Lightning arrester on-line rapid live detection system and method

Abstract

The invention belongs to the technical field of lightning arrester detection, and discloses an online rapid live detection system and method for a lightning arrester, and the system comprises a special high-frequency current sensor which is used for sensing a real-time high-frequency current signal generated by the lightning arrester according to a preset magnetic core parameter; the acquisition card is electrically connected with the special high-frequency current sensor and is used for sampling and converting the real-time high-frequency current signal sensed by the special high-frequency current sensor at a preset sampling frequency to obtain a digital real-time high-frequency current signal; the monitoring host is in communication connection with the acquisition card, the acquisition card uploads an abnormal signal of digital quantity to the monitoring host, the monitoring host takes a real-time high-frequency current signal of the digital quantity as a real-time original signal, preprocesses the real-time original signal, carries out defect identification on the preprocessed real-time original signal, and sends a defect identification result to the acquisition card; and obtaining the defect type of the lightning arrester, and visually displaying the defect type of the lightning arrester. The method has the advantages of high detection convenience and high defect detection accuracy.

Description

Technical Field

[0001] The invention belongs to the technical field of lightning arrester detection, and in particular relates to an online rapid energized detection system and method for a lightning arrester. Background Art

[0002] With the continuous expansion of urban and rural power grids, the use of zinc oxide arresters (MOA) is increasing. In recent years, zinc oxide arresters have been widely used in medium and high voltage and ultra-high voltage transmission and transformation systems due to their advantages such as low residual voltage, no follow-through, small action delay, and large current capacity. Arrester must be installed at the distribution network cable and overhead line combination pole, pole switch pole, arrester pole, substation and other locations. With the increase of commissioning time, the number of trips caused by arrester defects is becoming more and more frequent. These defects have a significant impact on the service life of zinc oxide arresters.

[0003] Typical defects of zinc oxide arresters are: internal moisture, internal insulation aging, external insulation contamination or aging, abnormal discharge, etc. Zinc oxide arresters have the dual characteristics of double Schottky barrier diodes and grain boundary conduction, that is, internal moisture and internal insulation defects have self-healing properties, which is manifested as tripping and then being able to close smoothly after a period of time. Therefore, this type of defect is very hidden and is called a metastable state. Therefore, it is urgent to study methods for rapid detection of zinc oxide arresters.

[0004] Traditional detection methods include: preventive testing, infrared temperature measurement, leakage current method and high-frequency partial discharge method.

[0005] The preventive test of zinc oxide lightning arrester is a pre-test before disconnection, which has significant disadvantages. It needs to be operated regularly under power outage with the use of cranes and lifts, which requires a large workload, long disassembly and assembly time, high cost, and may have certain consequences for personnel safety and equipment safety.

[0006] Domestic and foreign researchers have conducted research in different directions on the temperature rise changes caused by metastable MOA. The existing technology proposes the application of infrared technology in the detection of lightning arrester insulation defects. Through infrared temperature measurement, it was found that the temperature of 500kV MOA was abnormal. Disassembly analysis showed that it was caused by internal moisture. The temperature rise characteristics of MOA under different fault conditions and continuous current shocks were studied. According to the research results, when MOA is damp, the temperature rise in the damp area is large; with the increase of the damp area, the maximum temperature rise of MOA and the number of flanges with large surface temperature rise increase significantly; when different areas of MOA are short-circuited, the temperature rise is small. Through infrared detection technology, some mild moisture faults and all severe moisture faults can be identified. In addition, the existing technology has conducted a test on the influence of the return stroke current on the temperature rise of the valve plate. The long continuous current and the interval time of multiple return strokes have little effect on the temperature rise of MOA. The maximum temperature rise of MOA can reach 150.7℃ under 10 return strokes; after considering the heat accumulation effect, the maximum temperature rise increases by about 6.6℃; under the action of multiple return strokes, when a single-section side flash fault occurs in MOA, the long continuous current will make it difficult to extinguish the arc at the fault site, thereby causing the MOA temperature to rise rapidly and easily cause thermal collapse; experiments show that infrared temperature measurement cannot effectively detect the short-circuit failure defect of a single section of MOA.

