CN113297786B - Low-voltage fault arc sensing method based on semi-supervised machine learning - Google Patents

Low-voltage fault arc sensing method based on semi-supervised machine learning Download PDF

Info

Publication number
CN113297786B
CN113297786B CN202110491988.6A CN202110491988A CN113297786B CN 113297786 B CN113297786 B CN 113297786B CN 202110491988 A CN202110491988 A CN 202110491988A CN 113297786 B CN113297786 B CN 113297786B
Authority
CN
China
Prior art keywords
arc
fault
fault arc
current
low
Prior art date
Legal status (The legal status 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 status listed.)
Active
Application number
CN202110491988.6A
Other languages
Chinese (zh)
Other versions
CN113297786A (en
Inventor
陈思磊
王源丰
同向前
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Xian University of Technology
Original Assignee
Xian University of Technology
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 Xian University of Technology filed Critical Xian University of Technology
Priority to CN202110491988.6A priority Critical patent/CN113297786B/en
Publication of CN113297786A publication Critical patent/CN113297786A/en
Application granted granted Critical
Publication of CN113297786B publication Critical patent/CN113297786B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • 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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • 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
    • Y04S10/52—Outage or fault management, e.g. fault detection or location

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Software Systems (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Medical Informatics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Computer Hardware Design (AREA)
  • Geometry (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Testing Relating To Insulation (AREA)

Abstract

The invention discloses a low-voltage fault arc sensing method based on semi-supervised machine learning, which comprises the steps of reconstructing wavelet coefficients at corresponding nodes of a fault arc characteristic frequency band after wavelet packet decomposition of detection signals in an analysis period, obtaining characteristic values on the corresponding frequency band through absolute value summation, carrying out state classification on the characteristic values to a machine based on a safe semi-supervised support, judging the condition of fault arc generation in a low-voltage electric system, and identifying the type of the fault arc by combining the characteristic values obtained by carrying out short-time Fourier transform and specific frequency band component accumulation on the detection signals under the condition of sensing the fault arc generation. According to the method, under the condition of limited sample label training, the calculated amount of a classification model is reduced, relatively high fault electric arc sensing accuracy is obtained, and meanwhile, accurate judgment on series-connection and parallel-connection fault electric arcs is increased, so that arc extinguishing measures can be optimized to the maximum extent, and the operation and maintenance efficiency of a system is improved.

