Cable joint partial discharge detection method and system based on optical fiber voiceprint perception technology

Through fiber optic soundprint perception technology and improved signal processing methods, the noise interference and insufficient environmental adaptability in cable joint local discharge monitoring are solved, efficient and reliable monitoring of cable joint local discharge, and the accuracy of fault positioning is improved Sex and real-time.

CN119001364APending Publication Date: 2024-11-22MIANYANG POWER SUPPLY COMPANY STATE GRID SICHUANELECTRIC POWER
View PDF 0 Cites 3 Cited by

Patent Information

Application Number
CN202411305124.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-09-19
Publication Date
2024-11-22

AI Technical Summary

Technical Problem

The existing cable joint local discharge monitoring technology has obvious shortcomings in power supply and communication stability, environmental adaptability, signal processing and data analysis capabilities, and system deployment and maintenance, making it difficult to achieve efficient and reliable local discharge of cable joints. Monitoring leads to inaccurate fault location and low real-time performance.

Method used

Using a method based on fiber-optic soundprint perception technology, local sound wave signals are obtained through phase-sensitive photo-time domain reflection system, and the signal is feature extracted and classified and recognized by the Meier logarithmic spectrum analysis method and the improved convolutional neural network model to achieve the feature extraction and classification recognition of the signal. Accurate identification and fault location of locally distributed signals.

Benefits of technology

Overcome the problem of noise interference in high-frequency current sensors and pulse current monitoring signals, improve the accuracy and real-time of fault positioning, realize efficient and reliable monitoring of localized cable connectors, and enhance the safety and reliability of cable operation. sex.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119001364A_ABST
    Figure CN119001364A_ABST
Patent Text Reader

Abstract

The invention discloses a cable joint partial discharge detection method and system based on an optical fiber voiceprint perception technology, and the method comprises the steps: converting an optical signal of an optical fiber beam into a sound wave signal through a phase-sensitive optical time domain reflection system based on coherent detection, thereby obtaining a partial discharge sound wave signal at each cable joint; sound wave sensing is achieved through Rayleigh scattering phase change, electromagnetic interference is small, penetrability is good, and sensitivity is high. An optical signal is converted into a sound wave signal, and then voiceprint feature recognition is carried out on a partial discharge signal by adopting a Mel logarithm spectrum analysis method, so that the problems of inaccurate feature extraction and inaccurate recognition caused by the fact that a signal acquired by utilizing a high-frequency current sensor and pulse current monitoring contains a large amount of current noise at present are solved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of cable joint partial discharge detection, and in particular to a cable joint partial discharge detection method and system based on optical fiber voiceprint sensing technology. Background Art

[0002] In high-voltage power systems, cable joints are extremely critical components, and their reliability is directly related to the safe operation of the power system. However, due to the complex structure and harsh operating environment of cable joints, partial discharge often occurs. Partial discharge is an incomplete electrical discharge that will cause the gradual aging of cable insulation materials, which may eventually lead to cable failures and even large-scale power outages in the power system. Therefore, monitoring and early warning of partial discharge in cable joints are particularly important.

[0003] At present, the monitoring of partial discharge of cable joints mainly relies on broadband high-frequency current sensors and high-frequency pulse current partial discharge monitoring devices. These devices collect high-frequency current pulse data generated by partial discharge and upload them to the monitoring platform for analysis. However, these existing technologies have many defects and shortcomings in practical applications.

[0004] First, these monitoring devices require a stable power supply and communication environment, but high-voltage cables are usually laid in trough boxes or cable trenches in the central urban area, where the environment is harsh and the stability of power supply and communication is difficult to guarantee. Once the power supply is interrupted or the communication fails, the monitoring device will not be able to work properly, resulting in low coverage of partial discharge monitoring and inability to achieve real-time monitoring and early warning of cable joints.

[0005] Secondly, these monitoring devices have deficiencies in environmental adaptability such as waterproof and dustproof. Cable joints are often in humid and dusty environments, and traditional monitoring devices are easily damaged in such environments, affecting their normal operation and data accuracy. Therefore, how to improve the environmental adaptability of monitoring devices, especially waterproof performance, is an urgent problem to be solved.

[0006] In addition, existing partial discharge monitoring devices also have deficiencies in signal processing and data analysis. The signals collected by high-frequency current sensors and pulse current monitoring devices often contain a lot of noise. How to accurately extract the characteristics of partial discharge signals from these noises and achieve high-precision fault location is a technical challenge. The current monitoring system has not yet reached the ideal level in terms of signal processing algorithms and data analysis capabilities, resulting in low accuracy and real-time performance of fault location.

[0007] In addition, existing partial discharge monitoring technologies mostly use active equipment and need to rely on external power supply, which not only increases the complexity and maintenance cost of the system, but also is difficult to implement in some special environments. For example, in some underground cable joints far away from the power supply, traditional monitoring devices are difficult to deploy and maintain, resulting in many blind spots in partial discharge monitoring in these areas and inability to achieve full coverage.

[0008] In summary, the existing cable joint partial discharge monitoring technology has obvious deficiencies in terms of power supply and communication stability, environmental adaptability, signal processing and data analysis capabilities, and system deployment and maintenance, and it is difficult to meet the actual needs of high-voltage cable operation and maintenance. Therefore, a new partial discharge monitoring method and equipment is urgently needed to overcome the defects of the existing technology, realize efficient and reliable monitoring of cable joint partial discharge, and improve the safety and reliability of cable operation. Summary of the invention

[0009] Based on the problems raised by the above background technology, the purpose of the present invention is to provide a cable joint partial discharge detection method and system based on optical fiber voiceprint sensing technology, which solves the problem that the signals collected by high-frequency current sensors and pulse current monitoring devices often contain a lot of noise, and the existing technology cannot accurately extract the characteristics of partial discharge signals from these noises, thus leading to inaccurate fault location.

