Double-branch distributed partial discharge monitoring system based on wide and narrow double pulses

Abstract

The invention provides a double-branch distributed partial discharge monitoring system based on wide and narrow double pulses, and relates to the technical field of distributed optical fiber sensing. A continuous laser signal emitted by an ultra-narrow linewidth laser is divided into a first branch signal and a second branch signal through a beam splitter; the first branch signal passes through an acousto-optic modulator driven by an arbitrary waveform generator, so that the continuous laser signal is modulated into a wide-narrow double-pulse signal, and the wide-narrow double-pulse signal generates a backward Rayleigh scattering signal in the test optical fiber; the wide-narrow double-pulse signal is transmitted to the data demodulation module through the erbium-doped optical power amplifier, the first port of the optical circulator, the test optical fiber and the third port of the optical fiber ring optical circulator in sequence; and the second branch signal is transmitted to the data demodulation module through the beam splitter, the acoustic optical modulator, the erbium-doped amplifier, the test optical fiber, the optical fiber circulator and the coupler in sequence. According to the invention, the spatial positioning precision and the frequency response range of partial discharge monitoring can be improved.

Description

Technical Field

[0001] The present invention relates to the field of distributed optical fiber sensing technology, and particularly to a dual-branch distributed partial discharge monitoring system based on wide and narrow double pulses. Background Art

[0002] In underground tunnels, the safety of cables is of crucial importance. With the increasing demand for electricity, the application of underground cables is becoming more and more widespread. However, during long-term use, cables may suffer from structural damage due to material aging and external damage, which may further lead to partial discharge (PD) phenomena. Partial discharge may cause safety hazards such as insulation failure and equipment malfunction, and even trigger major accidents such as fires. Therefore, real-time online monitoring of cable partial discharge is an important measure to ensure the safe operation of tunnel cables.

[0003] Chinese Patent with Publication No. CN113899995A discloses a partial discharge detection method and device based on a distributed feedback fiber laser, including a distributed feedback fiber laser and a pump laser source; the detection light generated by the distributed feedback fiber laser under the excitation of the received pump light can enter a fiber interferometer after passing through a wavelength division multiplexer, and the fiber interferometer can obtain the first output light after interference and the second output light after interference for the received detection light; a photoelectric conversion device is used to receive the first output light after interference and the second output light after interference to obtain a first detection electrical signal and a second detection electrical signal after photoelectric conversion; a signal processing device is used to combine and process the first detection electrical signal and the second detection electrical signal to obtain partial discharge detection status information. However, the above solution cannot solve the problems of too long monitoring distance of tunnel cables and complex and diverse noises around tunnel cables, resulting in low spatial positioning accuracy. Therefore, it is very necessary to provide a dual-branch distributed partial discharge monitoring system based on wide and narrow double pulses, which helps to improve the spatial positioning accuracy of partial discharge monitoring. Summary of the Invention

[0004] In view of this, the present invention proposes a dual-branch distributed partial discharge monitoring system based on wide and narrow double pulses. By using an ultra-narrow linewidth laser and an acousto-optic modulator controlled by an arbitrary waveform generator, a continuous laser signal is modulated into a wide and narrow double pulse signal. The combination of the wide pulse and the narrow pulse enables the system to obtain higher spatial positioning accuracy while ensuring the detection distance.

[0005] The present invention provides a dual-branch distributed partial discharge monitoring system based on wide and narrow double pulses, including an ultra-narrow linewidth laser, a beam splitter, an acousto-optic modulator, an arbitrary waveform generator, an erbium-doped optical power amplifier, an optical circulator, a test optical fiber, an optical fiber loop, a coupler, and a data demodulation module, wherein,

[0006] The continuous laser signal emitted by the ultra-narrow linewidth laser is divided into a first branch signal and a second branch signal by the beam splitter;

[0007] The first branch signal passes through the acousto-optic modulator driven by the arbitrary waveform generator, so that the continuous laser signal is modulated into a wide-narrow double-pulse signal. The wide-narrow double-pulse signal generates a backward Rayleigh scattering signal in the test optical fiber. The wide-narrow double-pulse signal is sequentially transmitted to the data demodulation module through the erbium-doped optical power amplifier, the first port of the optical circulator, the test optical fiber, the optical fiber loop, and the third port of the optical circulator;

[0008] The second branch signal is sequentially transmitted to the data demodulation module through the beam splitter, the acousto-optic modulator, the erbium-doped amplifier, the test optical fiber, the optical fiber circulator, and the coupler.

[0009] On the basis of the above technical solution, preferably, the splitting ratio of the first output port to the second output port of the beam splitter is 9:1. The first output port of the beam splitter is connected to the acousto-optic modulator, and the second output port of the beam splitter is connected to the coupler.

