WO2020087752A1 - 基于自适应核函数和瞬时频率估计的光伏系统故障电弧检测方法 - Google Patents

基于自适应核函数和瞬时频率估计的光伏系统故障电弧检测方法 Download PDF

Info

Publication number
WO2020087752A1
WO2020087752A1 PCT/CN2018/125440 CN2018125440W WO2020087752A1 WO 2020087752 A1 WO2020087752 A1 WO 2020087752A1 CN 2018125440 W CN2018125440 W CN 2018125440W WO 2020087752 A1 WO2020087752 A1 WO 2020087752A1
Authority
WO
WIPO (PCT)
Prior art keywords
frequency
time
photovoltaic system
arc
fault arc
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.)
Ceased
Application number
PCT/CN2018/125440
Other languages
English (en)
French (fr)
Inventor
李兴文
陈思磊
孟羽
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 Jiaotong University
Original Assignee
Xian Jiaotong University
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 Jiaotong University filed Critical Xian Jiaotong University
Priority to US16/629,033 priority Critical patent/US11489490B2/en
Publication of WO2020087752A1 publication Critical patent/WO2020087752A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • H—ELECTRICITY
    • H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
    • H02S—GENERATION OF ELECTRIC POWER BY CONVERSION OF INFRARED RADIATION, VISIBLE LIGHT OR ULTRAVIOLET LIGHT, e.g. USING PHOTOVOLTAIC [PV] MODULES
    • H02S50/00—Monitoring or testing of PV systems, e.g. load balancing or fault identification
    • G—PHYSICS
    • G05—CONTROLLING; REGULATING
    • G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
    • G05B23/00—Testing or monitoring of control systems or parts thereof
    • G05B23/02—Electric testing or monitoring
    • G05B23/0205—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults
    • G05B23/0218—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults characterised by the fault detection method dealing with either existing or incipient faults
    • G05B23/0224—Process history based detection method, e.g. whereby history implies the availability of large amounts of data
    • G05B23/024—Quantitative history assessment, e.g. mathematical relationships between available data; Functions therefor; Principal component analysis [PCA]; Partial least square [PLS]; Statistical classifiers, e.g. Bayesian networks, linear regression or correlation analysis; Neural networks
    • G—PHYSICS
    • G05—CONTROLLING; REGULATING
    • G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
    • G05B23/00—Testing or monitoring of control systems or parts thereof
    • G05B23/02—Electric testing or monitoring
    • G05B23/0205—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults
    • G05B23/0259—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults characterized by the response to fault detection
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G06N20/10—Machine learning using kernel methods, e.g. support vector machines [SVM]
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N7/00—Computing arrangements based on specific mathematical models
    • G06N7/01—Probabilistic graphical models, e.g. probabilistic networks
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
    • Y02E10/00—Energy generation through renewable energy sources
    • Y02E10/50—Photovoltaic [PV] energy

