WO2022067562A1 - 一种故障电弧的诊断方法、装置和计算机可读存储介质 - Google Patents

一种故障电弧的诊断方法、装置和计算机可读存储介质 Download PDF

Info

Publication number
WO2022067562A1
WO2022067562A1 PCT/CN2020/119020 CN2020119020W WO2022067562A1 WO 2022067562 A1 WO2022067562 A1 WO 2022067562A1 CN 2020119020 W CN2020119020 W CN 2020119020W WO 2022067562 A1 WO2022067562 A1 WO 2022067562A1
Authority
WO
WIPO (PCT)
Prior art keywords
function
current signal
intrinsic mode
neural network
mutation
Prior art date
Application number
PCT/CN2020/119020
Other languages
English (en)
French (fr)
Inventor
夏雨
田中伟
Original Assignee
西门子股份公司
西门子（中国）有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 西门子股份公司, 西门子（中国）有限公司 filed Critical 西门子股份公司
Priority to PCT/CN2020/119020 priority Critical patent/WO2022067562A1/zh
Publication of WO2022067562A1 publication Critical patent/WO2022067562A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
    • G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
    • G01R31/12—Testing dielectric strength or breakdown voltage ; Testing or monitoring effectiveness or level of insulation, e.g. of a cable or of an apparatus, for example using partial discharge measurements; Electrostatic testing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
    • Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
    • Y04S10/00—Systems supporting electrical power generation, transmission or distribution
    • Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications

Definitions

  • the present invention relates to the technical field of power supply and distribution, and in particular, to an arc fault diagnosis method, device and computer-readable storage medium.
  • the arc fault is studied by means of simulation and experiments, and it is found that the electrical characteristics of the arc fault are related to the nature of the load, the location of occurrence, and the type of fault, and are intermittent, random and uncertain. Therefore, how to effectively detect arc faults has always been a hot research topic in the field of arcing.
  • Embodiments of the present invention provide an arc fault diagnosis method, device, and computer-readable storage medium.
  • a method for diagnosing an arc fault comprising:
  • the mutation features of the plurality of high frequency components are input into the trained neural network model to output the fault arc type from the neural network model.
  • the embodiment of the present invention realizes the multi-variable arc fault diagnosis, which can overcome the difficulty of threshold selection in the single-variable arc fault diagnosis, being susceptible to fault causes, arcing conditions, etc.
  • the technical defects affected by the nature of the load are improved, and the diagnostic accuracy is improved.
  • the multi-variable input neural network is easy to find the potential laws and mapping relationships between various fault currents, and can adaptively determine the potential fuzzy decision-making laws in the sample data, which can meet the actual requirements of fast and accurate action in fault arc protection technology. .
  • performing empirical mode decomposition on the acquired current signal to obtain a plurality of intrinsic mode functions of the current signal includes:
  • the interpolation method is a cubic spline interpolation method
  • the neural network model is a quantum neural network model
  • the obtaining abrupt features of multiple high-frequency components of the current signal based on multiple intrinsic mode functions includes:
  • a marginal spectral region variation feature of the eigenmode function is determined as a mutation feature corresponding to a high frequency component of the eigenmode function.
  • the Hilbert transform can be used to quickly determine the characteristic components in each frequency band.
  • the obtaining abrupt features of multiple high-frequency components of the current signal based on multiple intrinsic mode functions includes:
  • a normalized cumulative value of the abrupt amount of each eigenmode function is determined as an abrupt feature corresponding to the high frequency component of the eigenmode function.
  • the abrupt characteristics of the high frequency components can be determined.
  • each intrinsic mode function includes:
  • the dimensionless index of each eigenmode function is determined, and the dimensionless index is determined. is the abrupt feature corresponding to the high-frequency component of the natural mode function.
  • the method further includes:
  • the load characteristics as training data, the environmental parameters as training data, and the mutation characteristics of high-frequency components as training data are further input into the neural network model to train the neural network model;
  • the environmental parameters of the current signal collection point and the load characteristics of the current signal collection point are further input into the neural network model.
  • training the neural network model through the load characteristics and environmental parameters helps the quantum neural network to find the potential laws and mapping relationships between various fault currents, and can adaptively determine the potential fuzzy decision laws in the sample data.
  • a fault arc diagnostic device comprising:
  • the acquisition module is used to acquire the current signal
  • an empirical mode decomposition module configured to perform empirical mode decomposition on the current signal to obtain a plurality of intrinsic mode functions of the current signal
  • a mutation feature acquisition module configured to acquire mutation features of multiple high-frequency components of the current signal based on the multiple intrinsic mode functions
  • the type determination module is used for inputting the mutation features of the plurality of high-frequency components into the trained neural network model, so as to output the fault arc type from the neural network model.
  • the embodiment of the present invention realizes the multi-variable arc fault diagnosis, which can overcome the difficulty of threshold selection in the single-variable arc fault diagnosis, being susceptible to fault causes, arcing conditions, etc.
  • the technical defects affected by the nature of the load are improved, and the diagnostic accuracy is improved.
  • the empirical mode decomposition module is used to determine the local extreme point of the current signal; use an interpolation method to fit the local extreme point into an upper envelope and a lower envelope; based on The upper envelope and the lower envelope separate natural mode functions from the current signal.
  • the mutation feature acquisition module is configured to perform Hilbert transform on each intrinsic mode function to obtain a Hilbert spectrum of the intrinsic mode function, based on the Hilbert spectrum of each intrinsic mode function
  • the Hilbert spectrum determines the marginal spectral region variation characteristics of the eigenmode function as abrupt features corresponding to high frequency components of the eigenmode function.
  • the Hilbert transform can be used to quickly determine the characteristic components in each frequency band.
  • the mutation feature acquisition module is configured to extract the amplitude energy of each intrinsic mode function, determine the mutation amount of the amplitude energy of each intrinsic mode function, and determine the normalization of the mutation amount The accumulated value is used as the abrupt characteristic of the high-frequency component corresponding to the natural mode function.
  • the abrupt characteristics of the high frequency components can be determined.
  • the mutation feature acquisition module is configured to perform Hilbert transform on each intrinsic mode function to obtain a Hilbert spectrum of the intrinsic mode function, based on the Hilbert spectrum of each intrinsic mode function
  • the Hilbert spectrum determines the variation characteristics of the marginal spectral region of the eigenmode function; extracts the amplitude energy of each eigenmode function, determines the abrupt change of the amplitude energy of each eigenmode function, and determines each eigenmode function.
  • each eigenmode function is determined based on the variation characteristics of the marginal spectral region of each eigenmode function and the normalized cumulative value of the mutation amount of the eigenmode function
  • the dimensionless index is determined as the mutation characteristic corresponding to the high frequency component of the natural mode function.
  • the type determination module is further configured to, in the training phase of the neural network model, further input the load characteristics as training data, the environmental parameters as training data, and the mutation characteristics as high-frequency components of training data into the training data.
  • the neural network model is used to train the neural network model; in the application stage of the neural network model, the environmental parameters of the current signal collection point and the load characteristics of the current signal collection point are further input into the neural network model.
  • training the neural network model through the load characteristics and environmental parameters helps the quantum neural network to find the potential laws and mapping relationships between various fault currents, and can adaptively determine the potential fuzzy decision laws in the sample data.
  • An arc fault diagnosis device comprising a processor, a memory, and a computer program stored on the memory and running on the processor, the computer program being executed by the processor to achieve any of the above The diagnostic method of arc fault described above.
  • a computer-readable storage medium stores a computer program on the computer-readable storage medium, and when the computer program is executed by a processor, implements the method for diagnosing an arc fault according to any one of the above.
  • FIG. 1 is a flowchart of a method for diagnosing an arc fault according to an embodiment of the present invention.
  • FIG. 2 is a schematic diagram of current collection according to an embodiment of the present invention.
  • FIG. 3 is an exemplary schematic diagram of a diagnosis process of an arc fault according to an embodiment of the present invention.
  • FIG. 4 is a configuration diagram of an arc fault diagnosis apparatus according to an embodiment of the present invention.
  • FIG. 5 is an exemplary structural block diagram of an arc fault diagnosis apparatus with a memory-processor architecture according to an embodiment of the present invention.
  • the applicant's fault arc current diagnosis method based on Hilbert-Huang (Hilbert-Huang) transformation and neural network model (preferably a quantum neural network model)
  • Hilbert-Huang Hilbert-Huang
  • neural network model preferably a quantum neural network model
  • the change rule of the spectral characteristics of the fault arc current transient process is clarified, and the characteristic signals before and after the arc are constructed.
  • the potential fuzzy decision-making law in the data is extracted, and the uncertain potential law and relationship mapping of the fault arc current are extracted, so as to solve the problem that the threshold value is difficult to accurately determine in the existing single-variable criterion method of the fault arc.
  • the diagnosis of fault arc is essentially a pattern recognition problem.
  • the invention mainly includes three links: fault arc current signal acquisition and empirical mode decomposition (EMD), feature extraction and state identification, and fault arc identification, and the key lies in how to effectively extract the characteristics of each fault arc.
  • EMD empirical mode decomposition
  • feature extraction and state identification feature extraction and state identification
  • fault arc identification the key lies in how to effectively extract the characteristics of each fault arc.
  • the whole scheme On the basis of independent feature extraction, combined with quantum neural network artificial intelligence algorithm, the whole scheme has the ability to learn and adapt to unknown or uncertain systems. diagnosis.
  • FIG. 1 is a flowchart of a method for diagnosing an arc fault according to an embodiment of the present invention.
  • the method includes:
  • Step 101 Collect current signals.
  • current sensors can be arranged at any location in the power supply and distribution system where there may be an arc fault to acquire current signals.
  • FIG. 2 is a schematic diagram of current collection according to an embodiment of the present invention.
  • the commercial electrical load 51 is connected to the power distribution room 54 via the auxiliary access device 52 .
  • Industrial electrical loads 53 are connected to a distribution room 54 .
  • the power distribution room 54 is connected to the grid 56 .
  • a current sensor may be deployed around the transformer between the power distribution room 54 and the grid 56 .
  • current sensors may also be deployed at commercial electrical loads 51 , auxiliary access devices 52 , industrial electrical loads 53 , and power distribution rooms 54 , among others.
  • Step 102 Perform empirical mode decomposition on the current signal to obtain a plurality of intrinsic mode functions of the current signal.
  • performing empirical mode decomposition on the collected current signal to obtain the intrinsic mode function of the current signal includes: determining a local extreme point of the current signal; using an interpolation method to fit the local extreme point as an upper packet An envelope and a lower envelope; based on the upper and lower envelopes, an intrinsic mode function is separated from the current signal.
  • the interpolation method is a cubic spline interpolation method.
  • Step 103 Acquire abrupt change characteristics of multiple high-frequency components of the current signal based on the multiple natural mode functions.
  • the step 103 obtaining the mutation characteristics of the multiple high-frequency components of the current signal based on the multiple intrinsic mode functions includes: performing a Hilbert transform on each intrinsic mode function to obtain the intrinsic mode function.
  • Hilbert spectrum of the eigenmode function based on the Hilbert spectrum of each eigenmode function, the marginal spectral region variation characteristics of the eigenmode function are determined as abrupt changes corresponding to the high-frequency components of the eigenmode function feature. It can be seen that the Hilbert transform can be used to quickly determine the characteristic components in each frequency band.
  • step 103 acquiring the mutation characteristics of the multiple high-frequency components of the current signal based on the multiple intrinsic mode functions includes: extracting the amplitude energy of each intrinsic mode function; determining each intrinsic mode function The mutation amount of the amplitude energy of the function; the normalized accumulated value of the mutation amount of each eigenmode function is determined as the mutation characteristic corresponding to the high frequency component of the eigenmode function. Therefore, based on the direct calculation of the abrupt amount of the amplitude energy of the intrinsic mode function, the abrupt characteristics of the high frequency components can be determined.
  • the step 103 obtaining the mutation characteristics of the multiple high-frequency components of the current signal based on the eigenmode function includes: performing Hilbert transform on each eigenmode function to obtain the eigenmode function of the Hilbert spectrum, based on the Hilbert spectrum of each eigenmode function to determine the marginal spectral region variation characteristics of the eigenmode function; extract the amplitude energy of each eigenmode function to determine each eigenmode function The mutation amount of the amplitude energy of the function, the normalized cumulative value of the mutation amount of each intrinsic mode function is determined; based on the variation characteristics of the marginal spectral region of each intrinsic mode function and the mutation amount of the intrinsic mode function The normalized accumulated value of , determines the dimensionless index of each intrinsic mode function, and determines the dimensionless index as the mutation characteristic corresponding to the high-frequency component of the intrinsic mode function. It can be seen that by extracting a variety of dimensionless indicators as the input feature vector set of the diagnosis network, the problem that the arc detection based on the traditional threshold
  • Step 104 Input the mutation features of the plurality of high-frequency components into the trained neural network model, so as to output the fault arc type from the neural network model.
  • the neural network model is a quantum neural network (QNN) model.
  • QNN quantum neural network
  • Quantum neural network is composed of several quantum neurons according to a certain topology. It is a method that attempts to combine the artificial neural network model with the advantages of quantum information.
  • the method further includes: in the training phase of the neural network model, further inputting the load characteristics as training data, the environmental parameters as training data, and the mutation characteristics as high-frequency components of training data into the training data
  • the neural network model is used to train the neural network model; in the application stage of the neural network model, the environmental parameters of the current signal collection point and the load characteristics of the current signal collection point are further input into the neural network model.
  • training the neural network model through the load characteristics and environmental parameters helps the quantum neural network to find the potential laws and mapping relationships between various fault currents, and can adaptively determine the potential fuzzy decision laws in the sample data.
  • the following describes the fault arc diagnosis process of the embodiment of the present invention by taking the quantum neural network as an example.
  • FIG. 3 is a schematic diagram of a diagnosis process of an arc fault according to an embodiment of the present invention.
  • Arc fault current signal acquisition and empirical mode decomposition provide source signals and feature quantities for the overall detection process, and filter the initial signal, which is also the first step in the algorithm's start-up operation. Its input is the current signal of the sensor, which characterizes the dynamic characteristics of the measured circuit in real time, and its output is the feature extraction link of the fault arc state, so as to extract the multi-variable indicators that can accurately indicate the arc fault, and form a fault diagnosis vector matrix.