[0007] Through theoretical research, simulation, experiment and practice, it is proved that infrared temperature measurement is a very convenient and effective technical means for large MOAs with voltage levels of 110 kV and above. However, there is no actual research and experiment on 10 kV MOAs due to their small size, thin insulation layer and unclear heat accumulation effect. Practice shows that conventional infrared temperature measurement cannot detect the metastable state of 10 kV zinc oxide arresters.

[0008] Theoretical analysis and field experience show that the change of resistive current component in the arrester leakage current can reflect the degree of moisture or aging. The resistive current is the DC component obtained from the leakage current through Fourier or wavelet transform. However, due to factors such as the amplitude of milliamperes, on-site harmonic interference, and the superposition of leakage currents of other equipment, the sensor sensitivity does not meet the requirements, making the picked-up data inaccurate; at the same time, the weak DC component has different degrees of attenuation on the current line, sensor, transmission line and AD conversion module, which is easy to miss. Summary of the invention

[0009] The purpose of the present invention is to provide an online rapid live detection system and method for lightning arresters, so as to solve the problem that the sensor sensitivity does not meet the requirements and the picked-up data is inaccurate due to factors such as the amplitude of milliampere level, on-site harmonic interference, and the superposition of leakage current of other equipment; secondly, the current also has different degrees of attenuation and is prone to underreporting.

[0010] In order to achieve the above object, the present invention adopts the following technical solutions: In a first aspect, the present invention provides an online rapid energized detection system for a lightning arrester, the system comprising: A dedicated high-frequency current sensor is used to sense the real-time high-frequency current signal generated by the arrester with preset core parameters; The acquisition card is electrically connected to the dedicated high-frequency current sensor and is used to sample and convert the real-time high-frequency current signal sensed by the dedicated high-frequency current sensor at a preset sampling frequency to obtain a digital real-time high-frequency current signal; The monitoring host is communicatively connected with the acquisition card, and the acquisition card uploads the abnormal signal of the digital quantity to the monitoring host. The monitoring host is used to use the real-time high-frequency current signal of the digital quantity as the real-time original signal, preprocess the real-time original signal, identify defects on the preprocessed real-time original signal, obtain the defect type of the lightning arrester, and visualize the defect type of the lightning arrester. The defect types of the lightning arrester include: moisture inside the lightning arrester, aging of the internal insulation of the lightning arrester, contamination of the external insulation of the lightning arrester, aging of the external insulation of the lightning arrester, or abnormal discharge of the lightning arrester.

[0011] Preferably, the preset core parameters include: core inner diameter, core outer diameter, core height, core initial magnetic permeability, core saturation magnetic flux density, core Curie temperature, core magnetic path length, core magnetic flux area, core winding area and core magnetic flux volume.

[0012] Preferably, the preset sampling frequency of the acquisition card is 200 MHz, the bandwidth of the acquisition card is 1 kHz-100 MHz, and the resolution of the acquisition card is 12 bits.

[0013] Preferably, the system further comprises: a coaxial cable and a data line, the dedicated high-frequency current sensor is electrically connected to the acquisition card via the coaxial cable, and the acquisition card is communicatively connected to the monitoring host via the data line.

[0014] In a second aspect, the present invention provides an online rapid energization detection method for a lightning arrester, the method being applied to an online rapid energization detection system for a lightning arrester, the method comprising: Acquire the real-time high-frequency current signal of the arrester, use the acquired real-time high-frequency current signal as the real-time original signal, pre-process the real-time original signal to obtain the pre-processed real-time original signal, and the pre-processing includes: wavelet denoising processing; Extract features from the preprocessed real-time original signal to obtain the feature quantity to be identified; Based on the pre-trained classification model, the identification feature quantity is treated to obtain the classification result; The defect type of the arrester is determined according to the classification result, and the defect type of the arrester includes: moisture inside the arrester, aging of the internal insulation of the arrester, contamination of the external insulation of the arrester, aging of the external insulation of the arrester, or abnormal discharge of the arrester.

