Acousto-optic combined medium-voltage switch cabinet partial discharge online monitoring method and system

Abstract

According to the acousto-optic combined medium-voltage switch cabinet partial discharge online monitoring method and system, a switch cabinet is provided with a multispectral sensor, an ultrasonic sensor and a central control machine, the multispectral sensor and the ultrasonic sensor are in data connection with the central control machine, and the multispectral sensor is obtained by covering the surface of a multi-channel photoelectric sensor with a band-pass optical filter; the ultrasonic sensor is arranged on the outer side of the switch cabinet, the central control machine comprehensively processes and analyzes signals collected by the sensor, and the fault type and strength are analyzed through clustering analysis of acousto-optic combined signals, so that early warning of detailed discharge conditions is carried out. The advantages of an optical method and an acoustic method are fused, the optical-acoustic combined detection method suitable for remote online monitoring of partial discharge of the medium-voltage switch cabinet is provided, the richness of detection signals is improved, the signal randomness and interference influence are reduced, the partial discharge detection range is enlarged, the specific discharge mode is recognized, and the detection accuracy is improved. Accurate detection of partial discharge in the switch cabinet is realized.

Description

Technical Field

[0001] The invention belongs to the technical field of electric power, relates to a device for detecting arc and discharge, and is an acoustic-optical combined remote online monitoring device for partial discharge of a medium-voltage switch cabinet. Background Art

[0002] Switchgear is one of the important equipment in the power system. Its main function is to achieve control and protection in the process of power generation, transmission, distribution and power conversion. Insulation failure of switchgear is mostly caused by abnormal discharge, and partial discharge or intermittent creepage caused by the insulator interface is an important cause of insulation flashover inside the switchgear. Early monitoring of partial discharge, corona discharge and surface discharge can detect insulation defects, timely repair and prevent defects from further developing into faults.

[0003] Discharge is accompanied by many physical phenomena such as electricity, magnetism, light, and sound. In the past, researchers at home and abroad have proposed monitoring methods such as ultra-high frequency method, pulse current method, geomagnetic wave method, space radio frequency method, and ultrasonic method. For example, Chinese application CN118275833A uses ultrasonic method to monitor partial discharge, and CN118294795A uses pulse current method for monitoring. However, due to the serious electromagnetic noise and sound wave interference on site, the electromagnetic radiation spectrum and sound wave intensity of the three types of discharge, partial discharge, corona discharge and surface discharge, are significantly different in medium-voltage switchgear. It is difficult to solve the problem of partial discharge monitoring of medium-voltage switchgear in one way, and it is impossible to achieve high reliability and low false alarm rate medium-voltage switchgear discharge monitoring.

[0004] The spectrum of partial discharge has rich physical information. The switch cabinet is in a dark environment for a long time, which makes the optical signal have the advantages of strong specificity and low interference. The optical detection method can achieve detection with much higher sensitivity than the ultrasonic method. Optical detection has the advantages of strong anti-interference and high detection confidence. It has attracted much attention in the field of abnormal discharge detection of power equipment, spectral information fusion diagnosis, and multi-spectral band joint diagnosis. Usually, discharge is not only accompanied by rapid charge proliferation and transfer processes, but also accompanied by light radiation. The optical image characteristics carried by the light radiation have good intrinsic properties in discharge diagnosis, which can reflect the discharge development mode and the level of discharge energy. In addition, the optical image coupling is not affected by electromagnetic and acoustic interference in space and has a high confidence. However, light propagates in a straight line. Using only optical detection methods in a switch cabinet with a complex structure will face the difficulty of not being able to guarantee the detection of optical signals of partial discharge in corners.

[0005] The prior art also proposes a scheme for jointly detecting partial discharge in multiple ways, such as CN116231838A "A partial discharge early warning system and method based on acoustic, optical and gas sensing technology", which uses the monitoring data of optical signals, ultrasonic signals and gas signals in the partial discharge process, and considers the correlation between different types of sensor signals in the early warning analysis process of partial discharge to warn whether partial discharge occurs. However, this scheme cannot determine the type and intensity of the discharge fault, and in the early warning of whether to send a partial discharge, different units of sensors based on different principles make independent judgments on whether there is a partial discharge based on different models, and finally make a fusion decision on the binary result of the judgment: the presence / absence of partial discharge, and make a judgment on whether there is a partial discharge. In fact, for discharge modes with different intensities and discharge events with different locations, sensors based on different principles cannot detect them at the same time. For example, the optical method can detect weak discharges that the acoustic method cannot detect, and the acoustic method can monitor discharges in the blind area of ​​the optical method. This may make the decision result of the early warning provided by the CN116231838A scheme inaccurate. Summary of the invention

[0006] The problem to be solved by the present invention is that the main technical routes of the current medium voltage switchgear partial discharge monitoring solutions are pulse current method, ultra-high frequency method, space radio frequency method, geomagnetic wave method, ultrasonic method and optical detection method, and each of these methods has disadvantages:

[0007] (1) The solution based on the pulse current method requires the switch cabinet to be disconnected from the operating environment for detection. It does not have the conditions for online and long-term monitoring, and it is impossible to locate the fault location.