[0010] On the basis of the above technical solution, preferably, the test optical fiber responds to an external ultrasonic signal to change the phase of the backward Rayleigh scattered light.

[0011] More preferably, the data demodulation module includes a first photodetector, a first acquisition unit, a data adjustment unit, a second acquisition unit, and a second photodetector. The first photodetector is respectively connected to the third port of the optical circulator and the first acquisition unit. The data adjustment unit is electrically connected to the first acquisition unit and the second acquisition unit respectively. The second photodetector is respectively connected to the coupler and the second acquisition unit.

[0012] More preferably, the coupler is a 3×3 coupler, and the second photodetector is a 3×3 photodetector. The first port of the 3×3 coupler is connected to the beam splitter, the second port of the 3×3 coupler is left empty, the third port of the 3×3 coupler is connected to the optical fiber loop, and the fourth, fifth, and sixth ports of the 3×3 coupler are all connected to the 3×3 photodetector.

[0013] More preferably, the frequency response range of the first branch signal is expressed as:

[0014]

[0015] where f maxrepresents the maximum detectable frequency in the first branch signal, f sample represents the sampling rate of the dual-branch distributed partial discharge monitoring system, c represents the speed of light in vacuum, n represents the refractive index of the test optical fiber, and L represents the length of the test optical fiber.

[0016] More preferably, both the first acquisition unit and the second acquisition unit are connected to the partial discharge defect identification module, and the partial discharge defect identification module is built-in with a preset double-layer one-dimensional convolutional neural network.

[0017] More preferably, the double-layer one-dimensional convolutional neural network includes a first convolutional module, a second convolutional module, and a fully connected module, where

[0018] The first convolutional module includes a first convolutional layer, a first batch normalization layer, a first activation function layer, and a first pooling layer connected in sequence. The first convolutional layer extracts the initial dimensional feature map from the partial discharge ultrasonic signal waveform data. The first batch normalization layer performs batch normalization on the initial dimensional feature map. The first activation function layer is used to enhance the non-linear ability of the features output by the first convolutional layer. The first pooling layer is used to reduce the dimension of the initial dimensional feature map and extract local features to obtain the first dimensional feature map;

[0019] The second convolutional module includes a second convolutional layer, a second batch normalization layer, a second activation function layer, and a second pooling layer connected in sequence. The second convolutional layer extracts the transition dimensional feature map from the first dimensional feature map. The second batch normalization layer performs batch normalization on the transition dimensional feature map. The second activation function layer is used to enhance the non-linear ability of the features output by the second convolutional layer. The second pooling layer is used to reduce the dimension of the transition dimensional feature map and extract local features to obtain the second dimensional feature map;

[0020] The fully connected module includes a first fully connected layer, a third activation function layer, a dropout layer, a second fully connected layer, a probability activation function layer, and a classification layer connected in sequence. The first fully connected layer is used to integrate the second dimensional feature map and form a global feature vector. The dropout layer is used to randomly discard some neurons in the double-layer one-dimensional convolutional neural network to obtain a generalized feature vector. The second fully connected layer is used to map the generalized feature vector to the output space of the classification task to obtain the class score corresponding to the generalized feature vector. The probability activation function layer is used to convert the class score into a probability distribution so that the classification layer outputs the classification result of the partial discharge defect.

[0021] More preferably, the data of the first branch signal is restored based on the NPS demodulation algorithm to obtain the partial discharge ultrasonic signal waveform corresponding to the test optical fiber in response to the external ultrasonic signal.

[0022] More preferably, the sampling rate of the second acquisition unit is 1 MHz / s, the pulse frequency of the arbitrary waveform generator is 100 kHz, and the pulse width is 50 ns.

[0023] The dual-branch distributed partial discharge monitoring system based on wide and narrow double pulses provided by the present invention has the following beneficial effects compared with the prior art:

[0024] (1) By using an ultra-narrow linewidth laser and an acousto-optic modulator controlled by an arbitrary waveform generator, the continuous laser signal is modulated into a wide and narrow double pulse signal. Through the backward Rayleigh scattering in the test optical fiber, weak partial discharge signals can be captured, optimizing the pulse energy distribution and improving the detection ability of the dual-branch distributed partial discharge monitoring system for weak signals and the overall signal-to-noise ratio. The combination of the wide pulse and the narrow pulse enables the system to obtain higher spatial positioning accuracy while ensuring the detection distance. At the same time, by dividing the laser signal into two branches to form different modulation signal paths respectively, complementary detection of different pulse signals can be realized, which can not only perform redundant detection on the signals to improve data reliability, but also help reduce the influence of external interference on the system, improving the overall system stability and robustness. The signal distribution of the two branches enables the frequency response range to be improved even when the sensing distance is long, successfully breaking the limitation of the sensing distance on the frequency response range.