[0015] Preferably, the method further comprises: constructing a classification model, comprising: Constructing defect samples of lightning arresters, and randomly selecting a first preset number of samples from the defect samples of lightning arresters as target samples; Extract features from the target sample to obtain multiple time-frequency domain feature quantities; Performing normalization and dimensionality reduction processing on multiple time-frequency domain feature quantities to obtain multiple learning sample feature quantities; The support vector machine model is trained using multiple learning sample feature quantities to determine the optimal parameters of the support vector machine model, wherein the optimal parameters include an optimal penalty factor and an optimal bias; and the support vector machine model corresponding to the optimal parameters is used as a classification model.

[0016] Preferably, the kernel function of the support vector machine model adopts a radial basis Gaussian function.

[0017] Preferably, multiple time-frequency domain feature quantities are normalized and dimensionally reduced to obtain multiple learning sample feature quantities, including: Normalizing the multiple time-frequency domain feature quantities to obtain multiple normalized time-frequency domain feature quantities; Based on the principal component analysis method, multiple normalized time-frequency domain feature quantities are subjected to dimensionality reduction processing to obtain multiple learning sample feature quantities.

[0018] Preferably, the identification feature quantity is treated based on a pre-trained classification model to obtain a classification result, including: Randomly selecting a second preset number of time-frequency domain feature quantities from a plurality of time-frequency domain feature quantities and mixing them with the feature quantity to be identified to obtain a mixed feature quantity; The mixed feature quantities are classified based on the classification model to obtain the classification results.

[0019] Preferably, the method further comprises: Determine the high-frequency pulse time domain waveform according to the real-time high-frequency current signal; Visualize the high-frequency pulse time domain waveform.

[0020] Beneficial effects: 1. The detection system of the present invention is composed of hardware such as a dedicated high-frequency current sensor, an acquisition card and a monitoring host, which is easy to carry and improves the convenience of the detection task; 2. When the present invention detects the arrester, there is no need to shut down the power supply, disconnect the parallel connection, or apply a withstand voltage. It only needs to pick up the high-frequency signal on the grounding flat iron of the arrester through a special high-frequency current sensor, thereby improving the convenience of signal acquisition; 3. The present invention collects real-time high-frequency current signals through a dedicated high-frequency current sensor, and the monitoring host performs defect identification on the real-time high-frequency current signals, thereby identifying the defect type of the arrester, for example: moisture inside the arrester, aging of the internal insulation of the arrester, contamination of the external insulation of the arrester, aging of the external insulation of the arrester, or abnormal discharge of the arrester, thereby improving the accuracy of defect detection of the arrester. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The accompanying drawings are used to provide a further understanding of the embodiments of the present invention and constitute a part of the specification. Together with the following specific embodiments, they are used to explain the embodiments of the present invention, but do not constitute a limitation on the embodiments of the present invention. In the accompanying drawings: Figure 1 It is a block diagram of an online rapid live detection system for a lightning arrester provided by an embodiment of the present invention; Figure 2 This is an actual product diagram of an online rapid live detection system for lightning arresters provided by an embodiment of the present invention; Figure 3 is an equivalent circuit diagram of a dedicated high-frequency current sensor provided by one embodiment of the present invention; Figure 4 The present invention is a flowchart of a method for online rapid energized detection of a lightning arrester provided in one embodiment of the present invention. DETAILED DESCRIPTION

[0022] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the present invention will be briefly introduced below in combination with the drawings and the description of the embodiments or the prior art. Obviously, the following description of the structure of the drawings is only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. It should be noted that the description of these embodiments is used to help understand the present invention, but does not constitute a limitation of the present invention.

[0023] Embodiment 1 Figure 1 is a block diagram of an arrester online rapid live detection system provided by an embodiment of the present invention, such as Figure 1As shown, this embodiment provides an online rapid live detection system for lightning arresters, the system comprising: a dedicated high-frequency current sensor, an acquisition card and a monitoring host, wherein the acquisition card is integrated with an AD conversion module, an FPGA controller, etc., the monitoring host can be a PC, a laptop computer, etc., and the monitoring host preferably uses a laptop computer. The detection system also includes: a coaxial cable and a data cable, wherein the coaxial cable uses a 50Ω coaxial cable, and the data cable uses a USB data cable. The dedicated high-frequency current sensor is electrically connected to the acquisition card via a 50Ω coaxial cable, and the acquisition card is then communicatively connected to the monitoring host via a USB data cable; the detection system of the present invention is composed of hardware such as a dedicated high-frequency current sensor, an acquisition card and a monitoring host, which is easy to carry and improves the convenience of the detection task. The actual product diagram of the detection system of this embodiment is shown in FIG. Figure 2 As shown, in Figure 2 In the figure, number 1 is a special high-frequency current sensor, number 2 is a coaxial cable, number 3 is an acquisition card, number 4 is a USB data cable, number 5 is a laptop computer, and number 6 is a software program running on the laptop computer. The software program is used to perform defect identification tasks. A large number of practices have shown that the center frequency of the high-frequency current signal excited by the defects of zinc oxide lightning arresters has its own properties.