  • the arc fault state feature extraction link receives the time-frequency domain feature information from the arc fault current signal acquisition and empirical mode decomposition links, forms a eigenvector matrix sample set, and inputs it to the quantum neural network of the fault arc for its training and learning. Appropriate smoothing parameters to continuously improve and build a fault diagnosis model. Finally, the diagnosis decision of the scheme is output by the fault arc quantum neural network model link, and the fault type of the current waveform signal currently collected is identified.
  • the three links of the system workflow all contain time-frequency domain multi-scale analysis and self-learning control ideas, which changes the existing detection scheme that only uses a single frequency domain or time domain analysis method to extract the amplitude of non-stationary fault current signals obtained on site. Due to the limitations of value fault characteristics, the working principles of each link in the system are as follows:
  • the Hilbert-Huang transform is a new method of nonlinear non-stationary signal analysis, and its process is mainly composed of two parts: empirical mode decomposition method and Hilbert spectrum analysis.
  • the current signal 31 is decomposed into a series of linear sums of a reasonable number of intrinsic mode functions (IMFs) using empirical mode decomposition, and then Hilbert transform is performed on each intrinsic mode function to obtain the Hilbert time spectrogram to analyze the signal.
  • IMFs intrinsic mode functions
  • the empirical mode decomposition method is the basis of the Hilbert-Huang transformation.
  • the empirical mode decomposition decomposes the current signal 31 into the sum of a finite number of intrinsic mode functions, wherein any two intrinsic mode functions are independent of each other, and the intrinsic mode functions satisfy the following two conditions:
  • the number of zero-crossing points is equal to the number of extreme points (maximum value or small value) or the difference is at most 1;
  • the empirical modal decomposition steps of the fault arc current signal are as follows:
  • m 1 (t) as the average value of the upper envelope and the lower envelope ((v 1 (t))+(v 2 (t)))/2, then h 1 ( t):
  • h 1 (t) if h 1 (t) meets the requirements of the intrinsic mode function, then h 1 (t) is the first component of x(t), otherwise h 1 (t) is repeated as the original data to formula (1) , until after repeating the cycle k, h 1k (t) meets the requirements of the natural mode function, denoted as c 1 (t) as the first component that satisfies the natural mode function conditions.
  • c i (t) contains the component changes in different frequency bands from high frequency to low frequency of the original signal, and the residual component represents the central trend of the signal.
  • Hilbert transform is performed on each intrinsic modal function component c i (t) respectively (block 37 in Figure 3), and its instantaneous amplitude, The instantaneous phase and instantaneous frequency, the algorithm process is as follows:
  • the instantaneous frequency can be further extracted:
  • Hilbert transform can be applied to the analysis of nonlinear and non-stationary signals, and can simultaneously characterize the distribution law of arc fault signal energy in time and frequency.
  • the Hilbert transform yields the corresponding Hilbert transform spectrum, ie each intrinsic mode function is represented in the joint time-frequency domain.
  • marginal spectral region variation features can be extracted as mutation features (block 38 of FIG. 3 ).
  • the extraction of the amplitude energy of the intrinsic mode function can be performed in parallel with the Hilbert-Huang transform (as shown in block 35 in Figure 3), and the normalized cumulative value of the mutation is calculated (as shown in Figure 3). 3) as a mutation feature.
  • the multivariate signature analysis 39 a predetermined number of mutation signatures can be extracted.
  • the arc fault current is transformed by Hilbert Huang, and the prominent change of the natural modal function when the arc fault occurs can be used as the basis for fault detection.
  • the basis for fault type diagnosis Considering that the arc fault is affected by uncertain factors such as the cause of the fault, the nature of the load, and the arcing conditions, as well as the incomplete and inaccurate waveform data that may occur due to hardware reasons during the data acquisition process, although from the inherent modal function
  • the characteristic information of the fault arc can be extracted, but due to the limitations of the empirical mode decomposition method, the number of intrinsic mode functions obtained by the decomposition of different signals and the frequency band of each intrinsic mode function are not fixed, so only a certain one is analyzed.
  • the present invention adopts the multivariable natural mode function components to represent the distortion of the signal, realizes the diagnosis of the fault, and weakens the influence of the variation of the electrical parameters of the signal on the detection accuracy.
  • load characteristics 40 and environmental parameters 41 are further provided as state inputs 42 together with a predetermined number of abrupt signatures output from the multivariate signature analysis 39 .
  • a dimensionless index 44 of the characteristic signal is further provided (which is obtained from the change feature of the marginal spectral region as a mutation feature and the normalized cumulative value of the mutation amount, for example, the dimensionless index is specifically the change characteristic of the marginal spectral region and the normalized mutation amount
  • the ratio of the accumulated values, or the product of the ratio and a predetermined coefficient, etc.) to form the arc fault diagnosis vector matrix 44 is input into the quantum neural network 60 .
  • the seven inputs shown by E1-E7 are collectively used as the diagnostic input of the quantum neural network.
  • the signal types of E1-E7 include load characteristics 40 , environmental parameters 41 and dimensionless indicators 44 .
  • the signal types of E1-E7 include load characteristics 40, environmental parameters 41, variation characteristics of marginal spectral regions, and normalized cumulative values of mutation amounts.
  • the specific number of any one of the load characteristic 40, the environmental parameter 41 and the dimensionless index 44 input into the quantum neural network may be plural.
  • the present invention utilizes the activation function of the quantum neuron in the quantum neural network with multiple quantum energy levels, and can adaptively determine the potential ambiguity in the sample data.
  • the decision rule is used as an identification model to solve the arc fault type.
  • the structure of the quantum neural network is shown in Figure 3. Among them, E1 to E7 are the input layers 46, which are extracted from the previous analysis, and can represent the relevant information of the fault arc in multiple scales. The second layer is the hidden layer 47 .
  • the excitation function of each node in the hidden layer of the network is the sigmoid function f, the number of discrete levels of each hidden layer node is n, the steepness factor is a, the connection weight from the input layer to the hidden layer is ⁇ ij , the threshold value of each node is is a j , the quantum interval is ⁇ jr , let ⁇ jH be the output of the jth node in the hidden layer:
  • the third layer is the output layer 48 .
  • the linear function of each node of this layer is g
  • the connection weight from the hidden layer to the output layer is v jk
  • the threshold b k of each node, then Ok O is the output of the kth node of the output layer:
  • the actual output values of OkO can be represented as y1 and y2.
  • a quantum neural network with linear superposition of multilayer activation functions has been established as an arc fault type identification model.
  • the actual output value of the quantum neural network, and the corresponding decoding and mapping rules of the binary code of the electric shock fault type include:
  • the embodiment of the present invention realizes the mapping relationship between the arc fault type and the arc current, which is difficult to express with an accurate mathematical model.
  • a multi-variable arc fault diagnosis method takes line current as the target of arc fault identification and fault line location, uses Hilbert-Huang time-frequency analysis to obtain characteristic components in each frequency band, and extracts characteristic components that can characterize arc faults before and after the occurrence of arc faults to construct characteristic waveforms according to the signal correlation principle.
  • a fault diagnosis model based on the multivariate eigenvector set of the eigensignal is constructed, and a variety of dimensionless indicators of the eigenwaveform are extracted as the input eigenvector set of the diagnosis network.
  • the threshold method detects that the arc is affected by the size of the load, the nature of the load and the state of the arc.
  • the potential laws and mapping relationships between various fault currents are found by applying the quantum neural network.
  • the hidden layer of the network uses quantum neurons with multiple quantum energy levels, which has high computational efficiency and can adaptively determine the sample.
  • the potential fuzzy decision-making law in the data can meet the actual requirements of fast and accurate action in fault arc protection technology.
  • an embodiment of the present invention provides an arc fault diagnosis device.
  • FIG. 4 is a configuration diagram of an arc fault diagnosis apparatus according to an embodiment of the present invention.
  • the diagnostic device 400 for arc fault includes:
  • the acquisition module 401 is used to acquire the current signal
  • an empirical mode decomposition module 402 configured to perform empirical mode decomposition on the current signal to obtain a plurality of intrinsic mode functions of the current signal
  • a mutation feature acquisition module 403 configured to acquire mutation features of multiple high-frequency components of the current signal based on the multiple intrinsic mode functions
  • the type determination module 404 is used for inputting the mutation features of the plurality of high-frequency components into the trained neural network model, so as to output the fault arc type from the neural network model.
  • the empirical mode decomposition module 402 is used to determine the local extreme point of the current signal; use an interpolation method to fit the local extreme point into an upper envelope and a lower envelope; An envelope and a lower envelope to separate the intrinsic mode functions from the current signal.
  • the mutation feature acquisition module 403 is configured to perform Hilbert transform on each intrinsic mode function to obtain the Hilbert spectrum of the intrinsic mode function; based on each intrinsic mode function
  • the Hilbert spectrum of the eigenmode function determines the marginal spectral region variation characteristics of the eigenmode function as abrupt features corresponding to the high-frequency components of the eigenmode function.
  • the mutation feature acquisition module 403 is configured to extract the amplitude energy of each intrinsic mode function; determine the mutation amount of the amplitude energy of each intrinsic mode function; determine the normalization value of the mutation amount The normalized accumulated value is used as the abrupt characteristic of the high frequency component corresponding to the natural mode function.
  • the mutation feature acquisition module 403 is configured to perform Hilbert transform on each intrinsic mode function to obtain a Hilbert spectrum of the intrinsic mode function, based on each intrinsic mode function
  • the Hilbert spectrum of the eigenmode function determines the variation characteristics of the marginal spectral region of the eigenmode function; extracts the amplitude energy of each eigenmode function, determines the mutation amount of the amplitude energy of each eigenmode function, and determines each eigenmode function.
  • each eigenmode is determined based on the variation characteristics of the marginal spectral region of each eigenmode function and the normalized cumulative value of the mutation amount of the eigenmode function
  • the dimensionless index of the function is determined as a mutation characteristic corresponding to the high frequency component of the natural mode function.
  • the type determination module 404 is further configured to, in the training phase of the neural network model, further input the load characteristic as the training data, the environmental parameter as the training data, and the mutation characteristic as the high-frequency component of the training data into the input data.
  • the neural network model is used to train the neural network model; in the application stage of the neural network model, the environmental parameters of the current signal collection point and the load characteristics of the current signal collection point are further input into the neural network model.
  • the embodiments of the present invention also provide an arc fault diagnosis device with a memory-processor architecture.
  • FIG. 5 is an exemplary structural block diagram of an arc fault diagnosis apparatus with a memory-processor architecture according to an embodiment of the present invention.
  • an arc fault diagnosis device 500 includes a processor 501 , a memory 502 and a computer program stored in the memory 502 and executable on the processor 501 , and the computer program is implemented when the processor 501 executes it.
  • the method for diagnosing an arc fault according to any one of the above.
  • the memory 502 can be specifically implemented as various storage media such as Electrically Erasable Programmable Read-Only Memory (EEPROM), Flash Memory (Flash memory), Programmable Program Read-Only Memory (PROM).
  • the processor 501 may be implemented to include one or more central processing units or one or more field programmable gate arrays, wherein the field programmable gate arrays integrate one or more central processing unit cores.
  • the central processing unit or the central processing unit core may be implemented as a CPU or an MCU or a DSP or the like.
  • a module can be implemented by multiple modules, and the functions of multiple modules can also be implemented by the same module. These modules can be located in the same device. , or in a different device.
  • the hardware modules in various embodiments may be implemented mechanically or electronically.
  • a hardware module may include specially designed permanent circuits or logic devices (eg, special purpose processors, such as FPGAs or ASICs) for performing specific operations.
  • Hardware modules may also include programmable logic devices or circuits (eg, including general-purpose processors or other programmable processors) temporarily configured by software for performing particular operations.
  • programmable logic devices or circuits eg, including general-purpose processors or other programmable processors
  • temporarily configured circuit for example, configured by software
  • the present invention also provides a machine-readable storage medium storing instructions for causing a machine to perform the method as described herein.
  • a system or device equipped with a storage medium on which software program codes for realizing the functions of any one of the above-described embodiments are stored, and make the computer (or CPU or MPU of the system or device) ) to read and execute the program code stored in the storage medium.
  • a part or all of the actual operation can also be completed by an operating system or the like operating on the computer based on the instructions of the program code.
  • the program code read out from the storage medium can also be written into the memory provided in the expansion board inserted into the computer or into the memory provided in the expansion unit connected to the computer, and then the instructions based on the program code cause the device to be installed in the computer.
  • the CPU on the expansion board or the expansion unit or the like performs part and all of the actual operations, thereby realizing the functions of any one of the above-mentioned embodiments.
  • Embodiments of storage media for providing program code include floppy disks, hard disks, magneto-optical disks, optical disks (eg, CD-ROM, CD-R, CD-RW, DVD-ROM, DVD-RAM, DVD-RW, DVD+RW), Magnetic tapes, non-volatile memory cards and ROMs.
  • the program code may be downloaded from a server computer or cloud over a communications network.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Computational Linguistics (AREA)
  • General Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Testing And Monitoring For Control Systems (AREA)

