CN107154783B - The method for detecting photovoltaic system failure electric arc using independent component analysis and S-transformation - Google Patents

Abstract

The invention discloses a method for detecting a fault arc in a photovoltaic system by using independent component analysis and S-transformation. Based on the output current signal of the photovoltaic system, the independent main source signal of the current is obtained through independent component analysis, and the frequency of the signal after Fourier transform is The information is subjected to variance processing to obtain the first feature quantity, the current is processed through the S-transformation, and the time and high-frequency components of the obtained time-frequency matrix are integrated to form the second feature quantity. After the feature value is compared with its corresponding set threshold, the weight coefficient is used to weight the output judgment results of the two feature values on the decision-making layer to complete the real-time detection of the fault arc of the photovoltaic system. Through the two decision-making results of dynamic threshold comparison and weight coefficient weighting, the present invention can obviously excavate the essential difference of photovoltaic system fault arc occurring in the system process, and can quickly and accurately cut off the fault arc of photovoltaic system under coupling conditions, improving the photovoltaic system safe and stable operation capability.

Description

A Method for Detecting Arc Faults in Photovoltaic Systems Using Independent Component Analysis and S-Transform

technical field

The invention belongs to the technical field of photovoltaic electrical fault detection, and specifically relates to a method of applying independent component analysis and S transformation to obtain two characteristic values, comparing the characteristic values with corresponding dynamic setting thresholds to obtain two decision results, and using dynamically set The weight coefficient weights the decision results of the two characteristic values to detect the fault arc of the photovoltaic system in real time, and obviously excavates the essential difference of the fault arc of the photovoltaic system that occurs in the system process, and improves the rapidity and reliability of the fault arc detection of the photovoltaic system in the case of coupling In order to ensure that the photovoltaic system can operate stably, safely and economically at any time.

Background technique

The global energy crisis and climate warming are becoming more and more serious, making photovoltaic, wind power, fuel cells and other new green and renewable energy sources more and more widely used. In recent years, with the continuous reduction of the cost of photovoltaic products, the photovoltaic industry at home and abroad has grown rapidly. The expansion of the scale of the photovoltaic power generation system has increased the output voltage of the DC terminal of the photovoltaic system, generally from tens of volts to hundreds of volts, and large-scale photovoltaic power stations can even reach thousands of volts of DC high voltage. The aging degree of insulation has been increased, making photovoltaic system faults more and more frequent, and the fault arc on the DC side of the photovoltaic system is one of them. Once a photovoltaic system arc fault occurs, it is more dangerous because there is no zero-crossing point of the AC fault arc. If the photovoltaic system arc fault protection measures cannot be taken in time, it will cause huge damage to photovoltaic modules and transmission lines and even cause fires, resulting in serious accidents. Security issues such as economic losses and casualties. The earliest photovoltaic system fault arc can be traced back to the Mont Soleil photovoltaic power station in Sweden in the 1990s. Especially since 2006, more and more fire accidents in photovoltaic systems have been reported by the media. Fires occurred in residential photovoltaic facilities and commercial photovoltaic facilities. facilities and large-scale photovoltaic power plants. The photovoltaic system fire accident that occurred in 2006 was caused by a photovoltaic system fault arc in the BP solar junction box. BP also caused huge economic losses due to the recall and replacement of most defective photovoltaic modules. Therefore, it is of great significance to ensure the safe and reliable operation of photovoltaic power generation systems to comprehensively and effectively detect fault arcs in photovoltaic systems and control them in the bud.

At present, the relevant research objects at home and abroad are all aimed at the uncoupled photovoltaic system arc fault, that is, there is no system process when the photovoltaic system arc fault occurs. However, in actual testing, the photovoltaic system frequently experiences transient processes such as power changes and startups originating from the photovoltaic array side or the load side, which causes the photovoltaic system power to experience frequent transient changes. The occurrence time of fault arc in photovoltaic system is uncontrollable, so there is a certain probability that fault arc in photovoltaic system will occur in these system processes. For example, during the start-up of the photovoltaic system and the increase of system power, the output current of the photovoltaic system continues to increase. On the other hand, the fault arc of the series photovoltaic system will reduce the output current of the photovoltaic system. Therefore, in the photovoltaic system arc fault caused by system process coupling, the fault output current of the photovoltaic system will not be different from the normal output current of the photovoltaic system macroscopically, and the requirements for the fault arc detection algorithm of the photovoltaic system are more stringent. If the existing detection algorithm correctly determines the normal output current of the photovoltaic system, the fault arc of the photovoltaic system cannot be detected in time, and the fault arc detection device on the DC side of the corresponding photovoltaic system will refuse to operate, and the fault arc of the photovoltaic system that cannot be eliminated will cause Photovoltaic system fire accidents, resulting in loss of life and property; from the perspective of correctly determining the fault output current of the photovoltaic system, the normal operation of the photovoltaic system will cause misjudgment, and the corresponding photovoltaic system DC side fault arc detection device will malfunction. These misjudgments The normal state of the photovoltaic system will cause the shutdown of the photovoltaic system and reduce the power generation efficiency of the system. Therefore, the detection algorithm must extract the fundamental characteristics of the fault arc of the photovoltaic system under the coupling of the system process, accurately identify the time when the fault arc of the photovoltaic system occurs, and accurately, reliably and quickly identify the output current state of the photovoltaic system, thereby realizing the installation of the DC side of the photovoltaic system. Functional requirements for arc fault detection devices.

Contents of the invention

The purpose of the present invention is to accurately, reliably and quickly identify the fault arc of the photovoltaic system that occurs in the system process, and provides a method for detecting the fault arc of the photovoltaic system by applying independent component analysis and S-transformation.

To achieve the above object, the present invention adopts the following technical solutions:

1) Through multiple current sensors, the output current signal of the photovoltaic system is sampled point by point according to the sampling frequency f _s to obtain multiple current signals x _i,j , where i is the serial number of the current sensor, i∈N and i>1, j is the serial number of the analysis period, j∈N ₊ , for any two different values of i, when j takes the same value, x _{i and j} have the same number of sampling points N, when N meets the requirements of the analysis period, turn to To step 2) carry out the characteristic analysis of the first arc fault;

2) Form a high-dimensional mixed signal matrix X=[x _1,j ,x _2,j ,…, _xi,j ] ^T from the collected multi-channel current signals, and perform de-meaning and whitening processing on the obtained mixed signal matrix, After fast independent component analysis, the unmixing matrix W can be obtained, and the source signal matrix S=WX=[s _1,j ,s _2,j ,…,s _i,j ] ^T is calculated, where S contains the effective source signal and noise signal, select an effective independent main source signal s _1,j , perform fast Fourier transform on s _1,j , calculate the variance of the one-dimensional frequency matrix in the frequency domain, and obtain the first characteristic value r _1,j , go to step 3);

3) Set the threshold of the first feature value in the current analysis period to A ₁ ×μ _1,j -A ₂ ×σ _1,j , where μ _1,j is all the first features from the first analysis period to the current analysis period σ _1,j is the standard deviation of all the first feature values from the first analysis period to the current analysis period, A ₁ ∈ Z, A ₂ ∈ Z, the first feature value and the set Comparing the threshold value of the first feature value, output the corresponding level judgment result: if r _1,j ≥ A ₁ ×μ _1,j -A ₂ ×σ _1,j , then output the judgment result 0, and store it in the first fault arc Judgment matrix out ₁ [j]; if r _1,j <A ₁ ×μ _1,j –A ₂ ×σ _1,j , output the judgment result 1 and store it in the first arc fault judgment matrix out ₁ [j] , go to step 4) to carry out the second fault arc characteristic analysis;

4) Select one of the multiple current signals to perform S-transformation to obtain a two-dimensional complex time-frequency matrix in the time-frequency domain, calculate the integral of the absolute value of the high-frequency component in the frequency dimension along time, and obtain the second characteristic value r _{2, j} , go to step 5);

5) Set the threshold value of the second feature quantity in the current analysis period to A ₃ ×μ _2,j -A ₄ ×σ _2,j , where μ _2,j is all the second features from the first analysis period to the current analysis period σ _2,j is the standard deviation of all second feature values from the first analysis period to the current analysis period, A ₃ ∈ Z, A ₄ ∈ Z, the second feature value and the set Comparing the threshold value of the second feature quantity, output the corresponding level judgment result: if r _2,j ≥ A ₃ ×μ _2,j -A ₄ ×σ _2,j , then output the judgment result 0 and store it in the second fault arc Judgment matrix out ₂ [j]=0; if r _2,j <A ₃ ×μ _2,j -A ₄ ×σ _2,j , output the judgment result 1 and store it in the second fault arc judgment matrix out ₂ [ j]=1, go to step 6) carry out weighted processing of the output determination results on the two feature quantity decision-making layers;

6) Use dynamic weight coefficients to weight the output judgment results of independent component analysis and S-transformation, and obtain the weighted result outtemp _j = C _1,j ×out ₁ [j]+C _2,j ×out ₂ [j], and then perform preliminary State judgment: if outtemp _j >n, where n is the weighted result threshold, then output the judgment result 1 and store it in the preliminary state judgment result matrix outt[j]; otherwise, output the judgment result 0 and store it in the preliminary state judgment result matrix outt[j], go to step 7) to distinguish the status of the photovoltaic system;

7) Set the judgment accuracy p, and judge the state of the photovoltaic system every p time period: the matrix outt of the preliminary state judgment result is from the j-3pth element to the j-pth element, and from the j-2pth element to the jth element The number of elements is 1, if the value of the counted number is greater than p, it is confirmed that the fault arc of the photovoltaic system occurs within the j-2pth to j-pth period, and the corresponding protection measures for the fault arc of the photovoltaic system are taken; Otherwise, it is considered that the photovoltaic system is in normal operation during the period j-2p to j-p, and returns to step 1) to analyze the current signal in the next analysis period.