[0024] Among them, the dedicated high-frequency current sensor is used to sense the real-time high-frequency current signal generated by the arrester with preset magnetic core parameters. The arrester of this embodiment is used in 10kV transmission lines. The 10kV transmission line tower is mainly composed of a pole, an arrester crossarm, a 10kV arrester, a grounding flat iron, a 10kV conductor and an insulator. One end of the grounding flat iron is connected to the earth, and the other end of the grounding flat iron is connected to the 10kV arrester. The dedicated high-frequency current sensor can directly pick up the high-frequency signal on the grounding flat iron, which is the real-time high-frequency current signal generated by the arrester. Therefore, when the present invention detects the arrester, there is no need to shut down the power, decouple, or apply a withstand voltage. It only needs to pick up the high-frequency signal on the arrester grounding flat iron through a dedicated high-frequency current sensor, which improves the convenience of signal acquisition.

[0025] In this embodiment, the preset core parameters include: core inner diameter, core outer diameter, core height, core initial magnetic permeability, core saturation flux density, core Curie temperature, core magnetic path length, core flux area, core winding area and core flux volume.

[0026] In order to accurately determine the above-mentioned preset magnetic core parameters, this embodiment adopts the following method: The dedicated high-frequency current sensor is a passive sensor, and its sensitivity and bandwidth directly affect the accuracy of the collected data. If the high-frequency signal stimulated by the cable defect is used to collect the high-frequency signal of the 10kV zinc oxide arrester by the dedicated high-frequency current sensor, the upper and lower cutoff frequencies of the dedicated high-frequency current sensor do not match, which will cause the collected signal to be very weak. A large number of experiments have shown that the center frequency of the high-frequency signal stimulated by the internal defects of the arrester is about 1MHz.

[0027] The dedicated high-frequency current sensor measures the current based on the law of electromagnetic induction. Therefore, the equivalent circuit of the dedicated high-frequency current sensor is constructed, such as Figure 3 As shown, in Figure 3 Where M is the mutual inductance between the coil and the current-carrying conductor placed in the middle of the coil. L s is the self-inductance of the coil, R s The internal resistance of my coil, C s is the stray capacitance of the coil, R L It is the load of the dedicated high frequency current sensor.

[0028] Assume that the sum of all currents in the center of the coil is , then according to Ampere's circuit law: (1); If the current in the current-carrying conductor passing through the center of the coil is i 1, then the above formula becomes: (2); The magnetic core skeleton of the special high-frequency current sensor of the present application is a ring-shaped one, which carries the current i The magnetic field intensity H and magnetic induction intensity generated by the conductor of 1 at a distance r from the axis They are: (3); (4); In formula (4), is the magnetic permeability in vacuum, which is , is the relative magnetic permeability of the coil skeleton.

[0029] The magnetic flux through a single turn of a rectangular coil is: (5); In formula (5), is the magnetic flux of a single turn of the rectangular coil, h is the height of the cross section of the dedicated high-frequency current sensor, and s is the area of ​​the cross section of the dedicated high-frequency current sensor.

[0030] Substituting formula (4) into formula (5), we can obtain: (6); In formula (6), is the inner diameter of the dedicated high-frequency current sensor, is the outer diameter of the dedicated high-frequency current sensor.

[0031] In the coil skeleton cross section and When the turns are evenly distributed, the total magnetic flux in the dedicated high-frequency current sensor is: (7); In formula (7), is the total magnetic flux in the dedicated high-frequency current sensor.