Abstract

一种故障电弧的诊断方法，包括：采集电流信号（101）；对电流信号执行经验模式分解，以获取电流信号的多个固有模态函数（102）；基于多个固有模态函数获取电流信号的多个高频分量的突变特征（103）；将多个高频分量的突变特征输入已训练的神经网络模型，以由神经网络模型输出故障电弧类型（104）。还包括故障电弧的诊断装置和计算机可读存储介质。实现了故障电弧多变量诊断，能够较早地检测到故障电弧的发生，并能够区分不同类型的故障电弧，满足快速准确动作的实际要求。

Description

一种故障电弧的诊断方法、装置和计算机可读存储介质 技术领域
本发明涉及供配电技术领域，特别是涉及一种故障电弧的诊断方法、装置和计算机可读存储介质。
背景技术
随着国民经济的高速发展，电力消费急剧增长。电及电气设备已经成为日常生活中不可缺少的一部分。各种电器设备不断推陈出新。这一系列发展提高了人们的生活质量，同时也导致我们的生活空间布满各种安全隐患，若不能及时排查安全隐患，会导致故障线路继续运行。电弧故障为气火灾的罪魁祸首。当线路过载、短路以及接触不良等故障存在时，都能导致故障点的温度持续上升，继而引发电弧的燃烧温度，温度往往高达2000℃或者3000℃，若故障点附近存在可燃物体时，便会引发火灾的风险。
通过仿真、实验等手段对故障电弧进行研究，发现故障电弧的电气特性与负载性质、发生位置、故障类型有关，且具间歇性、随机性和不确定性。因此，如何有效检测故障电弧一直是电弧领域研究的热点。
目前，常用的故障电弧检测方法大多是基于电弧的物理现象和电气特性。利用弧声、弧光、电磁辐射等电弧物理特性检测方法在一些特定的场合得到较多应用(如开关柜故障电弧检测)。然而受故障电弧发生位置、传感器检测精度、环境干扰等限制，该方法难以推广到配电终端进行保护。因此，基于电弧电压、电流的时域、频域、时频域特性获取故障电弧征参数，研究其时域参数、傅里叶变换、小波分析等故障电弧检测方法是目前故障电弧检测的常用方法。
在现有技术中，单纯依赖单变量特征阈值诊断故障电弧。然而，单变量诊断容易出现特征阈值波动、正常状态与故障信息交叉重复等问题，从而导致误判或漏判。
发明内容
本发明实施方式提出一种故障电弧的诊断方法、装置和计算机可读存储介质。
本发明实施方式的技术方案如下：
一种故障电弧的诊断方法，包括：
采集电流信号；
对所述电流信号执行经验模式分解，以获取所述电流信号的多个固有模态函数；
基于所述多个固有模态函数获取所述电流信号的多个高频分量的突变特征；
将所述多个高频分量的突变特征输入已训练的神经网络模型，以由所述神经网络模型输出故障电弧类型。
可见，不同于现有技术中单变量的故障电弧诊断，本发明实施方式实现了多变量的故障电弧诊断，可以克服单变量故障电弧诊断中阈值选择困难、易受故障原因、燃弧工况、负载性质等影响的技术缺陷，提 高诊断准确率。
另外，多变量输入的神经网络便于找到多种故障电流之间的潜在规律及映射关系，可自适应地确定样本数据中潜在的模糊决策规律，能够满足故障电弧保护技术中快速准确动作的实际要求。
在一个实施方式中，所述对采集的电流信号执行经验模式分解，以获取所述电流信号的多个固有模态函数包括：
确定所述电流信号的局部极值点；
利用插值方法将所述局部极值点拟合为上包络线和下包络线；
基于所述上包络线和所述下包络线，从所述电流信号分离出多个固有模态函数。
因此，通过对采集的电流信号执行经验模式分解，可以获得多个固有模态函数，为后续的多变量诊断提供了应用基础。
在一个实施方式中，所述插值方法为三次样条插值方法，所述神经网络模型为量子神经网络模型。
因此，采用三次样条插值方法以及具有多个量子能级的量子神经元的量子神经网络模型，可以提高运算效率。
在一个实施方式中，所述基于多个固有模态函数获取所述电流信号的多个高频分量的突变特征包括：
对每个固有模态函数执行希尔伯特变换，得到该固有模态函数的希尔伯特谱；
基于每个固有模态函数的希尔伯特谱确定该固有模态函数的边际谱区域变化特征，以作为对应于该固有模态函数的高频分量的突变特征。
因此，可以利用希尔伯特变换快速确定各个频段下的特征分量。
在一个实施方式中，所述基于多个固有模态函数获取所述电流信号的多个高频分量的突变特征包括：
提取每个固有模态函数的幅值能量；
确定每个固有模态函数的幅值能量的突变量；
确定每个固有模态函数的所述突变量的归一化累计值，以作为对应于该固有模态函数的高频分量的突变特征。
因此，基于直接计算固有模态函数的幅值能量的突变量，可以确定高频分量的突变特征。
在一个实施方式中，所述基于固有模态函数获取所述电流信号的多个高频分量的突变特征每个固有模态函数的包括：
对每个固有模态函数执行希尔伯特变换，得到该固有模态函数的希尔伯特谱，基于每个固有模态函数的希尔伯特谱确定该固有模态函数的边际谱区域变化特征；
提取每个固有模态函数的幅值能量，确定每个固有模态函数的幅值能量的突变量，确定每个固有模态函数的所述突变量的归一化累计值；
基于每个固有模态函数的边际谱区域变化特征以及该固有模态函数的突变量的归一化累计值，确定所 述每个固有模态函数的无量纲指标，将所述无量纲指标确定为对应于该固有模态函数的高频分量的突变特征。
可见，通过提取多种无量纲指标作为诊断网络的输入特征向量集，克服了基于传统阈值方法检测电弧受负载大小、负载性质和燃弧状态影响的问题。
在一个实施方式中，该方法还包括：
在所述神经网络模型的训练阶段，进一步将作为训练数据的负载特性、作为训练数据的环境参量和作为训练数据的高频分量的突变特征输入所述神经网络模型以训练所述神经网络模型；
在所述神经网络模型的应用阶段，进一步将电流信号采集点的环境参量和电流信号采集点的负载特性输入所述神经网络模型。
因此，通过负载特性和环境参量训练神经网络模型，有助于量子神经网络找到多种故障电流之间存在的潜在规律及映射关系，可自适应地确定样本数据中潜在的模糊决策规律。
一种故障电弧的诊断装置，包括：
采集模块，用于采集电流信号；
经验模式分解模块，用于对所述电流信号执行经验模式分解，以获取所述电流信号的多个固有模态函数；
突变特征获取模块，用于基于所述多个固有模态函数获取所述电流信号的多个高频分量的突变特征；
类型确定模块，用于将所述多个高频分量的突变特征输入已训练的神经网络模型，以由所述神经网络模型输出故障电弧类型。
可见，不同于现有技术中单变量的故障电弧诊断，本发明实施方式实现了多变量的故障电弧诊断，可以克服单变量故障电弧诊断中阈值选择困难、易受故障原因、燃弧工况、负载性质等影响的技术缺陷，提高诊断准确率。
在一个实施方式中，所述经验模式分解模块，用于确定所述电流信号的局部极值点；利用插值方法将所述局部极值点拟合为上包络线和下包络线；基于所述上包络线和所述下包络线，从所述电流信号分离出固有模态函数。
因此，通过对采集的电流信号执行经验模式分解，可以获得多个固有模态函数，为后续的多变量诊断提供了应用基础。
在一个实施方式中，所述突变特征获取模块，用于对每个固有模态函数执行希尔伯特变换，得到该固有模态函数的希尔伯特谱，基于每个固有模态函数的希尔伯特谱确定该固有模态函数的边际谱区域变化特征，以作为对应于该固有模态函数的高频分量的突变特征。
因此，可以利用希尔伯特变换快速确定各个频段下的特征分量。
在一个实施方式中，所述突变特征获取模块，用于提取每个固有模态函数的幅值能量，确定每个固有 模态函数的幅值能量的突变量，确定所述突变量的归一化累计值，以作为所述对应于该固有模态函数的高频分量的突变特征。
因此，基于直接计算固有模态函数的幅值能量的突变量，可以确定高频分量的突变特征。
在一个实施方式中，所述突变特征获取模块，用于对每个固有模态函数执行希尔伯特变换，得到该固有模态函数的希尔伯特谱，基于每个固有模态函数的希尔伯特谱确定该固有模态函数的边际谱区域变化特征；提取每个固有模态函数的幅值能量，确定每个固有模态函数的幅值能量的突变量，确定每个固有模态函数的所述突变量的归一化累计值；基于每个固有模态函数的边际谱区域变化特征以及该固有模态函数的突变量的归一化累计值，确定每个固有模态函数的无量纲指标，将所述无量纲指标确定为对应于该固有模态函数的高频分量的突变特征。
可见，通过提取多种无量纲指标作为诊断网络的输入特征向量集，克服了基于传统阈值方法检测电弧受负载大小、负载性质和燃弧状态影响的问题。
在一个实施方式中，类型确定模块，还用于在神经网络模型的训练阶段，进一步将作为训练数据的负载特性、作为训练数据的环境参量和作为训练数据的高频分量的突变特征输入所述神经网络模型以训练所述神经网络模型；在所述神经网络模型的应用阶段，进一步将电流信号采集点的环境参量和电流信号采集点的负载特性输入所述神经网络模型。
因此，通过负载特性和环境参量训练神经网络模型，有助于量子神经网络找到多种故障电流之间存在的潜在规律及映射关系，可自适应地确定样本数据中潜在的模糊决策规律。
一种故障电弧的诊断装置，包括处理器、存储器及存储在所述存储器上并可在所述处理器上运行的计算机程序，所述计算机程序被所述处理器执行时实现如上任一项所述的故障电弧的诊断方法。
一种计算机可读存储介质，所述计算机可读存储介质上存储计算机程序，所述计算机程序被处理器执行时实现如上任一项所述的故障电弧的诊断方法。
附图说明
图1为本发明实施方式的故障电弧的诊断方法的流程图。
图2为本发明实施方式的电流采集示意图。
图3为本发明实施方式的故障电弧的诊断过程的示范性示意图。
图4为本发明实施方式的故障电弧的诊断装置的结构图。
图5为本发明实施方式的具有存储器-处理器架构的故障电弧的诊断装置的示范性结构框图。
其中，附图标记如下：
标号 含义
100 故障电弧的诊断方法
101～104 步骤
51 商业用电负荷
52 辅助接入装置
53 工业用电负荷
54 配电房
55 电流传感器
56 电网
31 电流信号
32 确定局部极值点
33 拟合上、下包络线
34 经验模式分解(EMD)
35 提取固有模态函数(IMF)幅值能量
36 计算突变量归一化累计值
37 固有模态函数的希尔伯特变换
38 提取边际谱区域变化特征
39 多变量特征分析
40 负载特性
41 环境参量
42 状态输入
43 特征信号的无量纲指标
44 电弧故障诊断向量矩阵
45 能量特征提取
46 输入层
47 隐含层
48 输出层
49 网络输出
60 量子神经网络
500 故障电弧的诊断装置
501 处理器
502 存储器
具体实施方式
