A photovoltaic dc series arc fault detection method

Abstract

The application provides a photovoltaic DC series arc fault detection method, solves the problem that a DC signal has no cycle and is not easy to be segmented; a self-correlation coefficient of a normal current signal is calculated, a window size is selected according to periodicity and a peak value; the rationality of the window size is verified by calculating the mean and variance of the normal current signal and each segmented window; the arc current signal is windowed, and a time domain waveform feature is extracted; the arc current signal is subjected to Fourier transform, and a low frequency feature in a frequency domain is extracted; the arc current signal is subjected to modal decomposition, the optimal high frequency modal component is selected by combining a Pearson coefficient and a spectral kurtosis method, the high frequency modal component is normalized and windowed, and a high frequency feature in a time-frequency domain is extracted; the features are subjected to equal weight fusion, and a support vector machine is used to realize series arc fault detection. The method comprehensively considers arc features, uses an equal weight fusion strategy, and realizes accurate and reliable detection of series arc faults by using a support vector machine.

Description

Technical Field

[0001] This invention relates to the field of current signal detection technology, and in particular to a method for detecting photovoltaic DC series arc faults. Background Technology

[0002] As the scale of photovoltaic power generation systems expands, the problem of arc faults has become increasingly prominent. Series arc faults, caused by poor contact between silicon chips inside the photovoltaic panel, within junction boxes, and between connecting wires, can often lead to electrical fires if reliable arc fault detection and protection measures are not implemented promptly. Therefore, effectively identifying series arc faults is crucial for ensuring the safety and reliability of photovoltaic power generation systems.

[0003] Existing arc fault detection methods are mainly divided into three types: the first type is based on the characteristics of arc light, arc sound, and electromagnetic radiation; the second type is based on the time-frequency domain characteristics of arc current and voltage; and the third type is based on pattern recognition learning algorithms. Among them, the detection method based on the time-frequency domain characteristics of arc current is currently the mainstream method for DC arc fault detection. The core idea is to extract the time-frequency domain features of arc current and use a threshold method to determine whether an arc fault has occurred. This method has the advantages of strong real-time performance and fast response speed.

[0004] The detection method based on the time-frequency domain characteristics of arc current is currently the most widely used method for DC arc fault detection. However, in distributed photovoltaic power generation systems, the scenarios are complex and there are many components. Normal switching actions and load changes can lead to misjudgments, and the threshold cannot adapt to changes in the environment. Traditional feature extraction methods show insufficient adaptability in the face of complex and variable environments, making it difficult to provide efficient and robust DC arc fault detection performance. At the same time, using a single feature for arc fault detection has the problem of poor universality. Due to the diversity of arc faults, relying on a single feature for detection is easily affected by interference and cannot fully capture the diverse manifestations of arc faults, resulting in a lack of comprehensiveness and accuracy in the detection results. Wu et al. [Wu Chunhua, Xu Wenxin, Li Zhihua et al. Identification of arc type of fault in DC side of photovoltaic system and circuit protection [J]. Proceedings of the CSEE, 2017, 37(17): 5028-5036+5222.] proposed to extract the time-domain features of the signal by standard deviation. This feature extraction method has the problem of insufficient adaptability, and the extracted features are insufficient to characterize arc faults. Zhang et al. [Zhang Guanying, Li Changwei, Zhao Yuan et al. DC fault arc detection based on periodic mean change rate [J]. Journal of Electrical Engineering, 2016, 11(09):44-47+54.] proposed a method to detect arc faults by using the periodic mean change rate of arc current. Although this method can effectively capture the significant changes in the arc current signal in the time domain, arc faults are diverse, and relying solely on this feature for arc detection has poor universality. Therefore, this invention proposes a photovoltaic DC series arc fault detection method based on windowed feature extraction and equal-weight fusion. This method can effectively extract significant arc features and comprehensively considers the diversity of arc features. It uses an equal-weight fusion strategy and employs a support vector machine to realize series arc fault detection, thereby improving the accuracy and reliability of arc detection. Summary of the Invention

[0005] The purpose of this invention is to provide a photovoltaic DC series arc fault detection method. This method can effectively capture changes in the current signal caused by the arc fault, extract significant arc features, and solve the problem of insufficient characterization of traditional features. Simultaneously, considering arc features comprehensively, it uses an equal-weight fusion strategy and employs a support vector machine to achieve series arc fault detection, solving the problems of low accuracy and poor reliability in existing arc fault detection methods. This achieves rapid and accurate arc fault detection, ensuring the safety of the photovoltaic power generation system. The purpose of this invention is achieved as follows:

[0006] A method for detecting photovoltaic DC series arc faults includes the following steps:

[0007] Step S1: Sample the current signal to obtain the arc current signal and the normal current signal;

[0008] Step S2: Calculate the autocorrelation coefficient of the normal current signal using the autocorrelation function method. Select the window size based on the periodicity and peak value of the autocorrelation coefficient. Calculate the mean and variance of the normal current signal and its segmented windows to verify the rationality of the selected window size.

[0009] Step S3: Window the arc current signal and extract time-domain waveform features;

[0010] Step S4: Perform Fourier transform on the arc current signal to extract low-frequency features in the frequency domain;

[0011] Step S5: Perform mode decomposition on the arc current signal, combine Pearson coefficient and spectral kurtosis method to select the optimal high-frequency mode component, normalize it and window it to extract high-frequency features in the time-frequency domain.

[0012] Step S6: Using time-domain waveform features, frequency-domain low-frequency features, and time-frequency-domain high-frequency features, a feature dataset is constructed using an equal-weight fusion strategy. Support vector machines are then used to detect series arc faults. If the detection result is abnormal, it is judged as an arc fault.

[0013] In one embodiment of the present invention, step S2 further includes,

[0014] Step S21, calculate the mean and variance of the normal current signal:

[0015]

[0016]

[0017] Where x(m) represents the normal current signal, μ represents the mean of x(m), and σ 2 Let N represent the variance of x(m), and let N represent the total number of sample points of x(m). i Let x(m) represent the i-th sample point.

[0018] In one embodiment of the present invention, step S2 further includes,

[0019] Step S22: Calculate the autocorrelation coefficient of the normal current signal using the autocorrelation function method, plot the autocorrelation coefficient image, identify periodic patterns and peak values ​​on the autocorrelation coefficient image, and determine the window size.

[0020]

[0021] K = min{T|ACF(T) = ACF} peak} Formula (4)

[0022] Where ACF(K) represents the autocorrelation coefficient of x(m) at a delay of K sample points, and K represents the window size, x iLet represent the i-th sample point of x(m), N represent the number of sample points of x(m), μ represent the mean of x(m), T represent the period of the autocorrelation coefficient, and ACF. peak This represents the peak value of the autocorrelation coefficient.

[0023] In one embodiment of the present invention, step S2 further includes,

[0024] Step S23: Window the normal current signal and calculate the mean and variance of each segment window of the normal current signal.

[0025]

[0026]

[0027] Where, μ j Let represent the mean of the j-th window. Let x(j) represent the variance of the j-th window. i Let represent the i-th sample point in the j-th window, s represent the number of windows for x(m), and K represent the number of sample points in the window.

[0028] In one embodiment of the present invention, step S2 further includes,

[0029] Step S24: Compare the mean and variance of the normal current signal and its segmented windows to verify the rationality of the added window size.

[0030] In one embodiment of the present invention, step S3 further includes,

[0031] By windowing the arc current signal and studying the differences between adjacent windows, the time-domain waveform features of the arc current signal are extracted.

[0032]

[0033]

[0034] α = max(z) Formula (9)

[0035] Where x(n) represents the arc current signal, s represents the number of windows for x(n), and y(j) represents the j-th difference window. i Let z(j) represent the i-th sample point in the j-th difference window, z(j) represent the absolute value of the mean of the j-th difference window, K represent the number of sample points in the window, and α represent the maximum absolute value of the mean of the difference window.

