A low voltage ac series arc fault detection method

Abstract

The application provides a low-voltage alternating current series arc fault detection method. In the arc fault component extraction, a background period difference method is used, the current signal of the current period is taken as a reference, the mean value of the current signals of the four most adjacent previous periods is calculated as a period background signal; the current signal of the current period is subjected to a difference operation with the period background signal to extract the significant difference information in the current signal of the current period. Aiming at the significant difference information, comprehensive feature expression is carried out in the time domain, the frequency domain and the time-frequency domain respectively to obtain feature indexes. Then, the Fisher Score algorithm is adopted to sort the Fisher scores of each feature, so as to screen out the two features with the largest values as the final feature expression of the current period current signal. Then, the two features with the largest values are taken as the horizontal and vertical coordinates to draw a feature space distribution graph, and a linear division line is designed in the feature space distribution graph through a logistic regression model to realize the classification of arc fault features and non-fault features.

Description

Technical Field

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

[0002] With the increasing variety of equipment in low-voltage power supply systems, the possibility of arc faults caused by aging lines, loose contacts, and other reasons has significantly increased, leading to a substantial rise in the probability of electrical fires. Therefore, researching an effective method for detecting arc faults has become extremely crucial.

[0003] Currently, arc fault detection methods are mainly divided into three categories: detection based on arc mathematical models, detection based on arc physical phenomena, and detection based on arc voltage or current waveforms. Among these methods, detection based on arc mathematical models relies on assumptions made during the modeling process and cannot fully reflect all characteristics of arc faults; while detection based on arc physical phenomena is only applicable to arc faults in specific locations. Therefore, current research on arc fault detection mainly focuses on detection based on arc voltage or current waveforms. Among arc voltage or current waveform detection methods, current methods primarily rely on arc current waveform detection, while methods for detecting arc voltage waveforms are relatively limited. Detection methods based on arc current waveforms achieve arc fault protection for downstream branches by setting monitoring points upstream of the line. This method has greater applicability and flexibility, and therefore has received wider attention from researchers.

[0004] Arc current waveform-based detection methods typically process current signal data through current signal acquisition, fault information extraction, fault feature analysis and description, fault feature selection, and fault detection. Traditional detection methods often lose important information during signal processing and struggle to capture subtle changes or sudden events in the fault signal, leading to difficulties in effectively and completely expressing arc fault characteristics and achieving effective arc fault detection using existing detection algorithms. Ding et al. [Ding Xin, Zhu Hongwei, Yin Haonan, et al. Fast arc detection method for AC electrical appliances based on Fast Fourier Transform (FFT) [J]. Electrical Appliances and Energy Efficiency Management Technology, 2015, (21): 8-12.] proposed using Fast Fourier Transform for spectral analysis based on the original current data and selecting three feature values ​​as criteria for whether an arc has occurred. This method processes the original current data, making it difficult to effectively and completely express arc fault characteristics. Zhang et al. [Zhang Guanying, Zhang Xiaoliang, Liu Hua, et al. Online detection method for arc faults in low-voltage systems [J]. Journal of Electrical Engineering, 2016, 31(08):109-115.] proposed an online detection method for arc faults in low-voltage systems. Although it obtained common thresholds for arc characteristic quantities under different loads and had good classification effects, the fixed thresholds were difficult to effectively handle complex arc characteristic quantity distributions, resulting in poor applicability. Therefore, this invention adopts a method based on periodic background difference and linear dividing lines. By describing the significant difference components obtained from periodic background difference, it achieves an effective and complete expression of arc fault characteristics. At the same time, an optimized linear dividing line is created in the two-dimensional feature space distribution map constructed by periodic background difference through a logistic regression model to accurately segment complex current signal characteristics, realizing the classification of arc fault characteristics and non-fault characteristics for linear and nonlinear loads, and improving the real-time performance and accuracy of arc fault detection. Summary of the Invention

