A zero-sleep and time-frequency feature fusion electric arc detection method and system for electric energy meter burning prevention and control

Abstract

This invention discloses a zero-time and time-frequency feature fusion arc detection method for preventing and controlling electricity meter burnout. The method includes: S1, synchronously acquiring current and voltage signals from the electricity meter circuit and extracting a detection window containing two consecutive power frequency cycles; S2, calculating the zero-time-voltage joint feature coefficient A1, the time-frequency domain similarity comprehensive feature coefficient A2, and the harmonic content coefficient A3 for a specific frequency band based on the current and voltage signals within the detection window; S3, performing a fusion decision on the feature vector [A1, A2, A3] calculated for the current detection window to obtain the arc detection result. This invention reduces the dimensionality of features through a fusion and decision mechanism, thereby reducing redundancy and computational complexity, achieving a highly reliable, low-computational-load arc detection method, thus effectively supporting the prevention and control of electricity meter burnout.

Description

Technical Field

[0001] This invention relates to a zero-downtime and time-frequency characteristic fusion arc detection method and system for preventing and controlling meter burnout, belonging to the field of intelligent meter prevention and control technology. Background Technology

[0002] With the continuous improvement of the intelligence level of power systems, smart meters, as key equipment for user-side electricity metering and consumption management, have been widely used in various power supply and consumption scenarios. However, in actual operation, meter burn-out incidents occur frequently due to various unexpected operating conditions. These incidents are generally caused by abnormal external electrical conditions, such as short circuits, grounding faults, or applied high-voltage pulses, leading to heat loss, combustion, or failure of meter components. Meter burn-out not only causes the meter's metering function to fail but may also trigger electrical safety accidents, threatening power grid operation and electricity safety. Therefore, effective technical means for prevention and early warning are urgently needed.

[0003] Studies have shown that abnormal overheating and arcing caused by wiring faults are among the major contributing factors to the burnout of electricity meters. An electric arc is a high-energy-density transient discharge phenomenon that is easily generated during poor contact, conductor gaps, insulation degradation, or fault switching. Its instantaneous temperature and local energy density are very high, capable of causing metal melting, insulation carbonization, and even localized fire within a very short time, thereby damaging the meter casing, measuring elements, or wiring terminals. Electric arcs cause significant distortion of voltage and current waveforms, accompanied by broadband high-frequency components and non-stationary transient characteristics, providing a theoretical basis for arc detection based on electrical quantities.

[0004] Currently, domestic and international technologies for detecting arc faults in electricity meters mainly include direct detection methods based on optical, acoustic, or thermal sensors, and indirect detection methods based on voltage and current signal analysis. Among these, electrical signal analysis methods have gradually become a research hotspot due to their advantages such as requiring no additional sensors and ease of integration. This method identifies faults by analyzing the voltage and current signal characteristics of the circuit during arc occurrence. In practical applications, due to the large variation in load location, it is difficult to effectively sample voltage signals; therefore, current signal detection is the primary method. Existing current signal analysis methods include:

[0005] (1) Single time-domain or frequency-domain analysis method: Analyze the time-domain characteristics such as average value, kurtosis, and zero-rest time of the current signal, or the frequency-domain characteristics such as harmonic spectrum and harmonic energy ratio. Existing studies use the zero-rest time or zero-crossing rest characteristics of the current to characterize the arc extinction and reignition behavior, but relying solely on a single current zero-rest characteristic for discrimination has obvious shortcomings: On the one hand, the occurrence of current zero rest is closely related to the load characteristics, power supply voltage waveform, and operating status of control equipment, and some nonlinear loads may also produce similar zero-rest phenomena during normal operation; on the other hand, current zero rest caused by the arc is often accompanied by changes in voltage amplitude and waveform distortion, but existing methods have failed to effectively characterize the intrinsic correspondence between current zero rest characteristics and voltage amplitude changes. It can be seen that the detection method with a single feature domain often has the problem of high false alarm rate or high false alarm rate when facing complex load backgrounds, and it is difficult to simultaneously meet the requirements of sensitivity and reliability.