[0025] (2) The first convolution module extracts preliminary features from the partial discharge ultrasonic signal through convolution operations, adjusts the feature distribution through batch normalization, enhances the non-linear ability through activation functions, and finally uses pooling to reduce the dimension and highlight the main local information. The second convolution module further abstracts on the basis of the preliminary features to extract higher-level discriminant features, and adopts similar normalization, activation, and pooling processes to optimize the feature expression. The fully connected module integrates the local features extracted by the double-layer convolution module to construct a global feature vector, and at the same time introduces a dropout mechanism to effectively prevent overfitting and improve the generalization ability of the model to new data. The global features are converted into class scores through the mapping and probability activation layers to achieve robust and accurate classification of partial discharge defects. Description of the Drawings

[0026] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the accompanying drawings required for the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0027] Figure 1 It is a schematic structural diagram of a dual-branch distributed partial discharge monitoring system based on wide and narrow double pulses provided by the present invention;

[0028] Figure 2 It is a comparison diagram of the positioning effects of three different positioning methods provided by the present invention;

[0029] Figure 3 It is a test diagram of the positioning effect for the distributed partial discharge fault provided by the present invention;

[0030] Figure 4 It is a schematic diagram of an optical fiber loop, the driving signal of the optical fiber loop, and the response sensitivity curve of the optical fiber loop provided by the present invention;

[0031] Figure 5 It is a schematic diagram of a partial discharge simulation system, a partial discharge signal, and the frequency band of the partial discharge signal provided by the present invention;

[0032] Figure 6 It is a signal diagram of four types of partial discharge defects provided by the present invention;

[0033] Figure 7 It is a schematic structural diagram of a double-layer one-dimensional convolutional neural network provided by the present invention;

[0034] Figure 8 It is a confusion matrix for pattern recognition of partial discharge defects provided by the present invention.

[0035] Explanation of reference numerals: 1. Ultra-narrow linewidth laser; 2. Beam splitter; 3. Acousto-optic modulator; 4. Arbitrary waveform generator; 5. Erbium-doped optical power amplifier; 6. Optical circulator; 7. Test optical fiber; 8. Optical fiber loop; 9. Coupler; 10. Data demodulation module; 101. First photodetector; 102. First acquisition unit; 103. Data adjustment unit; 104. Second acquisition unit; 105. Second photodetector. Detailed implementation manners

[0036] Next, in combination with the embodiments of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0037] Unless otherwise defined, the technical terms or scientific terms used in the present invention shall have the ordinary meanings understood by those of ordinary skill in the art to which the present invention belongs. The "first", "second" and similar terms used in the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. Similarly, the terms such as "a" or "one" do not indicate a quantity limit, but indicate that there is at least one. The terms such as "connected" or "coupled" are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. The terms such as "upper", "lower", "left" and "right" are only used to represent relative position relationships. When the absolute position of the object being described changes, the relative position relationship also changes accordingly.

[0038] Referring Figure 1 , the present invention provides a dual-branch distributed partial discharge monitoring system based on wide and narrow double pulses, including an ultra-narrow linewidth laser 1, a beam splitter 2, an acousto-optic modulator 3, an arbitrary waveform generator 4, an erbium-doped optical power amplifier 5, an optical circulator 6, a test optical fiber 7, an optical fiber loop 8, a coupler 9 and a data demodulation module 10. Among them,

[0039] The continuous laser signal emitted by the ultra-narrow linewidth laser 1 is divided into a first branch signal and a second branch signal by the beam splitter 2. The splitting ratio of the first output port to the second output port of the beam splitter 2 is 9:1. The first output port of the beam splitter 2 is connected to the acousto-optic modulator 3, and the second output port of the beam splitter 2 is connected to the coupler 9.

[0040] The first branch signal passes through the acousto-optic modulator 3 driven by the arbitrary waveform generator 4, so that the continuous laser signal is modulated into a wide and narrow double pulse signal. The wide and narrow double pulse signal generates a backward Rayleigh scattering signal in the test optical fiber 7. The wide and narrow double pulse signal sequentially passes through the erbium-doped optical power amplifier 5 and the first port of the optical circulator 6, the test optical fiber 7, and the third port of the optical fiber loop 8 and the optical circulator 6 and is transmitted to the data demodulation module 10. The test optical fiber 7 responds to the external ultrasonic signal to change the phase of the backward Rayleigh scattered light.

[0041] The second branch signal sequentially passes through the beam splitter 2, the acousto-optic modulator 3, the erbium-doped amplifier, the test optical fiber 7, the optical fiber loop 8 and the circulator and the coupler 9 and is transmitted to the data demodulation module 10.

