WO2021136053A1 - 一种故障电弧的识别方法、装置、设备及介质 - Google Patents

一种故障电弧的识别方法、装置、设备及介质 Download PDF

Info

Publication number
WO2021136053A1
WO2021136053A1 PCT/CN2020/138919 CN2020138919W WO2021136053A1 WO 2021136053 A1 WO2021136053 A1 WO 2021136053A1 CN 2020138919 W CN2020138919 W CN 2020138919W WO 2021136053 A1 WO2021136053 A1 WO 2021136053A1
Authority
WO
WIPO (PCT)
Prior art keywords
arc
target
sampling
neural network
dispersion
Prior art date
Application number
PCT/CN2020/138919
Other languages
English (en)
French (fr)
Inventor
王华荣
王建华
马越
Original Assignee
青岛鼎信通讯股份有限公司
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 青岛鼎信通讯股份有限公司 filed Critical 青岛鼎信通讯股份有限公司
Priority to US17/769,770 priority Critical patent/US11831138B2/en
Priority to CA3153759A priority patent/CA3153759C/en
Priority to EP20909461.4A priority patent/EP4024063B1/en
Publication of WO2021136053A1 publication Critical patent/WO2021136053A1/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
    • H—ELECTRICITY
    • H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
    • H02H—EMERGENCY PROTECTIVE CIRCUIT ARRANGEMENTS
    • H02H1/00—Details of emergency protective circuit arrangements
    • H02H1/0007—Details of emergency protective circuit arrangements concerning the detecting means
    • H02H1/0015—Using arc detectors
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
    • G01R23/00—Arrangements for measuring frequencies; Arrangements for analysing frequency spectra
    • G01R23/02—Arrangements for measuring frequency, e.g. pulse repetition rate; Arrangements for measuring period of current or voltage
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
    • G01R23/00—Arrangements for measuring frequencies; Arrangements for analysing frequency spectra
    • G01R23/02—Arrangements for measuring frequency, e.g. pulse repetition rate; Arrangements for measuring period of current or voltage
    • G01R23/06—Arrangements for measuring frequency, e.g. pulse repetition rate; Arrangements for measuring period of current or voltage by converting frequency into an amplitude of current or voltage
    • 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
    • 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
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
    • G01R19/00—Arrangements for measuring currents or voltages or for indicating presence or sign thereof
    • G01R19/25—Arrangements for measuring currents or voltages or for indicating presence or sign thereof using digital measurement techniques
    • G01R19/2506—Arrangements for conditioning or analysing measured signals, e.g. for indicating peak values ; Details concerning sampling, digitizing or waveform capturing
    • G01R19/2509—Details concerning sampling, digitizing or waveform capturing
    • 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
    • G01R31/1227—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 of components, parts or materials
    • G01R31/1263—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 of components, parts or materials of solid or fluid materials, e.g. insulation films, bulk material; of semiconductors or LV electronic components or parts; of cable, line or wire insulation
    • G01R31/1272—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 of components, parts or materials of solid or fluid materials, e.g. insulation films, bulk material; of semiconductors or LV electronic components or parts; of cable, line or wire insulation of cable, line or wire insulation, e.g. using partial discharge measurements
    • 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
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks

Definitions

  • the present invention relates to the technical field of power and electricity, and in particular to a method, device, equipment and medium for identifying fault arcs.
  • Fault arc is one of the important causes of electrical fires.
  • the arc is usually sampled and detected based on low-frequency sampling frequency.
  • this detection method cannot effectively observe the high-frequency characteristics in the arc, and it is easy to be confused with the frequency characteristics of household appliances. Therefore, misjudgments and missed judgments of fault arcs often occur.
  • the purpose of the present invention is to provide a method, device, equipment and medium for identifying arc faults, so as to further improve the accuracy and reliability of the results of arc fault recognition.
  • the specific plan is as follows:
  • a method for identifying arc faults including:
  • the process of creating the neural network model includes:
  • the neural network model is established by using the training data.
  • the process of sampling the target arc at a high frequency to obtain a high frequency sampling signal includes:
  • the target arc is sampled at a sampling rate of 1 GHz to obtain the high-frequency sampling signal.
  • the process of preprocessing the high-frequency sampling signal to obtain the processing sampling signal includes:
  • the process of performing feature extraction on the processed sample signal to obtain the target arc feature includes:
  • Ten sampling points are selected from the target sampling points in chronological order to obtain a target sampling sequence; wherein, the target sampling sequence includes sampling points D, E, F, G, H, I, J, M, P, and S;
  • a 1
  • V ij of i and j representing the serial number of the sampling points within the target sequence of samples
  • V DF max (y D , y E, y F) -min (y D, y E, y F )
  • V FH max(y F ,y G ,y H )-min(y F ,y G ,y H )
  • V HJ max(y H ,y I ,y J )-min(y H ,y I ,y J )
  • V JP max(y J ,y M ,y P )-min(y J ,y M ,y P )
  • y D , y E , y F , y G , y H , y I , Y J , y M , and y P are the ordinate values of sampling points D, E, F, G, H, I, J, M and P in the world coordinate system respectively;
  • a 3
  • i and j in VS ij respectively represent the serial number of each sampling point in the target sampling sequence
  • VS DF
  • , VS FH
  • , VS HJ
  • , VS JP
  • T 1 is the time interval between sampling points D and F
  • T 2 is the time interval between sampling points F and H
  • T 3 is the time interval between sampling points J and H
  • T 4 is the sampling point The time interval between P and J
  • a 7
  • a 8
  • a 9
  • X FH
  • , X DF
  • , X HJ
  • , X JP
  • , X EF
  • , X DE
  • , X FG
  • , X GH
  • , X HI
  • , X IJ
  • , X JM
  • , X MP
  • ; y D , y E , y F , y G , y H , y I , y J , y M , and y P are the ordinate values of the sampling points D, E, F, G, H, I, J, M, and P in the world coordinate system, respectively.
  • the process of establishing the neural network model based on the neural network algorithm and using the training data includes:
  • the neural network model is established by using the training data.
  • the process of inputting the target arc characteristic into the neural network model to obtain the target output result includes:
  • the target arc characteristic is converted into a normalized characteristic matrix, and the normalized characteristic matrix is input to the neural network model to obtain the target output result.
  • the present invention also discloses a fault arc recognition device, including:
  • the signal sampling module is used to sample the target arc at high frequencies to obtain high frequency sampling signals
  • a signal processing module for preprocessing the high-frequency sampling signal to obtain a processed sampling signal
  • the feature extraction module is used to perform feature extraction on the processed sampled signal to obtain the target arc feature
  • the result judgment module is used to input the target arc characteristic into the neural network model to obtain the target output result, and judge whether the target arc is a fault arc according to the target output result;
  • the process of creating the neural network model includes:
  • the neural network model is established by using the training data.
  • the present invention also discloses a fault arc recognition equipment, including:
  • Memory used to store computer programs
  • the processor is used to implement the steps of the method for identifying arc faults disclosed in the foregoing when executing the computer program.
  • the present invention also discloses a computer-readable storage medium having a computer program stored on the computer-readable storage medium, and when the computer program is executed by a processor, it realizes the identification of a fault arc as disclosed above. Method steps.
  • the arc characteristics corresponding to the normal arc and the fault arc are obtained in advance, and the training data is obtained. Then, based on the neural network algorithm, the neural network model is established by using the training data; in the process of identifying the fault of the target arc In the process, the target arc is first sampled at high frequency to obtain the high frequency sampling signal, and then the high frequency sampling signal is preprocessed to obtain the processed sampling signal, and feature extraction is performed on the processed sampling signal to obtain the target arc feature. Finally, input the target arc characteristics into the neural network model to obtain the target output result, and judge whether the target arc is a fault arc according to the target output result.
  • the device, equipment and medium for identifying arc faults provided by the present invention also have the above-mentioned beneficial effects.
  • Fig. 1 is a flowchart of a method for identifying arc faults provided by an embodiment of the present invention
  • Figure 2 is a schematic diagram of the waveform of an electric drill when an arc fault occurs
  • Figure 3 is a schematic diagram of the waveform of the switching power supply when an arc fault occurs
  • Figure 4 is a schematic diagram of the waveform of the air compressor when an arc fault occurs
  • Figure 5 is a schematic diagram of the waveform of the vacuum cleaner when an arc fault occurs
  • Figure 6 is a schematic diagram of the waveform of the vacuum cleaner when an arc fault occurs
  • Figure 7 is a schematic diagram of the waveform of the dimmer when an arc fault occurs
  • Figure 8 is a schematic diagram of the waveform of the processed sample signal obtained after the target arc is preprocessed
  • FIG. 9 is a schematic diagram of waveforms when feature extraction is performed on the processed sample signal of the target arc
  • Figure 10 is a schematic diagram of a neural network model created based on a convolutional neural network algorithm
  • FIG. 11 is a schematic diagram of a target arc during fault identification according to an embodiment of the present invention.
  • FIG. 12 is a structural diagram of an arc fault identification device provided by an embodiment of the present invention.
  • Fig. 13 is a structural diagram of an arc fault identification device provided by an embodiment of the present invention.
  • FIG. 1 is a flowchart of a method for identifying arc faults according to an embodiment of the present invention.
  • the identifying method includes:
  • Step S11 sampling the target arc at high frequency to obtain a high frequency sampling signal
  • Step S12 preprocessing the high-frequency sampling signal to obtain a processed sampling signal
  • Step S13 Perform feature extraction on the processed sample signal to obtain the target arc feature
  • Step S14 Input the target arc characteristic into the neural network model to obtain the target output result, and judge whether the target arc is a fault arc according to the target output result;
  • the creation process of the neural network model includes:
  • a method for identifying arc faults is provided, which can significantly improve the accuracy and reliability of arc fault detection results.
  • this identification method firstly, the arc characteristics corresponding to the normal arc and the fault arc are obtained in advance to obtain the training data, and then, based on the neural network algorithm, the neural network model is established by using the training data.
  • the model to be trained is first established based on the neural network algorithm, and then the normal arc and the fault arc are sampled at high frequency to obtain the high frequency sampling signal and the fault arc corresponding to the normal arc.
  • the high-frequency sampling signal corresponding to the fault arc; then, the high-frequency sampling signal corresponding to the normal arc and the high-frequency sampling signal corresponding to the fault arc are preprocessed to obtain the processing sampling signal corresponding to the normal arc and the fault arc respectively
  • the corresponding processing sampling signal, and finally, the processing sampling signal corresponding to the normal arc and the processing sampling signal corresponding to the fault arc are extracted to obtain the arc characteristics corresponding to the normal arc and the arc characteristics corresponding to the fault arc. That is, training data; after that, using the training data to train the model to be trained, the neural network model can be obtained. Obviously, when the neural network model is obtained, it is equivalent to obtaining a mathematical model that can classify arc faults.
  • the target arc is first sampled with high frequency to obtain the high frequency sampling signal. It is conceivable that when the target arc is sampled with high frequency, more arc characteristics related to the target arc can be obtained.
  • Figure 2 is a schematic diagram of the waveform of the electric drill when an arc fault occurs; the upper part of Figure 2 is the arc characteristic collected by the low-frequency sampling frequency when the electric drill has an arc fault. The lower part of 2 is the arc characteristics collected by the high-frequency sampling frequency when the electric drill fails; Figure 3 is the waveform diagram of the switching power supply when an arc fault occurs; among them, the upper half of Figure 3 is the switching power supply When a fault arc occurs, the arc characteristics collected by the low-frequency sampling frequency are used.
  • Figure 3 The lower part of Figure 3 is the arc characteristics collected by the high-frequency sampling frequency when the switching power supply fails;
  • Figure 4 shows the occurrence of an air compressor Schematic diagram of the waveform at the time of arc fault; among them, the upper part of Fig. 4 is the arc characteristics collected by the low-frequency sampling frequency when the air compressor has an arc fault.
  • the lower part of Fig. 4 is the air compressor in the event of a fault.
  • Figure 5 is a schematic diagram of the waveform of the vacuum cleaner when an arc fault occurs; among them, the upper part of Figure 5 is the vacuum cleaner using the low-frequency sampling frequency when the arc fault occurs Arc characteristics, the lower part of Figure 5 is the arc characteristics collected by the high-frequency sampling frequency when the vacuum cleaner fails;
  • Figure 6 is a schematic diagram of the waveform of the vacuum cleaner when an arc fault occurs; among them, the upper half of Figure 6 is The vacuum cleaner uses the arc characteristics collected by the low-frequency sampling frequency when an arc fault occurs in the vacuum cleaner.
  • the lower part of Figure 6 is the arc characteristics collected by the high-frequency sampling frequency when the vacuum cleaner fails;
  • Figure 7 shows the arc characteristics collected by the high-frequency sampling frequency when the vacuum cleaner fails.
  • the arc characteristics collected by the high-frequency sampling frequency are used. It can be seen from Figure 2 to Figure 7 that the use of high-frequency sampling frequency can collect more obvious and distinguishable arc characteristics from the target arc.
