AC fault arc detection method and system

Through empirical modal decomposition and CNN-LSTM hybrid neural network, the Hurst index, interharmonic variance and wavelet energy entropy characteristics of aviation cables are extracted, and the problems of low accuracy and high calculation cost in traditional methods are solved in identifying arc faults, and the problem of low accuracy and high calculation cost is achieved. Quick and accurate detection and circuit protection of aviation cable AC faults.

CN114609475BActive Publication Date: 2024-11-08CIVIL AVIATION UNIV OF CHINA
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202210277340.3
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-03-18
Publication Date
2024-11-08
Estimated Expiration
2042-03-18

AI Technical Summary

Technical Problem

Traditional aeronautical cable AC fault arc detection methods are difficult to identify arc faults in specific environments, especially when the cable connection is loose or the water beads between lines condense, the voltage and current amplitude is lower than the threshold, resulting in low recognition accuracy and complex neural network structure calculations. The cost is high, making it difficult to achieve industrial online testing.

Method used

The empirical modal decomposition algorithm is used to decompose the current signal into modal components of each order. The three-dimensional characteristic data is extracted through Hurst index, interharmonic variance and wavelet energy entropy, and a CNN-LSTM hybrid neural network is constructed for fault detection, combined with correlation coefficient selection Suitable modal components reconstruct current signals for fast and accurate fault identification.

Benefits of technology

It improves the accuracy and robustness of AC arc fault detection in aviation cables, simplifies the neural network structure, realizes rapid fault recognition, and provides effective circuit protection actions.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114609475B_ABST
    Figure CN114609475B_ABST
Patent Text Reader

Abstract

The present invention discloses an AC fault arc detection method and system, including: obtaining a discrete current signal in an aviation cable, using an empirical mode decomposition algorithm to decompose the discrete current signal into various order modal components, and reconstructing the current signal, dividing the reconstructed current signal by period, obtaining the Hurst index, interharmonic variance, and wavelet energy entropy three-dimensional feature data of each period, constructing a CNN-LSTM hybrid neural network, and inputting the Hurst index, interharmonic variance, and wavelet energy entropy three-dimensional features into the CNN-LSTM hybrid neural network to realize AC arc fault detection. The method can enhance the AC arc fault recognition capability of aviation cables, improve the accuracy and robustness of arc fault detection, and the structure of the CNN-long short-term memory neural network LSTM is simpler, making the arc fault recognition speed faster.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent detection of fault arcs in aviation power distribution systems, and more specifically to an alternating current fault arc detection method and system. Background Art

[0002] Aviation cables are the main components of electrical wire interconnection systems and are mainly used for power transmission and signal transmission. As aircraft work in complex environments such as vibration, friction, humidity, and temperature changes for a long time, cables are prone to cracks and breakage, and the electrical connections of cables are prone to loosening, which can easily lead to the occurrence of AC fault arcs. When an AC arc occurs, the temperature at the center of the arc can reach 5000-15000℃. If it is not identified and detected in time and protective measures are not taken, the fault arc can easily cause cable combustion or even fire. Therefore, it is of great significance to study an efficient, simple and accurate AC fault arc identification method and apply it to the aviation power distribution system, timely identify arc faults and perform effective circuit protection, prevent cable combustion, fire, and protect people's lives and property.

[0003] Traditional thermal protection circuit breakers and solid-state power controllers detect arcs by setting thresholds. Once the threshold is fixed, fault arcs with voltage and current amplitudes exceeding the threshold can be detected and circuit protected. However, in certain environments, loose cable connections or condensation of water droplets between wires can cause the voltage and current amplitudes to be lower than the fixed threshold when a fault arc occurs, making it difficult to effectively identify arc faults by simply adjusting the threshold. In addition, the loads connected to aviation cables are complex and may be switched at any time, which may also cause changes in the voltage and current amplitudes. In addition, traditional neural networks have complex structures, high computational costs, and low recognition accuracy, making it difficult to realize the application of rapid online detection of AC fault arcs in industrial aviation cables. Summary of the invention