The current sensors are not required to be of the same type, but their bandwidth should be greater than 100kHz, and they should be installed in different positions of the photovoltaic system to display the difference between the sampled current signals, considering the principle of accurately obtaining current independent main source signals while reducing hardware detection costs , the value range of the current sensor is 2 to 4; the sampling frequency f _s should be greater than twice the upper limit of the fault arc characteristic frequency band of the photovoltaic system, and the value range is 200 to 500kHz; consideration should be given to quickly and accurately obtain the fault arc characteristics of the photovoltaic system In principle, the range of sampling points N is 8000-12000.

The various parameters in the fast independent component analysis (based on the maximization of negentropy) are determined by quickly obtaining the significant characteristics of the photovoltaic system fault arc that occurs in the system process, and the non-linear function can be selected as g ₁ (u)=u ³ , g ₂ (u)=u ² , g ₃ (u)=arctan(q ₁ ×u), g ₄ (u)=u×e^(-q ² ₂ ×u ² /2), where, q ₁ and q ₂ is a constant, preferably g ₁ (u)=u ³ , the maximum iteration number ranges from 950 to 1050, and the iteration precision ranges from 0.00006 to 0.00015.

The number of independent main source signals obtained by the fast independent component analysis is the number of sampling current signals. Based on the principle of the strongest signal impact, an effective independent main source signal is selected for subsequent fast Fourier transform processing: calculate each independent main source signal The peak-to-peak difference of the main source signal within the analysis period, select the independent main source signal with the largest difference as the effective independent main source signal; based on reducing the spectrum leakage as much as possible and reducing the first characteristic quantity to detect the fault arc of the photovoltaic system in the case of coupling The negative impact of the negative effect, the value of the number of transformation points in the fast Fourier transform is selected as the value corresponding to the number of sampling points N.

Based on the principle of discovering the time-frequency difference of the photovoltaic system fault arc in the system process to the greatest extent, one of the multiple current signals is selected as the input of the S-transformation: the current signal corresponding to the current sensor with the highest sensitivity is preferably selected. When such current sensors are more than When there is only one, it is preferable to select the current signal corresponding to the current sensor closest to the location where the fault arc occurs in the photovoltaic system. The current signal corresponding to the current sensor with the least number of components; based on the same principle, the window width adjustment factor in the S-transform is preferably 1.

The absolute value processing is performed on the elements of the two-dimensional complex time-frequency matrix obtained after the S transformation, and the principle of constructing the second feature quantity based on the frequency dimension component of the time-frequency matrix is that there is a significant downward trend when the photovoltaic system fault arc occurs, and the relatively The small amplitude form shows the difference between the fault arc of the photovoltaic system and the previous system process. The characteristic frequency band of the fault arc of the photovoltaic system is selected as 40-100kHz and is not related to the value of the sampling frequency f _s .

The first feature quantity threshold A ₁ ×μ _1,j -A ₂ ×σ _1,j is related to the first feature quantity values of all previous analysis periods and follows the dynamic change of the first feature quantity r ₁ in real time, wherein the coefficient A ₁ and A ₂ are related to the output characteristic of the first characteristic quantity, depending on the fact that the corresponding photovoltaic system state can be correctly obtained by comparing the set first characteristic quantity threshold value with the first characteristic quantity value, A ₁ and A ₂ are preferably 1; Mean value estimation μ _1,j and standard deviation σ _1,j are corrected in real time according to the output judgment result of the first characteristic quantity: for the first characteristic quantity value r _1,1 obtained in the first analysis period, the correction value rtemp _{1, 1} ＝r _1,1 , mean value estimate μ _1,1 ＝r _1,1 , standard deviation σ _1,1 ＝0; for the first feature value r _1,j of the j-th analysis period, where, j∈N And j>1, if the first feature value in the current analysis period is greater than or equal to the threshold of the first feature value in the previous analysis period, set the correction value rtemp _1,j = r _1,j , the calculation formula of mean value estimation and standard deviation is

Among them, k is the sequence number of the analysis period in the accumulation process, k=1, 2...j, j∈N and j>1, if the first feature value in the current analysis period is less than the first feature value threshold of the previous analysis period, Let the correction amount rtemp _1,j ＝μ _1,j-1 –σ _1,j-1 , the calculation formula of mean value estimation and standard deviation is

The second feature quantity threshold value A ₃ ×μ _2,j -A ₄ ×σ _2,j is related to the second feature quantity values of all previous analysis periods and follows the dynamic change of the second feature quantity r ₂ in real time, wherein the coefficient A ₃ and A4 are related to the output characteristics of the second characteristic quantity, depending _on the fact that the corresponding photovoltaic system state can be correctly obtained by comparing the set second characteristic quantity threshold value with the second characteristic quantity value, A3 and _A4 are preferably ₁ ; Mean value estimation μ _2,j and standard deviation σ _2,j are corrected in real time according to the output judgment result of the second characteristic quantity: for the second characteristic quantity value r _2,1 obtained in the first analysis period, let the correction quantity rtemp _{2, 1} =r _2,1 , mean value estimate μ _2,1 =r _2,1 , standard deviation σ _2,1 =0; for the second feature value r _2,j of the j-th analysis period, where, j∈N And j>1, if the second feature value in the current analysis period is greater than or equal to the threshold value of the second feature value in the previous analysis period, set the correction value rtemp _2,j = r _2,j , the calculation formula of mean value estimation and standard deviation is

Among them, k is the serial number of the analysis period in the accumulation process, k=1, 2...j, j∈N and j>1, if the second characteristic quantity value in the current analysis period is less than the second characteristic quantity threshold value of the previous analysis period, Let the correction amount rtemp _2,j ＝μ _2,j-1 –σ _2,j-1 , the calculation formula of mean value estimation and standard deviation is

Based on the principle of quickly calculating each set threshold in the current analysis period, the calculation formula of the mean value estimation and standard deviation in the current analysis period is obtained by using the recursive relationship as follows:

Among them, μ _m,j , σ _m,j are the mean estimate and standard deviation in the current analysis period, respectively, μ _m,j-1 , σ _m,j-1 are the mean estimate and standard deviation in the previous analysis period , rtemp _m,j is the correction amount in the current analysis period, where m is the serial number of the characteristic quantity, and the value is 1 or 2, j∈N and j>1.

Dynamic weight coefficients are used to weight the output judgment results after comparing the two feature quantities with the set threshold, and the weight coefficients of the output judgment results of each feature quantity are determined according to the statistical characteristics of the feature quantity for the correctness of the state judgment in the historical analysis period, that is, the feature The more analysis periods the quantity makes correct state judgments for the historical analysis period, the greater the weight coefficient obtained by the feature quantity in the current analysis period. Specifically, construct the first feature quantity and the second feature quantity according to the following formulas: Weight coefficients C _1,j and C _2,j :

Among them, σ ² _out1 and σ ² _out2 are respectively the variance of the first arc fault judgment matrix and the second arc fault judgment matrix from the first element to the jth element, namely

Among them, out ₁ and out ₂ are the first arc fault judgment matrix and the second arc fault judgment matrix respectively, k is the counting number of matrix elements, k=1,2...j, j∈N and j>1, and are respectively the mean value estimation of the first arc fault judgment matrix and the second arc fault judgment matrix from the first element to the jth element; if the first arc fault judgment matrix and the second arc fault judgment matrix are from the first element to the jth element The j elements are all 0, that is, both feature quantities determine that all analysis periods are in the normal operating state, directly assign the value C _1,j = 0, C _2,j =0, and then weight the two feature quantities for subsequent PV system fault arc detection ; If the photovoltaic system is in the normal operation state during the j-2p to j-p period, the element interaction between the positions of the corresponding elements of the first fault arc judgment matrix and the second fault arc judgment matrix in this p period For processing.

Based on the principle of accurately identifying photovoltaic system fault arcs in the system process, the weighted result threshold n ranges from 0.45 to 0.55; based on the principles of reliability and quickness of photovoltaic system fault arc detection, avoid excessively small p values causing The misoperation phenomenon of the system process and the untimely action phenomenon of the fault arc of the photovoltaic system caused by the excessive p value, the value range of the judgment accuracy p is 2-5.