[0032] According to the law of electromagnetic induction, the induced electromotive force of the coil is: (8).

[0033] In addition, the induced electromotive force in the coil It can be determined by the following formula: (9).

[0034] From equations (8) and (9), the mutual inductance coefficient between the coil and the current-carrying conductor placed in the middle of the coil is: (10).

[0035] After a similar reasoning process as equations (4) to (10), the self-inductance coefficient of the coil is obtained as: (11).

[0036] According to formula (10) and formula (11), it can be seen that the mutual inductance coefficient of the dedicated high-frequency current sensor transformer is proportional to the number of turns of the coil, and the self-inductance coefficient is proportional to the square of the number of turns of the coil. That is, .Increasing the number of coil turns can significantly increase its self-inductance coefficient.

[0037] List the voltage and current equations in terms of equivalent circuits: (12); (13).

[0038] The system transfer function is: (14).

[0039] Assume that the denominator of the above equation has two roots and ,and (15).

[0040] Where: (16); (17).

[0041] As a necessary condition for self-integration, the load resistance Very small, in When , the denominator of the above formula has two unequal real roots, and equation (14) can be rewritten as: (18).

[0042] Amplitude-frequency characteristic curve at cut-off frequency and There is a frequency-independent region in between. The larger the value, the wider the area. When , the lower cutoff frequency is: (19); Upper cutoff frequency: (20); Considering , so equations (19) and (20) can be simplified to: (twenty one); (twenty two); Therefore, the upper and lower cutoff frequencies and the operating bandwidth of the measurement circuit are determined by the following equations: (twenty three); The amplitude-frequency characteristic curve obtained by formula (23) has a certain error, and the actual effective working frequency of the sensor is smaller than that of the above formula. If the relative error of 1% is considered, the actual effective working frequency range is, (twenty four).

[0043] It can be seen from equation group (23) that the various electrical parameters of the sensor itself determine its theoretical measurement frequency band. Analyzing equation group (23), it can be seen that to widen the frequency band of the sensor, on the one hand, the cutoff frequency of the high frequency band should be increased, and on the other hand, the cutoff frequency of the low frequency band should be reduced. The measures to increase the cutoff frequency of the high frequency band include reducing the resistance of the sensor coil itself. and stray capacitance In addition, measures to reduce the low-frequency cutoff frequency include increasing the inductance of the sensor coil itself. , reduce its own resistance and use a smaller load resistor wait.

[0044] Therefore, according to the derivation process from formula (1) to formula (23), the inner diameter and outer diameter of the magnetic core are designed according to the actual situation, and other parameters can be derived at this time. The optimal magnetic core parameters of this embodiment are: the inner diameter of the magnetic core is 55.5 mm, the outer diameter of the magnetic core is 85.7 mm, the height of the magnetic core is 19.05 mm, the initial magnetic permeability of the magnetic core is 5000 H / m, the saturation magnetic flux density of the magnetic core is 430 mT, the Curie temperature of the magnetic core is 145°C, the magnetic path length of the magnetic core is 214.9 mm, and the magnetic flux area of ​​the magnetic core is 283 mm. 2 , the core winding area is 24.2mm 2 , the core flux volume is 6874mm 3 .

[0045] Among them, the acquisition card is electrically connected to the dedicated high-frequency current sensor, and is used to sample and convert the real-time high-frequency current signal sensed by the dedicated high-frequency current sensor at a preset sampling frequency to obtain a real-time high-frequency current signal of a digital quantity. The preset sampling frequency of the acquisition card in this embodiment is 200MHz, the bandwidth of the acquisition card is 1kHz-100MHz, and the resolution of the acquisition card is 12 bits.

[0046] The monitoring host is communicatively connected with the acquisition card, and the acquisition card uploads the abnormal signal of the digital quantity to the monitoring host. The monitoring host is used to use the real-time high-frequency current signal of the digital quantity as the real-time original signal, preprocess the real-time original signal, identify defects on the preprocessed real-time original signal, obtain the defect type of the lightning arrester, and visualize the defect type of the lightning arrester. The defect types of the lightning arrester include: moisture inside the lightning arrester, aging of the internal insulation of the lightning arrester, contamination of the external insulation of the lightning arrester, aging of the external insulation of the lightning arrester, or abnormal discharge of the lightning arrester.

