A UHF partial discharge identification method based on shape, energy and statistical characteristics

Abstract

The invention discloses a method for identifying ultra-high frequency partial discharge based on shape, energy and statistical features, including ultra-high frequency partial discharge feature calibration based on shape, energy and statistics: including reading ultra-high frequency partial discharge data and data processing, extracting and calibrating ultra-high frequency partial discharge feature quantity; calculating ultra-high frequency partial discharge feature, identifying and locating fault abnormality type: including collecting ultra-high frequency partial discharge signal and signal processing; extracting ultra-high frequency partial discharge feature quantity; logical discrimination of normal operation data and interference data; identifying and locating ultra-high frequency partial discharge type and logical reasoning of ultra-high frequency partial discharge feature quantity. The invention is suitable for ultra-high frequency partial discharge identification of primary equipment of power grid, discovering the existing partial discharge site in the early stage of short-circuit fault, taking precautions in advance, avoiding partial discharge from further evolving into power grid accident, and improving the automation level; using three different dimensions based on shape, energy and statistics to extract time domain and frequency features of ultra-high frequency partial discharge respectively, different objects have different feature sensitivity, and are widely applicable to objects; the invention uses three different logics of "AND logic", "three-out-of-two logic" and "three-out-of-one logic", and has high flexibility.

Description

Technical Field

[0001] The invention relates to a method for identifying ultra-high frequency partial discharge types, and in particular to a method for identifying ultra-high frequency partial discharge based on shape, energy and statistical characteristics. Background Art

[0002] Transformers, GIS (Gas Insulated Switchgear), GIL (Gas Insulated Line), switchgear, etc. are the main equipment of the power grid. The safe operation of the equipment is related to the reliability level of the power system. If a failure occurs, it will cause a local or even large-scale power outage. With the increasing number of substations, the operation and maintenance costs of substation equipment are getting higher and higher, and the demand for online monitoring of substation main equipment is becoming more and more urgent.

[0003] The current power grid scale is constantly growing, and partial discharge failures of transformers, GIS, GIL, switchgear, etc. occur frequently. If partial discharge is not handled in time, it will develop into a short circuit failure, and the equipment failure will cause power outages in the power grid. Therefore, it is particularly important to promptly discover and identify partial discharge of transformers, GIS, GIL, switchgear and other equipment. The characteristic of partial discharge is a wide frequency band, among which partial discharge in the ultra-high frequency band (frequency of 300-3000MHz) is a key and important feature.

[0004] The method for extracting characteristic parameters of ultra-high frequency partial discharge detection spectrum and the abnormality detection method of application number 201710866621.1 describe the use of statistical methods to extract local discharge pattern characteristics and construct a characteristic vector to characterize the statistical distribution characteristics of the spectrum as the basis for classification of local discharge categories, with the focus on extracting the statistical characteristics based on the discharge amount to identify the ultra-high frequency discharge type; the method and device for calculating the time difference of ultra-high frequency partial discharge signal based on short-time energy of application number 202310293838.3 describe the conversion of the voltage waveform of the short-time energy signal into an energy accumulation curve, and determine the inflection point on the energy accumulation curve where the ultra-high frequency partial discharge signal is greater than the background noise as the starting time of the ultra-high frequency partial discharge signal, with the focus on noise reduction processing and the identification of the inflection point, thereby identifying the partial discharge.

[0005] The current single method used to identify UHF partial discharge has problems of low accuracy or missed identification. Summary of the invention

[0006] Purpose of the invention: The purpose of the present invention is to provide a method for identifying ultra-high frequency partial discharge based on shape, energy and statistical characteristics, which can improve the accuracy of identification and avoid missed and wrong judgments through the judgment of comprehensive methods.

[0007] Technical solution: The present invention includes: shape-, energy-, and statistical UHF partial discharge feature calibration: including reading UHF partial discharge data and signal processing, UHF partial discharge feature extraction and calibration; UHF partial discharge feature calculation, fault abnormality type identification and positioning: including collecting UHF partial discharge signals and signal processing; UHF partial discharge feature extraction; normal operation data and interference data logical discrimination; UHF partial discharge type identification and positioning and UHF partial discharge feature logical reasoning.

