Method for identifying partial discharge mode of high-voltage switch cabinet based on pulse current method

Abstract

The invention discloses a method for identifying a partial discharge mode of a high-voltage switch cabinet based on a pulse current method. The method comprises the following steps of: 1, constructing a typical insulation defect physical model for a partial discharge simulation experiment aiming at a common discharge phenomenon in the high-voltage switch cabinet; 2, building a partial discharge simulation experiment platform, and collecting a one-dimensional time domain signal when the typical insulation defect physical model generates a partial discharge phenomenon; 3, drawing PRPD spectrograms of different defect models at different moments according to the one-dimensional time domain signal, and establishing a data set; step 4, after a data set is established, extracting characteristic parameters in the PRPD spectrogram; and 5, carrying out classification identification on the characteristic parameters by using a BP neural network algorithm, and realizing identification of the partial discharge mode of the high-voltage switch cabinet based on a pulse current method. According to the invention, the partial discharge data acquired based on the pulse current sensor can be identified so as to discriminate the partial discharge type in the high-voltage switch cabinet, and the identification accuracy can reach 99.53%.

Description

Technical Field

[0001] The present invention belongs to the technical field of partial discharge detection of high-voltage switch cabinets, and in particular relates to a method for identifying partial discharge patterns of high-voltage switch cabinets based on a pulse current method. Background Art

[0002] High-voltage switchgear, a crucial foundational element of power distribution networks, primarily receives and distributes electrical energy. With the development of power systems, the number of switchgear connected to the grid continues to increase, making it the most widely used and numerous switchgear device. Therefore, the safe operation of high-voltage switchgear is crucial to the stability of the entire power system.

[0003] However, during the design, manufacturing, transportation, installation, commissioning, and operation of high-voltage switchgear, typical insulation defects such as metal residue (burrs), insulator air gaps, metal particles, insulator surface contamination, poor contact, or magnetic short circuits inevitably occur within it. These defects distort the original electric field and create localized areas of high field strength. When the electric field strength in a local area exceeds the breakdown field strength, partial discharge (PD) occurs in that area while other areas maintain their insulation properties, resulting in a phenomenon called partial discharge.

[0004] Partial discharge does not form a complete discharge channel and is considered non-penetrating discharge. The impact of partial discharge on the overall electrical performance of insulation can usually be ignored. However, if timely measures are not taken, long-term partial discharge will further damage the insulation structure and cause insulation aging. The insulation failure area will gradually expand from a small area to a large area, eventually developing into a penetrating breakdown or surface flashover, causing sudden insulation failure.

[0005] According to a statistical analysis of fault types in switchgear equipment below 40.5 kV conducted by the China Electric Power Research Institute, insulation and current-carrying faults (including obstructions) account for 30% to 53%. Statistics from the Guangdong Power Grid Company on switchgear equipment failure types show that insulation and current-carrying faults account for as high as 66%. Both of these faults are closely related to partial discharge (PD). Therefore, research on partial discharge monitoring technology for high-voltage switchgear is crucial for reducing failures.

[0006] Currently, the main methods for monitoring PD in high-voltage switchgear include high-frequency current, ultra-high frequency, ultrasonic, and transient ground voltage methods. The high-frequency current method requires the installation of a high-frequency sensor on the ground wire of the cable body or cable terminal connector. However, in practice, due to the switchgear structure and installation location, it is difficult for inspectors to enter the cable trench, and there are safety risks. This makes high-frequency PD detection difficult. Ultra-high frequency, ultrasonic, and transient ground voltage methods all require external testing and are generally susceptible to external acoustic and electrical interference. Furthermore, high-voltage switchgear is well sealed, making it difficult for internal discharge signals to propagate, potentially leading to missed PD detection.