为了使本发明的技术方案及优点更加清楚明白，以下结合附图及实施方式，对本发明进行进一步详细说明。应当理解，此处所描述的具体实施方式仅仅用以阐述性说明本发明，并不用于限定本发明的保护范 围。
为了描述上的简洁和直观，下文通过描述若干代表性的实施方式来对本发明的方案进行阐述。实施方式中大量的细节仅用于帮助理解本发明的方案。但是很明显，本发明的技术方案实现时可以不局限于这些细节。为了避免不必要地模糊了本发明的方案，一些实施方式没有进行细致地描述，而是仅给出了框架。下文中，“包括”是指“包括但不限于”，“根据……”是指“至少根据……，但不限于仅根据……”。由于汉语的语言习惯，下文中没有特别指出一个成分的数量时，意味着该成分可以是一个也可以是多个，或可理解为至少一个。
申请人发现：目前故障电弧的检测和识别由于电力系统的运行环境、负载多样性、系统复杂性，在对其进行准确判断的过程中遇到了技术瓶颈，亟待融入更为先进的技术理论，开发和研制一种能够有效诊断和识别故障电弧的新方法。可以考虑结合智能计算、自适应控制算法、迷糊控制理论和逻辑策略对故障电弧进行识别。但是，由于供配电系统中存在与故障电弧相似的波形，如何在复杂环境下准确地识别故障电弧、开关电器正常的操作电弧和其他干扰信号是故障检测的难点。同时，故障电弧产生的环境复杂且具有不可预见性，受传感器选型、硬件检测电路、负载性质和容量、工作环境等影响，采集到的电压、电流波形往存在不完整、不精确、模糊、随机性等问题。
有鉴于此，申请人一种基于希尔伯特-黄(Hilbert-Huang)变换和神经网络模型(优选为量子神经网络模型)的故障电弧电流诊断方法，在实现对燃弧前后电流信号时频分析的基础上，明确了故障电弧电流暂态过程频谱特性的变化规律，构建燃弧前后的特征信号，利用量子神经网络中量子神经元具有多个量子能级的激活函数，自适应地确定样本数据中潜在的模糊决策规律，提取故障电弧电流所具有的不确定的潜在规律及关系映射，从而解决现有的故障电弧的单变量判据方法中阈值难以准确确定的问题。
故障电弧的诊断实质上是一个模式识别问题。本发明主要包括三个环节：故障电弧电流信号采集与经验模式分解(EMD)、特征提取与状态识别以及故障电弧辨识，其关键在于如何有效提取各故障电弧的特征。在独立完成特征量提取的基础上，结合量子神经网络人工智能算法，使整个方案具有学习并适应未知或不确定系统的能力，在不同运行环境、不同负载模式下，实现对多类故障状态的诊断。
图1为本发明实施方式的故障电弧的诊断方法的流程图。
如图1所示，该方法包括：
步骤101：采集电流信号。
在这里，可以在供配电系统中可能存在故障电弧的任意位置处布置电流传感器，以采集电流信号。
图2为本发明实施方式的电流采集示意图。由图2可见，商业用电负荷51经由辅助接入装置52连接到配电房54。工业用电负荷53连接到配电房54。配电房54与电网56连接。
其中：可以在配电房54与电网56之间的变压器周边部署电流传感器。示范性地，还可以在商业用电负荷51、辅助接入装置52、工业用电负荷53和配电房54等处部署电流传感器。
以上示范性描述了电流传感器的部署位置，本领域技术人员可以意识到，这种描述仅是示范性的，并不用于限定本发明实施方式的保护范围。
步骤102：对所述电流信号执行经验模式分解，以获取所述电流信号的多个固有模态函数。
在这里，对采集的电流信号执行经验模式分解，以获取所述电流信号的固有模态函数包括：确定电流信号的局部极值点；利用插值方法将所述局部极值点拟合为上包络线和下包络线；基于上包络线和下包络线，从所述电流信号分离出固有模态函数。优选的，所述插值方法为三次样条插值方法。
步骤103：基于所述多个固有模态函数获取所述电流信号的多个高频分量的突变特征。
在一个实施方式中，步骤103中基于多个固有模态函数获取所述电流信号的多个高频分量的突变特征包括：对每个固有模态函数执行希尔伯特变换，得到该固有模态函数的希尔伯特谱；基于每个固有模态函数的希尔伯特谱确定该固有模态函数的边际谱区域变化特征，以作为对应于该固有模态函数的高频分量的突变特征。可见，可以利用希尔伯特变换快速确定各个频段下的特征分量。
在一个实施方式中，步骤103中基于多个固有模态函数获取所述电流信号的多个高频分量的突变特征包括：提取每个固有模态函数的幅值能量；确定每个固有模态函数的幅值能量的突变量；确定每个固有模态函数的所述突变量的归一化累计值，以作为对应于该固有模态函数的高频分量的突变特征。因此，基于直接计算固有模态函数的幅值能量的突变量，可以确定高频分量的突变特征。
在一个实施方式中，步骤103基于固有模态函数获取所述电流信号的多个高频分量的突变特征包括：对每个固有模态函数执行希尔伯特变换，得到该固有模态函数的希尔伯特谱，基于每个固有模态函数的希尔伯特谱确定该固有模态函数的边际谱区域变化特征；提取每个固有模态函数的幅值能量，确定每个固有模态函数的幅值能量的突变量，确定每个固有模态函数的所述突变量的归一化累计值；基于每个固有模态函数的边际谱区域变化特征以及该固有模态函数的突变量的归一化累计值，确定所述每个固有模态函数的无量纲指标，将所述无量纲指标确定为对应于该固有模态函数的高频分量的突变特征。可见，通过提取多种无量纲指标作为诊断网络的输入特征向量集，克服了基于传统阈值方法检测电弧受负载大小、负载性质和燃弧状态影响的问题。
步骤104：将所述多个高频分量的突变特征输入已训练的神经网络模型，以由所述神经网络模型输出故障电弧类型。
优选的，所述神经网络模型为量子神经网络(QNN)模型。量子神经网络由若干个量子神经元按一定的拓扑结构构成，它是试图将人工神经网络模型与量子信息优势相结合的一种方法。
在一个实施方式中，该方法还包括：在所述神经网络模型的训练阶段，进一步将作为训练数据的负载特性、作为训练数据的环境参量和作为训练数据的高频分量的突变特征输入所述神经网络模型以训练所述神经网络模型；在所述神经网络模型的应用阶段，进一步将电流信号采集点的环境参量和电流信号采集点的负载特性输入所述神经网络模型。
因此，通过负载特性和环境参量训练神经网络模型，有助于量子神经网络找到多种故障电流之间存在的潜在规律及映射关系，可自适应地确定样本数据中潜在的模糊决策规律。
下面以量子神经网络为例，描述本发明实施方式的故障电弧诊断过程。
图3为本发明实施方式的故障电弧的诊断过程的示意图。
在图3中，整个系统本质上可以看作一个闭环网络，具备三个信号处理环节。故障电弧电流信号采集与经验模式分解为整体检测过程提供源信号和特征量，并对初始信号进行滤波处理，也是算法启动运行的第一步。其输入为传感器的电流信号，实时表征所测电路的动态特征，其输出为故障电弧状态特征提取环节，以此提取能够准确指示电弧故障的多变量指标，形成故障诊断向量矩阵。故障电弧状态特征提取环节接收来自故障电弧电流信号采集与经验模式分解环节的时频域特征信息，形成特征向量矩阵样本集，输入至故障电弧的量子神经网络，以供其进行训练学习，通过配置合适的平滑参数，不断完善并构建故障诊断模型。最后，方案的诊断决策由故障电弧量子神经网络模型环节输出，识别出当前采集到电流波形信号的故障类型。
系统工作流程的三个环节中均蕴含了时频域多尺度分析以及自学习的控制思路，改变了现有检测方案仅用单一频域或时域分析方法提取现场获取的非平稳故障电流信号幅值故障特征的局限，系统中各个环节的工作原理如下：
(1)、基于Hilbert-Huang变换电弧电流时频特征提取：
Hilbert-Huang变换是一种非线性非平稳信号分析的新方法，其过程主要由2部分组成：经验模态分解方法和希尔伯特谱分析。首先用经验模态分解将电流信号31分解为一系列合理数目的固有模态函数(IMF)的线性和，然后对每个固有模态函数进行希尔伯特变换得到希尔伯特时频谱图来分析信号。
经验模态分解方法是Hilbert-Huang变换的基础。经验模态分解将电流信号31分解为有限个固有模态函数之和，其中任何2个固有模态函数之间是相互独立，固有模态函数满足以下2个条件：
①、在整个数据中，过零点数目与极值点(极大值或小值)数目相等或至多相差为1；
②、信号上任意一点，由局部极大值构成的上络线与由局部极小值构成的下包络线均值为0。
因此。根据固有模态函数的定义，故障电弧电流信号的经验模态分解步骤如下：
(a)、确定原始信号x(t)所具有的全部局部极值点(如图3的方框32)，然后利用3次样条插值方法将所有的局部极大值点拟合为该信号的上包络线v 1(t)，通过局部极小值点拟合得到下包络线v 2(t)；(如图3的方框33)。
(b)、定义m 1(t)为上包络线和下包络线的平均值((v 1(t))+(v 2(t)))/2，即可计算得到h 1(t)：
h 1(t)＝x(t)-m 1(t)；
(c)、如果h 1(t)满足固有模态函数要求，则h 1(t)是x(t)的第1个分量，否则将h 1(t)作为原始数据重复到式(1)，直到重复循环k后，h 1k(t)满固有模态函数的要求，记为c 1(t)为第1个满足固有模态函数条件的 分量。
(d)、从原信号x(t)中分离出满足固有模态函数条件c 1(t)，可得r 1(t)。
r 1(t)＝x(t)-c 1(t)；
将r 1(t)作为原始数据重复，其中重复n次可得到原信号x(t)的n个满足固有模态函数要求的分量。当残余分量r n(t)成为一个单调函数，且不能再从中提取新分量时，分解结束。此时，给定原信号x(t)可以表示为：
Figure PCTCN2020119020-appb-000001
经验模态分解(如图3的方框34)后，c i(t)包含了随原信号从高频到低频不同频率段的成分变化，残余分量则表示了该信号的中心趋势。原信号x(t)经经验模态分解后，分别对每一个固有模态函数分量c i(t)进行希尔伯特变换(如图3的方框37)，计算得到其瞬时幅值、瞬时相位和瞬时频率，算法过程如下：
对于每一个c i(n)(i＝1，…，n)进行希尔伯特变换得到：
Figure PCTCN2020119020-appb-000002
通过构造一个解析信号a i(t)exp(jθ(t))，计算得到幅值函数的振幅a i(t)：
Figure PCTCN2020119020-appb-000003
进一步可提取出瞬时频率：
Figure PCTCN2020119020-appb-000004
希尔伯特变换能适用于非线性非平稳信号的分析，可同时表征故障电弧信号能量在时间和频率上分布规律。希尔伯特变换得到相应希尔伯特变换谱，即将每个固有模态函数表示在联合的时频域中。基于希尔伯特变换谱，可以提取边际谱区域变化特征以作为突变特征(如图3的方框38)。而且，经验模态分解后，还可以与希尔伯特黄变换并行地执行提取固有模态函数幅值能量(如图3的方框35)，并计算突变量归一化累计值(如图3的方框36)以作为突变特征。在多变量特征分析39中，可以提取预定数目的突变特征。