Description

Low-voltage fault arc sensing method based on semi-supervised machine learning
Technical Field
The invention belongs to the technical field of electrical fault detection, and particularly relates to a low-voltage (below 1000V) fault arc sensing method based on semi-supervised machine learning.
Background
The occurrence of a fault arc can result in a serious electrical fire disaster on the low-voltage distribution line. Due to the characteristics of randomness, concealment, complexity and the like, the fault arc is difficult to detect. In particular, the detection difficulty of the series fault arc is higher than that of the parallel fault arc, so that the technical difficulty of the fault protection system is solved. Traditional fault arc detection methods fall into the following categories: the application condition of the detection method based on the simulation model is easy to limit, and the load of the model parameters mostly stays in the simulation stage; physical property based detection methods sensor locations are subject to limitations; the detection method based on the time domain characteristics is easily influenced by the arc-like waveform; the detection method based on the frequency domain characteristics is easy to be interfered by a system to cause misjudgment.
The detection method designed on the basis of carrying out intelligent informatization on the fault arc characteristics by utilizing technologies such as intelligent calculation, adaptive control algorithm and the like can effectively break through the limitation of the traditional detection method. The unsupervised learning does not use prior information, and similar samples are gathered by using the characteristic distribution rule of the label-free samples, but the accuracy of the model is difficult to ensure. After supervised learning is trained by using a small number of samples with labels, the generalization performance of a learner is often low, and a large amount of unlabelled sample data resources are wasted. Therefore, semi-supervised learning using a large number of unlabeled exemplars to improve learning performance using a small number of labeled exemplars as guidance has been the focus of research. Semi-supervised learning includes semi-supervised support vector machines (S3 VM) and direct-push support vector machines (TSVM), and in order to make the calculation faster, meanS S3VM is proposed, but with some loss of accuracy. In order to prevent performance degradation when using non-labeled data, a secure semi-supervised support vector machine (S4 VM) is proposed.
In the field of ac fault arc detection, arc fault circuit breakers (AFCI) are produced by major manufacturers in accordance with the UL1699 standard, such as eaton, general electric, texas instruments, etc., but arc fault circuit breakers designed for this purpose are only suitable for typical 120V power distribution systems. The main reasons for the poor quality of low-voltage ac power system fault arc detection related products, such as arc fault protection devices (AFDDs), are that the low-voltage ac power system has diversified power supply scenarios, each power supply scenario differs according to the type of load and the number of loads, and the fault arc identification difficulty is greater (especially for power supply scenarios with various interference loads). In the case where a plurality of loads are simultaneously connected, it is also difficult to locate a fault line after the arc fault is cut off. Meanwhile, the parallel fault arc generated on the front side of the circuit breaker cannot be effectively protected by the conventional AFDD, and the risks of difficult extinguishment and fire accidents are still existed.
China CN110376497A discloses a low-voltage distribution system series fault arc identification method based on full-phase deep learning, which carries out full-phase discrete Fourier transform on collected current signals and carries out deep learning training on full-phase frequency spectrum characteristic quantities under different loads and different running states based on a Logistic regression deep learning neural network model. However, the patent cannot simultaneously realize accurate detection of different types of fault arcs under various load disturbance conditions.
Disclosure of Invention
The invention aims to provide a low-voltage fault arc sensing method based on semi-supervised machine learning, so as to realize accurate, reliable and quick identification of fault arcs and arc-like working conditions (such as error tripping working conditions) of series connection and parallel connection of a low-voltage alternating-current power utilization system.
In order to achieve the purpose, the invention adopts the following technical scheme:
a low-voltage fault arc sensing method based on semi-supervised machine learning comprises the following steps:
1) In a low-voltage alternating-current power system, current signals of different branches are sampled point by frequency f, or current signals of different branches and a superior bus thereof are sampled point by point, and the sampling is carried out according to the time window length T s Respectively extracting N current signal sampling points corresponding to the current analysis period to obtain a detection signal x n Wherein N =1, \ 8230;, N, go to step 2);
2) For the detection signal x n Decomposing the wavelet packet to obtain a detection signal x n Turning to step 3) for the wavelet packet tree;
3) Selecting nodes corresponding to frequency range dimensions (namely fault arc characteristic frequency ranges) capable of displaying fault arc time-frequency characteristics from all decomposition layers of the wavelet packet tree, and performing wavelet coefficient reconstruction on the nodes to obtain y n N, summing absolute values of N numbers obtained by reconstruction to obtain a characteristic value (value of characteristic quantity) I of an analysis period, and turning to step 4);
4) And (3) judging whether a fault arc event exists by using the output value of the S4VM classifier: inputting the characteristic value I into an S4VM classifier, if the S4VM classifier outputs a value (for example, the output value is a high level of 1) corresponding to a fault arc event occurring in the system, judging that fault arc is likely to occur in the system, and turning to the step 5) to perform further analysis, otherwise, judging that the system is in normal operation, and returning to the step 1) to continue to perform detection signal analysis in the next analysis period (time window);
5) And judging whether the number of the output values of the S4VM classifier meets the cut-off standard or not: if the S4VM classifier outputs values corresponding to fault arc events in the system in K continuous analysis periods (for example, the fault arc cutting standard is that the classifier outputs 1 and K in K continuous time windows are used as fault arc cutting signal trigger thresholds), judging that fault arcs occur in corresponding branches or buses in the system, namely the number of the output values reaches the fault arc cutting standard, and sending out control signals for cutting off the corresponding fault arc branches or fault arc buses; otherwise, judging that the arc-like working condition occurs in the system, and returning to the step 1) to continue to analyze the detection signal in the next analysis period (time window).
Preferably, under the permission of a sampling device, the effective frequency band obtained by sampling can contain a fault arc characteristic frequency band (the fault arc characteristic frequency band reflects the fundamental difference between a fault arc and a false tripping working condition; the sampling frequency is twice of the maximum frequency in the fault arc characteristic frequency band), and the value range of f is 24 kHz-400 kHz; the relation between the time window length and the sampling frequency is T s And = N/f, wherein N is the number of sampling points of the detection signal in the time window, the selection principle of the number of the sampling points is that the detection signal in the time window with the determined length can reflect the effective time-frequency characteristics of the fault arc, and the value range of N is 240-4000.
Preferably, each parameter used in wavelet packet transformation (including wavelet packet decomposition and reconstruction) is based on maximum separation of series fault electric arcs, parallel fault electric arcs, false tripping working conditions and normal states, and by combining the consideration of the size of wavelet basis occupation storage and the like, the wavelet basis adopted in wavelet packet decomposition is Haar, the characteristic frequency band of the fault electric arcs is 9 kHz-12 kHz, and the characteristic frequency band is unrelated to the sampling frequency f.