[0004] An embodiment of the present invention provides an AC fault arc detection method, comprising:

[0005] Obtain discrete current signals in aviation cables;

[0006] The empirical mode decomposition algorithm is used to decompose the discrete current signal into various modal components and reconstruct the current signal;

[0007] The reconstructed current signal is divided into periods to obtain the three-dimensional characteristic data of Hurst index, interharmonic variance and wavelet energy entropy for each period;

[0008] Construct a CNN-LSTM hybrid neural network;

[0009] The three-dimensional features of Hurst exponent, interharmonic variance and wavelet energy entropy are input into the CNN-LSTM hybrid neural network to realize AC arc fault detection.

[0010] Furthermore, the step of selecting a suitable modal component to reconstruct the current signal by comparing the normalized correlation coefficient between each modal component and the original discrete current signal with a preset threshold value includes:

[0011] Initialize counter i=1 and set value r 预设 , number of cycles i max ;

[0012] Correlate each modal component with the original discrete current signal

[0013]

[0014] Among them, IMFi represents each modal component of the discrete current signal, Represents a discrete current signal;

[0015] Comparison i With r 预设 , if r i ≥r 预设 , then save r i Otherwise, the counter i is incremented by one and r is recalculated. i , until i reaches the cycle number i max ;

[0016] The stored modal components are added together to obtain a reconstructed current signal;

[0017] Furthermore, the Hurst exponential characteristic calculation steps include:

[0018] Divide x(i) into M subsequences of length n and calculate the mean of the subsequences Wherein, x(i) is the time series of length N divided into several periods by the reconstructed current signal;

[0019] Calculate the mean adjusted series D i,m =x(i,m)-μ n,m ,i=1,2,...,n;

[0020] Calculate the cumulative deviation of the subsequence i=1,2,...,n;

[0021] Calculate the subsequence range

[0022] Calculate the standard deviation of a subsequence

[0023] Calculate each rescaling range R d / S d and the average of all M subsequences rescaled ranges

[0024] Compare n with N / 10. If n>N / 32, pass The linear fitting estimates the Hurst exponent, and the operation is completed, where R represents the subsequence range, S represents the standard deviation of the subsequence, H represents the Hurst exponent, and C represents a constant;

[0025] Otherwise, set n=n / 2 and return to continue the calculation.

[0026] Furthermore, the interharmonic variance feature calculation step includes:

[0027] Compute the discrete Fourier transform of the time series x(i)

[0028] Select the intermediate harmonic components of X(k) to form a new discrete Fourier transform set Among them, β is the upper limit of the frequency space considered;

[0029] Calculate the interharmonic variance of time series in Represents the interharmonic mean of the time series.

[0030] Furthermore, the wavelet energy entropy feature calculation steps include:

[0031] Calculate the energy E of the i-th sampling point of the time series x(i) at the a-th scale a (i);

[0032] Calculate the total energy of the time series x(i) at scale a

[0033] Calculate the total energy of all scales of the time series x(i)

[0034] Calculate p a =E(a) / E;

[0035] Calculate the energy entropy of the wavelet at scale a Where m represents the number of dimension layers under wavelet transform.

[0036] Furthermore, it also includes testing and training the CNN-LSTM hybrid neural network, the steps of which include:

[0037] The three-dimensional feature data sets of Hurst index, interharmonic variance, and wavelet energy entropy are divided into training set and test set at a ratio of 8:2;

[0038] The grid search method is used to find the optimal initial learning rate, batch size and other parameters, determine the number of iterations W of the CNN-LSTM hybrid neural network, and initialize the number of training times v = 1;

[0039] The convolutional neural network (CNN) is used to train the model and extract the characteristic information of the three-dimensional feature data of Hurst index, interharmonic variance, and wavelet energy entropy.

[0040] Use long short-term memory neural network LSTM to learn feature information;

[0041] Use error back propagation to update the parameters of CNN-LSTM hybrid neural network;

[0042] Compare the number of training times v and W. If v>W, then end the training;

[0043] Otherwise, continue iterating the training operation;

[0044] Input the test set, perform fault arc identification, and output the detection results.

