A series arc fault detection method based on VMD hilbert marginal spectrum multi-feature fusion

Abstract

This invention belongs to the field of power system automation technology, specifically proposing a fault arc detection method based on variational mode decomposition (VMD) and Hilbert-Huang transform (HHT) multi-feature fusion. The method first performs VMD on the acquired current signal to extract multiple intrinsic mode functions (IMFs) of different frequencies. Then, it performs marginal spectrum analysis on each MIF component. For the selection of MIF marginal spectra, a method based on multi-feature fault discriminative power is proposed, selecting features with high fault discriminative power for each MIF marginal spectrum. To reduce feature redundancy, linear component analysis is used for feature fusion, and finally, the results are imported into a neural network for classification to verify the detection accuracy.

Description

Technical Field

[0001] This invention relates to the field of electrical circuit technology, and more specifically, to a method for detecting arc faults in series circuits. Background Technology

[0002] According to statistics from the Fire and Rescue Bureau of the Ministry of Emergency Management of my country in 2021, the number of electrical fires caused by fault arcs is far higher than the number of electrical fires caused by other types of causes. Therefore, fault arcs can be considered an important cause of electrical fires.

[0003] In recent years, numerous theoretical researchers and engineering experts both domestically and internationally have proposed various fault arc detection technologies. Fault arc detection based on electrical characteristics is currently a mainstream and widely applicable method. It primarily achieves this by studying the current and voltage waveforms within the electrical characteristics, extracting features such as the mean, variance, kurtosis, and waveform factor of the current or voltage to realize fault arc detection.

[0004] Series fault arcs are characterized by their strong concealment and randomness, relatively small current amplitude which is easily masked by load current, high correlation with load properties, and susceptibility to environmental interference. This makes traditional extraction methods and single-time-domain feature processing ineffective, hindering arc detection. Recent research shows that decomposing the current signal can separate noise from useful information, allowing for targeted feature analysis and processing, thereby improving the accuracy and precision of arc fault detection. Furthermore, methods such as EMD, VMD, and wavelet decomposition can extract multiple sub-signals with different frequency components. These sub-signals are more beneficial for analyzing high-frequency features and other frequency domain characteristics, increasing the feature dimensions during fault diagnosis and further improving feature discrimination and identification. However, wavelet transform generally requires signal preprocessing, such as selecting appropriate wavelet basis functions and thresholds, and it still has limitations in suppressing impulse interference. While the EMD method adaptively divides the signal into modal components, it may be affected by factors such as noise interference in some cases, resulting in too many or too few modal components, which is difficult to control. In contrast, the VMD method has good adaptability to changes in the scale and amplitude of the signal, can quickly and accurately monitor changes in the signal, and can set the modal components to be decomposed based on the statistical characteristics of the signal to be decomposed, ensuring higher accuracy.

[0005] Since arc signals often have non-stationary and nonlinear characteristics, traditional time-frequency analysis methods such as STFT (Short-Time Fourier Transform) and WT (Wavelet Transform) may not be able to fully analyze the time-frequency information contained in the signal. Hilbert-Huang transform (HHT) is to obtain the marginal spectrum of the intrinsic mode function (IMF) after the signal is decomposed, which can more accurately reflect the non-stationary and nonlinear characteristics inside the signal. In addition, HHT has adaptability

[14] , especially for small sample signals, thus avoiding the problem that wavelet transform will result in shorter data length as the signal is decomposed. It has more advantages in resolution and feature processing analysis of small sample signals. This paper introduces a combined VMD and HHT approach into the field of series fault arc signal processing. The VMD method decomposes the signal into multiple intrinsic mode functions (IMFs) and extracts the marginal spectrum of each IMF to obtain multi-scale time-frequency information. This decomposition method can compensate for the EMD mode aliasing problem in the HHT algorithm. The Hilbert transform can obtain the time-frequency distribution of the intrinsic mode components (IMFs), such as instantaneous frequency and instantaneous energy. The combination of both is more conducive to the extraction of arc fault signals. To maximize the extraction of fault features, this paper proposes selecting different VMD mode components based on the fault discrimination of each feature quantity. Feature extraction is performed on the selected mode components, and then LDA based on singular values ​​is used for feature dimensionality reduction and fusion of multiple features. Finally, a Generalized Regression Neural Network (GRNN) is used for experimental testing on the selected single-load and multi-load signals. Summary of the Invention