(2)、故障电弧状态特征提取：
故障电弧电流经希尔伯特黄变换，在电弧故障发生时固有模态函数的突出变化，可作为故障检测的依据，并利用每个固有模态函数中各幅值占有率和相关性系数作为故障类型诊断的依据。考虑到故障电弧受故障原因、负载性质、燃弧工况等不确定因素的影响以及在数据采集过程中因硬件原因可能出现的波形数据不完整、不精确等问题，虽然从固有模态函数中可以提取故障电弧特征信息，但是由经验模态分解方法的局限性，对于不同信号其分解得到的固有模态函数数目和每个固有模态函数的频段并非是固定的，因此 仅仅分析某1个或几个固有模态函数作为故障识别的依据，可能导致诊断阈值区分度小，诊断准确率降低。因此，本发明采用多变量的固有模态函数分量表征信号的畸变情况，实现对故障的诊断，减弱信号电参量变化对检测精度的影响。
优选地，进一步提供负载特性40和环境参量41，与多变量特征分析39输出的、预定数目的突变特征共同作为状态输入42。而且，进一步提供特征信号的无量纲指标44(由边际谱区域变化特征以作为突变特征与突变量归一化累计值共同得到，比如无量纲指标具体为边际谱区域变化特征与突变量归一化累计值的比值，或该比值与预定系数的乘积，等等)，以形成电弧故障诊断向量矩阵44输入到量子神经网络60中。如图3所示，由E1-E7所示的7个输入共同作为量子神经网络的诊断输入。优选地，E1-E7的信号种类包含负载特性40、环境参量41和无量纲指标44。可选地，E1-E7的信号种类包含负载特性40、环境参量41、边际谱区域变化特征与突变量归一化累计值。而且，输入到量子神经网络中的负载特性40、环境参量41和无量纲指标44的任一种的具体数目可以为多个。
(3)、基于量子神经网络的触电故障类型识别
由于电弧故障类型与电弧电流之间映射关系，难以用精确数学模型表达，本发明利用量子神经网络中量子神经元具有多个量子能级的激活函数，可自适应地确定样本数据中潜在的模糊决策规律，将其作为一种解决电弧故障类型识别模型。量子神经网络的结构如图3中所示。其中E1～E7为输入层46，由前面分析提取而来，可在多尺度表征故障电弧的相关信息。第二层为隐含层47。该网络隐含层各节点的激励函数为sigmoid函数f，每个隐含层节点的离散级别数为n，陡度因子为a，输入层到隐含层的连接加权为ω ij，各节点阈值为a j，量子间隔为θ jr，令θ jH为隐含层第j个节点的输出为：
Figure PCTCN2020119020-appb-000005
式中
Figure PCTCN2020119020-appb-000006
第三层为输出层48。该层各节点的线性函数为g，隐含层到输出层的连接加权为v jk，各节点的阈值b k，则Ok O为输出层第k节点的输出为：
Figure PCTCN2020119020-appb-000007
当输出层48具有2个节点时，Ok O各实际输出值可表示为y1和y2。至此建立了一种多层激活函数线性叠加的量子神经网络，作为电弧故障类型识别模型。量子神经网络的实际输出值，所对应触电故障类型二进制编码的解码映射规则包括：
(1)、当网络输出y1≥y2时，对应电弧故障类型编码为10，表示特殊电弧故障；
(2)、当网络输出y1＜y2时，对应电弧故障类型编码为01，表示常见介质电弧故障。
因此，本发明实施方式实现了电弧故障类型与电弧电流之间，难以用精确数学模型表达的映射关系。
可见，在本发明实施方式至少具有如下优点：
(1)、针对基于单变量的故障电弧诊断方法中阈值选择困难，易受故障原因、燃弧工况、负载性质等影响的问题，提出一种故障电弧多变量诊断方法。该方法以线路电流为故障电弧识别和故障线路定位的目标，利用Hilbert-Huang时频分析获取各个频段下特征分量，并根据信号相关原理提取能够表征电弧故障发生前后的特征分量构建特征波形。
(2)、为实现对故障线路的准确定位，构建了基于特征信号多变量特征向量集的故障诊断模型，提取特征波形的多种无量纲指标作为诊断网络的输入特征向量集，克服了基于传统阈值方法检测电弧受负载大小、负载性质和燃弧状态影响的问题。
(3)、应用量子神经网络找到了多种故障电流之间存在的潜在规律及映射关系，该网络隐含层采用多个量子能级的量子神经元，运算效率高，可自适应地确定样本数据中潜在的模糊决策规律，能够满足故障电弧保护技术中快速准确动作的实际要求。
基于上述描述，本发明实施方式提出了一种故障电弧的诊断装置。
图4为本发明实施方式的故障电弧的诊断装置的结构图。
如图4所示，故障电弧的诊断装置400，包括：
采集模块401，用于采集电流信号；
经验模式分解模块402，用于对所述电流信号执行经验模式分解，以获取所述电流信号的多个固有模态函数；
突变特征获取模块403，用于基于所述多个固有模态函数获取所述电流信号的多个高频分量的突变特征；
类型确定模块404，用于将所述多个高频分量的突变特征输入已训练的神经网络模型，以由所述神经网络模型输出故障电弧类型。
在一个实施方式中，所述经验模式分解模块402，用于确定电流信号的局部极值点；利用插值方法将所述局部极值点拟合为上包络线和下包络线；基于上包络线和下包络线，从所述电流信号分离出固有模态函数。
在一个实施方式中，所述突变特征获取模块403，用于对每个固有模态函数执行希尔伯特变换，得到该固有模态函数的希尔伯特谱；基于每个固有模态函数的希尔伯特谱确定该固有模态函数的边际谱区域变化特征，以作为对应于该固有模态函数的高频分量的突变特征。
在一个实施方式中，所述突变特征获取模块403，用于提取每个固有模态函数的幅值能量；确定每个固有模态函数的幅值能量的突变量；确定所述突变量的归一化累计值，以作为所述对应于该固有模态函数的高频分量的突变特征。
在一个实施方式中，所述突变特征获取模块403，用于对每个固有模态函数执行希尔伯特变换，得到 该固有模态函数的希尔伯特谱，基于每个固有模态函数的希尔伯特谱确定该固有模态函数的边际谱区域变化特征；提取每个固有模态函数的幅值能量，确定每个固有模态函数的幅值能量的突变量，确定每个固有模态函数的所述突变量的归一化累计值；基于每个固有模态函数的边际谱区域变化特征以及该固有模态函数的突变量的归一化累计值，确定每个固有模态函数的无量纲指标，将所述无量纲指标确定为对应于该固有模态函数的高频分量的突变特征。
在一个实施方式中，类型确定模块404，还用于在神经网络模型的训练阶段，进一步将作为训练数据的负载特性、作为训练数据的环境参量和作为训练数据的高频分量的突变特征输入所述神经网络模型以训练所述神经网络模型；在所述神经网络模型的应用阶段，进一步将电流信号采集点的环境参量和电流信号采集点的负载特性输入所述神经网络模型。
基于上述描述，本发明实施方式还提出了具有存储器-处理器架构的故障电弧的诊断装置。
图5为本发明实施方式的具有存储器-处理器架构的故障电弧的诊断装置的示范性结构框图。
如图5所示，故障电弧的诊断装置500包括处理器501、存储器502及存储在存储器502上并可在处理器501上运行的计算机程序，所述计算机程序被所述处理器501执行时实现如上任一项所述故障电弧的诊断装置方法。其中，存储器502具体可以实施为电可擦可编程只读存储器(EEPROM)、快闪存储器(Flash memory)、可编程程序只读存储器(PROM)等多种存储介质。处理器501可以实施为包括一或多个中央处理器或一或多个现场可编程门阵列，其中现场可编程门阵列集成一或多个中央处理器核。具体地，中央处理器或中央处理器核可以实施为CPU或MCU或DSP等等。
需要说明的是，上述各流程和各结构图中不是所有的步骤和模块都是必须的，可以根据实际的需要忽略某些步骤或模块。各步骤的执行顺序不是固定的，可以根据需要进行调整。各模块的划分仅仅是为了便于描述采用的功能上的划分，实际实现时，一个模块可以分由多个模块实现，多个模块的功能也可以由同一个模块实现，这些模块可以位于同一个设备中，也可以位于不同的设备中。各实施方式中的硬件模块可以以机械方式或电子方式实现。例如，一个硬件模块可以包括专门设计的永久性电路或逻辑器件(如专用处理器，如FPGA或ASIC)用于完成特定的操作。硬件模块也可以包括由软件临时配置的可编程逻辑器件或电路(如包括通用处理器或其它可编程处理器)用于执行特定操作。至于具体采用机械方式，或是采用专用的永久性电路，或是采用临时配置的电路(如由软件进行配置)来实现硬件模块，可以根据成本和时间上的考虑来决定。
本发明还提供了一种机器可读的存储介质，存储用于使一机器执行如本申请所述方法的指令。具体地，可以提供配有存储介质的系统或者装置，在该存储介质上存储着实现上述实施例中任一实施方式的功能的软件程序代码，且使该系统或者装置的计算机(或CPU或MPU)读出并执行存储在存储介质中的程序代码。此外，还可以通过基于程序代码的指令使计算机上操作的操作系统等来完成部分或者全部的实际操作。还可以将从存储介质读出的程序代码写到插入计算机内的扩展板中所设置的存储器中或者写到与计算机 相连接的扩展单元中设置的存储器中，随后基于程序代码的指令使安装在扩展板或者扩展单元上的CPU等来执行部分和全部实际操作，从而实现上述实施方式中任一实施方式的功能。用于提供程序代码的存储介质实施方式包括软盘、硬盘、磁光盘、光盘(如CD-ROM、CD-R、CD-RW、DVD-ROM、DVD-RAM、DVD-RW、DVD+RW)、磁带、非易失性存储卡和ROM。可选择地，可以由通信网络从服务器计算机或云上下载程序代码。
在本文中，“示意性”表示“充当实例、例子或说明”，不应将在本文中被描述为“示意性”的任何图示、实施方式解释为一种更优选的或更具优点的技术方案。为使图面简洁，各图中的只示意性地表示出了与本发明相关部分，而并不代表其作为产品的实际结构。另外，以使图面简洁便于理解，在有些图中具有相同结构或功能的部件，仅示意性地绘示了其中的一个，或仅标出了其中的一个。在本文中，“一个”并不表示将本发明相关部分的数量限制为“仅此一个”，并且“一个”不表示排除本发明相关部分的数量“多于一个”的情形。在本文中，“上”、“下”、“前”、“后”、“左”、“右”、“内”、“外”等仅用于表示相关部分之间的相对位置关系，而非限定这些相关部分的绝对位置。
以上所述，仅为本发明的较佳实施例而已，并非用于限定本发明的保护范围。凡在本发明的精神和原则之内，所作的任何修改、等同替换、改进等，均应包含在本发明的保护范围之内。