[0007] In recent years, researchers have proposed an online partial discharge monitoring method for high-voltage switchgear based on the pulse current method and capacitor ceramic insulators. This method, developed for switchgear and ring main units, exploits the fact that the "capacitance" within the capacitor ceramic insulators used in high-voltage switchgear provides a circuit for the pulse current generated by partial discharge. This method utilizes a "direct" measurement method and is capable of accurate quantitative measurement. However, it is worth noting that differences in the nature of insulation defects can lead to different types of partial discharge, posing varying degrees of threat to the insulation condition of the equipment. Therefore, accurately identifying partial discharge patterns is crucial for guiding maintenance work. Summary of the Invention

[0008] The present invention aims to address the shortcomings of existing technologies by providing a method for identifying partial discharge patterns in high-voltage switchgear based on a pulse current method. This method extracts features from partial discharges collected by a pulse current partial discharge sensor using a statistical feature parameter method, and then classifies and identifies these features using a BP neural network method to achieve partial discharge pattern recognition in high-voltage switchgear.

[0009] In order to achieve the above object, the present invention adopts the following technical solutions:

[0010] A method for identifying partial discharge patterns of high-voltage switchgear based on a pulse current method comprises the following steps:

[0011] Step 1: Based on the common surface discharge, tip discharge, internal discharge, and suspension discharge inside the high-voltage switchgear, a typical insulation defect physical model for partial discharge simulation experiments was constructed;

[0012] Step 2: Build a partial discharge simulation experimental platform to collect one-dimensional time domain signals when partial discharge occurs in a typical insulation defect physical model;

[0013] Step 3: Draw PRPD spectra of different defect models at different times based on the one-dimensional time domain signal collected in step 2, and establish a high-voltage switchgear partial discharge dataset based on the pulse current method;

[0014] Step 4: After the data set is established, the characteristic parameters in the PRPD spectrum are extracted;

[0015] Step 5: Use the BP neural network algorithm to classify and identify the characteristic parameters, and realize the identification of the partial discharge mode of the high-voltage switchgear based on the pulse current method.

[0016] As a further illustration of the present invention, in step 2, the partial discharge simulation experimental platform includes a voltage-stabilized power supply, a test transformer control box, an industrial frequency test transformer, a protective resistor, an artificial simulation defect module, a capacitor ceramic insulator, a detection impedance and a pulse current partial discharge sensor; the voltage-stabilized power supply, the test transformer control box and the industrial frequency test transformer are electrically connected in sequence; the artificial simulation defect module is connected to the industrial frequency test transformer through the protective resistor; and the pulse current partial discharge sensor is connected to the artificial simulation defect module through the capacitor ceramic insulator and the detection impedance.

[0017] In the present invention, the sensor for collecting partial discharge data is a pulse current partial discharge sensor with an online monitoring function. Compared with non-invasive and indirect partial discharge detection methods such as ultra-high frequency detection method, ultrasonic detection method, chemical detection method, and optical detection method, it has many advantages such as high sensitivity and quantitative detection. It is also the only partial discharge detection method currently with the national standard "GB / T7354-2018 High Voltage Test Technology-Partial Discharge Measurement" as a constraint basis.

[0018] As a further illustration of the present invention, the pulse current partial discharge sensor is provided with a LoRa communication antenna.

[0019] As a further illustration of the present invention, step three specifically involves performing a "matrix reorganization" transformation on the time domain waveform of the one-dimensional time domain signal collected in step two, transforming it from a 1×3600 array to a 50×72 array, and then plotting "phase resolved partial discharge" (PRPD) spectra corresponding to different defect models when partial discharge occurs.

[0020] As a further illustration of the present invention, in step 3, the PRPD spectrum drawn is Spectrum.

[0021] PRPD spectrum is a statistical spectrum, also known as model, which in turn led to Spectrum, Spectrum, Spectrum, Different discharge types have different performances on each spectrum. The discharge type can be determined by analyzing the difference in shape and contour of the positive and negative half-cycles of each spectrum. The discharge type is determined by the shape difference characteristics and contour difference characteristics of the positive and negative half-cycles of the spectrum.