[0010] The present invention is achieved through the following technical solutions:

[0011] The first aspect of the present invention provides a cable joint partial discharge detection method based on optical fiber voiceprint sensing technology, comprising the following steps:

[0012] Step S1, using a phase-sensitive optical time domain reflectometry system to obtain a partial discharge acoustic wave signal at each cable joint;

[0013] Step S2, using the Mel logarithmic spectrum analysis method to extract features of the partial discharge sound wave signal to obtain partial discharge soundprint features;

[0014] Step S3, constructing an improved convolutional neural network model, and using the improved convolutional neural network model to classify and identify the partial discharge voiceprint features;

[0015] Step S4: locating the cable joint with the partial discharge fault according to the classification and identification result.

[0016] In the above technical solution, the optical fiber is not only a medium for transmitting optical signals, but also a sensor for sensing external physical quantities. During the transmission of optical signals along the optical fiber, backscattered light is generated at each position. By collecting these backscattered lights and monitoring their parameter changes, changes in external physical quantities can be sensed. When sound waves act on the optical fiber, they will cause axial strain in the optical fiber, thereby changing the phase of the Rayleigh scattered signal in the optical fiber. Sound wave perception is achieved by measuring the optical phase change caused by the change in the axial strain of the optical fiber.

[0017] This method uses the above principle to convert the optical signal of the optical fiber beam into an acoustic signal through a phase-sensitive optical time-domain reflection system based on coherent detection, thereby obtaining the partial discharge acoustic signal at each cable joint and realizing acoustic wave perception by using the phase change of Rayleigh scattering. The partial discharge process is accompanied by various physical and chemical phenomena, such as current pulses, electromagnetic radiation, ultrasound, local overheating, dielectric loss, strong light, etc. The method of using optical fiber to detect partial discharge ultrasound of high-voltage cables has the advantages of little electromagnetic interference, good penetration, and high sensitivity.

[0018] The optical signal is converted into an acoustic signal, and then the Mel-log spectrum analysis method is used to perform voiceprint feature recognition on the partial discharge signal. This overcomes the problem of inaccurate feature extraction and inaccurate recognition caused by a large amount of current noise contained in the signals currently collected by high-frequency current sensors and pulse current monitoring.

[0019] Then, the improved convolutional neural network algorithm is used to extract the characteristics of the acoustic wave signal, more accurately identify different partial discharge soundprint patterns, and obtain the accurate partial discharge signal type and partial discharge signal location. Combining the partial discharge signal type and partial discharge signal location with the optical fiber length, the partial discharge fault information with distance information can be determined, thereby determining the cable joint with the fault.

[0020] In an optional embodiment, the step S1 includes:

[0021] Injecting narrowband pulse light generated by a pulse light source into a detection optical fiber, the pulse light generates Rayleigh scattering when propagating in the detection optical fiber, and obtaining a scattered light signal;

[0022] Calculating a phase change signal of the scattered light signal using a phase-sensitive optical time domain reflectometry system based on coherent detection;

[0023] The phase change signal is demodulated to convert the phase change signal into a partial discharge acoustic wave signal in the time domain.

[0024] In an optional embodiment, the step S12 includes the following steps:

[0025] Step S121, introducing the scattered light signal into a light detector, and the light detector converts the scattered light signal into an electrical signal;

[0026] Step S122, using a preamplifier to sequentially amplify and filter the electrical signal to obtain a processed electrical signal;

[0027] Step S123, using a lock-in amplifier to demodulate the processed electrical signal to obtain a demodulated signal, and extracting phase information and amplitude information from the processed electrical signal;

[0028] Step S123, sampling the demodulated signal to obtain a digital signal, and extracting the frequency characteristics of the digital signal by using fast Fourier transform;

[0029] Step S124: Calculate the frequency signal and the phase signal to obtain a partial discharge acoustic wave signal in the time domain.

[0030] In an optional embodiment, the step S2 includes:

[0031] Step S21, using an improved Mel logarithmic spectrum analysis method to convert the local discharge acoustic wave signal in the time domain into a frequency domain signal, and extracting the frequency domain features of the frequency domain signal;

[0032] Step S22, removing the noise features in the frequency domain features to obtain key features; wherein the key features include the frequency, amplitude and phase of the partial discharge voiceprint signal;

[0033] Step S23: normalize the key features, and select partial discharge voiceprint features from the normalized key features.

[0034] In an optional embodiment, step S23 includes the following steps:

[0035] Step S231: Use the information gain algorithm to evaluate the importance of the normalized key features to obtain feature information gain;

[0036] Step S232: selecting important features from the normalized key features according to the feature information gain;

[0037] Step S233, weighting the important features, and expanding the weighted important features using sliding window technology and data enhancement methods to obtain expanded features;

[0038] Step S234: perform feature fusion on the extended features to obtain partial release voiceprint features.

[0039] In an optional embodiment, the improved convolutional neural network model in step S3 includes: a multi-scale adaptive convolution module, a dynamic time-frequency fusion module and a hierarchical attention mechanism;

[0040] The multi-scale adaptive convolution module includes multiple parallel convolution kernels, and the multiple parallel convolution kernels extract time-frequency domain joint features, frequency domain features and time domain features from the local discharge voiceprint features;

[0041] The dynamic time-frequency fusion module fuses the time-frequency domain joint feature, the frequency domain feature and the time domain feature through an adaptive filter to obtain a fusion feature;

[0042] The hierarchical attention mechanism performs attention training on the fused features to obtain trained features, and the trained features are used to determine the type and position of the partial discharge signal.

[0043] In an optional embodiment, the size and step size of the convolution kernel are dynamically adjusted according to the frequency characteristics of the partial discharge voiceprint feature, wherein the dynamic adjustment process is:

[0044]

[0045] in, is the weight of the jth convolution kernel of the i-th layer at the m-th time-frequency point, α is the adjustment amplitude, β is the frequency adjustment factor, is the phase shift, f adjust (f input (t,f)) is the adjustment function of the input signal frequency characteristics, f input (t,f) is the time-frequency domain representation of the input signal. t represents the time variable of the input signal, which is the change of the signal in time. f represents the frequency variable of the input signal, which is the change of the signal in frequency.

[0046] In an optional embodiment, the adaptive filter dynamically adjusts filter parameters according to the partial discharge voiceprint feature, wherein the adjustment process of dynamically adjusting the filter parameters is as follows:

[0047] H filter (t,f)=H0(t,f)·(1+δ·cos(2πηt+θ)·g adjust (f));

[0048] Among them, H filter (t, f) is the transfer function of the adaptive filter, H0(t, f)· is the initial transfer function, δ is the filter adjustment amplitude, η is the time adjustment factor, θ is the phase adjustment, g adjust (f) is the adjustment function of the frequency characteristics.

