Photovoltaic DC series arc fault identification method and system

Abstract

The invention discloses a photovoltaic direct current series arc fault identification method and system, and relates to the technical field of arc fault identification, and the method comprises the steps: obtaining a bus current signal of a photovoltaic power generation system; decomposing the bus current signal by using a variational mode decomposition technology to obtain K mode components; performing feature extraction on each modal component from M dimensions, and combining sub-features extracted from each modal component to obtain M first feature vectors; selecting N first feature vectors of which the relevance ranks top N from the M first feature vectors; performing feature fusion processing on the N first feature vectors by using a robust principal component analysis technology to obtain fused features; and inputting the fusion features into a fault identification model to obtain a fault type corresponding to the bus current signal, the fault type comprising a branch arc fault or a bus arc fault. The method can determine whether the fault is a bus arc fault or a branch arc fault.

Description

Technical Field

[0001] This application relates to the technical field of arc fault identification, and particularly to a method and system for identifying photovoltaic DC series arc faults. Background Art

[0002] With the large-scale application of photovoltaic power generation systems, the stable operation of photovoltaic power generation systems has become increasingly crucial. During the operation of a photovoltaic power generation system, a series arc fault (SAF) may occur. A series arc fault refers to an arc fault where the arc current flows through a load and is in series with the load. Generally speaking, when a phase wire in a circuit has a break or poor contact, the air at the break or poor contact is broken down and a discharge spark occurs, forming a series arc.

[0003] The detection and location of series arc faults have become key issues in ensuring the safe and stable operation of photovoltaic systems. However, traditional series arc fault detection methods have difficulties in location when facing complex working conditions, and it is difficult to meet the actual business requirements. Summary of the Invention

[0004] This application provides a method and system for identifying photovoltaic DC series arc faults, which can determine whether it is a bus arc fault or a branch arc fault in the presence of a series arc fault.

[0005] To achieve the above object, this application adopts the following technical solutions: In a first aspect, this application provides a method for identifying photovoltaic DC series arc faults, the method including: Obtain the bus current signal of the photovoltaic power generation system; Use the variational mode decomposition (VMD) technique to decompose the bus current signal to obtain K modal components IMF; Extract features from each modal component from M dimensions respectively, and combine the sub-features extracted from each modal component to obtain M first feature vectors; Select N first feature vectors with the top N correlation rankings from the M first feature vectors; Use the robust principal component analysis (RPCA) technique to perform feature fusion processing on the N first feature vectors to obtain a fusion feature; Input the fusion feature into a fault identification model to obtain the fault type corresponding to the bus current signal, where the fault type includes a branch arc fault or a bus arc fault.

[0006] Optionally, the fault identification model is trained through the following method: Obtain a sample bus current signal and the sample label fault type corresponding to the sample bus current signal; Using the variational mode decomposition (VMD) technique, decompose the busbar current signal to obtain K sample mode components IMF; Extract features from each sample mode component respectively from M dimensions, and combine the sub-features extracted from each sample mode component to obtain M sample first feature vectors; Select N sample first feature vectors with the top N correlation rankings from the M sample first feature vectors; Using the robust principal component analysis (RPCA) technique, perform feature fusion processing on the N sample first feature vectors to obtain sample fusion features; Use the sample fusion features and the sample label fault types to train the model to obtain a fault identification model.

[0007] Optionally, the fault identification model includes: an input layer, a hidden layer, and an output layer; the hidden layer includes two corresponding gated recurrent units and an activation layer; the input layer is used to input the fusion features, the hidden layer is used to process the fusion features, and the output layer is used to output the fault types.

[0008] Optionally, the extracting features from each mode component respectively from M dimensions includes: Judge whether the change trend of the k-th mode component is a preset trend, where k is preset; If the change trend of the k-th mode component is the preset trend, extract features from each mode component respectively from M dimensions.

[0009] Optionally, the two corresponding gated recurrent units and the activation layer include a first gated recurrent unit, a first activation layer, a second gated recurrent unit, and a second activation layer. The first gated recurrent unit includes L1 neurons, and the second gated recurrent unit includes L2 neurons, where L1, L2, and the learning rate are determined in the following way: Generate an initial population, which includes an initial L1, an initial L2, and an initial learning rate; Judge whether the current iteration number has reached the preset iteration number; If the current iteration number has not reached the preset iteration number, process the sample fusion features according to the fault identification model composed of L1, L2, and the learning rate in the population corresponding to the current iteration number to obtain the sample predicted fault type corresponding to the current iteration number. According to the sample predicted fault type corresponding to the current iteration number and the sample label fault type corresponding to the sample fusion features, determine the loss value corresponding to the current iteration number; Judge whether the loss value corresponding to the current iteration number is less than the preset loss value; If the loss value corresponding to the current iteration number is less than or equal to the preset loss value, then determine the L1, L2, and learning rate in the population corresponding to the current iteration number as the final L1, final L2, and final learning rate; If the loss value corresponding to the current iteration number is greater than the preset loss value, then determine the L1, L2, and learning rate in the population corresponding to the next iteration number according to the L1, L2, and learning rate in the population corresponding to the current iteration number.

[0010] Optionally, the method further includes: If the current iteration number reaches the preset iteration number, determine the L1, L2, and learning rate in the population corresponding to the current iteration number as the final L1, final L2, and final learning rate.

[0011] Optionally, the determining the L1, L2, and learning rate in the population corresponding to the next iteration number according to the L1, L2, and learning rate in the population corresponding to the current iteration number includes:

[0012] Wherein, represents the L1, L2, and learning rate in the population corresponding to the next iteration number, represents the locally optimal L1, L2, and learning rate in the population after the end of this iteration; b represents a factor that changes with the iteration number and is used to determine the direction of the population spiral trajectory, represents the globally optimal L1, L2, and learning rate in the population after reaching the preset iteration number; S is a constant, is a random vector representing a normal distribution; represents the L1, L2, and learning rate in the population corresponding to the current iteration number; represents the operator of random walk, represents the operator of Brownian motion; is a step size control factor for controlling the size of R L and R B ; n and are the current iteration number and the preset iteration number respectively; is the moving step size.