[0022] Because the pulse current partial discharge sensor belongs to a linear system, the detection impedance Z d The pulse voltage signal U collected at both ends d It is proportional to the apparent discharge quantity q of the test object; therefore, The spectrum can be evolved into Spectrum.

[0023] As a further illustration of the present invention, in step 4, the extracted characteristic parameters include: The skewness Sk of the spectrum (including the skewness of the positive and negative half cycles and the entire power frequency cycle), The steepness Ku of the spectrum (including the steepness of the positive and negative half cycles and the entire power frequency cycle), The phase median Mv of the spectrum (including the phase median of the positive and negative half-power frequency cycles), The number of partial discharge peak points Pe of the spectrum (including the number of partial discharge peak points of positive and negative half-power frequency cycles), The cross-correlation coefficient cc and the modified cross-correlation coefficient mcc of the spectrum.

[0024] The skewness Sk of the spectrum reflects the left-right skewness of the spectrum shape relative to the normal distribution. That is, if Sk = 0, it means that the spectrum shape is symmetrical; if Sk > 0, it means that the spectrum shape is skewed to the left relative to the normal distribution shape; if Sk < 0, it means that the spectrum shape is skewed to the right relative to the normal distribution shape. Positive half cycle Skewness of the spectrum Sk + for:

[0025]

[0026] Where W + =36 is the number of phase windows in the positive half cycle; x j+ =j×Δx-Δx / 2 is the phase of the jth phase window in the positive half cycle, Δx=5° is the phase window width; p j+ is the probability density, that is, the proportion of the value on the jth phase window in the total value of the positive half cycle; μ + is the expectation of the probability density distribution; δ + is the variance of the probability density distribution.

[0027]

[0028] Where yj+ It is the vertical coordinate value of each spectrum on the j-th phase window under the positive half cycle of the power frequency.

[0029] The steepness of the spectrum Ku is used to describe the degree of protrusion of a certain shape compared to the normal distribution shape. That is, if Ku = 0, it means that the spectrum profile is a standard normal distribution; if Ku > 0, it means that the spectrum profile is sharper and steeper than the normal distribution profile; if Ku < 0, it means that the spectrum profile is flatter than the normal distribution profile. Positive half cycle The steepness of the spectrum Ku + for:

[0030]

[0031] The phase median Mv of the spectrum refers to the median value calculated for the extracted discharge frequency and the phase corresponding to the discharge frequency. From the perspective of probability distribution, it reflects the location of the point with the densest average distribution. Positive half cycle The phase median Mv of the spectrum + The calculation formula is the same as the positive half cycle The steepness of the spectrum Ku + Calculation formula.

[0032] The number of local discharge peak points Pe of the spectrum is used to describe the number of local peaks on the spectrum profile. For the partial discharge fingerprint, each phase window (x i ,y i ) is a local peak can be determined according to the following formula:

[0033] and

[0034] Convert the above formula into difference form, then:

[0035] and

[0036] Considering that the phase is an increasing sequence, the above formula can be simplified to:

[0037] y i -y i-1 >0 and y i+1 -y i <0.

[0038] The cross-correlation coefficient cc of the spectrum reflects the similarity of the shape of the spectrum in the positive and negative half cycles, that is, if cc = 1, The profiles of the positive and negative half cycles of the spectrum are exactly the same; if cc=0, The contours of the positive and negative half cycles of the spectrum are completely inconsistent.

[0039]

[0040] Where, W=72 is the number of phase windows in the entire cycle; W + =36 is the number of phase windows in the positive half cycle; W － =36 is the number of phase windows in the negative half cycle.

[0041] The correlation coefficient of the spectrum reflects the linear correlation between the positive and negative half cycles. The discharge amount and discharge frequency of the positive and negative half cycles are also different. Therefore, the operator F is introduced to reflect the difference in the amount of discharge activity in the positive and negative half cycles:

[0042]

[0043] (k) Furthermore, the modified cross-correlation coefficient mcc is:

[0044] mcc=F×cc.