[0049] In an optional embodiment, the hierarchical attention mechanism calculates the convolutional layer attention weight and the pooling layer attention weight respectively according to the weight calculation value of the fusion feature, wherein the calculation process is as follows:

[0050]

[0051] Among them, γ m (t,f) is the attention weight of the mth convolutional feature, e m (t,f) is the weight calculation value of the mth feature, λ n (t,f) is the attention weight of the nth pooling feature, e n (t,f) is the calculated weight value of the nth feature.

[0052] The second aspect of the present invention provides a cable joint partial discharge detection system based on optical fiber voiceprint sensing technology, comprising:

[0053] An information acquisition module, wherein the information acquisition module is used to acquire a partial discharge acoustic wave signal at each cable joint using a phase-sensitive optical time domain reflectometry system;

[0054] A feature extraction module, wherein the feature extraction module is used to extract features of the partial discharge sound wave signal by using a Mel logarithmic spectrum analysis method to obtain a partial discharge soundprint feature;

[0055] A classification and recognition module, wherein the classification and recognition module is used to construct an improved convolutional neural network model, and utilize the improved convolutional neural network model to classify and recognize the partial discharge voiceprint features;

[0056] A fault location module is used to locate the cable joint with partial discharge fault according to the result of classification and identification.

[0057] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0058] 1. Convert the optical signal into an acoustic signal, and then use the Mel logarithmic spectrum analysis method to identify the voiceprint features of the partial discharge signal, which overcomes the problem of inaccurate feature extraction and inaccurate identification caused by a large amount of current noise in the signals collected by high-frequency current sensors and pulse current monitoring.

[0059] 2. The optical fiber itself is a sound sensor. Optical fiber sensing itself does not require power. The coherent scattering echo is transmitted back along the detection optical fiber body. No additional transmission equipment is required, thus achieving passive cable partial discharge sensing and data feedback. In addition, the optical fiber can still transmit light waves in water, is not affected by water vapor, and can cope with harsh environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] In order to more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present invention and should not be regarded as limiting the scope. For those of ordinary skill in the art, other relevant drawings can be obtained based on these drawings without creative work. In the drawings:

[0061] Figure 1 A schematic diagram of the internal logic structure of a DAS module provided in Embodiment 1 of the present invention;

[0062] Figure 2 A schematic diagram of the process of the Mel-logarithmic spectrum analysis method provided in Example 1 of the present invention;

[0063] Figure 3 A schematic diagram of a process for classifying and identifying the partial discharge voiceprint features using an improved convolutional neural network model provided in Example 1 of the present invention;

[0064] Figure 4 The improved convolutional neural network model training iterative process provided in Example 1 of the present invention. DETAILED DESCRIPTION

[0065] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with embodiments and drawings. The exemplary embodiments of the present invention and their description are only used to explain the present invention and are not intended to limit the present invention.

[0066] Example 1

[0067] Embodiment 1 of the present invention provides a method for detecting partial discharge of a cable joint based on optical fiber voiceprint sensing technology, the method comprising the following steps:

[0068] Step S1, using a phase-sensitive optical time domain reflectometry system to obtain a partial discharge acoustic wave signal at each cable joint;

[0069] Step S2, using the Mel logarithmic spectrum analysis method to extract features of the partial discharge sound wave signal to obtain partial discharge soundprint features;

[0070] Step S3, constructing an improved convolutional neural network model, and using the improved convolutional neural network model to classify and identify the partial discharge voiceprint features;

[0071] Step S4: locating the cable joint with the partial discharge fault according to the classification and identification result.

[0072] It should be noted that since the signals collected by high-frequency current sensors and pulse current monitoring devices often contain a lot of noise, the existing technology cannot accurately extract the characteristics of partial discharge signals from these noises, resulting in inaccurate fault location.

[0073] In view of the above problems, the embodiment of the present invention proposes a cable joint partial discharge detection method based on optical fiber voiceprint sensing technology, in which the optical fiber is not only a medium for transmitting optical signals, but also a sensor for sensing external physical quantities; during the transmission of optical signals along the optical fiber, backscattered light is generated at each position, and the changes in external physical quantities can be sensed by collecting these backscattered lights and monitoring their parameter changes. When sound waves act on the optical fiber, the optical fiber will be strained in the axial direction, thereby changing the phase of the Rayleigh scattered signal in the optical fiber, and sound wave perception is achieved by measuring the light phase change caused by the change in the axial strain of the optical fiber.

[0074] This method uses the above principle to convert the optical signal of the optical fiber beam into an acoustic signal through a phase-sensitive optical time-domain reflection system based on coherent detection, thereby obtaining the partial discharge acoustic signal at each cable joint and realizing acoustic wave perception by using the phase change of Rayleigh scattering. The partial discharge process is accompanied by various physical and chemical phenomena, such as current pulses, electromagnetic radiation, ultrasound, local overheating, dielectric loss, strong light, etc. The method of using optical fiber to detect partial discharge ultrasound of high-voltage cables has the advantages of little electromagnetic interference, good penetration, and high sensitivity.

[0075] The optical signal is converted into an acoustic signal, and then the Mel logarithmic spectrum analysis method is used to identify the voiceprint features of the partial discharge signal. This overcomes the problem of inaccurate feature extraction and inaccurate identification caused by a large amount of current noise contained in the signals collected by high-frequency current sensors and pulse current monitoring.

[0076] Then, the improved convolutional neural network algorithm is used to extract the characteristics of the acoustic wave signal, more accurately identify different partial discharge soundprint patterns, and obtain the accurate partial discharge signal type and partial discharge signal location. Combining the partial discharge signal type and partial discharge signal location with the optical fiber length, the partial discharge fault information with distance information can be determined, thereby determining the cable joint with the fault.