[0008] The ultra-high frequency partial discharge feature extraction includes shape-based feature extraction, energy-based feature extraction and statistics-based feature extraction.

[0009] The shape-based feature extraction includes shape-based time domain feature extraction, shape-based frequency domain feature extraction, and shape-based period extraction.

[0010] The energy-based feature extraction includes energy-based time domain feature extraction, energy-based frequency domain feature extraction, and energy-based period extraction.

[0011] The statistical feature extraction includes statistical time domain feature extraction and statistical frequency domain feature extraction.

[0012] The ultra-high frequency partial discharge characteristic quantity calibration includes shape-based time domain characteristic calibration, shape-based frequency domain characteristic calibration, shape-based period calibration, energy-based time domain characteristic calibration, energy-based frequency domain characteristic calibration, energy-based period calibration, statistics-based time domain characteristic calibration, and statistics-based frequency domain characteristic calibration.

[0013] The ultra-high frequency partial discharge type identification and positioning are specifically as follows: an error comparison algorithm, a similarity algorithm, and a classification algorithm are used to identify the ultra-high frequency partial discharge type based on shape-based time domain features, shape-based frequency domain features, shape-based periodic features, energy-based time domain features, energy-based frequency domain features, energy-based periodic features, and statistical time domain features, and positioning is performed according to the size of ultra-high frequency signals collected at different positions, and spatial positioning is performed according to ultra-high frequency signals collected at different positions after time synchronization.

[0014] The ultra-high frequency partial discharge fault abnormality judgment logic includes a shape feature-based logic module, an energy feature-based logic module, and a statistical feature-based logic module.

[0015] The ultra-high frequency partial discharge fault abnormality judgment logic adopts three different logic judgments: "AND logic", "three-out-of-two logic" and "three-out-of-one logic".

[0016] A UHF partial discharge identification device based on shape, energy and statistical characteristics comprises a UHF partial discharge signal acquisition module, a UHF partial discharge signal denoising processing module, a UHF partial discharge feature quantity extraction module, a UHF partial discharge feature quantity discrimination algorithm module and a UHF partial discharge feature quantity logic reasoning module.

[0017] Beneficial effects: The present invention is suitable for ultra-high frequency partial discharge identification of primary equipment in power grids, and can discover the existing partial discharge site in the early stage of short-circuit faults, take precautions before it happens, and avoid partial discharge from further evolving into power grid accidents, thereby improving the level of automation. The time domain and frequency characteristics of ultra-high frequency partial discharge are extracted based on three different dimensions, namely shape, energy, and statistics. Different objects have different feature sensitivities and are widely applicable. The recognition accuracy is improved through the judgment of a comprehensive method, and missed judgments and wrong judgments are avoided. The present invention adopts three different logics, namely "AND logic", "three-out-of-two logic", and "three-out-of-one logic", and is highly flexible. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 This is a flow chart of the calibration of ultra-high frequency partial discharge characteristics based on shape, energy and statistics of the present invention;

[0019] Figure 2 It is a flow chart of UHF partial discharge characteristic calculation, fault abnormality type identification and location of the present invention;

[0020] Figure 3 It is the UHF partial discharge fault abnormality judgment logic diagram (and logic conditions) of the present invention;

[0021] Figure 4 It is a logic diagram for judging abnormality of UHF partial discharge fault of the present invention (two out of three logic conditions);

[0022] Figure 5 It is a UHF partial discharge fault abnormality judgment logic diagram (one out of three logic conditions) of the present invention. DETAILED DESCRIPTION

[0023] The present invention will be further described below in conjunction with the accompanying drawings.

[0024] like Figure 1 and Figure 2 As shown, the UHF partial discharge identification method based on shape, energy and statistical characteristics of the present invention includes: UHF partial discharge feature calibration based on shape, energy and statistics, UHF partial discharge feature calculation, fault abnormality type identification and positioning.