[0006] This invention aims to solve the problems of the prior art. It proposes a series fault arc detection method based on VMD Hilbert marginal spectrum multi-feature fusion. This algorithm mainly solves the limitations of single feature extraction difficulties, low detection accuracy, and limited frequency domain resolution for non-stationary signals. The technical solution of this invention is as follows: To achieve the above objectives, the technical solution adopted by this invention is as follows:

[0007] A series fault arc detection method based on VMD Hilbert marginal spectrum multi-feature fusion, the detection algorithm scheme is as follows:

[0008] A series fault arc detection method based on VMD Hilbert marginal spectrum multi-feature fusion is characterized by including the acquisition of normal and fault currents, VMD fault arc signal decomposition based on greedy algorithm optimization, and the principle and verification of fault arc detection algorithm based on Hilbert marginal spectrum.

[0009] The acquisition of normal and fault currents involves using current transformers and oscilloscopes or other waveform acquisition devices in a series AC system to acquire current signals from different combinations of typical single and multiple loads in the circuit, forming current datasets for single and multiple loads.

[0010] The decomposition of VMD fault arc signal based on greedy algorithm optimization is as follows: First, the decomposition level and penalty factor of VMD are optimized using a greedy algorithm to determine the decomposition parameters. Then, the current data of normal and fault states are decomposed into multiple modes based on the optimized parameters.

[0011] Furthermore, the fault arc detection algorithm based on Hilbert marginal spectrum analyzes the signal after VMD decomposition using Hilbert marginal spectrum analysis, then performs fault discrimination analysis on each modal component using multi-feature value fault discrimination, selects the modal component with high fault discrimination, calculates the feature quantity with high fault discrimination corresponding to the selected modal component, performs feature fusion on the selected feature quantity, and finally imports it into a generalized regression neural network (GRNN) to detect and verify fault arcs under different load scenarios.

[0012] This invention proposes a modal component selection method based on fault discrimination. This involves calculating multiple feature values ​​for each modal component, determining the feature ratio between fault and normal states for each feature value of the current modal component, and selecting features with the largest differences between fault and normal state features. To make the selection results more intuitive, this paper uses the fault discrimination of features to reflect the distinguishing weight of each feature in each modal component. Since fault features generally have larger feature values ​​than normal features, and the greater the difference, the larger the feature ratio, the fault discrimination formula is defined as follows:

[0013] ;

[0014] In the formula, D represents the fault discrimination index, Fa is the fault characteristic value, and Fn is the normal characteristic value. As can be seen from the formula, the closer the fault and normal characteristic values ​​are, the less obvious the fault characteristics are, and the closer the D value is to 0. The more obvious the fault characteristics are, the larger the D value is.

[0015] The LDA algorithm based on singular value decomposition is used to perform feature fusion on the data. This invention performs dimensionality reduction fusion on selected multiple features. It uses singular value decomposition technology to decompose the intra-class scatter matrix and the inter-class scatter matrix, and calculates their eigenvectors and eigenvalues. The k largest eigenvectors are selected as the projection vector matrix, where k represents the dimension of the data after projection dimensionality reduction.

[0016] Compared with the prior art, the present invention has the following beneficial effects:

[0017] (1) This invention is designed for arc fault detection in single-load circuit, dual-load circuit, and quad-load circuit, and is highly applicable to arc fault situations in real life.

[0018] (2) This invention introduces the method of combining variational mode decomposition and Hilbert marginal spectrum into the detection of series fault arcs. This method analyzes each frequency band of the arc signal, eliminating the limitation of frequency domain resolution in Fourier transform when analyzing random and non-stationary signals, thus extracting fault features more effectively.