Definitions

  • the invention belongs to the field of photovoltaic system electrical fault detection, and particularly relates to a method for finding a time when the spectrum energy increases based on a nonlinear frequency modulation wavelet transform, and inputting multiple time-frequency characteristics obtained based on an adaptive optimal kernel time-frequency distribution to Naive Bay
  • the yes model is a method to judge the operating state of the photovoltaic system, thereby ensuring that the fault arc can be accurately judged under the interference conditions of various arc conditions, so that the corresponding fault arc can be extinguished and the safe operation of the photovoltaic system is guaranteed.
  • the purpose of the present invention is to provide a fault arc detection method for photovoltaic systems based on adaptive kernel function and instantaneous frequency estimation, to solve the problem of accurate, reliable and rapid identification of fault arcs in the case of arc-like interference in the photovoltaic system.
  • the present invention adopts the following technical solutions:
  • the detection signal x t is equally divided into y time periods, and go to step 3) for further judgment; otherwise, it is judged that the photovoltaic system is in a normal working state at the sampling time, and the correspondence at the next sampling time is continued Comparison of feature changes until When the window between the detection signals are judged to be in normal operation for the photovoltaic system returns to step 1) starts detecting the signal acquisition and determination of the next time window;
  • the adaptive optimal kernel time-frequency distribution is started from the next period of the time period when the spectrum energy increases, and the corresponding complex time-frequency matrix in the time-frequency domain is obtained, and r effective frequency bands in the frequency dimension are selected.
  • the real part of the element is subjected to the square sum operation of the time dimension to obtain the corresponding column vector, and then the frequency dimension integration operation is performed on each selected frequency band to construct r feature quantities as input vectors, and go to step 4);
  • step 5 Use the trained Naive Bayes model and the corresponding r feature quantity values to classify the input vector. If the output value is 0, it is determined that the fault arc is present in this period, and go to step 5); If the output value is 1, it is determined to be arc-like interference, the count variable is cleared, and return to step 1) to start the detection signal collection and judgment of the next time window;
  • the current signal is a signal sampled by a Hall sensor and filtered by a high-pass filter or a signal directly sampled by a current transformer.
  • the selection principle of the time window length T NCT is that the detection signal within the time window of a determined length can reflect the effective time-frequency characteristics of the fault arc, and the value range of T NCT is 4-40 ms; the number of divided time periods y The value ranges from 2 to 10.
  • the parameters of the multi-time-frequency transform are based on the significant time-frequency characteristics of the maximum separation of the fault arc and the arc-like operating conditions; wherein, the nonlinear frequency modulation wavelet transform
  • the value range of the selected polynomial order is 10 to 30; the iteration termination condition is that the iteration stops when it reaches a predetermined number of times, and the value range of the number of iterations is 3 to 8; the output frequency of the adaptive optimal kernel time-frequency distribution is divided into scales The value ranges from 256 to 8192.
  • the selection principle of the effective iteration number (ie u) and frequency components of the nonlinear frequency modulation wavelet transform is to accurately indicate the time window where the fault arc occurs at the time of the pulse, and display the fault arc in a larger amplitude form
  • the effective iteration number u ranges from 2 to 5
  • the frequency component is specified according to the fault arc characteristic frequency band, and the selected fault arc characteristic frequency band at different iteration times
  • the number of fault arcs is the same or different, and the characteristic frequency bands of fault arcs are all selected in the range of 5-50kHz.
  • the frequency bands of fault arcs selected by different iterations partially overlap or are continuously divided or exist intervals; constructed according to the number of effective iterations and frequency band components
  • the total number of feature values n ranges from 5 to 10.
  • a threshold value judgment processing method is adopted for the n feature quantities NC i (t) based on the nonlinear frequency modulation wavelet transform.
  • the threshold setting principle in the threshold column vector Y is that the signal can be acquired at each feature quantity level.
  • different characteristic quantities NC i (t) correspond to set thresholds that are the same or different, and the threshold number m used to determine whether the spectrum energy increases at the current sampling time has a value range of n-2 to n.
  • the principle of selecting r frequency bands that are effective in the frequency dimension is that the time window where the fault arc occurs can be accurately indicated in the form of a pulse, with a larger amplitude
  • the form shows the difference between the fault arc state and the arc-like state and the difference is consistent; r frequency bands are selected according to the fault arc characteristic frequency band, and the fault arc characteristic frequency band is selected in the range of (0,50) kHz, and the selected
  • the characteristic bands of the fault arc partially overlap or are continuously divided or there is a gap; the value range r of the corresponding corresponding constructed feature quantity is 3-7.
  • the output state of the Naive Bayes model during the learning and training process is marked as follows: the arc voltage in the corresponding period is zero, then it is marked as 1, and the arc voltage in the corresponding period is non-zero, then it is marked as 0;
  • the training sample size ranges from 1500 to 3000.
  • the p is a fault arc triggering threshold
  • the selection principle is based on the classification of arc operating conditions for quickly removing the fault arc and not mis-acting in the area, and the value range is 18-120.
  • the nonlinear frequency modulation wavelet transform in the present invention makes the calculation result more accurate, can describe the change law of complex frequency in the signal more accurately, has good time-frequency energy concentration of the fault arc, and can obtain more than linear time-frequency transform Good fault arc time-frequency characteristic aggregation improves the ability of fault arc characteristics to resist noise interference;
  • the adaptive time-frequency analysis in the present invention intercepts the signal auto-correlation function, and details the details of the signal, so that there is no mutual restriction between the signal time-frequency resolution, and a better time-frequency localization is obtained As a result, the accuracy requirements of the time-frequency distribution required by the algorithm analysis are met, which is helpful to find the unique characteristics of the fault arc that is different from the arc-like.
  • the problem of cross-term interference caused by the rate increase;
  • the present invention extracts multiple feature quantities for the detection signal, and filters out the normal state and most arc-like states through the feature set of nonlinear frequency modulation wavelet transform. Only when the fault arc state and a few arc-like states need to be adapted. Frequency analysis and the calculation of Naive Bayes model, nonlinear FM wavelet transform and adaptive time-frequency analysis need to store very few intermediate matrices, and the corresponding feature quantities can be replaced by real-time calculation modes, which greatly reduces the consumption of detection algorithms. Memory, reducing the dependence of fault arc detection on hardware;
  • the nonlinear frequency modulation wavelet transform and its characteristic quantities are calculated, each time a wider time window is used to calculate the normal state current signal, the precise time scale of the transformation itself is beneficial to locate the moment when the fault arc occurs. After the suspected fault arc occurs, the state current signal is calculated using a narrow and wide analysis period.
  • the interleaved calculation timing of nonlinear frequency modulation wavelet transform and adaptive time-frequency analysis effectively improves the calculation speed of the detection algorithm and meets the fault arc Requirements for rapid testing;
  • the present invention proposes a feature quantity group composed of multiple feature quantities at each level of nonlinear frequency modulation wavelet transform and adaptive time-frequency analysis.
  • a certain feature quantity may fail, other features within the feature quantity group
  • the feature quantity can still ensure the effective judgment of the system state.
  • the cooperation between the two levels can greatly improve the accuracy of the classification of fault arc or arc-like working conditions, and meet the reliability requirements of fault arc detection.
  • the present invention makes full use of the advantages of nonlinear frequency modulation wavelet transform and adaptive time-frequency analysis to refine their respective optimal time-frequency characteristics, and the constructed detection algorithm can be reliable, accurate, and fast while preventing system-like arc interference Earth fault arcs ensure that the fault arcs are accurately distinguished from arc-like operating conditions caused by factors such as load switching, system startup, etc., while ensuring safe power consumption and fully guaranteeing the power supply stability of the photovoltaic system.
  • FIG. 1 is a schematic diagram of a hardware implementation principle of a photovoltaic system fault arc detection algorithm in an embodiment of the present invention.
  • FIG. 2 is a flowchart of a photovoltaic system fault arc detection algorithm in an embodiment of the present invention.
  • Figure 3a is the current signal of the fault arc.
  • Figure 3b, Figure 3c, Figure 3d, Figure 3e, Figure 3f, and Figure 3g are the characteristic waveforms of the photovoltaic system fault arc detection using nonlinear frequency modulation wavelet transform, and the frequency band used for each characteristic is selected as NC 1 : second 8.3kHz ⁇ 12.4kHz in the time-frequency graph obtained in the second iteration; NC 2 : 12.5kHz ⁇ 16.6kHz in the time-frequency graph obtained in the second iteration; NC 3 : 29.2kHz ⁇ 33.3 in the time-frequency graph obtained in the second iteration kHz; NC 4 : 8.3kHz to 12.4kHz in the time-frequency graph obtained by the third iteration; NC 5 : 25kHz to 29.1kHz in the time-frequency graph obtained by the third iteration; NC 6 : Time-frequency graph obtained by the third iteration Within 29.2kHz ⁇ 33.3kHz.
  • Figure 3h, Figure 3i, Figure 3j, Figure 3k, and Figure 3l are the characteristic waveforms of photovoltaic system fault arc detection using adaptive optimal nuclear time-frequency distribution.
  • the frequency band used for each characteristic is selected as AOK 1 : 24.4kHz ⁇ 30.4kHz; AOK 2: 18.3kHz ⁇ 24.3kHz; AOK 3: 12.2kHz ⁇ 18.2kHz; AOK 4: 6.1kHz ⁇ 12.1kHz; AOK 5: 0.1 ⁇ 6.0kHz.
  • Fig. 3m is a system state judgment output signal (for Fig. 3a) for applying the present invention to detect a fault arc of a photovoltaic system.
  • Figure 4a is the arc-like current signal for load switching.
  • Figures 4b, 4c, 4d, 4e, 4f, and 4g are the characteristic waveforms of photovoltaic system fault arc detection using nonlinear frequency modulation wavelet transform, and the frequency bands used for each characteristic are selected as NC 1 and NC. 2 , NC 3 , NC 4 , NC 5 and NC 6 .
  • Fig. 4h is a system state judgment output signal (for Fig. 4a) for applying the present invention to detect a fault arc of a photovoltaic system.
  • Figure 5a is the arc-like current signal for system startup.
  • Figures 5b, 5c, 5d, 5e, 5f, and 5g are the characteristic waveforms of the photovoltaic system fault arc detection using nonlinear frequency modulation wavelet transform, and the frequency bands used for each characteristic quantity are selected as NC 1 and NC. 2 , NC 3 , NC 4 , NC 5 and NC 6 .
  • Figures 5h, 5i, 5j, 5k, and 5l are the characteristic waveforms of photovoltaic system fault arc detection using adaptive optimal nuclear time-frequency distribution, and the frequency bands used for each characteristic are selected as AOK 1 and AOK 2 , AOK 3 , AOK 4 and AOK 5 .
  • FIG. 5m is a system state judgment output signal (for FIG. 5a) for applying the present invention to detect a fault arc of a photovoltaic system.
  • the current signal of the photovoltaic system under different arc conditions and fault arc conditions is sampled according to the time window. If the current signal is collected by the Hall sensor, it is followed by high-pass filtering (the purpose of filtering is to remove the DC component). Feature layer processing; if the current signal is collected by a current transformer, subsequent feature layer processing is directly performed.
  • the Hall sensor or current transformer is installed in the photovoltaic string to be monitored or the photovoltaic array bus DC bus, and an inverter can also be used to collect the system current. As shown in Figure 1, the collected current signal is input to the DC fault arc detection device, and is input to the DSP processing module through the high-pass filter and A / D conversion module.
  • the current signal undergoes wavelet transform based on nonlinear frequency modulation To obtain the corresponding two-dimensional matrix M, which is used to judge the moment when the spectrum energy increases.
  • the current signal under the current time window length is divided into a finer scale to form multiple time periods, and the adaptive optimal kernel time frequency is made from the next time period of energy increase Distribution, get the corresponding multiple eigenvalues as the input vector of the Naive Bayes model.
  • the current sampling signal in the input time window can be extracted into multiple time-frequency features based on the adaptive optimal kernel time-frequency distribution, and then input to the trained Naive Bayes model for state judgment.
  • the Naive Bayes model can output the 0/1 judgment result of whether a fault arc occurs in the photovoltaic system in real time, output 0 when a fault arc occurs, and output 1 when the system is in normal operation.
  • the fault arc removal signal is triggered, that is, the control signal output by the DSP processing module is driven by the D / A conversion module to drive the relay to control the corresponding support of the photovoltaic system.
  • the circuit breaker at the road is cut off; before the Naive Bayes model has reached the number of cycles of the predetermined output 0, as long as it has a 1 output, it is considered that this time is caused by the interference of arc-like operating conditions rather than a real fault arc.
  • Working condition Only when the Naive Bayesian model continuously outputs 0 to reach the specified number of times, the fault arc removal signal is triggered, that is, the control signal output by the DSP processing module is driven by the D / A conversion module to drive the relay to control the corresponding support of the photovoltaic system.
  • the signal x t is acquired with a time window of T NCT length, and the iterative time-frequency map of x t is obtained through nonlinear frequency modulation wavelet transform, and the feature quantity is constructed based on the frequency components in the selected iterative time-frequency map to determine the increase in spectrum energy.
  • Time after finding that there is a sudden change in energy increase (energy increase time), the adaptive optimal kernel time-frequency distribution is obtained to obtain the corresponding matrix distribution form of x t in the time-frequency domain.