[0045] The present invention also provides an AC fault arc detection system, comprising:

[0046] The signal conditioning module is used to collect the current signal in the aviation cable through the current transformer and input it into the signal conditioning module to collect and amplify the current signal, and then perform analog-to-digital conversion through the analog-to-digital converter to obtain a discrete current signal;

[0047] The interference filtering module is used to decompose the current signal in discrete form into modal components IMF of each order through the empirical mode decomposition algorithm, and then select the modal component to reconstruct the current according to the normalized correlation coefficient and the preset coefficient threshold value, and divide the obtained reconstructed current signal by period, and successively obtain the Hurst index, interharmonic variance, and wavelet energy entropy three-dimensional feature data of the current signal of each period;

[0048] The fault arc identification module is used to input the three-dimensional feature data of Hurst index, interharmonic variance and wavelet energy entropy into the fault identification module, and train and test the CNN-LSTM hybrid neural network to realize AC arc fault detection.

[0049] Furthermore, it also includes:

[0050] Aviation power supply;

[0051] AC arc fault circuit breaker;

[0052] Cable under test;

[0053] Current transformer;

[0054] Aviation load;

[0055] The aviation power supply is connected to one end of an AC fault arc circuit breaker, the AC fault arc circuit breaker is connected to a cable to be tested, a line of the cable to be tested passes through a current transformer, and the cable to be tested is connected to an aviation load.

[0056] The embodiment of the present invention provides an AC fault arc detection method, which has the following beneficial effects compared with the prior art:

[0057] The beneficial effects of the aviation cable AC fault arc identification method and system provided by the present invention are: acquiring fault information from the three dimensions of HIW can enhance the aviation cable AC arc fault identification capability, improve the arc fault detection accuracy and robustness, and the structure of the CNN-long short-term memory neural network LSTM is simpler, making the arc fault identification speed faster, thereby providing an effective basis for the next circuit protection action. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 Flowchart of intelligent identification of AC fault arc based on CNN-Long Short-Term Memory Neural Network LSTM;

[0059] Figure 2 It is a schematic diagram of the structure of the aviation cable AC fault arc identification system used in the method of the present invention;

[0060] Figure 3 Schematic diagram of the internal structure of the current transformer;

[0061] Figure 4 Interference filtering flow chart based on modal decomposition;

[0062] Figure 5 HIW 3D feature extraction flow chart;

[0063] Figure 6 The original current waveform of the example test circuit under linear load with arc fault;

[0064] Figure 7 (a)-(c) are three-dimensional characteristic waveform diagrams of the example test current signal HIW;

[0065] Figure 8 This is the absolute error curve of the training of the example test current signal based on the CNN-LSTM network;

[0066] Fig. 9 This is the AC arc fault detection accuracy curve of the method of the present invention. DETAILED DESCRIPTION

[0067] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0068] See also Figures 1 to 9 The embodiment of the present invention provides an AC fault arc detection method, the method comprising:

[0069] The aviation cable AC fault arc detection method provided by the present invention comprises the following steps performed in sequence:

[0070] Step 1) construct an aviation cable AC fault arc detection system; Figure 2 As shown, the fault detection system includes a system to be tested 1 and a host computer 2; wherein: the system to be tested 1 includes an aviation power supply 3, an AC fault arc circuit breaker 4, a cable to be tested 5, a current transformer 6 and an aviation load 7, wherein the aviation power supply 3 is connected to one end of the AC fault arc circuit breaker 4, the AC fault arc circuit breaker 4 is connected to the cable to be tested 5, a line of the cable to be tested 5 passes through the current transformer 6, and the cable to be tested 5 is connected to the aviation load 7; the host computer 2 includes a signal conditioning module 8, an interference filtering module 9 and a fault arc identification module 10, the output of the current transformer 6 is connected to the input of the signal conditioning module 8, the output of the signal conditioning module 8 is connected to the input end of the interference filtering module 9, the output of the interference filtering module 9 is connected to the input of the fault arc identification module 10, and the output of the fault arc identification module 10 is connected to one end of the AC fault arc circuit breaker 4.