  • the high-frequency sampling signal is preprocessed to obtain the processed sampling signal to remove the impurity signal in the high-frequency sampling signal and facilitate the processing of the subsequent process.
  • feature extraction is performed on the processed sampling signal to obtain the target arc
  • input the target arc characteristics into the neural network model and then it can be judged whether the target arc is a fault arc according to the output result of the neural network model.
  • the method of setting a threshold is generally used to classify arcs.
  • this classification method is difficult to adapt to different load environments.
  • the neural network model not only has strong data learning capabilities, but also has good data classification capabilities. Therefore, when the neural network model is used to identify and judge the target arc, it can further improve the process of identifying the target arc. Accuracy and reliability.
  • the arc characteristics corresponding to the normal arc and the fault arc are obtained in advance to obtain the training data, and then based on the neural network algorithm, the neural network model is established by using the training data; in the fault recognition of the target arc In the process, first sample the target arc with high frequency to obtain the high frequency sampling signal, and then preprocess the high frequency sampling signal to obtain the processed sampling signal, and perform feature extraction on the processed sampling signal to obtain the target arc feature Finally, input the target arc characteristics into the neural network model to obtain the target output result, and judge whether the target arc is a fault arc according to the target output result.
  • the neural network model has good data classification capabilities, when the neural network model is used for the target arc When making a judgment, the accuracy and reliability of the fault arc detection result can be further improved.
  • sampling the target arc at high frequency to obtain a high frequency sampling signal includes:
  • the target arc is sampled at a sampling rate of 1 GHz to obtain a high-frequency sampling signal.
  • the target arc is sampled at a sampling rate of 1 GHz to obtain a high-frequency sampling signal, because the sampling rate of 1 GHz is much higher than the sampling rate for sampling the target arc in the prior art. This will be more conducive to the feature extraction process of the target arc in the subsequent process.
  • ASIC Application Specific Integrated Circuit
  • preprocessing the high-frequency sampled signal to obtain the process of processing the sampled signal includes:
  • the high-frequency sampling signal is filtered to obtain the filtered sampling signal.
  • Adaptive gain adjustment is performed on the filtered sample signal, and the low frequency signal in the filtered sample signal is filtered out to obtain the processed sample signal.
  • a specific implementation method for preprocessing the high-frequency sampling signal is provided, that is, in the process of preprocessing the high-frequency sampling signal, the high-frequency sampling signal is first filtered.
  • the impurity signal in the high-frequency sampling signal is removed to obtain the filtered sampling signal, and then adaptive gain adjustment is performed on the filtered sampling signal, and the low frequency signal in the filtered sampling signal is filtered out to obtain the processed sampling signal.
  • the process of extracting the characteristics of the processed sampled signal to obtain the target arc characteristic includes:
  • the target sampling sequence includes sampling points D, E, F, G, H, I, J, M, P, and S;
  • a 1
  • V ij represent the serial number of each sampling point in the target sampling sequence
  • V DF max(y D ,y E ,y F )-min(y D ,y E ,y F )
  • V FH max(y F ,y G ,y H )-min(y F ,y G ,y H )
  • V HJ max(y H ,y I ,y J )-min(y H ,y I , y J )
  • V JP max(y J ,y M ,y P )-min(y J ,y M ,y P )
  • y D , y E , y F , y G , y H , y I , y J , y M , and y P are the ordinate values of sampling points D, E, F, G, H, I, J, M, and P in the world coordinate system, respectively;
  • a 3
  • i and j in VS ij represent the serial number of each sampling point in the target sampling sequence
  • VS DF
  • , VS FH
  • , VS HJ
  • , VS JP
  • T 1 is the time interval between sampling points D and F
  • T 2 is the time interval between sampling points F and H
  • T 3 is the time interval between sampling points J and H
  • T 4 is the sampling point The time interval between P and J
  • length(y) is the length of the target sampling sequence
  • a 7
  • a 8
  • a 9
  • X FH
  • , X DF
  • , X HJ
  • , X JP
  • , X EF
  • , X DE
  • , X FG
  • , X GH
  • , X HI
  • , X IJ
  • , X JM
  • , X MP
  • ; y D , y E , y F , y G , y H , y I , y J , y M , and y P are the ordinate values of the sampling points D, E, F, G, H, I, J, M, and P in the world coordinate system, respectively.
  • Figure 8 is a schematic diagram of the waveform of the processed sample signal obtained after the target arc is preprocessed.
  • 9 is a schematic diagram of the waveform when feature extraction is performed on the processed sample signal of the target arc.
  • the target sampling sequence includes sampling points D, E, F, G, H, I, J, M, P, and S; finally, nine arc features in the target sampling sequence are extracted, that is, the amplitude Value and dispersion, amplitude ratio dispersion, constant point amplitude and dispersion, constant point amplitude ratio dispersion, time ratio dispersion, number of waveforms, slope dispersion, slope difference dispersion, and points and dispersion.
  • the process of establishing a neural network model based on the neural network algorithm and using training data includes:
  • the neural network model is built using training data.
  • BP Back Propagation
  • RBF Random Basis Function, radial basis function
  • perceptron neural network algorithm perceptron neural network algorithm
  • self-organization Neural network algorithms and so on.
  • the neural network model is established based on the convolutional neural network algorithm and training data, because compared to other types of neural network algorithms, the convolutional neural network algorithm not only processes high-dimensional data There is no great pressure in the process, and there is no need to manually extract feature parameters using the convolutional neural network algorithm. The best classification effect can be achieved as long as the weights in the algorithm are set. Therefore, in this embodiment, it is based on the volume Integrate neural network algorithms and training data to build neural network models.
  • Figure 10 is a schematic diagram of a neural network model created based on a convolutional neural network algorithm.
  • the neural network model is composed of a convolutional layer, a pooling layer, and a fully connected layer.
  • the convolution layer uses three 3*3 convolution kernels to perform convolution operations on the target arc features.
  • the pooling layer is used to perform dimensionality reduction operations on the convolution results, that is, the pooling layer will
  • the target arc feature is reduced to a one-dimensional vector and provided to the fully connected layer for fault identification of the target arc.
  • the above-mentioned step: inputting the target arc characteristic into the neural network model to obtain the target output result includes:
  • the target arc characteristic is converted into a normalized characteristic matrix, and the normalized characteristic matrix is input to the neural network model to obtain the target output result.
  • the target arc characteristic is converted into a normalized characteristic matrix, because the normalized characteristic matrix can Significantly improve the computational convergence speed of the neural network model. Therefore, when the normalized feature matrix is input into the neural network model, the neural network model can further accelerate the recognition speed of the target arc fault recognition.
  • FIG. 11 is a schematic diagram of a target arc during fault identification according to an embodiment of the present invention.
  • the target current is sampled by the sampling circuit to obtain the target arc, and then the target arc is sampled at the sampling frequency of 1GHz to obtain the high-speed sampling signal.
  • the high-speed sampling signal is passed through the band Pass filter 1, band pass filter 2, and band pass filter 3 perform filtering.
  • the band pass frequency bands of band pass filter 1, band pass filter 2 and band pass filter 3 are 500KHz ⁇ 50MHz and 50MHz ⁇ respectively. 100MHz and 100MHz ⁇ 200MHz, and use this to extract the high frequency sampling signal of the target arc in different frequency bands.
  • each half wave of 10ms after filtering is equally divided into 500 segments to obtain a segmented waveform with a duration of about 20us, and the gain adjustment and time-sharing processing are performed on each segmented waveform respectively to obtain the processed sample signal, and then , And then extract the amplitude and dispersion of each processed sample signal, amplitude ratio dispersion, constant point amplitude and dispersion, constant point amplitude ratio dispersion, time ratio dispersion, number of waveforms, slope dispersion, Slope differential dispersion and points and dispersion.
  • the neural network model can be used to identify the target arc fault.
  • 500-dimensional feature vectors will be obtained, and each dimensional feature vector has 9 arc features, so use this 500*9 arc characteristics can form a normalized characteristic matrix, and then input the normalized characteristic matrix into the neural network model, and then it can be judged whether the target arc is a fault arc according to the output result of the neural network model.
  • FIG. 12 is a structural diagram of an arc fault identification device provided by an embodiment of the present invention, and the identification device includes:
  • the signal sampling module 21 is used to sample the target arc at a high frequency to obtain a high frequency sampling signal
  • the signal processing module 22 is used to preprocess the high-frequency sampling signal to obtain the processed sampling signal
  • the feature extraction module 23 is used to perform feature extraction on the processed sampled signal to obtain the target arc feature
  • the result judgment module 24 is used to input the target arc characteristic into the neural network model to obtain the target output result, and judge whether the target arc is a fault arc according to the target output result;
  • the creation process of the neural network model includes:
  • the device for identifying an arc fault provided by an embodiment of the present invention has the beneficial effects of the method for identifying an arc fault disclosed above.
  • FIG. 13 is a structural diagram of an arc fault identification device provided by an embodiment of the present invention, and the identification device includes:
  • the memory 31 is used to store computer programs
  • the processor 32 is configured to implement the steps of a method for identifying arc faults as disclosed above when executing a computer program.
  • the device for identifying an arc fault provided by an embodiment of the present invention has the beneficial effects of the method for identifying an arc fault disclosed above.
  • the embodiment of the present invention also discloses a computer-readable storage medium, and a computer program is stored on the computer-readable storage medium.
  • the computer program is executed by a processor, the method for identifying a fault arc as disclosed above is implemented. step.
  • the computer-readable storage medium provided by the embodiment of the present invention has the beneficial effects of the previously disclosed method for identifying arc faults.

