak correlation with frequency are obtained and the dimensionality reduction is completed. Finally, the support vector machine optimized based on the particle swarm algorithm is used to identify the aging stage of the high-frequency transformer.

Claims

1. A method for constructing a diagnostic model for thermal aging of oil-paper insulation under high-frequency pulse voltage, characterized in that: The steps of the method include: Step 1: Build a partial discharge experimental platform under high-frequency pulses, conduct partial discharge experiments on samples with insulation defects, and collect and denoise the high-frequency partial discharge signals at the rising / falling edges based on the established sliding phase window model; Step 2: obtaining the phase window normalized partial discharge spectra at different frequencies based on the denoised rising edge / falling edge partial discharge signals; Step 3: Extract the universal frequency characteristic quantities of aging such as phase jump and phase concentration from the phase window normalized partial discharge spectrum; Step 4: construct a correlation matrix by calculating the correlation between different through-band aging characteristic quantities and frequency, obtain the principal components that are weakly correlated with the frequency and complete the dimensionality reduction; Step 5: The support vector machine optimized based on the particle swarm algorithm is used to identify the different stages of thermal aging of oil-paper insulation under high-frequency pulses.

2. The method for constructing a diagnostic model for thermal aging of oil-paper insulation under high-frequency pulse voltage according to claim 1, characterized in that: The phase window normalized partial discharge spectra at different frequencies in step 1 can be obtained by establishing a sliding phase window model, including: First, the square wave signal is detected by the voltage divider probe, and then based on the rising edge / falling edge interval of the high-frequency pulse square wave ( t 1, t 2) Extract the local discharge signal within this interval. The phase range of the collected local discharge signal is (360 * t 1* f， 360 * t 2* f ); then determine the number of phase windows of the rising edge / falling edge with 1kHz as the base frequency, and then fix the number of phase windows of each rising edge / falling edge. When the frequency changes, adjust the phase window size by changing the frequency so that the number of phase windows of the rising edge / falling edge of different frequencies is consistent, thereby obtaining partial discharge spectra with normalized phase windows at different frequencies.

3. The method for constructing a diagnostic model for thermal aging of oil-paper insulation under high-frequency pulse voltage according to claim 1, characterized in that: The phase jump in step 2 is a characteristic quantity that describes the jump of the partial discharge signal during the phase change process, and its expression is: （1）； in is the phase angle after the phase jump, is the phase angle before the phase jump; Phase concentration is a characteristic quantity that describes the concentration of phase distribution of partial discharge signal, and its expression is: （2）； in C is the phase concentration; N is the sample size; P ( φ i ) is the probability density of each phase angle; Since the high-frequency changes are large in amplitude and fast in speed, aging characteristic quantities such as phase jump and phase concentration are used to reflect the changes in the phase window normalized partial discharge spectrum.

4. The method for constructing a diagnostic model for thermal aging of oil-paper insulation under high-frequency pulse voltage according to claim 1, characterized in that: Step 3 includes, firstly, performing a correlation analysis on the relationship between the characteristic quantity extracted from the phase window normalized partial discharge spectrum and the frequency, and calculating the correlation coefficient between each characteristic quantity and the frequency according to formula (3); （3）； in: ρ is the resolution coefficient, x 0(k) is the array corresponding to the feature quantity, x i (k) is the frequency correspondence array. The correlation degree is calculated according to formula (4): （4）； The grey correlation between each aging characteristic and the frequency is obtained, and a correlation matrix is ​​constructed. Based on the constructed correlation matrix, the principal component with a weak correlation with the frequency is obtained, and the dimensionality reduction processing of the through-band aging characteristic is completed.

5. The method for constructing a diagnostic model for thermal aging of oil-paper insulation under high-frequency pulse voltage according to claim 1, characterized in that: In step 4, the particle swarm algorithm is used to optimize the support vector machine classification algorithm, and the fitness of all sample particles is evaluated according to formula (5): （5）； in: e i For the i Cross-validation misclassification rate; h e is the number of correctly classified samples; h w is the number of misclassified samples; f accuracy The value is determined by the comprehensive mean classification error rate; calculate the fitness of all particle swarms. If the conditions are not met, it can be updated according to formula (5). Update the particle swarm and repeat the steps. When the set number of iterations is met, the algorithm ends and the optimal ( C , γ )combination.

6. The method for constructing a diagnostic model for thermal aging of oil-paper insulation under high-frequency pulse voltage according to claim 1, characterized in that: The particle swarm algorithm iteratively calculates the optimal ( C , γ ) combination to optimize the performance of the support vector machine classifier.

Images