[0005] The purpose of this invention is to provide a method for detecting low-voltage AC series arc faults. First, using the current cycle current signal as a reference, the average of the currents from the previous four cycles is used as the cycle background signal. Second, the current cycle current signal and the cycle background signal are differentially analyzed to obtain significant difference information contained in the current cycle current signal. Then, this significant difference information is comprehensively expressed in the time domain, frequency domain, and time-frequency domain, and the two features with the highest Fisher scores are selected as the horizontal and vertical axes to plot a feature space distribution map. This method not only highlights the current changes caused by arc faults but also enhances the sensitivity of the detection algorithm to minute current changes, thereby enhancing the feature representation capability of arc faults. Finally, an optimized boundary is created in the two-dimensional feature space distribution map constructed through cycle background differential using a logistic regression model to accurately segment complex current signal features, achieving the classification of arc fault features and non-fault features for linear and nonlinear loads, improving the real-time performance and accuracy of arc fault detection.

[0006] The objective of this invention is achieved as follows:

[0007] A method for detecting low-voltage AC series arc faults includes the following steps:

[0008] Step S1: Sample the current signal and construct a current dataset;

[0009] Step S2: Using the current cycle current signal in the current data set as a reference, calculate the average value of the current signal by taking the current of the four nearest previous cycles of the current signal, and use it as the cycle background signal.

[0010] Step S3: Extract significant difference information contained in the current periodic current signal by performing a difference operation between the current periodic current signal and the periodic background signal.

[0011] Step S4 involves performing feature analysis and description on the extracted significant difference information in the time domain, frequency domain, and time-frequency domain.

[0012] Step S5: Sort the extracted features by Fisher Score algorithm and select the two features with the highest Fisher scores.

[0013] Step S6: Construct a feature space distribution map for the two features with the largest Fisher scores, and design a linear dividing line in the feature space distribution map using a logistic regression model to classify arc fault features and non-fault features.

[0014] In one embodiment of the present invention, step S2 further includes step S21, where, for the initial data of the current signal sequence, the first four cycles cannot construct a periodic background signal. The present invention employs an interpolation algorithm to supplement the periodic current signal. For the initial four cycles of the current signal sequence, linear interpolation is performed between two adjacent cycles. The formula for linear interpolation is:

[0015]

[0016] Where, x n The representative is the number of periods, C n Representing two adjacent periods x n With x n+1 The periodic current signal inserted between, y n This is the periodic current signal corresponding to the nth period, where x represents any value inserted between two adjacent periods. After supplementing the periodic current signal using an interpolation algorithm, the periodic background signal for the first four periods can be calculated. The periodic background signal for the first four periods is as follows:

[0017]

[0018] Where M1, M2, M3, and M4 represent the periodic background signals of the first four cycles, and y1, y2, y3, and y4 represent the periodic current signals corresponding to the first four cycles.

[0019] In one embodiment of the present invention, step S2 further includes step S22, which involves calculating the average of the current signal sequence after the 5th cycle, using the current cycle current signal as a reference, and employing the current signals of the four nearest preceding cycles to obtain the corresponding periodic background signal. The formula for the periodic background signal is as follows:

[0020]

[0021] Among them, M n Represented as the periodic background signal starting from the 5th period, y i This is represented as the periodic current signal corresponding to the i-th period.

[0022] In one embodiment of the present invention, step S3 includes integrating the obtained periodic background signals, wherein the calculation formula for integrating the periodic background signals is as follows:

[0023] M∈{M1,M2,M3,L,M n}Formula (4)

[0024] Where M represents the periodic background signal of the current signal sequence, and n represents the number of periods in the current signal sequence. The periodic background signal and the periodic current signal are differentially analyzed to obtain the significant difference information contained in the current signal of the current period. The formula for calculating the significant difference information is as follows:

[0025] X = MY (Formula 5)

[0026] Where X represents the significant difference information contained in the periodic current signal, M represents the periodic background signal of the current signal sequence, and Y represents the original current signal sequence.

[0027] In one embodiment of the present invention, step S4 further includes performing comprehensive feature representation on the extracted significant difference information in the time domain, frequency domain, and time-frequency domain to obtain 23 feature indicators. In the time domain, 12 features are selected: maximum value, minimum value, mean, peak value, root mean square amplitude, variance, standard deviation, skewness, kurtosis, waveform factor, impulse factor, and margin factor. In the frequency domain, 8 features are selected: average frequency, centroid frequency, mean square frequency, root mean square frequency, frequency variance, frequency standard deviation, band energy, and power spectral entropy. In the time-frequency domain, 3 features are selected: average wavelet coefficient, standard deviation of wavelet coefficient, and skewness of wavelet coefficient.