[0025] Among them, the calibration of ultra-high frequency partial discharge characteristics based on shape, energy, and statistics specifically includes the following steps:

[0026] S1. Read the UHF PD data files from the site and the experiment

[0027] Read n groups of UHF PD data files collected on site and in experiments. The data files include normal data, abnormal data, and interference data. Different types of data and different types of PD data are calibrated respectively.

[0028] S2. De-noising of ultra-high frequency partial discharge signals: Use de-noising algorithm to filter out white noise.

[0029] S3. Extraction and calibration of ultra-high frequency partial discharge features. The following different feature extraction methods are used to extract ultra-high frequency partial discharge features, and data calibration is performed based on the same type of data of different objects. The data calibration includes shape-based time domain feature calibration, shape-based frequency domain feature calibration, shape-based period calibration, energy-based time domain feature calibration, energy-based frequency domain feature calibration, energy-based period calibration, statistics-based time domain feature calibration, and statistics-based frequency domain feature calibration. Specifically including:

[0030] S31, shape-based feature extraction; specifically including:

[0031] S311. Shape-based time domain feature extraction

[0032] The shape feature extraction in the time domain includes but is not limited to the envelope, DC bias value, peak-to-peak value, short-time average amplitude, long-time average amplitude, variance, skewness, and kurtosis. The calculation formula of the envelope is as follows:

[0033] Known curve function:

[0034] y=f(x)

[0035] The family of normals for a curve is represented as follows:

[0036]

[0037] Where (X,Y) are the moving coordinates of the normal, (x,y) are the points on the curve, and y is a function of x; y′ is the first derivative of y, and y″ is the second derivative of y.

[0038] The envelope of the normal family satisfies the equation system:

[0039]

[0040] The parametric equation of the envelope is obtained from the above solution:

[0041]

[0042] Substituting y′ and y″,

[0043]

[0044] Get the parametric equation of the envelope

[0045]

[0046] S312, shape-based frequency feature extraction

[0047] The shape feature extraction of frequency includes but is not limited to the number of frequency peaks, the center frequency value of the first spectrum peak, the first spectrum peak amplitude, the center frequency value of the second spectrum peak, the second spectrum peak amplitude, the center frequency value of the third spectrum peak, the third spectrum peak amplitude, the first power spectrum, the first power spectrum density, the second power spectrum, the second power spectrum density, the third power spectrum, and the third power spectrum density. The frequency peak threshold is set, and the statistics greater than the frequency peak threshold are the frequency peaks. The fast Fourier transform is used to obtain the center frequency value and amplitude, and the three frequency center values ​​with the largest amplitude are recorded. The calculation formulas for power spectrum and power spectrum density are as follows:

[0048]

[0049] Among them, P f is the power spectrum, DFT is the Fourier transform function that calculates the amplitude of frequency f, N is the number of signal samples, the above is the power spectrum of each sample at a certain frequency, and the calculation result is multiplied by the number of signal samples per unit time to calculate the power spectrum per unit time.

[0050]

[0051] Among them, P d is the power spectral density, P f is the power spectrum, DFT is the Fourier transform function that calculates the amplitude of frequency f, and N is the number of signal samples. is the cumulative sum of the power spectra of each frequency, and its value is also equal to That is, the accumulation of sampled data, x(i) is the sampled data, the above is the total power spectrum of each sample, and the calculation result is multiplied by the number of samples of the signal per unit time to calculate the total power spectrum per unit time.

[0052] S313. Shape-based periodic feature extraction

[0053] The shape-based period calculation formula is as follows:

[0054]

[0055] When the error between T1 and T2 is within the allowable range (the error range can be set, the default value is 1%), T is the calculation period. As a shape-based period, it must simultaneously satisfy the following conditions: the calculation results errors of the peak-to-peak value, short-time average amplitude and long-time average amplitude of the time domain-based shape at the start time t0 and the start time t0+T are within a certain range; and the calculation results errors of the number of frequency peaks, the center frequency value of the first spectrum peak and the first spectrum peak amplitude of the frequency-based shape at the start time t0 and t0+T are within the allowable range (the error range can be set, the default value is 1%).