Preferably, according to the obtained arc characteristic sample data (the characteristic value I), and on the basis of the principle that the classification effect is optimal under different conditions (including household power systems and power systems with similar applications and the like) and the generalization performance is guaranteed, the kernel function of the S4VM classifier is a radial basis kernel function, and according to the principle that the accuracy is higher and the running time is shortest, the sample time is 10-50, the smaller the sample time is, the shorter the program running time is, the parameter C1 is 50-70, and the parameter C2 is 0.001-0.08.
Preferably, based on separating the series fault arc, the parallel fault arc, the error tripping working condition and the normal state under various conditions to the maximum extent, and according to the priority of the reliability principle under the condition of the low-current arc and the priority of the quick-action principle under the condition of the high-current arc, two conditions are set for the value of K:
in the case of a low current (I.ltoreq.63A) arc, then K is calculated according to the following formula:
Figure BDA0003052796620000031
wherein, I represents the current of the corresponding branch or bus;
under the condition of large-current (I > 63A) arc, K takes the value as follows: k =4, 5.
Preferably, the fault arc sensing method further comprises the following steps: 6) Based on detection signals before and after occurrence of a fault arc (including detection signals x extracted respectively in one or more analysis periods before and after an analysis period for determining occurrence of a fault arc n ) And corresponding wavelet characteristic shape (according to each detection signal x) n Respectively obtained characteristic values I), applying and generating a countermeasure network to establish a correlation model of the load and the arc, identifying the external interference load type of a branch where the fault arc is located (namely, a fault arc branch) according to the correlation model, and determining and displaying load information required by fault arc position troubleshooting work for operating and maintaining after the fault arc is cut off according to the identified interference load type.
Preferably, in order to make a determination on the type of the fault arc when it is determined that the fault arc occurs in the system, the fault arc sensing method further includes the following steps:
7) For the detection signal x n Carrying out short-time Fourier transform while carrying out wavelet packet transform to obtain detection signal x n In a matrix distribution form (time-frequency domain component matrix) in a time-frequency domain, turning to the step 8);
8) Selecting specific components (corresponding to matrix elements) of frequency band dimensions capable of displaying time-frequency characteristics of the fault arc from the matrix elements obtained by short-time Fourier transform, accumulating to obtain a characteristic value II of an analysis period, and turning to the step 9);
9) Storing the characteristic values I and II obtained by the two conversions according to analysis periods, and emptying the stored numerical values (the characteristic values I and II) every 300-800 (for example, 500) analysis periods if the fault arc is judged not to occur according to the step 5); if the fault arc is judged to occur according to the step 5), turning to the step 10) for further analysis;
10 Respectively normalizing the stored characteristic values I and II of each analysis period (at least comprising the current analysis period, and at least one analysis period before the current analysis period when the analysis period is not emptied), then solving the Euclidean distance between the normalized characteristic values I and II, and turning to the step 11);
11 Comparing the obtained Euclidean distance with a given threshold, if the comparison result is smaller than the threshold, judging that the type of the fault arc occurring in the current analysis period is a parallel fault arc, and if the comparison result is larger than or equal to the threshold, judging that the type of the fault arc occurring in the current analysis period is a series fault arc.
Preferably, according to the principle that parallel fault arcs and serial fault arcs can be distinguished, the frequency range of matrix element accumulation in the frequency dimension of a matrix obtained through short-time Fourier transform is the fault arc characteristic frequency band (9 kHz-12 kHz); the value range of the threshold is 3.1-3.5 according to the principle that the fault arc type can be distinguished.
Preferably, the circuit breaker of the corresponding fault arc branch or fault arc bus is immediately cut off after the parallel fault arc is judged to occur in the system, if the fault arc can still be sensed after the circuit breaker is cut off, the parallel fault arc is determined to occur at the front side of the circuit breaker, and at the moment, the arc is extinguished by cutting off the power supply of the fault arc branch or fault arc bus; and immediately cutting off the circuit breaker of the corresponding fault arc branch or fault arc bus after judging that the series fault arc occurs in the system.
An arc fault protection electrical appliance based on semi-supervised machine learning comprises a current signal sampling device and a low-voltage fault arc sensing module;
the current signal sampling device is used for sampling a current signal of a branch or a bus where the arc fault protection electric appliance is located, and extracting N corresponding current signal sampling points as a detection signal x according to a time window n (ii) a I.e. the current signal sampling means may be used to perform step 1);
the low-voltage fault arc sensing module is used for sensing the low-voltage fault arc according to the detection signal x n Respectively combining wavelet packet transformation and short-time Fourier transformation to obtain a characteristic value I and a characteristic value II which reflect time-frequency characteristics of the fault arc, judging the occurrence of the fault arc on a branch or a bus by applying an S4VM classifier and a fault arc cutting standard according to the characteristic value I, and judging the type of the fault arc according to the stored Euclidean distance between the characteristic value I and the characteristic value II of each analysis period and a given threshold; i.e., the low-voltage fault arc sensing module may be used to perform steps 2) through 5) above, and to perform steps 7) through 11).
Preferably, the arc fault protection electrical appliance further comprises an interference load type identification module, which is used for identifying the type of the interference load according to the established association model of the load and the arc, and determining load information required by the fault arc position troubleshooting work for operating and maintaining after the fault arc is cut off; i.e. the interference load type identification module may be used to perform step 6).
Preferably, the current signal sampling device, the low-voltage fault arc sensing module and the interference load type identification module are integrated in a circuit breaker on a branch or a bus.
The invention has the beneficial effects that:
the invention provides a method for detecting a fault arc of a low-voltage alternating-current power system based on a time-frequency characteristic quantity obtained by wavelet packet transformation and combined with an S4VM classifier and a fault arc cut-off standard, which can obtain a more excellent fault arc sensing effect (compared with an SVM) by using fewer samples with labels for state classification, effectively distinguish the false tripping situation of normal start-stop operation of a load, avoid false operation of an arc fault protection electric appliance under the false tripping working condition, and simultaneously ensure that the consumption of manpower and time is reduced in the acquisition and analysis processes of detection signals.
Furthermore, the method can realize accurate identification of the low-voltage fault arc type under load interference conditions in different external environments by jointly applying the time-frequency characteristic quantities obtained by short-time Fourier transform and based on the calculation of Euclidean distance between the characteristic quantities and threshold judgment.
Furthermore, according to the comparison result of the accuracy, the parameter value ranges of C1, C2 and sample time are determined, so that the problem that the parameters need to be optimized and adjusted under different conditions is solved, the influence on the accuracy of the final S4VM classifier is eliminated, and the generalization performance is better.