  • the present invention can accurately identify the fault arc in the photovoltaic system through multiple effective time-frequency characteristics, and can also ensure that it does not malfunction under various arc working conditions, thereby improving the ability of the photovoltaic system to operate safely and stably.
  • the core of the photovoltaic system fault arc detection algorithm is a photovoltaic system fault arc detection method based on adaptive kernel function and instantaneous frequency estimation.
  • the specific steps of the algorithm are as follows (Figure 2):
  • Step 1 The parameter initialization process includes time window length T NCT , threshold column vector Y, m (spectral energy increase trigger threshold), fault arc trigger threshold p, and nonlinear frequency modulation wavelet transform and adaptive optimal kernel time-frequency distribution.
  • T NCT time window length
  • m spectral energy increase trigger threshold
  • fault arc trigger threshold p fault arc trigger threshold
  • nonlinear frequency modulation wavelet transform and adaptive optimal kernel time-frequency distribution various parameters in a time-frequency analysis tool (polynomial order selected for nonlinear FM wavelet transform, number of iterations, output frequency division scale of adaptive optimal kernel time-frequency distribution), etc.
  • the current signal is collected with the set time window length T NCT .
  • the current signal here is a signal detected by a Hall sensor and filtered by a high-pass filter or a signal without a DC component measured by a current transformer or the like.
  • the time window is too short, it will result in the frequency division accuracy of the two-dimensional complex time-frequency matrix obtained by the nonlinear FM wavelet transform is not enough to respond
  • the fault arc is different from the basic characteristics of arc-like; if the time window is too long, it will increase the requirements of the algorithm on the detection device hardware in memory and main frequency. Therefore, the value of the time window length T NCT ranges from 4 to 40 ms.
  • Step 2 Analyze the current detection signal x t collected under the t-th time window by using nonlinear frequency modulation wavelet transform to obtain the complex matrix time-frequency distribution form after multiple iterations of the output current signal of the photovoltaic system under the time window.
  • the effective number of iterations u (the effective number of iterations is equal to The selected number of the iterated complex time-frequency matrix is less than or equal to the number of iterations at the end of the iteration).
  • the value range is 2 to 5.
  • the number of the specified frequency components selected on the selected effective iteration number may be different, but they are all selected within the range of the fault arc characteristic frequency band 5-50kHz, which is constructed by the frequency band components under each effective iteration number (u)
  • the value range of the total number of features n is 5 to 10.
  • the n feature quantities form a two-dimensional matrix M within the time window, and compare the corresponding elements along the time dimension with the threshold column vector Y. If the two-dimensional matrix M (M is a two-dimensional matrix of n ⁇ T NCT ) corresponds to the feature If the value of the quantity (because the characteristic value itself is large, the change of the value of the corresponding characteristic quantity compared with the previous moment is actually selected) is greater than the corresponding threshold more than m, then the current signal of the time window is judged to have a spectrum If the energy increases, go to step three for further judgment, otherwise it is judged that the sampling time is in the normal working state, and the corresponding characteristic value comparison at the next sampling time is continued until the detection signals of the time window are all the normal working state of the photovoltaic system, return Step one continues sampling. Among them, the value range of m is n-2 ⁇ n.
  • Step 3 Divide the current detection signal x t into y segments (the value range of the divided period y is 2 to 10), and adopt the method of adaptive optimal kernel time-frequency distribution from the next period of energy increase.
  • the corresponding period of x t is analyzed segment by segment, and the complex matrix time-frequency distribution form corresponding to the output current signal of the photovoltaic system in the time-frequency domain under this period is obtained.
  • the real part of the corresponding element that is, the effective frequency band
  • the integration operation of the frequency dimension in the effective frequency band obtains the value of multiple feature quantities as the input vector, and then proceeds to step 4 for the recognition of the Naive Bayes model.
  • the value range of the output frequency division scale of the adaptive optimal kernel time-frequency distribution is 256 to 8192.
  • the frequency band of the fault arc is considered to be selected in the range of (0,50) kHz, and the effective frequency band is determined according to the selected result.
  • the total value of the constructed characteristic quantity r ranges from 3 to 7.
  • Step four Use the trained Naive Bayes model to classify the input vector. If the fault arc state is displayed during this period, the output value is 0, go to step five; otherwise, the output value is 1, it is determined to be arc-like interference , Clear the count variable and return to step 1 to analyze the detected signal in the next time window.
  • the mark of the Naive Bayes model in the learning and training process the arc voltage signal in the corresponding period is zero, then the mark is 1, and the arc voltage in the corresponding period is non-zero, then the mark is 0, the value of the training sample capacity
  • the range is from 1500 to 3000.
  • Step 5 Count the number of time slots whose output value calculated by the Naive Bayes model is 0. When the number of time slots reaches p, it is judged that a fault arc has occurred in the photovoltaic system, and a control signal for circuit breaking is issued. Extinguish the fault arc, otherwise, return to step 3 to analyze the detection signal in the next period.
  • the fault arc trigger threshold p has a value range of 18 to 120.
  • the feature values obtained based on the nonlinear frequency modulation wavelet transform may have threshold misjudgment within certain time windows and lose the normal judgment ability of arc-like operating conditions, and based on the adaptive optimal kernel time-frequency
  • the eigenvalues obtained by the distribution are not misjudged in the arc-like operating conditions within these time windows, and present eigenvalues different from the fault state.
  • the Naive Bayes model can still use the learned arc fault statistical rules to more accurately determine the arc-like operating conditions as normal, which reflects that the detection algorithm of the present invention focuses on multiple feature quantities to improve the reliability of arc-like operating conditions. Identify the advantages of arc fault capability.
  • the present invention has a strong fault arc identification capability, which not only avoids the accidental factors causing the erroneous operation of the DC fault arc detection device, but also ensures the rapidity of the fault arc branch cutoff signal.
  • the photovoltaic system outputs a current detection signal.
  • the photovoltaic system was in a normal working state, the fault arc occurred in the 0.6648s system, and a large-value pulse appeared in the current signal. It has been in the fault arc state since then.
  • the current signal is analyzed by nonlinear frequency modulation wavelet transform, the elements of the specified frequency components in the two-dimensional complex matrix in the obtained time-frequency domain are modulo, and the integral processing method (that is, the sum of squares) in the frequency dimension is used to build Six feature quantities of nonlinear frequency modulation wavelet transform (as shown in Figure 3b ⁇ Figure 3g).
  • Each feature has a large-value pulse indication at the time of the fault arc, and the change of the feature value is greater than the corresponding set threshold (the corresponding threshold column vector Y is set to [30, 35, 75, 60, 40, 30]), all It can be determined that there is an increase in spectrum energy.
  • the current signal is analyzed through the adaptive optimal kernel time-frequency distribution.
  • the real part of the corresponding element in the two-dimensional complex matrix in the time-frequency domain is processed by the square sum operation along the time dimension, and the integration is processed in the frequency dimension. Way, get 5 feature quantities based on adaptive optimal kernel time-frequency distribution.
  • the adaptive optimal kernel time-frequency distribution calculation results of the current signal in all periods are given, as shown in Figure 3h to Figure 3l.
  • each characteristic value has a short, extremely large pulse indication at the time of the fault arc, and the characteristic value after the fault arc occurs is larger than the normal working state as a whole. This consistent large value state is conducive to the accuracy of the fault arc Identify.
  • the five time-frequency characteristic values calculated above are input to the trained Naive Bayes model to determine whether there is a fault arc in the photovoltaic system.
  • the Naive Bayes model outputs 1, it is judged that the photovoltaic system is in an arc-like state during this period.
  • the aforementioned nonlinear frequency modulation wavelet transform has a misjudgment, and the state detection of the output current signal of the photovoltaic system in the next period continues; when Naive When the Bayesian model outputs 0, it is judged that a fault arc may have occurred in the photovoltaic system during this period, and the occurrence of the fault arc must be further confirmed by the set fault arc removal standard (P), that is, the period of continuous output 0 reaches 100 After that, it is determined that a fault arc has occurred in the photovoltaic system, and a signal to cut off the fault arc branch is issued to the corresponding circuit breaker.
  • P set fault arc removal standard
  • the detection algorithm can give a correct output in the face of normal operating current (judging that it does not meet the requirements of the fault arc removal standard), and can give a correct output for the fault state current signal (judging that it reaches the fault (Arc cutting standard requirements).
  • the photovoltaic system outputs a current detection signal.
  • the photovoltaic system was in a normal working state, and a load switching process occurred in 1.093s, forming a similar current mutation process in the time domain.
  • the current signal is analyzed by nonlinear frequency modulation wavelet transform, the elements of the two-dimensional complex matrix in the obtained time-frequency domain are modulo, and the integral processing method (that is, the sum of squares) in the frequency dimension is used to build a nonlinear frequency modulation
  • the six feature quantities of wavelet transform are shown in Figs. 4b to 4g. Among them, the amplitude of each characteristic value is significantly reduced when the system switches the load, and the change of the characteristic value is less than the corresponding given threshold in Y, and it can be determined that there is no increase in spectrum energy.
  • the threshold ( m) It is determined that the correct output can be given, it is determined that the photovoltaic system is in a normal working state within the time window, and the state detection of the output current signal of the photovoltaic system within the next time window is continued.
  • the photovoltaic system output current detection signal.
  • the photovoltaic system was in a normal shutdown state.
  • the system started the startup process, forming a smaller time-domain current amplitude level similar to that in Figure 4a, 2.586s
  • the system ends the startup process, forms a larger time-domain current amplitude level, and enters a normal working state.
  • the current signal is analyzed by nonlinear frequency modulation wavelet transform, the elements of the two-dimensional complex matrix in the obtained time-frequency domain are modulo, and the integral processing method (that is, the sum of squares) in the frequency dimension is used to build a nonlinear frequency modulation Six feature quantities of wavelet transform.
  • the arc-like operating condition at 2.256s is similar to the arc-like operating condition shown in Fig. 4a, so the corresponding characteristic amplitude amplitude based on the nonlinear frequency modulation wavelet transform will also show a significant reduction process as shown in Figs. 4b to 4g.
  • the value change is less than the corresponding threshold, and the threshold is determined to be in a normal working state, although the feature quantity constructed based on the adaptive optimal kernel time-frequency distribution analysis shown in Figures 5h to 5l appears the same fault arc mode as shown in Figures 3h to 3l, However, based on the nonlinear frequency modulation wavelet transform, the correct judgment can be given in advance, and the subsequent adaptive optimal kernel time-frequency distribution analysis is not performed.
  • the non-linear FM wavelet transform level here only gives the corresponding current analysis results for subsequent adaptive optimal kernel time-frequency distribution and naive Bayesian model analysis, as shown in Figures 5b to 5g.
  • each feature based on the nonlinear frequency modulation wavelet transform has a continuous amplitude increase process as shown in Figures 3b to 3g, it is determined that there is an increase in spectrum energy, and the subsequent adaptive optimal kernel time-frequency Distribution analysis steps.
  • the current signal is analyzed through the adaptive optimal kernel time-frequency distribution.
  • the real part of the corresponding element in the two-dimensional complex matrix in the time-frequency domain is processed by the square sum operation along the time dimension, and the integration is processed in the frequency dimension.
  • the Naive Bayes model fails to reach the trigger standard P due to the continuous 1 output. It is determined that arc-like conditions have occurred in the photovoltaic system at this time. The circuit breaker control signal is never issued, and the state detection of the output current signal of the photovoltaic system within the next time window is continued. As shown in the result shown in Figure 5m, the detection algorithm can give a correct output (it is judged that it does not meet the requirements of the fault arc removal standard) without causing misjudgment.
  • the photovoltaic fault arc detection method provided by the present invention is divided into two steps of nonlinear frequency modulation wavelet transform-threshold judgment and adaptive optimal kernel time-frequency distribution-naive Bayes model classification.
  • the previous step has high calculation resolution and is suitable for faults.
  • the occurrence time of the arc can be accurately captured. Once the current analysis time window has an energy increase trend, it can quickly enter the next step.
  • the interleaved sequence of the last step and the previous step makes the whole algorithm consume less memory and high calculation efficiency. Multiple time-frequency feature quantities are constructed in both steps.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Software Systems (AREA)
  • Mathematical Physics (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Computing Systems (AREA)
  • Medical Informatics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Automation & Control Theory (AREA)
  • Computational Mathematics (AREA)
  • Pure & Applied Mathematics (AREA)
  • Mathematical Optimization (AREA)
  • Mathematical Analysis (AREA)
  • Algebra (AREA)
  • Probability & Statistics with Applications (AREA)
  • Photovoltaic Devices (AREA)