[0077] This method is based on the fiber optic voiceprint sensing technology, which requires two parts of equipment: an intelligent algorithm bin for collecting fiber optic partial discharge voiceprints and a fiber optic voiceprint pickup device. The intelligent algorithm bin for collecting fiber optic partial discharge voiceprints is installed in the computer room. The single-mode optical cable for detecting voiceprints is laid along the cable and terminated in the computer room and at each cable joint. The detection optical cable is connected to the intelligent algorithm bin for collecting fiber optic partial discharge voiceprints on the computer room side. At the cable joint, the fiber optic voiceprint pickup device is connected to the front and rear sections of the detection optical cable.

[0078] The optical fiber partial discharge soundprint collection intelligent algorithm warehouse injects narrowband pulse light into the detection optical cable. Every time the pulse light passes through a fiber optic soundprint pickup device, it will be affected by the partial discharge sound wave of the corresponding cable joint, causing the echo light phase to change, thereby obtaining the partial discharge soundprint waveform on the cable joint and the corresponding optical fiber length. After partial discharge soundprint recognition algorithm analysis, partial discharge faults are diagnosed, and the optical fiber partial discharge soundprint collection intelligent algorithm warehouse pushes partial discharge alarm information with distance information to the partial discharge monitoring platform through the MQTT protocol.

[0079] In order to obtain accurate data, it should be noted in the specific implementation process that the optical fiber itself is a sound sensor, so the main component of the optical fiber voiceprint pickup device is the bare optical fiber. The detection optical cable can be laid around the cable joint to detect the minimum partial discharge signal of 150pC, but the initial partial discharge of the cable joint is usually only 10pC, so the detection optical cable needs to be improved.

[0080] Specifically, single-mode optical fiber is selected, and its specifications conform to the ITU-T G.652 standard, which is suitable for long-distance transmission and detection. According to the cable laying path and the distribution of connectors, a fiber laying path map is formulated to determine the laying route of the optical fiber from the machine room to each cable connector, to ensure that the optical fiber covers all the cable connectors to be monitored, and to arrange the single-mode optical fiber along the cable path. The optical fiber is fixed in the cable trough or cable trench. During the laying of the optical fiber, sharp bends and mechanical damage should be avoided to ensure the integrity and transmission performance of the optical fiber. While laying the optical fiber, the optical fiber is protected by using an optical fiber protection tube or protective sleeve to avoid the influence of water vapor, corrosion, and mechanical impact on the optical fiber. The optical fiber is terminated in the machine room and each cable connector, and FC / PC or SC / APC optical fiber connectors are used. The optical fiber voiceprint pickup device is installed at the cable connector. The pickup device includes a bottom shell, a bare optical fiber and an upper cover. The bottom shell has a circle of L-shaped hooks in the middle for winding the bare optical fiber, and the upper cover has a sound insulation sponge to reduce environmental noise interference. The bare optical fiber of the pickup device is connected to the detection optical fiber laid along the cable to ensure the continuity of the optical fiber at the connection point and the integrity of the optical signal transmission. The optical fiber soundprint pickup device is tightly fixed on the cable joint with adhesive tape and cable ties, and with honeycomb holes, the optical fiber soundprint pickup device can not only collect partial discharge soundprints through air vibration, but also has a bone conduction effect, which can improve the collection sensitivity of partial discharge soundprints. In addition, the 20-meter winding bare optical fiber can greatly improve the perception of sound waves, and the partial discharge detection capability can be reduced to 10pC.

[0081] After completing the fiber optic layout and installing the pickup device, debug the entire fiber optic system to ensure the transmission quality of the optical signal and the normal operation of the voiceprint pickup device, and to ensure that the local discharge acoustic wave signal at the cable joint can be obtained.

[0082] It should be noted that fiber optic sensing itself does not require power supply. The coherent scattering echo is transmitted back along the detection fiber body, and no additional transmission equipment is required, thus achieving passive cable partial discharge sensing and data feedback. In addition, the optical fiber can still transmit light waves in water, is not affected by water vapor, and can cope with harsh environments.

[0083] In the optical fiber partial discharge soundprint collection intelligent algorithm warehouse in the computer room, narrowband pulse light is injected into the detection optical fiber. Specifically, a narrowband pulse light source with a central wavelength of 1550nm, a pulse width of 10ns, and a repetition frequency of 1kHz is selected; an optical fiber partial discharge soundprint collection intelligent algorithm warehouse is arranged in the computer room. The optical fiber partial discharge soundprint collection intelligent algorithm warehouse integrates a DAS module, an FPGA module, an edge analysis gateway, and a power module. The optical fiber is used as a sensor, and the DAS module is used for soundprint collection. The DAS module is as follows: Figure 1 As shown, the DAS module then transmits the collected voiceprint information to the FPGA module for data processing. The intelligent algorithm warehouse contains a pulse light source, a fiber coupler, a light detector, and a data processing module. The pulse light source is connected to the detection fiber through a fiber coupler. The intelligent algorithm is used to monitor the signal transmission in the optical fiber in real time and dynamically adjust the output parameters of the pulse light source, including pulse width, frequency, and power, so that the pulse light can be stably transmitted under different environmental conditions:

[0084]

[0085] Where P(t) is the power of the pulse light, P0 is the initial power of the pulse light, σ is the pulse width, ω is the angular frequency, The initial phase is set; an environmental sensor is set in the intelligent algorithm warehouse to monitor the temperature, humidity and vibration of the machine room and the optical fiber path in real time, and the output of the pulse light source is dynamically adjusted according to environmental changes, so that the pulse light can still maintain the preset signal-to-noise ratio in harsh environments:

[0086]

[0087] Where ΔP is the change in pulse light power, The rate of change of temperature to pulse light power, ΔT is the temperature change, is the rate of change of humidity to pulse light power, ΔH is the humidity change, is the rate of change of vibration to pulse light power, and ΔV is the vibration change.

