An active arc warning method and system based on multispectral frequency bands

Through multi-spectral band analysis and nuclear limit learning machine algorithm, high confidence warning of switch cabinet arc faults is achieved, and the problem of inability to accurately warning and quickly deal with arc faults in the existing technology is solved, and the safety and diagnosis of equipment are improved. efficiency.

CN115598470BActive Publication Date: 2024-11-22WUXI POWER SUPPLY BRANCH OF STATE GRID JIANGSU ELECTRIC POWER CO LTD
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202211078546.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-09-05
Publication Date
2024-11-22
Estimated Expiration
2042-09-05

AI Technical Summary

Technical Problem

The prior art cannot accurately warn and quickly deal with arc faults in switch cabinets, resulting in equipment damage and safety hazards, and lack of quantitative means to judge the severity of discharge.

Method used

Adopting the active arc warning method based on multispectral bands, the discharge spectral characteristics are extracted through multispectral analysis technology, combined with the nuclear limit learning machine algorithm to train the arc discharge prediction model, collect multispectral band data in real time for arc stage prediction, and compare it with the fault database. Proactively warn.

Benefits of technology

It realizes the distinction between different types of local discharge and arc faults, provides high confidence warning, improves diagnostic efficiency and equipment safety, and can make early warnings before arc discharge occurs.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN115598470B_ABST
    Figure CN115598470B_ABST
Patent Text Reader

Abstract

An arc active early warning method and system based on multi-spectral bands, using the correspondence between multi-spectral bands and discharge types and discharge sequences to establish historical discharge feature information as input of a kernel extreme learning machine algorithm, training an arc discharge prediction model for high-voltage switch cabinet equipment to obtain an arc discharge prediction model; using multi-spectral bands of high-voltage switch cabinet equipment collected in real time as input, using the trained arc discharge prediction model to give an arc stage prediction result of the switch cabinet equipment; comparing the arc stage prediction result with data in a fault database to determine whether an arc discharge is about to occur, and performing an active early warning if it is determined to be an arc discharge. The method proposed in the present invention has the advantages of high confidence, strong applicability, high diagnostic efficiency, no need for phase information, etc., and can be applied to active early warning of arc discharge of switch cabinet equipment.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of power equipment status monitoring and fault diagnosis, and is applicable to an arc active early warning technology for switch cabinet equipment, and more specifically, to an arc active early warning method and system based on multi-spectral frequency bands. Background Art

[0002] In recent years, with the continuous expansion of economic construction, the demand for electric energy in all walks of life is also increasing, and the capacity of the power grid is also expanding. It should be noted that once a 10kV switchgear equipment fails, it will threaten the stable operation and safety of the power system, and the busbar insulation will be damaged, the equipment will be aged, etc., resulting in the burning of the equipment, and even seriously threatening the lives of staff and related personnel. During assembly, transportation or long-term operation, switchgear will inevitably have fault defects and be accompanied by partial discharge, and even develop into the arc stage with partial discharge. Long-term partial discharge will damage the insulation performance of the equipment, shorten the life of the equipment, and even cause power accidents. The arc stage is a destructive failure, which will cause irreversible damage to the switchgear.

[0003] In the existing technology, the safety protection means for switch cabinets mainly include status monitoring and arc protection. The former mainly relies on partial discharge live detection to complete regular inspections. However, traditional partial discharge detection means such as geoelectric waves or ultrasound are easily affected by on-site interference, and the defect detection rate and detection confidence rate are low. The diagnostic conclusion is vague, and there is a lack of quantitative means to accurately judge the initial discharge, intermittent creepage and flashover arcs. Therefore, it is impossible to achieve accurate warning before the fault and rapid processing after the fault, and it is even more impossible to form active protection logic to prevent accidents. At present, in the field of arc protection of switch cabinets, the protection measures taken for arc discharge are usually to use arc sensors to detect common fault areas, and diagnose and identify the discharge type based on the characteristic quantity extracted by the dual criteria of single-channel optical discharge signal and current amplitude. Although the above method is simple to operate and has a high confidence level, this method belongs to post-event protection, that is, to judge the characteristic data after the arc discharge occurs. This method has actually caused a certain degree of damage to the switch cabinet. In addition, the single-channel optical discharge signal is affected by the gas phase propagation distance. Moreover, since the arc monitoring sensor only targets the arc light of busbar faults, it has low optical sensitivity and cannot respond to the initial discharge before the fault. It only compares the setting value after the short-circuit fault occurs, and the equipment still has safety hazards. In summary, although the partial discharge and arc monitoring of switchgear have been applied to a certain extent, there is currently no protection method that can quantify the severity of discharge in the switchgear and give accurate and high-confidence early warnings.

[0004] In fact, there are differences in the spectra produced by different energy discharges. The spectra contain both characteristic spectral lines that can directly reflect the electron energy distribution during the discharge process and characteristic bands that can characterize the discharge driving mechanism. The interpretation of spectral information can reveal the microscopic process of discharge, and can achieve refined diagnosis and accurate warning of discharge in the switch cabinet. Therefore, it is urgent to study an active arc warning method and system based on multi-spectral bands. Summary of the invention

[0005] In order to solve the deficiencies in the prior art, the purpose of the present invention is to provide an arc active warning method and system based on multi-spectral bands, which makes full use of the spectral information contained in the discharge light signal, combines multi-spectral analysis technology, and extracts the discharge spectral characteristics to achieve the distinction between different types of local discharges and arc faults and find high-confidence criteria for the discharge pre-breakdown stage, thereby achieving active warning of power equipment.