Abstract

一种基于自适应核函数和瞬时频率估计的光伏系统故障电弧检测方法：以T NCT长度的时间窗采集信号x t，经过非线性调频小波变换得到x t的迭代时频图，基于所选迭代时频图内的频率分量构造特征量，用以判断频谱能量增大的时刻；在发现存在能量增大的突变点后，经过自适应最优核时频分布得到x t在时频域内对应的矩阵分布形式，对该矩阵按时间维度求平方和得列向量，选定多个频段，对每个频段进行频率维度的积分运算，得到多个特征值输入至训练好的朴素贝叶斯模型，判断当前时段内的光伏系统状态。通过多个有效的时频特征准确辨识光伏系统内故障电弧的同时还能确保多种类弧工况下不误动，使光伏系统安全稳定运行。

Description

基于自适应核函数和瞬时频率估计的光伏系统故障电弧检测方法 技术领域
本发明属于光伏系统电气故障检测领域，具体涉及一种基于非线性调频小波变换寻找频谱能量增大的时刻，并将基于自适应最优核时频分布得到的多个时频特征输入至朴素贝叶斯模型，以判断光伏系统运行状态的方法，由此确保多种类弧工况干扰情况下也能准确判断故障电弧，从而可以熄灭相应故障电弧，保障光伏系统的安全运行。
背景技术
目前电力电子半导体技术的发展令直流变压可通过直流变换器实现，占比越来越高的广义直流负荷通过直流变换器可连接至非水可再生发电系统。这种直流微型电网架构可通过减少变换环节降低供电系统运营成本、提升系统供电效率。对于典型非水可再生发电系统之一的太阳能光伏发电系统而言，一旦系统内出现线路绝缘老化、线路破损或者连接松动等现象，所引发的故障电弧现象会引起火灾事故，威胁着系统供电和人身用电安全。
目前，光伏系统故障电弧的特征捕获往往利用其电特性。然而，光伏系统在日出时必然会经历启动过程，在运行时也必然会经历直流负载的正常投切过程，这些正常的系统暂态过渡过程也会对采集的系统电量信号产生影响，形成与故障电弧相同的时域形态，这些工况统称为类弧。类弧工况会形成一些与故障电弧相似的特征表达，干扰着故障电弧的正确判断，会引发故障电弧检测装置的误动，对光伏系统造成频繁的不必要停机状态，严重影响着系统供电稳定性，大幅降低了光伏系统的运行效率。
现阶段应用于光伏系统的故障电弧保护产品尚未成熟，对其故障电弧特性的研究也处于起步阶段，研究如何有效区分故障电弧与类弧对故障电弧特性的理论研究有着至关重要的价值。因此，研究光伏系统故障电弧的独特特征，基于这些独有的故障电弧特性构建完备的故障电弧检测算法，使之在类弧扰动的情形下还能准确、可靠、快速地辨识故障电弧，对光伏系统可靠供电和维护用电安全有着极其重要的意义。
发明内容
本发明的目的在于提供一种基于自适应核函数和瞬时频率估计的光伏系统故障电弧检测方法，解决光伏系统内存在类弧干扰的情况下准确、可靠、快速辨识故障电弧的问题。
为达到上述目的，本发明采用了以下技术方案：
1)对光伏系统内的电流信号，以T NCT为时间窗长度进行逐点采样，得到检测信号x t，转至步骤2)；
2)对检测信号x t作非线性调频小波变换，得到检测信号x t多次迭代的复数时频矩阵，通过选定u个迭代复数时频矩阵，以及对选定的迭代复数时频矩阵中指定频率分量处元素的模进行频率维度(频率轴方向)的平方和运算，构造n个特征量NC i(t)，i＝1,2,3…,n(n表示所构造的特征量总数)，并形成n×T NCT的二维矩阵M，将M沿时间维度(时间轴方向)求取后一时刻(后一列)相对前一时刻(前一列)的特征量变化，并与阈值列向量Y进行相应元素比较，若该时间窗内某采样时刻(不含时间窗内第一个时刻)对应的特征量变化的值大于相应阈值的个数超过m个，则判断该时间窗的电流信号存在频谱能量增大，将检测信号x t等分为y个时段，转至步骤3)作进一步的判断；否则，判断该采样时刻光伏系统处于正常工作状态，继续进行下一采样时刻的对应特征量变化的比较，直至根据该时间窗的检测信号均判断为光伏系统处于正常工作状态时，返回步骤1)开始进行下一时间窗的检测信号采集与判断；
3)从频谱能量增大时刻所在时段的下一时段开始作自适应最优核时频分布，得到在时频域内对应的复数时频矩阵，选定频率维度上r个有效的频段，对相应元素的实部进行时间维度的平方和运算得相应列向量，然后对选定的每个频段进行频率维度的积分运算，构造得到r个特征量作为输入向量，转至步骤4)；
4)利用训练好的朴素贝叶斯模型及对应的r个特征量的值对该输入向量进行状态分类，若输出值为0，则判定该时段呈现故障电弧状态，转至步骤5)；若输出值为1，则判定是类弧干扰，清零计数变量，返回步骤1)开始进行下一时间窗的检测信号采集与判断；
5)对经所述朴素贝叶斯模型进行状态分类计算所得的输出值为0的时段个数利用计数变量进行计数，当输出值为0的时段个数达到p个时，则判断光伏系统发生故障电弧，发出电路开断的控制信号以熄灭故障电弧，否则继续按照步骤3)进行下一时段的检测信号分析。
优选的，所述电流信号为由霍尔传感器采样并经高通滤波后的信号或者为直接由电流互感器采样的信号。
优选的，所述时间窗长度T NCT的选取原则为确定长度的时间窗内检测信号能反映出有效的故障电弧时频特征，T NCT的取值范围为4～40ms；所划分时段数y的取值范围为2～10。
优选的，所述步骤2)及步骤3)中，多时频变换的各项参数基于最大程度地分离故障电弧区别于类弧工况的显著时频特征而定；其中，非线性调频小波变换所选用的多项式 阶次的取值范围为10～30；迭代终止条件为迭代达到既定次数即停止，迭代次数的取值范围为3～8；自适应最优核时频分布的输出频率划分尺度的取值范围为256～8192。
优选的，所述非线性调频小波变换的有效迭代次数(即u)和频率分量的选取原则为能以脉冲形式准确指示故障电弧发生时刻所在的时间窗，以较大的幅值形式显示故障电弧状态与类弧状态的差异且该差异具有一致性；有效迭代次数u的取值范围为2～5；所述频率分量根据故障电弧特征频段进行指定，不同迭代次数上选定的故障电弧特征频段个数相同或不同，故障电弧特征频段均在5～50kHz的范围内选定，不同迭代次数所选定的故障电弧特征频段部分重叠或连续分割或存在间隔；根据有效迭代次数及频段分量所构造的特征量总数n的取值范围为5～10。
优选的，对基于非线性调频小波变换构造的n个特征量NC i(t)采用阈值判断处理方式，所述阈值列向量Y中的阈值设定原则为能够获取信号在各特征量层面上的幅值变化模式，不同特征量NC i(t)对应设定的阈值相同或不同，用于判断当前采样时刻下频谱能量是否增大的阈值个数m的取值范围为n-2～n。
优选的，所述自适应最优核时频分布完成后，对频率维度上有效的r个频段进行选取的原则为能以脉冲形式准确指示故障电弧发生所在的时间窗，以较大的幅值形式显示故障电弧状态与类弧状态的差异且该差异具有一致性；r个频段根据故障电弧特征频段选定，故障电弧特征频段在(0,50]kHz的范围进行选定，所选定的故障电弧特征频段部分重叠或连续分割或存在间隔；所构造的对应特征量总数r的取值范围为3～7。
优选的，所述朴素贝叶斯模型在学习训练过程中的输出状态标记方式为：对应时段内的电弧电压为零，则标记为1，对应时段内的电弧电压非零，则标记为0；训练样本容量的取值范围为1500～3000。
优选的，所述p为故障电弧触发阈值，选取原则为以快速切除故障电弧、不误动地区分类弧工况而定，取值范围为18～120。
本发明具有如下有益的技术效果：
1)本发明中的非线性调频小波变换使计算结果更精确，可更为准确地描述信号中复杂频率的变化规律，对故障电弧的时频能量集中度好，能较线性时频变换获得更好的故障电弧时频特征聚集性，提升了故障电弧特征的抗噪声干扰能力；
2)本发明中的自适应时频分析对信号自相关函数进行截取，细致刻画了信号的细节部分，使得信号时频分辨率之间不存在相互制约关系，获得了较好的时频局部化结果，达到了算法分析所需的时频分布精度要求，有利于发现故障电弧差别于类弧的独特特征，该 分析方法折中控制了交叉项的抑制与自项的衰减，解决了时频分辨率提升时引发的交叉项干扰问题；
3)本发明对检测信号提取了多重特征量，通过非线性调频小波变换的特征量组筛除正常状态和大部分类弧状态，只有故障电弧状态和少部分类弧状态才需要进行自适应时频分析和朴素贝叶斯模型的计算，非线性调频小波变换和自适应时频分析需要存储的中间矩阵极少，相应的特征量均可采用实时计算替换模式，大幅减少了检测算法所需消耗的内存，降低了故障电弧检测对硬件的依赖；
4)本发明在非线性调频小波变换及其特征量计算时，每次采用较宽的时间窗对正常状态电流信号进行运算，变换本身的精细时间尺度有利于定位故障电弧发生时刻，在检测到疑似故障电弧发生时刻后，采用较窄宽的分析时段对该状态电流信号进行运算，非线性调频小波变换和自适应时频分析的交错计算时序切实提升了检测算法的计算速度，满足了故障电弧检测的快速性要求；
5)本发明在非线性调频小波变换和自适应时频分析的每个层级均提出多个特征量构成的特征量组，在可能造成某个特征量失效的情况下，特征量组内的其他特征量仍能确保系统状态的有效判断，两个层级之间的相互配合可大幅提升故障电弧或类弧工况分类的准确性，满足了故障电弧检测的可靠性要求。
进一步的，本发明充分利用非线性调频小波变换和自适应时频分析的优势，提炼各自的最优时频特征，所构建的检测算法能在防范系统类弧干扰的情况下可靠、准确、快速地动作故障电弧，确保准确区分故障电弧和由负载投切、系统启动等因素引发的类弧工况，在保障安全用电的同时充分保障了光伏系统的供电稳定性。
附图说明
图1为本发明实施例中的光伏系统故障电弧检测算法的硬件实现原理框架图。
图2为本发明实施例中的光伏系统故障电弧检测算法流程图。
图3a为故障电弧的电流信号。
图3b、图3c、图3d、图3e、图3f、图3g分别为应用非线性调频小波变换进行光伏系统故障电弧检测的特征量波形，各特征量所采用的频段选为NC 1：第二次迭代所得时频图内的8.3kHz～12.4kHz；NC 2：第二次迭代所得时频图内的12.5kHz～16.6kHz；NC 3：第二次迭代所得时频图内的29.2kHz～33.3kHz；NC 4：第三次迭代所得时频图内的8.3kHz～12.4kHz；NC 5：第三次迭代所得时频图内的25kHz～29.1kHz；NC 6：第三次迭代所得时频图内的29.2kHz～33.3kHz。
图3h、图3i、图3j、图3k、图3l分别为应用自适应最优核时频分布进行光伏系统故障电弧检测的特征量波形，各特征量所采用的频段选为AOK 1：24.4kHz～30.4kHz；AOK 2：18.3kHz～24.3kHz；AOK 3：12.2kHz～18.2kHz；AOK 4：6.1kHz～12.1kHz；AOK 5：0.1～6.0kHz。
图3m为应用本发明进行光伏系统故障电弧检测的系统状态判断输出信号(针对图3a)。
图4a为负载投切的类弧电流信号。
图4b、图4c、图4d、图4e、图4f、图4g分别为应用非线性调频小波变换进行光伏系统故障电弧检测的特征量波形，各特征量所采用的频段分别选为NC 1、NC 2、NC 3、NC 4、NC 5、NC 6。
图4h为应用本发明进行光伏系统故障电弧检测的系统状态判断输出信号(针对图4a)。
图5a为系统启动的类弧电流信号。
图5b、图5c、图5d、图5e、图5f、图5g分别为应用非线性调频小波变换进行光伏系统故障电弧检测的特征量波形，各特征量所采用的频段分别选为NC 1、NC 2、NC 3、NC 4、NC 5、NC 6。
图5h、图5i、图5j、图5k、图5l分别为应用自适应最优核时频分布进行光伏系统故障电弧检测的特征量波形，各特征量所采用的频段选为AOK 1、AOK 2、AOK 3、AOK 4、AOK 5。
图5m为应用本发明进行光伏系统故障电弧检测的系统状态判断输出信号(针对图5a)。
具体实施方式
下面结合附图和实施例对本发明进行详细说明。
(一)本发明提出的光伏系统故障电弧检测算法的硬件实现
首先对不同类弧工况和故障电弧工况下的光伏系统电流信号按时间窗采样，若所述电流信号为霍尔传感器采集得到，经高通滤波(滤波的目的是去除直流分量)后进行后续的特征层处理；若所述电流信号为电流互感器采集得到，则直接进行后续的特征层处理。所述霍尔传感器或电流互感器安装在所需监测的光伏串内或光伏阵列汇流直流母线上，亦可共用逆变器进行系统电流采集的装置。如图1所示，采集到的电流信号输入至直流故障电弧检测装置，经高通滤波、A/D转换模块输入至DSP处理模块，在DSP模块处理过程中，电流信号经过基于非线性调频小波变换的多特征量处理得到对应的二维矩阵M，用以判 断频谱能量增大的时刻。在发现存在能量增大的突变点后，对当前时间窗长度下的电流信号进行更为精细尺度的划分，形成多个时段，从能量增大的下一时段开始作自适应最优核时频分布，得到相应的多特征值作为朴素贝叶斯模型的输入向量。