[0047] The present invention collects real-time high-frequency current signals when a dedicated high-frequency current sensor is used, and the monitoring host performs defect identification on the real-time high-frequency current signals, thereby identifying the defect type of the lightning arrester, for example: moisture inside the lightning arrester, aging of the internal insulation of the lightning arrester, contamination of the external insulation of the lightning arrester, aging of the external insulation of the lightning arrester, or abnormal discharge of the lightning arrester, thereby improving the accuracy of defect detection of the lightning arrester.

[0048] Embodiment 2 Figure 4 is a flow chart of a method for online rapid energized detection of a lightning arrester provided by an embodiment of the present invention, such as Figure 4 As shown, this embodiment provides a method for online rapid detection of lightning arrester charging, which is applied to the online rapid detection system of lightning arrester charging of embodiment 1, and uses a high-frequency current pulse method for lightning arrester charging detection, and the method includes: Step S10: acquiring the real-time high-frequency current signal of the arrester, taking the acquired real-time high-frequency current signal as the real-time original signal, and preprocessing the real-time original signal to obtain the preprocessed real-time original signal, wherein the preprocessing includes: wavelet denoising processing.

[0049] In this embodiment, a dedicated high-frequency current sensor is used to collect real-time high-frequency current signals, and then an acquisition card samples and converts the real-time high-frequency current signals to obtain digital real-time high-frequency current signals, which are used as real-time original signals.

[0050] Step S20: extracting features from the preprocessed real-time original signal to obtain feature quantities to be identified.

[0051] Step S30: Obtain classification results based on the pre-trained classification model for the feature quantity to be identified.

[0052] Step S40: Determine the defect type of the arrester according to the classification result, the defect type of the arrester includes: internal moisture of the arrester, internal insulation aging of the arrester, external insulation contamination of the arrester, external insulation aging of the arrester, or abnormal discharge of the arrester.

[0053] As a further optimization of this embodiment, the method further includes: constructing a classification model, including: Step a10: construct defect samples of the lightning arrester, and randomly select a first preset number of samples from the defect samples of the lightning arrester as target samples.

[0054] In this embodiment, in a laboratory environment, MOA internal moisture and valve plate crack defect samples were prepared respectively, and then 60% of all the prepared samples were used as target samples.

[0055] Step a20: extract features from the target sample to obtain multiple time-frequency domain feature quantities.

[0056] In this embodiment, it is necessary to perform feature extraction for each target sample. Each target sample corresponds to multiple time-frequency domain feature quantities. The multiple time-frequency domain feature quantities of each target sample include eight dimensional feature quantities: center frequency, number of valid positive-sequence pulses within 20 ms, number of negative-sequence pulses, rising edge of main peak of a single pulse, duration of decline, number of continuous peaks and tail wave.

[0057] Step a30: performing normalization and dimensionality reduction processing on multiple time-frequency domain feature quantities to obtain multiple learning sample feature quantities.

[0058] Specifically, a plurality of time-frequency domain feature quantities are first normalized to obtain a plurality of normalized time-frequency domain feature quantities; and then a dimensionality reduction process is performed on the plurality of normalized time-frequency domain feature quantities based on a principal component analysis method to obtain a plurality of learning sample feature quantities.

[0059] The feature quantities of multiple learning samples (excerpt) are shown in Table 1: Table 1 Learning sample characteristic scale (excerpt) Serial number Frequency domain / kHz Positive order / pieces Negative sequence / pieces Rising edge / ms Falling edge / ms Duration / ms Crest / pieces Wake / pieces 1 1 200 2 2 0.020 0.030 0.030 8 1 2 1 200 2 2 0.030 0.020 0.020 9 1 3 1 200 2 2 0.050 0.040 0.050 7 2 4 1 200 2 2 0.060 0.060 0.050 8 1 Step a40: train the support vector machine model using multiple learning sample feature quantities to determine the optimal parameters of the support vector machine model, wherein the optimal parameters include: an optimal penalty factor and an optimal bias; and use the support vector machine model corresponding to the optimal parameters as the classification model.