[0045] Once the characteristic parameters are confirmed, a classification algorithm can be used to achieve partial discharge pattern recognition. Commonly used classifiers for pattern recognition include Bayesian classifiers, linear classifiers, and fuzzy recognition classifiers. Recent research on partial discharge pattern recognition has shown the widespread application of artificial neural networks. Therefore, the present invention preferably employs a BP neural network algorithm for partial discharge pattern recognition.

[0046] As a further illustration of the present invention, in step 5, the classification and identification of characteristic parameters using the BP neural network algorithm includes:

[0047] Step 1), build a BP neural network;

[0048] Step 2), divide the data set into training set and test set;

[0049] Step 3) Input the training set into the BP neural network to perform model training. During the training process, the model parameters are continuously adjusted to obtain the optimal network model.

[0050] In step 4), the test set is input into the optimal network model for testing, and the confusion matrix is analyzed to verify the feasibility of using the BP neural network to identify the partial discharge type.

[0051] In practical applications, the new data collected by the pulse current partial discharge sensor can be directly used to draw the PRPD spectrum. After the characteristic parameters are extracted, they are sent to the BP neural network for discrimination.

[0052] In the present invention, the BP neural network construction steps are:

[0053] The BP neural network consists of an input layer, a hidden layer, and an output layer. The characteristic parameters extracted in step 4 of the method indicate that the number of neurons in the input layer is 12, i.e., the 12 statistical characteristic parameters in Table 1. The number of neurons in the output layer is 5, i.e., normal, tip discharge, suspended discharge, surface discharge, and internal discharge, for a total of 5 discharge modes. The number of neurons in the hidden layer can be determined using the following empirical formula:

[0054]

[0055] Where N n is the number of neurons in the hidden layer; N i is the number of neurons in the input layer; N o is the number of neurons in the output layer; ε is a parameter ranging from 1 to 10, and ε=10 is taken in the present invention.

[0056] The relationship between the hidden layer and the input layer can be expressed as:

[0057] [Sk + Sk - L cc]v 1 +b 1

[0058] Where, [Sk + Sk - L cc] is the input layer matrix; v 1 is the weight between the first layer of neurons and the second layer of neurons

[0059] matrix; b 1 is the bias matrix between the first layer neurons and the second layer neurons;

[0060]

[0061] The hidden layer activation function is chosen to be tansig(x). Therefore, the output of the first neuron in the hidden layer is:

[0062]

[0063] Similarly, the relationship between the hidden layer and the output layer can be expressed as:

[0064] [U1 U2 LU n ]v 2 +b 2 ;

[0065] Where, v 2 is the weight matrix between the second layer neurons and the third layer neurons; b 2 For the second layer of neurons and the third layer

[0066] Bias matrix between neurons;

[0067]

[0068] The activation function of the output layer is logsig(x). Therefore, the output of the first neuron in the output layer of the BP neural network is:

[0069]

[0070] In the present invention, the step 2) specifically includes: dividing the partial discharge data collected by the partial discharge simulation experimental platform into 80% as a training set and 20% as a test set.

[0071] Beneficial effects of the present invention:

[0072] The method of the present invention can realize the recognition of partial discharge data collected by the pulse current sensor to determine the type of partial discharge in the high-voltage switch cabinet, and the recognition accuracy can be as high as 99.53%. BRIEF DESCRIPTION OF THE DRAWINGS

[0073] Figure 1 It is a typical insulation defect physical model;

[0074] Figure 2 It is a partial discharge simulation experimental platform;

[0075] Figure 3 The partial discharge mode spectrum is obtained by analyzing the time domain waveform and phase when there is no discharge, i.e. normal state;