[0006] (2) Time-frequency analysis method. For series fault arc signals, existing technologies widely use characteristic parameters such as current peak value, kurtosis, skewness, harmonic content, and maximum rate of change for detection. These parameters collectively reflect the fault arc characteristics of the signal and are strongly correlated with each other. For example, wavelet transform algorithm is used to perform time-frequency transformation on the current, and the indices of the decomposed dimensionless components are used as fault features. However, existing time-frequency analysis fails to effectively utilize the correlation between different characteristic parameters for feature fusion, resulting in high feature dimensions, increased redundancy, and high computational complexity, making it difficult to operate stably for a long time on terminals with limited computing resources, such as electricity meters. In addition, existing feature fusion methods often directly splice features from different physical levels, resulting in a lack of physical consistency in the fused features, making the fusion results insufficient in physical interpretation and discrimination in fault arc detection application scenarios.

[0007] (3) Intelligent algorithms that autonomously extract signal features, such as feature mapping networks, multilayer convolutional neural networks, and long short-term memory networks, have high computational complexity. When directly deployed on resource-constrained terminals such as electricity meters, they have problems of high computational burden and high power consumption.

[0008] Based on the above, it is evident that traditional detection methods typically rely solely on zero-time, time-domain, or frequency-domain features for series fault arc identification, neglecting the multi-dimensional and multi-physical-level comprehensive characteristics exhibited by the fault arc during actual operation. In complex electrical environments, these methods are highly sensitive to noise, changes in operating conditions, and transient interference, easily leading to misjudgments or missed detections. Furthermore, they lack the ability to characterize the inherent correlations between different features, making it difficult to simultaneously ensure both detection sensitivity and reliability. Therefore, how to fully explore and integrate the time-frequency characteristics of fault arcs in the electrical signals of electricity meters, while ensuring controllable computational complexity, to improve the accuracy and reliability of arc detection, has become an urgent problem to be solved. Summary of the Invention

[0009] Based on the above, the present invention provides a zero-downtime and time-frequency feature fusion arc detection method and system for the prevention and control of meter burnout, in order to solve the problems of insufficient accuracy, robustness and feasibility of series fault arc detection due to the limited computing resources of the meter terminal, the incompleteness of a single criterion, the high redundancy of features and lack of physical meaning, and the sensitivity of the decision mechanism to instantaneous interference.

[0010] The technical solution of this invention is:

[0011] In a first aspect, the present invention provides a zero-downtime and time-frequency characteristic fusion arc detection method for preventing and controlling meter burnout, comprising the following steps:

[0012] S1. Synchronously acquire the current and voltage signals of the energy meter circuit and capture a detection window containing two consecutive power frequency cycles.

[0013] S2. Based on the current signal and voltage signal within the detection window, calculate the zero-voltage joint characteristic coefficient A1, the time-frequency domain similarity comprehensive characteristic coefficient A2, and the harmonic content coefficient A3 for a specific frequency band.

[0014] S3, set the threshold values ​​for the zero-voltage joint characteristic coefficient (T1), the time-frequency domain similarity comprehensive characteristic coefficient (T2), the harmonic content coefficient threshold value for a specific frequency band (T3), and the continuous period decision counting threshold (P). th For the feature vector [A1, A2, A3] calculated in the current detection window, if A1 < T1, A2 < T2, and A3 > T3 simultaneously, then it is determined that there is an arc feature in this window, and the fault counter p is incremented by 1; otherwise, the fault counter p is cleared to zero; when the accumulated fault counter p reaches the counting threshold P th When a continuous series fault arc is detected, a warning signal is triggered.

[0015] Specifically, the continuous cycle cumulative decision includes two parts: first, threshold discrimination, and second, fault arc counting.

[0016] 01) Threshold discrimination: Based on the characteristic values ​​of normal operating conditions and arc operating conditions, select the following thresholds respectively: zero-rest voltage joint characteristic coefficient threshold T1, time-frequency domain similarity comprehensive characteristic threshold T2, specific frequency band harmonic content coefficient threshold T3, and continuous period decision counting threshold P. th The coefficient thresholds T1, T2, and T3 are selected within a range higher than the 95th percentile of normal operating conditions and lower than the mean of arc operating conditions to ensure that there are clear intervals between the feature distinction intervals; the counting threshold P th The selection should be based on the actual application standards, which may include, but are not limited to, GB 14287.4—2014.

[0017] 02) Fault Arc Counting: For a detection window containing two consecutive power frequency cycles obtained through real-time sampling, calculate its feature vector [A1, A2, A3]. If A1 is satisfied simultaneously... <T1、A2<T2、A3> T3 indicates an abnormal zero-current behavior, accompanied by voltage amplitude distortion, significantly reduced consistency between time and frequency domain waveforms, and significantly enhanced harmonic energy in a specific high-frequency band. This indicates the presence of a series fault arc within this window, and the fault count p is incremented by 1. If any of the above conditions are not met, the system is considered normal, and the fault count p is reset to zero. To prevent false alarms caused by transient interference, a continuous accumulation decision strategy is adopted. When the accumulated fault count p reaches P... th When a continuous series fault arc is detected, a fault arc warning signal is immediately triggered, and the counter p is reset to zero.