Drawings
Fig. 1 is a flow chart of low-voltage fault arc sensing in an embodiment of the invention.
Fig. 2a is a circuit diagram of a series fault arc simulation in an embodiment of the present invention.
Fig. 2b is a circuit diagram of a parallel fault arc simulation circuit in an embodiment of the present invention.
FIG. 3a is a system output current signal for series fault arc sensing under vacuum cleaner load conditions using the present invention.
FIG. 3b is a characteristic output waveform for low voltage fault arc sensing using wavelet packet transformation.
Fig. 3c is a system state real-time judgment output signal for sensing the fault arc of the low-voltage system by applying the invention.
FIG. 4a is a system output current signal for parallel fault arc sensing under 75A current limited arc generation conditions using the present invention.
FIG. 4b is a characteristic output waveform for low voltage fault arc sensing using wavelet packet transformation.
Fig. 4c is a system state real-time judgment output signal for sensing the fault arc of the low-voltage system by applying the invention.
Fig. 5a is a system output current signal for arc-like interference sensing under the condition of a switching power supply load by applying the invention.
FIG. 5b is a characteristic output waveform for low voltage system arc-like interference sensing using wavelet packet transformation.
Fig. 5c is a system state real-time judgment output signal for sensing the arc-like interference of the low-voltage system by applying the invention.
Fig. 6 shows the fault arc type determination for sensing the fault arc in the low-voltage system by applying the invention.
FIG. 7a is a graph of the effect of tuning parameter C1 on the S4VM classifier.
FIG. 7b is a graph of the effect of adjusting parameter C2 on the S4VM classifier.
FIG. 7c is a graph of the effect of adjusting the parameter sample time on the S4VM classifier.
Detailed Description
The present invention will be described in further detail with reference to the drawings and examples.
The invention combines the safe semi-supervised support vector machine with the characteristic quantity which can be used for distinguishing the fault arc from the normal working condition to realize accurate fault arc recognition, and after comparing the fault arc recognition effect with the fault arc recognition effect of the traditional supervised machine learning, the invention reflects the advantage of adopting the semi-supervised machine learning, and also embodies that the semi-supervised machine learning has greater application value and potential in the fault arc field. The invention also provides measures for arc quenching and troubleshooting of relevance.
Low-voltage fault arc detection algorithm framework
The method comprises the steps of firstly collecting and extracting characteristic values of currents under the conditions of series fault arcs, parallel fault arcs and false tripping, marking some samples according to the different conditions to form samples with labels, and then inputting the samples with the labels and the samples without the labels into an S4VM classifier for training. And finally, extracting characteristic values by combining real-time sampling and sampled signals and inputting the characteristic values into a trained S4VM classifier to finish judgment.
When whether a fault arc occurs in a low-voltage system is actually analyzed, only a current signal in a time window to be identified needs to be used as input of a time-frequency transformation tool, a time-frequency characteristic value is obtained and then input into an S4VM classifier to be judged, the S4VM classifier can output a-1/1 judgment result of whether the fault arc occurs in the system, the output is 1 when the fault arc possibly occurs, and the output is-1 when the system normally operates. The judgment of the fault arc cut-off signal trigger threshold is carried out only when the output of the S4VM classifier is 1, otherwise, the normal operation of the low-voltage system is confirmed, and no fault arc occurs, which also indicates that the time window of the system in most normal states can be used for directly carrying out the analysis of the detection signal in the next time window by carrying out the fault arc judgment process by one step, and is beneficial to the improvement of the fault arc detection speed of the low-voltage fault arc detection algorithm. If the S4VM classifier continuously outputs high level 1 in a plurality of analysis periods (time windows), and before the period number of the set output 1 is not reached, as long as one low level-1 output exists, the fault is considered to be caused by the error tripping working condition rather than the real fault arc working condition. And after the high-level output cycle number reaches the trigger threshold, confirming that the fault arc occurs, and finally sending a fault arc branch cutting signal by the detection algorithm to ensure that the low-voltage system is not damaged by the fault arc working condition.
Low-voltage fault arc sensing method combining (II) semi-supervised machine learning with time-frequency characteristics
Referring to fig. 1, the specific steps are as follows:
the parameter initialization process comprises the steps of setting the sampling frequency f of a detection signal sampling device on a current signal, the number N of sampling points in a time window, a fault arc cutting signal triggering threshold K, parameters in a wavelet packet isochronous frequency analysis tool and the like.
During the operation of the low-voltage system, the output current signal of the system is sampled point by the frequency f, and the length T of a set time window is used s Analyzing the current signal, and performing subsequent steps on the current detection signal x in a time window n And performing a time-frequency analysis process. Considering that on the one hand the time window is too small requires constant considerationTime-frequency analysis is performed and on the other hand the time window is too long, which may lead to a time lag in detecting a faulty arc. Therefore, the number of sample points N =244 within the time window is selected. In order to reduce the hardware realization requirement of the detection signal sampling device and reflect the fault arc characteristic frequency band of the difference between the fault arc and the error tripping working condition, the sampling frequency f of the output current signal of the low-voltage system is 24.3kHz.
Aiming at the maximum breaking time under the current of a small arc and the maximum half wave number within 0.5s under the current of a large arc in the national standard GB _ T31143-2014, two conditions are set according to the priority of the reliability principle under the condition of the small arc and the priority of the quick action principle under the condition of the large arc:
Figure BDA0003052796620000071
K＝4(I>63A)。
the effective filtering effect of wavelet packet transformation and the storage length of a filter are considered, the Haar is selected as a wavelet base, the transformation parameters selected above can save storage space, and normal and fault states under the working conditions of series connection, parallel connection fault electric arcs and error tripping can be accurately distinguished.
Step two, adopting a wavelet packet conversion method to detect the current detection signal x n And analyzing to obtain wavelet packet trees of the wavelet packets on each decomposition layer and node, and selecting the wavelet nodes corresponding to 9-12 kHz to reconstruct the coefficients.
For the wavelet packet tree, each node represents a time-frequency characteristic quantity on a frequency band, and the current signal is decomposed into 4 frequency bands (the nodes are respectively numbered as 0,1, 2 and 3) by selecting the number of decomposition layers as 2. According to the Nyquist sampling law, effective information is distributed in a range from 0kHz to 12.15kHz when the sampling frequency is 24.3kHz, and observation shows that a wavelet coefficient of a fault arc characteristic frequency band from 9kHz to 12kHz should be selected as a reconstruction object, the series fault arc time-frequency characteristic smaller than the frequency band is influenced by a direct current component to become inaccurate, and the 3A resistive load sudden-adding moment and the fault arc occurrence moment cannot be effectively distinguished. In particular in [2,3 ]]Wavelet coefficients at nodes (i.e., node numbered 3)Line processing, namely performing wavelet coefficient reconstruction on the node to obtain y n 。
And step three, obtaining a time-frequency characteristic value under each time window through absolute value summation.
Characteristic quantity formed according to the formula:
Figure BDA0003052796620000072