[0006] The present invention adopts the following technical solution.

[0007] The present invention proposes an active arc warning method based on multi-spectral bands, comprising:

[0008] Step 1, collecting multi-spectral bands during the discharge process; using the correspondence between the multi-spectral bands and the discharge type and discharge sequence to establish historical discharge characteristic information;

[0009] Step 2, using the kernel extreme learning machine algorithm, taking the historical discharge feature information as input, to train the arc discharge prediction model of the high-voltage switchgear equipment, using the minimum norm least squares method during training and introducing the damping parameter to solve the arc discharge prediction model to obtain the least squares estimate of the output layer weight matrix of the extreme learning machine network, and using the least squares estimate of the output layer weight matrix of the extreme learning machine network to obtain the arc discharge prediction model;

[0010] Among them, the least squares estimate of the output layer weight matrix of the extreme learning machine network satisfies In the formula, is the least squares estimate of the output layer weight matrix of the extreme learning machine network, I is a diagonal matrix, B is the regularization coefficient, and μ is the damping parameter;

[0011] Step 3, using the multi-spectral band of the high-voltage switchgear equipment collected in real time as input, and using the trained arc discharge prediction model to give the arc stage prediction result of the switchgear;

[0012] Step 4: Compare the arc stage prediction result with the data in the fault database to determine whether arc discharge is about to occur. If it is determined to be an arc discharge, an active warning is issued.

[0013] In step 1, the multi-spectral band includes: ultraviolet light, visible light, and infrared light; the multi-spectral band simultaneously passes through filters of different bands to obtain light signals of different bands, wherein the intensity ratio of light signals of different bands corresponds to the discharge type one by one; the light signal is converted into an electrical signal using photoelectric conversion technology, and the discharge timing of the electrical signal is synchronously collected.

[0014] Step 2 includes:

[0015] Step 2.1: Extract discharge timing from historical discharge feature information As the input signal of the extreme learning machine network; where, is the discharge time t i The intensity ratio of light signals in different frequency bands under , i = 1, 2, ... m, m is the number of discharge moments;

[0016] Step 2.2, set the number of hidden nodes L and activation function of the extreme learning machine network l=1,2,…,L;ω l is the weight of the lth hidden node, b l is the bias of the lth hidden node;

[0017] Step 2.3, the arc discharge prediction model based on the extreme learning machine network satisfies the following relationship:

[0018] Hβ=T

[0019] In the formula, H is the hidden layer output matrix of the extreme learning machine network, β is the output layer weight matrix of the extreme learning machine network, and T is the sample expected output matrix, that is, the output matrix of the predicted time series data;

[0020] Among them, the hidden layer output matrix satisfies the following relationship:

[0021]

[0022] Step 2.4, use the minimum norm least squares method to solve the arc discharge prediction model and obtain In the formula, is the least squares estimate of the output layer weight matrix of the extreme learning machine network, I is a diagonal matrix, B is the regularization coefficient, and μ is the damping parameter; where the initial value of the damping parameter μ 0 Satisfies the following relationship:

[0023]

[0024] In the formula, To construct the matrix A 0 The diagonal elements of 0 satisfy: in is the loss function of the least squares estimate of the output layer weight matrix of the extreme learning machine network; the damping parameter μ is obtained after multiple iterations.

[0025] Step 2.5, discharge timing As the input signal of the arc discharge prediction model, the arc discharge prediction model outputs the prediction time series data Satisfies the following relationship:

[0026]

[0027] In the formula,

[0028] is the element in the prediction time series data output matrix of the arc discharge prediction model,

[0029] It is the element in the hidden layer output matrix of the arc discharge prediction model;

[0030] Step 2.6, define the kernel function matrix Ω ELM =HH T , the elements in the kernel function matrix are radial basis kernel functions Where i≠k and satisfies the following relationship:

[0031]

[0032] Where δ is the radial basis kernel function The kernel parameters, and The discharge time t i The intensity ratio of light signals in different frequency bands and the discharge time t k The intensity ratio of light signals in different frequency bands under ;

[0033] Use radial basis kernel function instead of random mapping relationship;

[0034] When the discharge time series data y is used as the input signal of the arc discharge prediction model, the discharge prediction data f(y) output by the arc discharge prediction model satisfies the following relationship:

[0035]

[0036] In the formula, is the discharge time t m The intensity ratio characteristic data of light signals in different frequency bands.

[0037] In step 3, when the local discharge of the high-voltage switchgear equipment develops into a through-going arc fault, the discharge timing data of the multi-spectral band collected in real time is used as the input of the arc discharge prediction model, and the discharge prediction data f(y) output by the arc discharge prediction model is used; the discharge prediction data f(y) is used to obtain the discharge prediction waveform.

[0038] In step 4, the discharge prediction waveform is compared with the discharge defect characteristic waveform, the fault stage of the switching device is determined according to the comparison result, and an early warning is issued.