[0008] (2) Solutions based on the geostationary wave method and the space radio frequency method can only locate the approximate location of the discharge, but cannot accurately locate the specific location of the discharge, nor can they quantitatively indicate the severity of the discharge. In addition, the monitored signal frequency is similar to that of the communication base station, and is extremely sensitive to interference from the communication signal.

[0009] (3) The ultrasonic method has a small effective detection range for discharge and is extremely sensitive to mechanical vibration. Due to this limitation, the ultrasonic method can only detect local discharge faults with higher intensity, while local discharge faults with lower intensity will be drowned out by the noise of equipment operation and environmental noise.

[0010] (4) When conducting partial discharge detection inside the switch cabinet, the optical detection method is subject to the complex structure of the switch cabinet and the linear propagation of light, and it is difficult to ensure that all possible partial discharge faults in the switch cabinet are fully covered. For example, a discharge fault occurring in an air gap defect inside the insulation cannot be detected by optical detection technology because it is completely enclosed.

[0011] The technical solution of the present invention is: an acoustic-optical combined medium-voltage switch cabinet partial discharge online monitoring method, the switch cabinet is equipped with a multi-spectral sensor, an ultrasonic sensor and a central control machine, the multi-spectral sensor, the ultrasonic sensor and the central control machine are data connected, the multi-spectral sensor is obtained by covering the surface of a multi-channel photoelectric sensor with a bandpass filter, the ultrasonic sensor is arranged outside the switch cabinet, and the central control machine processes the signal collected by the sensor as follows:

[0012] Preprocess the optical signal to obtain the multi-channel signal of the multi-spectral sensor I i , determine the weakest attenuation channel according to the light intensity of each channel, and set the output of the weakest attenuation channel to be I t , based on I t Normalize the signals of other channels;

[0013] Preprocess the ultrasonic signal, set the noise threshold according to the ultrasonic signal when the switch cabinet is in normal operation, perform Fourier transform on the real-time signal, extract its 50Hz component, and then perform inverse Fourier transform to obtain the waveform of the fundamental frequency vibration, calculate the maximum value, root mean square, 50Hz, and 100Hz component amplitude of the signal in each cycle, and subtract the noise threshold, record the points where the reading is greater than 0, calculate the mean output once every N cycles, and obtain the maximum mean, root mean square mean, 50Hz component mean, and 100Hz component mean of the denoised signal;

[0014] Collect data from multispectral sensors and ultrasonic sensors under different partial discharge modes of the switch cabinet, and preprocess to obtain the multi-channel normalized light intensity and I of the optical signal. t , and the mean output of the ultrasonic signal, train the clustering model based on Gaussian mean clustering, and obtain the clustering parameters for acoustic and optical joint detection in different modes for the switch cabinet;

[0015] For the switch cabinet signal actually measured, the corresponding optical signal and ultrasonic signal are preprocessed and input into the clustering model. According to the clustering result, it is confirmed whether the current switch cabinet signal is fault-free or in a certain fault mode, and the discharge intensity is obtained according to the light intensity information.

[0016] Furthermore, the preprocessing of the optical signal is as follows: let the number of the weakest attenuation channel be t, and calculate the signal ratio between the two channels of the multispectral sensor And the ratio of the light intensity integral between the two channels Get the conversion coefficient between two channels For all channels, β(i,t)=β(i,k)×β(k,t), where k is the channel number other than number t. There is a value of t such that all β(i,t)≤1. Then the channel corresponding to the value of t is the weakest attenuation channel. The ratio of the light intensity of other channels to the weakest attenuation channel is As a normalized result, the absolute signal intensity I of channel t is t Together they serve as the preprocessing results of the optical signal.

[0017] Furthermore, the clustering model uses Gaussian mean clustering, specifically:

[0018] Assume that the cluster is a Gaussian mixture distribution consisting of K mixture components, each with a mean of μ k , the covariance matrix is ​​Σ k The m-dimensional Gaussian distribution accounts for α in the mixed distribution k , then each sample point has a certain probability α k Belonging to any cluster, the sample point S n is a collection of the preprocessing results of optical signals and ultrasonic signals. The Gaussian clustering model is as follows:

[0019] The prior probability of generating a single sample point is:

[0020]

[0021] in

[0022]

[0023] p(S n |μ k ,∑ k ) is the sample point S n By Gaussian distribution (μ k ,∑ k ) is generated;

[0024] The goal of clustering is to find a set of parameters (α k ,μ k ,Σ k ), maximize the total probability of the samples used, and use the maximum likelihood estimate to maximize the following likelihood based on the prior probability expression above:

[0025]

[0026] Let the random variable z n ∈{1,2,…K}, indicating the generation of sample S n The Gaussian mixture component number, γ nk Indicates S n By the z n The posterior probability generated by Gaussian components is:

[0027]

[0028] Find α k ,μ k ,Σk The maximum likelihood estimate of :

[0029]

[0030] Collect sample point data S of different modes n , train according to the following steps to determine the model clustering parameter μ for the joint detection of different modes of sound and light in the air switch cabinet k ,∑ k :

[0031] 1) Input the number of clustering patterns K, where K corresponds to the number of patterns to be detected in the switch cabinet, and initialize the parameters of the mixed Gaussian distribution {(α k ,μ k ,Σ k )|k∈(1,…,K)};