[0056] S32, energy-based feature extraction; specifically including:

[0057] S321. Energy-based time domain feature extraction

[0058] Energy feature extraction in the time domain includes but is not limited to long-term average energy, short-term average energy, root mean square, amplitude accumulation, discharge time, etc. Long-term average energy is the energy average value calculated over a long period of time, and the time length can be set. Short-term average energy is the energy average value calculated over a short period of time that meets the discharge trigger condition. The discharge trigger condition is that the discharge power is greater than a certain threshold. The discharge time is the continuous discharge time when the energy value is greater than a certain threshold. The calculation formula for the discharge time is as follows:

[0059] When x(t1)x(t1)≥Setting, t1 is the discharge start time.

[0060] When x(t2)x(t2)≤Setting, t2 is the end time of discharge.

[0061] t discharge =t2-t1

[0062] In the formula, t1 is the start time of discharge, t2 is the end time of discharge, t discharge is the discharge time.

[0063] S322. Energy-based frequency feature extraction

[0064] The energy feature extraction of the frequency includes but is not limited to the first spectrum peak energy value, the second spectrum peak energy value, the third spectrum peak energy value, etc. The spectrum peak energy calculation formulas are as follows:

[0065] E f =DFT(x)·DFT(x)

[0066] Among them, E fis the spectrum peak energy value, x is the sampling data, which means the sampling data within a sampling time. DFT is the Fourier transform function that calculates the amplitude of the frequency f. If the sampling data x takes a long window time, the calculation result is the long-time spectrum peak energy value. If the sampling data x takes a short window time, the calculation result is the short-time spectrum peak energy value.

[0067] S323. Energy-based periodic feature extraction

[0068] The energy-based cycle calculation formula is as follows:

[0069] T1=DFT(x)·DFT(x)

[0070] T2=DFT(x+T)·DFT(x+T)

[0071] When the errors of T1 and T2 of the first spectrum peak energy value, the second spectrum peak energy value, and the third spectrum peak energy value are all within the allowable range (the error range can be set, and the default value is 1%), T is the calculation period, the sampling start time of the x data is t0, and the sampling start time of the x+T data is t0+T.

[0072] S33, feature extraction based on statistics; specifically including:

[0073] S331. Extraction of time domain features based on statistics

[0074] The statistical feature extraction in the time domain includes but is not limited to the initial phase angle of the positive and negative half-wave discharge, the duration of the positive and negative half-wave discharge, the number of positive and negative half-wave discharge pulses, the short-time zero-crossing rate, etc. The phase angle at which the positive and negative half-wave discharge starts is when the discharge amount is greater than a certain set value, and the discharge duration is greater than the set threshold (the set threshold can be adjusted), and the discharge duration is greater than the allowable error range (the error range can be set, the default value is 1%), that is, it is judged as the start time of the discharge, and the time from the start time of the discharge to the phase angle 0° signal is converted into the phase angle, which is the initial phase angle of the discharge. The initial phase angles of the positive and negative half-wave discharge are calculated separately. The calculation formula for the initial phase angle of the discharge is as follows:

[0075] When x(t1)≥Setting_dischage, t1 is the discharge start time.

[0076] When x(t2)≤Setting_dischage, t2 is the end time of discharge.

[0077] t discharge =t2-t1, t discharge Continuous discharge time

[0078] t discharge≥Setting_pulse, the continuous discharge time is greater than the set value Setting_pulse

[0079] t0 is the signal zero crossing time, T cycle Signal period.

[0080] S332. Frequency feature extraction based on statistics

[0081] The statistical feature extraction in the frequency domain includes but is not limited to power spectrum band statistics, frequency spectrum band information entropy statistics, etc. The calculation formula based on power spectrum band information entropy is as follows:

[0082]

[0083] Among them, E k is the band energy value, P k is the frequency band power spectrum, F n is the power spectrum band information entropy.

[0084] The calculation of UHF partial discharge characteristics, identification and location of fault anomaly types specifically include the following steps:

[0085] S1. Collect UHF partial discharge signals: Collect UHF partial discharge data of primary equipment of power grid on site.