[0039] The discharge defect characteristic waveforms include: corona discharge defect characteristic waveform, flash discharge defect characteristic waveform, suspended discharge defect characteristic waveform, pre-breakdown defect characteristic waveform and arc discharge defect characteristic waveform.

[0040] Step 4 includes:

[0041] Step 4.1: When the discharge prediction waveform and the corona discharge defect characteristic waveform have the highest degree of fit, it is determined that the switchgear is in the primary stage of partial discharge before the arc occurs; at this time, no warning is issued;

[0042] Step 4.2: When the discharge prediction waveform has the highest degree of fit with the characteristic waveform of the flash discharge defect and the characteristic waveform of the suspended discharge defect, it is determined that the switchgear is in the intermediate stage of partial discharge before the arc occurs; at this time, a fault location signal is sent to the operation and maintenance system;

[0043] Step 4.3: When the discharge prediction waveform and the pre-breakdown discharge defect characteristic waveform have the highest degree of fit, it is determined that the switchgear is in the transition stage from partial discharge to arc discharge; at this time, a fault processing demand signal is sent to the operation and maintenance system;

[0044] Step 4.4, when the fit between the discharge prediction waveform and the arc discharge defect characteristic waveform is the highest, it is determined that the switchgear is in the arc discharge stage; at this time, a protection action signal is sent to the relay protection system.

[0045] The fault database is constructed using the characteristic waveform of discharge defects.

[0046] On the other hand, the present invention also proposes an arc active warning method based on a multi-spectral band and an arc active warning system based on a multi-spectral band, the system comprising: a discharge sensor module, a multi-channel digital acquisition module, and a host computer;

[0047] The discharge sensing module includes: a filter and a discharge sensor array; the discharge sensing module is used to collect discharge fault light, and simultaneously pass the discharge fault light through filters of different bands to obtain light signals of different spectral bands, and the discharge fault light of different bands is input into the discharge sensor array, and the discharge sensor array synchronously converts the light signals of different spectral bands and synchronously outputs them into electrical signals; the electrical signal output by the discharge sensing module is used as the input of the multi-channel digital acquisition module; wherein the electrical signal includes a plurality of discharge pulse data;

[0048] Multi-channel digital acquisition module, used for synchronous acquisition of discharge pulse data; within the sampling time, the discharge pulse data and the sampling time together constitute the discharge timing information, which serves as the input of the host computer;

[0049] The host computer includes: a discharge prediction unit and a fault stage determination unit;

[0050] A discharge prediction unit is used to train an arc discharge prediction model for high-voltage switchgear equipment based on a kernel extreme learning machine classification algorithm and with historical discharge feature information as input. During training, the arc discharge prediction model is solved by using a minimum norm least squares method and introducing a damping parameter to obtain a least squares estimate of an output layer weight matrix of an extreme learning machine network. The arc discharge prediction model is obtained by using the least squares estimate of an output layer weight matrix of an extreme learning machine network.

[0051] The fault stage determination unit compares the arc stage prediction result with the data in the fault database to determine whether arc discharge is about to occur. If it is determined that arc discharge is about to occur, an active warning is issued.

[0052] The host computer also includes: a historical discharge characteristic information acquisition unit; the historical discharge characteristic information acquisition unit is used to establish historical discharge characteristic information based on a clustering algorithm using the correspondence between multi-spectral bands and discharge types and discharge timings.

[0053] The filters include: ultraviolet light filter, visible light filter, infrared light filter; among them,

[0054] The wavelength range of the ultraviolet filter is 251-398nm; the wavelength range of the visible light filter is 400-780nm; the wavelength range of the infrared filter is not less than 760nm;

[0055] The central transmittance of each filter is not less than 90%.

[0056] The discharge sensor array includes a plurality of mutually independent single-photon discharge sensor array units, each array unit is used to detect the optical signal in the local discharge stage before the arc occurs, and synchronously convert the optical signal into an electrical signal, and then synchronously output the obtained electrical signal;

[0057] The performance parameters of each single-photon discharge sensor array unit are: quantum efficiency not less than 45%, driving voltage not more than 30V, single-chip sensor array micro-element size not more than 4×4mm 2 , the maximum current is not more than 10mA, and the gain is not less than 4×10 6 , dark count rate is no more than 50kHz / mm 2 , crosstalk is no more than 10%, and residual pulse is no more than 1%.

[0058] The performance parameters of the multi-channel digital acquisition unit are: bandwidth not less than 80MHz, sampling rate not less than 320MS / s, acquisition channels not less than 4, minimum time base range not more than 100ns / div, pulse response not more than 1% overshoot.

[0059] The beneficial effect of the present invention is that, compared with the prior art, it can determine whether an arc discharge is about to occur by means of a high-confidence criterion before the arc discharge occurs, and make a corresponding active warning. This method has the advantages of high confidence, strong applicability, high diagnostic efficiency, and no need for phase information, and can be applied to the active warning of arc discharge in switch cabinet equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 It is a flowchart of the steps of an active arc warning method based on multi-spectral bands of the present invention;

[0061] Figure 2 is a schematic diagram of filter transmittance parameter requirements at different frequency bands in an embodiment of the present invention;

[0062] Figure 3 Schematic diagram of the whole process of partial discharge developing into through-arc discharge in an embodiment of the present invention;

[0063] Figure 4 is a schematic structural diagram of an arc active early warning system based on multi-spectral bands in an embodiment of the present invention;

[0064] Figure 4 The reference numerals in the figures are described as follows:

[0065] 1-discharge sensor module; 2-multi-channel digital acquisition module; 3-host computer. DETAILED DESCRIPTION

[0066] The present application is further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the protection scope of the present application.