The present invention has following beneficial technical effect:

1) This method improves the ability of the photovoltaic system to identify the normal state of the current, and solves the faulty arc detection device on the DC side of the photovoltaic system that is generated by the photovoltaic system fault output current perspective detection algorithm in the face of transient processes such as system power changes and startup. Problem, by correctly determining the system process as a normal operating state, the uptime of the photovoltaic system is greatly extended, the power generation efficiency of the photovoltaic system is significantly improved, and the stability of the normal operation of the photovoltaic system is enhanced;

2) This method can accurately grasp the fundamental characteristics of the fault arc of the photovoltaic system that occurs in the system process, and solves the problem of the DC fault of the photovoltaic system generated by the fault arc of the photovoltaic system that occurs immediately after the system process using the detection algorithm from the perspective of the normal output current of the photovoltaic system. The fault arc detection device on the side of the photovoltaic system refuses to operate. By correctly judging the fault arc of the photovoltaic system under the coupling condition as a fault state, the effectiveness of detection of the photovoltaic system arc fault is guaranteed, and the photovoltaic fire accident caused by the fault arc of this type of photovoltaic system is eliminated in time. , loss of life and property and other hazards, expanding the scope of application of the current photovoltaic system fault arc detection method;

3) This method has a wide detection range of photovoltaic system arc faults, and is not affected by the change trend direction of photovoltaic system output current caused by photovoltaic system fault arcs under coupling conditions. Larger, smaller, or unchanged, the detection algorithm can reliably and accurately detect the fault arc of the photovoltaic system in the process of the system;

4) This method has strong quickness. The single analysis period for detecting the fault arc of the photovoltaic system is 40-60ms, and the setting accuracy is 2-5. That is, the longest time for detecting the fault arc of the photovoltaic system is 0.3s, and the shortest It takes 0.08s to send out the fault arc line cut-off control signal, and the judgment time of photovoltaic system fault arc in the case of reliable detection of coupling is much shorter than the 2s standard stipulated in the current American standard UL1699B;

5) This method uses the mean value estimation and standard deviation of the feature quantity to construct the feature quantity threshold value, uses the feature quantity value and the threshold value comparison process to realize the normalization of the output of each feature quantity, and solves the problem of multi-feature quantity detection caused by the magnitude difference of different feature quantity outputs. The interference of photovoltaic system fault arc is also conducive to the realization of subsequent multi-feature quantity decision-making layer weighting, the dynamic change processing of threshold and weight coefficient in different analysis cycles, and the establishment of photovoltaic system fault arc standards are conducive to the detection algorithm It is more reliable to give the correct judgment result of the system state in each analysis period;

6) This method does not have high requirements for the hardware change of the existing photovoltaic system arc fault detection. It only needs to reasonably lay the required current sensors in the original photovoltaic system, and add a fault arc detection device to the original photovoltaic system DC side. Detect the signal input port, and then the software program of the DC side fault arc detection device of the photovoltaic system only needs to calculate the characteristic quantities under the two methods, set the dynamic threshold value, and calculate the dynamic weighting coefficient, and finally realize the weighting of the two characteristic quantities on the decision-making layer Photovoltaic system fault arc detection, programming principle is simple, and the implementation cost is low.

Furthermore, when it is determined that a photovoltaic system fault arc occurs, the calculation of the mean value estimate and standard deviation in the threshold setting needs to be corrected to avoid large fluctuations in the threshold value due to large changes in characteristic quantities; when it is determined that the photovoltaic system is operating normally and the two When the characteristic quantity output judgment results are not equal, the corresponding unequal elements of the two fault arc judgment matrices are exchanged to realize the correct dynamic change of the weight coefficient, eliminate the problem of normal operation misoperation caused by external interference, and effectively improve the photovoltaic system. The reliability of system fault arc detection increases the economic benefits of photovoltaic system operation.

Description of drawings

Fig. 1a is a flow chart of the photovoltaic system arc fault detection method of the present invention;

Fig. 1b is a flow chart of dynamic threshold value setting in the photovoltaic system arc fault detection method of the present invention;

Fig. 2 is a schematic block diagram of the present invention in the implementation of specific photovoltaic system application hardware including a photovoltaic system DC side fault arc detection device integrated in the bus;

Fig. 3a is the output current signal of the photovoltaic system with a constant trend at the moment of the fault arc detected by the photovoltaic system under the coupling situation of the present invention;

Fig. 3b is the characteristic quantity and its set threshold waveform of the fault arc detection of the photovoltaic system in the case of applying independent component analysis for coupling;

Fig. 3c is the characteristic quantity and its set threshold waveform of fault arc detection in the photovoltaic system in the case of applying S-transform for coupling;

Fig. 3d is the system state judgment output signal of photovoltaic system arc fault detection under the coupling situation of applying the present invention;

Fig. 4a is the output current signal of the photovoltaic system with a decreasing trend at the moment of the fault arc detected by the photovoltaic system under the coupling condition of the present invention;

Figure 4b is the characteristic quantity and its set threshold waveform of the fault arc detection of the photovoltaic system in the case of coupling by independent component analysis;

Fig. 4c is the characteristic quantity and its set threshold waveform of the fault arc detection of the photovoltaic system in the case of applying S-transform for coupling;

Fig. 4d is the system state judgment output signal of photovoltaic system arc fault detection under the coupling situation of applying the present invention;

Fig. 5a is the output current signal of the photovoltaic system with an increasing trend at the moment of the fault arc detected by the photovoltaic system under the coupling condition of the present invention;

Figure 5b is the characteristic quantity and its set threshold waveform of the fault arc detection of the photovoltaic system in the case of applying independent component analysis for coupling;

Fig. 5c is the characteristic quantity and its set threshold waveform of the fault arc detection of the photovoltaic system in the case of applying S-transformation for coupling;

Fig. 5d is the system state judgment output signal of photovoltaic system arc fault detection in the case of applying the present invention for coupling;

In the figure: 1. Photovoltaic system; 2. DC side fault arc detection device of photovoltaic system; 3. Tripping device; 4. Circuit breaker; 5. Load; 6. Current sensor; 7. Photovoltaic system arc fault; 8. Photovoltaic module .

Detailed ways

The method of the present invention will be described in detail below in conjunction with the accompanying drawings and embodiments.

With reference to Fig. 1a, the steps of the arc fault method of photovoltaic system under the application of independent component analysis and S-transform detection system process coupling according to the present invention are described in detail.

Step 1. The parameter initialization process includes setting the sampling frequency f _s of the current sensor for the current signal, the number of sampling points N in the analysis period, the judgment accuracy p, the weighted result threshold n, clearing the variables for the mean value estimation and standard deviation, independent Various parameters in the component analysis and S-transformation arc fault characteristic analysis tools, etc. The current sensor performs parallel sampling on the multi-channel current signals required by the fault arc detection device on the DC side of the photovoltaic system according to the predetermined sampling frequency f _s , and obtains the multi-channel current signals x _i (the serial number of the current sensor is i∈N and i>1) , once the number of sampling points of these current signals reaches N, they are input to the fault arc detection device on the DC side of the photovoltaic system through multiple ports, and go to step 2 to extract the multi-faceted characteristics of the photovoltaic system arc fault.

The current sensors used in the present invention are not required to be of the same type, as long as the bandwidth parameter of the selected current sensor is greater than 100kHz to ensure that it can obtain the characteristic frequency band of the fault arc of the photovoltaic system. Multiple current sensors should be installed in different locations of the photovoltaic system to reflect the influence of the fault arc of the photovoltaic system on the output current signal of the photovoltaic system at different sampling points. After the optimized arrangement of current sensors in the photovoltaic system, while accurately obtaining the current independent main source signal, the number of current sensors used can be reduced as much as possible, thereby reducing the hardware detection cost of the entire photovoltaic system arc fault. Four current sensors are used in this embodiment.

During the working process of the fault arc detection device on the DC side of the photovoltaic system, the output current signal of the photovoltaic system is sampled point by point with the sampling frequency f _s . If the sampling frequency is too high, the hardware cost of the current sensor will be increased, and if the sampling frequency is too low, it cannot cover the current signal. Reflected photovoltaic system fault arc characteristic frequency. Therefore, under the selection that the upper limit of the fault arc characteristic frequency band of the photovoltaic system concerned by the present invention is 100kHz, in order to reduce the hardware implementation requirements of the current sensor, the sampling frequency f _s =200kHz determined in this embodiment. Note that the current signal collected by the i-th sensor in the j-th analysis period is x _{i, j} (the representation sequence number of the current sensor is i∈N and i>1; the representation sequence number of the analysis period is j∈N ₊ ), for any two different In terms of the value of i, when j takes the same value, x _{i and j} both have the same number of sampling points N, that is, the photovoltaic system arc fault detection algorithm proposed by the present invention analyzes multiple current signals at equal time intervals. If the number of sampling points N is too large, it will increase the analysis and running time of the photovoltaic system arc fault detection algorithm, which is not conducive to the rapid detection of photovoltaic system arc faults. If the value of sampling points N is too small, it is not enough to ensure accurate detection of photovoltaic system arc faults during the system process detection effect. Therefore, this embodiment considers the principle of quickly and accurately obtaining the fault arc characteristics of the photovoltaic system, and the determined number of sampling points is N=10000.