[0086] S2. De-noising of UHF partial discharge signals: filtering out the white noise of UHF partial discharge data of primary equipment of power grid collected on-site.

[0087] S3. Extraction of ultra-high frequency partial discharge characteristics: same as described above.

[0088] S4. Logical judgment of normal operation data: judge whether it belongs to normal operation data according to the normal operation data calibration of different objects.

[0089] S5. Logical judgment of interference data: judge whether it is interference data according to the different interference data calibration of different objects.

[0090] S6. Identification and positioning of UHF partial discharge types: Error comparison algorithm, similarity algorithm and classification algorithm are used to identify the type of UHF partial discharge based on sub-modules such as shape-based time domain features, shape-based frequency domain features, shape-based periodic features, energy-based time domain features, energy-based frequency domain features, energy-based periodic features and statistics-based time domain features. The UHF partial discharge is positioned based on the size of the UHF signals collected at different locations, and spatial positioning is performed based on the UHF signals collected at different locations after time synchronization.

[0091] The error comparison algorithm is specifically as follows: calibrate the above calculation result data, compare the ultra-high frequency partial discharge data collected on-site or in experiments with the above calibration data, and use the error algorithm. When it meets the error range of the above calibration data, it is judged as this type of ultra-high frequency partial discharge.

[0092] The similarity algorithm is: calibrate the above calculation result data, compare the ultra-high frequency partial discharge data collected on-site or in experiments with the above calibration data, and use the similarity algorithm. When the error of the calculation result of the similarity algorithm is within the allowable range, it is judged as this type of ultra-high frequency partial discharge.

[0093] The classification algorithm is: calibrate the above calculation result data, compare the ultra-high frequency partial discharge data collected on-site or in the test with the above calibration data, and use a classification method, including but not limited to linear discriminant method, distance discriminant method, Bayesian classifier, decision tree, neural network ANN, support vector machine SVM and other classification methods. When the classification conditions of the classification method are met, it is judged as this type of ultra-high frequency partial discharge.

[0094] S7. Logical reasoning of UHF partial discharge characteristic quantities

[0095] The UHF partial discharge fault abnormality judgment logic includes three major logic modules, namely, the shape feature-based logic module, the energy feature-based logic module, and the statistical feature-based logic module. Among them, the shape feature-based logic module includes the shape-based time domain feature conditions, the shape-based frequency feature conditions, and the shape feature-based periodic conditions; the energy feature-based logic module includes the energy-based time domain feature conditions, the energy-based frequency feature conditions, and the energy feature-based periodic conditions; the statistical feature-based logic module includes the statistical-based time domain feature conditions and the statistical-based frequency feature conditions.

[0096] like Figure 3 As shown, the strong logic judgment condition is: the shape feature logic module, the energy feature logic module, and the statistical feature logic module use the "AND logic" condition judgment, that is, the three modules must meet the conditions to judge it as partial discharge.

[0097] like Figure 4 As shown, the strong logic judgment condition is: "three out of two logic" condition judgment based on shape feature logic module, energy feature logic module and statistical feature logic module, that is, two or more modules of the three modules meet the conditions and are judged as partial discharge.

[0098] like Figure 5As shown, the weak logic judgment condition is: "three-out-of-one logic" condition judgment based on the shape feature logic module, the energy feature logic module, and the statistical feature logic module. If any one of the three modules meets the condition, it is judged as partial discharge, that is, if one, two or three modules meet the condition, it is judged as partial discharge.

[0099] A device for identifying ultra-high frequency partial discharge types based on shape features, energy features and statistical features includes the following modules:

[0100] Module 1: UHF partial discharge signal acquisition module;

[0101] The main hardware circuits for UHF partial discharge signal acquisition include UHF receiving antenna, signal amplifier, filter, A / D analog-to-digital conversion module, etc.

[0102] The main software for collecting UHF partial discharge signals processes analog-to-digital conversion, converting analog quantities into digital quantities. To ensure that the analog quantities of UHF signals are not distorted, the sampling rate must be high enough.