[0042] In one example, the data demodulation module 10 includes a first photodetector 101, a first acquisition unit 102, a data conditioning unit 103, a second acquisition unit 104, and a second photodetector 105. The first photodetector 101 is respectively connected to the third port of the optical circulator 6 and the first acquisition unit 102. The data conditioning unit 103 is electrically connected to the first acquisition unit 102 and the second acquisition unit 104 respectively. The second photodetector 105 is respectively connected to the coupler 9 and the second acquisition unit 104. The coupler 9 is a 3×3 coupler 9, and the second photodetector 105 is a 3×3 photodetector. The first port of the 3×3 coupler 9 is connected to the beam splitter 2, the second port of the 3×3 coupler 9 is left empty, the third port of the 3×3 coupler 9 is connected to the optical fiber loop 8, and the fourth, fifth, and sixth ports of the 3×3 coupler 9 are all connected to the 3×3 photodetector. Both the first acquisition unit 102 and the second acquisition unit 104 are acquisition cards.

[0043] It can be understood that the dual-branch distributed partial discharge monitoring system includes an MZI branch and a Φ-OTDR branch. The Φ-OTDR branch transmits the first branch signal. The signal in the Φ-OTDR branch sequentially passes through the beam splitter 2, the acousto-optic modulator 3, the erbium-doped amplifier, the first port of the optical circulator 6, the test optical fiber 7, the third port of the optical fiber loop 8, the first photodetector 101, and the first acquisition unit 102. The MZI branch transmits the second branch signal. The signal in the MZI branch sequentially passes through the beam splitter 2, the acousto-optic modulator 3, the erbium-doped amplifier, the test optical fiber 7, the optical fiber loop 8, the coupler 9, and the second photodetector 105.

[0044] The frequency response range of the first branch signal is expressed as:

[0045]

[0046] where f max represents the maximum detectable frequency in the first branch signal, f sample represents the sampling rate of the dual-branch distributed partial discharge monitoring system, c represents the speed of light in a vacuum, n represents the refractive index of the test optical fiber 7, and L represents the length of the test optical fiber 7.

[0047] Among them, for the Φ-OTDR branch to perform positioning, an ultrasonic signal is applied to the optical fiber, which changes the refractive index of the optical fiber, thereby changing the phase of the backward Rayleigh scattered light in the optical fiber. The change in the optical phase will cause a change in the optical intensity. Therefore, the method of moving differential averaging can superimpose the change amount of the optical intensity at this point to achieve positioning. However, in an underground tunnel, the environment is complex and there are various environmental noises, which have a huge impact on the positioning effect. Therefore, an adaptive threshold wavelet denoising technique is adopted, and the soft threshold method is selected to eliminate the noise, significantly improving the positioning signal-to-noise ratio. SeeFigure 3 Meanwhile, the pulse width of the pulse determines the positioning accuracy. Therefore, a pulse with a pulse width of 50 ns is used as a narrow pulse.

[0048] Furthermore, the ultra-narrow linewidth laser 1 is connected to the 90:10 port of the beam splitter 2. 90% of the light enters the acousto-optic modulator 3. The arbitrary waveform generator 4 modulates the pulse into a narrow and wide double-pulse signal to drive the acousto-optic modulator 3 to form the target double pulse. The double pulse is then amplified by the erbium-doped optical power amplifier 5, and the optical power enters from the first port of the optical fiber loop circulator 8. The second port of the optical circulator 6 is connected to the test optical fiber 7 and the optical fiber loop 8, and then accesses the 3×3 coupler 9 to interfere with the 10% of the light split by the beam splitter 2. One of the input ports of the 3×3 coupler 9 is left vacant. The three output ports of the 3×3 coupler 9 are connected to the 3×3 photodetector, and then collected by the second acquisition unit 104. At the same time, the backward Rayleigh scattering signal generated in the test optical fiber 7 is output from the third port of the optical circulator 6, accessed into the second photodetector 105, and then collected by the first acquisition unit 102.

[0049] The sampling rate of the second acquisition unit 104 is 1 MHz / s, the pulse frequency of the arbitrary waveform generator 4 is 100 kHz, and the pulse width is 50 ns. Driven by the arbitrary waveform generator 4, the acousto-optic modulator 3 modulates to generate a narrow pulse and a wide pulse closely combined to form a double pulse. This is because when the sensing distance reaches 941 m, in order to ensure that another pulse is emitted only after the pulse has completed a full round trip in the optical fiber, the pulse period must be greater than the time for a round trip in the optical fiber. Therefore, the pulse period is set to 10 μs, and the pulse width of the narrow pulse is 50 ns. Therefore, the pulse width of the wide pulse can be maintained at 9.95 μs (10 μs - 50 ns). The frequencies of the narrow and wide pulses are the same, both set to 100 kHz. Therefore, the pulse width of the wide pulse is much larger than that of the narrow pulse. Therefore, the double pulse can be regarded as continuous light in the MZI branch. Since the frequency response range of continuous light detection in the interferometer is only related to the sampling rate of the acquisition card, the sampling rate of the acquisition card is set to 1 MHz / s. According to the sampling theorem, the maximum detection frequency is less than half of the sampling rate. Therefore, the frequency response bandwidth of the double-branch distributed partial discharge monitoring system is theoretically increased to 500 kHz.