and step four, inputting the obtained characteristic value (wavelet characteristic value) of a time window into the S4VM classifier, judging whether a fault arc exists or not according to the output value of the S4VM classifier, outputting-1 to represent that the system in the time window is in a normal operation state, outputting 1 to represent that the system in the time window is possible to generate the fault arc, and turning to step five.
Due to the used time-frequency characteristics and the S4VM classifier, the characteristic values can be not normalized, and the model parameters of machine learning can be not optimized by multiple parameters, so that the detection time of the fault arc of the system is shortened, the judgment process of the fault arc detection algorithm in the system is simplified, and the method is suitable for the real-time processing process.
Step five, preliminarily judging the current running state of the system according to the output value of the S4VM classifier, judging that the system in the time window is in a normal running state if the output value is-1, and returning to the step one to analyze the output current signal of the system in the next time window; if 1 is output, the system is judged to possibly generate the fault arc in the time window, and the generation of the fault arc is further judged and confirmed through the following standards: whether the number of the cycles of the continuous output 1 reaches a fault arc cutting signal triggering threshold value or not, if yes, determining that fault arc occurs in the system, and turning to the sixth step; and if not, judging that the error tripping working condition of the system forms insufficient continuous 1 outputs, and returning to the first step to analyze the output current signal of the system in the next time window.
The characteristic quantity used by the invention has stronger fault arc identification capability, and the design of the characteristic quantity not only avoids misoperation caused by accidental factors, but also ensures the rapidity of sending out fault arc branch circuit cutting signals.
Step six, to the time windowInternal current detection signal x n And performing short-time Fourier transform processing, setting the window length to be the same as wavelet packet decomposition, and obtaining the matrix distribution form of the detection signal in the time-frequency domain, wherein the window type is a rectangular window. And selecting specific components corresponding to 9-12 kHz on the frequency dimension of the matrix elements, accumulating to obtain characteristic quantity, and calling the characteristic value corresponding to a time window as a Fourier characteristic value. The sequence of characteristic values (stored) obtained by the two transformations is used<Feature values of 500 time windows) are respectively normalized, and the feature values are distributed in [0,1 ]]Between the ranges:
Figure BDA0003052796620000081
wherein, S is the normalization result of the characteristic value X, max (X) is the maximum characteristic value in the sequence, and min (X) is the minimum characteristic value in the sequence.
And (3) solving Euclidean distances between the normalized Fourier characteristic value and two sequences of the wavelet characteristic values obtained in the step three:
Figure BDA0003052796620000082
wherein M is the number of characteristic values in each sequence, S 1 Representing a sequence of wavelet characteristic values, S 2 Representing a sequence of fourier eigenvalues.
And then judging the type of the fault arc: comparing the obtained Euclidean distance with a given threshold, if the Euclidean distance is smaller than the threshold, judging that the arc is a parallel fault, and if the Euclidean distance is larger than or equal to the threshold, judging that the arc is a series fault; the value range of the threshold is 3.1-3.5.
According to the type of the fault arc obtained by judgment, a targeted correlation measure can be taken, namely if the parallel fault arc occurs, the circuit breaker of the corresponding fault arc branch is cut off, so that the fault arc can still be sensed, the parallel fault arc is considered to occur at the front side of the circuit breaker, and the arc is extinguished by cutting off the branch power supply; and if the series fault arc occurs, the circuit breaker of the corresponding fault arc branch is cut off.
And seventhly, identifying the external interference load type of the branch where the fault arc is located by using the generation countermeasure network according to the detection signals before and after the fault arc occurs and the wavelet characteristic forms, realizing the primary positioning of the fault arc in the branch, and displaying the load information for the operation and maintenance troubleshooting work after the fault arc is cut off.
(III) simulation of fault arc and collection of current data
Referring to fig. 2a, switches S1, S2, S3, S4 and AFDD are in the closed position, allowing stable operation of the inhibitory load, regulated by the resistive load. And then the switch S2 is disconnected, the switch S4 is suddenly disconnected, a prepared cable sample connected with the load in series is accessed, the rated voltage is applied to generate series fault arc, and the analog circuit is disconnected after enough sample data is acquired.
Referring to fig. 2b, the switches S1, S2, S3 and S4 are in the closed position, the test current is adjusted to a fixed value through the line impedance Z in the circuit, the switches S2, S3 and S4 are opened, the AFDD and the switch S1 are closed, then the switch S3 is closed suddenly to generate a parallel fault arc, and the analog circuit is opened after enough sample data is acquired.
The low-voltage fault arc detection algorithm can be applied to accurate identification of fault arcs and false tripping working conditions under different power supply scenes by changing the external interference load types and updating the fault arc learning sample database.
(III) the sensing result of the invention under the conditions of series fault arc, parallel fault arc and false tripping
3.1 perception of series-connected Fault arcs with shielded loads (vacuum cleaners) engaged
As shown in fig. 3a, the current detection signal outputted in the case of the analog circuit (fig. 2 a) is acquired at a sampling frequency of 24.3kHz, and the current is initially large during the starting process of the cleaner and then gradually decreases to a current value corresponding to the normal operation state. After 2.78s, the combined turn-on of the 3A resistive load results in an increase in loop current. After a period of operation of 2.891s, the occurrence of series fault arcing was simulated, and then the maximum value of the current waveform was reduced and arc-induced current amplitude instability was exhibited. After 3.93s, the AC switch is switched off, the load stops working, the fault arc is extinguished, and the loop current is reduced to zero.
The time-frequency characteristic quantity subjected to wavelet packet transformation and absolute value summation is shown in fig. 3b, and the amplitude of the time-frequency characteristic quantity begins to become large after the fault arc occurs, so that the time-frequency characteristic quantity has a significant distinguishing level from the characteristic value in a normal state. The characteristic value is not affected by the normal increase of the current amplitude caused by introducing a new load, and the characteristic value is restored to a normal level after the alternating current switch is switched off, so that the accuracy of S4VM classification is ensured. As shown in fig. 3c, the semi-supervised machine learning real-time status identification signal starts to continuously output high levels after 2.891s, sensing and arc extinguishing of the fault arc are completed after reaching the trigger threshold, and a plurality of discrete high levels caused by switching operation are output after the fault arc occurs, and false triggering cannot be caused if the discrete high levels do not reach the trigger threshold.
3.2 perception of parallel fault arcs generated in 75A Current limiting arc Generation mode
As shown in fig. 4a, the current detection signal outputted in the case of the analog circuit (fig. 2 b) was acquired at a sampling frequency of 24.3kHz, and the circuit was in a normal state before 2.76s, and no fault arc was generated. And between 2.76s and 3.14s, the collected current becomes larger due to the generation of fault arc. After 3.14s, the opening of the ac switch restores the loop current to zero.