[0032] 2) Calculate the prior probability p generated by each sample point according to formula (1) (2) M (S n );

[0033] 3) Update the parameters (α) of the Gaussian distribution according to formula (5) k ,μ k ,Σ k );

[0034] 4) Repeat steps 2) and 3) until (α k ,μ k ,Σ k )convergence;

[0035] 5) The final γ nk , that is, for each sample point S n Clustering by Gaussian distribution (α k ,μ k ,Σ k ) generates the posterior probability, assuming that γ nk The largest k is the clustering mode number to which the sample point actually belongs;

[0036] For the actual measured switch cabinet signal, obtain the real-time sample point S n , clustering calculation belongs to mode k, the mode with the largest probability is the mode corresponding to the current sound and light signal.

[0037] Furthermore, the process of online monitoring of partial discharge of medium voltage switchgear is as follows:

[0038] The first step is that a partial discharge fault occurs, and the discharge process generates a light signal or an acoustic signal. The strength of the discharge process is characterized by the discharge amount and the discharge current. The stronger the discharge, the higher the discharge amount, the greater the current, and the stronger the light emission, the more it will cause the equipment to vibrate and generate ultrasonic waves.

[0039] In the second step, the optical and acoustic signals generated during the discharge process are collected by multi-spectral sensors and ultrasonic sensors, and the collected signals are transmitted to the central control computer for fault analysis.

[0040] The third step is to collect, preprocess and analyze the discharge signal data. The discharge process in the switch cabinet is manifested in different discharge modes. The central control computer integrates the light intensity information collected by the multi-spectral sensor and the waveform of the ultrasonic vibration to analyze the fault type and intensity.

[0041] The fourth step is post-processing of discharge faults. For the discharge faults found, the central control computer records them in real time, outputs the fault records, submits them to the host computer and issues an alarm to remind the user to handle them in time.

[0042] The present invention also proposes an acoustic-optical combined medium-voltage switchgear partial discharge online monitoring system, comprising a multi-spectral sensor, an ultrasonic sensor and a central control machine, wherein the central control machine comprises a spectral signal acquisition signal preprocessing module, an ultrasonic signal acquisition and signal preprocessing module, a signal comprehensive analysis central control system and a signal feature statistical database module.

[0043] The multispectral sensor is arranged in the switch cabinet, and its position and orientation try to cover all corners of the switch cabinet, and the multispectral sensor is connected to the central control computer data;

[0044] The ultrasonic sensor is arranged inside or on the surface of the cabinet and connected to the central control machine to collect the ultrasonic signal emitted by partial discharge. The ultrasonic sensor response covers the frequency band of 35 to 45 kHz.

[0045] The spectral signal acquisition signal preprocessing module and the ultrasonic signal acquisition and signal preprocessing module of the central control machine implement the above-mentioned optical signal preprocessing and ultrasonic signal preprocessing respectively. The signal comprehensive analysis central control system implements the above-mentioned clustering model. The signal feature statistical database module is used to store the data processed and output by the central control machine.

[0046] Furthermore, the multi-spectral sensor is obtained by covering the surface of a multi-channel photoelectric sensor with a bandpass filter. The bandpass filter is selected according to two different standards. Standard 1 uses a narrowband filter with a central wavelength being the wavelength of a spectral line of luminescence from an insulating gas discharge in a switch cabinet to obtain the light intensity of a local discharge at a certain characteristic wavelength; Standard 2 uses a bandpass filter to obtain the average light intensity of a local discharge in a certain band, and is selected according to the spectral characteristics of each abnormal discharge mode of concern in the actual gas environment of the switch cabinet.

[0047] The present invention has the following beneficial effects:

[0048] In response to the problem that traditional partial discharge detection methods represented by ground wave and ultra-high frequency methods are easily interfered by radio frequency electromagnetic signals from mobile communication base stations, the optical sensor used in this patent is sensitive to the visible light band and is installed in the closed metal cavity of the switch cabinet and will not be affected by external sunlight; therefore, the detection scheme adopted by this patent has strong resistance to external interference.

[0049] In order to solve the problem that traditional partial discharge detection methods represented by the pulse current method require the equipment to be offline and cannot detect equipment failure in time, the optical sensor used in this patent is arranged in the switch cabinet and can work for a long time. This solution can realize long-term monitoring of partial discharge failures in the switch cabinet.

[0050] In order to address the problem that traditional partial discharge detection methods are unable to determine the fault location and fault discharge type, the optical sensor used in this patent can restore the partial discharge fault type and fault intensity and estimate the location where the discharge occurs through the layout position, signal delay, and collected multi-spectral information.

[0051] In view of the problem that optical detection methods may have detection blind spots, the solution proposed in this patent combines ultrasonic method with synchronous detection of partial discharge faults to avoid the problem that discharge occurs in the blind spot of the optical sensor represented by the inside of the insulation, which makes the optical method ineffective.