[0050] In this embodiment, by using an ultra-narrow linewidth laser 1 and an acousto-optic modulator 3 controlled by an arbitrary waveform generator 4, a continuous laser signal is modulated into a narrow-wide double-pulse signal. By measuring the backscattered Rayleigh scattering in the test optical fiber 7, weak partial discharge signals can be captured, optimizing the pulse energy distribution and improving the detection ability of the dual-branch distributed partial discharge monitoring system for weak signals and the overall signal-to-noise ratio. The combination of the wide pulse and the narrow pulse enables the system to obtain higher spatial positioning accuracy while ensuring the detection distance. At the same time, by dividing the laser signal into two branches to form different modulated signal paths, complementary detection of different pulse signals is achieved. This not only enables redundant detection of signals to improve data reliability but also helps reduce the impact of external interference on the system, enhancing the overall system stability and robustness. By combining the OTDR branch and the MZI branch, when the sensing distance reaches 941 m, the frequency response range is increased to 300 kHz, successfully breaking the limitation of the sensing distance on the frequency response range.

[0051] In one example, using the arbitrary waveform generator 4, the pulse mode is selected, the pulse frequency is set to 100 kHz, and the pulse width is set to 50 ns. At the same time, the high and low level knobs are adjusted so that the ratio of the high level to the low level is 10, and then this signal is output to the acousto-optic modulator 3. At this time, the high level of this signal is regarded as a narrow pulse, and the low level is regarded as a wide pulse. The modulated signal can be regarded as a double-pulse signal in which the narrow pulse and the wide pulse are closely combined. By using one acousto-optic modulator 3 and one arbitrary waveform generator 4, the modulation of the narrow-wide double-pulse signal is achieved. Compared with other frequency-division multiplexing methods, the frequency response range of the OTDR system is increased. This method uses fewer instruments to obtain the target signal, achieving the purpose of increasing the frequency response range.

[0052] Furthermore, when conducting a partial discharge localization test experiment on the dual-branch distributed partial discharge monitoring system, an optical fiber can be wound around a piezoelectric ceramic (PZT). The arbitrary waveform generator 4 provides a driving signal to drive the PZT to generate an ultrasonic signal of 150 kHz, which is collected by the first acquisition unit 102 for 1000 sets of scattering curves. Three localization methods are respectively used for the data, namely moving average difference (MAD), a localization method with variance as the detection variable (VM), and adaptive threshold wavelet denoising combined with moving average difference (ATWD+MAD).

[0053] As Figure 2 shown, Figure 2(a) shows the variation of the amplitude of the unprocessed original signal with distance. The signal contains obvious noise, and the partial discharge signal is submerged by the noise, making it difficult to directly identify the discharge location. The signal amplitude fluctuates between 0.57 and 0.66, without obvious spike features, indicating that the signal-to-noise ratio of the original signal is low, that is, the noise in the original signal is strong, and the signal-to-noise ratio (SNR) is low, making it impossible to directly locate the position of the partial discharge signal. Figure 2 (b) shows the signal processed by the adaptive threshold wavelet denoising (ATWD) and moving differential averaging (MAD) methods. The partial discharge signal is significantly enhanced near 650 m, and the difference between the signal amplitude and the background noise reaches 14.1 dB. The ATWD method removes most of the noise through wavelet transform and retains the main features of the signal. The MAD method further smooths the signal and enhances the significance of the partial discharge signal. The SNR improvement effect of the ATWD+MAD method is the best, and it can clearly locate the position of the partial discharge signal. Figure 2 (c) shows the signal processed only by the moving differential averaging (MAD) method. The partial discharge signal is enhanced near 650 m, but the difference between the signal amplitude and the background noise is only 3.9 dB. The smoothing process of the MAD method effectively reduces some noise, but fails to completely remove the high-frequency noise, and the SNR improvement is limited. The significance of the partial discharge signal is not as good as that of the ATWD-MAD method. Figure 2 (d) shows the signal processed by the vector averaging (VM) method. The partial discharge signal is enhanced near 650 m, but the difference between the signal amplitude and the background noise is only 2.7 dB. The processing effect of the VM method on the signal is weak, the noise suppression ability is limited, the significance of the partial discharge signal is low, and the SNR improvement effect is not as good as that of the ATWD-MAD and MAD methods.