[0088] The injected pulsed light is amplified by an optical amplifier to ensure that the acoustic wave signal can be obtained at each cable joint. The local discharge acoustic wave signal at each cable joint is obtained by a phase-sensitive optical time domain reflectometry system based on coherent detection, and the acoustic wave perception is realized by using the phase change of Rayleigh scattering:

[0089]

[0090] in, is the phase change of the pulsed light, λ is the wavelength of the pulsed light, and ΔL is the change in the length of the optical fiber caused by the partial discharge acoustic wave. The edge computing module in the intelligent algorithm warehouse analyzes the optical signal transmitted in real time, intelligently adjusts the parameters of the pulsed light source and the optical amplifier according to the analysis results, and trains the system's response speed and detection accuracy to sudden partial discharge signals:

[0091]

[0092] Among them, ΔP adj is the adjustment amount of pulse light power, K is the feedback gain coefficient, is the measured phase change, is the reference phase change

[0093] In an optional embodiment, the step S1 includes:

[0094] Step S11, injecting narrowband pulse light generated by a pulse light source into a detection optical fiber, and generating Rayleigh scattering when the pulse light propagates in the detection optical fiber to obtain a scattered light signal;

[0095] Step S12: Calculate the phase change signal of the scattered light signal using a phase-sensitive optical time-domain reflectometry system based on coherent detection, demodulate the phase change signal, and convert the phase change signal into a partial discharge acoustic wave signal in the time domain.

[0096] It should be noted that when the narrowband pulse light generated by the pulse light source is injected into the detection optical fiber, Rayleigh scattering occurs when the pulse light propagates in the optical fiber. The pulse light electric field intensity is expressed as:

[0097]

[0098] Among them, E p (t) is the pulsed optical electric field intensity, A p is the pulse light amplitude, f p is the pulse light frequency, is the initial phase of the pulse light.

[0099] The scattered light signal contains the phase information of all the partial discharge sound waves along the way, and the scattered light electric field intensity is expressed as:

[0100]

[0101] Among them, E r (t) is the scattered light electric field intensity, A r is the scattered light amplitude, f r is the scattered light frequency, is the initial phase of the scattered light.

[0102] By using the phase-sensitive characteristics of optical fiber and measuring the phase change of Rayleigh scattered light, the partial discharge acoustic wave signal at the cable joint is sensed, and the change of the acoustic wave signal is reflected in the phase change of the optical fiber. The acquired phase change signal is demodulated to restore the partial discharge acoustic wave signal at the cable joint. The phase change signal is converted into an acoustic wave signal in the time domain using an optical detector and a data processing module. The demodulation process is as follows:

[0103]

[0104] Among them, S(t) is the restored sound wave signal, A i is the amplitude of each frequency component, f i is the frequency of each frequency component, is the initial phase of each frequency component.

[0105] In an optional embodiment, the step S12 includes the following steps:

[0106] Step S121, introducing the scattered light signal into a light detector, and the light detector converts the scattered light signal into an electrical signal;

[0107] Step S122, using a preamplifier to sequentially amplify and filter the electrical signal to obtain a processed electrical signal;

[0108] Step S123, using a lock-in amplifier to demodulate the processed electrical signal to obtain a demodulated signal, and extracting phase information and amplitude information from the processed electrical signal;

[0109] Step S123, sampling the demodulated signal to obtain a digital signal, and extracting the frequency characteristics of the digital signal by using fast Fourier transform;

[0110] Step S124: Calculate the frequency signal and the phase signal to obtain a partial discharge acoustic wave signal in the time domain.

[0111] It should be noted that the Rayleigh scattered signal transmitted through the optical fiber is introduced into the optical detector, which receives and converts the optical signal into an electrical signal:

[0112]

[0113] Where I(t) is the received optical signal strength, I0 is the DC component, and I1 is the amplitude of the AC component. is the signal phase.

[0114] The received electrical signal is amplified by the preamplifier so that the signal strength meets the input requirements of the signal processing module:

[0115] Iamp (t) = G·I(t);

[0116] Among them, I amp (t) is the amplified electrical signal, and G is the gain of the amplifier.

[0117] The amplified electrical signal is filtered to remove high-frequency noise and low-frequency interference signals. The filtered signal is expressed as:

[0118]

[0119] Among them, I filt (t) is the electric signal after filtering, H i (f) is the filter transfer function, and N is the number of filters.

[0120] The filtered signal is demodulated using a lock-in amplifier to extract the phase and amplitude information of the signal:

[0121]

[0122] Among them, V demod (t) is the demodulated signal voltage, is the demodulation reference signal.

[0123] The demodulated signal is sampled by the data acquisition module to convert the analog signal into a digital signal. The sampled signal is expressed as:

[0124] S[n]=V demod (t n );

[0125] Among them, S[n] is the sampled digital signal, t n is the time of the nth sampling point.

[0126] The sampled digital signal is processed and the frequency characteristics of the signal are extracted through fast Fourier transform:

[0127]

[0128] Among them, X(k) is the frequency domain signal, N is the number of sampling points, and k is the frequency point.

[0129] According to the frequency characteristics and phase information, the acoustic wave signals at each position of the partial discharge sensing optical fiber are restored, and the time domain signal is restored using the inverse fast Fourier transform. The restored time domain signal is the partial discharge soundprint signal:

[0130]

[0131] Among them, s(t) is the restored time domain signal, X(k) is the frequency domain signal, and N is the number of frequency points.

[0132] In an optional embodiment, the step S2 includes:

[0133] Step S21, using an improved Mel logarithmic spectrum analysis method to convert the local discharge acoustic wave signal in the time domain into a frequency domain signal, and extracting the frequency domain features of the frequency domain signal;

[0134] Step S22, removing the noise features in the frequency domain features to obtain key features; wherein the key features include the frequency, amplitude and phase of the partial discharge voiceprint signal;

[0135] Step S23: normalize the key features, and select partial discharge voiceprint features from the normalized key features.

[0136] It should be noted that in the present invention, the improved Mel logarithmic spectrum analysis method is used to extract features. The extraction process is as follows: Figure 2 As shown, the local discharge acoustic wave signal in the time domain is converted into a frequency domain signal through fast Fourier transform, and the frequency is segmented, the segment amplitude of each segment is calculated, and the Mel logarithmic spectrum is calculated according to the segment amplitude.

[0137] Specifically, the calculation process is as follows:

[0138]

[0139] Among them, MFSC is a new spectrum feature, S(f m ) is the amplitude of the mth frequency component, f m is the mth frequency component, f0 is the reference frequency, M is the total number of frequency components, and p is the parameter for adjusting the spectrum characteristics.