利用电流信号对应的电弧电压信号这一先验知识在输入向量最后一行附加类别标记，生成朴素贝叶斯模型的训练学习样本，在朴素贝叶斯模型学习完毕，并测试训练所得模型进行状态分类的准确率后，便可对输入时间窗内的电流采样信号提取到多个基于自适应最优核时频分布的时频特征，输入至训练好的朴素贝叶斯模型中进行状态判断。朴素贝叶斯模型可实时输出光伏系统内是否发生故障电弧的0/1判定结果，发生故障电弧时输出0，判断系统正常运行时输出1。只有在朴素贝叶斯模型连续输出0到达指定时段个数时，才进行故障电弧切除信号的触发，即经DSP处理模块输出的控制信号经D/A转换模块驱动继电器动作，控制光伏系统相应支路处的断路器进行切断动作；在朴素贝叶斯模型未达到既定输出0的周期数前，只要其有一个1输出，则认为此时是类弧工况的干扰引发而不是真正的故障电弧工况。
(二)本发明提出的光伏系统故障电弧检测算法流程
以T NCT长度的时间窗采集信号x t，经过非线性调频小波变换得到x t的迭代时频图，基于所选迭代时频图内的频率分量构造特征量，用以判断频谱能量增大的时刻；在发现存在能量增大的突变点(能量增大的时刻)后，经过自适应最优核时频分布得到x t在时频域内对应的矩阵分布形式，对该矩阵按行(时间维度)求平方和得列向量，选定多个频段，对每个频段进行频率维度的积分运算，得到多个特征值输入至训练好的朴素贝叶斯模型，便可判断当前时段内的光伏系统状态。本发明通过多个有效的时频特征准确辨识光伏系统内故障电弧的同时，还能确保多种类弧工况下不误动，由此提升了光伏系统安全稳定运行的能力。
该光伏系统故障电弧检测算法的核心为基于自适应核函数和瞬时频率估计的光伏系统故障电弧检测方法，算法的具体步骤如下(图2)：
步骤一、参数初始化过程包括时间窗长度T NCT、阈值列向量Y、m(频谱能量增大触发阈值)、故障电弧触发阈值p，以及非线性调频小波变换及自适应最优核时频分布两种时频分析工具内的各项参数(非线性调频小波变换所选用的多项式阶次，迭代次数、自适应最优核时频分布的输出频率划分尺度)等。
光伏系统运行中，以设定的时间窗长度T NCT对电流信号进行采集。这里的电流信号是采用霍尔传感器检测并经高通滤波后的信号或者由电流互感器等所测不含直流分量的 信号。考虑非线性调频小波变换所得的时间、频率分辨率之间的相互约束关系，若时间窗取的过短，会导致非线性调频小波变换所得二维复数时频矩阵的频率划分精度不够，无法反应故障电弧差异于类弧的根本特征；若时间窗取的过长，则加重了所述算法对检测装置硬件在内存和主频上的要求。因此，所述时间窗长度T NCT的取值范围为4～40ms。
步骤二、采用非线性调频小波变换的方法对第t个时间窗下采集的电流检测信号x t进行分析，得到该时间窗下光伏系统输出电流信号多次迭代后的复数矩阵时频分布形式。选定u个迭代复数时频矩阵，对指定频率分量处元素的模进行频率维度的平方和运算(即对选定迭代次数下的时频矩阵按所选故障电弧特征频段沿频率维度作能量积分)，构造n个特征量。其中，通过对比正常、类弧与故障电弧状态下的幅值差异，并考虑在保证故障电弧检测可靠性的前提下尽可能减少迭代及频率分量的计算，确定有效迭代次数u(有效迭代次数等于迭代复数时频矩阵选定个数，是小于或等于迭代终止时的迭代次数的)的取值范围为2～5。所选择的有效迭代次数上所选定的指定频率分量个数可不同，但均在故障电弧特征频段5～50kHz的范围内选定，由各有效迭代次数(u)下的频段分量所构造的特征量总数n的取值范围为5～10。
这n个特征量在该时间窗内形成二维矩阵M，将其沿时间维度与阈值列向量Y进行相应元素比较，若二维矩阵M(M为n×T NCT的二维矩阵)对应特征量的值(由于特征值本身较大，故实际选取其相较于前一时刻的对应特征量的值的变化)大于相应阈值的个数超过m个，则判断该时间窗的电流信号存在频谱能量增大，转至步骤三作进一步的判断，否则判断该采样时刻处于正常工作状态，继续进行下一采样时刻的对应特征值比较直至该时间窗的检测信号均为光伏系统正常工作状态，返回步骤一继续采样。其中，m的取值范围为n-2～n。
步骤三、将电流检测信号x t等分为y段(所划分时段y的取值范围为2～10)，从能量增大的下一时段开始采用自适应最优核时频分布的方法对x t的对应时段逐段进行分析，得到该时段下光伏系统输出电流信号在时频域内对应的复数矩阵时频分布形式。为尽可能在保证故障电弧检测可靠性的前提下减少计算数量、提高判断效率和速度，对相应元素(即有效的频段)的实部进行时间维度的平方和运算得相应列向量，对每个有效频段进行频率维度的积分运算得到多特征量的值作为输入向量，转至步骤四进行朴素贝叶斯模型识别。
其中，自适应最优核时频分布的输出频率划分尺度的取值范围为256～8192。在构建特征量时考虑故障电弧特征频段均在(0,50]kHz的范围进行选定，根据选定结果确定有效频段，所构造的特征量总数r的取值范围为3～7。
步骤四、利用训练好的朴素贝叶斯模型对该输入向量进行状态分类，若该时段呈现故障电弧状态，则输出值为0，转至步骤五；否则输出值为1，判定是类弧干扰，清零计数变量，返回步骤一进行下一时间窗内的检测信号分析。
其中，朴素贝叶斯模型在学习训练过程中的标记：对应时段内的电弧电压信号为零，则标记为1，对应时段内的电弧电压非零，则标记为0，训练样本容量的取值范围为1500～3000。
步骤五、对经朴素贝叶斯模型计算得到的输出值为0的时段个数进行计数，当时段个数达到p个时，则判断光伏系统内发生故障电弧，发出电路开断的控制信号以熄灭故障电弧，否则，返回步骤三进行下一时段内的检测信号分析。
其中，为快速切除故障电弧、不误动地区分类弧工况，所述故障电弧触发阈值p的取值范围为18～120。
在某些类弧工况下，基于非线性调频小波变换所得的特征量在某些时间窗内可能发生阈值误判而失去类弧工况的正常判定能力，而基于自适应最优核时频分布所得的特征量在这些时间窗内的类弧工况则不发生误判，呈现与故障态不同的特征值。朴素贝叶斯模型仍能利用所学习到的故障电弧统计规律较为准确地将类弧工况判定为正常态，由此体现了本发明检测算法关注多重特征量对于提高类弧工况干扰下可靠辨识故障电弧能力的优势。本发明依据提出的多特征量具有较强的故障电弧识别能力，既避免了偶然因素引起直流故障电弧检测装置的误动作，又保证了发出故障电弧支路切断信号的快速性。
(三)上述光伏系统故障电弧检测算法对故障电弧工况的辨识效果
如图3a所示的光伏系统输出电流检测信号，在0.6648s以前，光伏系统处于正常工作状态，0.6648s系统出现故障电弧，电流信号出现大幅值脉冲，此后一直处于故障电弧状态。
通过非线性调频小波变换对电流信号进行分析，对所得到的时频域内二维复数矩阵内指定频率分量的元素取模，在频率维度上采用积分的处理方式(即平方和运算)，构建基于非线性调频小波变换的6个特征量(如图3b～图3g所示)。各个特征量均在故障电弧发生时刻有大幅值的脉冲指示，特征值变化大于相应所设定的阈值(对应阈值列向量Y设为[30，35，75，60，40，30])，均可判定存在频谱能量增大。由于判断为频谱能量增大的特征量个数大于初始给定阈值(M＝5)，对图3a所示故障电弧脉冲后的电流信号进行后续的自适应最优核时频分布及朴素贝叶斯模型分析。
通过自适应最优核时频分布对电流信号进行分析，所得到的时频域内二维复数矩阵内相应元素的实部沿时间维度采用平方和运算的处理方式，在频率维度上采用积分的处理方式，得到基于自适应最优核时频分布的5个特征量。为了显示正常与故障电弧状态的差异，给出全部时段电流信号的自适应最优核时频分布计算结果，如图3h～图3l所示。其中，各个特征量均在故障电弧发生时刻有短暂、极大幅值的脉冲指示，在故障电弧发生后的特征值较正常工作状态整体变大，这种一致的大幅值状态有利于故障电弧的准确辨识。
将上述计算所得的5个时频特征值输入至训练好的朴素贝叶斯模型，判断光伏系统内是否存在故障电弧。当朴素贝叶斯模型输出1时，则判断该时段内光伏系统处于类弧状态，前述非线性调频小波变换发生了误判，继续进行下一时段内光伏系统输出电流信号的状态检测；当朴素贝叶斯模型输出0时，则判断该时段内光伏系统可能发生了故障电弧，还需通过所设定的故障电弧切除标准(P)进一步确认故障电弧的发生，即连续输出0的周期达到100个后，则确定光伏系统内发生了故障电弧，给相应的断路器发出切断故障电弧支路信号。如图3m所示的结果，检测算法面对正常工作电流能够给出正确的输出(判断其未达到故障电弧切除标准的要求)，对故障态电流信号能够给出正确的输出(判断其达到故障电弧切除标准的要求)。
(四)上述光伏系统故障电弧检测算法对光伏系统内多种源于正常操作形成的类弧工况的辨识效果
4.1对于光伏系统内通过负载切换所形成类弧的辨识效果
如图4a所示的光伏系统输出电流检测信号，在1.093s以前，光伏系统处于正常工作状态，1.093s发生负载切换过程，形成时域上类似的电流突变过程。
通过非线性调频小波变换对电流信号进行分析，对所得到的时频域内二维复数矩阵内各元素取模，在频率维度上采用积分的处理方式(即平方和运算)，构建基于非线性调频小波变换的6个特征量如图4b～图4g所示。其中，各个特征量均在系统切换负载时刻幅值显著降低，特征值变化小于Y中相应给定阈值，均可判定不存在频谱能量增大。由于判断频谱能量增大的特征量个数小于设定阈值m，电流信号不进行后续的自适应最优核时频分布及朴素贝叶斯模型分析步骤，如图4h所示的结果，阈值(m)判定能够给出正确的输出，判断该时间窗内光伏系统处于正常工作状态，继续进行下一时间窗内光伏系统输出电流信号的状态检测。
4.2对光伏系统内通过系统启动所形成类弧的辨识效果
如图5a所示的光伏系统输出电流检测信号，在2.256s以前，光伏系统处于正常停机状态，2.256s时，系统开始启动过程，形成类似于图4a的较小时域电流幅值等级，2.586s时，系统结束启动过程，形成较大的时域电流幅值等级，进入正常工作状态。
通过非线性调频小波变换对电流信号进行分析，对所得到的时频域内二维复数矩阵内各元素取模，在频率维度上采用积分的处理方式(即平方和运算)，构建基于非线性调频小波变换的6个特征量。2.256s时刻的类弧工况与图4a所示类弧工况类似，故而相应基于非线性调频小波变换所构建的特征量幅值也会出现如图4b～图4g一样的显著降低过程，特征值变化小于相应阈值，阈值判定处于正常工作状态，尽管如图5h～图5l所示基于自适应最优核时频分布分析所构建的特征量出现如图3h～图3l一样的故障电弧模式，但基于非线性调频小波变换能够预先给出正确的判断，不进行后续的自适应最优核时频分布分析。
这里的非线性调频小波变换层级仅给出进行后续自适应最优核时频分布及朴素贝叶斯模型分析的对应电流分析结果，如图5b～图5g所示。2.586s时，基于非线性调频小波变换的各特征量均出现如图3b～图3g所示的持续幅值增大过程，判定存在频谱能量增大，转至后续的自适应最优核时频分布分析步骤。通过自适应最优核时频分布对电流信号进行分析，所得到的时频域内二维复数矩阵内相应元素的实部沿时间维度采用平方和运算的处理方式，在频率维度上采用积分的处理方式，得到基于自适应最优核时频分布的5个特征量，全部时段电流信号的计算结果如图5h～图5l所示。各个特征量在2.586s之前的特征值均幅值极小，其幅值状态明显区别于故障电弧发生之前的正常工作状态(如图3h～图3l内0.6s之前的幅值等级和图5h～图5l内2.8s以后的幅值等级)，故而系统启动过程与正常工作状态是存在显著的模式差别的，这也证实了基于自适应最优核时频分布所构建的特征量是具有区分故障电弧和类弧工况能力的。2.256s及2.586s时刻分别由非线性调频小波变换、自适应最优核时频分布发挥作用完成了正确的系统状态判断过程，这充分说明了多元方法综合判断对准确判断故障电弧的重要性。
将五个计算所得的时频特征值输入至朴素贝叶斯模型，朴素贝叶斯模型因持续出现的1输出而达不到触发标准P，判定此时光伏系统内发生了类弧工况，从而不发出断路器控制信号，继续进行下一时间窗内光伏系统输出电流信号的状态检测。如图5m所示的结果，检测算法能够给出正确的输出(判断其未达到故障电弧切除标准的要求)，不造成误判。
总之，本发明所提供的光伏故障电弧检测方法分为非线性调频小波变换-阈值判断和自适应最优核时频分布-朴素贝叶斯模型分类两步，前一步计算分辨率高，对于故障电弧 的发生时刻捕捉精准，一旦识别当前分析时间窗内有能量增大趋势能够迅速进入后一步。后一步与前一步的交错时序令整个算法消耗内存少，计算效率高。两步中均构建了多个时频特征量，即使个别特征量失效，仍能依靠其他有效特征量正确完成故障电弧与类弧的有效辨识，提高了故障电弧或类弧工况辨识的可靠性，解决了类弧工况干扰下准确、可靠、快速动作光伏系统内故障电弧的问题，有效防止了故障电弧给光伏系统运行、人身财产带来的安全威胁。