Landscapes

  • Physics & Mathematics (AREA)
  • Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Artificial Intelligence (AREA)
  • Biomedical Technology (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Emergency Protection Circuit Devices (AREA)
  • Feedback Control In General (AREA)
  • Locating Faults (AREA)
  • Complex Calculations (AREA)
  • Testing Relating To Insulation (AREA)
  • Measurement Of Current Or Voltage (AREA)

Abstract

一种故障电弧的识别方法、装置、设备及介质，方法包括：以高频对目标电弧进行采样，得到高频采样信号(S11)；对高频采样信号进行预处理，得到处理采样信号(S12)；对处理采样信号进行特征提取，得到目标电弧特征(S13)；将目标电弧特征输入至神经网络模型，得到目标输出结果，并根据目标输出结果判断目标电弧是否为故障电弧(S14)；因为通过高频来对目标电弧进行采样时，可以获取得到目标电弧中数量更多的电弧特征，并且，由于神经网络模型具有良好的数据分类能力，所以，当利用神经网络模型来对目标电弧进行判断时，可以提高故障电弧检测结果的准确性与可靠性。

Description

一种故障电弧的识别方法、装置、设备及介质
本申请要求于2020年01月02日提交中国专利局、申请号为202010000565.5、发明名称为“一种基于电流高频特征的故障电弧识别方法”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本发明涉及电力电气技术领域，特别涉及一种故障电弧的识别方法、装置、设备及介质。
背景技术
故障电弧是引起电气火灾的重要原因之一。目前，在对故障电弧进行检测时，通常是基于低频采样频率来对电弧进行采样检测，但是，此种检测方法不能有效观测出电弧中的高频率特征，容易和家用电器的频率特征发生混淆，所以，经常会出现故障电弧的误判和漏判。目前，针对这一技术问题，还没有较为有效的解决办法。
由此可见，如何进一步提高故障电弧检测结果的准确性与可靠性，是本领域技术人员亟待解决的技术问题。
发明内容
有鉴于此，本发明的目的在于提供一种故障电弧的识别方法、装置、设备及介质，以进一步提高故障电弧识别结果的准确性与可靠性。其具体方案如下：
一种故障电弧的识别方法，包括：
以高频对目标电弧进行采样，得到高频采样信号；
对所述高频采样信号进行预处理，得到处理采样信号；
对所述处理采样信号进行特征提取，得到目标电弧特征；
将所述目标电弧特征输入至神经网络模型，得到目标输出结果，并根据所述目标输出结果判断所述目标电弧是否为故障电弧；
其中，所述神经网络模型的创建过程包括：
预先获取正常电弧和故障电弧所对应的电弧特征，得到训练数据；
基于神经网络算法，利用所述训练数据建立所述神经网络模型。
优选的，所述以高频对目标电弧进行采样，得到高频采样信号的过程，包括：
以1GHz的采样速率对所述目标电弧进行采样，得到所述高频采样信号。
优选的，所述对所述高频采样信号进行预处理，得到处理采样信号的过程，包括：
对所述高频采样信号进行滤波，得到滤波采样信号。
对所述滤波采样信号进行自适应增益调整，并滤除所述滤波采样信号中的低频信号，得到所述处理采样信号。
优选的，所述对所述处理采样信号进行特征提取，得到目标电弧特征的过程，包括：
剔除所述处理采样信号中的非局部极值点，得到目标采样点；
按照时间顺序依次从所述目标采样点中选取十个采样点，得到目标采样序列；其中，所述目标采样序列包括采样点D、E、F、G、H、I、J、M、P和S；
对所述目标采样序列进行特征提取，得到幅值和离散度、幅值比离散度、始终点幅值和离散度、始终点幅值比离散度、时间比离散度、波形个数、斜率离散度、斜率差分离散度以及点和离散度；
其中，所述幅值和离散度的计算表达式为：
A 1＝|V DF|+|V FH|+|V HJ|+|V JP|；
所述幅值比离散度的计算表达式为：
Figure PCTCN2020138919-appb-000001
式中，V ij中的i和j分别代表所述目标采样序列内各采样点的序列号，V DF＝max(y D,y E,y F)-min(y D,y E,y F)，V FH＝max(y F,y G,y H)-min(y F,y G,y H)，V HJ＝max(y H,y I,y J)-min(y H,y I,y J)，V JP＝max(y J,y M,y P)-min(y J,y M,y P)，y D、 y E、y F、y G、y H、y I、y J、y M、y P分别为采样点D、E、F、G、H、I、J、M和P在世界坐标系中的纵坐标值；
所述始终点幅值和离散度的计算表达式为：
A 3＝|VS DF|+|VS FH|+|VS HJ|+|VS JP|；
所述始终点幅值比离散度的计算表达式为：
Figure PCTCN2020138919-appb-000002
式中，VS ij中的i和j分别代表所述目标采样序列内各采样点的序列号，VS DF＝|y D-y F|，VS FH＝|y F-y H|，VS HJ＝|y H-y J|，VS JP＝|y J-y P|，V DF＝max(y D,y E,y F)-min(y D,y E,y F)，V FH＝max(y F,y G,y H)-min(y F,y G,y H)，V HJ＝max(y H,y I,y J)-min(y H,y I,y J)，V JP＝max(y J,y M,y P)-min(y J,y M,y P)，y D、y E、y F、y G、y H、y I、y J、y M、y P分别为采样点D、E、F、G、H、I、J、M和P在世界坐标系中的纵坐标值；
所述时间比离散度的计算表达式为：
Figure PCTCN2020138919-appb-000003
式中，T 1为采样点D和F之间的时间间隔，T 2为采样点F和H之间的时间间隔，T 3为采样点J和H之间的时间间隔，T 4为采样点P和J之间的时间间隔；
所述波形个数的计算表达式为：
Figure PCTCN2020138919-appb-000004
式中，
Figure PCTCN2020138919-appb-000005
表示向下取整，length(y)为所述目标采样序列的长度；
所述斜率离散度的计算表达式为：
A 7＝|X EF-X DE|+|X FG-X EF|+|X GH-X FG|+|X HI-X GH|+|X IJ-X HI|+|X JM-X IJ|+|X MP-X JM|
所述斜率差分离散度的计算表达式为：
A 8＝|X FH-X DF|+|X HJ-X FH|+|X JP-X HJ|；
所述点和离散度的计算表达式为：
A 9＝|y D|+|y E|+|y F|+|y G|+|y H|+|y I|+|y J|+|y M|+|y P|；
X FH＝|X GH-X FG|、X DF＝|X EF-X DE|、X HJ＝|X IJ-X HI|、X JP＝|X MP-X JM|、X EF＝|y E-y F|，X DE＝|y D-y E|，X FG＝|y F-y G|，X GH＝|y G-y H|，X HI＝|y H-y I|，X IJ＝|y I-y J|，X JM＝|y J-y M|，X MP＝|y M-y P|；y D、y E、y F、y G、y H、y I、y J、y M、y P分别为采样点D、E、F、G、H、I、J、M和P在世界坐标系中的纵坐标值。
优选的，所述基于神经网络算法，利用所述训练数据建立所述神经网络模型的过程，包括：
基于卷积神经网络算法，利用所述训练数据建立所述神经网络模型。
优选的，所述将所述目标电弧特征输入至神经网络模型，得到目标输出结果的过程，包括：
将所述目标电弧特征转换为归一化特征矩阵，并将所述归一化特征矩阵输入至所述神经网络模型，得到所述目标输出结果。
相应的，本发明还公开了一种故障电弧的识别装置，包括：
信号采样模块，用于以高频对目标电弧进行采样，得到高频采样信号；
信号处理模块，用于对所述高频采样信号进行预处理，得到处理采样信号；
特征提取模块，用于对所述处理采样信号进行特征提取，得到目标电弧特征；
结果判断模块，用于将所述目标电弧特征输入至神经网络模型，得到目标输出结果，并根据所述目标输出结果判断所述目标电弧是否为故障电弧；
其中，所述神经网络模型的创建过程包括：
预先获取正常电弧和故障电弧所对应的电弧特征，得到训练数据；
基于神经网络算法，利用所述训练数据建立所述神经网络模型。
相应的，本发明还公开了一种故障电弧的识别设备，包括：
存储器，用于存储计算机程序；
处理器，用于执行所述计算机程序时实现如前述所公开的一种故障电弧的识别方法的步骤。
相应的，本发明还公开了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有计算机程序，所述计算机程序被处理器执行时实现如前述所公开的一种故障电弧的识别方法的步骤。
可见，在本发明中，是预先获取正常电弧和故障电弧所对应的电弧特征，得到训练数据，之后，再基于神经网络算法，利用训练数据建立神经网络模型；在对目标电弧进行故障识别的过程中，首先是以高频对目标电弧进行采样，得到高频采样信号，然后，再对高频采样信号进行预处理，得到处理采样信号，并对处理采样信号进行特征提取，得到目标电弧特征，最后，再将目标电弧特征输入至神经网络模型，得到目标输出结果，并根据目标输出结果判断目标电弧是否为故障电弧。显然，因为通过高频来对目标电弧进行采样时，可以获取得到目标电弧中数量更多的电弧特征，并且，由于神经网络模型具有良好的数据分类能力，所以，当利用神经网络模型来对目标电弧进行判断时，就可以进一步提高故障电弧检测结果的准确性与可靠性。相应的，本发明所提供的一种故障电弧的识别装置、设备及介质，同样具有上述有益效果。