[0076] Figure 4 Partial discharge pattern spectrum for tip discharge time domain waveform and phase analysis;

[0077] Figure 5 Partial discharge pattern spectrum for suspension discharge time domain waveform and phase analysis;

[0078] Figure 6 Partial discharge pattern spectrum for surface discharge time domain waveform and phase analysis;

[0079] Figure 7 Partial discharge pattern spectrum for internal discharge time domain waveform and phase analysis;

[0080] Figure 8 It is a summary of statistical characteristic parameters;

[0081] Figure 9 A partial discharge pattern recognition model for high-voltage switchgear based on the pulse current method;

[0082] Figure 10 This is the training flow chart of BP neural network;

[0083] Figure 11is the confusion matrix.

[0084] exist Figure 2 In the diagram, 1- voltage regulated power supply, 2- power frequency test transformer T, 3- protection resistor R r , 4-artificial simulation defect module, 5-capacitor ceramic insulator, 6-detection impedance Z d , 7-Pulse current partial discharge sensor M, 8-LoRa communication antenna. DETAILED DESCRIPTION

[0085] The present invention will be further described below with reference to the accompanying drawings.

[0086] Example:

[0087] A method for identifying partial discharge patterns of high-voltage switchgear based on a pulse current method comprises the following steps:

[0088] Step 1: For the common surface discharge, tip discharge, internal discharge and suspension discharge inside the high-voltage switchgear, a typical insulation defect physical model for partial discharge simulation experiment is constructed (such as Figure 1 shown);

[0089] Step 2: Build a partial discharge simulation experimental platform (such as Figure 2 As shown in Figure 2, one-dimensional time domain signals are collected when partial discharge occurs in a typical insulation defect physical model;

[0090] Step 3: Draw the PRPD spectra of different defect models at different times based on the one-dimensional time domain signal collected in step 2 (such as Figure 3-Figure 7 As shown in Figure 2), a partial discharge dataset of high-voltage switchgear was established based on the pulse current method;

[0091] Step 4: After the data set is established, the characteristic parameters in the PRPD spectrum are extracted;

[0092] Step 5: Use the BP neural network algorithm to classify and identify the characteristic parameters, and realize the identification of the partial discharge mode of the high-voltage switchgear based on the pulse current method.

[0093] This embodiment further illustrates that in step 2, if Figure 2 As shown, the partial discharge simulation experimental platform includes a voltage stabilized power supply 1, a test transformer control box, a power frequency test transformer T2, a protective resistor R r 3. Artificial simulation defect module 4. Capacitor ceramic insulator 5. Detection impedance Z d 6 and pulse current partial discharge sensor M 7; the voltage stabilized power supply 1, test transformer control box, power frequency test transformer T2 are electrically connected in sequence; the artificial simulation defect module 4 is connected through the protective resistor R r3 is connected to the power frequency test transformer T2; the pulse current partial discharge sensor M7 ... d 6 is connected to the artificial simulation defect module 4.

[0094] This embodiment further illustrates that the pulse current partial discharge sensor M 7 is provided with a LoRa communication antenna 8 .

[0095] This embodiment further illustrates that step three specifically involves performing a "matrix reorganization" transformation on the time domain waveform of the one-dimensional time domain signal collected in step two, transforming it from a 1×3600 array to a 50×72 array, and then plotting "phase resolved partial discharge" (PRPD) spectra corresponding to different defect models when partial discharge occurs.

[0096] PRPD spectrum is a statistical spectrum, also known as model, which in turn led to Spectrum, Spectrum, Spectrum, Spectrum. Different discharge types have different performances on each spectrum. The discharge type can be determined by analyzing the difference in shape and contour of the positive and negative half-cycles of the above spectra. The discharge type is determined by the shape difference characteristics and contour difference characteristics of the positive and negative half-cycles of the spectrum.