Step 2: Form a high-dimensional mixed-signal matrix X=[x _1,j ,x _2,j ,… _xi,j ] ^T through the collected multi-channel current signals, and perform de-meaning processing on the obtained mixed-signal matrix, that is, x ' _1,j ＝x _1,j –E(x _1,j ), where E(x _1,j ) represents the mean estimation of x _1,j ; then whiten the obtained zero-mean signal, let E=( e ₁ , e ₂ ,...e _n ) is a matrix whose columns are the unit norm eigenvectors of the covariance matrix C=E(x _1,j x ^T _1,j ), let D=diag(e ₁ , e ₂ ,...e _n ) is a diagonal matrix whose eigenvalues of the covariance matrix C are diagonal elements. After linear transformation, V=D ^-1/2 E ^T , and the whitened signal after transformation is z=Vx' _1,j . The unmixing matrix W can be obtained after fast independent component analysis based on the maximum negative entropy, and the source signal matrix S=WX=[s _1,j ,s _2,j ,…s _i,j ] ^T is calculated, where S includes For the effective source signal and the noise signal, select the effective independent main source signal s _1,j , perform fast Fourier transform on it, calculate the variance of the one-dimensional frequency matrix in the frequency domain, and obtain the first characteristic value r _1,j . Select one of the signals x _i,j among the multiple current signals, and perform S-transformation on the signal to obtain a two-dimensional complex matrix distribution in the time-frequency domain, calculate the integral of the absolute value of the high-frequency component in the frequency dimension along time, and obtain the second characteristic For the value r _2,j , go to step 3 and compare it with the corresponding set threshold to obtain the judgment result of each feature quantity in the current analysis period.

The fast independent component analysis determined in this embodiment is the fast independent component analysis based on negentropy maximization: a suitable unmixing matrix W is found through the negentropy maximization algorithm, and finally the independent main source signals of the current are obtained. In order to find a suitable unmixing matrix as soon as possible to speed up the fault arc detection process of the photovoltaic system, the non-linear function selected in this embodiment is u ³ , and the iteration accuracy value of the end iteration process is determined to be 0.0001, and the maximum number of iterations is 1000. The method of fast independent component analysis is used to analyze multi-channel current signals, and the number of independent main source signals is the number of analyzed current signal channels. Calculate the peak-to-peak difference of these independent main source signals within the analysis period, and select the largest The independent main source signal corresponding to the peak-to-peak difference is an effective independent main source signal. Perform fast Fourier transform processing on this effective independent main source signal. Too many transformation points will cause distortion of the original current spectrum analysis, and too few transformation points will cause serious spectrum leakage. These factors are not conducive to The first characteristic quantity accurately detects the fault arc of the photovoltaic system. Therefore, the number of transform points in the fast Fourier transform in this embodiment is selected as 10000. After a photovoltaic system fault arc occurs in the system process, the spectrum energy migration makes the distribution of the spectrum matrix more uniform in the current analysis period, so its variance peaks at the time of the photovoltaic system fault arc occurrence, and the overall fault arc state has a smaller value than the normal operating state. Therefore, it has the ability to accurately discover the photovoltaic system fault arc hidden in the system process, and is selected as the first characteristic quantity.

Different from independent component analysis, the input of S-transform has only one current signal. Therefore, it is most appropriate to compare and select the current signal that most effectively reflects the fault arc difference of the photovoltaic system among the multiple current signals as the input signal. Comparing the distance between the installation position of the sensor and the fault arc of the photovoltaic system, the sensitivity of the current sensor is used as a selection index with higher priority, that is, when there is more than one current sensor with a certain type of high sensitivity, the current sensor closest to the fault arc occurrence position of the photovoltaic system is preferred The corresponding current signal is used as the input of S transform. The matching optimization of each parameter in the S-transform is also based on the characteristics of the fault arc of the photovoltaic system in the process of separating the system to the greatest extent, so as to identify the fault arc of the photovoltaic system more reliably. After S-transformation, the current signal becomes a two-dimensional complex matrix in the time-frequency domain, and its real part, imaginary part or phase angle are not as effective as absolute value processing in indicating fault arcs in photovoltaic systems. The absolute value has a good consistency in the frequency dimension of 40 ~ 100kHz, and the overall amplitude has a significant drop after the fault arc of the photovoltaic system occurs, and the influence of the system process on this frequency band is often weak, so it occurs in the system The photovoltaic system fault arc in the process has a good separation effect in this frequency band, and is selected as the second characteristic quantity. In order to improve the reliability of photovoltaic system arc fault detection, the characteristic frequency band of 40-100kHz photovoltaic system arc fault under this time-frequency conversion tool is integrated along the time axis for superposition processing. The characteristic frequency band has nothing to do with the value of the sampling frequency f _s . Therefore, when using the technical solution of the present invention, the sampling frequency must not be lower than 200kHz.

Step 3: After the current signal is analyzed and processed by the above two methods, two feature values are obtained on the feature layer in each analysis period, and by comparing the first feature value, the second feature value with the corresponding threshold, the The output results of each feature quantity are normalized to the decision-making layer. The threshold value is dynamically changed in different analysis periods. When it is determined that an arc fault in the photovoltaic system occurs within the analysis period, the threshold value is set after the calculation and correction of the mean value estimate and standard deviation, and the first arc fault judgment matrix out ₁ , For the second fault arc determination matrix out ₂ , go to step 4 for weighting processing.

In order to overcome the order of magnitude differences generated by each feature quantity, each feature quantity matches its own unique set threshold. Here, the comparison between the first feature value and its set threshold is taken as an example to illustrate this step. Compare the first feature value with the set first feature value threshold, and output the corresponding level judgment result: if the first feature value is greater than the set threshold, then output the judgment result 0 and store it in the first fault arc Judgment matrix out ₁ [j]; if the first eigenvalue is smaller than the set threshold, the judgment result 1 is output and stored in the first arc fault judgment matrix out ₁ [j]. Therefore, the threshold comparison process is equivalent to normalizing each feature quantity, so that the weighting process does not produce significant amplitude fluctuations. The second arc fault judgment matrix out ₂ can be obtained using a similar method for the second feature quantity and the second feature quantity threshold.

In order to adapt to the feature quantity fluctuations generated in the normal analysis period, the set threshold is related to the mean value estimate and standard deviation of the corresponding feature quantity, that is, the first feature quantity threshold in the current analysis period is set to A ₁ ×μ _1,j –A ₂ ×σ _1,j , where, μ _1,j is the mean value estimation of all the first feature values from the first analysis period to the current analysis period, σ _1,j is all the first feature values from the first analysis period to the current analysis period The standard deviation of the characteristic value, A ₁ ∈ Z, A ₂ ∈ Z. When it is determined that a photovoltaic system fault arc occurs during the current analysis period, it is necessary to correct the calculation of the mean value estimate and standard deviation to avoid large fluctuations in the threshold value due to large changes in characteristic quantities, and to eliminate normal operation errors caused by external system process interference. It effectively improves the reliability of photovoltaic system fault arc detection under coupling conditions, and increases the economic benefits of photovoltaic system operation.

With reference to FIG. 1 b , taking the first feature value threshold setting process as an example, the dynamic threshold setting process in the photovoltaic system arc fault detection method is specifically described.

The first feature quantity threshold A ₁ ×μ _1,j -A ₂ ×σ _1,j is related to the first feature quantity values of all previous analysis periods and follows the dynamic change of the first feature quantity r ₁ in real time, wherein the coefficient A ₁ and A ₂ are related to the output characteristics of the first characteristic quantity, depending on whether the corresponding photovoltaic system state can be correctly obtained by comparing the set first characteristic quantity threshold value with the first characteristic quantity value. This embodiment is based on independent component analysis. The characteristic of the first feature quantity, the determination coefficients A ₁ =1, A ₂ =1, that is, the threshold value of the first feature quantity is μ _1,j –σ _1,j ; the mean value estimation μ _1,j and standard deviation σ _1,j are based on The output judgment result of the first characteristic quantity is corrected in real time: for the first characteristic quantity r _1,1 obtained in the first analysis period, set the correction quantity rtemp _1,1 =r _1,1 , and the mean value estimate μ _1,1 =r _1,1 , standard deviation σ _1,1 ＝0, the corresponding output threshold is μ _1,1 –σ _1,1 ; for the first feature quantity r _1,j of the jth analysis period, where, j∈ N and j>1, if the first feature value in the current analysis period is greater than or equal to the first feature value threshold of the previous analysis period, that is, when it is initially determined that the current analysis period is normal, the correction value rtemp _1,j = r _{1 ,j} , the calculation formula of the mean estimate and standard deviation is

Among them, k is the serial number of the analysis period in the accumulation process, k=1, 2...j, j∈N and j>1, if the feature quantity value in the current analysis period is less than the first feature quantity threshold value of the previous analysis period, that is, preliminary When it is determined that the current analysis period is in a fault state, another set of threshold setting schemes is required to ensure the correct judgment of the fault arc detection algorithm of the photovoltaic system, and the correction value rtemp _1,j = μ _1,j-1 –σ _{1,j- 1.} The calculation formulas for mean estimation and standard deviation are:

Based on the principle of fast calculation of each set threshold in the current analysis period, using the existing mean value estimation and standard deviation initial value μ _1,1 , σ _1,1 and the correction value rtemp _1,j are calculated according to the following recursive relationship The mean value estimation and standard deviation of each analysis period starting from the second analysis period, and then the set threshold value of the corresponding output is μ _1,j –σ _1,j :

Among them, μ _1,j , σ _1,j are the mean estimate and standard deviation in the current analysis period, respectively, μ _1,j-1 , σ _1,j-1 are the mean estimate and standard deviation in the previous analysis period , rtemp _{1, j} is the correction amount in the current analysis period, j∈N and j>1.