Claims (15)

  1. 一种故障电弧的诊断方法(100)，其特征在于，包括：
    采集电流信号(101)；
    对所述电流信号执行经验模式分解，以获取所述电流信号的多个固有模态函数(102)；
    基于所述多个固有模态函数获取所述电流信号的多个高频分量的突变特征(103)；
    将所述多个高频分量的突变特征输入已训练的神经网络模型，以由所述神经网络模型输出故障电弧类型(104)。
  2. 根据权利要求1所述的故障电弧的诊断方法(100)，其特征在于，所述对采集的电流信号执行经验模式分解，以获取所述电流信号的多个固有模态函数(102)包括：
    确定所述电流信号的局部极值点；
    利用插值方法将所述局部极值点拟合为上包络线和下包络线；
    基于所述上包络线和所述下包络线，从所述电流信号分离出多个固有模态函数。
  3. 根据权利要求2所述的故障电弧的诊断方法(100)，其特征在于，所述插值方法为三次样条插值方法，所述神经网络模型为量子神经网络模型。
  4. 根据权利要求1所述的故障电弧的诊断方法(100)，其特征在于，所述基于多个固有模态函数获取所述电流信号的多个高频分量的突变特征(103)包括：
    对每个固有模态函数执行希尔伯特变换，得到该固有模态函数的希尔伯特谱；
    基于每个固有模态函数的希尔伯特谱确定该固有模态函数的边际谱区域变化特征，以作为对应于该固有模态函数的高频分量的突变特征。
  5. 根据权利要求1所述的故障电弧的诊断方法(100)，其特征在于，所述基于多个固有模态函数获取所述电流信号的多个高频分量的突变特征(103)包括：
    提取每个固有模态函数的幅值能量；
    确定每个固有模态函数的幅值能量的突变量；
    确定每个固有模态函数的所述突变量的归一化累计值，以作为对应于该固有模态函数的高频分量的突变特征。
  6. 根据权利要求1所述的故障电弧的诊断方法(100)，其特征在于，所述基于固有模态函数获取所述电流信号的多个高频分量的突变特征每个固有模态函数的包括：
    对每个固有模态函数执行希尔伯特变换，得到该固有模态函数的希尔伯特谱，基于每个固有模态函数的希尔伯特谱确定该固有模态函数的边际谱区域变化特征；
    提取每个固有模态函数的幅值能量，确定每个固有模态函数的幅值能量的突变量，确定每个固有模态函数的所述突变量的归一化累计值；
    基于每个固有模态函数的边际谱区域变化特征以及该固有模态函数的突变量的归一化累计值，确定所 述每个固有模态函数的无量纲指标，将所述无量纲指标确定为对应于该固有模态函数的高频分量的突变特征。
  7. 根据权利要求1所述的故障电弧的诊断方法(100)，其特征在于，该方法(100)还包括：
    在所述神经网络模型的训练阶段，进一步将作为训练数据的负载特性、作为训练数据的环境参量和作为训练数据的高频分量的突变特征输入所述神经网络模型以训练所述神经网络模型；
    在所述神经网络模型的应用阶段，进一步将电流信号采集点的环境参量和电流信号采集点的负载特性输入所述神经网络模型。
  8. 一种故障电弧的诊断装置(400)，其特征在于，包括：
    采集模块(401)，用于采集电流信号；
    经验模式分解模块(402)，用于对所述电流信号执行经验模式分解，以获取所述电流信号的多个固有模态函数；
    突变特征获取模块(403)，用于基于所述多个固有模态函数获取所述电流信号的多个高频分量的突变特征；
    类型确定模块(404)，用于将所述多个高频分量的突变特征输入已训练的神经网络模型，以由所述神经网络模型输出故障电弧类型。
  9. 根据权利要求8所述的故障电弧的诊断装置(400)，其特征在于，
    所述经验模式分解模块(402)，用于确定所述电流信号的局部极值点；利用插值方法将所述局部极值点拟合为上包络线和下包络线；基于所述上包络线和所述下包络线，从所述电流信号分离出固有模态函数。
  10. 根据权利要求8所述的故障电弧的诊断装置(400)，其特征在于，
    所述突变特征获取模块(403)，用于对每个固有模态函数执行希尔伯特变换，得到该固有模态函数的希尔伯特谱，基于每个固有模态函数的希尔伯特谱确定该固有模态函数的边际谱区域变化特征，以作为对应于该固有模态函数的高频分量的突变特征。
  11. 根据权利要求8所述的故障电弧的诊断装置(400)，其特征在于，
    所述突变特征获取模块(403)，用于提取每个固有模态函数的幅值能量，确定每个固有模态函数的幅值能量的突变量，确定所述突变量的归一化累计值，以作为所述对应于该固有模态函数的高频分量的突变特征。
  12. 根据权利要求8所述的故障电弧的诊断装置(400)，其特征在于，
    所述突变特征获取模块(403)，用于对每个固有模态函数执行希尔伯特变换，得到该固有模态函数的希尔伯特谱，基于每个固有模态函数的希尔伯特谱确定该固有模态函数的边际谱区域变化特征；提取每个固有模态函数的幅值能量，确定每个固有模态函数的幅值能量的突变量，确定每个固有模态函数的所述突变量的归一化累计值；基于每个固有模态函数的边际谱区域变化特征以及该固有模态函数的突变量的归一 化累计值，确定每个固有模态函数的无量纲指标，将所述无量纲指标确定为对应于该固有模态函数的高频分量的突变特征。
  13. 根据权利要求1所述的故障电弧的诊断装置(400)，其特征在于，
    类型确定模块(404)，还用于在神经网络模型的训练阶段，进一步将作为训练数据的负载特性、作为训练数据的环境参量和作为训练数据的高频分量的突变特征输入所述神经网络模型以训练所述神经网络模型；在所述神经网络模型的应用阶段，进一步将电流信号采集点的环境参量和电流信号采集点的负载特性输入所述神经网络模型。
  14. 一种故障电弧的诊断装置(500)，其特征在于，包括处理器(501)、存储器(502)及存储在所述存储器(502)上并可在所述处理器(501)上运行的计算机程序，所述计算机程序被所述处理器(501)执行时实现如权利要求1至7中任一项所述的故障电弧的诊断方法(100)。
  15. 一种计算机可读存储介质，其特征在于，所述计算机可读存储介质上存储计算机程序，所述计算机程序被处理器执行时实现如权利要求1至7中任一项所述的故障电弧的诊断方法(100)。
PCT/CN2020/119020 2020-09-29 2020-09-29 一种故障电弧的诊断方法、装置和计算机可读存储介质 WO2022067562A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
PCT/CN2020/119020 WO2022067562A1 (zh) 2020-09-29 2020-09-29 一种故障电弧的诊断方法、装置和计算机可读存储介质

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2020/119020 WO2022067562A1 (zh) 2020-09-29 2020-09-29 一种故障电弧的诊断方法、装置和计算机可读存储介质

Publications (1)

Publication Number Publication Date
WO2022067562A1 true WO2022067562A1 (zh) 2022-04-07

Family

ID=80949409

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/119020 WO2022067562A1 (zh) 2020-09-29 2020-09-29 一种故障电弧的诊断方法、装置和计算机可读存储介质

Country Status (1)

Country Link
WO (1) WO2022067562A1 (zh)