附图说明
为了更清楚地说明本发明实施例或现有技术中的技术方案，下面将对实施例或现有技术描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是本发明的实施例，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据提供的附图获得其他的附图。
图1为本发明实施例所提供的一种故障电弧的识别方法的流程图；
图2为电钻在发生电弧故障时的波形示意图；
图3为开关电源在发生电弧故障时的波形示意图；
图4为空压机在发生电弧故障时的波形示意图；
图5为吸尘器在发生电弧故障时的波形示意图；
图6为吸尘器在发生电弧故障时的波形示意图；
图7为调光器在发生电弧故障时的波形示意图；
图8为目标电弧经过预处理之后，得到处理采样信号的波形示意图；
图9为对目标电弧的处理采样信号进行特征提取时的波形示意图；
图10为基于卷积神经网络算法所创建的神经网络模型的示意图；
图11为本发明实施例所提供的一种在对目标电弧进行故障识别时的示意图；
图12为本发明实施例所提供的一种故障电弧的识别装置的结构图；
图13为本发明实施例所提供的一种故障电弧的识别设备的结构图。
具体实施方式
下面将结合本发明实施例中的附图，对本发明实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例仅仅是本发明一部分实施例，而不是全部的实施例。基于本发明中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例，都属于本发明保护的范围。
请参见图1，图1为本发明实施例所提供的一种故障电弧的识别方法的流程图，该识别方法包括：
步骤S11：以高频对目标电弧进行采样，得到高频采样信号；
步骤S12：对高频采样信号进行预处理，得到处理采样信号；
步骤S13：对处理采样信号进行特征提取，得到目标电弧特征；
步骤S14：将目标电弧特征输入至神经网络模型，得到目标输出结果，并根据目标输出结果判断目标电弧是否为故障电弧；
其中，神经网络模型的创建过程包括：
预先获取正常电弧和故障电弧所对应的电弧特征，得到训练数据；
基于神经网络算法，利用训练数据建立神经网络模型。
在本实施例中，是提供了一种故障电弧的识别方法，利用该识别方法可以显著提高故障电弧检测结果的准确性与可靠性。在该识别方法中，首先是预先获取正常电弧和故障电弧所对应的电弧特征，得到训练数据，然后，再基于神经网络算法，利用训练数据建立神经网络模型。
具体的，在创建神经网络模型的过程中，首先是基于神经网络算法建立待训练模型，然后，再以高频对正常电弧和故障电弧进行采样，分别得 到正常电弧所对应的高频采样信号和故障电弧所对应的高频采样信号；然后，再对正常电弧所对应的高频采样信号和故障电弧所对应的高频采样信号进行预处理，分别得到正常电弧所对应的处理采样信号和故障电弧所对应的处理采样信号，最后，再对正常电弧所对应的处理采样信号和故障电弧所对应的处理采样信号进行特征提取，得到正常电弧所对应的电弧特征和故障电弧所对应的电弧特征，也即，训练数据；之后，再利用训练数据对待训练模型进行训练，就可以得到神经网络模型。显然，当获取得到神经网络模型之后，就相当于是得到了一个能够对电弧进行故障分类的数学模型。
在对目标电弧进行故障识别的过程中，首先是以高频对目标电弧进行采样，得到高频采样信号。能够想到的是，当利用高频对目标电弧进行采样时，就可以获取得到有关目标电弧数量更多的电弧特征。
具体请参见图2到图7，图2为电钻在发生电弧故障时的波形示意图；其中，图2的上半部分为电钻在发生故障电弧时，利用低频采样频率所采集到的电弧特征，图2的下半部分为电钻在发生故障时，利用高频采样频率所采集到的电弧特征；图3为开关电源在发生电弧故障时的波形示意图；其中，图3的上半部分为开关电源在发生故障电弧时，利用低频采样频率所采集到的电弧特征，图3的下半部分为开关电源在发生故障时，利用高频采样频率所采集到的电弧特征；图4为空压机在发生电弧故障时的波形示意图；其中，图4的上半部分为空压机在发生故障电弧时，利用低频采样频率所采集到的电弧特征，图4的下半部分为空压机在发生故障时，利用高频采样频率所采集到的电弧特征；图5为吸尘器在发生电弧故障时的波形示意图；其中，图5的上半部分为吸尘器在发生故障电弧时，利用低频采样频率所采集到的电弧特征，图5的下半部分为吸尘器在发生故障时，利用高频采样频率所采集到的电弧特征；图6为吸尘器在发生电弧故障时的波形示意图；其中，图6的上半部分为吸尘器在发生故障电弧时，利用低频采样频率所采集到的电弧特征，图6的下半部分为吸尘器在发生故障时，利用高频采样频率所采集到的电弧特征；图7为调光器在发生电弧故障时的波形示意图；其中，图7的上半部分为调光器在发生故障电弧时， 利用低频采样频率所采集到的电弧特征，图7的下半部分为调光器在发生故障时，利用高频采样频率所采集到的电弧特征。从图2到图7可以看出，利用高频采样频率可以从目标电弧中采集得到更为明显、更具有区分度的电弧特征。
然后，再对高频采样信号进行预处理，得到处理采样信号，以将高频采样信号中的杂质信号去除，并方便后续流程的处理，之后，再对处理采样信号进行特征提取，得到目标电弧特征，最后，再将目标电弧特征输入到神经网络模型中，就可以根据神经网络模型的输出结果判断出目标电弧是否为故障电弧。
可以理解的是，在现有技术中，对目标电弧进行分类时，一般是利用设定阈值的方法来对电弧进行分类，但是，此种分类方法很难适应不同的负载环境。而神经网络模型不仅具有强大的数据学习能力，而且，还具有良好的数据分类能力，所以，当利用神经网络模型来对目标电弧进行识别判断时，就可以进一步提高在对目标电弧进行识别过程中的准确性与可靠性。
可见，在本实施例中，是预先获取正常电弧和故障电弧所对应的电弧特征，得到训练数据，之后，再基于神经网络算法，利用训练数据建立神经网络模型；在对目标电弧进行故障识别的过程中，首先是以高频对目标电弧进行采样，得到高频采样信号，然后，再对高频采样信号进行预处理，得到处理采样信号，并对处理采样信号进行特征提取，得到目标电弧特征，最后，再将目标电弧特征输入至神经网络模型，得到目标输出结果，并根据目标输出结果判断目标电弧是否为故障电弧。显然，因为通过高频来对目标电弧进行采样时，可以获取得到目标电弧中数量更多的电弧特征，并且，由于神经网络模型具有良好的数据分类能力，所以，当利用神经网络模型对目标电弧进行判断时，就可以进一步提高故障电弧检测结果的准确性与可靠性。
基于上述实施例，本实施例对技术方案作进一步的说明与优化，作为一种优选的实施方式，上述步骤：以高频对目标电弧进行采样，得到高频采样信号的过程，包括：
以1GHz的采样速率对目标电弧进行采样，得到高频采样信号。
具体的，在本实施例中，是以1GHz的采样速率来对目标电弧进行采样，得到高频采样信号，因为1GHz的采样速率远远高于现有技术中对目标电弧进行采样的采样速率，这样就会更加有利于后续过程中对目标电弧的特征提取过程。
另外，在实际应用中，可以利用ASIC(Application Specific Integrated Circuit，专用集成电路)来以1GHz的采样速率来对目标电弧进行采样，因为通过这样的采样方式不仅可以提取到目标电弧中频率更高的采样成分，而且，ASIC相比于其它集成芯片而言，具有更强的抗干扰能力和更低的系统功耗。
基于上述实施例，本实施例对技术方案作进一步的说明与优化，作为一种优选的实施方式，上述步骤：对高频采样信号进行预处理，得到处理采样信号的过程，包括：
对高频采样信号进行滤波，得到滤波采样信号。
对滤波采样信号进行自适应增益调整，并滤除滤波采样信号中的低频信号，得到处理采样信号。
在本实施例中，是提供了一种对高频采样信号进行预处理的具体实现方法，也即，在对高频采样信号进行预处理的过程中，首先是对高频采样信号进行滤波，去除高频采样信号中的杂质信号，得到滤波采样信号，之后，再对滤波采样信号进行自适应增益调整，并滤除滤波采样信号中的低频信号，得到处理采样信号。
显然，通过本实施例所提供的技术方案，就可以相对避免杂质信号对电弧故障识别结果的影响与干扰。
基于上述实施例，本实施例对技术方案作进一步的说明与优化，作为一种优选的实施方式，上述步骤：对处理采样信号进行特征提取，得到目标电弧特征的过程，包括：
剔除处理采样信号中的非局部极值点，得到目标采样点；
按照时间顺序依次从目标采样点中选取十个采样点，得到目标采样序列；其中，目标采样序列包括采样点D、E、F、G、H、I、J、M、P和S；
对目标采样序列进行特征提取，得到幅值和离散度、幅值比离散度、始终点幅值和离散度、始终点幅值比离散度、时间比离散度、波形个数、斜率离散度、斜率差分离散度以及点和离散度；
其中，幅值和离散度的计算表达式为：
A 1＝|V DF|+|V FH|+|V HJ|+|V JP|；
幅值比离散度的计算表达式为：
Figure PCTCN2020138919-appb-000006
式中，V ij中的i和j分别代表目标采样序列内各采样点的序列号，V DF＝max(y D,y E,y F)-min(y D,y E,y F)，V FH＝max(y F,y G,y H)-min(y F,y G,y H)，V HJ＝max(y H,y I,y J)-min(y H,y I,y J)，V JP＝max(y J,y M,y P)-min(y J,y M,y P)，y D、y E、y F、y G、y H、y I、y J、y M、y P分别为采样点D、E、F、G、H、I、J、M和P在世界坐标系中的纵坐标值；
始终点幅值和离散度的计算表达式为：
A 3＝|VS DF|+|VS FH|+|VS HJ|+|VS JP|；
始终点幅值比离散度的计算表达式为：