[0013] In a second aspect, the present application provides a photovoltaic DC series arc fault identification system, and the system includes an acquisition module for acquiring the bus current signal of the photovoltaic power generation system; a decomposition module for decomposing the bus current signal by using the variational mode decomposition (VMD) technique to obtain K modal components IMF; A feature extraction module, which is used to extract features from each modal component in M dimensions respectively, combine the sub-features extracted from each modal component, and obtain M first feature vectors; A screening module, which is used to select N first feature vectors with the top N correlation rankings from the M first feature vectors; A fusion module, which is used to perform feature fusion processing on the N first feature vectors by using the Robust Principal Component Analysis (RPCA) technology to obtain fusion features; An identification module, which is used to input the fusion features into a fault identification model to obtain the fault type corresponding to the bus current signal, and the fault type includes branch arc fault or bus arc fault.

[0014] Optionally, the fault identification model is trained in the following manner: Obtain a sample bus current signal and the sample label fault type corresponding to the sample bus current signal; use the Variational Mode Decomposition (VMD) technology to decompose the bus current signal to obtain K sample modal components IMF; extract features from each sample modal component in M dimensions respectively, combine the sub-features extracted from each sample modal component, and obtain M sample first feature vectors; select N sample first feature vectors with the top N correlation rankings from the M sample first feature vectors; use the Robust Principal Component Analysis (RPCA) technology to perform feature fusion processing on the N sample first feature vectors to obtain sample fusion features; use the sample fusion features and the sample label fault type to perform model training to obtain a fault identification model.

[0015] Optionally, the fault identification model includes: an input layer, a hidden layer, and an output layer; the hidden layer includes two groups of corresponding gated recurrent units and an activation layer; the input layer is used to input fusion features, the hidden layer is used to process the fusion features, and the output layer is used to output the fault type.

[0016] In a third aspect, the present application provides a computing device, including a memory and a processor; Wherein, one or more computer programs are stored in the memory, and the one or more computer programs include instructions; when the instructions are executed by the processor, the computing device is enabled to execute the method according to any one of the first aspect.

[0017] In a fourth aspect, the present application provides a computer-readable storage medium, which is used to store a computer program, and the computer program is used to execute the method according to any one of the first aspect.

[0018] It can be seen from the above technical solutions that the present application has at least the following beneficial effects: An embodiment of the present application provides a photovoltaic DC series arc fault identification method, in which the bus current signal of the photovoltaic power generation system is first collected, and then processed using variational mode decomposition (VMD) technology to decompose the complex bus current signal into K modal components (IMFs). This decomposition method can adapt to the diversity and non-stationarity of signals under complex working conditions, effectively separate the information of different characteristic scales and frequency components in the bus current signal, provide a more detailed and accurate basis for subsequent fault feature extraction, and help to more accurately capture the signal characteristics related to series arc faults; feature extraction is performed on each IMF from M dimensions, and M first eigenvectors are combined to obtain. This multi-dimensional feature extraction method can comprehensively mine the fault information contained in the bus current signal, and can more fully reflect the various characteristic manifestations of series arc faults under complex working conditions, improve the richness and completeness of the features, and thus enhance the characterization ability of the fault; N first eigenvectors ranked top N in terms of relevance are selected from the M first eigenvectors, effectively removing redundant and irrelevant feature information. Under complex working conditions, there may be a large number of interference factors in the bus current signal. Through this feature screening mechanism, the most representative and discriminative features for fault detection and positioning can be retained, the feature dimension is reduced, the amount of calculation is reduced, and the quality and effectiveness of the features are improved, making subsequent fault identification more accurate and efficient. The robust principal component analysis (RPCA) technology is used to perform feature fusion processing on the N first eigenvectors to obtain fused features. This technology can further integrate the filtered feature information, explore the potential relationship between different features, and generate more robust and discriminative fused features. Under complex working conditions, this fused feature can better resist the influence of noise and interference, improve the stability and reliability of fault features, and thus improve the accuracy of fault detection and location. The fused features are input into the fault identification model for identification, which can accurately determine the fault type corresponding to the bus current signal (including branch line arc fault or target arc fault).

[0019] The overall solution is optimized for complex working conditions in all aspects from signal acquisition to fault identification. Whether it is the decomposition of complex signals, multi-dimensional feature extraction, feature screening by VMD technology, or feature fusion by RPCA technology, the solution can better adapt to various changes and interferences under complex working conditions. Compared with traditional methods, it has stronger adaptability and robustness, and can more effectively detect and locate series arc faults, that is, identify whether it is a branch line arc fault or a target arc fault, which improves the operational stability and safety of the photovoltaic system in complex environments.

[0020] It should be understood that the description of technical features, technical solutions, beneficial effects or similar language in this application does not imply that all features and advantages can be achieved in any single embodiment. On the contrary, it can be understood that the description of features or beneficial effects means that specific technical features, technical solutions or beneficial effects are included in at least one embodiment. Therefore, the description of technical features, technical solutions or beneficial effects in this specification does not necessarily refer to the same embodiment. Furthermore, the technical features, technical solutions and beneficial effects described in this embodiment can be combined in any appropriate manner. Those skilled in the art will understand that an embodiment can be implemented without one or more specific technical features, technical solutions or beneficial effects of a specific embodiment. In other embodiments, additional technical features and beneficial effects can also be identified in specific embodiments that do not embody all embodiments. Description of the Drawings