[0052] The present invention adopts an optical-acoustic joint detection method, comprehensively utilizing the advantages of high detection sensitivity of optical sensors and small restrictions of ultrasonic sensors on the internal structure of the switch cabinet, configuring a central control machine, and integrating the signals collected by the two types of sensors into the central control program for comprehensive analysis, thereby improving the richness of detection signals, reducing signal randomness and interference effects, increasing the local discharge detection range, and improving the ability to analyze the severity of local discharge faults. Through comprehensive sound and light analysis, early warning of local discharge of the switch cabinet and identification of discharge mode and discharge intensity are achieved, thereby realizing accurate detection of local discharge in the switch cabinet. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 This is a typical structural diagram of the multi-channel detection unit of a multi-spectral sensor.

[0054] Figure 2 It is a structural schematic diagram of the partial discharge detection device of the present invention.

[0055] Figure 3 It is a schematic diagram of the flow of the local room detection method of the present invention. DETAILED DESCRIPTION

[0056] The present invention proposes an acoustic-optic combined medium-voltage switchgear partial discharge online monitoring method and system, wherein a multispectral sensor, an ultrasonic sensor and a central control machine are configured for the switchgear, and the multispectral sensor, the ultrasonic sensor and the central control machine are data-connected, and the central control machine performs comprehensive processing and analysis on the signals collected by the sensors, analyzes the fault type and intensity through cluster analysis of the acoustic-optic combined signals, and performs early warning of detailed discharge conditions.

[0057] The optical detection method used in the present invention is based on a type of optical sensor, the basic principle of which is to cover the surface of a multi-channel photoelectric sensor, such as a silicon photomultiplier (SiPM), with a bandpass filter. The characteristics of the multi-channel sensor are that the light intensity of different channels can be output to the data processing module separately, and the channels are arranged on the same plane, and the detection efficiency of electromagnetic waves in the visible light and ultraviolet bands is good.

[0058] Furthermore, the bandpass filter is selected according to two different standards. Standard 1 uses a narrowband filter with a central wavelength of the common spectral line wavelength of the insulating gas discharge light in the switch cabinet. The insulating gas is generally air. The light intensity of the partial discharge at a certain characteristic wavelength is obtained by the method of standard 1; standard 2 uses a bandpass filter to obtain the average light intensity of the partial discharge in a certain band. The actual scheme adopted is determined according to the spectral characteristics of each abnormal discharge mode of concern in the application gas environment.

[0059] For ultrasonic detection, a series of ultrasonic sensors are arranged on the outside of the switch cabinet equipment at the same time. The arranged ultrasonic sensors can be any sensors with the function of collecting ultrasonic vibration signals generated by partial discharge, and the typical frequency response range covers 35 to 45 kHz.

[0060] like Figure 2 As shown, the arranged multi-spectral sensors and ultrasonic sensors are connected to the central control machine, and the central control machine is equipped with spectral signal and ultrasonic signal preprocessing, data processing and fault identification modules, including spectral signal acquisition signal preprocessing module, ultrasonic signal acquisition and signal preprocessing module, signal comprehensive analysis central control system and signal feature statistical database module, which can complete the work of denoising and normalizing the signal to extract the real signal, as well as the analysis process of identifying the fault mode and fault intensity through the real discharge signal.

[0061] In view of the problems that the traditional detection method of partial discharge has poor ability to locate discharge faults, poor ability to quantify intensity, and is affected by electromagnetic or vibration interference in the daily production and living environment, the present invention uses an optical detection method to solve the problem. For extremely low-intensity partial discharges, the emitted visible light and ultraviolet light are incident on each channel of the above-mentioned photoelectric sensor through a bandpass filter, and the light intensity of each typical band and spectrum line is obtained. The light intensity information received by each channel of the multi-spectral sensor is converted into a multi-spectral digital signal, which is uploaded to the signal analysis and processing system of the central control machine after signal preprocessing. The signal analysis and processing system integrates the signals returned by each sensor. If the discharge is detected by one sensor, the type, intensity, and distance from the sensor of the discharge fault are inferred based on the pre-collected database and the trained classification model; if the discharge is detected by multiple sensors at the same time, the location of the discharge fault is inferred based on the time difference of the collected signals, the direction of the sensor layout, and the internal structure of the switch cabinet. Figure 1 The typical structure of the detection unit of the multi-channel sensor used for optical detection, 64 units each have the ability to independently output the detected light intensity information. The input units with the same letter label cover the same filter, and the multi-spectral sensor shown in the embodiment of the present invention has the ability to detect light intensity of 8 wavelengths / bands.

[0062] The optical method is limited by the complex structure of the switch cabinet and the rectilinear propagation of light, and it is difficult to ensure complete coverage of partial discharge faults that may occur in all locations in the switch cabinet. The present invention combines an ultrasonic sensor for comprehensive analysis. The ultrasonic sensor can collect corresponding signals and make judgments on discharge faults. According to the location of the ultrasonic sensor and the time delay of the collected signal, when the optical method cannot diagnose the partial discharge fault, the fault can also be detected, and the location and intensity of the fault can be roughly estimated, so as to remind the system user to conduct targeted inspections on specific locations.

[0063] The signal analysis and processing by the central control machine of the present invention is described in detail below.