[0054] Finally, the distributed positioning of partial discharge is tested. A high-frequency signal of 150 kHz is provided by adding one and two PZTs to the optical fiber to be measured respectively, and the positioning effect is as Figure 3 shown. The results show that the system realizes distributed positioning. Figure 3 (a) shows the variation of the amplitude of the unprocessed original signal with distance. The signal contains obvious noise, and the partial discharge signal is masked by the noise, making it difficult to directly identify the discharge location. Figure 3 (b) The partial discharge signal is significantly enhanced at 937 m, the difference between the signal amplitude and the background noise is obvious, and the characteristics of the partial discharge signal are amplified, showing the spike shape of the signal. Figure 3 (c) The overall fluctuation amplitude of the signal decreases, but the characteristics of the partial discharge signal are still not obvious. The signal amplitude fluctuates between 0.58 and 0.60, without obvious spike features. Figure 3In (d), the partial discharge signals are enhanced at 919 m and 941 m. The difference between the signal amplitude and the background noise is relatively obvious, and the characteristics of the partial discharge signals are amplified, showing two spike signals. The positions (919 m and 941 m) where multiple partial discharge signals can be detected simultaneously have a good SNR improvement effect, but compared with Figure 3 the enhancement effect of the signal in Figure (b) is slightly weaker.

[0055] In one example, the waveform restoration method of partial discharge in a dual-branch distributed partial discharge monitoring system includes: winding and preparing an optical fiber loop 8 to improve the detection ability of weak partial discharge signals; improving the frequency response range through wide pulses; obtaining the original interference signal through the MZI branch, and then restoring the partial discharge ultrasonic signal waveform by the NPS loan note algorithm. However, the detection ability of ordinary single-mode optical fiber for weak partial discharge signals is limited. Therefore, winding the optical fiber loop 8 is adopted to improve the sound pressure sensitivity. The diameter of the optical fiber loop 8 is 20 mm, the height is 5 mm, and the length of the wound optical fiber loop 8 is 5 m (the positioning accuracy is 5 m). The sensitivity of the optical fiber loop 8 is tested. Seven ultrasonic signals with calibrated sound pressure are emitted by an ultrasonic microphone, and the microphone is fixed 10 mm away from the optical fiber loop 8. Record the demodulated signal amplitude after each signal is applied, repeat 5 times, and take the average value to obtain the sound pressure sensitivity curve. The sensitivity is 4.03 rad / Pa. The optical fiber loop 8 is as Figure 4 shown.

[0056] Furthermore, a partial discharge simulation system is built, which consists of a discharge defect, a DC power supply, and a high-voltage coil, to test the waveform restoration of partial discharge signals. The interference signal is collected by the MZI branch, and then the partial discharge signal is demodulated by the NPS demodulation algorithm. The waveform is as Figure 5 shown in (c). At the same time, considering that most current research on optical fiber sensors for partial discharge only focuses on 20 kHz - 300 kHz, the present invention removes the influence of other noises through a band-pass filter (20 kHz - 300 kHz). Figure 5 The frequency domain diagram of the partial discharge signal in (d) shows that there is a response between 20 kHz and 300 kHz. The energy of the signal is mainly concentrated between 20 kHz and 110 kHz, and there is an obvious peak at 56 kHz.

[0057] Customize the discharge electrode, and construct four types of partial discharge defect types to obtain the ultrasonic signal waveforms of four types of partial discharges. There are 1260 groups of data for each type of partial discharge, and a partial discharge defect data set is established. As Figure 6 shown, Figure 6 (a), (b), (c), and (d) in represent one type of discharge defect respectively. The data set is divided according to the ratio of 9:1 for the training set and the test set, and then the defect recognition is carried out by a double-layer one-dimensional convolutional neural network (1D CNN).

[0058] Both the first acquisition unit 102 and the second acquisition unit 104 are connected to the partial discharge defect recognition module, and the partial discharge defect recognition module is built-in with a preset double-layer one-dimensional convolutional neural network. As Figure 7 shown, the double-layer one-dimensional convolutional neural network includes a first convolutional module, a second convolutional module, and a fully connected module, where

[0059] The first convolutional module includes a first convolutional layer, a first batch normalization layer, a first activation function layer, and a first pooling layer connected in sequence. The first convolutional layer extracts the initial dimension feature map in the partial discharge ultrasonic signal waveform data. The first batch normalization layer performs batch normalization on the initial dimension feature map. The first activation function layer is used to enhance the non-linear ability of the features output by the first convolutional layer. The first pooling layer is used to reduce the dimension of the initial dimension feature map and extract local features to obtain the first dimension feature map;

[0060] The second convolutional module includes a second convolutional layer, a second batch normalization layer, a second activation function layer, and a second pooling layer connected in sequence. The second convolutional layer extracts the transition dimension feature map in the first dimension feature map. The second batch normalization layer performs batch normalization on the transition dimension feature map. The second activation function layer is used to enhance the non-linear ability of the features output by the second convolutional layer. The second pooling layer is used to reduce the dimension of the transition dimension feature map and extract local features to obtain the second dimension feature map;