[0036] In one embodiment of the present invention, step S4 further includes,

[0037] By performing a Fourier transform on the arc current signal, and considering the characteristics and generation mechanism of DC arcs, a low-frequency range is selected to extract the low-frequency features in the arc current signal.

[0038]

[0039]

[0040]

[0041] Where X(k) represents the signal after Fourier transform of x(n), k represents the frequency component, ESD(k) represents the energy spectral density of x(n), and E represents the frequency band energy within the frequency band [k1,k2].

[0042] In one embodiment of the present invention, step S5 further includes,

[0043] By performing mode decomposition on the arc current signal, and combining the Pearson coefficient and spectral kurtosis methods, the optimal high-frequency mode components are selected, normalized, and windowed to extract the time-frequency domain high-frequency features of the arc current signal.

[0044]

[0045]

[0046] PSD(k) = |F(IMF(n))| 2 Formula (15)

[0047]

[0048]

[0049] α = max(E) Formula (18)

[0050] Where f represents the number of modes decomposition layers, IMF i (n) represents the i-th modal component, R(n) represents the residual term, and ρ xy The Pearson coefficients represent the relationship between x(n) and the modal components. i Let x(n) be the i-th sample point. Let y represent the average value of x(n). i This represents the i-th sample point of the modal component. Let F(IMF(n)) represent the average value of the modal components, F(IMF(n)) represent the Fourier transform of IMF(n), PSD(k) represent the energy spectral density of IMF(n), PQD represent the spectral kurtosis, k represent the frequency component, E(j) represent the energy of the j-th window of the optimal modal component, and imf(j) represent the energy of the modal components. iLet represent the i-th sample point in the j-th window of the optimal modal component, s represent the number of windows for the optimal modal component, K represent the number of sample points in the window, and α represent the maximum value of the window energy.

[0051] In one embodiment of the present invention, step 6 further includes,

[0052] By utilizing time-domain waveform features, low-frequency features in the frequency domain, and high-frequency features in the time-frequency domain, a feature dataset is constructed using an equal-weight fusion strategy. Support vector machines are then used to detect series arc faults, and abnormal detection results are identified as arc faults.

[0053] This invention provides a photovoltaic DC series arc fault detection method based on windowed feature extraction and equal-weighted fusion. This method achieves the extraction of significant arc features, overcoming the problem of insufficient adaptability of traditional arc feature extraction methods. Simultaneously, by comprehensively considering arc features and employing an equal-weighted fusion strategy, a support vector machine is used to realize series arc fault detection, overcoming the problem of poor universality of single-feature detection methods and further improving the accuracy and reliability of arc fault detection. Attached Figure Description

[0054] Figure 1 This is a flowchart of the photovoltaic DC series arc fault detection method of the present invention.

[0055] Figure 2 This is a schematic diagram of the arc fault current waveform.

[0056] Figure 3 This is a graph showing the normal current signal and its autocorrelation results.

[0057] Figure 4 This is a spectrum diagram of the low-frequency band of normal current signal and arc current signal.

[0058] Figure 5 This is a diagram showing the mode decomposition results of the arc current signal. Detailed Implementation

[0059] The following specific embodiments illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification.

[0060] It should be noted that the accompanying drawings are only used to complement the content disclosed in this specification, so that those skilled in the art can understand and read them, and are not intended to limit the conditions under which the present invention can be implemented, and therefore have no substantial technical significance.

[0061] Figure 1 The flowchart of the photovoltaic DC series arc fault detection method of the present invention is as follows: Figure 1 As shown, the process of the photovoltaic DC series arc fault detection method provided by the present invention is as follows.

[0062] Step S1: Sample the current signal to obtain the arc current signal and the normal current signal. The arc current signal is denoted as x(n), and the normal current signal is denoted as x(m). Figure 2 This is a schematic diagram of the arc fault current waveform.

[0063] Step S2: Calculate the autocorrelation coefficient of the normal current signal using the autocorrelation function method. Select the window size based on the periodicity and peak value of the autocorrelation coefficient. Calculate the mean and variance of the normal current signal and its segmented windows to verify the rationality of the selected window size.

[0064] Step S3: Window the arc current signal and extract time-domain waveform features.

[0065] Step S4: Perform Fourier transform on the arc current signal to extract low-frequency features in the frequency domain.

[0066] Step S5: Perform mode decomposition on the arc current signal, combine Pearson coefficient and spectral kurtosis method to select the optimal high-frequency mode component, normalize it and window it to extract high-frequency features in the time-frequency domain.

[0067] Step S6: Using time-domain waveform features, frequency-domain low-frequency features, and time-frequency-domain high-frequency features, a feature dataset is constructed using an equal-weight fusion strategy. Support vector machines are then used to detect series arc faults. If the detection result is abnormal, it is judged as an arc fault.

[0068] The photovoltaic DC series arc fault detection method of the present invention has the following characteristics.

[0069] First, the autocorrelation coefficient of the normal current signal is calculated using the autocorrelation function method. Based on the periodicity and peak value of the autocorrelation results, the window size is determined.

[0070] Second, the arc current signal is windowed, and the time-domain waveform features of the arc current signal are extracted based on the differences between adjacent windows.

[0071] Third, Fourier transform is performed on the arc current signal. Based on the characteristics and generation mechanism of DC arc, a low-frequency range is selected to extract the low-frequency characteristics of the arc current signal in the frequency domain.

[0072] Fourth, modal decomposition is performed on the arc current signal. By combining the Pearson coefficient and spectral kurtosis methods, the optimal high-frequency modal components are selected, normalized, and windowed to extract the time-frequency domain high-frequency features of the arc current signal.

[0073] Fifth, taking into account the characteristics of the electric arc, an equal-weight fusion strategy is used, and a support vector machine is employed to detect series electric arc faults.

[0074] To address this, the autocorrelation function method was used to calculate the autocorrelation coefficient of the normal current signal. The periodicity and peak value of the autocorrelation coefficient were observed to determine the appropriate window size, overcoming the difficulty in segmenting DC signals due to their non-periodic nature. The arc current signal was windowed, and the differences between adjacent windows were studied to better extract time-domain waveform features. The Fourier transform method was used to process the arc current signal. Considering the characteristics and generation mechanism of DC arcs, a low-frequency range was selected to observe the differences between the normal current signal and the fault current signal, thus better extracting low-frequency features in the frequency domain. The mode decomposition method was used to process the arc current signal. Combining the Pearson coefficient and spectral kurtosis methods, the optimal high-frequency mode components were selected, normalized, and windowed to better extract high-frequency features in the time-frequency domain. Simultaneously, the time-domain waveform features, low-frequency features, and high-frequency features of the arc were comprehensively used, and an equal-weight fusion strategy was employed to construct a feature dataset, forming a complete representation of arc fault features. Support vector machines were used to implement series arc fault detection, improving the accuracy and reliability of arc detection.

[0075] Step S2 uses the autocorrelation function method to calculate the autocorrelation coefficient of the normal current signal, selects the window size based on the periodicity and peak value of the autocorrelation coefficient, calculates the mean and variance of the normal current signal and its segmented windows, and verifies the rationality of the window size. Specifically, step S2 includes the following steps.

[0076] Step S21, calculate the mean and variance of the normal current signal:

[0077]

[0078]

[0079] Where x(m) represents the normal current signal, μ represents the mean of x(m), and σ 2 Let N represent the variance of x(m), and let N represent the total number of sample points of x(m). i Let x(m) represent the i-th sample point.

[0080] Step S22: Calculate the autocorrelation coefficient of the normal current signal using the autocorrelation function method, plot the autocorrelation coefficient image, identify periodic patterns and peak values ​​on the autocorrelation coefficient image, and determine the window size using the following formula:

[0081]

[0082] K = min{T|ACF(T) = ACF} peak} Formula (4)

[0083] Where ACF(K) represents the autocorrelation coefficient of x(m) at a delay of K sample points, and K represents the window size, x iLet represent the i-th sample point of x(m), N represent the number of sample points of x(m), μ represent the mean of x(m), T represent the period of the autocorrelation coefficient, and ACF. peak This represents the peak value of the autocorrelation coefficient. Figure 3 The image shows a normal current signal and its autocorrelation coefficient. It's easy to see that the autocorrelation coefficient exhibits periodicity. Considering its periodicity and peak value, the window size is selected. The autocorrelation function method overcomes the difficulty of segmenting DC signals due to their aperiodic nature. Windowing facilitates a better study of the differences between adjacent windows of the arc current, allowing for the extraction of the time-domain waveform characteristics of the arc current signal.

[0084] Step S23: Window the normal current signal and calculate the mean and variance of each segment window of the normal current signal.

[0085]

[0086]

[0087] Where, μ j Let represent the mean of the j-th window. Let x(j) represent the variance of the j-th window. i Let represent the i-th sample point in the j-th window, s represent the number of windows for x(m), and K represent the number of sample points in the window.

[0088] Step S24: Compare the mean and variance of the normal current signal and its segmented windows to verify the rationality of the added window size.

[0089] Step S3 involves windowing the arc current signal and extracting time-domain waveform features, more specifically including:

[0090] By windowing the arc current signal and studying the differences between adjacent windows, the time-domain waveform features of the arc current signal are extracted.

[0091]

[0092]

[0093] α = max(z) Formula (9)

[0094] Where x(n) represents the arc current signal, s represents the number of windows for x(n), and y(j) represents the j-th difference window. i Let z(j) represent the i-th sample point in the j-th difference window, z(j) represent the absolute value of the mean of the j-th difference window, K represent the number of sample points in the window, and α represent the maximum absolute value of the mean of the difference window.

[0095] Step S4 involves performing a Fourier transform on the arc current signal to extract low-frequency features in the frequency domain, and more specifically includes:

[0096] By performing a Fourier transform on the arc current signal, and considering the characteristics and generation mechanism of DC arcs, a low-frequency range is selected to extract the low-frequency features in the arc current signal.

[0097]

[0098]

[0099]

[0100] Where X(k) represents the signal after Fourier transform of x(n), k represents the frequency component, ESD(k) represents the energy spectral density of x(n), and E represents the frequency band energy within the frequency band [k1,k2]. Figure 4 The images show the low-frequency spectra of normal current signals and arc current signals. It is evident that the occurrence of an arc leads to an increase in the low-frequency spectrum of the current signal, exhibiting a significant difference compared to the normal current signal. This is because the formation of a DC arc involves complex processes such as plasma dynamics and gas recombination and reionization. These factors interact and limit the speed of the plasma response, resulting in a relatively low frequency distribution for the DC arc.

[0101] Step S5 involves modal decomposition of the arc current signal. Combining Pearson coefficients and spectral kurtosis methods, the optimal high-frequency modal components are selected, normalized, and windowed to extract high-frequency features in the time-frequency domain. More specifically, this includes:

[0102] By performing mode decomposition on the arc current signal, and combining the Pearson coefficient and spectral kurtosis methods, the optimal high-frequency mode components are selected, normalized, and windowed to extract the time-frequency domain high-frequency features of the arc current signal.

[0103]

[0104]

[0105] PSD(k) = |F(IMF(n))| 2 Formula (15)

[0106]

[0107]

[0108] α = max(E) Formula (18)

[0109] Where f represents the number of modes decomposition layers, IMF i(n) represents the i-th modal component, R(n) represents the residual term, and ρ xy The Pearson coefficients represent the relationship between x(n) and the modal components. i Let x(n) be the i-th sample point. Let y represent the average value of x(n). i This represents the i-th sample point of the modal component. Let F(IMF(n)) represent the average value of the modal components, F(IMF(n)) represent the Fourier transform of IMF(n), PSD(k) represent the energy spectral density of IMF(n), PQD represent the spectral kurtosis, k represent the frequency component, E(j) represent the energy of the j-th window of the optimal modal component, and imf(j) represent the energy of the modal components. i Let represent the i-th sample point in the j-th window of the optimal modal component, s represent the number of windows for the optimal modal component, K represent the number of sample points in the window, and α represent the maximum value of the window energy. Figure 5 The image shows the result of processing the arc current signal using the mode decomposition method. The mode decomposition method can decompose the signal into modal components with different frequency characteristics. The Pearson correlation coefficient between each modal component and the original signal is calculated, and the component corresponding to the largest coefficient value is selected as the fundamental frequency component. The remaining modal components are then high-frequency components. The spectral kurtosis of the remaining modal components is then calculated, and the modal component with the largest spectral kurtosis value is selected as the optimal high-frequency modal component. It is then normalized and windowed to extract the time-frequency domain high-frequency features of the arc current signal.

[0110] Step S6 utilizes time-domain waveform features, low-frequency features in the frequency domain, and high-frequency features in the time-frequency domain to construct a feature dataset using an equal-weight fusion strategy. A support vector machine is then used to detect series arc faults. Abnormal detection results are identified as arc faults. More specifically, this includes:

[0111] By utilizing the time-domain waveform features, low-frequency features, and high-frequency features of the extracted arc fault current signal, a feature dataset is constructed using an equal-weight fusion strategy to form a complete representation of the arc fault features. M1 samples are randomly selected as training samples, and the remaining M2 samples are used as test samples. A support vector machine is used to detect series arc faults; abnormal detection results are identified as arc faults.

[0112] The proposed photovoltaic DC series arc fault detection method can extract significant arc features, overcoming the negative impact of insufficient adaptability of traditional feature extraction methods. Furthermore, it employs an equal-weight fusion strategy and a support vector machine to achieve rapid and stable arc fault detection.

[0113] This invention provides a photovoltaic DC series arc fault detection method. The method first calculates the autocorrelation coefficient of the normal current signal using the autocorrelation function method, and selects the window size based on the periodicity and peak value of the autocorrelation coefficient. The rationality of the window size is verified by calculating the mean and variance of the normal current signal and its segmented windows. Then, the arc current signal is windowed, and the time-domain waveform features of the arc current signal are extracted based on the differences between adjacent windows. The arc current signal is processed using Fourier transform, and its low-frequency features in the frequency domain are extracted by combining arc characteristics and generation mechanisms. The arc current signal is then processed using mode decomposition, and the optimal high-frequency mode components are selected by combining Pearson coefficients and spectral kurtosis methods. These components are then normalized and windowed to extract the high-frequency features in the time and frequency domains of the arc current signal. Finally, an equal-weighted fusion strategy is used, and a support vector machine is employed to realize series arc fault detection, effectively identifying series arc faults. This method can extract significant arc features and, by comprehensively considering arc characteristics, achieve rapid and stable detection of arc faults.

[0114] Although the present invention has been disclosed above by way of embodiments, it is not intended to limit the present invention. Anyone skilled in the art can make some modifications and refinements without departing from the spirit and scope of the present invention. Therefore, the scope of protection of the present invention shall be determined by the appended claims.

Claims

1. A method for detecting photovoltaic DC series arc faults, characterized in that, Includes the following steps: Step S1: Sample the current signal to obtain the arc current signal and the normal current signal; Step S2: Calculate the autocorrelation coefficient of the normal current signal using the autocorrelation function method. Select the window size based on the periodicity and peak value of the autocorrelation coefficient. Calculate the mean and variance of the normal current signal and its segmented windows to verify the rationality of the selected window size. Step S2 further includes calculating the autocorrelation coefficient of the normal current signal using the autocorrelation function method, plotting the autocorrelation coefficient image, identifying periodic patterns and peak values ​​on the autocorrelation coefficient image, and determining the window size: in, This indicates a normal current signal. express In delay The autocorrelation coefficient for each sample point Indicates the window size. express The One sample point, express The number of sample points, express The mean, The period representing the autocorrelation coefficient. Indicates the peak value of the autocorrelation coefficient; Step S3: Window the arc current signal according to the selected window size and extract time-domain waveform features; Step S4: Perform Fourier transform on the arc current signal to extract low-frequency features in the frequency domain; Step S5: Perform mode decomposition on the arc current signal, combine Pearson coefficient and spectral kurtosis method to select the optimal high-frequency mode component, normalize it and perform windowing processing according to the selected window size to extract high-frequency features in the time-frequency domain. Step S6: Using time-domain waveform features, frequency-domain low-frequency features, and time-frequency-domain high-frequency features, a feature dataset is constructed using an equal-weight fusion strategy. Support vector machines are then used to detect series arc faults. If the detection result is abnormal, it is judged as an arc fault.

2. The photovoltaic DC series arc fault detection method according to claim 1, characterized in that, Step S2 also includes, Calculate the mean and variance of the normal current signal: Official (1) Official (2) in, This indicates a normal current signal. express The mean, express variance express Total sample points, express The 1 sample point.

3. The photovoltaic DC series arc fault detection method according to claim 2, characterized in that, Step S2 also includes, Window the normal current signal and calculate the mean and variance of each segment window of the normal current signal: Official (5) Official (6) in, Indicates the first The mean of each window, Indicates the first The variance of each window, Indicates the first The first window One sample point, express The number of windows, This indicates the number of sample points in the window.

4. The photovoltaic DC series arc fault detection method according to claim 3, characterized in that, Step S2 also includes, By comparing the mean and variance of the normal current signal and its segmented windows, the rationality of the added window size can be verified.

5. The photovoltaic DC series arc fault detection method according to claim 1, characterized in that, Step S3 also includes By windowing the arc current signal and studying the differences between adjacent windows, the time-domain waveform features of the arc current signal are extracted. Official (7) Official (8) Official (9) in, Represents the arc current signal. Indicates the first The first window One sample point, express The number of windows, Indicates the first A difference window, Indicates the first The first difference window One sample point, Indicates the first The absolute value of the mean of the difference windows, This indicates the number of sample points in the window. This represents the maximum absolute value of the mean of the difference window.

6. The photovoltaic DC series arc fault detection method according to claim 1, characterized in that, Step S4 also includes, By performing a Fourier transform on the arc current signal, and considering the characteristics and generation mechanism of DC arcs, a low-frequency range is selected to extract the low-frequency features in the arc current signal. Official (10) Official (11) Official (12) in, Represents the arc current signal. express The signal after Fourier transform Represents frequency components, express energy spectral density, Indicates frequency band The energy of the frequency band within.

7. The photovoltaic DC series arc fault detection method according to claim 1, characterized in that, Step S5 also includes, By performing mode decomposition on the arc current signal, and combining the Pearson coefficient and spectral kurtosis methods, the optimal high-frequency mode components are selected, normalized, and windowed to extract the time-frequency domain high-frequency features of the arc current signal. Official (13) Official (14) Official (15) Official (16) Official (17) Official (18) in, Represents the arc current signal. Indicates the number of modes decomposition layers. Indicates the first One modal component, Indicates the remaining terms. express The Pearson coefficient between the modal components, express The One sample point, express The average value, The first modal component represents the modal component. One sample point, This represents the average value of the modal components. express Fourier transform, express energy spectral density, Indicates spectral kurtosis, Represents frequency components, Represents the optimal modal component. The energy of a window, Represents the optimal modal component. The first window One sample point, The number of windows representing the optimal modal components. This indicates the number of sample points in the window. This represents the maximum value of the window energy.

Images