[0140] The extracted frequency domain features are screened, and irrelevant features and noise features are eliminated, leaving only the key features representing the partial discharge voiceprint. The key features in the partial discharge voiceprint are manifested as the enhancement of specific frequency components and the change of phase, from which irrelevant features and noise features can be eliminated.

[0141] Furthermore, the feature processing function can be used for screening, and the key features obtained after processing are expressed as follows:

[0142] F filtered =H(F);

[0143] Among them, F filtered is the processed feature, and H(F) is the feature processing function.

[0144] The key features are normalized to ensure that the eigenvalues ​​of the key features are within the same dimensional range. In the embodiment of the present invention, the frequency, amplitude and phase of the partial discharge voiceprint signal are normalized by the following formula:

[0145]

[0146] Among them, x i ′ is the normalized eigenvalue, x i is the original eigenvalue, min(x) and max(x) are the minimum and maximum values ​​of the feature set, respectively.

[0147] The partial discharge voiceprint features representing the partial discharge voiceprint are selected from the normalized key features.

[0148] In an optional embodiment, step S23 includes the following steps:

[0149] Step S231: Use the information gain algorithm to evaluate the importance of the normalized key features to obtain feature information gain;

[0150] Step S232: selecting important features from the normalized key features according to the feature information gain;

[0151] Step S233, weighting the important features, and expanding the weighted important features using sliding window technology and data enhancement methods to obtain expanded features;

[0152] Step S234: perform feature fusion on the extended features to obtain partial release voiceprint features.

[0153] It should be noted that the information gain algorithm is used to calculate the contribution of each normalized key feature to the partial discharge voiceprint recognition, that is, the information gain value of the key feature for the frequency component, amplitude and phase changes. Since the partial discharge voiceprint feature is a high-frequency and short-term signal, the frequency, amplitude and phase are used as target variables, and the information gain of the key feature for each target variable is calculated using the following formula:

[0154] IG(T,X)=H(T)-H(T|X);

[0155] Among them, IG(T,X) is the information gain of the key feature X to the target variable T, H(T) is the entropy of the target variable, and H(T|X) is the entropy of the target variable under the condition of feature X.

[0156] The information gain of each key feature is calculated, and the key features with higher information gain are retained as important features. For partial discharge voiceprint signals, the selected features can effectively reflect the physical properties of partial discharge activities. The selected important features are as follows:

[0157] S={x i ′ IG(T,x i′ )≥θ};

[0158] Among them, S is the selected feature set, and θ is the threshold of information gain, which ensures that only features that contribute significantly to local release voiceprint recognition are retained.

[0159] The important features are weighted to further highlight their impact on partial discharge voiceprint recognition. The feature weighting formula is as follows:

[0160]

[0161] Among them, w i is feature x i ′ The weight of , n is the total number of features.

[0162] The sliding window technology and data enhancement method are used to expand the weighted important features to increase the diversity and robustness of the features. The expansion of the partial discharge voiceprint features helps to capture the dynamic changes of the signal and improve the generalization ability of the model. The expansion is either left or right expansion or frame skipping expansion. The expansion formula is as follows:

[0163]

[0164] in, is the expanded feature value, and m is the size of the sliding window.

[0165] The expanded features are fused to obtain the partial discharge voiceprint feature, which combines a variety of feature information to ensure the diversity and representativeness of the partial discharge voiceprint, and helps to improve the accuracy and stability of partial discharge voiceprint recognition.

[0166] Furthermore, the formula for fusion processing is as follows:

[0167]

[0168] Among them, F is the fused feature set, w i is the weight of the feature, is the extended eigenvalue.

[0169] Through step S231 to step S234, the partial discharge voiceprint features are refined and optimized, thereby improving the representativeness and recognition accuracy of the features. By combining the high-frequency and short-term partial discharge signal characteristics, using information gain, sliding window, data enhancement and feature fusion technology, the problem of inaccurate partial discharge voiceprint feature recognition is solved, and the accuracy and real-time performance of cable joint partial discharge detection are improved.

[0170] In an optional embodiment, the improved convolutional neural network model in step S3 includes: a multi-scale adaptive convolution module, a dynamic time-frequency fusion module and a hierarchical attention mechanism;

[0171] The multi-scale adaptive convolution module includes multiple parallel convolution kernels, and the multiple parallel convolution kernels extract time-frequency domain joint features, frequency domain features and time domain features from the local discharge voiceprint features;

[0172] The dynamic time-frequency fusion module fuses the time-frequency domain joint feature, the frequency domain feature and the time domain feature through an adaptive filter to obtain a fusion feature;

[0173] The hierarchical attention mechanism performs attention training on the fused features to obtain trained features, and the trained features are used to determine the type and position of the partial discharge signal.

[0174] In an optional embodiment, the size and step size of the convolution kernel are dynamically adjusted according to the frequency characteristics of the partial discharge voiceprint feature, wherein the dynamic adjustment process is:

[0175]

[0176] in, is the weight of the jth convolution kernel of the i-th layer at the m-th time-frequency point, α is the adjustment amplitude, β is the frequency adjustment factor, is the phase shift, f adjust (f input (t,f)) is the adjustment function of the input signal frequency characteristics, f input (t,f) is the time-frequency domain representation of the input signal, t represents the time variable of the input signal, which is the change of the signal in time, and f represents the frequency variable of the input signal, which is the change of the signal in frequency.

[0177] It should be noted that the frequency characteristics of the cable joint partial discharge signal will change with time. By introducing the dynamic adjustment parameters α, β and The convolution kernel can adapt to different signal characteristics, thereby improving the sensitivity and accuracy of detection, and adjusting the function f adjust (f input (t,f)) is adjusted according to the frequency characteristics of the input signal so that the convolution kernel can capture the characteristics of the partial discharge signal more accurately.

[0178] The adjusted multi-scale adaptive convolution module can capture features in different frequency ranges, where the capture process is as follows:

[0179]

[0180] in, is the time-frequency domain feature after the i-th convolution, K i is the total number of convolution kernels in the i-th layer, M i is the total number of time-frequency points in the i-th layer, is the response of the jth convolution kernel at the mth time-frequency point, Represents the response of the convolution kernel at different frequency points.