[0061] The fully connected module includes a first fully connected layer, a third activation function layer, a dropout layer, a second fully connected layer, a probability activation function layer, and a classification layer connected in sequence. The first fully connected layer is used to integrate the second dimension feature map and form a global feature vector. The dropout layer is used to randomly discard some neurons in the double-layer one-dimensional convolutional neural network to obtain a generalized feature vector. The second fully connected layer is used to map the generalized feature vector to the output space of the classification task to obtain the class score corresponding to the generalized feature vector. The probability activation function layer is used to convert the class score into a probability distribution so that the classification layer outputs the classification result of the partial discharge defect.

[0062] In one example, the first convolutional module uses a large 15×1 convolutional kernel to capture long-period features, while the second convolutional module uses a small 5×1 convolutional kernel to focus on short-period transient features. This design enhances the fault characterization ability of the photodetector and reduces parameter redundancy. The exponential linear unit (ELU) is selected as the activation function because it can alleviate the vanishing gradient, enhance the robustness to noise, and show higher anti-noise ability than ReLU, Sigmoid, and Tanh in the partial discharge defect pattern recognition task, with specific parameter settings.

[0063] In this embodiment, the first convolutional layer is used to directly extract low-level and multi-scale initial features from the partial discharge ultrasonic signals. Subsequently, batch normalization is performed to reduce the internal covariate shift. Then, the activation function is used to enhance the non-linear expression of the features. Finally, the pooling operation is carried out to reduce the feature dimension and retain the local main information. The function of this module is to capture the details and local patterns of the original signal and provide rich low-level information for the subsequent steps. Based on the first module, the features are further abstracted, and the second convolutional layer is used to extract features of the transition dimension, thereby forming a more abstract and discriminative feature representation. The subsequent same normalization, activation, and pooling steps help highlight the key features and suppress the noise at the same time, which is equivalent to converting the original low-level features into higher-level feature maps that can be used for classification. The fully connected module combines the local abstract features extracted by the second convolutional module into a global feature vector. The first fully connected layer realizes the integration of the features, and then globalizes the signal feature information. The dropout layer, as a regularization means, helps prevent overfitting and enhances the generalization ability of the model on unknown data. The second fully connected layer and the subsequent probability activation layer map the global features to the target classification space and convert the output into a probability distribution. Finally, the specific partial discharge defect recognition result is obtained through the classification layer. Since batch normalization, non-linear activation, and dropout strategies are introduced into the network, the negative impacts of noise and outliers in the input signal on model training are effectively suppressed, and at the same time, the consistency of the output distributions of each layer is improved. In addition, the dropout layer prevents the double-layer one-dimensional convolutional neural network from over-relying on specific features and enhances the generalization ability of the double-layer one-dimensional convolutional neural network in a new environment or for new types of partial discharge defects.

[0064] Please refer to Figure 8 , considering that the dataset is large enough, the dataset is divided according to the ratio of 9:1 for the training set and the test set. Then, the two-layer one-dimensional convolutional neural network is used for defect recognition. It can be seen from the confusion matrix that after processing 4536 groups of random PD type data vectors, the CNN model reaches a classification accuracy of 96.32% on the remaining 504 PD signal vectors.

[0065] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A dual-branch distributed partial discharge monitoring system based on wide and narrow dual pulses, characterized in that: The invention comprises an ultra-narrow linewidth laser (1), a beam splitter (2), an acousto-optic modulator (3), an arbitrary waveform generator (4), an erbium-doped optical power amplifier (5), an optical circulator (6), a test optical fiber (7), an optical fiber ring (8), a coupler (9) and a data demodulation module (10), wherein: The continuous laser signal emitted by the ultra-narrow linewidth laser (1) is divided into a first branch signal and a second branch signal by the beam splitter (2); The first branch signal passes through the acousto-optic modulator (3) driven by the arbitrary waveform generator (4), so that the continuous laser signal is modulated into a wide and narrow dual pulse signal, the wide and narrow dual pulse signal generates a backward Rayleigh scattering signal in the test optical fiber (7), and the wide and narrow dual pulse signal is sequentially transmitted through the erbium-doped optical power amplifier (5) and the first port of the optical circulator (6), the test optical fiber (7), the optical fiber ring (8), and the third port of the optical circulator (6) to the data demodulation module (10); The second branch signal is transmitted to the data demodulation module (10) through the beam splitter (2), the acousto-optic modulator (3), the erbium-doped amplifier, the test optical fiber (7), the optical fiber ring (8) and the coupler (9) in sequence.