Step 4: Match the corresponding weight coefficients to the output judgment results of independent component analysis and S-transformation, weight the fault arc judgment matrix of the two characteristic quantities in the current analysis period, and obtain the weighted result outtemp _j =C _1,j ×out ₁ [j]+C _2,j ×out ₂ [j], and then make a preliminary state judgment: if outtemp _j >n, where n is the weighted result threshold, then output the judgment result 1 and store it in the preliminary state judgment result matrix outt [j]; otherwise, output the judgment result 0, store it in the preliminary state judgment result matrix outt[j], and go to step 5 to judge whether to issue the fault arc cutting control signal.

Based on the principle of accurately identifying the photovoltaic system fault arc in the system process, the weighted result threshold n determined in this embodiment is 0.5. Dynamic weight coefficients are used to weight the output judgment results after the comparison between the two feature values and the threshold value, and the weight coefficients of the output judgment results of each feature value are determined according to the statistical characteristics of the feature value to the correctness of the state judgment in the historical analysis period, that is, the feature value is The more analysis periods in which the correct state judgment is made in the historical analysis period, the greater the weight coefficient obtained by the feature quantity in the current analysis period. Specifically, the weights of the first feature quantity and the second feature quantity are respectively constructed based on the following formulas Coefficients C _1,j and C _2,j :

Among them, σ ² _out1 and σ ² _out2 are respectively the variance of the first arc fault judgment matrix and the second arc fault judgment matrix from the first element to the jth element, namely

Among them, out ₁ and out ₂ are the first arc fault judgment matrix and the second arc fault judgment matrix respectively, k is the counting number of matrix elements, k=1,2...j, j∈N and j>1, and are the mean value estimates from the first element to the jth element of the first arc fault judgment matrix and the second arc fault judgment matrix, respectively.

If the first arc fault judgment matrix and the second arc fault judgment matrix are all 0 from the first element to the jth element, that is, both feature quantities determine that all analysis periods are in the normal operating state. In this case, the final determination of the photovoltaic system The operating status must be normal. In order to avoid complex calculations of weighting coefficients, directly assign values C _1,j =0, C _2,j =0, and then weight the two characteristic quantities for subsequent photovoltaic system fault arc detection.

Step 5: Set the judgment accuracy p, and make a judgment on the fault arc discrimination result of the photovoltaic system every p time periods. The corresponding judgment principle is: to count the number of 1s from the j–3pth element to the j–pth element, and from the j–2pth element to the jth element of the preliminary state judgment result matrix outt. If the value of the number is greater than p, it is confirmed that the fault arc of the photovoltaic system occurs during the period j–2p to j–p, and the corresponding protection measures for the arc fault of the photovoltaic system are taken; otherwise, the photovoltaic system is considered to be in normal operation during this period state, return to the step to sample and analyze the current signal in the next analysis period. If the photovoltaic system is in normal operation during the j-2p to j-p period, then the photovoltaic system is considered to be in normal operation during this period. For this p period, if the first arc fault judgment matrix, the first The output of the corresponding elements of the two fault arc judgment matrix is a combination of 0/1 or 1/0, that is, when the values of the corresponding elements are not equal, the two elements at the corresponding positions are exchanged to ensure the validity of the weighting coefficient in any case .

If the setting judgment accuracy p is set too small, it may lead to misjudgment when the photovoltaic system has a system process, and the corresponding fault arc detection device on the DC side of the photovoltaic system will malfunction, resulting in unnecessary power loss of the photovoltaic system; if p is set If it is too large, the fault line cannot be removed in time after the fault arc of the photovoltaic system occurs, and the corresponding fault arc detection device on the DC side of the photovoltaic system will refuse to operate, causing major economic losses and serious security threats. Based on the principles of the reliability and quickness of fault arc detection in photovoltaic systems, the system process misoperation caused by too small p value and the untimely action phenomenon of photovoltaic system fault arc caused by too large p value are avoided. The judgment accuracy determined in this embodiment p is 5.

The method of the present invention applied to an actual photovoltaic system is described, as shown in FIG. 2 , illustrating the operation process of the method of the present invention in an actual photovoltaic system. The photovoltaic system 1 composed of photovoltaic modules 8 outputs DC power, which is input to the load 5 through the current sensor 6 and the circuit breaker 4 .

The multi-channel photovoltaic system output current signal is input to the photovoltaic system DC side arc fault detection device 2 through multiple current sensors 6 to perform the above photovoltaic system arc fault identification process. Due to the randomness of the location of the fault arc in the photovoltaic system, a plurality of current sensors 6 need to be reasonably arranged in the photovoltaic system to minimize the blind area of the photovoltaic system fault arc missed detection. Here, this embodiment describes the method of selecting one output current signal from multiple photovoltaic systems as the S-transform input. Assuming that 6A～6D are selected as the same type of current sensors, select a suitable output current signal of the photovoltaic system for the fault arc 7A of the photovoltaic system. The system outputs the current signal as the input of S transformation. For the fault arc 7B of the photovoltaic system, select an appropriate output current signal of the photovoltaic system. Assuming that there are enough photovoltaic modules 8 on the photovoltaic string where the current sensor 6C is located, the distance between the fault arc 7B of the photovoltaic system and the current sensor 6C is much greater than that of the current sensor 6C. The distance of the current sensor 6B, according to the principle of the sensor type first and then the distance proposed by the present invention, should select the photovoltaic system output current signal collected by the current sensor 6B as the input of the S transformation; assuming that the photovoltaic module on the photovoltaic string where the current sensor 6C is located 8 is enough, so that the distance between the fault arc 7B of the photovoltaic system and the current sensor 6B is far greater than the distance between it and the current sensor 6C. The output current signal of the photovoltaic system is used as the input of the S transformation; assuming that the photovoltaic module 8 on the photovoltaic string where the current sensor 6C is located just makes the distance between the fault arc 7B of the photovoltaic system and the current sensor 6B equal to the distance between it and the current sensor 6C, according to this According to the principle of distance first and then wiring complexity proposed by the invention, the output current signal of the photovoltaic system collected by the current sensor 6B should be selected as the input of the S-transformation.

In actual testing, in addition to experiencing the transient process of starting the photovoltaic system every morning in the photovoltaic system, there will also be a transient process of power regulation due to normal system operation, which makes the photovoltaic system power experience frequent transient changes . The occurrence time of the photovoltaic system arc fault is uncontrollable, so there is a certain probability that the photovoltaic system arc fault will occur in these system processes, so it will eventually lead to the photovoltaic system arc fault under the coupling effect. For example, during the start-up of the photovoltaic system and the increase of system power, the output current of the photovoltaic system continues to increase. On the other hand, the fault arc of the series photovoltaic system will reduce the output current of the photovoltaic system. Therefore, in the fault arc of the photovoltaic system in the case of coupling, the fault output current of the photovoltaic system will appear immediately after the normal output current of the photovoltaic system, and the requirements for the fault arc detection algorithm of the photovoltaic system are more stringent.

The corresponding photovoltaic system arc fault detection algorithm must grasp the fundamental characteristics of the photovoltaic system fault arc that is different from the system process, and accurately, reliably and quickly identify the moment when the photovoltaic system fault arc occurs, so as to complete the detection of the photovoltaic system DC side fault arc detection Functional requirements of device 2. The specific requirements for correct and reliable detection of photovoltaic system arc faults in the process of the system are: when the photovoltaic system is operating normally, the low level output by the photovoltaic system DC side arc fault detection device 2 does not operate the circuit breaker 4, and the photovoltaic system 1 remains Stably output electric energy to the load 5; if the photovoltaic system DC side fault arc detection device 2 detects the photovoltaic system fault arc 7 that occurs in the system process, it can quickly and accurately send a control signal to cut off the corresponding branch circuit to the tripping device 3 , and finally control the circuit breaker 4 to cut off the entire photovoltaic system circuit, the load stops working, extinguishes the fault arc of the photovoltaic system and eliminates the threat to the operation safety of the photovoltaic system, and avoids the fault arc detection of the DC side of the photovoltaic system caused by the fault arc of the photovoltaic system The problem of device refusal to operate can avoid the problem of faulty arc detection device on the DC side of the photovoltaic system caused by the normal operation of the photovoltaic system, thereby expanding the application range of the photovoltaic system detection algorithm and solving the problem of photovoltaic system fault arc in the process of facing the system Potential failures may occur that threaten the stable and safe operation of the photovoltaic system.