[0181] In the detection of partial discharge at cable joints, the convolution kernel is dynamically adjusted according to the characteristic changes of the partial discharge signal at different times and frequencies to better capture the characteristics of the partial discharge signal and effectively extract signal features in complex environments.

[0182] In an optional embodiment, the adaptive filter dynamically adjusts filter parameters according to the partial discharge voiceprint feature, wherein the adjustment process of dynamically adjusting the filter parameters is as follows:

[0183] H filter (t,f)=H0(t,f)·(1+δ·cos(2πηt+θ)·g adjust (f));

[0184] Among them, H filter (t, f) is the transfer function of the adaptive filter, H0(t, f)· is the initial transfer function, δ is the filter adjustment amplitude, η is the time adjustment factor, θ is the phase adjustment, g adjust (f) is the adjustment function of the frequency characteristics.

[0185] It should be noted that the adaptive filter can adjust the filter parameters according to the real-time signal characteristics in the cable joint partial discharge detection. The filter transfer function H filter (t,f) can be optimized according to the change of the signal. By adjusting the parameter filter, the noise can be better filtered out and the key features of the partial discharge signal can be retained, thereby improving the detection effect.

[0186] The three features are fused by using an adaptive filter after dynamically adjusting the filter parameters to generate a comprehensive vector feature. Specifically, the fusion process is as follows:

[0187] F fused (t,f)=α(t,f)·F time (t)+β(t,f)·F freq (f)+γ(t,f)·F joint (t,f);

[0188] Among them, F fused (t,f) is the feature after time-frequency fusion, α(t,f) and β(t,f) are the fusion weights of time and frequency features, γ(t,f) is the fusion weight of joint features, and F time (t) is the time domain feature, Ffreq (f) is the frequency domain feature, F joint (t,f) is the joint feature in time and frequency domains.

[0189] Furthermore, in order to better detect the partial discharge signal of the cable joint, the fusion of time domain and frequency domain information is the key. The fusion weights α(t,f) and β(t,f) are dynamically adjusted to adapt to different signal characteristics to ensure that the comprehensive feature vector F fused (t,f) can accurately reflect the characteristics of the partial discharge signal, and the joint feature F joint (t,f) combines the information in the time and frequency domains to improve the robustness of signal recognition.

[0190] In an optional embodiment, the hierarchical attention mechanism calculates the convolutional layer attention weight and the pooling layer attention weight respectively according to the weight calculation value of the fusion feature, wherein the calculation process is as follows:

[0191]

[0192] Among them, γ m (t,f) is the attention weight of the mth convolutional feature, e m (t,f) is the weight calculation value of the mth feature, λ n (t,f) is the attention weight of the nth pooling feature, e n (t,f) is the calculated weight value of the nth feature.

[0193] It should be noted that the calculation of attention weights ensures the importance of key features in cable joint partial discharge detection. m (t,f) and λ n (t,f) weights each feature to ensure that the most important feature gets the highest weight in the time-frequency domain, thereby improving the effect and accuracy of signal detection.

[0194] The present invention introduces a hierarchical attention mechanism between the convolution layer and the pooling layer to enhance the weight of key features and suppress irrelevant noise. The attention mechanism is calculated based on the determination of the convolution layer attention weight and the pooling layer attention weight. Specifically, the calculation process is as follows:

[0195]

[0196] Among them, F attention (t,f) is the time-frequency domain feature after attention mechanism processing, γ m (t,f) is the attention weight of the mth convolutional feature, is the mth convolution feature, λ n (t,f) is the attention weight of the nth pooling feature, is the nth pooling feature.

[0197] In the detection of partial discharge at cable joints, the introduction of a hierarchical attention mechanism can effectively improve the weight of key features and suppress irrelevant noise. By calculating the attention weight, it ensures that key features receive higher attention at different times and frequencies, thereby improving the accuracy and robustness of feature extraction.

[0198] The process of step S3 of the present invention is as follows Figure 3 As shown in the figure, by designing a multi-scale adaptive convolution module, a dynamic time-frequency fusion module, and a hierarchical attention mechanism, the characteristics of the cable joint partial discharge signal are accurately captured and fused, ensuring high-precision partial discharge signal recognition in complex environments. According to the fiber length information, the cable joint where the partial discharge fault occurs is located, and the fault location and partial discharge characteristic information are uploaded to the monitoring platform through the MQTT protocol for real-time monitoring and fault diagnosis.

[0199] Figure 4 The improved convolutional neural network model training iterative process provided in Example 1 of the present invention is as follows: Figure 4 As shown, in order to verify the accuracy of voiceprint recognition, the trained CNN model was used to identify the 20 partial discharge voiceprint signals 500 times each, and the voiceprint recognition accuracy of each segment was recorded to be over 95%.

[0200] Example 2

[0201] Figure 2 A cable joint partial discharge detection system based on optical fiber voiceprint sensing technology is provided for embodiment 2 of the present invention, comprising:

[0202] An information acquisition module, wherein the information acquisition module is used to acquire a partial discharge acoustic wave signal at each cable joint using a phase-sensitive optical time domain reflectometry system;

[0203] A feature extraction module, wherein the feature extraction module is used to extract features of the partial discharge sound wave signal by using a Mel logarithmic spectrum analysis method to obtain a partial discharge soundprint feature;

[0204] A classification and recognition module, wherein the classification and recognition module is used to construct an improved convolutional neural network model, and utilize the improved convolutional neural network model to classify and recognize the partial discharge voiceprint features;

[0205] A fault location module is used to locate the cable joint of the partial discharge fault according to the classification and identification results.

[0206] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A cable joint partial discharge detection method based on optical fiber voiceprint sensing technology, characterized in that: The steps include: Step S1, using a phase-sensitive optical time domain reflectometry system to obtain a partial discharge acoustic wave signal at each cable joint; Step S2, using the Mel logarithmic spectrum analysis method to extract features of the partial discharge sound wave signal to obtain partial discharge soundprint features; Step S3, constructing an improved convolutional neural network model, and using the improved convolutional neural network model to classify and identify the partial discharge voiceprint features; Step S4: locating the cable joint with the partial discharge fault according to the classification and identification result.