[0060] In this embodiment, the kernel function of the support vector machine model adopts a radial basis Gaussian function, which can map the learning sample feature quantities in 8 dimensions to a high-dimensional space, making classification in the high-dimensional space simple.

[0061] The expression of the radial basis Gaussian function is: (25); In formula (25), exp() is the natural exponential function, It is a parameter that controls the width of the Gaussian kernel function, that is, controls the influence range of the kernel function. The larger the value, the closer the data points are to the hyperplane, which may lead to overfitting.

[0062] Will Set to: (26); In formula (26), the optimal parameter τ = 0.81 is obtained through practice.

[0063] In the training process of the support vector machine model, it is necessary to determine the key parameters of the support vector machine model, namely the optimal penalty factor and the optimal bias. Among them, the penalty factor C determines the tolerance of the model to misclassification. The larger the value, the more the model tends to find a hyperplane that perfectly classifies but may overfit the data. Choosing a suitable C is crucial.

[0064] Assume p nonlinear samples {(x 1 , d 1 ), (x 2 , d 2 ),…,(x p , d p )}, for any input sample x p , the expected output is d p= ±1 (representing two-class category output).

[0065] From the linearly separable hyperplane equation: (27); Determine the optimal hyperplane pair weights in a nonlinear separable system The constraints on the bias b are: (28); in, is a slack variable, and ≥0.

[0066] Under the constraints of formula (28), we can get the following equations about weight W and slack variables: The minimum cost function : (29); in, C is the selected positive parameter penalty factor, T To transpose, according to the dual problem of nonlinear separable data, given the training samples, the objective function is obtained as: (30); In formula (30), is the Lagrange coefficient of the conjugate vector set, is the expected output of the conjugate data set, formula (30) makes the above The Lagrange coefficient of the maximum value }, and the following constraints are met: (31); That is, the constraint condition is .

[0067] The optimal solution of W and b must satisfy the Kuhn Tucker optimization condition: (32).

[0068] Through a large number of tests, in the two samples of the given arrester damp and internal insulation damage, the damp output is positive and the internal damp is negative; the respective penalty factors are determined as: C + =3.17, C - =21.17.

[0069] Therefore, in step S30, based on the pre-trained classification model, the identification feature quantity is treated to obtain the classification result, including: Step S301: randomly selecting a second preset number of time-frequency domain feature quantities from a plurality of time-frequency domain feature quantities and mixing them with the feature quantity to be identified to obtain a mixed feature quantity.

[0070] Step S302: classify the mixed feature quantities based on the classification model to obtain a classification result.

[0071] In this embodiment, 40% of the feature quantities are randomly selected from multiple time-frequency domain feature quantities and mixed with the feature quantities to be identified to obtain mixed feature quantities, and then the mixed feature quantities are classified using the classification model to obtain the final classification results. Through feature mixing, the classification quality can be improved and the classification error can be reduced.

[0072] As a further optimization of this embodiment, the method further includes: determining the high-frequency pulse time domain waveform according to the real-time high-frequency current signal; and visually displaying the high-frequency pulse time domain waveform.

[0073] Among them, the function expression of the high-frequency pulse time domain waveform is: (33); In formula (33), q is the total discharge capacity, Z is the impedance, t is the pulse propagation time, σ is the time factor; To test the high-frequency pulse time domain waveform function; then the high-frequency pulse time domain waveform is displayed on the human-machine display interface of the laptop, so that users can observe the high-frequency pulse time domain waveform under different defects. According to the high-frequency pulse time domain waveform, it can be identified whether the arrester defect is moisture, zinc oxide valve plate insulation damage or other.

[0074] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0075] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1A device that provides the functions specified in a block or multiple blocks.

[0076] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included within the scope of the claims of the present application.

Claims

1. An online rapid live detection system for lightning arresters, characterized in that: The system comprises: A dedicated high-frequency current sensor is used to sense the real-time high-frequency current signal generated by the arrester with preset core parameters; The acquisition card is electrically connected to the dedicated high-frequency current sensor and is used to sample and convert the real-time high-frequency current signal sensed by the dedicated high-frequency current sensor at a preset sampling frequency to obtain a digital real-time high-frequency current signal; The monitoring host is communicatively connected with the acquisition card, and the acquisition card uploads the abnormal signal of the digital quantity to the monitoring host. The monitoring host is used to use the real-time high-frequency current signal of the digital quantity as the real-time original signal, preprocess the real-time original signal, perform defect identification on the preprocessed real-time original signal, obtain the defect type of the lightning arrester, and visualize the defect type of the lightning arrester. The defect types of the lightning arrester include: moisture inside the lightning arrester, aging of the insulation inside the lightning arrester, and abnormal discharge outside the lightning arrester.