Claims (9)

  1. 一种基于自适应核函数和瞬时频率估计的光伏系统故障电弧检测方法，其特征在于：该光伏系统故障电弧检测方法包括以下步骤：
    1)对光伏系统内的电流信号，以T NCT为时间窗长度进行逐点采样，得到检测信号x t，转至步骤2)；
    2)对检测信号x t作非线性调频小波变换，得到检测信号x t多次迭代的复数时频矩阵，通过选定u个迭代复数时频矩阵，以及对选定的迭代复数时频矩阵中指定频率分量处元素的模进行频率维度的平方和运算，构造n个特征量NC i(t)，i＝1,2,3…,n，并形成n×T NCT的二维矩阵M，将M沿时间维度求取后一时刻相对前一时刻的特征量变化，并与阈值列向量Y进行相应元素比较，若该时间窗内某采样时刻对应的特征量变化的值大于相应阈值的个数超过m个，则判断该时间窗的电流信号存在频谱能量增大，将检测信号x t等分为y个时段，转至步骤3)；否则，判断该采样时刻光伏系统处于正常工作状态，继续进行下一采样时刻的对应特征量变化的比较，直至该时间窗的检测信号均为光伏系统正常工作状态时，返回步骤1)进行下一时间窗的检测信号采集；
    3)从频谱能量增大的下一时段开始作自适应最优核时频分布，得到在时频域内对应的复数时频矩阵，选定频率维度上r个频段，对相应元素的实部进行时间维度的平方和运算得相应列向量，然后对选定的每个频段进行频率维度的积分运算，构造得到r个特征量作为输入向量，转至步骤4)；
    4)利用训练好的朴素贝叶斯模型及对应的r个特征量的值对该输入向量进行状态分类，若输出值为0，则判定该时段呈现故障电弧状态，转至步骤5)；若输出值为1，则判定是类弧干扰，清零计数变量，返回步骤1)进行下一时间窗的检测信号采集；
    5)对经所述朴素贝叶斯模型进行状态分类所得的输出值为0的时段个数利用计数变量进行计数，当输出值为0的时段个数达到p个时，则判断光伏系统发生故障电弧，发出电路开断的控制信号以熄灭故障电弧，否则继续按照步骤3)进行下一时段的检测信号分析。
  2. 根据权利要求1所述一种基于自适应核函数和瞬时频率估计的光伏系统故障电弧检测方法，其特征在于：所述电流信号为由霍尔传感器采样并经高通滤波后的信号或者为直接由电流互感器采样的信号。
  3. 根据权利要求1所述一种基于自适应核函数和瞬时频率估计的光伏系统故障电弧检测方法，其特征在于：所述T NCT的取值范围为4～40ms；y的取值范围为2～10。
  4. 根据权利要求1所述一种基于自适应核函数和瞬时频率估计的光伏系统故障电弧检测方法，其特征在于：所述非线性调频小波变换所选用的多项式阶次的取值范围为10～30；迭代终止条件为迭代达到既定次数即停止，迭代次数的取值范围为3～8；自适应最优核时频分布的输出频率划分尺度的取值范围为256～8192。
  5. 根据权利要求1所述一种基于自适应核函数和瞬时频率估计的光伏系统故障电弧检测方法，其特征在于：所述u的取值范围为2～5；所述频率分量根据故障电弧特征频段进行指定，不同迭代次数上选定的故障电弧特征频段个数相同或不同，故障电弧特征频段均在5～50kHz的范围内选定，所选定的故障电弧特征频段部分重叠或连续分割或存在间隔；n的取值范围为5～10。
  6. 根据权利要求1所述一种基于自适应核函数和瞬时频率估计的光伏系统故障电弧检测方法，其特征在于：所述阈值列向量Y中的阈值设定原则为能够获取信号在各特征量层面上的幅值变化模式，不同特征量NC i(t)对应设定的阈值相同或不同；m的取值范围为n-2～n。
  7. 根据权利要求1所述一种基于自适应核函数和瞬时频率估计的光伏系统故障电弧检测方法，其特征在于：所述r个频段根据故障电弧特征频段选定，故障电弧特征频段在(0,50]kHz的范围进行选定，所选定的故障电弧特征频段部分重叠或连续分割或存在间隔；r的取值范围为3～7。
  8. 根据权利要求1所述一种基于自适应核函数和瞬时频率估计的光伏系统故障电弧检测方法，其特征在于：所述朴素贝叶斯模型在学习训练过程中的输出状态标记方式为：对应时段内的电弧电压为零，则标记为1，对应时段内的电弧电压非零，则标记为0；训练样本容量的取值范围为1500～3000。
  9. 根据权利要求1所述一种基于自适应核函数和瞬时频率估计的光伏系统故障电弧检测方法，其特征在于：所述p的取值范围为18～120。
PCT/CN2018/125440 2018-10-29 2018-12-29 基于自适应核函数和瞬时频率估计的光伏系统故障电弧检测方法 Ceased WO2020087752A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US16/629,033 US11489490B2 (en) 2018-10-29 2018-12-29 Arc fault detection method for photovoltaic system based on adaptive kernel function and instantaneous frequency estimation