Figure PCTCN2020138919-appb-000007
式中，VS ij中的i和j分别代表目标采样序列内各采样点的序列号，VS DF＝|y D-y F|，VS FH＝|y F-y H|，VS HJ＝|y H-y J|，VS JP＝|y J-y P|，V DF＝max(y D,y E,y F)-min(y D,y E,y F)，V FH＝max(y F,y G,y H)-min(y F,y G,y H)，V HJ＝max(y H,y I,y J)-min(y H,y I,y J)，V JP＝max(y J,y M,y P)-min(y J,y M,y P)，y D、 y E、y F、y G、y H、y I、y J、y M、y P分别为采样点D、E、F、G、H、I、J、M和P在世界坐标系中的纵坐标值；
时间比离散度的计算表达式为：
Figure PCTCN2020138919-appb-000008
式中，T 1为采样点D和F之间的时间间隔，T 2为采样点F和H之间的时间间隔，T 3为采样点J和H之间的时间间隔，T 4为采样点P和J之间的时间间隔；
波形个数的计算表达式为：
Figure PCTCN2020138919-appb-000009
式中，
Figure PCTCN2020138919-appb-000010
表示向下取整，length(y)为目标采样序列的长度；
斜率离散度的计算表达式为：
A 7＝|X EF-X DE|+|X FG-X EF|+|X GH-X FG|+|X HI-X GH|+|X IJ-X HI|+|X JM-X IJ|+|X MP-X JM|
斜率差分离散度的计算表达式为：
A 8＝|X FH-X DF|+|X HJ-X FH|+|X JP-X HJ|；
点和离散度的计算表达式为：
A 9＝|y D|+|y E|+|y F|+|y G|+|y H|+|y I|+|y J|+|y M|+|y P|；
X FH＝|X GH-X FG|、X DF＝|X EF-X DE|、X HJ＝|X IJ-X HI|、X JP＝|X MP-X JM|、X EF＝|y E-y F|，X DE＝|y D-y E|，X FG＝|y F-y G|，X GH＝|y G-y H|，X HI＝|y H-y I|，X IJ＝|y I-y J|，X JM＝|y J-y M|，X MP＝|y M-y P|；y D、y E、y F、y G、y H、y I、y J、y M、y P分别为采样点D、E、F、G、H、I、J、M和P在世界坐标系中的纵坐标值。
在本实施例中，是提供了一种对处理采样信号进行特征提取的具体实施方法，请参见图8和图9，图8为目标电弧经过预处理之后，得到处理采样信号的波形示意图，图9为对目标电弧的处理采样信号进行特征提取时的波形示意图。
在获取得到目标电弧的处理采样信号之后，首先是剔除处理采样信号中的非局部极值点，得到目标采样点，然后，再按照时间顺序依次从目标采样点中选取十个采样点，得到目标采样点序列；其中，目标采样序列包括采样点D、E、F、G、H、I、J、M、P和S；最后，再提取目标采样序列中的九个电弧特征，也即，幅值和离散度、幅值比离散度、始终点幅值和离散度、始终点幅值比离散度、时间比离散度、波形个数、斜率离散度、斜率差分离散度以及点和离散度。
基于上述实施例，本实施例对技术方案作进一步的说明与优化，作为一种优选的实施方式，上述步骤：基于神经网络算法，利用训练数据建立神经网络模型的过程，包括：
基于卷积神经网络算法，利用训练数据建立神经网络模型。
可以理解的是，基于神经网络算法衍生出的算法多种多样，比如：BP(Back Propagation)神经网络算法、RBF(Radial Basis Function，径向基)神经网络算法、感知器神经网络算法以及自组织神经网络算法等等。
具体的，在本实施例中，是基于卷积神经网络算法和训练数据来建立神经网络模型，因为卷积神经网络算法相比于其它类型的神经网络算法而言，不仅对高维数据的处理过程无较大压力，而且，利用卷积神经网络算法也无需对特征参数进行手动提取，只要设置好算法中的权重就可以取得最佳的分类效果，所以，在本实施例中，是基于卷积神经网络算法和训练数据来建立神经网络模型。
请参见图10，图10为基于卷积神经网络算法所创建的神经网络模型的示意图，其中，该神经网络模型是由卷积层、池化层和全连接层所构成，并且，在该神经网络模型中，卷积层是使用三个3*3的卷积核对目标电弧特征进行卷积运算，之后，是利用池化层来对卷积结果进行降维运算，也即，池化层将目标电弧特征降维成一个一维向量提供给全连接层对目标电弧进行故障识别。
显然，通过本实施例所提供的技术方案，就可以进一步提高在对目标电弧进行故障识别时的准确性。
基于上述实施例，本实施例对技术方案作进一步的说明与优化，作为一种优选的实施方式，上述步骤：将目标电弧特征输入至神经网络模型，得到目标输出结果的过程，包括：
将目标电弧特征转换为归一化特征矩阵，并将归一化特征矩阵输入至神经网络模型，得到目标输出结果。
在本实施例中，为了进一步提高神经网络模型对目标电弧的识别速度，在获取得到目标电弧的目标电弧特征之后，是将目标电弧特征转换为归一化特征矩阵，因为归一化特征矩阵能够显著提高神经网络模型的计算收敛速度，所以，当将归一化特征矩阵输入到神经网络模型中时，就可以进一步加快神经网络模型对目标电弧进行故障识别时的识别速度。
为了使得本领域技术人员能够更为清楚、明白本发明的实现原理，本实施例通过一个具体例子进行说明。请参见图11，图11为本发明实施例所提供的一种在对目标电弧进行故障识别时的示意图。
在实际操作过程中，首先是利用采样电路对目标电流进行采样，得到目标电弧，之后，再以1GHz的采样频率对目标电弧进行采样，得到高速采样信号，其次，再将高速采样信号分别通过带通滤波器1、带通滤波器2和带通滤波器3进行滤波，其中，带通滤波器1、带通滤波器2和带通滤波器3的带通频段分别为500KHz～50MHz、50MHz～100MHz和100MHz～200MHz，并以此来提取目标电弧在不同频段时的高频采样信号。
然后，再将滤波之后的每一个10ms的半波等分为500段，得到时长约为20us的分段波形，并分别对各个分段波形进行增益调整和分时处理，得到处理采样信号，之后，再提取每段处理采样信号的幅值和离散度、幅值比离散度、始终点幅值和离散度、始终点幅值比离散度、时间比离散度、波形个数、斜率离散度、斜率差分离散度以及点和离散度。与此同时，再利用过零检测电路所触发的过零信号进行特征统计，并将统计后的电弧特征输入至神经网络模型中，就可以利用神经网络模型对目标电弧进行故障识别。
需要说明的是，在本实施例中，当获取得到500段处理采样信号时，就会得到数量为500维的特征向量，而每一维的特征向量具有9个电弧特征，所以，这样利用这500*9个电弧特征就可以组成一个归一化特征矩阵，之后，再将归一化特征矩阵输入至神经网络模型，就可以根据神经网络模型的输出结果判断出目标电弧是否为故障电弧。
显然，因为通过高频来对目标电弧进行采样时，就可以获取得到目标电弧中数量更多的电弧特征，并且，由于神经网络模型具有良好的数据分类能力，所以，当利用神经网络模型对目标电弧进行判断时，就可以进一步提高故障电弧检测结果的准确性与可靠性。
请参见图12，图12为本发明实施例所提供的一种故障电弧的识别装置的结构图，该识别装置包括：
信号采样模块21，用于以高频对目标电弧进行采样，得到高频采样信号；
信号处理模块22，用于对高频采样信号进行预处理，得到处理采样信号；
特征提取模块23，用于对处理采样信号进行特征提取，得到目标电弧特征；
结果判断模块24，用于将目标电弧特征输入至神经网络模型，得到目标输出结果，并根据目标输出结果判断目标电弧是否为故障电弧；
其中，神经网络模型的创建过程包括：
预先获取正常电弧和故障电弧所对应的电弧特征，得到训练数据；
基于神经网络算法，利用训练数据建立神经网络模型。
本发明实施例所提供的一种故障电弧的识别装置，具有前述所公开的一种故障电弧的识别方法所具有的有益效果。
请参见图13，图13为本发明实施例所提供的一种故障电弧的识别设备的结构图，该识别设备包括：
存储器31，用于存储计算机程序；
处理器32，用于执行计算机程序时实现如前述所公开的一种故障电弧的识别方法的步骤。
本发明实施例所提供的一种故障电弧的识别设备，具有前述所公开的一种故障电弧的识别方法所具有的有益效果。
相应的，本发明实施例还公开了一种计算机可读存储介质，计算机可读存储介质上存储有计算机程序，计算机程序被处理器执行时实现如前述所公开的一种故障电弧的识别方法的步骤。
本发明实施例所提供的一种计算机可读存储介质，具有前述所公开的一种故障电弧的识别方法所具有的有益效果。
本说明书中各个实施例采用递进的方式描述，每个实施例重点说明的都是与其它实施例的不同之处，各个实施例之间相同或相似部分互相参见即可。对于实施例公开的装置而言，由于其与实施例公开的方法相对应，所以描述的比较简单，相关之处参见方法部分说明即可。
最后，还需要说明的是，在本文中，诸如第一和第二等之类的关系术语仅仅用来将一个实体或者操作与另一个实体或操作区分开来，而不一定要求或者暗示这些实体或操作之间存在任何这种实际的关系或者顺序。而且，术语“包括”、“包含”或者其任何其他变体意在涵盖非排他性的包含，从而使得包括一系列要素的过程、方法、物品或者设备不仅包括那些要素，而且还包括没有明确列出的其他要素，或者是还包括为这种过程、方法、物品或者设备所固有的要素。在没有更多限制的情况下，由语句“包括一个……”限定的要素，并不排除在包括所述要素的过程、方法、物品或者设备中还存在另外的相同要素。
以上对本发明所提供的一种故障电弧的识别方法、装置、设备及介质进行了详细介绍，本文中应用了具体个例对本发明的原理及实施方式进行了阐述，以上实施例的说明只是用于帮助理解本发明的方法及其核心思想；同时，对于本领域的一般技术人员，依据本发明的思想，在具体实施方式及应用范围上均会有改变之处，综上所述，本说明书内容不应理解为对本发明的限制。