[0097] Because the pulse current partial discharge sensor belongs to a linear system, the detection impedance Z d The pulse voltage signal U collected at both ends d It is proportional to the apparent discharge quantity q of the test object; therefore, The spectrum can be evolved into Spectrum.

[0098] In this embodiment, in step 4, if Figure 8 As shown in Figure 2, the extracted feature parameters include: The skewness Sk of the spectrum (including the skewness of the positive and negative half cycles and the entire power frequency cycle), The steepness Ku of the spectrum (including the steepness of the positive and negative half cycles and the entire power frequency cycle), The phase median Mv of the spectrum (including the phase median of the positive and negative half-power frequency cycles), The number of partial discharge peak points Pe of the spectrum (including the number of partial discharge peak points of positive and negative half-power frequency cycles), The cross-correlation coefficient cc and the modified cross-correlation coefficient mcc of the spectrum.

[0099] The skewness Sk of the spectrum reflects the left-right skewness of the spectrum shape relative to the normal distribution. That is, if Sk = 0, it means that the spectrum shape is symmetrical; if Sk > 0, it means that the spectrum shape is skewed to the left relative to the normal distribution shape; if Sk < 0, it means that the spectrum shape is skewed to the right relative to the normal distribution shape. Positive half cycle Skewness of the spectrum Sk + for:

[0100]

[0101] Where W + =36 is the number of phase windows in the positive half cycle; x j+ =j×Δx-Δx / 2 is the phase of the jth phase window in the positive half cycle, Δx=5° is the phase window width; p j+ is the probability density, that is, the proportion of the value on the jth phase window in the total value of the positive half cycle; μ + is the expectation of the probability density distribution; δ + is the variance of the probability density distribution.

[0102]

[0103] Where y j+ It is the vertical coordinate value of each spectrum on the j-th phase window under the positive half cycle of the power frequency.

[0104] The steepness of the spectrum Ku is used to describe the degree of protrusion of a certain shape compared to the normal distribution shape. That is, if Ku = 0, it means that the spectrum profile is a standard normal distribution; if Ku > 0, it means that the spectrum profile is sharper and steeper than the normal distribution profile; if Ku < 0, it means that the spectrum profile is flatter than the normal distribution profile. Positive half cycle The steepness of the spectrum Ku + for:

[0105]

[0106] The phase median Mv of the spectrum refers to the median value calculated for the extracted discharge frequency and the phase corresponding to the discharge frequency. From the perspective of probability distribution, it reflects the location of the point with the densest average distribution. Positive half cycle The phase median Mv of the spectrum + The calculation formula is the same as the positive half cycle The steepness of the spectrum Ku + Calculation formula.

[0107] The number of local discharge peak points Pe of the spectrum is used to describe the number of local peaks on the spectrum profile. For the partial discharge fingerprint, each phase window (xi ,y i ) is a local peak can be determined according to the following formula:

[0108] and

[0109] Convert the above formula into difference form, then:

[0110] and

[0111] Considering that the phase is an increasing sequence, the above formula can be simplified to:

[0112] y i -y i-1 >0 and y i+1 -y i <0.

[0113] The cross-correlation coefficient cc of the spectrum reflects the similarity of the shape of the spectrum in the positive and negative half cycles, that is, if cc = 1, The profiles of the positive and negative half cycles of the spectrum are exactly the same; if cc=0, The contours of the positive and negative half cycles of the spectrum are completely inconsistent.

[0114]

[0115] Where, W=72 is the number of phase windows in the entire cycle; W + =36 is the number of phase windows in the positive half cycle; W － =36 is the number of phase windows in the negative half cycle.

[0116] The correlation coefficient of the spectrum reflects the linear correlation between the positive and negative half cycles. The discharge amount and discharge frequency of the positive and negative half cycles are also different. Therefore, the operator F is introduced to reflect the difference in the amount of discharge activity in the positive and negative half cycles:

[0117]

[0118] (k) Furthermore, the modified cross-correlation coefficient mcc is:

[0119] mcc=F×cc.