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201811269779.1A CN109560770B (zh) 2018-10-29 2018-10-29 基于自适应核函数和瞬时频率估计的光伏系统故障电弧检测方法
CN201811269779.1 2018-10-29

Publications (1)

Publication Number Publication Date
WO2020087752A1 true WO2020087752A1 (zh) 2020-05-07

Family

ID=65865434

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2018/125440 Ceased WO2020087752A1 (zh) 2018-10-29 2018-12-29 基于自适应核函数和瞬时频率估计的光伏系统故障电弧检测方法

Country Status (3)

Country Link
US (1) US11489490B2 (zh)
CN (1) CN109560770B (zh)
WO (1) WO2020087752A1 (zh)

Cited By (18)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113095173A (zh) * 2021-03-29 2021-07-09 西安交通大学 一种Adaboost融合多分类器的光伏系统故障电弧检测方法
CN113297786A (zh) * 2021-05-06 2021-08-24 西安理工大学 一种基于半监督式机器学习的低压故障电弧感知方法
CN113626352A (zh) * 2021-07-01 2021-11-09 珠海全志科技股份有限公司 内存控制器的读取校准方法、计算机装置和可读存储介质
CN114034960A (zh) * 2021-11-25 2022-02-11 国网重庆市电力公司电力科学研究院 一种基于弱电弧特征的故障电弧检测方法
CN114239639A (zh) * 2021-11-08 2022-03-25 深圳供电局有限公司 特征选择方法、装置、计算机设备和存储介质
CN114880627A (zh) * 2022-07-08 2022-08-09 西南交通大学 一种基于迭代解调时变滤波的自适应瞬时频率估计方法
CN115078934A (zh) * 2022-06-27 2022-09-20 安徽大学 基于小波lstm自编码器单类重构检测拉弧故障方法及系统
CN115144182A (zh) * 2022-09-01 2022-10-04 杭州景业智能科技股份有限公司 轴承健康状态监测方法、装置、计算机设备及存储介质
CN116522113A (zh) * 2023-04-28 2023-08-01 江苏大学 轮毂电机电动汽车驱动系统共性与故障特征协同提取方法
CN116610990A (zh) * 2023-04-06 2023-08-18 安徽继远软件有限公司 一种基于特征空间差异化的断路器隐患识别方法及装置
CN117216518A (zh) * 2023-06-30 2023-12-12 西安交通大学 一种建筑光伏系统直流故障电弧检测方法及系统
CN113659542B (zh) * 2021-07-26 2024-03-05 西安理工大学 一种低压直流配用电系统故障电弧保护方法
CN118747324A (zh) * 2024-06-11 2024-10-08 扬州大学 一种基于时频复合递归图的直流电弧故障检测方法和系统
CN118759417A (zh) * 2024-09-09 2024-10-11 四川蜀道电气化建设有限公司 电气化铁路馈线电缆接地保护智能监测方法
CN118861755A (zh) * 2024-06-27 2024-10-29 国网四川省电力公司电力科学研究院 基于人工智能可解释性方法的致火电弧故障实时检测方法
CN118937987A (zh) * 2024-10-11 2024-11-12 西安亚能电气有限责任公司 变电站隔离开关故障监测系统
CN119646490A (zh) * 2025-02-14 2025-03-18 上海宽域工业网络设备有限公司 电力系统宽频测量数据处理方法、装置、设备及介质
CN120103086A (zh) * 2025-04-02 2025-06-06 重庆邮电大学 一种基于线谱抑制的光伏直流系统故障检测方法及装置

Families Citing this family (47)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110045252B (zh) * 2019-05-06 2020-11-10 西安交通大学 一种利用频谱能量积分的串联直流电弧检测方法
CN111846095B (zh) * 2019-05-14 2022-05-17 北京骑胜科技有限公司 一种故障检测设备、电助力车及故障检测方法
CN112345891A (zh) * 2019-08-07 2021-02-09 青岛鼎信通讯股份有限公司 一种基于电流多通道时频特征提取的故障电弧检测方法
CN110969194B (zh) * 2019-11-21 2023-12-19 国网辽宁省电力有限公司电力科学研究院 基于改进型卷积神经网络的电缆早期故障定位方法
JP7546224B2 (ja) * 2020-08-26 2024-09-06 パナソニックＩｐマネジメント株式会社 異常検出装置、異常検出方法及びプログラム
CN112433128B (zh) * 2020-10-13 2022-12-02 西安理工大学 一种基于随机共振的直流故障电弧诊断方法及装置
CN113219374B (zh) * 2021-03-05 2024-04-30 国网江苏省电力有限公司徐州供电分公司 一种变压器电压故障检测方法
CN113298124B (zh) * 2021-05-08 2023-03-24 西安理工大学 一种基于参数匹配的多层次直流故障电弧检测方法
EP4363820A1 (en) * 2021-06-29 2024-05-08 General Electric Renovables España, S.L. Systems and methods for controlling an industrial asset of an asset family
CN113437723B (zh) * 2021-07-06 2022-08-12 丰郅(上海)新能源科技有限公司 故障电弧的检测方法
CN113740672B (zh) * 2021-07-14 2023-06-06 国网江苏省电力有限公司电力科学研究院 一种基于k线图的光伏组件阵列电弧故障的检测方法
CN114062880B (zh) * 2021-12-20 2022-12-13 西安交通大学 基于自适应信号处理及cart树集成学习的直流故障电弧检测方法
CN114566936B (zh) * 2022-02-14 2023-10-31 福建华电电力工程有限公司 一种火电厂低压交流系统继电保护脱扣触发方法与装置
CN114282689B (zh) * 2022-03-04 2022-05-31 苏芯物联技术(南京)有限公司 一种普适焊接起弧实时智能检测方法及系统
CN114814478A (zh) * 2022-03-08 2022-07-29 福州大学 一种电弧状态观测器及其建立方法
CN114636851B (zh) * 2022-03-09 2025-08-01 阳光电源股份有限公司 频段选择方法及装置、电子设备、计算机可读存储介质
CN114609475B (zh) * 2022-03-18 2024-11-08 中国民航大学 一种交流故障电弧检测方法和系统
CN114371342B (zh) * 2022-03-21 2022-05-27 国仪量子(合肥)技术有限公司 Fpga及基于其的实时信号测频方法以及锁相放大器
CN114584069B (zh) * 2022-05-05 2022-08-05 华中科技大学 一种光伏直流电弧检测方法及系统
CN114755546B (zh) * 2022-06-14 2022-08-26 锦浪科技股份有限公司 一种光伏系统直流故障电弧的检测方法、装置及光伏系统
CN115167364A (zh) * 2022-07-27 2022-10-11 山东科技大学 一种基于概率变换与统计特性分析的早期故障检测方法
CN115563466B (zh) * 2022-09-22 2025-08-29 中国人民解放军国防科技大学 面向多分量强时变信号的同步压缩自适应变换方法及装置
CN115293303B (zh) * 2022-10-10 2023-01-24 广东电网有限责任公司中山供电局 一种高压输电线网监测方法、系统、设备和介质
CN115603278A (zh) * 2022-10-19 2023-01-13 西安交通大学（Cn） 基于压力频率特征的油浸式变压器数字式非电量保护方法
CN115906492A (zh) * 2022-11-28 2023-04-04 深圳供电局有限公司 一种电弧宏微观参数关联模型构建方法及装置
CN115982604A (zh) * 2022-11-29 2023-04-18 国网山东省电力公司电力科学研究院 一种串联电弧故障检测方法及系统
CN115967349B (zh) * 2022-12-05 2026-04-21 上海电气分布式能源科技有限公司 一种光伏发电损失评估方法、装置、电子设备及存储介质
CN116231586B (zh) * 2022-12-30 2025-11-25 山东大学 基于瞬时差动功率时频能量提取的风电场升压变保护方法及系统
CN116068346B (zh) * 2023-02-16 2024-12-03 固德威技术股份有限公司 一种电弧故障确认方法、装置以及介质
CN116578837B (zh) * 2023-03-24 2025-05-13 重庆大学 一种大转速波动下的旋转机械无键相阶次跟踪方法
CN116132485B (zh) * 2023-04-04 2023-08-11 广州极联智能科技有限公司 一种智能化项目快捷调试部署方法及系统
CN116679163B (zh) * 2023-06-13 2025-12-19 沈阳工业大学 一种风电场线路故障识别方法、装置、介质及电子设备
CN117250443B (zh) * 2023-07-05 2026-01-09 西安理工大学 电力电子干扰条件下直流故障电弧检测与高效防护方法
CN117036825B (zh) * 2023-08-25 2026-02-17 众淼创新科技(青岛)股份有限公司 一种太阳能电池板检测方法、介质及系统
CN117277246B (zh) * 2023-11-15 2024-02-06 昆明理工大学 一种有源无源结合的配电网接地故障自适应熄弧方法
CN117763399B (zh) * 2024-02-21 2024-05-14 电子科技大学 一种自适应变长信号输入的神经网络分类方法
CN118226192B (zh) * 2024-03-11 2025-11-28 西安理工大学 基于s变换的柔性直流输电线路区内外故障判别方法
CN118584264B (zh) * 2024-05-21 2025-11-07 西安交通大学 基于同步压缩小波变换和arima算法的直流故障电弧检测方法
CN118501617B (zh) * 2024-07-17 2024-10-18 国网山东省电力公司聊城供电公司 一种直流配电系统线路故障识别方法
CN118840567B (zh) * 2024-08-26 2025-05-09 山东浪潮智慧能源科技有限公司 一种光伏逆变器的热故障识别方法、装置及存储介质
CN118937904B (zh) * 2024-09-10 2025-12-26 广东电网有限责任公司 一种配电网高阻接地故障检测方法和装置
CN119782873B (zh) * 2024-11-27 2025-10-21 国网江苏省电力有限公司连云港供电分公司 一种一二次融合集中式dtu设备智能模拟试验方法及系统
CN119936568B (zh) * 2025-04-08 2025-07-11 国网上海市电力公司 一种非在运电缆的断点定位方法及系统
CN120370063A (zh) * 2025-04-14 2025-07-25 利锐特电气有限公司 一种基于多源数据融合的配电箱运行状态监测方法及系统
CN120729410B (zh) * 2025-08-26 2025-12-16 内蒙古电力(集团)有限责任公司包头供电分公司 基于时域反射的配电网光纤故障定位方法及系统
CN120955559B (zh) * 2025-10-15 2025-12-23 浙江翔远电气有限公司 一种配电箱电弧故障预警系统
CN121239143B (zh) * 2025-12-02 2026-03-27 兰州理工大学 一种多运行工况下光伏系统直流电弧故障诊断方法及装置