[0071] The aviation power supply 3 is an AC power supply of 115V / 400Hz, and the current transformer adopts an industrial-grade clamp-on transformer, allowing the cable to be tested to pass through the center of the current transformer.

[0072] Step 2) The current signal in the aviation cable is collected by the current transformer 6. The internal structure of the current transformer 6 is as follows: Figure 3 As shown, the cable passes through the current transformer 6, and a sinusoidal AC signal flows through the cable to generate a changing magnetic field. When the clamping joint 201 of the current transformer 6 is closed, the coil 202 generates current and inputs the signal conditioning module 8 through the output terminal 203 to realize the collection and amplification of the current signal, and then performs analog-to-digital conversion through the analog-to-digital converter to obtain a discrete current signal.

[0073] Step 3) The specific method of interfering with the filtering module 9 is as follows Figure 4 As shown, the current signal is amplified by the signal conditioning module 8, and then input into the interference filtering module 9 for decomposition using the empirical mode decomposition algorithm, and the appropriate mode component is selected to reconstruct the current signal by comparing the normalized correlation coefficient of each mode component with the original signal and the preset threshold value to achieve interference filtering. The specific process is as follows:

[0074] ① Initialize i=1 and set the preset value r according to the actual situation 预设 , number of cycles i max ;

[0075] ② Perform empirical mode decomposition on the current signal to obtain a series of modal components;

[0076] ③ Perform correlation operation on each modal component and the original signal Get r i ;

[0077] ④ r i With r 预设 Compare. i ≥r 预设 , then save r i In c, otherwise, increment timer i by 1 and return to ③ to continue the calculation until i reaches the number of loops i max ;

[0078] ⑤Add the modal components in c to obtain the reconstructed signal y.

[0079] Step 4) The reconstructed current signal y obtained in step 3 is divided into a number of time series x(i) of length N according to the period, and the three-dimensional features of Hurst exponent, intermittent wave equation and wavelet energy entropy are extracted from the current signal of each period in turn. The process is as follows: Figure 5 As shown, the specific process is as follows:

[0080] Hurst index characteristics:

[0081] ① Divide x(i) into M subsequences of length n and calculate the mean of the subsequences

[0082] ② Calculate the mean adjusted sequence D i,m =x(i,m)-μ n,m ,i=1,2,...,n;

[0083] ③Calculate the cumulative deviation of the subsequence i=1,2,...,n;

[0084] ④Calculate the subsequence range

[0085] ⑤Calculate the standard deviation of the subsequence

[0086] ⑥ Calculate each rescaling range R d / S d and the average of all M subsequences rescaled ranges

[0087] ⑦ Compare n with N / 10. If n>N / 32, pass The linear fitting estimates H and the operation ends; otherwise, set n = n / 2 and return to ② to continue the operation.

[0088] Interharmonic variance characteristics:

[0089] ① Calculate the discrete Fourier transform of the time series x(i)

[0090] ② Select the intermediate harmonic components of X(k) to form a new discrete Fourier transform set Where β is the upper limit of the frequency space considered;

[0091] ③Calculate the harmonic variance between time series

[0092] Wavelet energy entropy characteristics:

[0093] ① Calculate the energy E of the i-th sampling point of the time series x(i) at the a-th scale a (i);

[0094] ② Calculate the total energy of the time series x(i) at scale a

[0095] ③ Calculate the total energy of all scales of the time series x(i)

[0096] ④ Calculate p a =E(a) / E;

[0097] ⑤Finally, calculate the energy entropy of the wavelet at scale a

[0098] Step 5) The calculation result obtained in step 4 is input into the fault arc identification module 10, and the CNN-LSTM (convolutional neural network combined with long short-term memory neural network) hybrid neural network is trained and tested to realize AC arc fault detection. The detection flow chart is as follows: Figure 1 As shown, the specific detection process is as follows:

[0099] ① Divide the feature data set into a training set and a test set in an 8:2 ratio, and label the data set;

[0100] ② Use the grid search method to find the optimal initial learning rate, batch size and other parameters, determine the number of algorithm iterations W of the recognition module, and initialize the number of training times v = 1;

[0101] ③ Use convolutional neural network (CNN) to train the model and extract the feature information of HIW three-dimensional feature data;

[0102] ④ Use long short-term memory neural network LSTM to learn feature information;

[0103] ⑤Use error back propagation to update network parameters;

[0104] ⑥Compare the number of training times v and W. If v>W, then end the training. Otherwise, increase the number of iterations by one and return to ③ to continue the training operation.