[0103] Module 2: UHF partial discharge signal denoising processing module;

[0104] The ultra-high frequency partial discharge signal denoising uses a software algorithm to remove white noise as much as possible, leaving only the effective high-frequency partial discharge characteristic signal.

[0105] Module 3: UHF partial discharge feature extraction module;

[0106] 3.1 Shape-based time domain feature extraction

[0107] The shape feature extraction in the time domain includes but is not limited to the envelope, DC bias value, peak-to-peak value, short-time average amplitude, long-time average amplitude, variance, skewness, and kurtosis. The calculation formula of the envelope is as follows:

[0108] Known curve function:

[0109] y=f(x)

[0110] The normal family of this curve is expressed as follows:

[0111]

[0112] Here, (X,Y) are the moving coordinates of the normal, (x,y) are the points on the curve, and y is a function of x. y' is the first derivative of y, and y" is the second derivative of y.

[0113] The normal family envelope satisfies the following equations:

[0114]

[0115] The parametric equation of the envelope is obtained from the above solution:

[0116]

[0117] Substituting y′ and y″,

[0118]

[0119] Get the parametric equation of the envelope

[0120]

[0121] 3.2 Shape-based frequency feature extraction

[0122] The shape feature extraction of frequency includes but is not limited to the number of frequency peaks, the center frequency value of the first spectrum peak, the first spectrum peak amplitude, the center frequency value of the second spectrum peak, the second spectrum peak amplitude, the center frequency value of the third spectrum peak, the third spectrum peak amplitude, the first power spectrum, the first power spectrum density, the second power spectrum, the second power spectrum density, the third power spectrum, and the third power spectrum density. The frequency peak threshold is set, and the statistics greater than the frequency peak threshold are the frequency peaks. The fast Fourier transform is used to obtain the center frequency value and amplitude, and the three frequency center values ​​with the largest amplitude are recorded. The power spectrum and power spectrum density calculation formulas are as follows:

[0123]

[0124] Among them, P f is the power spectrum, DFT is the Fourier transform function that calculates the amplitude of frequency f, N is the number of signal samples, the above is the power spectrum of each sample at a certain frequency, and the calculation result is multiplied by the number of signal samples per unit time to calculate the power spectrum per unit time.

[0125]

[0126] Among them, P d is the power spectral density, P f is the power spectrum, DFT is the Fourier transform function that calculates the amplitude of frequency f, and N is the number of signal samples.

[0127] is the cumulative sum of the power spectra of each frequency, and its value is also equal to That is, the accumulation of sampled data, x(i) is the sampled data, the above is the total power spectrum of each sample, and the calculation result is multiplied by the number of samples of the signal per unit time to calculate the total power spectrum per unit time.

[0128] 3.3 Shape-based periodic feature extraction

[0129] The shape-based period calculation formula is as follows:

[0130]

[0131] When the error between T1 and T2 is within a certain range, T is the calculation period. As a shape-based period, it must simultaneously satisfy the following conditions: the calculation results errors of the peak-to-peak value, short-time average amplitude, and long-time average amplitude of the time-domain shape at the start time t0 and the start time t0+T are within a certain range; and the calculation results errors of the number of frequency peaks, the center frequency value of the first spectrum peak, and the first spectrum peak amplitude of the frequency-based shape at the start time t0 and t0+T are within a certain range.

[0132] 3.4 Energy-based time-domain feature extraction

[0133] Energy feature extraction in the time domain includes but is not limited to long-term average energy, short-term average energy, root mean square, amplitude accumulation, discharge time, etc. Long-term average energy is the energy average value calculated over a long period of time, and the time length can be set. Short-term average energy is the energy average value calculated over a short period of time that meets the discharge trigger condition. The discharge trigger condition is that the discharge power is greater than a certain threshold. The discharge time is the continuous discharge time when the energy value is greater than a certain threshold. The calculation formula for the discharge time is as follows:

[0134] When x(t1)x(t1)≥Setting, t1 is the discharge start time.

[0135] When x(t2)x(t2)≤Setting, t2 is the end time of discharge.