[0067] The concept of the present invention is that, first, the discharge fault light passes through the ultraviolet, visible and infrared band filters to select the band light data and enter the independent single-photon discharge sensor array, and the multi-spectral band light signal is converted into an electrical signal through the photoelectric conversion of the sensor array, and input into the multi-channel digital acquisition unit for synchronous acquisition of the discharge data timing information, and finally the future change trend of the multi-spectral band data is predicted through the online detection and prediction method, and further compared with the fault database to determine whether the arc discharge is about to occur. If a fault is about to occur, the operation and maintenance personnel are notified to execute corresponding measures, otherwise, the equipment continues to be tested.

[0068] like Figure 1 , an arc active warning method based on multi-spectral bands, including steps 1 to 6.

[0069] Step 1: collect multi-spectral bands during the discharge process; and establish historical discharge characteristic information using the correspondence between the multi-spectral bands and the discharge type and discharge sequence.

[0070] In step 1, the multi-spectral band includes: ultraviolet light, visible light, and infrared light; the multi-spectral band simultaneously passes through filters of different bands to obtain light signals of different bands, wherein the intensity ratio of light signals of different bands corresponds to the discharge type one by one; the light signal is converted into an electrical signal using photoelectric conversion technology, and the discharge timing of the electrical signal is synchronously collected.

[0071] In the preferred embodiment of the present invention, the selected filters of ultraviolet, visible and infrared bands can realize the distinction of different types of partial discharge and arc discharge through the relative relationship of the discharge data of the three bands. The wavelength range of the ultraviolet filter is: 251-398nm, the wavelength range of the visible filter is: 400-780nm, and the infrared filter is: >760nm. The transmittance under the three bands should meet Figure 2 The parameters shown in the figure can enable each independent channel to obtain a more realistic optical signal amplitude and the optical signal amplitudes received by optical channels of different frequency bands can remain consistent for subsequent comparative analysis. Starting from the discharge type, the intensity information of the multi-spectral band can reflect the intrinsic characteristics of the discharge. Therefore, there will be different band intensity ratio information for different discharges. For example, the ultraviolet band information of corona discharge will be significantly higher than other discharges. Starting from the discharge timing, the ratio information of different frequency bands can reflect the evolution of different danger levels. For example, as the intensity of the discharge increases, the infrared band information of the discharge along the insulator will increase significantly.

[0072] In a preferred embodiment of the present invention, the discharge sensor array includes a plurality of mutually independent single-photon discharge sensor array units; the selected single-photon discharge sensor array can detect the local discharge phenomenon before the occurrence of the through-arc, such as Figure 3As shown, it can be clearly found that there are obvious differences in the multi-spectral band components of partial discharge and arc discharge. By capturing the entire process of partial discharge developing into arc, extracting the characteristic operator for judging the occurrence of arc discharge and the multi-spectral band criterion for distinguishing partial discharge and arc discharge, it is possible to predict whether an arc will occur in power equipment, thereby further realizing active arc warning of switch cabinet equipment. The performance parameters of the single-photon level sensor array are: quantum efficiency>45%, driving voltage<30V, and the micro-element size of the single-chip sensor array is less than<4×4mm 2 , maximum current <10mA, gain>4×10 6 , dark count rate <50kHz / mm 2 , crosstalk <10%, residual pulse <1%.

[0073] Specifically, a plurality of synchronous discharge pulse data are synchronously collected in a multi-channel digital acquisition unit with the same A / D sampling rate and analog bandwidth; the sampling time is used as the horizontal axis and the value of the discharge pulse data is used as the vertical axis. Under this coordinate axis, the continuous points obtained constitute the discharge timing information.

[0074] In this embodiment, the step of acquiring historical discharge characteristic information further includes:

[0075] Step A1, setting the number of cluster centers K of the K-means algorithm;

[0076] Step A2: Using the discharge timing information, a discharge timing information sample set D = {x 1 ,x 2 ,…,x m}, the input signal is clustered with the discharge timing information sample set; where x i is the discharge pulse data at the i-th sampling moment, each discharge pulse data is a set of vectors, each set of vectors includes electrical signals converted from optical signals of different spectral bands, i = 1, 2…m, m is the total number of sampling moments;

[0077] Step A3, through clustering iteration, obtain the cluster set C = {C 1 ,C 2 ,…,C K}; where C j is the jth cluster, and the jth cluster corresponds to the jth partial discharge type;

[0078] Step A4, using the error of the cluster set C as an indicator, calibrate each cluster center, wherein the error E of the cluster set C satisfies the following relationship:

[0079]

[0080] In the formula, x is the number of nodes belonging to the jth cluster C. jDischarge pulse data, μ j is the jth cluster C j The mean vector of , and satisfies the following relationship:

[0081]

[0082] It can be seen that the smaller the E value, the higher the similarity of samples within the cluster, and the more obvious the clustering effect of different discharge types.