[0120] Once the characteristic parameters are confirmed, a classification algorithm can be used to achieve partial discharge pattern recognition. Commonly used classifiers for pattern recognition include Bayesian classifiers, linear classifiers, and fuzzy recognition classifiers. Recent research on partial discharge pattern recognition has shown the widespread application of artificial neural networks. Therefore, this embodiment preferably employs a BP neural network algorithm for partial discharge pattern recognition.

[0121] Further explanation of this embodiment, in step five, if Figures 9-11 As shown, the classification and identification of characteristic parameters using the BP neural network algorithm includes:

[0122] Step 1), build a BP neural network;

[0123] Step 2), divide the dataset into training set (80%) and test set (20%);

[0124] Step 3) Input the training set into the BP neural network to perform model training. During the training process, the model parameters are continuously adjusted to obtain the optimal network model.

[0125] In step 4), the test set is input into the optimal network model for testing, and the confusion matrix is analyzed to verify the feasibility of using the BP neural network to identify the partial discharge type.

[0126] In practical applications, the new data collected by the pulse current partial discharge sensor can be directly used to draw the PRPD spectrum. After the characteristic parameters are extracted, they are sent to the BP neural network for discrimination.

[0127] The steps to build a BP neural network are:

[0128] like Figure 9 As shown, the BP neural network consists of an input layer, a hidden layer, and an output layer. The characteristic parameters extracted in step 4 of the method indicate that the number of neurons in the input layer is 12, i.e., the 12 statistical characteristic parameters in Table 1. The number of neurons in the output layer is 5, i.e., normal, tip discharge, suspended discharge, surface discharge, and internal discharge, for a total of 5 discharge modes. The number of neurons in the hidden layer can be determined using the following empirical formula:

[0129]

[0130] Where N n is the number of neurons in the hidden layer; N i is the number of neurons in the input layer; N o is the number of neurons in the output layer; ε is a parameter ranging from 1 to 10, and ε=10 is taken in the present invention.

[0131] The relationship between the hidden layer and the input layer can be expressed as:

[0132] [Sk + Sk - L cc]v 1 +b 1

[0133] Where, [Sk + Sk - L cc] is the input layer matrix; v1 is the weight between the first layer of neurons and the second layer of neurons

[0134] matrix; b 1 is the bias matrix between the first layer neurons and the second layer neurons;

[0135]

[0136] The hidden layer activation function is chosen to be tansig(x). Therefore, the output of the first neuron in the hidden layer is:

[0137]

[0138] Similarly, the relationship between the hidden layer and the output layer can be expressed as:

[0139] [U1 U2 LU n ]v 2 +b 2 ;

[0140] Where, v 2 is the weight matrix between the second layer neurons and the third layer neurons; b 2 For the second layer of neurons and the third layer

[0141] Bias matrix between neurons;

[0142]

[0143] The activation function of the output layer is logsig(x). Therefore, the output of the first neuron in the output layer of the BP neural network is:

[0144]

[0145] The confusion matrix of the recognition results on the test set is as follows Figure 11 As shown in the table, the model achieved an overall recognition accuracy of 99.53% for the test set, and 100.00% accuracy for tip discharge, suspended discharge, and internal discharge. Of the 122 normal samples, 1 was identified as creeping discharge; and of the 232 internal discharge samples, 1 was identified as normal and 1 as creeping discharge. Therefore, the proposed method has a high overall accuracy of 99.53% for partial discharge pattern recognition, effectively distinguishing several typical insulation defect types.

[0146] In subsequent practical applications of this embodiment, the new data collected by the pulse current partial discharge sensor can be directly used to draw a PRPD spectrum, and after the characteristic parameters are extracted, they are sent to the BP neural network for discrimination.