3a-3d, the photovoltaic system arc fault detection method of the present invention is applied to the photovoltaic system arc fault identification effect in the coupling situation with the characteristic that the fault arc occurrence time does not change.

The output current signals of multiple photovoltaic systems are acquired at a sampling frequency f _s =200 kHz, as shown in FIG. 3 a , and the input waveform is described by taking one of the photovoltaic system output current signals as an example. Before 3.53s, the current signal was in the normal state, and the photovoltaic system supplied power to the load through the closed circuit at this time; after 3.53s, the current signal was in the fault state, but at this time the fault current waveform did not change dynamically due to the occurrence of the photovoltaic system series fault arc Instead, it maintains the form of normal current at the moment of fault arc occurrence, and maintains a constant current change trend.

After the multi-channel current signals are averaged and whitened, the multi-channel current signals are analyzed by independent component analysis, an effective independent main source signal is selected, and the variance of the one-dimensional frequency matrix after the fast Fourier transform of the signal is calculated. The obtained first feature quantity is shown as the solid line in Fig. 3b. In order to better observe the final judgment result of the first feature quantity, the corresponding first feature quantity threshold value μ _1,j −σ _1,j is also shown in the form of a dotted line in Fig. 3b. It can be seen from the figure that the first characteristic quantity indicates the analysis period of the fault arc of the photovoltaic system in the form of a large pulse, and the overall low amplitude level presents the difference between the fault arc of the photovoltaic system and the previous normal operation, showing the first characteristic quantity Availability of arc fault detection for such photovoltaic systems. The first feature value is compared with the constructed threshold, and the corresponding level judgment result is output, which is stored in the first fault arc judgment matrix out ₁ . Analyze the current signal through the S-transform method to obtain the two-dimensional complex matrix distribution in the time-frequency domain. After performing absolute value processing on each element of the two-dimensional matrix, calculate the integral of the 40-100kHz component of the frequency dimension along time, and obtain The second feature quantity is shown by the solid line in Fig. 3c. In order to better observe the final determination result of the second feature quantity, the corresponding second feature quantity threshold value μ _2,j −σ _2,j is also shown in the form of a dotted line in Fig. 3c. It can be seen from the figure that the second characteristic quantity presents greater fluctuations in each analysis period than the first characteristic quantity, but it still presents the difference between the fault arc of the photovoltaic system and the previous normal operation at a lower amplitude level as a whole. It also shows the effectiveness of the second feature quantity for arc fault detection of this type of photovoltaic system. The second feature value is compared with the constructed threshold, and the corresponding level judgment result is output, which is stored in the second fault arc judgment matrix out ₂ .

After the dynamic threshold comparison of the two feature values, the output judgment results of the independent component analysis and S-transformation are obtained, and the weight coefficients are determined according to the statistical results of the correctness of the system state in the first j-1 analysis periods of each feature value. Then the outtemp _j is obtained after weighting with dynamic weight coefficients on the decision-making layer. By comparing the corresponding thresholds, weighting the two feature quantities to obtain the judgment results in each analysis period, and obtaining the preliminary state judgment result matrix outt. Count the number of preliminary state judgment result matrix outt from the j–3pth element to the j–pth element, from the j–2pth element to the jth element with 1, if the value of the counted number is greater than p , then it is confirmed that the photovoltaic system arc fault occurs in the j–pth to j–2pth period, and the final judgment result is output as 1, and the corresponding photovoltaic system arc fault protection measures are taken; otherwise, the photovoltaic system is considered to be in normal operation, and the output The final judgment result is 0. As shown in the results shown in Figure 3d, the detection algorithm can give a correct low-level indication for the normal operation of the photovoltaic system, and can give a correct high-level indication for the fault arc of the photovoltaic system without any change, so the detection algorithm The fault arc of the photovoltaic system that occurs in the process of the system can be quickly detected.

4a to 4d, the photovoltaic system arc fault detection method of the present invention is applied to the photovoltaic system arc fault identification effect in the coupling situation with the characteristic that the fault arc occurrence time becomes smaller.

The output current signals of multiple photovoltaic systems are acquired at a sampling frequency of f _s =200kHz, as shown in FIG. 4a , and the input waveform is described by taking one of the photovoltaic system output current signals as an example. Before 5.86s, the current signal is in a normal state, at this time, the photovoltaic system supplies power to the load through the closed circuit; after 5.86s, the current signal is in a fault state, at this time, a dynamically reduced fault arc current is generated due to a series fault arc on the photovoltaic system bus The waveform has a decreasing current variation trend at the time of the fault arc occurrence, but this fault arc current, which is lower than the normal current, cannot be maintained. The immediately increased power of the photovoltaic system makes the fault current waveform rise instantaneously, and then it is compared with the normal current level. A consistent fault current is maintained.

After the multi-channel current signals are averaged and whitened, the multi-channel current signals are analyzed by independent component analysis, an effective independent main source signal is selected, and the variance of the one-dimensional frequency matrix after the fast Fourier transform of the signal is calculated. The obtained first feature quantity is shown as the solid line in Fig. 4b. In order to better observe the final judgment result of the first feature quantity, the corresponding first feature quantity threshold value μ _1,j −σ _1,j is also shown in the form of a dotted line in FIG. 4b. It can be seen from the figure that the first characteristic quantity indicates the analysis period of the photovoltaic system fault arc occurrence and subsequent short-term changes in the form of large pulses, and the overall low amplitude level presents the difference between the stable photovoltaic system fault arc and the previous normal operation. The validity of the first feature quantity for arc fault detection of this type of photovoltaic system is tested. The first feature value is compared with the constructed threshold, and the corresponding level judgment result is output, which is stored in the first fault arc judgment matrix out ₁ . Analyze the current signal through the S-transform method to obtain the two-dimensional complex matrix distribution in the time-frequency domain. After performing absolute value processing on each element of the two-dimensional matrix, calculate the integral of the 40-100kHz component of the frequency dimension along time, and obtain The second feature quantity is shown by the solid line in Fig. 4c. In order to better observe the final determination result of the second feature quantity, the corresponding second feature quantity threshold value μ _2,j −σ _2,j is also shown in the form of a dotted line in Fig. 4c. It can be seen from the figure that the second characteristic quantity presents greater fluctuations in each analysis period than the first characteristic quantity, but it still presents the difference between the fault arc of the photovoltaic system and the previous normal operation at a lower amplitude level as a whole. It also shows the effectiveness of the second feature quantity for arc fault detection of this type of photovoltaic system. The second feature value is compared with the constructed threshold, and the corresponding level judgment result is output, which is stored in the second fault arc judgment matrix out ₂ .

After the dynamic threshold comparison of the two feature values, the output judgment results of the independent component analysis and S-transformation are obtained, and the weight coefficients are determined according to the statistical results of the correctness of the system state in the first j-1 analysis periods of each feature value. Then the outtemp _j is obtained after weighting with dynamic weight coefficients on the decision-making layer. By comparing the corresponding thresholds, weighting the two feature quantities to obtain the judgment results in each analysis period, and obtaining the preliminary state judgment result matrix outt. Count the number of preliminary state judgment result matrix outt from the j–3pth element to the j–pth element, from the j–2pth element to the jth element with 1, if the value of the counted number is greater than p , then it is confirmed that the photovoltaic system arc fault occurs in the j–pth to j–2pth period, and the final judgment result is output as 1, and the corresponding photovoltaic system arc fault protection measures are taken; otherwise, the photovoltaic system is considered to be in normal operation, and the output The final judgment result is 0. As shown in the results shown in Figure 4d, the detection algorithm can give the correct low-level indication for the normal operation of the photovoltaic system, and can give a correct low-level indication for the short-term decrease and increase of the fault arc transient and the subsequent fault arc of the photovoltaic system without any change. The correct high-level indication is given, so the detection algorithm can quickly detect the fault arc of the photovoltaic system that occurs in the system process.

5a-5d, the photovoltaic system arc fault detection method of the present invention is applied to the photovoltaic system arc fault identification effect in the coupling situation with the characteristic that the fault arc becomes larger at the moment of occurrence.

The output current signals of multiple photovoltaic systems are obtained at a sampling frequency of f _s =200kHz, as shown in FIG. 5 a , and the input waveform is described by taking one of the photovoltaic system output current signals as an example. Before 1.21s, the current signal was in a normal state, at this time the photovoltaic system supplies power to the load through the closed circuit; after 1.21s, the current signal was in a fault state, but at this time the fault arc occurred in the process of the system, and the power adjustment of the photovoltaic system failed The extent of the arc current is much greater than that of the photovoltaic system bus series fault arc, which has an increasing current change trend at the time of the fault arc occurrence, and then maintains a higher level of fault arc current.