[0083] Step A5, taking the cluster set C corresponding to the minimum value of the error E as the final clustering result; wherein each cluster corresponds to a partial discharge type, and the data in each cluster is the discharge characteristic data under the corresponding partial discharge type.

[0084] Step A6, in a cluster, that is, in a local discharge type, the sampling time is used as the horizontal axis and the value of the discharge characteristic data is used as the vertical axis. Under this coordinate axis, the continuous points obtained constitute the discharge characteristic time series information; the discharge characteristic time series information is used to form a historical discharge characteristic information library.

[0085] Step 2, using the kernel extreme learning machine algorithm, taking the historical discharge feature information as input, to train the arc discharge prediction model of the high-voltage switchgear equipment, using the minimum norm least squares method during training and introducing the damping parameter to solve the arc discharge prediction model to obtain the least squares estimate of the output layer weight matrix of the extreme learning machine network, and using the least squares estimate of the output layer weight matrix of the extreme learning machine network to obtain the arc discharge prediction model;

[0086] Among them, the least squares estimate of the output layer weight matrix of the extreme learning machine network satisfies In the formula, is the least squares estimate of the output layer weight matrix of the extreme learning machine network, I is a diagonal matrix, B is the regularization coefficient, and μ is the damping parameter.

[0087] Step 2 includes:

[0088] Step 2.1: Extract discharge timing from historical discharge feature information As the input signal of the extreme learning machine network; where, is the discharge time t i The intensity ratio of light signals in different frequency bands under , i = 1, 2, ... m, m is the number of discharge moments;

[0089] Step 2.2, set the number of hidden nodes L and activation function of the extreme learning machine network l=1,2,…,L;ω l is the weight of the lth hidden node, b l is the bias of the lth hidden node;

[0090] Step 2.3, the arc discharge prediction model based on the extreme learning machine network satisfies the following relationship:

[0091] Hβ=T

[0092] In the formula, H is the hidden layer output matrix of the extreme learning machine network, β is the output layer weight matrix of the extreme learning machine network, and T is the sample expected output matrix, that is, the output matrix of the predicted time series data;

[0093] Among them, the hidden layer output matrix satisfies the following relationship:

[0094]

[0095] Step 2.4, use the minimum norm least squares method to solve the arc discharge prediction model and obtain In the formula, is the least squares estimate of the output layer weight matrix of the extreme learning machine network, I is a diagonal matrix, B is the regularization coefficient, and μ is the damping parameter; where the initial value of the damping parameter μ 0 Satisfies the following relationship:

[0096]

[0097] In the formula, To construct the matrix A 0 The diagonal elements of 0 satisfy: in is the loss function of the least squares estimate of the output layer weight matrix of the extreme learning machine network; the damping parameter μ is obtained after multiple iterations.

[0098] The selection of regularization coefficient B and kernel parameter δ is the key parameter to improve the prediction of time series data. The prediction trend of discharge multi-spectral information is obtained by initializing and iterating the parameters of the arc discharge prediction model in the switchgear.

[0099] In this embodiment, in order to avoid overfitting, one means is to use regularization to limit the complexity of the model. Therefore, when using the minimum norm least squares method to solve the arc discharge prediction model, a regularization coefficient is also introduced. In order to further control the regularization coefficient within a certain trust region, thereby ensuring the convergence of the least squares method and obtaining the optimal solution, a damping parameter μ is introduced in this embodiment, where the initial value of the damping parameter μ 0 Satisfies the following relationship:

[0100]

[0101] In the formula, To construct the matrix A 0 The diagonal elements of 0 satisfy: in The loss function is the least squares estimate of the output layer weight matrix of the extreme learning machine network.

[0102] After multiple iterations, the damping parameter μ is obtained so that the least squares estimate of the output layer weight matrix of the extreme learning machine network satisfies the following relationship:

[0103]

[0104] Step 2.5, discharge timing As the input signal of the arc discharge prediction model, the arc discharge prediction model outputs the prediction time series data Satisfies the following relationship:

[0105]

[0106] In the formula,

[0107] is the element in the prediction time series data output matrix of the arc discharge prediction model,

[0108] It is the element in the hidden layer output matrix of the arc discharge prediction model;

[0109] Step 2.6, define the kernel function matrix Ω ELM =HH T , the elements in the kernel function matrix are radial basis kernel functions Where i≠k and satisfies the following relationship:

[0110]

[0111] Where δ is the radial basis kernel function The kernel parameters, and The discharge time t i The intensity ratio of light signals in different frequency bands and the discharge time t k The intensity ratio of light signals in different frequency bands under ;

[0112] Use radial basis kernel function instead of random mapping relationship;

[0113] When the discharge time series data y is used as the input signal of the arc discharge prediction model, the discharge prediction data f(y) output by the arc discharge prediction model satisfies the following relationship:

[0114]

[0115] In the formula, is the discharge time t m The intensity ratio characteristic data of light signals in different frequency bands.