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20120174961A1 (en) * 2011-01-12 2012-07-12 Schneider Electric USA, Inc. Arc fault mitigation for photovoltaic systems
US9722535B2 (en) * 2015-03-25 2017-08-01 Delta Electronics (Shanghai) Co., Ltd Method and device for ARC fault detection for photovoltaic inverter and photovoltaic inverter using the same
CN107086855A (zh) * 2017-04-25 2017-08-22 西安交通大学 一种机器学习融合多时频特征的光伏系统故障电弧检测方法
CN107154783A (zh) * 2017-04-18 2017-09-12 西安交通大学 应用独立成分分析和s变换检测系统过程耦合情况下光伏系统故障电弧的方法
CN107181460A (zh) * 2016-03-09 2017-09-19 上海英孚特电子技术有限公司 一种光伏系统电弧故障检测方法
CN107294494A (zh) * 2017-07-03 2017-10-24 上海岩芯电子科技有限公司 光伏系统直流侧电弧故障检测装置及其检测方法

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9945894B2 (en) * 2012-02-29 2018-04-17 Innovative Scientific Solutions, Inc. Arc fault detection
US9025287B2 (en) * 2012-12-19 2015-05-05 Stmicroelectronics S.R.L. Arc fault detection equipment and method using low frequency harmonic current analysis
CN106443297B (zh) * 2016-09-07 2018-05-22 江南大学 光伏二极管箝位型三电平逆变器的决策树svm故障诊断方法

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20120174961A1 (en) * 2011-01-12 2012-07-12 Schneider Electric USA, Inc. Arc fault mitigation for photovoltaic systems
US9722535B2 (en) * 2015-03-25 2017-08-01 Delta Electronics (Shanghai) Co., Ltd Method and device for ARC fault detection for photovoltaic inverter and photovoltaic inverter using the same
CN107181460A (zh) * 2016-03-09 2017-09-19 上海英孚特电子技术有限公司 一种光伏系统电弧故障检测方法
CN107154783A (zh) * 2017-04-18 2017-09-12 西安交通大学 应用独立成分分析和s变换检测系统过程耦合情况下光伏系统故障电弧的方法
CN107086855A (zh) * 2017-04-25 2017-08-22 西安交通大学 一种机器学习融合多时频特征的光伏系统故障电弧检测方法
CN107294494A (zh) * 2017-07-03 2017-10-24 上海岩芯电子科技有限公司 光伏系统直流侧电弧故障检测装置及其检测方法

Cited By (25)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113095173B (zh) * 2021-03-29 2024-04-02 西安交通大学 一种Adaboost融合多分类器的光伏系统故障电弧检测方法
CN113095173A (zh) * 2021-03-29 2021-07-09 西安交通大学 一种Adaboost融合多分类器的光伏系统故障电弧检测方法
CN113297786B (zh) * 2021-05-06 2023-04-07 西安理工大学 一种基于半监督式机器学习的低压故障电弧感知方法
CN113297786A (zh) * 2021-05-06 2021-08-24 西安理工大学 一种基于半监督式机器学习的低压故障电弧感知方法
CN113626352A (zh) * 2021-07-01 2021-11-09 珠海全志科技股份有限公司 内存控制器的读取校准方法、计算机装置和可读存储介质
CN113626352B (zh) * 2021-07-01 2024-04-30 珠海全志科技股份有限公司 内存控制器的读取校准方法、计算机装置和可读存储介质
CN113659542B (zh) * 2021-07-26 2024-03-05 西安理工大学 一种低压直流配用电系统故障电弧保护方法
CN114239639A (zh) * 2021-11-08 2022-03-25 深圳供电局有限公司 特征选择方法、装置、计算机设备和存储介质
CN114034960A (zh) * 2021-11-25 2022-02-11 国网重庆市电力公司电力科学研究院 一种基于弱电弧特征的故障电弧检测方法
CN115078934A (zh) * 2022-06-27 2022-09-20 安徽大学 基于小波lstm自编码器单类重构检测拉弧故障方法及系统
CN114880627B (zh) * 2022-07-08 2022-09-23 西南交通大学 一种基于迭代解调时变滤波的自适应瞬时频率估计方法
CN114880627A (zh) * 2022-07-08 2022-08-09 西南交通大学 一种基于迭代解调时变滤波的自适应瞬时频率估计方法
CN115144182A (zh) * 2022-09-01 2022-10-04 杭州景业智能科技股份有限公司 轴承健康状态监测方法、装置、计算机设备及存储介质
CN116610990A (zh) * 2023-04-06 2023-08-18 安徽继远软件有限公司 一种基于特征空间差异化的断路器隐患识别方法及装置
CN116522113A (zh) * 2023-04-28 2023-08-01 江苏大学 轮毂电机电动汽车驱动系统共性与故障特征协同提取方法
CN116522113B (zh) * 2023-04-28 2025-11-07 江苏大学 轮毂电机电动汽车驱动系统共性与故障特征协同提取方法
CN117216518A (zh) * 2023-06-30 2023-12-12 西安交通大学 一种建筑光伏系统直流故障电弧检测方法及系统
CN118747324A (zh) * 2024-06-11 2024-10-08 扬州大学 一种基于时频复合递归图的直流电弧故障检测方法和系统
CN118861755A (zh) * 2024-06-27 2024-10-29 国网四川省电力公司电力科学研究院 基于人工智能可解释性方法的致火电弧故障实时检测方法
CN118861755B (zh) * 2024-06-27 2025-05-06 国网四川省电力公司电力科学研究院 基于人工智能可解释性方法的致火电弧故障实时检测方法
CN118759417A (zh) * 2024-09-09 2024-10-11 四川蜀道电气化建设有限公司 电气化铁路馈线电缆接地保护智能监测方法
CN118937987A (zh) * 2024-10-11 2024-11-12 西安亚能电气有限责任公司 变电站隔离开关故障监测系统
CN119646490A (zh) * 2025-02-14 2025-03-18 上海宽域工业网络设备有限公司 电力系统宽频测量数据处理方法、装置、设备及介质
CN120103086A (zh) * 2025-04-02 2025-06-06 重庆邮电大学 一种基于线谱抑制的光伏直流系统故障检测方法及装置
CN120103086B (zh) * 2025-04-02 2026-01-16 重庆邮电大学 一种基于线谱抑制的光伏直流系统故障检测方法及装置

Also Published As

Publication number Publication date
US11489490B2 (en) 2022-11-01
CN109560770A (zh) 2019-04-02
US20210036656A1 (en) 2021-02-04
CN109560770B (zh) 2020-03-13

Similar Documents

Publication Publication Date Title
CN109560770B (zh) 基于自适应核函数和瞬时频率估计的光伏系统故障电弧检测方法
CN107086855B (zh) 一种机器学习融合多时频特征的光伏系统故障电弧检测方法
EP1861726B1 (en) Method and apparatus for generalized arc fault detection
CN112183628B (zh) 基于多种线性时频变换的交流电弧故障检测方法及系统
CN103913663A (zh) 一种直流系统电弧故障在线检测方法和保护装置
CN111650470A (zh) 微电网线路区段快速自适应故障检测与识别方法
CN107154783A (zh) 应用独立成分分析和s变换检测系统过程耦合情况下光伏系统故障电弧的方法
CN105093132A (zh) 一种大功率整流器开路故障快速在线诊断方法
CN108963962B (zh) 一种多层级电弧故障断路系统
CN114062880A (zh) 基于自适应信号处理及cart树集成学习的直流故障电弧检测方法
CN107015140A (zh) 一种有载分接开关弹簧动能不足故障的振动信号分析方法
CN109635430A (zh) 电网输电线路暂态信号监测方法和系统
CN110095680A (zh) 一种低压短路故障快速检测与短路电流峰值预测方法
CN104198893B (zh) 自适应故障电流检测方法
CN114859266A (zh) 一种chb光伏并网逆变器系统开路故障诊断方法
CN106324437A (zh) 一种基于电压极值的隐患放电电流的检测方法
CN118152942A (zh) 一种基于形状、能量及统计特征的特高频局放识别方法
CN117216518A (zh) 一种建筑光伏系统直流故障电弧检测方法及系统
CN110568300B (zh) 一种基于多源信息的配电网单相接地故障辨识方法
CN117060352A (zh) 一种基于修正电压的柔直电网新型暂态量保护方法
CN112362956B (zh) 基于三级分压的金属氧化物避雷器过电压检测方法及系统
CN120801934A (zh) 一种用于储能系统的直流拉弧检测系统及方法
CN120357241A (zh) 一种可远程控制的漏电安全保护插头
Venkatesh et al. Wavelet-ANN based classification of HVDC converter faults
CN113178847A (zh) 一种串联故障电弧在线检测方法及装置

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: 18938941

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 18938941

Country of ref document: EP

Kind code of ref document: A1

122 Ep: pct application non-entry in european phase

Ref document number: 18938941

Country of ref document: EP

Kind code of ref document: A1

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: PURSUANT TO RULE 112(1) EPC (EPO FORM 1205 DATED 13/12/2021)

122 Ep: pct application non-entry in european phase

Ref document number: 18938941

Country of ref document: EP

Kind code of ref document: A1