[0064] The central control machine receives the digital signal of the spectral sensor, receives the voltage signal of the acoustic sensor and converts it into a digital signal, which is input into the data preprocessing module. The output of the multi-spectral sensor is the sum of the output intensities of each channel covering the same filter, and the sampling rate of the analog-to-digital converter used by the ultrasonic sensor should be greater than 50kHz.

[0065] For spectral data, the central control machine is pre-loaded with the wavelength, band, and attenuation rate information of the filters used for each input channel, and the digital signal of the spectral sensor is normalized. The attenuation rate of the filter used by the multi-spectral sensor is calibrated by using the multi-spectral sensor and the spectrometer to measure discharges of known intensity.

[0066] Assume that the number of the weakest attenuation channel is t, and calculate the signal ratio between the two channels of the multispectral sensor And the ratio of the light intensity integral between the two channels Get the conversion coefficient between two channels If the corresponding filter detects the band average light intensity, the spectrometer calculates the light intensity integral corresponding to the wavelength range passed by the filter. I0(λ) is the spectrum captured by the spectrometer. If the corresponding filter detects the intensity of the spectral line, the three-point method is used to interpolate and calculate the intensity I of the characteristic spectral line. 0i =f[I(λ i ),I(λ i-1 ),I(λ i+1 ),λ].

[0067] During calibration, under various discharge modes such as corona, spark, surface flashover, arc discharge, etc., record at least 2 multi-spectral sensor channel output signals with obvious readings and the spectrum I0(λ) taken by the spectrometer. The multi-spectral sensor channels are numbered 1-8 respectively. When calibrating under four discharge modes, at least one of the 8 channels has an obvious reading. Otherwise, it means that the selection of the spectral channel is not typical and the spectral channel should be reselected.

[0068] According to the definition of the conversion coefficient β, it is easy to verify that β(i, t) = β(i, k) × β(k, t), and there is a value of t such that all β(i, t) ≤ 1. This value of t is called the "weakest attenuation channel", and the set of all β(i, k) at this time is the 8 conversion coefficients.

[0069] Finally, the multispectral sensor inputs the light intensity of 8 numbered channels and outputs the ratio of the light intensity of 7 channels to the weakest attenuation channel. The absolute signal strength I of channel t t , for access t, γ t,output Always equal to 1, no output.

[0070] For the signal collected by the acoustic unit, first measure it on a normal device, record the maximum value and root mean square of the signal every 30 seconds within 1 hour, and set 150% of the amplitude mean as the noise threshold. At the same time, use Fourier transform to extract the amplitude of its 50Hz and 100Hz components, and set 150% of the amplitude mean as the noise threshold.

[0071] In partial discharge monitoring, the input is an actual signal. First, perform Fourier transform on the signal, extract its 50Hz component, and then perform inverse Fourier transform to obtain the waveform of the fundamental frequency vibration. In the same way, calculate the maximum value, root mean square, 50Hz, and 100Hz component amplitude of the signal in each cycle, subtract the noise threshold, and record the points where the reading is greater than 0. For each point where the signal is the maximum value and greater than 0 in each cycle, record its phase relative to the fundamental frequency vibration. Output is performed every 1500 cycles, and the output is the maximum value, root mean square, 50Hz amplitude, the percentage of the number of cycles with 100Hz amplitude greater than 0 after all denoising, and the amplitude of each signal in each cycle.

[0072] The 7 normalized light intensities, I t There are eight optical signal data in total, and four sets of data including the maximum mean value, root mean square mean value, 50Hz component mean value, and 100Hz component mean value of the signal within 1500 cycles output by the acoustic module, a total of 12 sets of data, which are used as sample points for cluster analysis of the clustering model.

[0073] Clustering uses Gaussian mean clustering, the principle is as follows:

[0074] Assume that the entire cluster is a Gaussian mixture distribution: it consists of K mixture components, each with a mean of μ k , the covariance matrix is ​​∑ k The m-dimensional Gaussian distribution accounts for α in the mixed distribution k In this clustering method, there is no longer a strict clustering boundary, and each point has a certain probability α k Belong to any cluster: Each sample point is first classified by α k The probability of randomly determining which Gaussian distribution is used to generate it, and then generating it according to the corresponding Gaussian distribution.

[0075] According to the above clustering model, the prior probability of a single sample point is:

[0076]

[0077] in

[0078]

[0079] p(S n |μ k ,∑ k ) is the sample point S n By Gaussian distribution (μ k ,∑ k ) is generated with probability.

[0080] The goal of clustering is to find a set of parameters (α k ,μ k ,∑k ), so that the total probability of the samples used is maximized. To this end, the present invention uses maximum likelihood estimation, based on the above prior probability expression, to maximize the following likelihood:

[0081]

[0082] N represents the total number of sample points.

[0083] Let the random variable z n ∈{1,2,…K}, indicating the generation of sample S n The Gaussian mixture component number. γ nk Indicates S n By the z n The posterior probability generated by Gaussian components is:

[0084]

[0085] Find ɑ k ,μ k ,∑ k The maximum likelihood estimate of :

[0086]

[0087] The clustering algorithm based on Gaussian mixture model can be summarized as an iterative method, using the newly calculated ɑ k ,μ k ,∑ k Update the prior probability p M , and then recalculate the new α k ,μ k ,∑ k The algorithm is as follows:

[0088] 1) Input the number of clustering patterns K and initialize the parameters of the mixed Gaussian distribution {(α k ,μ k ,∑ k )|k∈(1,…,K)}. For example, take α k =1 / K,μ k Determined by K-means algorithm, ∑ k Take any diagonal matrix.