[0116] In step 3, when the local discharge of the high-voltage switchgear equipment develops into a through-going arc fault, the discharge timing data of the multi-spectral band collected in real time is used as the input of the arc discharge prediction model, and the discharge prediction data f(y) output by the arc discharge prediction model is used; the discharge prediction data f(y) is used to obtain the discharge prediction waveform.

[0117] Step 4: Compare the arc stage prediction result with the data in the fault database to determine whether arc discharge is about to occur. If it is determined to be an arc discharge, an active warning is issued.

[0118] Specifically, in step 4, the discharge prediction waveform is compared with the discharge defect characteristic waveform, the fault stage of the switching device is determined according to the comparison result, and an early warning is issued.

[0119] The discharge defect characteristic waveforms include: corona discharge defect characteristic waveform, flash discharge defect characteristic waveform, suspended discharge defect characteristic waveform, pre-breakdown defect characteristic waveform and arc discharge defect characteristic waveform.

[0120] In the present invention, a typical fault database is established, which is mainly obtained by setting typical defects of partial discharge developing into arc in typical switch cabinet equipment (including creepage along solid insulators, floating potential and tip protrusions, etc.), obtaining the intrinsic characteristics of multi-spectral band data of different defects, and then comparing with the results predicted by the nuclear extreme learning machine (KELM) algorithm to determine the fault development stage of the switch cabinet and guide the operation and maintenance personnel to take corresponding early warning measures.

[0121] Step 4 includes:

[0122] Step 4.1: When the discharge prediction waveform and the corona discharge defect characteristic waveform have the highest degree of fit, it is determined that the switchgear is in the primary stage of partial discharge before the arc occurs; at this time, no warning is issued;

[0123] Step 4.2: When the discharge prediction waveform has the highest degree of fit with the characteristic waveform of the flash discharge defect and the characteristic waveform of the suspended discharge defect, it is determined that the switchgear is in the intermediate stage of partial discharge before the arc occurs; at this time, a fault location signal is sent to the operation and maintenance system;

[0124] Step 4.3: When the discharge prediction waveform and the pre-breakdown discharge defect characteristic waveform have the highest degree of fit, it is determined that the switchgear is in the transition stage from partial discharge to arc discharge; at this time, a fault processing demand signal is sent to the operation and maintenance system;

[0125] Step 4.4, when the fit between the discharge prediction waveform and the arc discharge defect characteristic waveform is the highest, it is determined that the switchgear is in the arc discharge stage; at this time, a protection action signal is sent to the relay protection system.

[0126] The fault database is constructed using the characteristic waveform of discharge defects.

[0127] like Figure 4 , an arc active warning system based on multi-spectral bands, including: a discharge sensor module 1, a multi-channel digital acquisition module 2, and a host computer 3.

[0128] The discharge sensor module 1 comprises: a filter and a discharge sensor array; the discharge sensor module is used to collect discharge fault light, and simultaneously pass the discharge fault light through filters of different bands to obtain light signals of different spectral bands, the discharge fault light of different bands is input into the discharge sensor array, and the discharge sensor array synchronously converts the light signals of different spectral bands and synchronously outputs them into electrical signals; the electrical signal output by the discharge sensor module is used as the input of the multi-channel digital acquisition module; wherein the electrical signal comprises a plurality of discharge pulse data;

[0129] The multi-channel digital acquisition module 2 is used to synchronously acquire the discharge pulse data; within the sampling time, the discharge pulse data and the sampling time together constitute the discharge timing information as the input of the host computer;

[0130] The host computer 3 includes: a discharge prediction unit and a fault stage determination unit;

[0131] A discharge prediction unit is used to train an arc discharge prediction model for high-voltage switchgear equipment based on a kernel extreme learning machine classification algorithm and with historical discharge feature information as input. During training, the arc discharge prediction model is solved by using a minimum norm least squares method and introducing a damping parameter to obtain a least squares estimate of an output layer weight matrix of an extreme learning machine network. The arc discharge prediction model is obtained by using the least squares estimate of an output layer weight matrix of an extreme learning machine network.

[0132] The fault stage determination unit compares the arc stage prediction result with the data in the fault database to determine whether arc discharge is about to occur. If it is determined that arc discharge is about to occur, an active warning is issued.

[0133] The host computer also includes: a historical discharge characteristic information acquisition unit; the historical discharge characteristic information acquisition unit is used to establish historical discharge characteristic information based on a clustering algorithm using the correspondence between multi-spectral bands and discharge types and discharge timings.

[0134] The filters include: ultraviolet light filter, visible light filter, infrared light filter; among them,

[0135] The wavelength range of the ultraviolet filter is 251-398nm; the wavelength range of the visible light filter is 400-780nm; the wavelength range of the infrared filter is not less than 760nm;

[0136] The central transmittance of each filter is not less than 90%.

[0137] The discharge sensor array includes a plurality of mutually independent single-photon discharge sensor array units, each array unit is used to detect the optical signal in the local discharge stage before the arc occurs, and synchronously convert the optical signal into an electrical signal, and then synchronously output the obtained electrical signal;

[0138] The performance parameters of each single-photon discharge sensor array unit are: quantum efficiency not less than 45%, driving voltage not more than 30V, single-chip sensor array micro-element size not more than 4×4mm 2 , the maximum current is not more than 10mA, and the gain is not less than 4×10 6 , dark count rate is no more than 50kHz / mm 2 , crosstalk is no more than 10%, and residual pulse is no more than 1%.