After the multi-channel current signals are averaged and whitened, the multi-channel current signals are analyzed by independent component analysis, an effective independent main source signal is selected, and the variance of the one-dimensional frequency matrix after the fast Fourier transform of the signal is calculated. The obtained first feature quantity is shown by the solid line in Fig. 5b. In order to better observe the final determination result of the first feature quantity, the corresponding first feature quantity threshold value μ _1,j −σ _1,j is also shown in the form of a dotted line in FIG. 5b. It can be seen from the figure that the first characteristic quantity as a whole presents the difference between the fault arc state of all photovoltaic systems and the previous normal operation at a lower amplitude level. Although the increasing trend of the large pulse indication will affect the correct identification of the fault arc state, but The short continuous analysis period does not affect the overall identification of arc faults, which shows the effectiveness of the first feature quantity for arc fault detection in photovoltaic systems. The first feature value is compared with the constructed threshold, and the corresponding level judgment result is output, which is stored in the first fault arc judgment matrix out ₁ . Analyze the current signal through the S-transform method to obtain the two-dimensional complex matrix distribution in the time-frequency domain. After performing absolute value processing on each element of the two-dimensional matrix, calculate the integral of the 40-100kHz component of the frequency dimension along time, and obtain The second feature quantity is shown by the solid line in Fig. 5c. In order to better observe the final determination result of the second feature quantity, the corresponding second feature quantity threshold value μ _2,j −σ _2,j is also shown in the form of a dotted line in Fig. 5c. It can be seen from the figure that the second characteristic quantity presents greater fluctuations in each analysis period than the first characteristic quantity, but it still presents the difference between the arc fault state of all photovoltaic systems and the previous normal operation at a lower amplitude level as a whole. The feature makes it possible to correct the misjudgment of the first feature quantity in some analysis periods, emphasizing the necessity of using the weighting of the two feature quantities in the present invention. The second feature value is compared with the constructed threshold, and the corresponding level judgment result is output, which is stored in the second fault arc judgment matrix out ₂ .

After the dynamic threshold comparison of the two feature quantities, the determination results of independent component analysis and S-transformation are obtained. The weights are determined according to the statistical results of the correctness of the system state in the first j-1 analysis period of each feature quantity, and then in the decision-making process. Outtemp _j is obtained after weighting with dynamic weight coefficients on the layer. By comparing the corresponding thresholds, weighting the two feature quantities to obtain the judgment results in each analysis period, and obtaining the preliminary state judgment result matrix outt. Count the number of preliminary state judgment result matrix outt from the j–3pth element to the j–pth element, from the j–2pth element to the jth element with 1, if the value of the counted number is greater than p , then it is confirmed that the photovoltaic system arc fault occurs in the j–pth to j–2pth period, and the final judgment result is output as 1, and the corresponding photovoltaic system arc fault protection measures are taken; otherwise, the photovoltaic system is considered to be in normal operation, and the output The final judgment result is 0. As shown in Figure 5d, the detection algorithm can give correct low-level indications for the normal operation of the photovoltaic system, and for the long-term fault arc transients that increase, decrease and then increase, and the subsequent higher-amplitude fault arc steady state The correct high level indication can be maintained, so the detection algorithm can quickly detect the fault arc of the photovoltaic system that occurs in the system process.

As shown in Figures 1a to 1b, the photovoltaic fault arc detection method provided by the present invention uses the mean value estimation and standard deviation of the characteristic quantity to construct the threshold value of the characteristic quantity, and the threshold value is dynamically changed in different analysis periods. When a fault arc occurs in the system, the calculation of the mean value estimate and standard deviation needs to be corrected. The normalization of the output of each feature quantity is realized by using the comparison process of feature value and threshold value, which solves the interference of different feature quantity output orders of magnitude difference on the weighted multi-feature quantity detection arc fault, and is conducive to the realization of multi-feature weighting on the decision-making layer. The weight value is determined by the statistical law of correct identification of the historical system state by each characteristic quantity. When the photovoltaic system is determined to be in normal operation and the output judgment results of the two characteristic quantities are not equal, the corresponding unequal elements of the two fault arc judgment matrices are exchanged. It is beneficial to more reliably give the correct judgment result of the system state in each analysis period, effectively improves the reliability of the fault arc detection of the photovoltaic system, and increases the economic benefits of the photovoltaic system operation.

As shown in Figures 3a to 5d, the photovoltaic fault arc detection method provided by the present invention grasps the statistical laws and core characteristics of the photovoltaic system fault arc by weighting the weight coefficients on the two feature quantity decision-making layers, and improves the photovoltaic fault arc detection method. The ability of the system to identify the normal state of the current solves the problem of misoperation of the fault arc detection device on the DC side of the photovoltaic system caused by the fault output current perspective detection algorithm of the photovoltaic system in the face of transient processes such as system power adjustment and startup. It is judged as a normal operation state, which greatly prolongs the normal operation time of the photovoltaic system, significantly improves the power generation efficiency of the photovoltaic system, and enhances the stability of the normal operation of the photovoltaic system. The present invention can also accurately grasp the fundamental characteristics of photovoltaic system fault arcs that occur in the system process, and accurately identify photovoltaic system fault arcs that occur in the system process. Influenced by the change trend direction of the photovoltaic system, it expands the scope of application of the current photovoltaic system arc fault detection method, and solves the DC side fault arc of the photovoltaic system that is generated by the photovoltaic system fault arc that occurs immediately following the system process with the detection algorithm of the normal output current perspective of the photovoltaic system. The detection device refuses to move. By correctly judging the fault arc of the photovoltaic system under the coupling condition as a fault state, the effectiveness of detection of the fault arc of the photovoltaic system is ensured, and the photovoltaic fire accidents caused by the fault arc of the photovoltaic system, life and property are eliminated in time. hazards such as loss.

Claims (10)