[0028] In one embodiment of the present invention, step S5 further includes the following: the formula for the feature selection method Fisher Score is:

[0029]

[0030]

[0031]

[0032]

[0033]

[0034] Where F represents a feature among the feature indices; Fisher Score(F) is the Fisher score of feature F; x i x represents the eigenvalue of characteristic F under non-fault conditions. j The characteristic value of characteristic F under fault conditions is represented; N1 represents the number of cycles of the non-fault current signal, and N2 represents the number of cycles of the fault current signal. The eigenvalue x of characteristic F under non-fault conditions i The mean, The eigenvalue x of characteristic F under fault conditions j The mean; The eigenvalue x represents the characteristic F under non-fault conditions. i variance The eigenvalue x represents the characteristic F under fault conditions. j The variance is calculated, and the Fisher Score algorithm is used to sort the Fisher scores of each feature, thereby selecting the two features with the largest Fisher scores as the final feature representation of the current signal in the current cycle.

[0035] In one embodiment of the present invention, step S6 further includes constructing a feature vector using the two features Z1 and Z2 with the largest Fisher scores selected in step S5 as coordinates. The features of all current signals are used to form a feature set. Using the feature set as input, a logistic regression model is trained to obtain the model parameters β0′, β1′, and β2′ of the regression model. A linear dividing line is calculated using these three model parameters. The formula for calculating the linear dividing line is:

[0036]

[0037] In the constructed feature space distribution map, if seven or more data appear in the upper right corner of the linear dividing line, an arc fault is considered to have occurred, and the arc fault detection is completed.

[0038] The purpose of this invention is to provide a method for detecting low-voltage AC series arc faults. If the current period contains an arc fault component, significant arc fault information can be obtained through periodic background subtraction, achieving an effective and complete representation of arc fault characteristics. Simultaneously, the use of a linear segmentation line can better adapt to datasets with different types of loads, effectively improving the real-time performance and accuracy of fault detection, thereby enhancing the reliability of the system. Attached Figure Description

[0039] Figure 1 This is a flowchart of the low-voltage AC series arc fault detection method of the present invention.

[0040] Figure 2 The waveforms of the arc current during non-fault and fault conditions are shown.

[0041] Figure 3 This is a schematic diagram of the differential signal during a non-arc fault.

[0042] Figure 4 This is a schematic diagram of the differential signal during an arc fault.

[0043] Figure 5 This is an example diagram illustrating the classification of fault features and non-fault features based on a linear dividing line method. Detailed Implementation

[0044] 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.

[0045] 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.

[0046] Figure 1 This is a flowchart of the low-voltage AC series arc fault detection method of the present invention, combined with... Figure 1 The detection method of the present invention is as follows.

[0047] Step S1: Sample the current signal and denot it as y(t). Combine this with... Figure 2 The waveforms of the arc current under non-fault and fault conditions are given, and then a current dataset Y = {y1, y2, L, y n}

[0048] Step S2: Using the current cycle current signal in the current data set as a reference, calculate the average value of the current signal from the four nearest previous cycles and use it as the cycle background signal.

[0049] Step S3: Extract significant difference information contained in the current periodic current signal by performing a differential operation between the current periodic current signal and the periodic background signal.

[0050] Step S4 involves performing feature analysis and description on the extracted significant difference information in the time domain, frequency domain, and time-frequency domain.

[0051] Step S5: Sort the extracted features by Fisher Score algorithm and select the two features with the largest Fisher scores.

[0052] Step S6: Construct a feature space distribution map for the two features with the largest Fisher scores, and design a linear dividing line in the feature space distribution map using a logistic regression model to classify arc fault features and non-fault features.

[0053] The low-voltage AC series arc fault detection method of the present invention has the following characteristics.