[0139] The performance parameters of the multi-channel digital acquisition unit are: bandwidth not less than 80MHz, sampling rate not less than 320MS / s, acquisition channels not less than 4, minimum time base range not more than 100ns / div, pulse response not more than 1% overshoot.

[0140] The applicant of the present invention has made a detailed explanation and description of the implementation examples of the present invention in conjunction with the drawings in the specification. However, those skilled in the art should understand that the above implementation examples are only preferred implementation schemes of the present invention, and the detailed description is only to help readers better understand the spirit of the present invention, but not to limit the scope of protection of the present invention. On the contrary, any improvements or modifications based on the inventive spirit of the present invention should fall within the scope of protection of the present invention.

Claims

1. An active arc warning method based on multi-spectral bands, It is characterized in that The method comprises: Step 1, collecting multi-spectral bands during the discharge process; using the correspondence between the multi-spectral bands and the discharge type and discharge sequence to establish historical discharge characteristic information; the multi-spectral bands include: ultraviolet light, visible light, and infrared light; the wavelength range of the ultraviolet filter is: 251-398nm, the wavelength range of the visible filter is: 400-780nm, and the infrared filter is: >760nm, and the multi-spectral bands simultaneously pass through filters of different bands to obtain light signals of different bands, wherein the intensity ratio of the light signals of different bands corresponds to the discharge type one by one, and the relative relationship of the discharge data of the three bands is used to distinguish different local discharge types and arc discharge; Step 2: Use the kernel extreme learning machine algorithm to extract the discharge timing from the historical discharge feature information , as the input signal of the extreme learning machine network; among them, Discharge time The intensity ratio of light signals in different frequency bands under , is the number of discharge moments; Set the number of hidden nodes in the extreme learning machine network And the activation function , ; For the The weights of hidden nodes, For the The bias of hidden nodes is determined to obtain the arc discharge prediction model based on extreme learning machine network. The arc discharge prediction model of high-voltage switchgear equipment is trained. During the training, the minimum norm least squares method is used and the damping parameter is introduced to solve the arc discharge prediction model to obtain the least squares estimate of the output layer weight matrix of the extreme learning machine network. The arc discharge prediction model is obtained using the least squares estimate of the output layer weight matrix of the extreme learning machine network. Among them, the least squares estimate of the output layer weight matrix of the extreme learning machine network satisfies , where is the least squares estimate of the output layer weight matrix of the extreme learning machine network, is a diagonal matrix, is the hidden layer output matrix of the extreme learning machine network, is the sample expected output matrix, is the regularization coefficient, is the damping parameter; the initial value of the damping parameter Satisfies the following relationship: In the formula, To construct the matrix The diagonal elements of satisfy: ,in is the loss function of the least squares estimate of the output layer weight matrix of the extreme learning machine network; When the discharge timing data When the arc discharge prediction model outputs the discharge prediction data , satisfying the following relationship: In the formula, Discharge time The intensity ratio of light signals in different frequency bands under is the kernel function matrix, , radial basis kernel function is the kernel function matrix Elements in Step 3: When the local discharge of the high-voltage switchgear equipment develops into a through-going arc fault, the discharge time series data of the multi-spectral band collected in real time is used as the input of the arc discharge prediction model, and the arc discharge prediction model trained in step 2 outputs the discharge prediction data. ; Using discharge prediction data A discharge prediction waveform is obtained as an arc stage prediction result of the switching device; Step 4: Compare the arc stage prediction result with the data in the fault database to determine whether arc discharge is about to occur. If it is determined to be an arc discharge, an active warning is issued.

2. The arc active early warning method based on multi-spectral bands according to claim 1, It is characterized in that In step 1, the optical signal is converted into an electrical signal using photoelectric conversion technology, and the discharge timing of the electrical signal is synchronously collected.

3. The arc active early warning method based on multi-spectral bands according to claim 2, It is characterized in that Step 2 also includes: The arc discharge prediction model based on the extreme learning machine network satisfies the following relationship: In the formula, is the hidden layer output matrix of the extreme learning machine network, is the output layer weight matrix of the extreme learning machine network, is the sample expected output matrix, that is, the output matrix of the predicted time series data; Among them, the hidden layer output matrix satisfies the following relationship: Discharge timing As the input signal of the arc discharge prediction model, the arc discharge prediction model outputs the prediction time series data , satisfying the following relationship: In the formula, is the element in the prediction time series data output matrix of the arc discharge prediction model, It is the element in the hidden layer output matrix of the arc discharge prediction model; Radial Basis Kernel Function ,in , satisfying the following relationship: In the formula, is the radial basis kernel function The kernel parameters, and Discharge time The intensity ratio and discharge time of light signals in different frequency bands under The intensity ratio of light signals in different frequency bands under ; The radial basis kernel function is used to replace the random mapping relationship.

4. The arc active early warning method based on multi-spectral bands according to claim 3, It is characterized in that In step 4, the discharge prediction waveform is compared with the discharge defect characteristic waveform, the fault stage of the switching device is determined according to the comparison result, and an early warning is issued.