[0019] (3) To address the limitations of selecting modal components using a single feature, this invention proposes a method that utilizes multiple features to select different modal components, thereby better selecting modal components and extracting fault features from each modal component. Furthermore, to reduce the redundancy of multiple features, LDA based on singular value decomposition is used to fuse and reduce the dimensionality of the multiple features, and multiple neural network algorithms are used for verification experiments. In experimental tests, this invention showed high diagnostic rates with classic loads, other multi-load combinations, and unknown loads, providing a new approach for the accurate diagnosis of fault arcs in power systems. Attached Figure Description

[0020] Figure 1 This is a flowchart of the algorithm of the present invention.

[0021] Figure 2 This is a flowchart of the VMD algorithm optimization process of the present invention.

[0022] Figure 3 A comparison chart of marginal spectrum and FFT.

[0023] Figure 4 This represents the fault discrimination of each modal component in this invention.

[0024] Figure 5 IMF1 and IMF2 select feature quantity fault and normal comparison chart.

[0025] Figure 6 Comparison of faulty and normal features after feature fusion. Detailed Implementation

[0026] The present invention will be further described below with reference to the accompanying drawings and embodiments. The embodiments of the present invention include, but are not limited to, the following embodiments.

[0027] Example

[0028] like Figure 1First, variational mode decomposition is performed on the current signal to obtain multiple intrinsic mode functions at different frequencies. Then, Hilbert marginal spectrum analysis is performed on each mode component, and features with higher marginal spectra of the mode components are selected based on the multi-feature fault discrimination. To eliminate feature redundancy, linear component analysis is used for feature fusion. Finally, a neural network is used to verify the detection accuracy.

[0029] like Figure 2 The steps of the VMD decomposition algorithm optimized by the greedy algorithm are as follows:

[0030] (1) Initialize the values ​​of the decomposition level and the penalty factor α. Given the current decomposition level k and penalty factor α, calculate the values ​​of the objective function F1(k, α) and F2(k, α).

[0031] (2) For the current decomposition level n, add 1 level and calculate the corresponding objective function value. Select the scheme with the larger objective function F1 value to obtain the new decomposition level k_new.

[0032] (3) Based on the new decomposition level k_new, increase or decrease the value of the penalty factor λ, calculate the corresponding objective function binary, select the scheme with smaller objective function binary, and obtain the new penalty factor α_new.

[0033] (4) If the new decomposition level k_new and the penalty factor α_new are equal to the original values, then stop the search; otherwise, return to step (2).

[0034] The final decomposition level k and penalty factor α are the optimal solution.

[0035] like Figure 3 As shown, the Hilbert marginal spectrum is more sensitive to distinguishing arc faults than the FFT, with darker colors representing fault states. The Hilbert marginal spectrum is obtained by integrating the Hilbert spectrum on the time axis. It mainly describes the distribution of energy on the frequency axis. The presence of energy at a certain frequency in the marginal spectrum means that the original signal has a certain intensity at that frequency. Unlike the FFT, which is typically used for analyzing periodic signals and is more effective, the Hilbert marginal spectrum is more sensitive to distinguishing arc faults than the FFT.

[0036] like Figure 4As shown in the figure, black dots represent features with low discriminative power of a certain modal component. It can be seen from the figure that the modal components with low fault discriminative power of the selected load features are mainly concentrated in the IMF3, IM4, and IMF5 modal components. Based on this characteristic, this invention selects IMF1 and IMF2 as the modal components for feature extraction. Furthermore, it can be seen from the figure that among the feature components with low discriminative power in IMF1, features numbered 4, 5, 8, 10, and 11 have relatively low discriminative power, while the feature components with high fault discriminative power in IMF2 have high discriminative power for features numbered 4, 5, 11, and 8. Therefore, this paper selects features 1, 2, 3, 7, 9, 12, 13, and 14 from IMF1, and features 4, 5, 8, and 11 from IMF2, removing features 6 and 10 with low discriminative power. This method of selecting different feature quantities for different IMFs based on fault discriminative power can better reflect the extraction of fault information by each feature quantity and improve the accuracy of detection.

[0037] like Figure 5 As shown, for the selected IMF1, the amplitudes of the first 256 fault data are uneven, and the amplitudes are almost all higher than those of the last 256 normal states. The amplitudes of some fault states are even thousands of times higher than those of the normal states, which has a high degree of discrimination. The four sets of features selected by IMF2 also have this property. The first 256 fault features and the last 256 normal features can be clearly distinguished.