[0089] 2) Calculate the prior probability p of each sample point according to formula 4-6 M (S n ).

[0090] 3) Update the parameters of the Gaussian distribution (α) according to formula 4-9 k ,μ k ,∑ k ).

[0091] 4) Repeat steps 2 and 3 until (α k ,μ k ,∑ k ) is less than a certain value.

[0092] 5) γ output by the algorithm nk , that is, for each sample point S n Clustering by Gaussian distribution (α k ,μ k ,∑ k ) generated by the posterior probability. It can be considered that γ nk The largest k is the cluster number to which the sample point actually belongs.

[0093] In the embodiment, the number of clustering patterns K=6, the dimension m=12, corresponding to no fault, and five partial discharge fault modes of corona, surface, spark, floating potential, and arc, and the 12-dimensional data type of the sample point. For the artificial partial discharge in the switch cabinet and the vibration signal of the fault-free equipment, the 12-dimensional data S output of each discharge mode is collected. n , calculate according to steps 1-5 to determine the model clustering parameter μ for the four-mode acoustic and optical joint detection of air switch cabinets k ,∑ k .

[0094] For the actual measured partial discharge signal, input S n , calculate the probability that the partial discharge belongs to mode k at this time, and the mode with the highest probability is the mode corresponding to the current sound and light signal.

[0095] like Figure 3 As shown in the figure, when a partial discharge fault occurs, the processing procedure of using the partial discharge monitoring system to automatically diagnose the partial discharge fault is as follows:

[0096] In the first step, a partial discharge fault occurs, and the discharge process generates light or sound signals. The strength of the discharge process can be characterized by the discharge amount and discharge current: the stronger the discharge, the higher the discharge amount, the greater the current, and the stronger the light emission. Stronger discharges may also cause vibrations in equipment components and generate ultrasonic waves.

[0097] In the second step, the optical and acoustic signals generated by the discharge process are collected by the corresponding sensors. Even if the discharge is weak, an optical signal will be generated. The optical signal will enter one or more multispectral sensors facing the discharge fault, and pass through a broadband filter covering a certain band or a narrowband filter corresponding to a certain emission spectrum line to obtain the average light intensity of a corresponding band or the light intensity signal of a certain emission spectrum line, which will be converted and output into a light intensity digital signal by the spectrum sensor; for some discharges that are not in the coverage area of ​​the multispectral sensor, their acoustic signals can also be collected by the ultrasonic sensor and output as digital signals. The collected signals are transmitted to the central control machine through the signal line for fault analysis.

[0098] The third step is to collect, preprocess and analyze the discharge signal data. The discharge process in the switch cabinet is manifested in different discharge modes, such as corona discharge, suspended potential discharge, spark discharge, arc discharge, etc. The spectra and ultrasonic signals of different modes are significantly different. The central control computer integrates the light intensity information collected by different sensors and different pixels, and the waveform of ultrasonic vibration, and analyzes the fault type and intensity according to the above steps.

[0099] The fourth step is post-processing of discharge faults. For the discharge faults found, the central control computer records them in real time and outputs the fault records; for more serious faults with higher processing priority, they are submitted to the host computer and an alarm is issued to remind the user to handle them in time.

[0100] The acoustic-optic combined medium-voltage switchgear partial discharge online monitoring system proposed in the present invention comprises a multi-spectral sensor, an ultrasonic sensor and a central control machine, wherein the central control machine comprises a spectral signal acquisition signal preprocessing module, an ultrasonic signal acquisition and signal preprocessing module, a signal comprehensive analysis central control system and a signal feature statistical database module.

[0101] The multispectral sensor is arranged in the switch cabinet, and its position and orientation try to cover all corners of the switch cabinet, and the multispectral sensor is connected to the central control computer data;

[0102] The ultrasonic sensor is arranged inside or on the surface of the cabinet and connected to the central control machine to collect the ultrasonic signal emitted by partial discharge. The ultrasonic sensor response covers the frequency band of 35 to 45 kHz.

[0103] The spectral signal acquisition signal preprocessing module and the ultrasonic signal acquisition and signal preprocessing module of the central control machine implement the above-mentioned optical signal preprocessing and ultrasonic signal preprocessing respectively. The signal comprehensive analysis central control system implements the above-mentioned clustering model. The signal feature statistical database module is used to store the data processed and output by the central control machine.

[0104] As an embodiment, the system of the present invention is implemented as an air-insulated 110 kV switchgear device and is monitored.

[0105] The first step is to prepare an optical sensor. The optical sensor has multiple pixels and has a good response to the visible light band and the near-ultraviolet band. A typical example is the Hamamatsu S4111-16Q 16-pixel silicon photodiode array.