5. The arc active early warning method based on multi-spectral bands according to claim 4, It is characterized in that The discharge defect characteristic waveforms include: corona discharge defect characteristic waveform, flash discharge defect characteristic waveform, suspended discharge defect characteristic waveform, pre-breakdown defect characteristic waveform and arc discharge defect characteristic waveform.

6. The arc active early warning method based on multi-spectral bands according to claim 5, It is characterized in that Step 4 includes: Step 4.1: When the discharge prediction waveform and the corona discharge defect characteristic waveform have the highest degree of fit, it is determined that the switchgear is in the primary stage of partial discharge before the arc occurs; at this time, no warning is issued; Step 4.2: When the discharge prediction waveform has the highest degree of fit with the characteristic waveform of the flash discharge defect and the characteristic waveform of the suspended discharge defect, it is determined that the switchgear is in the intermediate stage of partial discharge before the arc occurs; at this time, a fault location signal is sent to the operation and maintenance system; Step 4.3: When the discharge prediction waveform and the pre-breakdown discharge defect characteristic waveform have the highest degree of fit, it is determined that the switchgear is in the transition stage from partial discharge to arc discharge; at this time, a fault processing demand signal is sent to the operation and maintenance system; Step 4.4, when the fit between the discharge prediction waveform and the arc discharge defect characteristic waveform is the highest, it is determined that the switchgear is in the arc discharge stage; at this time, a protection action signal is sent to the relay protection system.

7. The arc active early warning method based on multi-spectral bands according to claim 5, It is characterized in that The fault database is constructed using the characteristic waveform of discharge defects.

8. An arc active warning system based on multi-spectral bands, used to implement an arc active warning method based on multi-spectral bands as claimed in any one of claims 1 to 7, It is characterized in that The system includes: discharge sensor module, multi-channel digital acquisition module, and host computer; The discharge sensing module includes: a filter and a discharge sensor array; the discharge sensing module is used to collect discharge fault light, and simultaneously pass the discharge fault light through filters of different bands to obtain light signals of different spectral bands, and the discharge fault light of different bands is input into the discharge sensor array, and the discharge sensor array synchronously converts the light signals of different spectral bands and synchronously outputs them into electrical signals; the electrical signal output by the discharge sensing module is used as the input of the multi-channel digital acquisition module; wherein the electrical signal includes a plurality of discharge pulse data; Multi-channel digital acquisition module, used for synchronous acquisition of discharge pulse data; within the sampling time, the discharge pulse data and the sampling time together constitute the discharge timing information, which serves as the input of the host computer; The host computer includes: a discharge prediction unit and a fault stage determination unit; A discharge prediction unit is used to train an arc discharge prediction model for high-voltage switchgear equipment based on a kernel extreme learning machine classification algorithm and with historical discharge feature information as input. During training, the arc discharge prediction model is solved by using a minimum norm least squares method and introducing a damping parameter to obtain a least squares estimate of an output layer weight matrix of an extreme learning machine network. The arc discharge prediction model is obtained by using the least squares estimate of an output layer weight matrix of an extreme learning machine network. The fault stage determination unit compares the arc stage prediction result with the data in the fault database to determine whether arc discharge is about to occur. If it is determined that arc discharge is about to occur, an active warning is issued.

9. The arc active warning system based on multi-spectral bands according to claim 8, It is characterized in that The host computer also includes: a historical discharge characteristic information acquisition unit; the historical discharge characteristic information acquisition unit is used to establish historical discharge characteristic information based on a clustering algorithm using the correspondence between multi-spectral bands and discharge types and discharge sequences.

10. The arc active warning system based on multi-spectral bands according to claim 8, It is characterized in that The filters include: ultraviolet light filter, visible light filter, infrared light filter; among them, The wavelength range of the ultraviolet filter is 251-398nm; the wavelength range of the visible light filter is 400-780nm; the wavelength range of the infrared filter is not less than 760nm; The central transmittance of each filter is not less than 90%.

11. The arc active warning system based on multi-spectral bands according to claim 8, It is characterized in that The discharge sensor array includes a plurality of mutually independent single-photon discharge sensor array units, each array unit is used to detect the optical signal in the local discharge stage before the arc occurs, and synchronously convert the optical signal into an electrical signal, and then synchronously output the obtained electrical signal; The performance parameters of each single-photon discharge sensor array unit are: quantum efficiency not less than 45%, driving voltage not more than 30V, single-chip sensor array micro-element size not more than 4×4mm 2 , the maximum current is not more than 10mA, and the gain is not less than 4×10 6 , dark count rate is no more than 50kHz / mm 2 , crosstalk is no more than 10%, and residual pulse is no more than 1%.

12. The arc active warning system based on multi-spectral bands according to claim 8, It is characterized in that The performance parameters of the multi-channel digital acquisition unit are: bandwidth not less than 80MHz, sampling rate not less than 320MS / s, acquisition channels not less than 4, minimum time base range not greater than 100ns / div, pulse response not greater than 1% overshoot.

Citation Information

Cited By

  • Switch cabinet partial discharge intelligent diagnosis system integrating acousto-optic thermoelectric four-dimensional signals

    CN122386031A