The time-frequency characteristic quantity processed by wavelet packet transformation and absolute value summation is shown in fig. 4b, and has good response to the generation of the fault arc, and the value of the characteristic quantity is larger in the time of generating the fault arc in the middle. As shown in fig. 4c, the semi-supervised machine learning real-time state identification signal starts to continuously output high levels after 2.76s, sensing and arc extinguishing of the fault arc are completed after the trigger threshold is reached, and accidental single high levels occurring before and after the fault arc cannot cause false triggering because the accidental single high levels do not reach the trigger threshold.
3.3 perception of the false tripping condition under the shielding load condition of the switch-in power supply
As shown in fig. 5a, the shielding load in the test circuit (fig. 2 a) is replaced by a switching power supply, a current detection signal output by the system is obtained at a sampling frequency of 24.3kHz, a large starting current is generated due to starting at 1.07s, and after 1.31s, normal operation is resumed until 7.21s is ac-disconnected.
The time-frequency characteristic quantity subjected to wavelet packet transformation and absolute value summation processing is shown in fig. 5b, and it can be seen that high-amplitude characteristic quantity of 1 time window appears due to the influence of the starting state, and then the high-amplitude characteristic quantity is not significantly influenced by the normal operation of the switching power supply, the value of the corresponding characteristic quantity is not changed violently, and the accuracy of S4VM classification is ensured. As shown in fig. 5c, the semi-supervised machine learning real-time status recognition signal, although including a single high level occurring during the start-up process, does not reach the trigger threshold and does not cause false triggers.
(IV) Process for determining the type of Fault when a Fault arc occurs in the System
As shown in fig. 6, after the occurrence of the fault arc is determined, the stored feature values obtained through short-time fourier transform and wavelet packet transform are respectively normalized, euclidean distances are obtained for the two feature value sequences after normalization, and it is seen from fig. 6 that the euclidean distance of the series fault arc is greater than that of the parallel fault arc, which is also the basis by which the semi-supervised machine learning method can effectively distinguish the series fault arc from the parallel fault arc. After the type of the generated fault arc is determined, corresponding measures are taken to extinguish the arc, so that the fire accident caused by the fact that some parallel fault arcs are difficult to cut under the condition of the existing series fault arc protection action is avoided.
(V) influence of parameter selection on classification accuracy of S4VM
The influence of the model parameters on the perception accuracy is explored by selecting the S4VM kernel function as a radial basis kernel function or a linear kernel function and changing the values of the parameters C1, C2 and sample time, and the specific process is as follows:
in order to perform accuracy comparison with supervised machine learning (for example, SVM) under different parameters of S4VM, marking is performed according to the obtained samples, a small amount of samples are used as labeled samples, and a large amount of samples are used as unlabeled samples; changing parameters of C1, C2 and sample time to obtain the accuracy rate of the S4VM under the condition of the same label sample and the label-free sample; comparing with supervised machine learning, using the same labeled sample and unlabeled sample, wherein the ratio of the labeled sample for S4VM training is far smaller than that of the SVM, performing cross parameter optimization on the SVM, and comparing the accuracy; and S4, the proportion of label samples used by the VM is further reduced, and the accuracy of the label samples is continuously compared with the accuracy of the label samples.
Under the condition that the three parameters are changed, the accuracy rate of sensing the fault arc can be kept above 96%, and it can be known that the S4VM is not sensitive to the setting of the parameters. The reason is that the S4VM considers a plurality of low-density separators to generate a final prediction label, and the characteristic also enables the S4VM to have a wider application range, the label can be optimally distributed under the condition that the arc characteristics are weak, and a better perception effect can be obtained in a plurality of different application scenes by applying the same group of parameters. The kernel function type, g and c parameter values of the supervised SVM classifier have a great influence on the classification effect, and in the actual application process, global optimization is generally required to obtain a better classification result.
As shown in fig. 7a, 7b, and 7C, although the accuracies under different kernel functions are similar, the radial basis kernel function is a nonlinear kernel function, in order to ensure the generalization performance of the S4VM, the kernel function of the S4VM classifier is selected as a radial basis, the smaller the value of the sample time is, the shorter the program running time is, the shorter the running time is under the condition of ensuring the accuracy by comparison, the running time is made as short as possible, for this reason, the sample time is selected to be 10, the parameter C1 is selected to be 50 or 70, the parameter C2 is selected to be in the range of 0.001 to 0.08, and the ratio of the adopted label samples is 21%.
As shown in table 1, the SVM classification judgment is performed by using the labeled samples and the unlabeled samples which are the same as those of S4VM, the SVM model at this time is the optimal model parameter which has been obtained by using the cross validation method, and the corresponding perception accuracy is 95.47%, whereas under the condition that several typical waveforms in the above-mentioned section (third) are used as input, the perception accuracy of S4VM is 96.18%, which is higher than that of the SVM classifier. By reducing the proportion of label samples to 11%, comparing the fault arc perception results of the two classifiers under the condition of the same sample input, the S4VM still has higher accuracy, and the SVM perception accuracy at the moment is reduced by about 3 times of the S4VM perception accuracy. This shows that semi-supervised machine learning is more advantageous under less sample label training conditions.
Table 1.S4VM and SVM comparison results
Figure BDA0003052796620000121
The invention has the following characteristics:
1) According to the method, the safe semi-supervised support vector machine is used, the label-free data are fully utilized, so that the perception method has the advantage of higher accuracy compared with the traditional supervised machine learning under the condition of less sample label training, and the perception performance and efficiency of the fault arc are improved.
2) The S4VM applied by the method can keep higher accuracy under different parameter setting combination conditions, the sensing accuracy is insensitive to the set model parameters, and the method has better generalization performance, which means that the S4VM can obtain satisfactory sensing accuracy under various arc detection scenes without adjusting the model parameters, and the parameter optimization calculation of the classification model in the training process is greatly reduced.
3) The method has the advantages that the statistical law of a large number of electric arc situations is mastered by using semi-supervised machine learning, so that the method has a wider application range, can reliably and quickly act various fault electric arc working conditions, cannot generate misoperation in the normal running process of starting and stopping various loads, solves the problem of refusal action caused by the complexity of the fault electric arc working conditions, and effectively prevents the personal and property safety threats caused by fault electric arcs.
4) After the fault arc occurs, the type of the fault arc can be further identified, and the type of the external interference load can be judged, so that different arc quenching measures are taken for different types of fault arcs, and the fire risk is reduced to the maximum extent; the occurrence position of the fault arc in the electric branch can be preliminarily judged according to the given external interference load type, so that the quick positioning of the fault arc is facilitated, and the operation and maintenance efficiency of the system is improved.
5) The machine learning method for system state perception and fault arc type identification has two classification characteristics, and the machine learning method for judging the external interference load type of the branch where the fault arc is located has a multi-classification characteristic.