2. The arrester online rapid live detection system according to claim 1 is characterized in that: The preset magnetic core parameters include: magnetic core inner diameter, magnetic core outer diameter, magnetic core height, magnetic core initial magnetic permeability, magnetic core saturation magnetic flux density, magnetic core Curie temperature, magnetic core magnetic path length, magnetic core magnetic flux area, magnetic core winding area and magnetic core magnetic flux volume.

3. The arrester online rapid live detection system according to claim 1 is characterized in that: The preset sampling frequency of the acquisition card is 200 MHz, the bandwidth of the acquisition card is 1 kHz-100 MHz, and the resolution of the acquisition card is 12 bits.

4. The arrester online rapid live detection system according to claim 1 is characterized in that: The system further comprises: a coaxial cable and a data line, the dedicated high-frequency current sensor is electrically connected to the acquisition card via the coaxial cable, and the acquisition card is communicatively connected to the monitoring host via the data line.

5. A method for rapid online detection of lightning arrester charged, the method being applied to the system for rapid online detection of lightning arrester charged according to any one of claims 1 to 4, characterized in that: The method comprises: Acquire the real-time high-frequency current signal of the arrester, use the acquired real-time high-frequency current signal as the real-time original signal, pre-process the real-time original signal to obtain the pre-processed real-time original signal, and the pre-processing includes: wavelet denoising processing; Extract features from the preprocessed real-time original signal to obtain the feature quantity to be identified; Based on the pre-trained classification model, the identification feature quantity is treated to obtain the classification result; The defect type of the arrester is determined according to the classification result, and the defect type of the arrester includes: moisture inside the arrester, aging of insulation inside the arrester, and abnormal discharge outside the arrester.

6. The method for rapid online detection of lightning arrester power supply according to claim 5, characterized in that: The method further comprises: constructing a classification model, comprising: Constructing defect samples of lightning arresters, and randomly selecting a first preset number of samples from the defect samples of lightning arresters as target samples; Extract features from the target sample to obtain multiple time-frequency domain feature quantities; Performing normalization and dimensionality reduction processing on multiple time-frequency domain feature quantities to obtain multiple learning sample feature quantities; The support vector machine model is trained using multiple learning sample feature quantities to determine the optimal parameters of the support vector machine model, wherein the optimal parameters include an optimal penalty factor and an optimal bias; and the support vector machine model corresponding to the optimal parameters is used as a classification model.

7. The method for online rapid detection of lightning arrester charge according to claim 6, characterized in that: The kernel function of the support vector machine model adopts a radial basis Gaussian function.

8. The method for rapid online detection of lightning arrester power supply according to claim 6, characterized in that: Multiple time-frequency domain feature quantities are normalized and dimensionally reduced to obtain multiple learning sample feature quantities, including: Normalizing the multiple time-frequency domain feature quantities to obtain multiple normalized time-frequency domain feature quantities; Based on the principal component analysis method, multiple normalized time-frequency domain feature quantities are subjected to dimensionality reduction processing to obtain multiple learning sample feature quantities.

9. The method for online rapid detection of lightning arrester charging according to claim 5, characterized in that: Based on the pre-trained classification model, the identification feature quantity is treated to obtain the classification results, including: Randomly selecting a second preset number of time-frequency domain feature quantities from a plurality of time-frequency domain feature quantities and mixing them with the feature quantity to be identified to obtain a mixed feature quantity; The mixed feature quantities are classified based on the classification model to obtain the classification results.

10. The method for rapid online detection of lightning arrester electrification according to claim 5, characterized in that: The method further comprises: According to the real-time high-frequency current signal, the high-frequency pulse time domain waveform is determined; and the high-frequency pulse time domain waveform is visualized.

Images