[0136] t discharge =t2-t1

[0137] In the above formula, t1 is the discharge start time, t2 is the discharge end time, t discharge is the discharge time.

[0138] 3.5 Energy-based frequency feature extraction

[0139] The energy feature extraction of the frequency includes but is not limited to the first spectrum peak energy value, the second spectrum peak energy value, the third spectrum peak energy value, etc. The spectrum peak energy calculation formulas are as follows:

[0140] E f =DFT(x)·DFT(x)

[0141] Among them, E fis the spectrum peak energy value, x is the sampling data, which means the sampling data within a sampling time. DFT is the Fourier transform function that calculates the amplitude of the frequency f. If the sampling data x takes a long window time, the calculation result is the long-time spectrum peak energy value. If the sampling data x takes a short window time, the calculation result is the short-time spectrum peak energy value.

[0142] 3.6 Energy-based periodic feature extraction

[0143] The energy-based cycle calculation formula is as follows:

[0144] T1=DFT(x)·DFT(x)

[0145] T2=DFT(x+T)·DFT(x+T)

[0146] When the errors of T1 and T2 of the first spectrum peak energy value, the second spectrum peak energy value, and the third spectrum peak energy value are all within a certain range, T is the calculation period, the sampling start time of the x data is t0, and the sampling start time of the x+T data is t0+T.

[0147] 3.7 Statistical-based time domain feature extraction

[0148] The statistical feature extraction in the time domain includes but is not limited to the initial phase angle of the positive and negative half-wave discharge, the duration of the positive and negative half-wave discharge, the number of positive and negative half-wave discharge pulses, the short-time zero-crossing rate, etc. The phase angle of the positive and negative half-wave discharge is when the discharge amount is greater than a certain set value and the discharge duration is greater than a certain set value, which is the start time of the discharge. The time from the start time of the discharge to the phase angle 0° signal is converted into the phase angle, which is the initial phase angle of the discharge. The initial phase angles of the positive and negative half-wave discharge are calculated separately. The calculation formula for the initial phase angle of the discharge is as follows:

[0149] When x(t1)≥Setting_dischage, t1 is the discharge start time.

[0150] When x(t2)≤Setting_dischage, t2 is the end time of discharge.

[0151] t discharge =t2-t1, t discharge Continuous discharge time

[0152] t discharge ≥Setting_pulse, the continuous discharge time is greater than the set value Setting_pulse

[0153] t0 is the signal zero crossing time, T cycle Signal period

[0154] 3.8 Frequency feature extraction based on statistics

[0155] The statistical feature extraction in the frequency domain includes but is not limited to power spectrum band statistics, frequency spectrum band information entropy statistics, etc. The calculation formula based on power spectrum band information entropy is as follows:

[0156]

[0157] Where E k is the band energy value, P k is the frequency band power spectrum, F n is the power spectrum band information entropy.

[0158] Module 4: UHF partial discharge characteristic quantity identification algorithm module;

[0159] 4.1 Error comparison algorithm

[0160] The above calculation result data is calibrated, and the UHF partial discharge data collected on-site or in the test is compared with the above calibration data. An error algorithm is used. When it meets the error range of the above calibration data, it is judged as this type of UHF partial discharge.

[0161] 4.2 Similarity Algorithm

[0162] The above calculation result data is calibrated, and the UHF partial discharge data collected on-site or in the test is compared with the above calibration data. A similarity algorithm is used. When the error of the calculation result of the similarity algorithm is within the allowable range, it is judged as this type of UHF partial discharge.

[0163] 4.3 Classification Algorithm

[0164] The above calculation result data are calibrated, and the UHF partial discharge data collected on-site or in experiments are compared with the above calibration data. A classification method is used, including but not limited to linear discriminant method, distance discriminant method, Bayesian classifier, decision tree, neural network ANN, support vector machine SVM and other classification methods. When the classification conditions of the classification method are met, it is judged as this type of UHF partial discharge.