[0021] Figure 1 It is a flowchart of a photovoltaic DC series arc fault identification method provided by an embodiment of this application; Figure 2 It is an architecture diagram of a photovoltaic power generation system provided by an embodiment of this application; Figure 3 It is a schematic diagram of the process of processing the bus current signal provided by an embodiment of this application; Figure 4 It is a schematic diagram of a branch arc fault provided by an embodiment of this application; Figure 5 It is a schematic diagram of a bus arc fault provided by an embodiment of this application; Figure 6 It is a frequency-domain waveform of the bus current under different conditions provided by an embodiment of this application; Figure 7 It is a schematic diagram of the sum of the frequency-domain amplitudes of the bus current under different frequency bands provided by an embodiment of this application; Figure 8 It is a schematic diagram of a feature label matrix provided by an embodiment of this application; Figure 9 It is a schematic diagram of the modal components of the bus power provided by an embodiment of this application; Figure 10 It is an NMI bar chart provided by an embodiment of this application; Figure 11 It is a comparison schematic diagram provided by an embodiment of this application; Figure 12 It is a structural diagram of a fault identification model provided by an embodiment of this application; Figure 13 It is a schematic diagram of an improved dung beetle optimization algorithm provided by an embodiment of this application; Figure 14 Schematic diagram of a photovoltaic DC series arc fault identification system provided by an embodiment of the present application; Figure 15 Schematic diagram of a computing device provided by an embodiment of the present application. Specific embodiments

[0022] The terms "first", "second", "third", etc. in the specification and drawings of the present application are used to distinguish different objects, rather than to limit a specific order.

[0023] In the embodiments of the present application, words such as "exemplary" or "for example" are used to represent examples, illustrations or explanations. Any embodiment or design solution described as "exemplary" or "for example" in the embodiments of the present application should not be construed as being more preferred or having more advantages than other embodiments or design solutions. Rather, the use of words such as "exemplary" or "for example" is intended to present relevant concepts in a specific manner.

[0024] In order to make the technical solutions of the present application clearer and easier to understand, the technical solutions of the present application will be introduced below with reference to the accompanying drawings.

[0025] A photovoltaic DC series arc fault identification method provided by an embodiment of the present application. This method can be executed by a processing device, which can be a terminal or a server. The terminal includes, but is not limited to, a smart phone, a tablet computer, a laptop computer, a personal digital assistant, or a smart wearable device, etc. The server can be a cloud server, such as the central server in a central cloud computing cluster, or the edge server in an edge cloud computing cluster. Of course, the server can also be a server in a local data center. A local data center refers to a data center directly controlled by a user.

[0026] As Figure 1 shown, this figure is a flowchart of a photovoltaic DC series arc fault identification method provided by an embodiment of the present application. The method includes: S101. The processing device acquires the bus current signal of the photovoltaic power generation system.

[0027] As Figure 2 shown, this figure is an architecture diagram of a photovoltaic power generation system provided by an embodiment of the present application. The photovoltaic power generation system includes: a programmed DC voltage source 101, a decoupling unit 102, an impedance unit 103, an arc generator 104, a fuse 105, an inverter 106, a DC circuit breaker 107, an AC circuit breaker 108, an anti-backflow device 109, and a load 110. The bus current acquisition position is as Figure 2 shown at the midpoint A.

[0028] In the embodiments of the present application, when it is necessary to identify a photovoltaic DC series arc fault, the processing device can acquire atFigure 2 The busbar current signal collected at the midpoint A. In some examples, the processing device may collect the busbar current signal according to time periods, and the time length of the time period can be set according to requirements. For example, 30 seconds, 1 minute, etc.

[0029] S102. The processing device uses VMD technology to decompose the busbar current signal to obtain K IMFs.

[0030] Variational Mode Decomposition (VMD) is a signal decomposition method that can perform adaptive decomposition, accurately extract features, has strong anti-interference ability, and has no mode mixing.

[0031] Any complex non-linear and non-stationary signal can be decomposed into several modal components (Intrinsic Mode Function, IMF) and a residual component. Each IMF component represents the information of different time scales and frequency components in the original signal, arranged in order from high frequency to low frequency. This can more clearly analyze the characteristics of the original signal. For example, in signal processing, it helps to extract the key features of the signal and remove interference information such as noise.

[0032] As Figure 3 shown, this figure is a schematic diagram of the process of processing the busbar current signal provided by the embodiment of the present application.

[0033] After the processing device obtains the busbar current signal, it can first use VMD technology to decompose the busbar current signal to obtain K linear modal components, that is, IMF1~IMFK.

[0034] This decomposition method can adapt to the diversity and non-stationarity of signals under complex working conditions, effectively separate the information of different characteristic scales and frequency components in the busbar current signal, provide a more detailed and accurate basis for subsequent fault feature extraction, and help to more accurately capture the signal features related to series arc faults.

[0035] For example, the processing device can perform adaptive non-linear decomposition on the busbar current signal through VMD technology to capture the hidden internal fault information. Assume the busbar current signal is decomposed into modal components, then the constrained variational model is:

[0036] Among them, is the busbar current; is the th modal component, is the The central frequency of each modal component, and K represents the total number of modal components; is the Dirac function, represents the derivative operation with respect to time and " " is the convolution symbol; is the norm; represents the imaginary function; {·} represents the set of variables.

[0037] S103. The processing device extracts features from each modal component in M dimensions respectively, combines the sub-features extracted from each modal component, and obtains M first feature vectors.

[0038] Among them, the M dimensions can be variance, peak-to-peak value, skewness, kurtosis, peak factor, impact factor, sum of 10kHz frequency domain amplitudes, sum of 20kHz frequency domain amplitudes, sum of 60kHz frequency domain amplitudes, Shannon entropy, sample entropy, approximate entropy, fuzzy entropy, permutation entropy, weighted permutation entropy, wavelet entropy, wavelet-variance-DEO3S, VMD-energy entropy, VMD-sample entropy, VMD-approximate entropy, and VMD-fuzzy entropy. The above M dimensions are respectively derived from time domain feature quantities, frequency domain feature quantities, entropy feature quantities, and feature quantities mentioned in the literature.

[0039] The processing device extracts features from each modal component in M dimensions, then combines the sub-features extracted from each modal component, and obtains M first feature vectors. The M first feature vectors form an M×K order feature matrix :

[0040] This multi-dimensional feature extraction method can comprehensively mine the fault information contained in the bus current signal, can more fully reflect various characteristic manifestations of the series arc fault under complex working conditions, improves the richness and integrity of the features, and thus enhances the ability to characterize the fault.

[0041] In some embodiments, in the case of an arc fault, the change trend of the kth modal component is a preset trend. Based on this, it is possible to first determine whether there is an arc fault, and then determine whether it is a branch arc fault or a bus arc fault specifically.

[0042] The processing device determines whether the change trend of the k-th modal component is a preset trend, where k is preset in advance. If the change trend of the k-th modal component is the preset trend, feature extraction is performed on each modal component from M dimensions respectively. If the change trend of the k-th modal component is not the preset trend, there is no need to perform subsequent processing, that is, there is no arc fault, and there is no need to analyze where the specific fault is. Exemplarily, k = 1. Because, based on the inventor's experimental research, it is found that whether a branch arc fault or a bus arc fault occurs, the amplitude of IMF1 significantly attenuates, which will be introduced in detail later.

[0043] S104. The processing device selects N first feature vectors with the top N correlation rankings from the M first feature vectors.

[0044] After the processing device obtains the M first feature vectors, it can calculate the Normalized Mutual Information (NMI) between the M first feature vectors. NMI is an index widely used in fields such as information theory and machine learning, and is used to measure the correlation between two random variables. Then, N first feature vectors with the top N correlation rankings are selected from them to form an N×K order feature matrix. :

[0045] This screening method effectively removes redundant and irrelevant feature information. Under complex working conditions, there may be a large number of interference factors in the bus current signal. Through this feature screening mechanism, the features that are most representative and discriminative for fault detection and location can be retained, reducing the feature dimension, reducing the calculation amount, and at the same time improving the quality and effectiveness of the features, making the subsequent fault identification more accurate and efficient.

[0046] For the constraint of the above formula to be transformed into an unconstrained form, a penalty factor and a Lagrange multiplier are introduced to construct a Lagrangian function:

[0047] where is the bus current; is the Dirac symbol, representing the inner product of the Lagrange multiplier and the constraint condition error; is set to zero; The GWO algorithm is used to optimize the VMD parameters ( = 10 and = 1216).

[0048] S105. The processing device uses the Robust Principal Component Analysis (RPCA) technology to perform feature fusion processing on the N first feature vectors to obtain fused features.

[0049] Robust Principal Component Analysis (RPCA) technology is a data analysis method used to handle anomalies such as noise and outliers in data. For the filtered feature vectors , the alternating direction method is used to extract the low-rank matrix and the sparse matrix . Among them, the low-rank matrix represents the main part of the bus current signal, and the sparse matrix represents the noise, outliers, or sparse error part in the bus current signal.

[0050]

[0051] As a 1×K-dimensional feature vector, the rank is at most 1 because all rows (or columns) are linearly dependent. Among them, can be regarded as the feature vector corresponding to the t-th time period, that is, . The processing device can also perform similar processing on the bus current signal corresponding to the (t + 1)-th segment, and then obtain . By splicing and , a 2×K-dimensional feature vector is obtained. From , a low-rank matrix is extracted once. Principal Component Analysis (PCA) processing is performed on the extracted low-rank matrix , and the most representative first principal component is selected as the feature vector after dimensionality reduction. For example, the processing device can obtain the principal component from the low-rank matrix . Similarly, the processing device can sequentially process the subsequent features, and then obtain E:

[0052] Among them, E is the fused feature.

[0053] S106. The processing device inputs the fused feature into the fault identification model to obtain the fault type corresponding to the bus current signal.

[0054] Among them, the fault types include branch arc fault or bus arc fault.

[0055] After obtaining the fused feature, the processing device can input the fused feature into the fault identification model, and the fault identification model outputs the fault type corresponding to the bus current signal.

[0056] The training process of the fault identification model is introduced below: The processing device acquires a sample busbar current signal and the corresponding sample label fault type; uses the variational mode decomposition (VMD) technique to decompose the busbar current signal to obtain K sample mode components IMF; extracts features from each sample mode component in M dimensions respectively, combines the sub-features extracted from each sample mode component to obtain M sample first feature vectors; selects the top N sample first feature vectors with the highest correlation from the M sample first feature vectors; uses the robust principal component analysis (RPCA) technique to perform feature fusion processing on the N sample first feature vectors to obtain sample fusion features; uses the sample fusion features and the sample label fault type for model training to obtain a fault identification model.

[0057] In some embodiments, the processing device can acquire the sample busbar current signal and the corresponding sample label fault type in the following manner.

[0058] Considering three working conditions of branch arc fault, busbar arc fault and normal, two DC busbar experiment levels of 100V / 6A and 150V / 10A, and the array power generation situation under different illuminations, an experimental scheme is set as shown in Table 1.

[0059] Table 1:

[0060] Among them, each experimental scheme repeats the arcing experiment 5 times. Each arcing experiment can obtain 200ms of busbar current data, with a total of 40,000 sampling points. Window sliding operation is performed with a window width of 4000 sampling points and a step size of 1000 sampling points to obtain 1110 groups of busbar current experimental data sets of three working conditions, six experimental levels and different illuminations, denoted as X. Taking Schemes 4 and 5 as examples, typical busbar current waveforms are given, as Figure 4 shown. This figure is a schematic diagram of a branch arc fault provided by an embodiment of the present application, as Figure 5 shown. This figure is a schematic diagram of a busbar arc fault provided by an embodiment of the present application.

[0061] From Figure 4 and Figure 5 it can be seen that when a branch arc fault occurs, the busbar current fault is weak, and the busbar current decreases slightly after the fault occurs; when a busbar arc fault occurs, the busbar current fault is obvious, and the busbar current decreases after fluctuating slightly after the fault occurs.

[0062] As Figure 6 shown. This figure is the frequency-domain waveform of the busbar current under different conditions provided by an embodiment of the present application. As Figure 7As shown in the figure, this is a schematic diagram of the sum of the frequency-domain amplitudes of the busbar current in different frequency bands provided by an embodiment of the present application. With a sampling frequency of 200 kHz and 1024 sampling points, Fourier transform is performed on the busbar current under normal conditions, branch arc fault, and busbar arc fault to obtain its spectrum, as Figure 6 ; At intervals of every 10 kHz, calculate the percentage of the sum of the frequency-domain amplitudes of the busbar current in different frequency bands, as Figure 7 .

[0063] From Figure 6 and Figure 7 it can be seen that around 63 kHz and 95 kHz, the amplitude of the series fault arc current is higher than that in the normal situation. Among the sums of the frequency-domain amplitudes of the busbar current in different frequency bands, the percentage changes of the three groups of frequency band amplitudes with the largest changes are all lower than 2%, the discrimination is not high, and there is no obvious pattern. Considering comprehensively Figures 5 - 7 it can be seen that the time-frequency characteristics of the busbar current under normal conditions, branch arc fault, and busbar arc fault are weak and difficult to identify; in the working condition scenarios of multi-branch confluence, strong photovoltaic illumination, or rapid decrease in photovoltaic illumination, the identification difficulty is even greater.

[0064] In each link from signal acquisition to fault identification in the present application, optimized designs are carried out for complex working conditions. Whether it is the decomposition of complex signals by VMD technology, multi-dimensional feature extraction, feature screening, or the feature fusion of RPCA technology, it enables this solution to better adapt to various changes and interferences under complex working conditions. Compared with traditional methods, it has stronger adaptability and robustness, and can more effectively detect and locate series arc faults, that is, identify whether it is a branch arc fault or a target arc fault, improving the operation stability and safety of the photovoltaic system in a complex environment.

[0065] After obtaining the sample data, the processing device can construct a feature-label matrix, as Figure 8 shown in the figure, which is a schematic diagram of a feature-label matrix provided by an embodiment of the present application. The three states of normal condition, branch arc fault, and busbar arc fault are labeled, and the classification label values are assigned 1, 2, and 3 respectively; the sample data set is processed to obtain a feature-label matrix.

[0066] Continuing with the above example, =10 and =1216, the processing device can decompose the busbar current signal to obtain a schematic diagram of the modal classification results under 3 working conditions, as Figure 9As shown in the figure, this is a schematic diagram of the modal components of the busbar power in the embodiment of the present application. In the figure, 901 is the 150V / 10A experiment, and 902 is the 100V / 6A experiment. It can be seen from the figure that whether a branch arc fault or a busbar arc fault occurs, the amplitude of IMF1 decreases significantly, and an arc fault occurrence criterion can be designed based on this; after a branch arc fault and a busbar arc fault occur, the outer envelopes of the VMD modal components IMF2, IMF3, IMF7, IMF9, and IMF10 of the busbar current expand and almost cover the corresponding components under normal conditions. Appropriate features need to be extracted to distinguish between branch arc faults and busbar arc faults by quantity; compared with the 100V / 6A experiment level, the differences in the VMD modal components under the three working conditions at the 150V / 10A experiment level are smaller, and the fault characteristics are weak in systems with more busbars in parallel and during periods of high power generation. Appropriate features need to be selected to identify faults.

[0067] Calculate the feature-label matrices of variance, kurtosis, peak factor, impact factor, Shannon entropy, sample entropy, fuzzy entropy, and weighted permutation entropy of the original signal and the modal components respectively. For calculating the entropy features, the embedding dimension is 2, and the tolerance multiple is 0.2 times the standard deviation of the time series. Calculate their NMI as Figure 10 shown.

[0068] As Figure 10 can be seen, among the modal components, the sum of the three larger NMIs of IMF6 and IMF7 is the largest, which are 1.65 and 1.61 respectively; the smallest is IMF1, which is 0.63; among the state features, the order of the sum of the three larger NMIs from large to small is fuzzy entropy (2.03), sample (2.01), weighted permutation entropy (1.82), indicating that these entropies have strong characteristics after modal decomposition; for sample entropy, fuzzy entropy, and weighted permutation entropy, the NMI of the weighted permutation entropy of the original signal is the lowest, only 0.18. The NMIs of the modal components with the highest NMIs are increased by 227.8%, 14.9%, and 25% respectively compared with the original signal, with an overall increase of 89.2%; compared with the maximum NMI of the original signal features (sample entropy, 0.47), the kurtosis of IMF2, the sample entropy and fuzzy entropy of IMF6, IMF7, and IMF10, and the weighted permutation entropy of IMF6 and IMF7 are all increased, with an overall increase of 13.2%.

[0069] The NMIs of some VMD components have increased, but the effect is weak. If in an extreme working condition of multi-branch busbars in parallel, enhancement is still needed. Moreover, the three entropies of the 10 VMD components, a total of 30-dimensional feature vector data, have too high a dimension and are prone to overfitting. Feature enhancement and dimensionality reduction and reconstruction of NMI are required. Then, the processing device uses the RPCA technology to perform dimensionality reduction and reconstruction on the features selected based on NMI.

[0070] As Figure 11As shown in the figure, this is a comparison schematic diagram provided by an embodiment of the present application. The total NMI of the feature vectors obtained by RPCA processing reaches 2.02, which is a 21.7% increase compared to the total NMI of the modal components with the maximum NMI; it is a 5.1%, 9%, and 11.7% increase respectively compared to the NMI of the modal components with the highest NMI, with a comprehensive increase of 8.6%. The NMI of the feature vectors obtained by RPCA processing is a 244%, 55.3%, and 52.3% increase respectively compared to the original signals, with a comprehensive increase of 117.2%, proving that this method can effectively extract arc fault features and enhance arc fault-related features.

[0071] The above describes the data processing work required during model training, that is, determining the sample fusion features and the corresponding sample label fault types. Next, the structure of the fault identification model will be introduced, and the model training process will be described.

[0072] As Figure 12 shown in the figure, this is a structural diagram of a fault identification model provided by an embodiment of the present application. The fault identification model includes: an input layer 1101, a hidden layer 1102, and an output layer 1103; the hidden layer 1102 includes two corresponding gated recurrent units and an activation layer; the input layer 1101 is used to input the fusion features, the hidden layer 1102 is used to process the fusion features, and the output layer 1103 is used to output the fault types. Among them, the output layer 1103 includes a fully connected layer, a softmax function, and a classification result.

[0073] The two corresponding gated recurrent units and the activation layer include a first gated recurrent unit, a first activation layer, a second gated recurrent unit, and a second activation layer. The first gated recurrent unit includes L1 neurons, and the second gated recurrent unit includes L2 neurons. Among them, L1, L2, and the learning rate are determined by optimizing through an improved dung beetle algorithm.

[0074] The DBO algorithm is an intelligent optimization algorithm that simulates the behavior of dung beetles. Its process mainly includes four core behaviors: rolling the ball, breeding, foraging, and stealing. These four behaviors cooperate with each other, corresponding to the integration of global and local searches in optimization, enabling the algorithm to efficiently find the optimal solution in the search space. However, the traditional dung beetle optimization algorithm forms a population with relatively fixed position updates through random initial solutions, resulting in a reduction in population diversity and poor optimization effects, affecting the convergence speed of the model. Therefore, an improved dung beetle optimization algorithm based on Tent initialization and random walk enhancement is proposed.

[0075] As Figure 13As shown in the figure, this is a schematic diagram of an improved dung beetle optimization algorithm provided by an embodiment of the present application. Among them, the improved parts include segmented sine Tent initialization of the dung beetle population and non-linear stealing behavior. Specifically, the processing device first generates an initial overpopulation, and the initial population includes initial L1, initial L2, and an initial learning rate. Then, it determines whether the current iteration number has reached the preset iteration number. If the current iteration number has not reached the preset iteration number, according to the fault identification model composed of L1, L2, and the learning rate in the population corresponding to the current iteration number, the sample fusion features are processed to obtain the sample predicted fault type corresponding to the current iteration number. According to the sample predicted fault type corresponding to the current iteration number and the sample label fault type corresponding to the sample fusion features, the loss value corresponding to the current iteration number is determined; if the current iteration number reaches the preset iteration number, according to L1, L2, and the learning rate in the population corresponding to the current iteration number, they are determined as the final L1, final L2, and final learning rate.

[0076] Then the processing device determines whether the loss value corresponding to the current iteration number is less than the preset loss value; if the loss value corresponding to the current iteration number is less than or equal to the preset loss value, then L1, L2, and the learning rate in the population corresponding to the current iteration number are determined as the final L1, final L2, and final learning rate; if the loss value corresponding to the current iteration number is greater than the preset loss value, then according to L1, L2, and the learning rate in the population corresponding to the current iteration number, the L1, L2, and learning rate in the population corresponding to the next iteration number are determined.

[0077] Segmented sine Tent initialization of the dung beetle population: Regarding the problem that the chaotic mapping boundary values are relatively dense and affect the algorithm convergence speed, in order to improve the global search ability of the population and the model convergence ability, on the basis of the Tent chaotic mapping, each segment is multiplied by the sin function to make the population distribution more uniform. The formula for the population initialization method of the segmented sine Tent chaotic mapping is:

[0078] Among them, represents the (n + 1)-th generated initialization population, represents the n-th generated initialization population, and are vectors regarding L1, L2, and the learning rate, and The L1, L2, and learning rate represented may be different, .

[0079] Non-linear stealing behavior: In theft behavior, Levy flight (an operator of random walk) and Brownian motion (an operator of Brownian motion) are usually combined to balance global and local search and increase population diversity. However, linear combination is still prone to falling into local optimum in the initial stage of iteration and reducing the convergence speed in the final stage. This application proposes a Levy-Brownian hybrid variational mechanism based on a non-linear factor . In the initial stage of iteration, is close to 1, and the population mainly uses Levy flight with long-distance jumps; in the later stage, λ drops rapidly, and Brownian motion, which is good at local search, is mainly used. The improved formula for updating the position of the theft dung beetle is:

[0080] where, represents L1, L2 and the learning rate in the population corresponding to the next iteration number, represents L1, L2 and the learning rate of the local best in the population after the end of this iteration; b represents a factor that changes with the iteration number and determines the direction of the population spiral trajectory, represents L1, L2 and the learning rate of the global best in the population after reaching the preset iteration number; S is a constant, is a random vector following a normal distribution; represents L1, L2 and the learning rate in the population corresponding to the current iteration number; represents the operator of random walk, represents the operator of Brownian motion; is a step size control factor used to control the magnitudes of R L and R B ; n and are the current number and the preset iteration number respectively; is the moving step size. Among them, the best L1, L2 and learning rate are the values corresponding to the minimum loss value.

[0081] In some examples, it is assumed that the population size P = 30, the maximum number of iterations of the optimization algorithm is 10, and the proportion of the dung beetle population for the four behaviors is 3 (rolling the ball): 2 (reproduction): 2 (foraging): 3 (theft) in sequence. The results after parameter optimization are shown in Table 2.

[0082] Table 2:

[0083] Based on the above description, an embodiment of the present application provides a method for identifying photovoltaic DC series arc faults. This method can extract the characteristics of photovoltaic DC series arcs based on the modal component feature robust principal component analysis technology. After decomposing the bus current signal into several modal components, it uses NMI to preferentially select three kinds of entropy as the characteristics of the modal components of VMD. Finally, it uses RPCA for dimensionality reduction to obtain a feature matrix, and identifies the fault type through a fault identification model. The photovoltaic DC series arc feature extraction technology based on modal component feature robust principal component analysis proposed in this application, compared with traditional feature extraction methods, eliminates the process of screening modal components, effectively avoids the loss of feature information caused by unreasonable modal screening, and at the same time uses RPCA to effectively enhance the informativeness of the feature quantity. Experimental results show that the mutual information coefficient of the feature information extracted by this method has increased by up to 36.8% compared with the existing arc fault feature extraction technology, significantly improving the accuracy and stability of fault diagnosis. The dung beetle optimizer (DBO) is used to optimize the fault identification model. Compared with before optimization, based on the model parameters obtained by optimizing the DBO algorithm, the classification accuracy of the fault identification model can be improved by 0.48%. By using MSDBO to optimize the hyperparameters of the GRU neural network, the fault identification model constructed in this application shows higher diagnostic accuracy and stronger generalization ability in photovoltaic DC arc fault diagnosis. The comprehensive accuracy rate of the proposed algorithm reaches 99.6%, among which the detection rate of arc faults in branches reaches 100%, which is an average increase of 3.9% compared with the existing detection algorithms, effectively improving the diagnostic efficiency.

[0084] As described above in conjunction with Figures 1 - 13 a detailed introduction to the method for identifying photovoltaic DC series arc faults provided by the embodiments of the present application has been given. Next, the systems and devices provided by the embodiments of the present application will be introduced in conjunction with the accompanying drawings.

[0085] As Figure 14 shown, this figure is a schematic diagram of a photovoltaic DC series arc fault identification system provided by an embodiment of the present application. The system includes: An acquisition module 1301, configured to acquire the bus current signal of the photovoltaic power generation system; A decomposition module 1302, configured to decompose the bus current signal by using the variational mode decomposition (VMD) technology to obtain K modal components IMF; A feature extraction module 1303, configured to extract features from each modal component in M dimensions respectively, and combine the sub-features extracted from each modal component to obtain M first feature vectors; A screening module 1304, configured to select N first feature vectors with the top N correlation rankings from the M first feature vectors; A fusion module 1305, configured to perform feature fusion processing on N first feature vectors by using the robust principal component analysis (RPCA) technology to obtain fusion features; An identification module 1306, configured to input the fusion features into a fault identification model to obtain a fault type corresponding to the bus current signal, where the fault type includes a branch arc fault or a bus arc fault.

[0086] Optionally, the fault identification model is trained in the following manner: Obtain a sample bus current signal and a sample label fault type corresponding to the sample bus current signal; use the variational mode decomposition (VMD) technology to decompose the bus current signal to obtain K sample mode components IMF; extract features from each sample mode component in M dimensions respectively, and combine the sub-features extracted from each sample mode component to obtain M sample first feature vectors; select N sample first feature vectors with the top N correlation rankings from the M sample first feature vectors; use the robust principal component analysis (RPCA) technology to perform feature fusion processing on the N sample first feature vectors to obtain sample fusion features; use the sample fusion features and the sample label fault type to perform model training to obtain a fault identification model.

[0087] Optionally, the fault identification model includes an input layer, a hidden layer, and an output layer; the hidden layer includes two corresponding gated recurrent units and an activation layer; the input layer is configured to input fusion features, the hidden layer is configured to process the fusion features, and the output layer is configured to output a fault type.

[0088] The photovoltaic DC series arc fault identification system according to the embodiment of the present application can correspond to execute the method described in the embodiment of the present application, and the above other operations and / or functions of each module / unit of the photovoltaic DC series arc fault identification system are respectively for implementing Figure 1 the corresponding processes of the respective methods in the illustrated embodiments, and for the sake of brevity, they will not be described in detail here.

[0089] The embodiment of the present application further provides a computing device. As Figure 15 shown, this figure is a schematic diagram of a computing device provided by the embodiment of the present application. The computing device 1400 includes a bus 1401, a processor 1402, a communication interface 1403, and a memory 1404. The processor 1402, the memory 1404, and the communication interface 1403 communicate with each other through the bus 1401.

[0090] The bus 1401 can be a peripheral component interconnect (PCI) bus, an extended industry standard architecture (EISA) bus, or the like. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 15 it is only represented by a thick line in Figure 15 , but it does not mean that there is only one bus or one type of bus.

[0091] The processor 1402 can be any one or more of a central processing unit (CPU), a graphics processing unit (GPU), a microprocessor (MP), or a digital signal processor (DSP).

[0092] The communication interface 1403 is used for external communication.

[0093] The memory 1404 can include a volatile memory, such as a random access memory (RAM). The memory 1404 can also include a non-volatile memory, such as a read-only memory (ROM), a flash memory, a hard disk drive (HDD), or a solid state drive (SSD).

[0094] An executable code is stored in the memory 1404, and the processor 1402 executes the executable code to perform the foregoing photovoltaic DC series arc fault identification method.

[0095] Specifically, in the case of implementing Figure 14 the embodiments shown, and Figure 14 when each module or unit of the photovoltaic DC series arc fault identification system described in the embodiments is implemented by software, the software or program code required to execute the functions of each module / unit in Figure 14 can be partially or fully stored in the memory 1404. The processor 1402 executes the program code corresponding to each unit stored in the memory 1404 to perform the foregoing photovoltaic DC series arc fault identification method. Figure 14

[0096] ​The embodiments of the present application also provide a computer-readable storage medium. The computer-readable storage medium can be any available medium that can be stored by a computing device or a data storage device such as a data center containing one or more available media. The available medium can be a magnetic medium (such as a floppy disk, hard disk, magnetic tape), an optical medium (such as a DVD), or a semiconductor medium (such as a solid-state drive), etc. The computer-readable storage medium includes instructions that direct the computing device to execute the above-mentioned photovoltaic DC series arc fault identification method.

[0097] The embodiments of the present application also provide a computer program product, which includes one or more computer instructions. When the computer instructions are loaded and executed on a computing device, they wholly or partially generate the processes or functions described in the embodiments of the present application.

[0098] The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, or data center to another website, computer, or data center in a wired manner (such as coaxial cable, optical fiber, digital subscriber line) or a wireless manner (such as infrared, wireless, microwave, etc.).

[0099] When the computer program product is executed by a computer, the computer executes any one of the aforementioned photovoltaic DC series arc fault identification methods. The computer program product can be a software installation package. In the case where any one of the aforementioned photovoltaic DC series arc fault identification methods is needed, the computer program product can be downloaded and executed on the computer.

[0100] The descriptions of the processes or structures corresponding to the above-mentioned respective drawings each have their own emphases. For parts not detailed in a certain process or structure, reference can be made to the relevant descriptions of other processes or structures.

[0101] As described above, the above are only the specific embodiments of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions within the technical scope disclosed in the present application should be covered by the protection scope of the present application.