2. The cable joint partial discharge detection method based on optical fiber voiceprint sensing technology according to claim 1 is characterized in that: The step S1 comprises: Injecting narrowband pulse light generated by a pulse light source into a detection optical fiber, the pulse light generates Rayleigh scattering when propagating in the detection optical fiber, and obtaining a scattered light signal; Calculating a phase change signal of the scattered light signal using a phase-sensitive optical time domain reflectometry system based on coherent detection; The phase change signal is demodulated to convert the phase change signal into a partial discharge acoustic wave signal in the time domain.

3. The cable joint partial discharge detection method based on optical fiber voiceprint sensing technology according to claim 2 is characterized in that: The step S12 comprises the following steps: Step S121, introducing the scattered light signal into a light detector, and the light detector converts the scattered light signal into an electrical signal; Step S122, using a preamplifier to sequentially amplify and filter the electrical signal to obtain a processed electrical signal; Step S123, using a lock-in amplifier to demodulate the processed electrical signal to obtain a demodulated signal, and extracting phase information and amplitude information from the processed electrical signal; Step S123, sampling the demodulated signal to obtain a digital signal, and extracting the frequency characteristics of the digital signal by using fast Fourier transform; Step S124: Calculate the frequency signal and the phase signal to obtain a partial discharge acoustic wave signal in the time domain.

4. The cable joint partial discharge detection method based on optical fiber voiceprint sensing technology according to claim 1 is characterized in that: The step S2 comprises: Step S21, using an improved Mel logarithmic spectrum analysis method to convert the local discharge acoustic wave signal in the time domain into a frequency domain signal, and extracting the frequency domain features of the frequency domain signal; Step S22, removing the noise features in the frequency domain features to obtain key features; wherein the key features include the frequency, amplitude and phase of the partial discharge voiceprint signal; Step S23: normalize the key features, and select partial discharge voiceprint features from the normalized key features.

5. The cable joint partial discharge detection method based on optical fiber voiceprint sensing technology according to claim 4 is characterized in that: The step S23 comprises the following steps: Step S231: Use the information gain algorithm to evaluate the importance of the normalized key features to obtain feature information gain; Step S232: selecting important features from the normalized key features according to the feature information gain; Step S233, weighting the important features, and expanding the weighted important features using sliding window technology and data enhancement methods to obtain expanded features; Step S234: perform feature fusion on the extended features to obtain partial release voiceprint features.

6. The cable joint partial discharge detection method based on optical fiber voiceprint sensing technology according to claim 1 is characterized in that: The improved convolutional neural network model in step S3 includes: a multi-scale adaptive convolution module, a dynamic time-frequency fusion module and a hierarchical attention mechanism; The multi-scale adaptive convolution module includes multiple parallel convolution kernels, and the multiple parallel convolution kernels extract time-frequency domain joint features, frequency domain features and time domain features from the local discharge voiceprint features; The dynamic time-frequency fusion module fuses the time-frequency domain joint feature, the frequency domain feature and the time domain feature through an adaptive filter to obtain a fusion feature; The hierarchical attention mechanism performs attention training on the fused features to obtain trained features, and the trained features are used to determine the type and position of the partial discharge signal.

7. The cable joint partial discharge detection method based on optical fiber voiceprint sensing technology according to claim 6 is characterized in that: The size and step length of the convolution kernel are dynamically adjusted according to the frequency characteristics of the partial discharge voiceprint feature, wherein the dynamic adjustment process is: in, is the weight of the jth convolution kernel of the i-th layer at the m-th time-frequency point, α is the adjustment amplitude, β is the frequency adjustment factor, is the phase shift, f adjust (f input (t,f)) is the adjustment function of the input signal frequency characteristics, f input (t,f) is the time-frequency domain representation of the input signal. t represents the time variable of the input signal, which is the change of the signal in time. f represents the frequency variable of the input signal, which is the change of the signal in frequency.

8. The cable joint partial discharge detection method based on optical fiber voiceprint sensing technology according to claim 6 is characterized in that: The adaptive filter dynamically adjusts filter parameters according to the partial discharge voiceprint characteristics, wherein the adjustment process of dynamically adjusting the filter parameters is as follows: H filter (t,f)=H0(t,f)·(1+δ·cos(2πηt+θ)·g adjust (f)); Among them, H filter (t, f) is the transfer function of the adaptive filter, H0(t, f)· is the initial transfer function, δ is the filter adjustment amplitude, η is the time adjustment factor, θ is the phase adjustment, g adjust (f) is the adjustment function of the frequency characteristics.

9. The cable joint partial discharge detection method based on optical fiber voiceprint sensing technology according to claim 6 is characterized in that: The hierarchical attention mechanism calculates the convolutional layer attention weight and the pooling layer attention weight respectively according to the weight calculation value of the fusion feature, wherein the calculation process is as follows: Among them, γ m (t,f) is the attention weight of the mth convolutional feature, e m (t,f) is the weight calculation value of the mth feature, λ n (t,f) is the attention weight of the nth pooling feature, e n (t,f) is the calculated weight value of the nth feature.

10. The cable joint partial discharge detection system based on optical fiber voiceprint sensing technology is characterized by: include: An information acquisition module, wherein the information acquisition module is used to acquire a partial discharge acoustic wave signal at each cable joint using a phase-sensitive optical time domain reflectometry system; A feature extraction module, wherein the feature extraction module is used to extract features of the partial discharge sound wave signal by using a Mel logarithmic spectrum analysis method to obtain a partial discharge soundprint feature; A classification and recognition module, wherein the classification and recognition module is used to construct an improved convolutional neural network model, and utilize the improved convolutional neural network model to classify and recognize the partial discharge voiceprint features; A fault location module is used to locate the cable joint with partial discharge fault according to the result of classification and identification.

Citation Information

Cited By

  • Cable partial discharge signal multipath interference identification and compensation method, equipment and medium

    CN119881562A

  • Deepwater umbilical cable temperature monitoring data processing method and system

    CN120141681A

  • A method and system for processing deepwater umbilical cable temperature monitoring data

    CN120141681B