[0165] Module 5: UHF partial discharge characteristic quantity logic reasoning module;

[0166] The UHF partial discharge fault abnormality judgment logic includes three major logic modules, namely, a shape feature-based logic module, an energy feature-based logic module, and a statistical feature-based logic module. The shape feature-based logic module includes time domain feature conditions based on shape, frequency feature conditions based on shape, and periodic conditions based on shape features. The energy feature-based logic module includes time domain feature conditions based on energy, frequency feature conditions based on energy, and periodic conditions based on energy features. The statistical feature-based logic module includes time domain feature conditions based on statistics and frequency feature conditions based on statistics.

[0167] 4.1 Strong logical judgment conditions

[0168] The shape feature logic module, the energy feature logic module and the statistical feature logic module adopt the "AND logic" conditional judgment, that is, partial discharge is judged only when the three modules meet the conditions.

[0169] 4.2 Strong logical judgment conditions

[0170] The "two out of three logic" conditional judgment based on the shape feature logic module, the energy feature logic module, and the statistical feature logic module, that is, if two or more modules of the three modules meet the conditions, it is judged as partial discharge.

[0171] 4.3 Weak logical judgment conditions

[0172] Based on the "three-out-of-one logic" condition judgment of the shape feature logic module, the energy feature logic module and the statistical feature logic module, if any one of the three modules meets the conditions, it is judged as partial discharge, that is, if one, two or three of the three modules meet the conditions, it is judged as partial discharge.

[0173] The present invention also provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the steps of the above method are implemented when the processor executes the computer program.

[0174] The present invention also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the above method are implemented.

Claims

1. A method for identifying ultra-high frequency partial discharge based on shape, energy and statistical characteristics, characterized in that: include: UHF PD feature calibration based on shape, energy and statistics: including reading UHF PD data and signal processing, UHF PD feature extraction and calibration; UHF partial discharge feature calculation, fault anomaly type identification and positioning: including collecting UHF partial discharge signals and signal processing; UHF partial discharge feature extraction; normal operation data and interference data logical discrimination; UHF partial discharge type identification and positioning and UHF partial discharge feature logical reasoning; the UHF partial discharge feature extraction includes shape-based feature extraction, energy-based feature extraction and statistics-based feature extraction, the shape-based feature extraction includes shape-based time domain feature extraction, shape-based frequency domain feature extraction, shape-based period extraction, the energy-based feature extraction includes energy-based time domain feature extraction, energy-based frequency domain feature extraction, energy-based period extraction, the statistics-based feature extraction includes statistics-based time domain feature extraction and statistics-based frequency domain feature extraction; The ultra-high frequency partial discharge type identification and positioning are specifically as follows: an error comparison algorithm, a similarity algorithm, and a classification algorithm are used to identify the ultra-high frequency partial discharge type based on shape-based time domain features, shape-based frequency domain features, shape-based periodic features, energy-based time domain features, energy-based frequency domain features, energy-based periodic features, and statistical time domain features, and positioning is performed according to the size of ultra-high frequency signals collected at different positions, and spatial positioning is performed according to ultra-high frequency signals collected at different positions after time synchronization.

2. The method for identifying ultra-high frequency partial discharge based on shape, energy and statistical characteristics according to claim 1, characterized in that: The ultra-high frequency partial discharge characteristic quantity calibration includes shape-based time domain characteristic calibration, shape-based frequency domain characteristic calibration, shape-based period calibration, energy-based time domain characteristic calibration, energy-based frequency domain characteristic calibration, energy-based period calibration, statistics-based time domain characteristic calibration, and statistics-based frequency domain characteristic calibration.

3. The method for identifying ultra-high frequency partial discharge based on shape, energy and statistical characteristics according to claim 1, characterized in that: The ultra-high frequency partial discharge fault abnormality judgment logic includes a shape feature-based logic module, an energy feature-based logic module, and a statistical feature-based logic module.

4. The method for identifying ultra-high frequency partial discharge based on shape, energy and statistical characteristics according to claim 3 is characterized in that: The ultra-high frequency partial discharge fault abnormality judgment logic adopts three different logic judgments: "AND logic", "three-out-of-two logic" and "three-out-of-one logic".

Images