[0018] Unlike traditional methods that rely solely on zero-time, time-domain, or frequency-domain features for series fault arc identification, this invention utilizes a combination of zero-time-voltage characteristic threshold T1, time-frequency domain similarity comprehensive characteristic threshold T2, specific frequency band harmonic content coefficient threshold T3, and continuous periodic decision counting threshold P. th The system comprehensively judges series fault arcs, taking into account both time and frequency domain characteristics and preventing false alarms caused by instantaneous interference through fault counting accumulation, thereby improving the stability and interpretability of arc detection results.

[0019] In one example, in step S2, the zero-voltage joint characteristic coefficient A1 is calculated as follows:

[0020] Based on the current and voltage signals within the detection window, current waveform data x[n] and y[n] and their corresponding voltage waveform data u under normal operating conditions and the operating conditions to be detected are extracted respectively. x [n]、u y [n], where n = 0, 1, 2, ..., N-1, and N is the number of sampling points in one power frequency cycle;

[0021] Calculate the zero-rest time proportionality coefficient I0 based on the current waveform data. The load voltage amplitude variation coefficient V is calculated based on the voltage waveform data. d V d With load voltage waveform distortion coefficient V s V s , ;

[0022] For I0, V d and V s Using weight vectors Weighted fusion is performed to achieve dimensionality reduction, yielding the zero-voltage joint characteristic coefficient A1, whose calculation formula is as follows:

[0023] in, The preset weighting coefficients, It can be calculated by algorithms such as principal component analysis (PCA) and linear discriminant analysis (LDA).

[0024] By introducing the correspondence between the zero-current characteristic and the voltage amplitude change, a characteristic coefficient A1 is constructed to establish the relationship between the two, thus overcoming the problem of incomplete criterion based on a single zero-current characteristic.

[0025] In one example, the calculation process for the zero-rest time proportion coefficient I0 includes:

[0026] The threshold for determining zero current is set as a preset percentage (a%) of the effective value of the current waveform.

[0027] Count the number P of sampling points with amplitudes lower than the threshold a% in the normal operating current waveform x[n] and the current waveform y[n] under test. x and P y ;

[0028] Calculate zero-day work hours:

[0029]

[0030] Where f is the sampling frequency;

[0031] Calculate the zero-rest time ratio coefficient:

[0032] .

[0033] According to the formula, I0 approaches 0 during normal operation. When I0 approaches 0, it indicates that the current zero-rest behavior is inconsistent and random arc extinction may occur.

[0034] In one example, the load voltage amplitude variation coefficient V d The calculation formula is:

[0035]

[0036] Among them, U x U y These are the effective values ​​of the load voltage under normal operating conditions and fault arc conditions, respectively, derived from the corresponding voltage waveform data u. x [n]、u y [n] is calculated. V d Used to characterize the sudden voltage change phenomenon caused by electric arc.

[0037] In one example, the calculation process for the load voltage waveform distortion coefficient Vs is as follows:

[0038] For the voltage waveform under normal operating conditions u x [n] and the voltage waveform under the test condition u y [n] are subjected to discrete Fourier transforms respectively to obtain its spectrum U x [k] and U y [k], k=0,1,2,…,N-1;

[0039] Let the fundamental wave amplitudes be respectively , Calculate the total harmonic distortion rate:

[0040]

[0041]

[0042] Calculate the load voltage waveform distortion coefficient:

[0043] .

[0044] In one example, the calculation method for the time-frequency domain similarity comprehensive feature coefficient A2 in step S2 is as follows:

[0045] Based on the current signal within the detection window, current waveform data x[n] and y[n] under normal operating conditions and the operating conditions to be detected are extracted, and the maximum cross-correlation coefficient T with normalized absolute value in the time domain is calculated. s The maximum cross-correlation coefficient P of the normalized absolute value in the frequency domain s ;

[0046] For T s and P s Using weight vectors Weighted fusion is performed to achieve dimensionality reduction, resulting in the comprehensive feature coefficient A2 of time-frequency domain similarity. Its calculation formula is as follows:

[0047] in, The preset weighting coefficients, Weight vector Principal components can be calculated using algorithms such as Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA).

[0048] The maximum cross-correlation coefficient T of the time-domain normalized absolute value of the current signal within two cycles is calculated. s and the maximum cross-correlation coefficient P of the normalized absolute value in the frequency domain sA time-frequency domain similarity comprehensive feature coefficient A2 is constructed, which considers both the similarity of time-domain features and frequency-domain features of current signals with different cycles. The time-frequency features are fused and dimensionality reduced, thereby reducing the number of redundant features and the dimension of the feature space. While ensuring the effectiveness of arc detection, the computational load and storage requirements in the feature extraction and decision process are significantly reduced, making it easy to implement.

[0049] In one example, the time-domain normalized absolute value maximum cross-correlation coefficient T s The calculation process includes:

[0050] Calculate the cross-correlation function of two current waveforms x[n] and y[n]:

[0051]

[0052] Where m is the sampling delay of the current waveform y relative to the current waveform x, and y* represents the complex conjugate value;

[0053] Normalizing the cross-correlation function yields the normalized cross-correlation coefficient:

[0054]

[0055] Calculate the maximum cross-correlation coefficient of the normalized time domain absolute value:

[0056] .

[0057] This coefficient can be used to determine the similarity between two periodic current waveforms; the higher the similarity, the greater the similarity. s The closer it is to 1, the closer it is to 0.

[0058] In one example, the frequency domain normalized absolute value maximum cross-correlation coefficient P s The calculation process includes:

[0059] Perform discrete Fourier transforms on the current waveforms x[n] and y[n] respectively to obtain X(k) and Y(k);

[0060] Calculate the frequency domain cross-correlation coefficient:

[0061]

[0062] After normalization, the normalized frequency domain cross-correlation coefficient is obtained:

[0063]

[0064] Calculate the normalized absolute value frequency domain maximum cross-correlation coefficient:

[0065] .

[0066] In one example, the calculation method for the specific frequency band harmonic content coefficient A3 in step S2 is as follows:

[0067] The discrete Fourier transform is performed on the current waveform data y[n] of the working condition to be detected within the detection window to obtain the frequency domain amplitude spectrum I(n f0), where f0 is the fundamental frequency;

[0068] Take a specific frequency band [f a ,f b The sum of squares is calculated for the amplitudes of all discrete frequency points within the frequency band.

[0069]

[0070] Perform normalization and divide the resulting value by the fundamental frequency energy. The harmonic content coefficient A3 is obtained:

[0071] .

[0072] Secondly, the present invention provides a zero-downtime and time-frequency characteristic fusion arc detection system for the prevention and control of meter burnout, comprising:

[0073] The signal acquisition module is used to synchronously acquire the current and voltage signals of the energy meter circuit and capture a detection window containing two consecutive power frequency cycles.

[0074] The feature calculation module is used to calculate the zero-voltage joint feature coefficient A1, the time-frequency domain similarity comprehensive feature coefficient A2, and the harmonic content coefficient A3 of a specific frequency band based on the current signal and voltage signal within the detection window.

[0075] The decision output module is used to set the threshold values ​​for the zero-sleep-voltage joint characteristic coefficient (T1), the time-frequency domain similarity comprehensive characteristic coefficient (T2), the harmonic content coefficient for a specific frequency band (T3), and the continuous period decision count threshold (P). th For the feature vector [A1, A2, A3] calculated in the current detection window, if A1 < T1, A2 < T2, and A3 > T3 simultaneously, then it is determined that there is an arc feature in this window, and the fault counter p is incremented by 1; otherwise, the fault counter p is cleared to zero; when the accumulated fault counter p reaches the counting threshold P th When a continuous series fault arc is detected, a warning signal is triggered.

[0076] The beneficial effects of this invention are:

[0077] 1. To address the incompleteness of criteria based on a single zero-time characteristic, this invention introduces a correspondence between the current zero-time characteristic and the effective voltage value change, constructing a characteristic coefficient A1 for the relationship between the two, thus overcoming the problem of incompleteness based on a single zero-time characteristic. Specifically, A1 integrates the current zero-time ratio I0 and the load voltage amplitude change V. d and waveform distortion V s Three characteristics comprehensively characterize the strong correlation between zero-fire and voltage changes during arc occurrence from both current and voltage dimensions. Compared with existing technologies, this method can effectively distinguish between zero-fire caused by arc and zero-fire caused by normal operating conditions such as changes in equipment operating status, reduce the false alarm rate of fault arcs, and improve the reliability and stability of arc detection under complex load environments.

[0078] 2. To address the issues of high redundancy and lack of physical meaning in existing time-frequency domain detection methods, this invention fuses and reduces the dimensionality of the time-domain normalized absolute value maximum cross-correlation coefficient Ts and the frequency-domain normalized absolute value maximum cross-correlation coefficient Ps, which are at the same level in physical meaning. This results in a time-frequency domain similarity comprehensive feature coefficient A2, reflecting the similarity of current signals with different periods. This feature coefficient considers both the similarity of time-domain and frequency-domain features of current signals with different periods. While ensuring the effectiveness of arc detection, it significantly reduces the computational load and storage requirements in the feature extraction and decision-making process, improving the interpretability and feasibility of the method.

[0079] 3. To address the issues of traditional methods for identifying series fault arcs relying on single-dimensional criteria and being sensitive to transient interference, this invention constructs a comprehensive identification mechanism based on multi-feature threshold fusion and continuous periodic cumulative decision-making. This mechanism combines the zero-rest voltage joint feature threshold T1, the time-frequency domain similarity comprehensive feature threshold T2, the harmonic content coefficient threshold T3 for a specific frequency band, and the continuous periodic decision counting threshold P. th This mechanism combines information from both spatial (multi-feature) and temporal (continuous cycle) dimensions, effectively filtering out transient interference signals caused by load switching, instantaneous noise, etc., improving the stability of arc detection results, and enhancing the robustness and applicability of series fault arc identification. Attached Figure Description

[0080] Figure 1 This is a flowchart of the method described in this invention;

[0081] Figure 2 Waveform of series fault arc experiment;

[0082] Figure 3 Comparison of zero-voltage joint characteristic coefficients under normal and fault conditions

[0083] Figure 4Comparison of time-frequency domain similarity feature coefficients under normal and fault conditions

[0084] Figure 5 Comparison of harmonic content coefficients under normal and fault conditions

[0085] Figure 6 Experimental verification results. Detailed Implementation

[0086] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Many specific details are set forth in the following description to provide a thorough understanding of the present invention. However, the present invention can be practiced in many other ways different from those described herein, and those skilled in the art can make similar modifications without departing from the spirit of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0087] Example 1: A zero-downtime and time-frequency characteristic fusion arc detection method for the prevention and control of meter burnout.

[0088] Detection method reference Figure 1 This example uses a 64-bit Windows 11 operating system, an Intel(R) Core(TM) i7-12900H CPU, an NVIDIA GeForce RTX 3060 graphics card, 40GB of RAM, and Matlab software version R2021b. The specific implementation process is as follows:

[0089] Step S01: Construct an experimental platform for series fault arcing of electricity meters, including an AC power supply, a typical load, an arc generating device, and a smart electricity meter sampling module. Experimental parameter settings are shown in Table 1.

[0090]

[0091] Data was collected under normal steady-state operating conditions and under series fault arc conditions. A set of current and load voltage signal waveforms during a fault arc is shown below. Figure 2 The fault occurred at 0.05 seconds, and a sliding window was used to detect it. In the figure, window 1 represents the detection window under normal operating conditions, and window 2 represents the detection window under the condition to be detected.

[0092] Step S02: Collect data from window 1 and window 2 respectively, and calculate the current zero-rest time proportionality coefficient I0. Let the current waveform data sequences collected under normal operating conditions and within the window be x[n] and y[n], respectively. Set the current zero-rest judgment threshold a% to 5% of the effective value of the current waveform in that window. The threshold under normal operating conditions is: 5%I rms,x ≈2.3293A; the threshold within window 2 is 5%I. rms,y≈2.0506 A. Count the number P of sampling points in x[n] whose amplitude is below the threshold. x =12, the number of sampling points P in y[n] with amplitudes lower than A y =22. Therefore, the zero-day work week is:

[0093]

[0094]

[0095] Zero-day work ratio coefficient: .

[0096] Calculate the load voltage amplitude variation coefficient V d Harmonic distortion rate V s Acquire load voltage waveform data u sampled simultaneously with the current. x [n] and u y [n], calculate the load voltage RMS value variation coefficient:

[0097] Perform a Discrete Fourier Transform on the voltage signal to obtain its spectrum, and calculate the total harmonic distortion rate:

[0098]

[0099]

[0100] The voltage waveform distortion coefficient is: .

[0101] For I0, V d and V s Using weight vectors Weighted fusion is performed to achieve dimensionality reduction, yielding the zero-voltage joint characteristic coefficients A1. This embodiment uses Principal Component Analysis (PCA) to calculate the weight vector. The calculation steps are as follows:

[0102] Assume there are q sets of samples, each set containing the current zero-rest time proportionality coefficient I0 and the voltage amplitude variation coefficient V. d and voltage waveform distortion coefficient V s Three data points, with a data matrix L:

[0103]

[0104] Each column represents a feature, and each row represents a sample. The data is centered by calculating the mean of each feature and subtracting that feature mean from each data point.

[0105]

[0106]

[0107] Calculate the covariance matrix C, and the elements σ of the covariance matrix. ij Represents the covariance between the i-th feature and the j-th feature, for example: .

[0108]

[0109] Perform eigenvalue decomposition on the covariance matrix to obtain the eigenvalues. and the corresponding feature vector Sort by eigenvalues ​​from largest to smallest, assuming and ,but corresponding feature vector The direction of the first principal component represents the direction of change with the largest variance in the data, and the weight vector is selected accordingly. The zero-voltage joint characteristic coefficient of window 2 is then:

[0110]

[0111] Similarly, the zero-voltage joint characteristic coefficients of window 1 are obtained. The result is as follows Figure 3 As shown.

[0112] Step S03: Calculate the time-frequency domain similarity feature A2. First, calculate the time-domain cross-correlation coefficient T. s Calculate the cross-correlation sequence R of signals x[n] and y[n] under normal operating conditions and fault arc conditions. xy [m]. Taking the calculation of the normalized cross-correlation coefficient near zero delay as an example, the zero-delay normalized cross-correlation coefficient is: Temporal similarity features: .

[0113] Calculate the frequency domain cross-correlation coefficient P s Perform a DFT on x[n] and y[n] to obtain X[k] and Y[k]. Calculate the frequency domain cross-power spectrum and perform an inverse transform to obtain the frequency domain cross-correlation sequence. Through a normalization process similar to step S05, calculate the absolute value of the maximum normalized frequency domain cross-correlation coefficient. .

[0114] For T s and P s Using weight vectors Weighted fusion is performed to achieve dimensionality reduction, resulting in the comprehensive feature coefficient A2 of time-frequency domain similarity. This embodiment uses Principal Component Analysis (PCA) to calculate the weight vector. The calculation steps are the same as in step S02. The comprehensive feature coefficient of time-frequency domain similarity is:

[0115]

[0116] Similarly, the time-frequency domain similarity comprehensive feature coefficients of window 1 are obtained. The result is as follows Figure 4 As shown.

[0117] Step S04: Calculate the harmonic content coefficient A3. Perform spectral analysis on the current signal y[n]. Calculate the harmonic energy in the frequency band from 2kHz to 5kHz. Fundamental wave energy The harmonic content coefficient is then:

[0118]

[0119] Similarly, the harmonic content coefficient of window 1 is obtained. The result is as follows Figure 5 As shown.

[0120] Step S05: Continuous periodic cumulative judgment of fault arc

[0121] S0501: Set feature decision thresholds. Based on the feature statistics of samples from normal and arc conditions, set decision thresholds. In this example, select thresholds T1=0.8, T2=0.9, and T3=0.005.

[0122] S0502: Output logic based on continuous periodic cumulative decisions. In this example, according to Clause 5.4 of Part 4 of the national standard GB14287.4—2014, a continuous decision counting threshold is set. The current window feature vector calculated above is compared with the threshold: , , Since all three conditions are met simultaneously, the detection window is determined to have arc characteristics, and the fault counter p is incremented by 1. If subsequent consecutive windows also meet the conditions, p continues to accumulate. When p ≥ P th When a continuous series fault arc occurs, the system determines that an alarm signal has been issued and the counter has been reset. If any window does not meet the conditions, the counter p is cleared to zero.

[0123] To further verify the effectiveness and beneficial effects of the method of the present invention, under the same experimental platform, the same sampling parameters, and the same test data conditions as described in Table 1, several existing typical series fault arc detection methods were selected as control methods for comparative experimental analysis. The control methods include: Method A, a single zero-time detection method that makes a decision based solely on the current zero-time proportionality coefficient; Method B, a traditional time-frequency domain multi-feature detection method that makes a joint decision based on multiple time-frequency domain feature parameters, including current kurtosis, high-frequency harmonic energy, and wavelet high-frequency component energy; and Method C, a single frequency domain detection method that makes a decision based on current harmonic content or total harmonic distortion rate.

[0124] All the above methods use the same sampling frequency, detection window length, sliding step size and decision time reference, and are tested on the same set of normal operating condition data and series fault arc operating condition data to ensure the fairness and comparability of the comparative test.

[0125] In the comparative experiments, the detection accuracy, false alarm rate, false negative rate, and average detection delay of different methods in the series fault arc detection process were statistically analyzed. The comparative test results of each method are shown in Table 2.

[0126]

[0127] The experimental results show that Method A is prone to misjudging zero-time features under complex load and changing operating conditions, resulting in high false alarm and false negative rates. Although Method B improves detection accuracy, its average detection delay increases significantly due to its high feature dimension and complex calculation process, making it unsuitable for implementation in terminal devices with limited computing resources, such as electricity meters. Method C is sensitive to load changes and harmonic interference, and its detection stability and reliability are insufficient.

[0128] The detection results are shown in Table 2. In comparison, the arc detection method proposed in this invention, which integrates zero-time and time-frequency features, achieves feature dimensionality reduction while ensuring physical consistency of features by constructing a zero-time-voltage joint feature coefficient A1, a time-frequency domain similarity comprehensive feature coefficient A2, and a specific frequency band harmonic content coefficient A3. Combined with a continuous periodic cumulative decision mechanism, it effectively suppresses the impact of instantaneous disturbances and load switching on the detection results. Under the same experimental conditions, the method of this invention significantly outperforms the control method in key performance indicators such as detection accuracy, false alarm rate, and false negative rate, while maintaining a low detection delay. This verifies the high reliability, low false alarm rate, and good engineering applicability of the method in series fault arc detection, effectively supporting the application requirements for meter burnout prevention and control.

[0129] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A zero-downtime and time-frequency characteristic fusion arc detection method for the prevention and control of meter burnout, characterized in that, Includes the following steps: S1. Synchronously acquire the current and voltage signals of the energy meter circuit and capture a detection window containing two consecutive power frequency cycles. S2. Based on the current signal and voltage signal within the detection window, calculate the zero-voltage joint characteristic coefficient A1, the time-frequency domain similarity comprehensive characteristic coefficient A2, and the harmonic content coefficient A3 for a specific frequency band. S3, set the threshold values ​​for the zero-voltage joint characteristic coefficient (T1), the time-frequency domain similarity comprehensive characteristic coefficient (T2), the harmonic content coefficient threshold value for a specific frequency band (T3), and the continuous period decision counting threshold (P). th For the feature vector [A1, A2, A3] calculated for the current detection window, if A1 < T1, A2 < T2, and A3 > T3 simultaneously, then it is determined that there is an arc feature in this window, and the fault counter p is incremented by 1; otherwise, the fault counter p is cleared to zero. When the fault counter p accumulates to the counting threshold P th When a continuous series fault arc is detected, a warning signal is triggered.

2. The arc detection method based on zero-downtime and time-frequency characteristic fusion for the prevention and control of meter burnout according to claim 1, characterized in that, In step S2, the zero-stable voltage joint characteristic coefficient A1 is calculated using the following method: Based on the current and voltage signals within the detection window, current waveform data x[n] and y[n] and their corresponding voltage waveform data u under normal operating conditions and the operating conditions to be detected are extracted respectively. x [n]、u y [n], where n = 0, 1, 2, ..., N-1, and N is the number of sampling points in one power frequency cycle; Calculate the zero-rest time proportionality coefficient I0 based on the current waveform data. The load voltage amplitude variation coefficient V is calculated based on the voltage waveform data. d V d With load voltage waveform distortion coefficient V s V s ; For I0, V d and V s Weighted fusion is performed to obtain the zero-voltage joint characteristic coefficient A1, which is calculated using the following formula: ； in, The preset weighting coefficients, .

3. The arc detection method based on zero-downtime and time-frequency characteristic fusion for the prevention and control of meter burnout according to claim 1, characterized in that, The calculation process for the zero-rest time ratio coefficient I0 includes: The threshold for determining zero current is set as a preset percentage (a%) of the effective value of the current waveform. Count the number P of sampling points with amplitudes lower than the threshold a% in the normal operating current waveform x[n] and the current waveform y[n] under test. x and P y ; Calculate zero-day work hours: ； Where f is the sampling frequency; Calculate the zero-rest time ratio coefficient: 。 4. The arc detection method based on zero-downtime and time-frequency characteristic fusion for the prevention and control of meter burnout according to claim 1, characterized in that, The load voltage amplitude variation coefficient V d The calculation formula is: ； Among them, U x U y These are the effective values ​​of the load voltage under normal operating conditions and fault arc conditions, respectively, derived from the corresponding voltage waveform data u. x [n]、u y [n] is calculated.

5. The arc detection method based on zero-downtime and time-frequency characteristic fusion for the prevention and control of meter burnout according to claim 1, characterized in that, The calculation process for the load voltage waveform distortion coefficient Vs is as follows: For the voltage waveform under normal operating conditions u x [n] and the voltage waveform under the test condition u y [n] are subjected to discrete Fourier transforms respectively to obtain its spectrum U x [k] and U y [k], k=0,1,2,…,N-1; Let the fundamental wave amplitudes be respectively , Calculate the total harmonic distortion rate: ； ； Calculate the load voltage waveform distortion coefficient: 。 6. The arc detection method based on zero-downtime and time-frequency characteristic fusion for the prevention and control of meter burnout according to claim 1, characterized in that, In step S2, the calculation method for the time-frequency domain similarity comprehensive feature coefficient A2 is as follows: Based on the current signal within the detection window, current waveform data x[n] and y[n] under normal operating conditions and the operating conditions to be detected are extracted, and the maximum cross-correlation coefficient T with normalized absolute value in the time domain is calculated. s The maximum cross-correlation coefficient P of the normalized absolute value in the frequency domain s ; For T s and P s Weighted fusion is performed to obtain the comprehensive feature coefficient A2 of time-frequency domain similarity, and its calculation formula is as follows: ; in, The preset weighting coefficients, .

7. The arc detection method based on zero-downtime and time-frequency characteristic fusion for the prevention and control of meter burnout according to claim 6, characterized in that, The maximum cross-correlation coefficient T of the time-domain normalized absolute value s The calculation process includes: Calculate the cross-correlation function of two current waveforms x[n] and y[n]: ； Where m is the sampling delay of the current waveform y relative to the current waveform x, and y* represents the complex conjugate value; Normalizing the cross-correlation function yields the normalized cross-correlation coefficient: ； Calculate the maximum cross-correlation coefficient of the normalized time domain absolute value: 。 8. The arc detection method based on zero-downtime and time-frequency characteristic fusion for the prevention and control of meter burnout according to claim 6, characterized in that, The frequency domain normalized absolute value maximum cross-correlation coefficient P s The calculation process includes: Perform discrete Fourier transforms on the current waveforms x[n] and y[n] respectively to obtain X(k) and Y(k); Calculate the frequency domain cross-correlation coefficient: ； After normalization, the normalized frequency domain cross-correlation coefficient is obtained: ； Calculate the normalized absolute value frequency domain maximum cross-correlation coefficient: 。 9. The arc detection method based on zero-downtime and time-frequency characteristic fusion for the prevention and control of meter burnout according to claim 1, characterized in that, In step S2, the calculation method for the harmonic content coefficient A3 of the specific frequency band is as follows: The discrete Fourier transform is performed on the current waveform data y[n] of the working condition to be detected within the detection window to obtain the frequency domain amplitude spectrum I(n f0), where f0 is the fundamental frequency; Take a specific frequency band [f a ,f b The sum of squares is calculated for the amplitudes of all discrete frequency points within the frequency band. ； Perform normalization and divide the resulting value by the fundamental frequency energy. The harmonic content coefficient A3 is obtained: 。 10. A zero-downtime and time-frequency characteristic fusion arc detection system for the prevention and control of meter burnout, characterized in that, include: The signal acquisition module is used to synchronously acquire the current and voltage signals of the energy meter circuit and capture a detection window containing two consecutive power frequency cycles. The feature calculation module is used to calculate the zero-voltage joint feature coefficient A1, the time-frequency domain similarity comprehensive feature coefficient A2, and the harmonic content coefficient A3 of a specific frequency band based on the current signal and voltage signal within the detection window. The decision output module is used to set the threshold values ​​for the zero-sleep-voltage joint characteristic coefficient (T1), the time-frequency domain similarity comprehensive characteristic coefficient (T2), the harmonic content coefficient for a specific frequency band (T3), and the continuous period decision count threshold (P). th For the feature vector [A1, A2, A3] calculated for the current detection window, if A1 < T1, A2 < T2, and A3 > T3 simultaneously, then it is determined that there is an arc feature in this window, and the fault counter p is incremented by 1; otherwise, the fault counter p is cleared to zero. When the fault counter p accumulates to the counting threshold P th When a continuous series fault arc is detected, a warning signal is triggered.

Images