Claims (9)

  1. 一种故障电弧的识别方法，其特征在于，包括：
    以高频对目标电弧进行采样，得到高频采样信号；
    对所述高频采样信号进行预处理，得到处理采样信号；
    对所述处理采样信号进行特征提取，得到目标电弧特征；
    将所述目标电弧特征输入至神经网络模型，得到目标输出结果，并根据所述目标输出结果判断所述目标电弧是否为故障电弧；
    其中，所述神经网络模型的创建过程包括：
    预先获取正常电弧和故障电弧所对应的电弧特征，得到训练数据；
    基于神经网络算法，利用所述训练数据建立所述神经网络模型。
  2. 根据权利要求1所述的识别方法，其特征在于，所述以高频对目标电弧进行采样，得到高频采样信号的过程，包括：
    以1GHz的采样速率对所述目标电弧进行采样，得到所述高频采样信号。
  3. 根据权利要求1所述的识别方法，其特征在于，所述对所述高频采样信号进行预处理，得到处理采样信号的过程，包括：
    对所述高频采样信号进行滤波，得到滤波采样信号。
    对所述滤波采样信号进行自适应增益调整，并滤除所述滤波采样信号中的低频信号，得到所述处理采样信号。
  4. 根据权利要求1所述的识别方法，其特征在于，所述对所述处理采样信号进行特征提取，得到目标电弧特征的过程，包括：
    剔除所述处理采样信号中的非局部极值点，得到目标采样点；
    按照时间顺序依次从所述目标采样点中选取十个采样点，得到目标采样序列；其中，所述目标采样序列包括采样点D、E、F、G、H、I、J、M、P和S；
    对所述目标采样序列进行特征提取，得到幅值和离散度、幅值比离散度、始终点幅值和离散度、始终点幅值比离散度、时间比离散度、波形个数、斜率离散度、斜率差分离散度以及点和离散度；
    其中，所述幅值和离散度的计算表达式为：
    A 1＝|V DF|+|V FH|+|V HJ|+|V JP|；
    所述幅值比离散度的计算表达式为：
    Figure PCTCN2020138919-appb-100001
    式中，V ij中的i和j分别代表所述目标采样序列内各采样点的序列号，V DF＝max(y D,y E,y F)-min(y D,y E,y F)，V FH＝max(y F,y G,y H)-min(y F,y G,y H)，V HJ＝max(y H,y I,y J)-min(y H,y I,y J)，V JP＝max(y J,y M,y P)-min(y J,y M,y P)，y D、y E、y F、y G、y H、y I、y J、y M、y P分别为采样点D、E、F、G、H、I、J、M和P在世界坐标系中的纵坐标值；
    所述始终点幅值和离散度的计算表达式为：
    A 3＝|VS DF|+|VS FH|+|VS HJ|+|VS JP|；
    所述始终点幅值比离散度的计算表达式为：
    Figure PCTCN2020138919-appb-100002
    式中，VS ij中的i和j分别代表所述目标采样序列内各采样点的序列号，VS DF＝|y D-y F|，VS FH＝|y F-y H|，VS HJ＝|y H-y J|，VS JP＝|y J-y P|，V DF＝max(y D,y E,y F)-min(y D,y E,y F)，V FH＝max(y F,y G,y H)-min(y F,y G,y H)，V HJ＝max(y H,y I,y J)-min(y H,y I,y J)，V JP＝max(y J,y M,y P)-min(y J,y M,y P)，y D、y E、y F、y G、y H、y I、y J、y M、y P分别为采样点D、E、F、G、H、I、J、M和P在世界坐标系中的纵坐标值；
    所述时间比离散度的计算表达式为：
    Figure PCTCN2020138919-appb-100003
    式中，T 1为采样点D和F之间的时间间隔，T 2为采样点F和H之间的时间间隔，T 3为采样点J和H之间的时间间隔，T 4为采样点P和J之间的时间间隔；
    所述波形个数的计算表达式为：
    Figure PCTCN2020138919-appb-100004
    式中，
    Figure PCTCN2020138919-appb-100005
    表示向下取整，length(y)为所述目标采样序列的长度；
    所述斜率离散度的计算表达式为：
    A 7＝|X EF-X DE|+|X FG-X EF|+|X GH-X FG|+|X HI-X GH|+|X IJ-X HI|+|X JM-X IJ|+|X MP-X JM|
    所述斜率差分离散度的计算表达式为：
    A 8＝|X FH-X DF|+|X HJ-X FH|+|X JP-X HJ|；
    所述点和离散度的计算表达式为：
    A 9＝|y D|+|y E|+|y F|+|y G|+|y H|+|y I|+|y J|+|y M|+|y P|；
    X FH＝|X GH-X FG|、X DF＝|X EF-X DE|、X HJ＝|X IJ-X HI|、X JP＝|X MP-X JM|、X EF＝|y E-y F|，X DE＝|y D-y E|，X FG＝|y F-y G|，X GH＝|y G-y H|，X HI＝|y H-y I|，X IJ＝|y I-y J|，X JM＝|y J-y M|，X MP＝|y M-y P|；y D、y E、y F、y G、y H、y I、y J、y M、y P分别为采样点D、E、F、G、H、I、J、M和P在世界坐标系中的纵坐标值。
  5. 根据权利要求1所述的识别方法，其特征在于，所述基于神经网络算法，利用所述训练数据建立所述神经网络模型的过程，包括：
    基于卷积神经网络算法，利用所述训练数据建立所述神经网络模型。
  6. 根据权利要求1至5任一项所述的识别方法，其特征在于，所述将所述目标电弧特征输入至神经网络模型，得到目标输出结果的过程，包括：
    将所述目标电弧特征转换为归一化特征矩阵，并将所述归一化特征矩阵输入至所述神经网络模型，得到所述目标输出结果。
  7. 一种故障电弧的识别装置，其特征在于，包括：
    信号采样模块，用于以高频对目标电弧进行采样，得到高频采样信号；
    信号处理模块，用于对所述高频采样信号进行预处理，得到处理采样信号；
    特征提取模块，用于对所述处理采样信号进行特征提取，得到目标电弧特征；
    结果判断模块，用于将所述目标电弧特征输入至神经网络模型，得到目标输出结果，并根据所述目标输出结果判断所述目标电弧是否为故障电弧；
    其中，所述神经网络模型的创建过程包括：
    预先获取正常电弧和故障电弧所对应的电弧特征，得到训练数据；
    基于神经网络算法，利用所述训练数据建立所述神经网络模型。
  8. 一种故障电弧的识别设备，其特征在于，包括：
    存储器，用于存储计算机程序；
    处理器，用于执行所述计算机程序时实现如权利要求1至6任一项所述的一种故障电弧的识别方法的步骤。
  9. 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有计算机程序，所述计算机程序被处理器执行时实现如权利要求1至6任一项所述的一种故障电弧的识别方法的步骤。
PCT/CN2020/138919 2020-01-02 2020-12-24 一种故障电弧的识别方法、装置、设备及介质 WO2021136053A1 (zh)

Priority Applications (3)

Application Number Priority Date Filing Date Title
US17/769,770 US11831138B2 (en) 2020-01-02 2020-12-24 Fault-arc identification method, device and apparatus, and storage medium
CA3153759A CA3153759C (en) 2020-01-02 2020-12-24 Fault-arc identification method, device and apparatus, and storage medium
EP20909461.4A EP4024063B1 (en) 2020-01-02 2020-12-24 Fault-arc identification method, device and apparatus, and storage medium