[0106] The second step is to convert the optical sensor into a multispectral sensor. The conversion method is to cover each pixel of the optical sensor with a filter, and the filter can pass wavelengths that include (1) a certain characteristic wavelength range or (2) a certain characteristic spectral line of the local discharge luminescence. For both standards, at least one filter is selected based on the spectral characteristics of common local discharge modes in the current gas atmosphere.

[0107] For example, for partial discharge monitoring of air insulation, there are five different modes of air corona discharge, spark discharge, surface flashover, arc discharge, and suspended potential discharge under atmospheric pressure. The characteristic spectra with obvious differences are mainly generated by the transitions of N+, O, and N2. For the N+ spectrum, select filters with a bandwidth of 5nm and a center wavelength of 500.7nm and 567.7nm; for the O spectrum, select filters with a bandwidth of 5nm and a center wavelength of 777.2nm; for the N2 spectrum, select filters with a bandwidth of 5nm and a center wavelength of 336.9nm and 357.3nm. For the first positive band of N2, select filters with a bandwidth of 100nm and a center wavelength of 850nm; for the second positive band of N2, select filters with a center wavelength of 350nm and 450nm.

[0108] The third step is to place the multispectral sensor in the switch cabinet. The position and orientation of the sensor should cover all corners of the switch cabinet as much as possible, record the position of the sensor, and connect the sensor data output to the detection system control computer.

[0109] The fourth step is to arrange the ultrasonic sensor. Any commercially available ultrasonic probe, such as resonant sensor, ultrasonic sensor with self-adsorption device, built-in preamplifier sensor, broadband sensor, etc., can meet the requirements. The ultrasonic sensor is arranged inside or on the surface of the cabinet, connected to the control machine of the detection system, and collects the ultrasonic signal emitted by the partial discharge. The typical response of the acoustic sensor needs to cover the frequency band of 35 to 45kHz, and has the function of converting the acoustic signal into a voltage signal.

[0110] The fifth step is to configure the central control machine. The central control machine is equipped with a data collection module, a data preprocessing module, and a data processing module, which are responsible for collecting data from different spectral sensors and ultrasonic sensors, analyzing the data and synthesizing the data to determine whether a partial discharge fault has occurred. The central control machine is turned on when partial discharge monitoring is required. The feature of this system is that it can be turned on continuously to continuously monitor the equipment.

Claims

1. A method for online monitoring of partial discharge of medium-voltage switchgear by combining sound and light, characterized in that The switch cabinet is equipped with a multi-spectral sensor, an ultrasonic sensor and a central control computer. The multi-spectral sensor, the ultrasonic sensor and the central control computer are connected with data. The multi-spectral sensor is obtained by covering the surface of a multi-channel photoelectric sensor with a bandpass filter. The ultrasonic sensor is arranged outside the switch cabinet. The central control computer processes the signals collected by the sensor as follows: Preprocess the optical signal to obtain the multi-channel signal of the multi-spectral sensor I i , determine the weakest attenuation channel according to the light intensity of each channel, and set the output of the weakest attenuation channel to be I t , based on I t Normalize the signals of other channels; Preprocess the ultrasonic signal, set the noise threshold according to the ultrasonic signal when the switch cabinet is in normal operation, perform Fourier transform on the real-time signal, extract its 50Hz component, and then perform inverse Fourier transform to obtain the waveform of the fundamental frequency vibration, calculate the maximum value, root mean square, 50Hz, and 100Hz component amplitude of the signal in each cycle, and subtract the noise threshold, record the points where the reading is greater than 0, calculate the mean output once every N cycles, and obtain the maximum mean, root mean square mean, 50Hz component mean, and 100Hz component mean of the denoised signal; Collect data from multispectral sensors and ultrasonic sensors under different partial discharge modes of the switch cabinet, and preprocess to obtain the multi-channel normalized light intensity and I of the optical signal. t , and the mean output of the ultrasonic signal, train the clustering model based on Gaussian mean clustering, and obtain the clustering parameters for acoustic and optical joint detection in different modes for the switch cabinet; For the switch cabinet signal actually measured, the corresponding optical signal and ultrasonic signal are preprocessed and input into the clustering model. According to the clustering result, it is confirmed whether the current switch cabinet signal is fault-free or in a certain fault mode, and the discharge intensity is obtained according to the light intensity information.

2. The method for online monitoring of partial discharge of a medium voltage switch cabinet by combining sound and light according to claim 1 is characterized in that The preprocessing of optical signals is as follows: let the number of the weakest attenuation channel be t, and calculate the signal ratio between the two channels of the multispectral sensor. And the ratio of the light intensity integral between the two channels Get the conversion factor between two channels For all channels, β(i,t)=β(i,k)×β(k,t), where k is the channel number other than number t. There is a value of t such that all β(i,t)≤1. Then the channel corresponding to the value of t is the weakest attenuation channel. The ratio of the light intensity of other channels to the weakest attenuation channel is As a normalized result, the absolute signal intensity I of channel t is t Together they serve as the preprocessing results of the optical signal.