Cited By (14)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115077374A (zh) * 2022-06-30 2022-09-20 中国工程物理研究院电子工程研究所 微马达执行器驱动时绝对位置的动态检测装置及检测方法
CN115327373A (zh) * 2022-04-20 2022-11-11 岱特智能科技（上海）有限公司 基于bp神经网络的血透设备故障诊断方法及存储介质
CN115423221A (zh) * 2022-11-03 2022-12-02 北京建筑大学 一种设施运行趋势预测方法
CN115510925A (zh) * 2022-11-17 2022-12-23 中国人民解放军国防科技大学 一种机械故障诊断方法、装置、介质
CN115526215A (zh) * 2022-11-24 2022-12-27 光微半导体(吉林)有限公司 一种胶泵故障诊断维护分析优化方法及系统
CN115841082A (zh) * 2023-02-22 2023-03-24 天津佰焰科技股份有限公司 一种加气站异常诊断系统及方法
CN116070140A (zh) * 2023-04-03 2023-05-05 国网冀北电力有限公司 一种配电变电站安全运行状态监测系统及方法
CN116381406A (zh) * 2023-03-16 2023-07-04 武汉船舶职业技术学院 船舶电网故障定位方法、装置、设备及可读存储介质
CN116756619A (zh) * 2023-07-12 2023-09-15 常熟浦发第二热电能源有限公司 一种基于大数据的设备智能诊断方法及系统
CN116955995A (zh) * 2023-09-20 2023-10-27 深圳市嘉友锦磁科技有限公司 一种三相直流无刷电机逆变器故障诊断方法
CN117114516A (zh) * 2023-10-25 2023-11-24 湖南省水务规划设计院有限公司 一种长距离小断面引水隧洞安全评估方法
WO2023226113A1 (zh) * 2022-05-25 2023-11-30 中国南方电网有限责任公司超高压输电公司检修试验中心 直流电弧仿真方法、装置、计算机设备和存储介质
CN117633611A (zh) * 2023-10-23 2024-03-01 北京航天常兴科技发展股份有限公司 一种危险用电器及用电行为识别方法及系统
CN116381406B (zh) * 2023-03-16 2024-06-04 武汉船舶职业技术学院 船舶电网故障定位方法、装置、设备及可读存储介质

Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20020118022A1 (en) * 2001-02-27 2002-08-29 Dring Jonathan Samuel Arc detection
CN103064008A (zh) * 2012-12-26 2013-04-24 浙江大学 一种基于希尔伯特-黄变换的非线性模拟电路软故障诊断方法
CN108761287A (zh) * 2018-05-28 2018-11-06 深圳供电局有限公司 一种变压器局部放电类型识别方法
CN109521301A (zh) * 2018-11-30 2019-03-26 北京航空航天大学 一种故障电弧产生装置及其检测方法
CN109828227A (zh) * 2019-01-24 2019-05-31 东南大学 一种基于电流信息特征的电子式电流互感器故障诊断方法
CN109975673A (zh) * 2019-04-23 2019-07-05 辽宁工程技术大学 一种光伏微电网直流侧故障电弧识别方法
CN110069886A (zh) * 2019-05-05 2019-07-30 四川大学 基于vmd和cnn的电缆早期故障识别与分类方法
CN110398669A (zh) * 2019-06-11 2019-11-01 深圳供电局有限公司 电弧故障检测方法
CN111610416A (zh) * 2020-05-25 2020-09-01 南京航空航天大学 一种串联电弧故障智能断路器

Patent Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20020118022A1 (en) * 2001-02-27 2002-08-29 Dring Jonathan Samuel Arc detection
CN103064008A (zh) * 2012-12-26 2013-04-24 浙江大学 一种基于希尔伯特-黄变换的非线性模拟电路软故障诊断方法
CN108761287A (zh) * 2018-05-28 2018-11-06 深圳供电局有限公司 一种变压器局部放电类型识别方法
CN109521301A (zh) * 2018-11-30 2019-03-26 北京航空航天大学 一种故障电弧产生装置及其检测方法
CN109828227A (zh) * 2019-01-24 2019-05-31 东南大学 一种基于电流信息特征的电子式电流互感器故障诊断方法
CN109975673A (zh) * 2019-04-23 2019-07-05 辽宁工程技术大学 一种光伏微电网直流侧故障电弧识别方法
CN110069886A (zh) * 2019-05-05 2019-07-30 四川大学 基于vmd和cnn的电缆早期故障识别与分类方法
CN110398669A (zh) * 2019-06-11 2019-11-01 深圳供电局有限公司 电弧故障检测方法
CN111610416A (zh) * 2020-05-25 2020-09-01 南京航空航天大学 一种串联电弧故障智能断路器

Cited By (21)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115327373A (zh) * 2022-04-20 2022-11-11 岱特智能科技（上海）有限公司 基于bp神经网络的血透设备故障诊断方法及存储介质
CN115327373B (zh) * 2022-04-20 2023-04-18 岱特智能科技（上海）有限公司 基于bp神经网络的血透设备故障诊断方法及存储介质
WO2023226113A1 (zh) * 2022-05-25 2023-11-30 中国南方电网有限责任公司超高压输电公司检修试验中心 直流电弧仿真方法、装置、计算机设备和存储介质
CN115077374A (zh) * 2022-06-30 2022-09-20 中国工程物理研究院电子工程研究所 微马达执行器驱动时绝对位置的动态检测装置及检测方法
CN115423221A (zh) * 2022-11-03 2022-12-02 北京建筑大学 一种设施运行趋势预测方法
CN115510925A (zh) * 2022-11-17 2022-12-23 中国人民解放军国防科技大学 一种机械故障诊断方法、装置、介质
CN115510925B (zh) * 2022-11-17 2023-04-07 中国人民解放军国防科技大学 一种机械故障诊断方法、装置、介质
CN115526215A (zh) * 2022-11-24 2022-12-27 光微半导体(吉林)有限公司 一种胶泵故障诊断维护分析优化方法及系统
CN115526215B (zh) * 2022-11-24 2023-04-07 光微半导体(吉林)有限公司 一种胶泵故障诊断维护分析优化方法及系统
CN115841082A (zh) * 2023-02-22 2023-03-24 天津佰焰科技股份有限公司 一种加气站异常诊断系统及方法
CN116381406A (zh) * 2023-03-16 2023-07-04 武汉船舶职业技术学院 船舶电网故障定位方法、装置、设备及可读存储介质
CN116381406B (zh) * 2023-03-16 2024-06-04 武汉船舶职业技术学院 船舶电网故障定位方法、装置、设备及可读存储介质
CN116070140A (zh) * 2023-04-03 2023-05-05 国网冀北电力有限公司 一种配电变电站安全运行状态监测系统及方法
CN116756619A (zh) * 2023-07-12 2023-09-15 常熟浦发第二热电能源有限公司 一种基于大数据的设备智能诊断方法及系统
CN116756619B (zh) * 2023-07-12 2023-11-21 常熟浦发第二热电能源有限公司 一种基于大数据的设备智能诊断方法及系统
CN116955995A (zh) * 2023-09-20 2023-10-27 深圳市嘉友锦磁科技有限公司 一种三相直流无刷电机逆变器故障诊断方法
CN116955995B (zh) * 2023-09-20 2024-01-05 深圳市嘉友锦磁科技有限公司 一种三相直流无刷电机逆变器故障诊断方法
CN117633611A (zh) * 2023-10-23 2024-03-01 北京航天常兴科技发展股份有限公司 一种危险用电器及用电行为识别方法及系统
CN117633611B (zh) * 2023-10-23 2024-05-24 北京航天常兴科技发展股份有限公司 一种危险用电器及用电行为识别方法及系统
CN117114516A (zh) * 2023-10-25 2023-11-24 湖南省水务规划设计院有限公司 一种长距离小断面引水隧洞安全评估方法
CN117114516B (zh) * 2023-10-25 2024-05-28 湖南省水务规划设计院有限公司 一种长距离小断面引水隧洞安全评估方法

Similar Documents

Publication Publication Date Title
WO2022067562A1 (zh) 一种故障电弧的诊断方法、装置和计算机可读存储介质
Veerasamy et al. LSTM recurrent neural network classifier for high impedance fault detection in solar PV integrated power system
Rodríguez et al. Rule-based classification of power quality disturbances using S-transform
CN110879917A (zh) 一种基于迁移学习的电力系统暂态稳定自适应评估方法
CN112149510B (zh) 一种非侵入式负荷检测方法
CN111160241B (zh) 一种基于深度学习的配电网故障分类方法、系统和介质
Jia et al. Validity evaluation of transformer DGA online monitoring data in grid edge systems
Li et al. Physics-informed learning for high impedance faults detection
CN111275069B (zh) 一种非侵入式负荷监测方法
CN114006413A (zh) 一种基于图神经网络的电力系统暂态稳定控制方法及系统
Harish et al. Fault detection and classification for wide area backup protection of power transmission lines using weighted extreme learning machine
Zou et al. Series arc fault detection based on dual filtering feature selection and improved hierarchical clustering sensitive component selection
CN109635430B (zh) 电网输电线路暂态信号监测方法和系统
Cui et al. High-impedance fault detection method based on sparse data divergence discrimination in distribution networks
Gong et al. Series arc fault identification method based on wavelet transform and feature values decomposition fusion DNN
Singh et al. Power system fault diagnosis using fuzzy decision tree
Ma et al. Fault diagnosis of power electronic system based on fault gradation and neural network group
Liu et al. Classification and identification of electric shock current for safety operation in power distribution network
Rizvi et al. Real-time ZIP load parameter tracking using adaptive window and variable elimination with realistic synthetic synchrophasor data
Wilson et al. Uncertainty Quantification of Capacitor Switching Transient Location Using Machine Learning
Yue et al. A power system harmonic problem based on the BP neural network learning algorithm
Tian et al. Intelligent Detection Method of Power Grid Fault Big Data Based on Deep Learning
Wang et al. Impedance Ground Faults Detection and Classification Method for DC Microgrid
Ramadoss et al. Ensemble machine learning approach to identify excitation failure in synchronous generators
Hou et al. A series arc fault diagnosis method based on random forest model

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20955587

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20955587

Country of ref document: EP

Kind code of ref document: A1