Claims (8)

1. A low-voltage fault arc sensing method based on semi-supervised machine learning is characterized by comprising the following steps: the fault arc sensing method comprises the following steps:
1) In a low-voltage alternating-current power system, current signals of different branches are sampled point by frequency f, or current signals of different branches and superior buses of all branches are sampled point by point, and the sampling is carried out according to the time window length T s Respectively extracting N current signal sampling points corresponding to the current analysis period to obtain a detection signal x n Wherein N =1, \8230, N;
2) For the detection signal x n Decomposing the wavelet packet to obtain a detection signal x n The wavelet packet tree of (a);
3) Selecting a node corresponding to the frequency band dimension capable of displaying the time-frequency characteristics of the fault arc from the wavelet packet tree, and performing wavelet coefficient reconstruction on the node to obtain y n N, and then carrying out absolute value summation to obtain a characteristic value I of an analysis period;
4) Inputting the characteristic value I into an S4VM classifier, if the S4VM classifier outputs a value corresponding to a fault arc event in the system, turning to the step 5), otherwise, judging that the system is in normal operation, and returning to the step 1);
5) If the S4VM classifier outputs values corresponding to the fault arc events in the system in continuous K analysis periods, judging that fault arcs occur in corresponding branches or buses in the system; otherwise, judging that the arc-like working condition occurs in the system, and returning to the step 1); the value of K is as follows:
in the case of a low current arc, i.e., I' is ≦ 63A, then K is calculated according to the following equation:
Figure FDA0004077393770000011
wherein, I' represents the current of the corresponding branch or bus;
in case of a high current arc, i.e. I' >63A, K =4, 5;
the fault arc sensing method further comprises the following steps: for the detection signal x n Performing short-time Fourier transform while performing wavelet packet transform to obtain detection signal x n The time-frequency domain component matrix of (a); selecting corresponding matrix elements of frequency band dimensions capable of displaying time-frequency characteristics of the fault arc in the matrix to accumulate to obtain a characteristic value II of an analysis period, and judging the type of the fault arc according to the Euclidean distance between the stored characteristic value I and the stored characteristic value II of each analysis period;
the method for judging the type of the fault arc specifically comprises the following steps: respectively normalizing the stored characteristic values I and II of each analysis period, and then solving the Euclidean distance between the normalized characteristic values I and II; comparing the obtained Euclidean distance with a given threshold, if the comparison result is smaller than the threshold, judging that the type of the fault arc generated in the current analysis period is a parallel fault arc, and if the comparison result is larger than or equal to the threshold, judging that the type of the fault arc generated in the current analysis period is a series fault arc; the value range of the threshold is 3.1-3.5.
2. The low-voltage fault arc sensing method based on semi-supervised machine learning according to claim 1, characterized in that: the frequency f is 24 kHz-400kHz, and the N is 240-4000.
3. The low-voltage fault arc sensing method based on semi-supervised machine learning according to claim 1, characterized in that: the wavelet base adopted by wavelet packet decomposition is Haar, and the characteristic frequency band of the fault electric arc is 9 kHz-12 kHz.
4. The low-voltage fault arc sensing method based on semi-supervised machine learning according to claim 1, characterized in that: the kernel function of the S4VM classifier is a radial basis kernel function, the sample time is 10-50, the parameter C1 is 50-70, and the parameter C2 is 0.001-0.08.
5. The low-voltage fault arc sensing method based on semi-supervised machine learning according to claim 1, characterized in that: the fault arc sensing method further comprises the following steps: according to detection signals before and after the fault arc occurs and corresponding wavelet characteristic forms, the generation countermeasure network is applied to establish a correlation model of the load and the arc, and the external interference load type of a branch where the fault arc is located is identified according to the correlation model.
6. The utility model provides an electric arc fault protection electrical apparatus based on semi-supervised formula machine learning which characterized in that: the arc fault protection electric appliance comprises a current signal sampling device and a low-voltage fault arc sensing module;
the current signal sampling device is used for sampling current signals of a branch or a bus where the arc fault protection electric appliance is located, and extracting corresponding N current signal sampling points as detection signals x according to a time window n ；
The low-voltage fault arc sensing module is used for sensing the low-voltage fault arc according to the detection signal x n Respectively combining wavelet packet transformation and short-time Fourier transformation to obtain a characteristic value I and a characteristic value II which reflect time-frequency characteristics of the fault arc, judging the occurrence of the fault arc on a branch or a bus by applying an S4VM classifier according to the characteristic value I, and judging the type of the fault arc according to the Euclidean distance between the stored characteristic value I and the characteristic value II of each analysis period;
for the detection signal x n Decomposing the wavelet packet to obtain a detection signal x n Selecting a node corresponding to a frequency range dimension capable of displaying time-frequency characteristics of the fault arc from the wavelet packet tree, and performing wavelet coefficient reconstruction on the node to obtain y n Wherein N =1, \ 8230, N, howeverThen, absolute value summation is carried out to obtain a characteristic value I of an analysis period;
if the S4VM classifier outputs values corresponding to the fault arc events in the system in the continuous K analysis periods, judging that fault arcs occur in corresponding branches or buses in the system; otherwise, judging that an arc-like working condition occurs in the system; the value of K is as follows:
in the case of a low current arc, i.e., I' is ≦ 63A, then K is calculated according to the following equation:
Figure FDA0004077393770000021
wherein, I' represents the current of the corresponding branch or bus;
in case of a high current arc, i.e. I' >63A, K =4, 5;
for the detection signal x n Performing short-time Fourier transform while performing wavelet packet transform to obtain detection signal x n The time-frequency domain component matrix of (a); selecting corresponding matrix elements of frequency band dimensions capable of displaying time-frequency characteristics of the fault arc in the matrix for accumulation to obtain a characteristic value II of an analysis period;
respectively normalizing the stored characteristic values I and II of each analysis period, and then solving the Euclidean distance between the normalized characteristic values I and II; comparing the obtained Euclidean distance with a given threshold, if the comparison result is smaller than the threshold, judging that the type of the fault arc generated in the current analysis period is a parallel fault arc, and if the comparison result is larger than or equal to the threshold, judging that the type of the fault arc generated in the current analysis period is a series fault arc; the value range of the threshold is 3.1-3.5.
7. The arc fault protection appliance based on semi-supervised machine learning of claim 6, wherein: the arc fault protection appliance further comprises an interference load type identification module for identifying the type of the interference load according to the established association model of the load and the arc.
8. The arc fault protection appliance based on semi-supervised machine learning of claim 7, wherein: the current signal sampling device, the low-voltage fault arc sensing module and the interference load type identification module are integrated on a circuit breaker on a branch or a bus.
CN202110491988.6A 2021-05-06 2021-05-06 Low-voltage fault arc sensing method based on semi-supervised machine learning Active CN113297786B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110491988.6A CN113297786B (en) 2021-05-06 2021-05-06 Low-voltage fault arc sensing method based on semi-supervised machine learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110491988.6A CN113297786B (en) 2021-05-06 2021-05-06 Low-voltage fault arc sensing method based on semi-supervised machine learning