3. The method for online monitoring of partial discharge of a medium voltage switch cabinet by combining sound and light according to claim 1 is characterized in that The clustering model uses Gaussian mean clustering, specifically: Assume that the cluster is a Gaussian mixture distribution consisting of K mixture components, each with a mean of μ k , the covariance matrix is ​​∑ k The m-dimensional Gaussian distribution accounts for α in the mixed distribution k , then each sample point has a certain probability α k Belonging to any cluster, the sample point S n is a collection of the preprocessing results of optical signals and ultrasonic signals. The Gaussian clustering model is as follows: The prior probability of generating a single sample point is: in p(S n |μ k ,∑ k ) is the sample point S n By Gaussian distribution (μ k ,∑ k ) is generated; The goal of clustering is to find a set of parameters (α k ,μ k ,Σ k ), maximize the total probability of the samples used, and use the maximum likelihood estimate to maximize the following likelihood based on the prior probability expression above: Let the random variable z n ∈{1,2,…K}, indicating the generation of sample S n The Gaussian mixture component number, γ nk Indicates S n By the z n The posterior probability generated by Gaussian components is: Find α k ,μ k ,∑ k The maximum likelihood estimate of : Collect sample point data S of different modes n , train according to the following steps to determine the model clustering parameter μ for the joint detection of different modes of sound and light in the air switch cabinet k ,∑ k : 1) Input the number of clustering patterns K, where K corresponds to the number of patterns to be detected in the switch cabinet, and initialize the parameters of the mixed Gaussian distribution {(α k ,μ k ,Σ k )|k∈(1,…,K)}; 2) Calculate the prior probability p of each sample point according to formula (1) (2) M (S n ); 3) Update the parameters (α) of the Gaussian distribution according to formula (5) k ,μ k ,Σ k ); 4) Repeat steps 2) and 3) until (α k ,μ k ,Σ k )convergence; 5) The final γ nk , that is, for each sample point S n Clustering by Gaussian distribution (α k ,μ k ,Σ k ) generates the posterior probability, assuming that γ nk The largest k is the clustering mode number to which the sample point actually belongs; For the actual measured switch cabinet signal, obtain the real-time sample point S n , clustering calculation belongs to mode k, the mode with the largest probability is the mode corresponding to the current sound and light signal.

4. The method for online monitoring of partial discharge of a medium voltage switch cabinet by combining sound and light according to claim 1 is characterized in that The different partial discharge modes of switchgear include no fault, and five partial discharge fault modes: corona, surface, spark, floating potential, and arc.

5. The method for online monitoring of partial discharge of medium voltage switch cabinet by combining sound and light according to claim 1 is characterized in that The process of online monitoring of partial discharge of medium voltage switchgear is as follows: The first step is that a partial discharge fault occurs, and the discharge process generates a light signal or an acoustic signal. The strength of the discharge process is characterized by the discharge amount and the discharge current. The stronger the discharge, the higher the discharge amount, the greater the current, and the stronger the light emission, the more it will cause the equipment to vibrate and generate ultrasonic waves. In the second step, the optical and acoustic signals generated during the discharge process are collected by multi-spectral sensors and ultrasonic sensors, and the collected signals are transmitted to the central control computer for fault analysis. The third step is to collect, preprocess and analyze the discharge signal data. The discharge process in the switch cabinet is manifested in different discharge modes. The central control computer integrates the light intensity information collected by the multi-spectral sensor and the waveform of the ultrasonic vibration to analyze the fault type and intensity. The fourth step is post-processing of discharge faults. For the discharge faults found, the central control computer records them in real time, outputs the fault records, submits them to the host computer and issues an alarm to remind the user to handle them in time.

6. An acoustic-optical combined medium voltage switchgear partial discharge online monitoring system, characterized in that It includes multi-spectral sensors, ultrasonic sensors and central control machines. The central control machine includes spectral signal acquisition and signal preprocessing modules, ultrasonic signal acquisition and signal preprocessing modules, signal comprehensive analysis central control system and signal feature statistical database modules. The multispectral sensor is arranged in the switch cabinet, and its position and orientation try to cover all corners of the switch cabinet, and the multispectral sensor is connected to the central control computer data; The ultrasonic sensor is arranged inside or on the surface of the cabinet and connected to the central control machine to collect the ultrasonic signal emitted by partial discharge. The ultrasonic sensor response covers the frequency band of 35 to 45 kHz. The spectral signal acquisition signal preprocessing module and the ultrasonic signal acquisition and signal preprocessing module of the central control machine correspondingly implement the optical signal preprocessing and ultrasonic signal preprocessing described in any one of claims 1-5, the signal comprehensive analysis central control system implements the clustering model described in any one of claims 1-5, and the signal feature statistical database module is used to store the data processed and output by the central control machine.

7. The sound-light combined medium voltage switch cabinet partial discharge online monitoring system according to claim 6 is characterized by: The multi-spectral sensor is obtained by covering the surface of a multi-channel photoelectric sensor with a bandpass filter. The bandpass filter is selected according to two different standards. Standard 1 uses a narrowband filter with a central wavelength being the wavelength of the spectral line of the insulating gas discharge luminescence in the switch cabinet to obtain the light intensity of the local discharge at a certain characteristic wavelength; Standard 2 uses a bandpass filter to obtain the average light intensity of the local discharge in a certain band, which is selected according to the spectral characteristics of each abnormal discharge mode of concern in the actual gas environment of the switch cabinet.

Images