[0038] like Figure 6 As shown, in order to achieve a balance between the accuracy of fault arc identification and the simplification of feature dimensions, the number of features after feature fusion and dimensionality reduction was selected to be 6. This achieves multi-dimensional fusion and dimensionality reduction of sensitive features in the marginal spectrum time-frequency domain and entropy. The state differentiation effect of the 6-dimensional feature after dimensionality reduction is shown in the figure. The X-axis represents 8 groups of loads, each with 6 feature quantities, for a total of 48 groups of data. The y-axis represents 512 groups of experimental data, with the first 256 groups being fault data and the last 256 groups being normal data. It can be observed from the figure that after fitting, the waveform of the first 256 groups of normal state data is smoother and flatter, while the waveform of the last 256 groups of fault arc data is inconsistent in size and chaotic in amplitude. It can be seen that based on the extracted 6 features, the normal state and the fault state can be effectively distinguished.

[0039] The above embodiments are merely one of the preferred embodiments of the present invention and should not be used to limit the scope of protection of the present invention. Any modifications or refinements made to the main design concept and spirit of the present invention that are not of substantial significance, but solve the same technical problem as the present invention, should be included within the scope of protection of the present invention.

Claims

1. A method for detecting series fault arcs based on VMD Hilbert marginal spectrum multi-feature fusion, characterized in that, Includes the following steps: (1) Collect normal and fault current signals under different load scenarios in the series AC system. The collection is completed by current transformer and waveform acquisition equipment to form current datasets for single load and multiple load. (2) The greedy algorithm is used to optimize the number of decomposition layers and the penalty factor of VMD to obtain the optimal decomposition parameters. The parameters are then used to perform multimodal decomposition on the current data collected in step (1) to obtain multiple IMF components. (3) Perform Hilbert marginal spectrum analysis on each IMF component, and select high-discrimination modal components and corresponding feature quantities based on fault discrimination, and remove features with indiscriminate discrimination. (4) The LDA algorithm based on singular value decomposition is used to reduce the dimension of the filtered features and fuse them to obtain the feature vector and feature value; (5) Use generalized regression neural network (GRNN) to detect and verify fault arcs under different load scenarios.

2. The method for detecting series fault arcs based on VMD Hilbert marginal spectrum multi-feature fusion according to claim 1, characterized in that, The acquisition of normal and fault currents in step (1) is specifically as follows: In a series AC system, current transformers and oscilloscopes or other waveform acquisition devices are used to acquire current signals of various types of typical single loads and multiple loads in the circuit, forming current sample sets of different types of single loads and multiple loads.

3. The method for detecting series fault arcs based on VMD Hilbert marginal spectrum multi-feature fusion according to claim 1, characterized in that, The process of optimizing the decomposition level and penalty factor of VMD using a greedy algorithm in step (2) is as follows: taking objective function F1 and objective function F2 as the optimization basis, first initialize the values ​​of the decomposition level and penalty factor α, and calculate the values ​​of objective function F1 and F2 under the current parameters; To increase the current decomposition level by one level and calculate the corresponding objective function value, select the scheme with the larger objective function F1 value to obtain the new decomposition level k_new; based on the determined decomposition level k_new, adjust the value of the penalty factor α and calculate the corresponding objective function value, select the scheme with the smaller objective function F2 value to obtain the new penalty factor α_new; if the newly obtained decomposition level k_new and penalty factor α_new are consistent with the values ​​of the previous round, stop the search, otherwise repeat the above iterative steps.

4. The method for detecting series fault arcs based on VMD Hilbert marginal spectrum multi-feature fusion according to claim 1, characterized in that, The process of filtering high-discrimination modal components and corresponding feature quantities based on fault discrimination in step (3) is as follows: calculate multiple feature quantities for each modal component, calculate the fault discrimination degree D of each feature quantity based on the fault discrimination degree formula, and select the feature quantity with a larger fault discrimination degree D. The fault discrimination formula is: ； Where D is the fault discrimination index, Fa is the fault characteristic value, and Fn is the normal characteristic value.

Images