2. The dual-branch distributed partial discharge monitoring system based on wide and narrow dual pulses according to claim 1, characterized in that: The splitting ratio between the first output port and the second output port of the beam splitter (2) is 9:1, the first output port of the beam splitter (2) is connected to the acousto-optic modulator (3), and the second output port of the beam splitter (2) is connected to the coupler (9).

3. The dual-branch distributed partial discharge monitoring system based on wide and narrow dual pulses according to claim 1, characterized in that: The test optical fiber (7) responds to an external ultrasonic signal to change the phase of the backscattered Rayleigh light.

4. The dual-branch distributed partial discharge monitoring system based on wide and narrow dual pulses according to claim 1, characterized in that: The data demodulation module (10) comprises a first photodetector (101), a first acquisition unit (102), a data adjustment unit (103), a second acquisition unit (104) and a second photodetector (105); the first photodetector (101) is respectively connected to the third port of the optical circulator (6) and the first acquisition unit (102); the data adjustment unit (103) is respectively electrically connected to the first acquisition unit (102) and the second acquisition unit (104); and the second photodetector (105) is respectively connected to the coupler (9) and the second acquisition unit (104).

5. The dual-branch distributed partial discharge monitoring system based on wide and narrow dual pulses as claimed in claim 4, characterized in that: The coupler (9) is a 3×3 coupler (9), the second photodetector (105) is a 3×3 photodetector, the first port of the 3×3 coupler (9) is connected to the beam splitter (2), the second port of the 3×3 coupler (9) is left empty, the third port of the 3×3 coupler (9) is connected to the optical fiber ring (8), and the fourth port, the fifth port and the sixth port of the 3×3 coupler (9) are all connected to the 3×3 photodetector.

6. The dual-branch distributed partial discharge monitoring system based on wide and narrow dual pulses according to claim 1, characterized in that: The frequency response range of the first branch signal is expressed as: Among them, f max represents the maximum detectable frequency in the first branch signal, f sample represents the sampling rate of the dual-branch distributed partial discharge monitoring system, c represents the propagation speed of light in a vacuum, n represents the refractive index of the test optical fiber (7), and L represents the length of the test optical fiber (7).

7. The dual-branch distributed partial discharge monitoring system based on wide and narrow dual pulses according to claim 4, characterized in that: The first acquisition unit (102) and the second acquisition unit (104) are both connected to a partial discharge defect recognition module, and the partial discharge defect recognition module is equipped with a preset double-layer one-dimensional convolutional neural network.

8. The dual-branch distributed partial discharge monitoring system based on wide and narrow dual pulses as claimed in claim 7, characterized in that: The two-layer one-dimensional convolutional neural network includes a first convolutional module, a second convolutional module and a fully connected module, wherein: The first convolution module includes a first convolution layer, a first batch normalization layer, a first activation function layer and a first pooling layer connected in sequence, the first convolution layer extracts an initial dimensional feature map in the partial discharge ultrasonic signal waveform data, the first batch normalization layer performs batch normalization on the initial dimensional feature map, the first activation function layer is used to enhance the nonlinear ability of the output feature of the first convolution layer, and the first pooling layer is used to reduce the dimension of the initial dimensional feature map and extract local features to obtain a first dimensional feature map; The second convolution module includes a second convolution layer, a second batch normalization layer, a second activation function layer and a second pooling layer connected in sequence, the second convolution layer extracts a transition dimension feature map in the first dimensional feature map, the second batch normalization layer performs batch normalization on the transition dimension feature map, the second activation function layer is used to enhance the nonlinear capability of the output feature of the second convolution layer, and the second pooling layer is used to reduce the dimension of the transition dimension feature map and extract local features to obtain a second dimensional feature map; The fully connected module includes a first fully connected layer, a third activation function layer, a random dropout layer, a second fully connected layer, a probability activation function layer and a classification layer which are connected in sequence. The first fully connected layer is used to integrate the second dimensional feature map and form a global feature vector. The random dropout layer is used to randomly discard some neurons in the double-layer one-dimensional convolutional neural network to obtain a generalized feature vector. The second fully connected layer is used to map the generalized feature vector to the output space of the classification task to obtain a category score corresponding to the generalized feature vector. The probability activation function layer is used to convert the category score into a probability distribution so that the classification layer outputs a classification result of a partial discharge defect.

9. The dual-branch distributed partial discharge monitoring system based on wide and narrow dual pulses according to claim 1, characterized in that: The first branch signal is subjected to data restoration based on an NPS demodulation algorithm to obtain a partial discharge ultrasonic signal waveform corresponding to the test optical fiber (7) in response to an external ultrasonic signal.

10. The dual-branch distributed partial discharge monitoring system based on wide and narrow dual pulses according to claim 4, characterized in that: The sampling rate of the second acquisition unit (104) is 1 MHz / s, the pulse frequency of the arbitrary waveform generator (4) is 100 kHz, and the pulse width is 50 ns.

Images