1. A method for applying independent component analysis and S-transformation to detect photovoltaic system arc faults, characterized in that: the method for photovoltaic system arc faults under the detection system process coupling situation comprises the following steps: 1) Through multiple current sensors, the output current signal of the photovoltaic system is sampled point by point according to the sampling frequency f _s to obtain multiple current signals x _i,j , where i is the serial number of the current sensor, i∈N and i>1, j is the serial number of the analysis period, j∈N ₊ , for any two different values of i, when j takes the same value, x _{i and j} have the same number of sampling points N, when N meets the requirements of the analysis period, turn to To step 2) carry out the characteristic analysis of the first arc fault; 2) Form a high-dimensional mixed signal matrix X=[x _1,j ,x _2,j ,…, _xi,j ] ^T from the collected multi-channel current signals, and perform de-meaning and whitening processing on the obtained mixed signal matrix, After fast independent component analysis, the unmixing matrix W can be obtained, the source signal matrix S=WX=[s _1,j ,s _2,j ,…,s _i,j ] ^T is calculated, and the effective independent main source signal s is selected _1,j , perform fast Fourier transform on s _1,j , calculate the variance of the one-dimensional frequency matrix in the frequency domain, obtain the first characteristic value r _1,j , go to step 3); 3) Set the threshold of the first feature quantity in the current analysis period to be A ₁ × μ _1,j -A ₂ ×σ _1,j , where the coefficients A ₁ and A ₂ are based on the set first feature quantity threshold and the first A characteristic value comparison can correctly obtain the corresponding photovoltaic system state, μ _1,j is the mean value estimation of all the first characteristic values from the first analysis period to the current analysis period, σ _1,j is the mean value estimate from the first analysis period The standard deviation of all the first feature values up to the current analysis period, A ₁ ∈ Z, A ₂ ∈ Z, compare the first feature value with the set first feature value threshold, and output the corresponding level judgment result: if r _1,j ≥A ₁ ×μ _1,j –A ₂ ×σ _1,j , then output the judgment result 0 and store it in the first fault arc judgment matrix out ₁ [j]; if r _1,j <A ₁ ×μ _1,j -A ₂ ×σ _1,j , then output the judgment result 1, store it in the first fault arc judgment matrix out ₁ [j], and go to step 4) to analyze the second fault arc characteristics; 4) Select one of the multiple current signals to perform S-transformation to obtain a two-dimensional complex time-frequency matrix in the time-frequency domain, calculate the integral of the absolute value of the high-frequency component in the frequency dimension along time, and obtain the second characteristic value r _{2, j} , go to step 5); 5) Set the threshold value of the second feature quantity in the current analysis period as A ₃ ×μ _2,j -A ₄ ×σ _2,j , where the coefficients A ₃ and A ₄ are based on the set second feature quantity threshold and the first The comparison of the two characteristic quantities depends on whether the corresponding photovoltaic system state can be obtained correctly. μ _2,j is the mean value estimation of all the second characteristic quantities from the first analysis period to the current analysis period, and σ _2,j is the mean value estimate from the first analysis period The standard deviation of all the second feature values up to the current analysis period, A ₃ ∈ Z, A ₄ ∈ Z, compare the second feature value with the set second feature value threshold, and output the corresponding level judgment result: if r _2,j ≥A ₃ ×μ _2,j –A ₄ ×σ _2,j , then output the judgment result 0 and store it in the second fault arc judgment matrix out ₂ [j]; if r _2,j <A ₃ ×μ _2,j –A ₄ ×σ _2,j , then output the judgment result 1, store it in the second fault arc judgment matrix out ₂ [j], and go to step 6) to carry out the output judgment on the two feature quantity decision-making layers result weighting; 6) Use dynamic weight coefficients to weight the output judgment results of independent component analysis and S-transformation, and obtain the weighted result outtemp _j = C _1,j ×out ₁ [j]+C _2,j ×out ₂ [j], and then perform preliminary State judgment: if outtemp _j >n, where n is the weighted result threshold, then output the judgment result 1 and store it in the preliminary state judgment result matrix outt[j]; otherwise, output the judgment result 0 and store it in the preliminary state judgment result matrix outt[j], go to step 7) to distinguish the state of the photovoltaic system, C _1,j and C _2,j are the weight coefficients of the first feature quantity and the second feature quantity; 7) Set the judgment accuracy p, and judge the state of the photovoltaic system every p time period: the matrix outt of the preliminary state judgment result is from the j-3pth element to the j-pth element, and from the j-2pth element to the jth element The number of elements is 1, if the value of the counted number is greater than p, it is confirmed that the fault arc of the photovoltaic system occurs within the j-2pth to j-pth period, and the corresponding protection measures for the fault arc of the photovoltaic system are taken; Otherwise, it is considered that the photovoltaic system is in normal operation during the period j-2p to j-p, and returns to step 1) to analyze the current signal in the next analysis period. 2. A method for detecting arc faults in photovoltaic systems using independent component analysis and S-transformation according to claim 1, characterized in that: the bandwidth of the current sensor is greater than 100kHz, and is installed in different positions of the photovoltaic system to display the sampling current For the difference between signals, the value range of current sensors is 2-4; the value range of the sampling frequency f _s is 200-500kHz; the value range of the number of sampling points N is 8000-12000. 3. A method of applying independent component analysis and S-transformation to detect photovoltaic system arc faults according to claim 1, characterized in that: said fast independent component analysis is preferably fast independent component analysis based on negentropy maximization, fast independent Non-linear functions in component analysis can be selected g ₁ (u)=u ³ , g ₂ (u)=u ² , g ₃ (u)=arctan(q ₁ ×u), g ₄ (u)=u×e One of ^(-q ² ₂ ×u ² /2), where q ₁ and q ₂ are constants, the maximum number of iterations ranges from 950 to 1050, and the range of iteration precision is 0.00006 to 0.00015. 4. A method of applying independent component analysis and S-transformation to detect photovoltaic system arc faults according to claim 1, characterized in that: the number of independent main source signals obtained by said fast independent component analysis is the number of channels of sampling current signals , based on the principle of the strongest signal impact, select an effective independent main source signal for subsequent fast Fourier transform processing, that is, calculate the peak-to-peak difference of each independent main source signal within the analysis period, and select the independent main source signal with the largest difference. The main source signal is an effective independent main source signal; the number of transformation points of the fast Fourier transform is selected as the value corresponding to the number of sampling points N. 5. A method of applying independent component analysis and S-transform to detect arc faults in photovoltaic systems according to claim 1, characterized in that: the method of selecting one of the multiple current signals as the S-transform input is: preferentially select the one with the highest sensitivity The current signal corresponding to the current sensor; when there is more than one such current sensor, the current signal corresponding to the current sensor closest to the fault arc occurrence position of the photovoltaic system is preferred; when there are more than one current sensors closest to the fault arc occurrence position of the photovoltaic system, The current signal corresponding to the current sensor with the least number of components in the transmission path from the photovoltaic system fault arc to the current sensor is preferentially selected; the window width adjustment factor of the S transformation is preferably 1. 6. A method of applying independent component analysis and S-transform to detect arc faults in photovoltaic systems according to claim 1, characterized in that: the two-dimensional complex time-frequency matrix elements obtained by S-transform are subjected to absolute value processing to construct the second feature The frequency dimension component of the time-frequency matrix of the quantity is selected as 40~100kHz, and the characteristic frequency band of the fault arc of the photovoltaic system is not related to the value of the sampling frequency f _s . 7. A method for detecting arc faults in photovoltaic systems by applying independent component analysis and S-transformation according to claim 1, characterized in that: the first feature value threshold A ₁ ×μ _1,j -A ₂ ×σ _{1, j} is related to the first feature value of all previous analysis periods and follows the dynamic change of the first feature value _r1 in real time, wherein the coefficients A1 and _A2 are compared with the _first feature value according to the threshold value of the first feature value set by It depends on whether the corresponding photovoltaic system state can be obtained correctly; the mean value estimation μ _1,j and standard deviation σ _1,j are corrected in real time according to the output judgment result of the first characteristic quantity: for the first characteristic quantity obtained in the first analysis period r _1,1 , let the correction amount rtemp _1,1 =r _1,1 , the mean estimate μ _1,1 =r _1,1 , the standard deviation σ _1,1 =0; for the first feature quantity of the jth analysis period Value r _1,j , where j∈N and j>1, if the first feature value in the current analysis period is greater than or equal to the first feature value threshold in the previous analysis period, set the correction value rtemp _1,j =r _{1, j} , the calculation formula of the mean estimate and standard deviation is Among them, k is the sequence number of the analysis period in the accumulation process, k=1, 2...j, j∈N and j>1, if the first feature value in the current analysis period is less than the first feature value threshold of the previous analysis period, Let the correction amount rtemp _1,j ＝A ₁ ×μ _1,j-1 –A ₂ ×σ _1,j-1 , the calculation formula of mean value estimation and standard deviation is The second feature quantity threshold value A ₃ ×μ _2,j -A ₄ ×σ _2,j is related to the second feature quantity values of all previous analysis periods and follows the dynamic change of the second feature quantity r ₂ in real time, wherein the coefficient A ₃ and A ₄ depend on the fact that the corresponding photovoltaic system state can be obtained correctly by comparing the set threshold value of the _second characteristic _quantity with the second characteristic quantity value; Real-time correction of the output judgment result of : For the second feature value r _2,1 obtained in the first analysis period, the correction value rtemp _2,1 =r _2,1 , the mean value estimate μ _2,1 =r _2,1 , standard deviation σ _2,1 =0; for the second feature value r _2,j of the jth analysis period, where j∈N and j>1, if the second feature value in the current analysis period is greater than or equal to the above When the second characteristic quantity threshold of an analysis period is set, the correction value rtemp _2,j = r _2,j , and the calculation formulas for mean value estimation and standard deviation are: Among them, k is the serial number of the analysis period in the accumulation process, k=1, 2...j, j∈N and j>1, if the second characteristic quantity value in the current analysis period is less than the second characteristic quantity threshold value of the previous analysis period, Let the correction amount rtemp _2,j ＝A ₃ ×μ _2,j-1 –A ₄ ×σ _2,j-1 , the calculation formula of mean value estimation and standard deviation is . 8. A method of applying independent component analysis and S-transform to detect photovoltaic system arc faults according to claim 7, characterized in that: the calculation formula for mean value estimation and standard deviation in the current analysis period is obtained by using a recursive relationship: Among them, μ _m,j , σ _m,j are the mean estimate and standard deviation in the current analysis period, respectively, μ _m,j-1 , σ _m,j-1 are the mean estimate and standard deviation in the previous analysis period , rtemp _m,j is the correction amount in the current analysis period, where m is the serial number of the characteristic quantity, and the value is 1 or 2, j∈N and j>1. 9. A method of applying independent component analysis and S-transform to detect photovoltaic system fault arcs according to claim 1, characterized in that: when adopting the output judgment result of dynamic weight coefficient weighted independent component analysis and S-transform, two features The weight coefficient of the quantity output judgment result is determined according to the statistical characteristics of the correctness of the corresponding feature quantity to the state judgment of the historical analysis period, that is, the more analysis periods the feature quantity makes the correct state judgment for the historical analysis period, the more the feature quantity is in the current analysis period. The weight coefficient obtained is larger. Specifically, the weight coefficients C _1,j and C _2,j of the first feature quantity and the second feature quantity are respectively constructed based on the following formulas: Among them, σ ² _out1 and σ ² _out2 are respectively the variance of the first arc fault judgment matrix and the second arc fault judgment matrix from the first element to the jth element, namely Among them, out ₁ and out ₂ are the first arc fault judgment matrix and the second arc fault judgment matrix respectively, k is the counting number of matrix elements, k=1,2...j, j∈N and j>1, and are respectively the mean value estimation of the first arc fault judgment matrix and the second arc fault judgment matrix from the first element to the jth element; if the first arc fault judgment matrix and the second arc fault judgment matrix are from the first element to the jth element The j elements are all 0, that is, both feature quantities determine that all analysis periods are in normal operation state, and directly assign C _1,j = 0, C _2,j = 0; if the j–2p to j–p period The internal photovoltaic system is in a normal operating state, and the positions of the corresponding elements of the first fault arc judgment matrix and the second fault arc judgment matrix under this p period are exchanged for elements. 10. A method for detecting arc faults in photovoltaic systems using independent component analysis and S-transformation according to claim 1, characterized in that: the weighted result threshold n ranges from 0.45 to 0.55; the judgment accuracy p The value range is 2~5.