[0147] Obviously, the above embodiments are merely examples for the purpose of clearly illustrating the present invention and are not intended to limit the implementation of the present invention. Those skilled in the art will readily appreciate that other variations or modifications may be made based on the above description. It is not necessary and impossible to enumerate all possible implementations here. Obvious variations or modifications derived therefrom remain within the scope of protection of the present invention.

Claims

1. A method for identifying partial discharge patterns of high-voltage switchgear based on a pulse current method, characterized in that: The following steps are involved: Step 1: Based on the common surface discharge, tip discharge, internal discharge, and suspension discharge inside the high-voltage switchgear, a typical insulation defect physical model for partial discharge simulation experiments was constructed; Step 2: Build a partial discharge simulation experimental platform to collect one-dimensional time domain signals when partial discharge occurs in a typical insulation defect physical model; Step 3: Draw PRPD spectra of different defect models at different times based on the one-dimensional time domain signal collected in step 2, and establish a high-voltage switchgear partial discharge dataset based on the pulse current method; Step 4: After the data set is established, the characteristic parameters in the PRPD spectrum are extracted; Step 5: Use the BP neural network algorithm to classify and identify the characteristic parameters, and realize the identification of the partial discharge mode of the high-voltage switchgear based on the pulse current method.

2. The method for identifying partial discharge patterns of high-voltage switchgear based on the pulse current method according to claim 1, characterized in that: In step 2, the partial discharge simulation experimental platform includes a voltage-stabilized power supply, a test transformer control box, an industrial frequency test transformer, a protective resistor, an artificial simulation defect module, a capacitor ceramic insulator, a detection impedance and a pulse current partial discharge sensor; the voltage-stabilized power supply, the test transformer control box and the industrial frequency test transformer are electrically connected in sequence; the artificial simulation defect module is connected to the industrial frequency test transformer through the protective resistor; the pulse current partial discharge sensor is connected to the artificial simulation defect module through the capacitor ceramic insulator and the detection impedance.

3. The method for identifying partial discharge patterns of high-voltage switchgear based on the pulse current method according to claim 2, characterized in that: The pulse current partial discharge sensor is provided with a LoRa communication antenna.

4. The method for identifying partial discharge patterns of high-voltage switchgear based on the pulse current method according to claim 1, characterized in that: Step three specifically involves performing a "matrix reorganization" transformation on the time domain waveform of the one-dimensional time domain signal collected in step two, converting it from a 1×3600 array to a 50×72 array, and then plotting "phase analysis partial discharge pattern" spectra corresponding to different defect models when partial discharge occurs.

5. The method for identifying partial discharge patterns of high-voltage switchgear based on the pulse current method according to claim 4, characterized in that: In step 3, the PRPD spectrum drawn is Spectrum.

6. The method for identifying partial discharge patterns of high-voltage switchgear based on the pulse current method according to claim 5, characterized in that: In step 4, the extracted feature parameters include: The skewness Sk of the spectrum, The steepness Ku of the spectrum, The phase median Mv of the spectrum, The number of partial discharge peak points Pe in the spectrum, The cross-correlation coefficient cc and the modified cross-correlation coefficient mcc of the spectrum.

7. The method for identifying partial discharge patterns of high-voltage switchgear based on the pulse current method according to claim 1, characterized in that: In step 5, the classification and identification of characteristic parameters using the BP neural network algorithm includes: Step 1), build a BP neural network; Step 2), divide the data set into training set and test set; Step 3) Input the training set into the BP neural network to perform model training. During the training process, the model parameters are continuously adjusted to obtain the optimal network model. In step 4), the test set is input into the optimal network model for testing, and the confusion matrix is analyzed to verify the feasibility of using the BP neural network to identify the partial discharge type.

8. The method for identifying partial discharge patterns of high-voltage switchgear based on the pulse current method according to claim 7, characterized in that: The step 2) specifically includes dividing the partial discharge data collected by the partial discharge simulation experimental platform into 80% as a training set and 20% as a test set.

Images