Publications (2)

Publication Number Publication Date
CN113297786A CN113297786A (en) 2021-08-24
CN113297786B true CN113297786B (en) 2023-04-07

Family

ID=77321770

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110491988.6A Active CN113297786B (en) 2021-05-06 2021-05-06 Low-voltage fault arc sensing method based on semi-supervised machine learning

Country Status (1)

Country Link
CN (1) CN113297786B (en)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113866572B (en) * 2021-09-06 2024-05-28 西安理工大学 DC fault arc detection and positioning method under multi-power electronic device access condition
CN114142605B (en) * 2021-11-09 2022-07-15 广东工业大学 Pilot protection method, device and storage medium

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2004102760A1 (en) * 2003-05-19 2004-11-25 Ellenberger & Poensgen Gmbh Method and device for detecting disturbing arcs
CN107086855A (en) * 2017-04-25 2017-08-22 西安交通大学 The photovoltaic system fault arc detection method of many time-frequency characteristics is merged in a kind of machine learning
CN107154783A (en) * 2017-04-18 2017-09-12 西安交通大学 Using the method for photovoltaic system fault electric arc in the case of independent component analysis and S-transformation detecting system PROCESS COUPLING
CN109560770A (en) * 2018-10-29 2019-04-02 西安交通大学 Photovoltaic system fault arc detection method based on self-adaptive kernel function and instantaneous Frequency Estimation
CN112183628A (en) * 2020-09-28 2021-01-05 浙江零壹智能电器研究院有限公司 Alternating current arc fault detection method and system based on multiple linear time-frequency transformations

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108075728A (en) * 2016-11-15 2018-05-25 上海英孚特电子技术有限公司 A kind of photovoltaic system DC side arc fault type identification and protective device
CN112433128B (en) * 2020-10-13 2022-12-02 西安理工大学 Stochastic resonance-based direct-current fault arc diagnosis method and device

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2004102760A1 (en) * 2003-05-19 2004-11-25 Ellenberger & Poensgen Gmbh Method and device for detecting disturbing arcs
CN107154783A (en) * 2017-04-18 2017-09-12 西安交通大学 Using the method for photovoltaic system fault electric arc in the case of independent component analysis and S-transformation detecting system PROCESS COUPLING
CN107086855A (en) * 2017-04-25 2017-08-22 西安交通大学 The photovoltaic system fault arc detection method of many time-frequency characteristics is merged in a kind of machine learning
CN109560770A (en) * 2018-10-29 2019-04-02 西安交通大学 Photovoltaic system fault arc detection method based on self-adaptive kernel function and instantaneous Frequency Estimation
WO2020087752A1 (en) * 2018-10-29 2020-05-07 西安交通大学 Photovoltaic system arc fault detection method employing adaptive kernel function and instantaneous frequency estimation
CN112183628A (en) * 2020-09-28 2021-01-05 浙江零壹智能电器研究院有限公司 Alternating current arc fault detection method and system based on multiple linear time-frequency transformations

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
chen silei等.Wavelet-based protection strategy for series arc faults interfered by multicomponent noise signals in grid-connected photovoltaic systems.scienceDirect.2019,全文. *
丁鑫 ; 竺红卫 ; 殷浩楠 ; 王一闻 ; .基于机器学习方法的直流电弧故障检测.传感器与微系统.2017,(11),全文. *

Also Published As

Publication number Publication date
CN113297786A (en) 2021-08-24

Similar Documents

Publication Publication Date Title
US11489490B2 (en) Arc fault detection method for photovoltaic system based on adaptive kernel function and instantaneous frequency estimation
Gomes et al. High-sensitivity vegetation high-impedance fault detection based on signal's high-frequency contents
CN113049922B (en) Fault arc signal detection method adopting convolutional neural network
CN113297786B (en) Low-voltage fault arc sensing method based on semi-supervised machine learning
CA2554962C (en) High impedance fault detection
Xiong et al. Incipient fault identification in power distribution systems via human-level concept learning
WO2020015277A1 (en) Arc light fault identifying device and method based on panoramic information
CN108885237A (en) Method and apparatus for detecting the arc fault in electrical system
CN110398669A (en) Method for detecting arc
Wei et al. Fault detection based on the generalized S-transform with a variable factor for resonant grounding distribution networks
Ashok et al. Fault diagnosis scheme for cross-country faults in dual-circuit line with emphasis on high-impedance fault syndrome
CN114584069B (en) Photovoltaic direct-current arc detection method and system
Tripathy Power transformer differential protection based on neural network principal component analysis, harmonic restraint and Park's plots
Guillen et al. Hermite transform based algorithm for detection and classification of high impedance faults
Khond et al. A combined Data Mining and Machine Learning approach for High Impedance Fault Detection in Microgrids
CN110568300B (en) Power distribution network single-phase earth fault identification method based on multi-source information
CN113376474A (en) Neural network fault arc identification system and method based on generalized S transformation
Yao et al. Distribution high impedance fault detection using the fault signal reconstruction method
Adly et al. A novel wavelet packet transform based fault identification procedures in HV transmission line based on current signals
CN113298124B (en) Parameter matching-based multi-level direct current fault arc detection method
Meléndez et al. Analysis of sequences of events for the characterisation of faults in power systems
Sirojan et al. High Impedance Fault Detection by Convolutional Deep Neural Network
Heidari et al. Performance evaluation of HHT and WT for detection of HIF and CT saturationin smart grids
CN113917294B (en) Intelligent self-adaptive arc detection method based on wavelet decomposition and application device thereof
Varghese et al. Application of signal processing techniques and intelligent classifiers for high-impedance fault detection in ensuring the reliable operation of power distribution systems

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant