A method for locating fault arc branches based on load response characteristics

Abstract

This invention proposes a fault arc branch location method based on load response characteristics, belonging to the field of electrical fault location. First, the current signal of the monitored power network is acquired, and potential electrical equipment in the network is identified. Normal current signals without fault arc information and fault current signals with fault arc information are filtered to determine the fault arc feature matrix and load markers for the normal and fault current signals at the current moment. The fault arc feature matrix and load markers for 10 power frequency cycles before and after the fault arc occur are recorded to form a time series. Based on Euclidean distance, the hyperplane closest to the time series is determined, and the load branch where the fault arc is located is determined using the results of a non-intrusive load decomposition method as constraints. This invention achieves low-cost, high-reliability, and accurate fault arc branch location without requiring strict synchronous sampling and additional signal injection, thus bridging the key technological gap between "arc detection" and "fault isolation."

Description

Technical Field

[0001] This invention belongs to the field of electrical fault location, and particularly relates to a method for locating fault arc branches, specifically a method for locating fault arc branches based on load response characteristics. Background Technology

[0002] With the advancement of urbanization, the situation of electrical fires in residential buildings is becoming increasingly serious. Older communities, in particular, suffer from severe problems such as aging electrical wiring, haphazard connections, and an increasing number of electrical appliances, leading to long-term overload of the original power distribution network and a sharp increase in electrical fire hazards. Among these hazards, fault arcs caused by aging and damaged insulation, loose electrical connections, humid air, and a sudden increase in voltage and current not only fail to generate significant heat but also produce high-temperature sputtering material that can easily ignite surrounding flammable and explosive materials, posing a significant risk of electrical fires in older communities.

[0003] Currently, scholars both domestically and internationally have conducted extensive research on fault arc diagnosis technology. The main methods include feature extraction based on time-frequency analysis, which detects changes in arc faults and related indicators by observing various frequency components and line current ranges; fault identification based on chaotic fractal theory, which identifies faults by calculating the correlation dimension of the current waveform and nonlinear features such as the Lyapunov exponent; and classification algorithms based on artificial intelligence, which autonomously learn and extract deep fault features of current signals in the time and frequency domains through deep convolutional neural networks or long short-term memory networks to achieve high-precision and strong generalization ability diagnosis of fault arcs.

[0004] However, in complex, multi-branch power distribution systems, simply detecting fault arcs is far from sufficient. When the main protection unit or centralized arc fault circuit breaker detects a fault arc in the system, quickly and accurately locating the specific branch where the fault occurred is crucial for achieving precise fault isolation, preventing unexplained power outages in non-faulty circuits, and ensuring power supply reliability and safety. This is also a core challenge currently facing the technology. Summary of the Invention

[0005] This invention aims to propose a fault arc branch location method based on load response characteristics. It utilizes the coupling relationship between the power disturbance and the dynamic current response characteristics of each branch when a fault arc occurs to design a fault arc load identification algorithm. Without the need for strict synchronous sampling and additional signal injection, it achieves low-cost and high-reliability accurate fault arc branch location, thus bridging the key technical gap between "arc detection" and "fault isolation".

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0007] A method for locating fault arc branches based on load response characteristics includes the following steps:

[0008] Step 1: Acquire the current signal of the power network to be monitored at a time scale of 0.02 seconds, and identify potential electrical equipment in the power network to be monitored based on the non-intrusive load decomposition method;

[0009] Step 2: Use a multi-branch fault arc detection algorithm to filter normal current signals without fault arc information and fault current signals with fault arc information, and determine the fault arc feature matrix of the normal current signal at the current moment. and load marking Fault arc characteristic matrix of fault current signal and load marking ;

[0010] Step 3: Update the fault arc characteristic matrix and load markers of the current signal of the monitored power network in real time, and record the fault arc characteristics for 10 power frequency cycles before and after the fault arc occurs. , With 10 power frequency cycles , The time series of fault arc feature matrix and load marking are formed. ;

[0011] Step 4: Determine the relationship with the time series based on Euclidean distance. The nearest hyperplane, constrained by the results of a non-intrusive load decomposition method, determines the load branch where the fault arc is located.

[0012] As a preferred embodiment of the present invention, in step 1, the identification of the electrical equipment in the power network to be monitored is completed by calculating the time-series changes in the effective value of the current signal and the kurtosis of the local waveform. The specific steps are as follows:

[0013] The current signal of the power network to be monitored is acquired at a time scale of 0.02 seconds; the effective value of the current signal in each power frequency cycle is calculated. With local waveform kurtosis Extract valid values ​​within a 1-second time window. With local waveform kurtosis Two-dimensional time series The input is fed into the trained LSTM model to complete the load decomposition of the power network to be monitored and to identify the potential power devices in the power network to be monitored.

[0014] Further, in step 1, the effective value of the current signal The calculation formula is:

[0015]

[0016] in, This is the index value of the sampled current sequence. The length of the sampled current sequence;

[0017] The local waveform kurtosis The calculation includes the following steps:

[0018] S11: Calculate the derivative sequence of the current signal. ;

[0019] S12: Determine the derivative sequence of the current signal The index of the maximum value index value of the minimum value ;

[0020] S13: For current signals Gain processing is performed on the sequence to form a new current signal sequence. ,in:

[0021]

[0022] S14: Based on the new current signal sequence Calculate local waveform kurtosis:

[0023]

[0024] in, for The mean of the sequence, for Standard deviation of the sequence.

[0025] As a preferred embodiment of the present invention, in step 2, the multi-branch fault arc detection algorithm samples the high-order statistical features of the current signal in the time-frequency domain that are not affected by the background load as the fault arc feature matrix, and uses it as the basis for determining whether a fault arc has occurred; the extraction of the high-order statistical feature matrix of the current signal in the time-frequency domain includes the following steps:

[0026] S21: Select the sym4 wavelet function as the basis function to perform two-level wavelet decomposition on the acquired current signal to obtain the first-level wavelet detail coefficients. With the second layer wavelet detail coefficients ;

[0027] In the wavelet decomposition of the current signal, this invention employs an adjacent period truncation scheme to curb the non-convergence behavior of the signal boundary during the wavelet decomposition process. Specifically, during the wavelet transform of the current signal in the current period, this invention truncates 1 / 10 of the length of the current signal in the next period to supplement the current signal in the current period, and after the wavelet transform is completed, the target signal is correspondingly truncated to obtain the first-level wavelet detail coefficients. With the second layer wavelet detail coefficients To minimize the boundary non-convergence behavior caused by wavelet transform.

[0028] S22: Calculate the kurtosis of the first-level wavelet detail coefficients kurtosis of the second-level wavelet detail coefficients ,as well as and Coupling coefficient ;

[0029] S23: Calculate the skewness of the first-level wavelet detail coefficients Skewness with the second-level wavelet detail coefficients ,as well as and Coupling coefficient ;

[0030] S24: Calculate the coefficient of variation of the first-level wavelet detail coefficients. The coefficient of variation of the second-level wavelet detail coefficients ,as well as and Coupling coefficient ;

[0031] S25: Will As a feature matrix of the fault arc, it is uploaded to the multi-branch fault arc detection system to determine whether the current signal at the current moment carries fault arc information;

[0032] S26: Construct load markers that are strongly correlated with the severity of the fault arc based on the changes in load current signals in the low-frequency and high-frequency spaces before and after the occurrence of the fault arc. Its specific expression is:

[0033] .

[0034] Furthermore, in step S24 above:

[0035] cliff , cliff The calculation formulas are as follows:

[0036]

[0037]

[0038] Skewness skewness The calculation formulas are as follows:

[0039]

[0040] .

[0041] Due to the quasi-periodic oscillation of the wavelet detail coefficients of the current signal near zero... and The mean exhibits significant randomness. To address this issue, this invention makes targeted adjustments to the mathematical processing of the traditional coefficient of variation. Instead of using the mean in the coefficient of variation, it modifies the processing of the absolute value of the original signal to reduce the randomness of the features; this is called a quasi-coefficient of variation. Class of coefficients of variation The calculation formulas are as follows:

[0042]

[0043] .

[0044] Extracting high-order statistical features from the wavelet detail coefficients of the current signal can effectively reduce the interference of background load on the fault arc features. Coupled processing of the double-layer wavelet detail coefficient features can greatly reduce the interference of noise signals and ensure the convergence of the hyperplane of the marked load used in the subsequent fault arc branch location method.

[0045] As a preferred technical solution of the present invention, in step 4, the hyperplane representing the functional relationship between the fault arc feature matrix and the load mark under different loads is determined based on experimental data, and the branch location of the fault arc is completed by determining the hyperplane with the closest time series distance composed of the fault arc feature matrix and the load.

[0046] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0047] 1. This invention utilizes a non-intrusive load decomposition method to determine the potential electrical load in the monitored power network, thereby clarifying the boundary parameters of the fault arc branch location method. This not only enables the fault arc branch location method to present a more intuitive alarm effect but also reduces the risk of false alarms.

[0048] 2. This invention, based on a multi-branch fault arc detection algorithm, filters fault arc feature matrices and load markers in both normal and fault states of the monitored power network. This not only allows for branch location of fault arcs through the distance relationship between discrete time point groups and the hyperplane representing the load, but also assesses the disaster risk caused by fault arcs based on the changing patterns of fault arc feature values ​​and load marker values ​​over time, thereby further improving the reliability of the multi-branch fault arc detection algorithm.

[0049] 3. The multi-branch fault arc detection algorithm, non-intrusive load decomposition method and fault arc branch location method adopted in this invention complement each other and can better assist the intelligent decision-making of electrical fire monitoring system, and reduce the time for electrical fire hazard investigation and treatment. Attached Figure Description

[0050] Figure 1 This is a flowchart of a fault arc branch location method based on load response characteristics. Detailed Implementation

[0051] The present invention will be further described in detail below with reference to the embodiments and accompanying drawings.

[0052] This invention, based on computational fault arc characteristics, conducts fault arc experiments under different loads. By calculating the fault arc characteristics over a certain time period, corresponding data is obtained. A multi-branch fault arc detection algorithm is trained using machine learning and then ported to an embedded system, forming a multi-branch fault arc detection system. During the construction of the multi-branch fault arc detection system, current signals are experimentally acquired, the aforementioned characteristics are calculated, and a hyperplane representing the functional relationship between the fault arc characteristic matrix under different loads and the load markers is determined based on the test data. In other words, the obtained experimental data is used to train the model, thereby obtaining the multi-branch fault arc detection system, which is then used for actual detection.

[0053] like Figure 1 As shown in the figure, the fault arc branch location method based on load response characteristics proposed in this embodiment includes the following steps:

[0054] Step 1: Acquire the current signal of the power network to be monitored at a time scale of 0.02 seconds and calculate its effective value. .

[0055] Step 2: Calculate the derivative sequence of the current signal. Determine the derivative sequence of the current signal. The index of the maximum value index value of the minimum value .

[0056] Step 3: Process the current signal Gain processing is performed on the sequence to form a new current signal sequence. Based on the new current signal sequence Calculate local waveform kurtosis .

[0057] Step 4: Extract a two-dimensional time series of effective values ​​and local waveform kurtosis within a 1-second time window. The input is fed into the trained LSTM model to complete the load decomposition of the power network to be monitored and to identify the potential power devices in the power network to be monitored.

[0058] Step 5: Select the sym4 wavelet function as the basis function to perform two-level wavelet decomposition on the acquired current signal to obtain the first-level wavelet detail coefficients. With the second layer wavelet detail coefficients .

[0059] Step 6: Wavelet detail coefficients based on current signal and Calculate the fault arc characteristic matrix The fault arc feature matrix is ​​uploaded to the multi-branch fault arc detection system to determine whether the current signal at the current moment carries fault arc information.

[0060] Step 7: Construct load markers that are strongly correlated with the intensity of the fault arc based on current signals. .

[0061] Step 8: Based on the multi-branch fault arc detection algorithm, filter current signals that do not carry fault arc information to determine the fault arc feature matrix of the normal current signal at the current moment. and load marking Based on a multi-branch fault arc detection algorithm, the fault arc feature matrix of the fault current signal at the current moment is determined by filtering the current signals carrying fault arc information. and load marking .

[0062] Step 9: Record the power frequency cycles for 10 cycles before and after the occurrence of the fault arc. , With 10 power frequency cycles , The time series of fault arc feature matrix and load marking are formed. The time series of the fault arc feature matrix and load markers are determined based on Euclidean distance. The most recent hyperplane was used to locate the faulty arc branch.

Claims

1. A method for locating fault arc branches based on load response characteristics, characterized in that, Includes the following steps: Step 1: Acquire the current signal of the power network to be monitored at a time scale of 0.02 seconds, and identify potential electrical equipment in the power network to be monitored based on the non-intrusive load decomposition method; Step 2: Use a multi-branch fault arc detection algorithm to filter normal current signals without fault arc information and fault current signals with fault arc information, and determine the fault arc feature matrix of the normal current signal at the current moment. and load marking Fault arc characteristic matrix of fault current signal and load marking ; The multi-branch fault arc detection algorithm samples the high-order statistical features of the current signal in the time-frequency domain, which are not affected by the background load, as the fault arc feature matrix and serves as the basis for determining whether a fault arc has occurred. The extraction of the high-order statistical feature matrix of the current signal in the time-frequency domain includes the following steps: S21: Select the sym4 wavelet function as the basis function to perform two-level wavelet decomposition on the acquired current signal to obtain the first-level wavelet detail coefficients. With the second layer wavelet detail coefficients The two-layer wavelet decomposition method involves: during the wavelet transform of the current signal in the current cycle, truncating 1 / 10 of the length of the current signal in the next cycle to supplement the current signal in the current cycle; and after the wavelet transform, correspondingly pruning the target signal to obtain the first-layer wavelet detail coefficients. With the second layer wavelet detail coefficients ; S22: Calculate the kurtosis of the first-level wavelet detail coefficients kurtosis of the second-level wavelet detail coefficients ,as well as and Coupling coefficient ; cliff , cliff The calculation formulas are as follows: S23: Calculate the skewness of the first-level wavelet detail coefficients Skewness with the second-level wavelet detail coefficients ,as well as and Coupling coefficient ; Skewness skewness The calculation formulas are as follows: S24: Calculate the coefficient of variation of the first-level wavelet detail coefficients. The coefficient of variation of the second-level wavelet detail coefficients ,as well as and Coupling coefficient ; coefficient of variation Class of coefficients of variation The calculation formulas are as follows: S25: Will As a feature matrix of the fault arc, it is uploaded to the multi-branch fault arc detection system to determine whether the current signal at the current moment carries fault arc information; S26: Construct load markers that are strongly correlated with the severity of the fault arc based on the changes in load current signals in the low-frequency and high-frequency spaces before and after the occurrence of the fault arc. Its specific expression is: Step 3: Update the fault arc characteristic matrix and load markers of the current signal of the monitored power network in real time, and record the fault arc characteristics for 10 power frequency cycles before and after the fault arc occurs. , With 10 power frequency cycles , The time series of fault arc feature matrix and load marking are formed. ; Step 4: Determine the relationship with the time series based on Euclidean distance. The nearest hyperplane, constrained by the results of a non-intrusive load decomposition method, determines the load branch where the fault arc is located.

2. The fault arc branch location method based on load response characteristics as described in claim 1, characterized in that, In step 1, the identification of electrical equipment in the monitored power network is completed by calculating the time-series changes in the effective value of the current signal and the kurtosis of the local waveform. The specific steps are as follows: The current signal of the power network to be monitored is acquired at a time scale of 0.02 seconds; the effective value of the current signal in each power frequency cycle is calculated. With local waveform kurtosis Extract valid values ​​within a 1-second time window. With local waveform kurtosis Two-dimensional time series The input is fed into the trained LSTM model to complete the load decomposition of the power network to be monitored and to identify the potential power devices in the power network to be monitored.

3. The fault arc branch location method based on load response characteristics as described in claim 2, characterized in that, The effective value of the current signal The calculation formula is: in, This is the index value of the sampled current sequence. The length of the sampled current sequence; The local waveform kurtosis The calculation includes the following steps: S11: Calculate the derivative sequence of the current signal. ; S12: Determine the derivative sequence of the current signal The index of the maximum value index value of the minimum value ; S13: For current signals Gain processing is performed on the sequence to form a new current signal sequence. ,in: S14: Based on the new current signal sequence Calculate local waveform kurtosis: in, for The mean of the sequence, for Standard deviation of the sequence.

4. The fault arc branch location method based on load response characteristics as described in claim 1, characterized in that, In step 4, the hyperplane representing the functional relationship between the fault arc feature matrix and the load marker under different loads is determined based on experimental data. The branch location of the fault arc is completed by determining the hyperplane with the closest time series distance composed of the fault arc feature matrix and the load.

Images