[0054] First, using the current cycle current signal as a reference, the average of the four nearest cycle currents is calculated and used as the cycle background signal. Then, by performing a difference operation between the current cycle current signal and the cycle background signal, the significant difference information contained in the current cycle current signal is obtained.

[0055] Second, construct a feature space distribution map for the two features with the largest Fisher scores, and design a linear dividing line in this feature space distribution map using a logistic regression model to classify arc fault features and non-fault features.

[0056] To address this, the proposed arc fault feature extraction method utilizes periodic background differencing to obtain significant differences in the current current signal, achieving an effective and complete representation of arc fault features. Simultaneously, the arc fault detection method constructs a feature space distribution map using the two features with the highest Fisher scores. By designing a linear dividing line through a logistic regression model, it can adapt to low-voltage series arc current data from different types of loads, reducing reliance on threshold selection and providing more flexible classification boundaries, thus enabling accurate and rapid arc fault detection.

[0057] Step S2 uses the current cycle current signal as a reference and calculates the average value of the four nearest previous cycle currents of this signal as the cycle background signal, including:

[0058] Step S21: For the initial data of the current signal sequence, the first four periods cannot construct a periodic background signal. This invention uses an interpolation algorithm to supplement the periodic current signal. For the initial four periods of the current signal sequence, linear interpolation is performed between two adjacent periods. The formula for linear interpolation is:

[0059]

[0060] Where, x n The representative is the number of periods, C n Representing two adjacent periods x n With x n+1 The periodic current signal inserted between, y n This is the periodic current signal corresponding to the nth period, where x represents any value inserted between two adjacent periods. After supplementing the periodic current signal using an interpolation algorithm, the periodic background signal for the first four periods can be calculated. The periodic background signal for the first four periods is as follows:

[0061]

[0062] Where M1, M2, M3, and M4 represent the periodic background signals of the first four cycles, and y1, y2, y3, and y4 represent the periodic current signals corresponding to the first four cycles.

[0063] Step S22: For the current signal sequence after the 5th cycle, using the current cycle current signal as a reference, calculate the average value using the four nearest-neighbor current signals of that signal to obtain the corresponding periodic background signal. The formula for the periodic background signal is:

[0064]

[0065] Among them, M n Represented as the periodic background signal starting from the 5th period, y i This is represented as the periodic current signal corresponding to the i-th period.

[0066] Step S3 involves performing a difference operation between the current periodic current signal and the periodic background signal to obtain significant difference information contained in the current periodic current signal. More specifically, it includes integrating the obtained periodic background signals, and the calculation formula for the integration of the periodic background signals is as follows:

[0067] M∈{M1,M2,M3,L,M n}Formula (4)

[0068] Where M represents the periodic background signal of the current signal sequence, and n represents the number of periods of the current signal sequence.

[0069] The periodic background signal and the original periodic current signal are differentially analyzed to obtain the significant difference information contained in the current periodic current signal. The formula for calculating the significant difference information is as follows:

[0070] X = MY (Formula 5)

[0071] Where X represents the significant difference information contained in the periodic current signal, M represents the periodic background signal of the current signal sequence, and Y represents the original current signal sequence. Figure 3 This is a schematic diagram of the differential signal under non-arc fault conditions. Figure 4 This is a schematic diagram of the differential signal during an arc fault. (Combined with...) Figure 3 , Figure 4 The differential signals for non-arc faults and arc faults are presented respectively. This arc fault feature extraction method has stronger generalization ability and can achieve an effective and complete representation of arc fault features.

[0072] Step S4 involves comprehensively representing the extracted significant difference information in the time domain, frequency domain, and time-frequency domain to obtain feature indicators. More specifically, this includes comprehensively representing the extracted significant difference information in the time domain, frequency domain, and time-frequency domain to obtain 23 feature indicators. In the time domain, 12 features are selected: maximum value, minimum value, mean, peak value, root mean square amplitude, variance, standard deviation, skewness, kurtosis, waveform factor, impulse factor, and margin factor. In the frequency domain, 8 features are selected: average frequency, centroid frequency, mean square frequency, root mean square frequency, frequency variance, frequency standard deviation, band energy, and power spectral entropy. In the time-frequency domain, 3 features are selected: average wavelet coefficient, standard deviation of wavelet coefficient, and skewness of wavelet coefficient.

[0073] Step S5 uses the Fisher Score algorithm to sort the Fisher scores of each feature, thereby selecting the two features with the highest Fisher scores as the final feature representation of the current signal in the current cycle. More specifically, the formula for the feature selection method Fisher Score is as follows:

[0074]

[0075]

[0076]

[0077]

[0078]

[0079] Where F represents a feature among the 26 feature indicators; Fisher Score(F) is the Fisher score of feature F; x i x represents the eigenvalue of characteristic F under non-fault conditions. j The characteristic value of characteristic F under fault conditions is represented; N1 represents the number of cycles of the non-fault current signal, and N2 represents the number of cycles of the fault current signal. The eigenvalue x of characteristic F under non-fault conditions iThe mean, The eigenvalue x of characteristic F under fault conditions j The mean; The eigenvalue x represents the characteristic F under non-fault conditions. i variance The eigenvalue x represents the characteristic F under fault conditions. j The variance.

[0080] The Fisher Score algorithm is used to sort the Fisher scores of each feature, and the two features with the largest Fisher scores (average frequency and band energy) are selected as the final feature representation of the current signal in the current cycle.

[0081] In step S6, a feature vector is constructed using the two features Z1 and Z2 with the largest Fisher scores selected in step S5 as coordinates. This constructs a feature set from all current signal features. Using this feature set as input, a logistic regression model is trained to obtain the model parameters β0′, β1′, and β2′. The linear dividing line is calculated using these three model parameters. The formula for calculating the linear dividing line is:

[0082]

[0083] An arc current dataset Y was obtained experimentally and used as the input sample. S1 samples were randomly selected from Y as training samples to obtain a linear dividing line, and the remaining S2 samples were used as test samples. In the constructed feature space distribution map, if 7 or more data points appear above and to the right of the linear dividing line, an arc fault is considered to have occurred, thus completing the arc fault detection. Figure 5 This is an example diagram illustrating the classification of fault features and non-fault features based on a linear segmentation method, combined with... Figure 5 Taking a 400W induction cooker as an example, the effect of using a linear dividing line method to classify fault features and non-fault features is presented.

[0084] This invention provides a method for detecting low-voltage AC series arc faults. Firstly, in the extraction of arc fault components, a background period difference method is designed. This method uses the current periodic current signal as a reference and calculates the average of the four nearest-neighboring periodic currents as the periodic background signal. Further, the current periodic current signal and the periodic background signal are differentially analyzed to obtain significant difference information contained in the current periodic current signal. Secondly, for the extracted significant difference information, comprehensive feature representation is performed in the time domain, frequency domain, and time-frequency domain, obtaining 23 feature indicators. Then, the Fisher Score algorithm is used to sort the Fisher scores of each feature, thereby selecting the two features with the highest Fisher scores as the final feature representation of the current periodic current signal. Finally, a feature space distribution map is constructed using the two features with the highest Fisher scores, and a linear dividing line is designed on this feature space distribution map using a logistic regression model to classify arc fault features from non-fault features. This detection method performs feature description and fault discrimination on the obtained significant difference components, achieving a reliable representation of arc fault features and effectively improving the real-time performance and accuracy of fault detection.

[0085] 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 low-voltage AC series arc faults, characterized in that, Includes the following steps: Step S1: Sample the current signal and construct a current dataset; Step S2: Using the current cycle current signal in the current data set as a reference, calculate the average value of the current signal by taking the current of the four nearest previous cycles of the current signal, and use it as the cycle background signal. Step S3: Extract significant difference information contained in the current periodic current signal by performing a differential operation between the current periodic current signal and the periodic background signal. Step S4 involves performing feature analysis and description on the extracted significant difference information in the time domain, frequency domain, and time-frequency domain. Step S5: Sort the extracted features by Fisher Score algorithm and select the two features with the highest Fisher scores. Step S6: Construct a feature space distribution map for the two features with the largest Fisher scores, and design a linear dividing line in the feature space distribution map using a logistic regression model to classify arc fault features and non-fault features. Step S5 also includes, The formula for the feature selection method Fisher Score is: Official (1) Official (2) Official (3) Official (4) Official (5) in, F2 A feature among the feature indicators that represent a feature; Features F Fisher's fraction; Indicates characteristics under non-fault conditions F eigenvalues, Indicates characteristics under fault conditions F eigenvalues; N 1 indicates the number of cycles of the non-fault current signal. N 2 indicates the number of fault current signal cycles; Characteristics under non-fault conditions F eigenvalues The mean, Characteristics under fault conditions F eigenvalues The mean; Characteristics under non-fault conditions F eigenvalues variance Characteristics indicating fault conditions F eigenvalues The variance of the current signal is determined by sorting the Fisher scores of each feature using the Fisher Score algorithm, thereby selecting the two features with the largest Fisher scores as the final feature representation of the current signal in the current cycle. Step S6 also includes, The two features with the highest Fisher scores selected in step S5 , A feature vector is constructed for the coordinates. The features of all current signals form a feature set. Using this feature set as input, a logistic regression model is trained to obtain the model parameters of the regression model. , , The linear dividing line is calculated using three model parameters. The formula for calculating the linear dividing line is: Official (6) In the constructed feature space distribution map, if seven or more data appear in the upper right corner of the linear dividing line, it is considered that an arc fault has occurred, thus completing the classification of arc fault features and non-fault features.

2. The method for detecting low-voltage AC series arc faults according to claim 1, characterized in that, Step S2 also includes, Step S21: For the initial data of the current signal sequence, the first four periods cannot construct a periodic background signal. An interpolation algorithm is used to supplement the periodic current signal. For the initial four periods of the current signal sequence, linear interpolation is performed between two adjacent periods. The formula for linear interpolation is: Formula (7) in, The representative is the number of cycles. Representing two adjacent cycles and Periodic current signals inserted between It is the first The periodic current signal corresponding to the period, The representation is any value inserted between two adjacent cycles. After supplementing the periodic current signal using an interpolation algorithm, the periodic background signal for the first four cycles can be calculated. The periodic background signal for the first four cycles is as follows: Official (8) in, , , , These represent the periodic background signals for the first four cycles, respectively. , , , These represent the periodic current signals corresponding to the first four cycles.

3. The method for detecting low-voltage AC series arc faults as described in claim 2, characterized in that, Step S2 also includes, Step S22: For the current signal sequence after the 5th cycle, using the current cycle current signal as a reference, calculate the average value using the four nearest-neighbor current signals of that signal to obtain the corresponding periodic background signal. The formula for the periodic background signal is: Formula (9) in, This is represented as the periodic background signal starting from the 5th period. Represented as the first The periodic current signal corresponding to the period.

4. The method for detecting low-voltage AC series arc faults as described in claim 3, characterized in that, Step S3 also includes The obtained periodic background signals are integrated. The calculation formula for the integrated periodic background signals is as follows: Official (10) in, The periodic background signal is represented as a current signal sequence. This is represented by the number of periods in the current signal sequence. The periodic background signal and the periodic current signal are differentiated to obtain the significant difference information contained in the current signal of the current period. The formula for calculating the significant difference information is as follows: Formula (11) where, This represents the significant difference information contained in the periodic current signal. The periodic background signal is represented as a current signal sequence. This represents the original current signal sequence.

5. The method for detecting low-voltage AC series arc faults as described in claim 1, characterized in that, Step S4 also includes, For the extracted significant difference information, comprehensive feature representation was performed in the time domain, frequency domain, and time-frequency domain, resulting in 23 feature indicators. In the time domain, 12 features were selected: maximum value, minimum value, mean, peak value, root mean square amplitude, variance, standard deviation, skewness, kurtosis, waveform factor, impulse factor, and margin factor. In the frequency domain, 8 features were selected: average frequency, centroid frequency, mean square frequency, root mean square frequency, frequency variance, frequency standard deviation, band energy, and power spectral entropy. Three features were selected in the time-frequency domain: the mean value of the wavelet coefficients, the standard deviation of the wavelet coefficients, and the skewness of the wavelet coefficients.

Images