[0105] ⑦ Input the test set, perform fault arc identification, and output the detection results;

[0106] ⑧ Input the detection result of the fault arc identification module 10 into the circuit breaker 4 for action control, and the operation is completed.

[0107] Select as Figure 6 The example test current signal including normal and arc fault is shown in Figure 1. The HIW three-dimensional features are extracted using the method in step 4. Figure 7 (a)-(c) show, and the HIW three-dimensional features are input into the CNN-LSTM hybrid neural network in step 5 for training and testing. The absolute error of network training is shown in Figure 8 The test results are shown in Fig. 9 The experimental results show that the method of the present invention can effectively, quickly and accurately identify and detect the AC fault arc of aviation cables, and its network structure is simpler and the control performance is better.

[0108] Disclosed above are only a few specific embodiments of the present invention. Those skilled in the art may make various changes and modifications to the embodiments of the present invention without departing from the spirit and scope of the present invention. However, the embodiments of the present invention are not limited thereto. Any changes that can be conceived by those skilled in the art should fall within the scope of protection of the present invention.

Claims

1. A method for detecting an AC fault arc, characterized in that: include: Obtain discrete current signals in aviation cables; The empirical mode decomposition algorithm is used to decompose the discrete current signal into various modal components and reconstruct the current signal; The reconstructed current signal is divided into periods to obtain the three-dimensional characteristic data of Hurst index, interharmonic variance and wavelet energy entropy for each period; Construct a CNN-LSTM hybrid neural network; The three-dimensional features of Hurst index, interharmonic variance and wavelet energy entropy are input into the CNN-LSTM hybrid neural network to realize AC arc fault detection; The step of reconstructing the current signal comprises: Initialize counter i=1 and preset value r 预设 , number of cycles i max ; Each modal component of the discrete current signal decomposed by the empirical mode decomposition algorithm is correlated with the normalized correlation coefficient of the discrete current signal to obtain the normalized correlation coefficient r i : Among them, IMFi represents each modal component of the discrete current signal, Represents a discrete current signal; Comparison i With r 预设 , if r i ≥r 预设 , then save r i Otherwise, the counter i is incremented by one and r is recalculated. i , until i reaches the cycle number i max ; The stored modal components are added together to obtain a reconstructed current signal; It also includes testing and training the CNN-LSTM hybrid neural network, the steps of which include: The three-dimensional feature data sets of Hurst index, interharmonic variance, and wavelet energy entropy are divided into training set and test set at a ratio of 8:2; The grid search method is used to find the optimal initial learning rate and batch size parameters, determine the number of iterations W of the CNN-LSTM hybrid neural network, and initialize the number of training times v = 1; The convolutional neural network (CNN) is used to train the model and extract the characteristic information of the three-dimensional feature data of Hurst index, interharmonic variance, and wavelet energy entropy. Use long short-term memory neural network LSTM to learn feature information; Use error back propagation to update the parameters of CNN-LSTM hybrid neural network; Compare the number of training times v and W. If v>W, then end the training; Otherwise, continue iterating the training operation; Input the test set to perform arc fault identification.

2. An AC fault arc detection method as claimed in claim 1, characterized in that: The Hurst exponential characteristic calculation step includes: Divide x(i) into M subsequences of length n and calculate the mean of the subsequences Wherein, x(i) is the reconstructed current signal divided into several time series of length N according to the period; Calculate the subsequence mean adjustment sequence D i,m =x(i,m)-μ n,m ,i=1,2,...,n; Calculate the cumulative deviation of the subsequence i=1,2,...,n; Calculate the subsequence range Calculate the standard deviation of a subsequence Calculate each rescaling range R d / S d and the average of all M subsequences rescaled ranges Compare n with N / 10. If n>N / 32, pass The linear fitting estimates the Hurst exponential characteristics, and the operation is completed, where R represents the subsequence range, S represents the standard deviation of the subsequence, H represents the Hurst exponent, and C represents the constant; Otherwise, set n=n / 2 and return to continue the calculation.

3. An AC fault arc detection method as claimed in claim 2, characterized in that: The interharmonic variance feature calculation step comprises: Compute the discrete Fourier transform of the time series x(i) Select the intermediate harmonic components of X(k) to form a new discrete Fourier transform set Among them, β is the upper limit of the frequency space considered; Calculate the interharmonic variance of time series in Represents the interharmonic mean of the time series.

4. The AC fault arc detection method according to claim 2, characterized in that: The wavelet energy entropy feature calculation step includes: Calculate the energy E of the i-th sampling point of the time series x(i) at the a-th scale a (i); Calculate the total energy of the time series x(i) at scale a Calculate the total energy of all scales of the time series x(i) Calculate p a =E(a) / E; Calculate the energy entropy of the wavelet at scale a Where m represents the number of dimensions under wavelet transform.

5. An AC fault arc detection system, characterized in that: include: The signal conditioning module is used to collect the current signal in the aviation cable through the current transformer and input it into the signal conditioning module to collect and amplify the current signal, and then perform analog-to-digital conversion through the analog-to-digital converter to obtain a discrete current signal; The interference filtering module is used to decompose the current signal into modal components of various orders through the empirical mode decomposition algorithm, and then select the modal components to reconstruct the current according to the normalized correlation coefficient and the preset coefficient threshold value, and divide the obtained reconstructed current signal by period, and successively obtain the Hurst index, interharmonic variance, and wavelet energy entropy three-dimensional feature data of the current signal of each period; The fault arc identification module is used to input the three-dimensional feature data of Hurst index, interharmonic variance, and wavelet energy entropy into the fault identification module, and train and test the CNN-LSTM hybrid neural network to realize AC arc fault detection; The reconstructed current signal comprises: Initialize counter i=1 and preset value r 预设 , number of cycles i max ; Each modal component of the discrete current signal decomposed by the empirical mode decomposition algorithm is correlated with the normalized correlation coefficient of the discrete current signal to obtain the normalized correlation coefficient r i : Among them, IMFi represents each modal component of the discrete current signal, Represents a discrete current signal; Comparison i With r 预设 , if r i ≥r 预设 , then save r i Otherwise, the counter i is incremented by one and r is recalculated. i , until i reaches the cycle number i max ; The stored modal components are added together to obtain a reconstructed current signal; Also included is testing and training of CNN-LSTM hybrid neural networks, including: The three-dimensional feature data sets of Hurst index, interharmonic variance, and wavelet energy entropy are divided into training set and test set at a ratio of 8:2; The grid search method is used to find the optimal initial learning rate and batch size parameters, determine the number of iterations W of the CNN-LSTM hybrid neural network, and initialize the number of training times v = 1; The convolutional neural network (CNN) is used to train the model and extract the characteristic information of the three-dimensional feature data of Hurst index, interharmonic variance, and wavelet energy entropy. Use long short-term memory neural network LSTM to learn feature information; Use error back propagation to update the parameters of CNN-LSTM hybrid neural network; Compare the number of training times v and W. If v>W, then end the training; Otherwise, continue iterating the training operation; Input the test set to perform arc fault identification.

6. An AC fault arc detection system as claimed in claim 5, characterized in that: Also includes: Aviation power supply; AC arc fault circuit breaker; Cable under test; Current transformer; Aviation load; The aviation power supply is connected to one end of an AC fault arc circuit breaker, the AC fault arc circuit breaker is connected to a cable to be tested, a line of the cable to be tested passes through a current transformer, and the cable to be tested is connected to an aviation load.