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202010000565 2020-01-02
CN202010000565.5 2020-01-02

Publications (1)

Publication Number Publication Date
WO2021136053A1 true WO2021136053A1 (zh) 2021-07-08

Family

ID=74923375

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/138919 WO2021136053A1 (zh) 2020-01-02 2020-12-24 一种故障电弧的识别方法、装置、设备及介质

Country Status (5)

Country Link
US (1) US11831138B2 (zh)
EP (1) EP4024063B1 (zh)
CN (1) CN112505512B (zh)
CA (1) CA3153759C (zh)
WO (1) WO2021136053A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2024018119A1 (fr) 2022-07-19 2024-01-25 Hager-Electro Sas Procede, dispositif, appareillage de detection d'un defaut d'arc electrique

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113376474A (zh) * 2021-04-26 2021-09-10 国网天津市电力公司城南供电分公司 一种基于广义s变换的神经网络故障电弧识别系统及方法
CN114330452A (zh) * 2021-12-31 2022-04-12 河北工业大学 训练方法、电弧故障检测方法、装置、电子设备及介质
CN117076933B (zh) * 2023-10-16 2024-02-06 锦浪科技股份有限公司 电弧判断模型的训练、光伏直流电弧检测方法及计算设备

Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
GB2375244A (en) * 2001-02-27 2002-11-06 Smiths Group Plc Arc fault detection system
CN101706527A (zh) * 2009-10-30 2010-05-12 西安交通大学 基于电流高频分量时频特征的电弧故障检测方法
CN103543375A (zh) * 2013-08-26 2014-01-29 上海交通大学 基于小波变换和时域混合特征的交流故障电弧检测方法
CN103823179A (zh) * 2014-01-15 2014-05-28 云南电力试验研究院（集团）有限公司电力研究院 一种非接触式空间高频高压断路器动作特性测量方法
CN105223476A (zh) * 2015-09-16 2016-01-06 北京芯同汇科技有限公司 一种基于高频特征的故障电弧检测方法与装置
US20160370420A1 (en) * 2015-06-16 2016-12-22 Clinton Instrument Company Apparatus and method for spark fault detection and typing
CN108535589A (zh) * 2018-05-21 2018-09-14 彭浩明 一种故障电弧检测方法及装置
CN110320452A (zh) * 2019-06-21 2019-10-11 河南理工大学 一种串联故障电弧检测方法
CN110376489A (zh) * 2019-08-13 2019-10-25 南京东博智慧能源研究院有限公司 一种智能配电网故障电弧检测系统

Family Cites Families (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6876528B2 (en) * 2000-03-04 2005-04-05 Passi Seymour, Inc. Two winding resonating arc fault sensor which boosts arc fault signals while rejecting arc mimicking noise
US8295444B2 (en) * 2008-06-13 2012-10-23 Ikanos Communications, Inc. Systems and methods for performing loop analysis based on un-calibrated single-ended line testing
CN101696986B (zh) * 2009-10-26 2012-06-13 吴为麟 故障电弧检测方法及保护装置
CN204028283U (zh) * 2014-05-31 2014-12-17 浙江合极电气科技有限公司 一种新型的故障电弧检测装置
CN203981767U (zh) * 2014-06-14 2014-12-03 三科电器集团有限公司 故障电弧采样与检测装置
CN105425118B (zh) 2015-10-29 2018-09-04 山东建筑大学 一种多信息融合故障电弧检测方法及装置
CN105718958B (zh) * 2016-01-27 2018-09-21 国网江苏省电力有限公司检修分公司 基于线性判别分析和支持向量机的断路器故障诊断方法
CN108646149A (zh) * 2018-04-28 2018-10-12 国网江苏省电力有限公司苏州供电分公司 基于电流特征提取的故障电弧识别方法
CN112269110A (zh) * 2020-10-19 2021-01-26 合肥阳光新能源科技有限公司 一种电弧故障判定方法

Patent Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
GB2375244A (en) * 2001-02-27 2002-11-06 Smiths Group Plc Arc fault detection system
CN101706527A (zh) * 2009-10-30 2010-05-12 西安交通大学 基于电流高频分量时频特征的电弧故障检测方法
CN103543375A (zh) * 2013-08-26 2014-01-29 上海交通大学 基于小波变换和时域混合特征的交流故障电弧检测方法
CN103823179A (zh) * 2014-01-15 2014-05-28 云南电力试验研究院（集团）有限公司电力研究院 一种非接触式空间高频高压断路器动作特性测量方法
US20160370420A1 (en) * 2015-06-16 2016-12-22 Clinton Instrument Company Apparatus and method for spark fault detection and typing
CN105223476A (zh) * 2015-09-16 2016-01-06 北京芯同汇科技有限公司 一种基于高频特征的故障电弧检测方法与装置
CN108535589A (zh) * 2018-05-21 2018-09-14 彭浩明 一种故障电弧检测方法及装置
CN110320452A (zh) * 2019-06-21 2019-10-11 河南理工大学 一种串联故障电弧检测方法
CN110376489A (zh) * 2019-08-13 2019-10-25 南京东博智慧能源研究院有限公司 一种智能配电网故障电弧检测系统

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
See also references of EP4024063A4

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2024018119A1 (fr) 2022-07-19 2024-01-25 Hager-Electro Sas Procede, dispositif, appareillage de detection d'un defaut d'arc electrique

Also Published As

Publication number Publication date
CA3153759A1 (en) 2021-07-08
EP4024063A1 (en) 2022-07-06
CA3153759C (en) 2023-04-04
EP4024063A4 (en) 2023-01-11
US11831138B2 (en) 2023-11-28
CN112505512B (zh) 2023-01-10
EP4024063B1 (en) 2024-03-20
US20220390501A1 (en) 2022-12-08
CN112505512A (zh) 2021-03-16
EP4024063C0 (en) 2024-03-20

Similar Documents

Publication Publication Date Title
WO2021136053A1 (zh) 一种故障电弧的识别方法、装置、设备及介质
WO2021212891A1 (zh) 一种采用卷积神经网络的故障电弧信号检测方法
WO2021068454A1 (zh) 一种基于bp神经网络的微能源器件能量识别方法
Biswal et al. Detection and characterization of multiple power quality disturbances with a fast S-transform and decision tree based classifier
CN108030494A (zh) 基于交叉验证的心电信号错误标记训练样本识别方法
Zhang et al. A new method for on-line monitoring discharge pulse in WEDM-MS process
CN108508297B (zh) 一种基于突变系数和svm的故障电弧检测方法
CN106504760B (zh) 宽频带背景噪声与语音分离检测系统及方法
CN117113180A (zh) 基于vmd与卷积神经网络的直流配网高阻故障识别方法
JP4760614B2 (ja) 信号識別装置の学習データの選択方法
CN112284707B (zh) 一种断路器振动信号的处理方法
CN113920062A (zh) 一种红外热成像电力设备故障检测方法
CN115112989B (zh) 基于低频数据的非侵入式负荷监测方法
CN116913316A (zh) 一种基于Mosaic数据增强的电力变压器典型故障声纹诊断方法
CN115963350A (zh) 一种直流配电网故障定位方法及装置
Thirumala et al. A generalized empirical wavelet transform for classification of power quality disturbances
CN111382790A (zh) 一种基于小波变换的高压直流故障分类方法
CN112348159A (zh) 一种故障电弧神经网络优化训练方法
CN110647838B (zh) 一种用于设备工作状态识别的基于几何形状的一维信号特征提取方法
CN114155878B (zh) 一种人工智能检测系统、方法和计算机程序
CN112927710B (zh) 一种基于无监督方式的电力变压器工况噪声分离方法
CN116304975A (zh) 一种基于多特征融合的电弧模型判别方法、装置及介质
Naderian et al. An implementation of S-transform and type-2 fuzzy kernel based support vector machine algorithm for power quality events classification
CN113855051A (zh) 一种决策树式脑电噪声识别系统
Yan et al. An Integrated Target Recognition Method Based on Improved Faster-RCNN for Apple Detection, Counting, Localization, and Quality Estimation

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: 20909461

Country of ref document: EP

Kind code of ref document: A1

ENP Entry into the national phase

Ref document number: 3153759

Country of ref document: CA

ENP Entry into the national phase

Ref document number: 2020909461

Country of ref document: EP

Effective date: 20220331

NENP Non-entry into the national phase

Ref country code: DE