Claims

1. A photovoltaic DC series arc fault identification method, characterized in that: The method comprises: Obtain bus current signal of photovoltaic power generation system; The bus current signal is decomposed by using variational mode decomposition (VMD) technology to obtain K modal components (IMF); Extract features from each modal component from M dimensions respectively, combine the sub-features extracted from each modal component to obtain M first eigenvectors; Select the top N first eigenvectors with the highest correlation ranking from the M first eigenvectors; Using the robust principal component analysis (RPCA) technique, the N first eigenvectors are subjected to feature fusion processing to obtain fusion features. The fusion feature is input into a fault identification model to obtain a fault type corresponding to the bus current signal, where the fault type includes a branch arc fault or a bus arc fault.

2. The method according to claim 1, characterized in that The fault identification model is trained in the following way: Acquire a sample bus current signal and a sample label fault type corresponding to the sample bus current signal; Decomposing the bus current signal by using variational mode decomposition (VMD) technology to obtain K sample modal components (IMF); Extract features from each sample modal component from M dimensions respectively, combine the sub-features extracted from each sample modal component, and obtain M sample first feature vectors; Select the first N sample eigenvectors with the top N correlation rankings from the first eigenvectors of the M samples; Using the robust principal component analysis (RPCA) technique, the first feature vectors of N samples are subjected to feature fusion processing to obtain sample fusion features. The sample fusion features and the sample label fault types are used to perform model training to obtain a fault identification model.

3. The method according to claim 2, characterized in that The fault identification model includes: an input layer, a hidden layer and an output layer; the hidden layer includes two groups of corresponding gated recurrent units and an excitation layer; the input layer is used to input fusion features, the hidden layer is used to process the fusion features, and the output layer is used to output the fault type.

4. The method according to claim 1, characterized in that: The feature extraction of each modal component is performed from M dimensions respectively, including: Determine whether the change trend of the kth modal component is a preset trend, where k is preset; If the change trend of the kth modal component is a preset trend, feature extraction is performed on each modal component from M dimensions respectively.

5. The method according to claim 3, characterized in that: The two corresponding groups of gated recurrent units and excitation layers include a first gated recurrent unit, a first excitation layer, a second gated recurrent unit, and a second excitation layer, wherein the first gated recurrent unit includes L1 neurons, and the second gated recurrent unit includes L2 neurons, wherein L1, L2, and the learning rate are determined by the following method: Generate an initial population, wherein the initial population includes an initial L1, an initial L2, and an initial learning rate; Determine whether the current number of iterations reaches the preset number of iterations; If the current number of iterations does not reach the preset number of iterations, the sample fusion features are processed according to the fault identification model composed of L1, L2 and learning rate in the population corresponding to the current number of iterations to obtain the sample predicted fault type corresponding to the current number of iterations, and the loss value corresponding to the current number of iterations is determined according to the sample predicted fault type corresponding to the current number of iterations and the sample label fault type corresponding to the sample fusion features; Determine whether the loss value corresponding to the current number of iterations is less than a preset loss value; If the loss value corresponding to the current number of iterations is less than or equal to the preset loss value, L1, L2 and learning rate in the population corresponding to the current number of iterations are determined as the final L1, final L2 and final learning rate; If the loss value corresponding to the current number of iterations is greater than the preset loss value, the L1, L2 and learning rate in the population corresponding to the current number of iterations are determined according to the L1, L2 and learning rate in the population corresponding to the current number of iterations.

6. The method according to claim 5, characterized in that The method further comprises: If the current number of iterations reaches the preset number of iterations, the final L1, final L2 and final learning rate are determined according to the L1, L2 and learning rate in the population corresponding to the current number of iterations.

7. The method according to claim 5, characterized in that The step of determining L1, L2 and the learning rate in the population corresponding to the next number of iterations according to L1, L2 and the learning rate in the population corresponding to the current number of iterations includes: in, Represents L1, L2 and learning rate in the population corresponding to the next number of iterations, It represents the local optimal L1, L2 and learning rate in the population after the iteration. b represents the factor that changes with the number of iterations and is used to determine the direction of the spiral trajectory of the population. It represents the global optimal L1, L2 and learning rate in the population after reaching the preset number of iterations; S is a constant, represents a random vector following a normal distribution; Indicates L1, L2 and learning rate in the population corresponding to the current number of iterations; represents the random walk operator, An operator representing Brownian motion; is the step size control factor, used to control and The size of n and They are the current number of iterations and the preset number of iterations respectively; is the moving step length.

8. A photovoltaic DC series arc fault identification system, characterized in that: The system comprises An acquisition module, used for acquiring bus current signals of a photovoltaic power generation system; A decomposition module is used to decompose the bus current signal by using variational mode decomposition (VMD) technology to obtain K modal components (IMF); A feature extraction module is used to extract features from each modal component from M dimensions respectively, and combine sub-features extracted from each modal component to obtain M first feature vectors; A screening module, used for selecting N first eigenvectors with top N correlation rankings from the M first eigenvectors; A fusion module is used to perform feature fusion processing on the N first feature vectors using the robust principal component analysis (RPCA) technology to obtain fusion features; The identification module is used to input the fusion feature into a fault identification model to obtain the fault type corresponding to the bus current signal, and the fault type includes a branch arc fault or a bus arc fault.

9. The system according to claim 8, characterized in that The fault identification model is trained in the following way: Obtain a sample bus current signal and a sample label fault type corresponding to the sample bus current signal; decompose the bus current signal using variational mode decomposition (VMD) technology to obtain K sample modal components (IMF); extract features from each sample modal component from M dimensions, combine the sub-features extracted from each sample modal component, and obtain M sample first feature vectors; Select the top N first feature vectors of samples with the highest correlation from the first feature vectors of M samples; use the robust principal component analysis (RPCA) technology to perform feature fusion processing on the first feature vectors of N samples to obtain sample fusion features; use the sample fusion features and the sample label fault type to perform model training to obtain a fault identification model.

10. The system according to claim 9, characterized in that The fault identification model includes: an input layer, a hidden layer and an output layer; the hidden layer includes two groups of corresponding gated recurrent units and an excitation layer; the input layer is used to input fusion features, the hidden layer is used to process the fusion features, and the output layer is used to output the fault type.

Images