A dual mode multi-load circuit arc fault detection system

Abstract

The application discloses and realizes a dual-mode multi-load loop arc fault detection system, and aims at the problems of insufficient precision of existing real-time arc fault detection devices and high-load continuous operation of equipment, etc., divides the fault arc into two kinds of starting arc and process arc, and uses different modes of detection methods according to different conditions of starting fault and process fault, adopts a characteristic value partition detection method in the process mode, and makes the method tend to be simple under the condition of ensuring accuracy and real-time performance. The method takes an stm32H7 as a core microprocessor, is matched with a conditioning circuit, a power supply circuit, a data acquisition circuit and a wireless communication circuit to form a dual-mode fault diagnosis system. The application has the characteristics of high monitoring precision and high speed, and has strong popularization value and use value.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of electrical circuit, in particular, to a multi-load series arc fault detection system. BACKGROUND

[0002] Electricity is an indispensable part of human life, and ensuring power safety is an important part of ensuring personal safety. Arc fault is a key part of power safety. Arc is a kind of gas ionization discharge phenomenon. Arc on the line can be divided into two types: one is normal operating arc, called "good arc"; the other is fault arc, called "bad arc". "Good arc" refers to the arc generated when the motor rotates, the switch electrical appliance, and the plug-in power supply. "Bad arc" is also called fault arc. When fault arc occurs, it will produce extremely high temperature, which can easily ignite combustible materials, which indirectly leads to more than half of the total number of electrical fires caused by electrical line faults every few years.

[0003] Electrical fire accidents pose an increasing threat to people's lives and property. Since fault arc does not necessarily exhibit abnormal current, it has characteristics such as small current and intermittency, which makes traditional overcurrent protectors and circuit breakers unable to detect and protect it. Therefore, fault arc accidents cannot be effectively prevented, and real-time identification of arc faults is of great significance to fire prevention.

[0004] Fault arc current is usually small, especially for series arc fault, whose arc current is limited by line load and is generally smaller than normal working current of the line. The circuit breaker or fuse of the power supply system cannot cut off such arc faults. Domestic fault arc detection equipment is difficult to implement due to the high complexity of most detection algorithms. Threshold judgment is mainly used for detection. To ensure accuracy, current arc fault detection equipment algorithms are mainly tested for single load. Therefore, arc fault detection algorithms with strong applicability, high accuracy, and relative simplicity and ease of implementation have become the focus of current research.

[0005] Domestic research on arc fault is mainly based on decomposition algorithms such as wavelet decomposition, empirical mode decomposition (EMD), and variational mode decomposition to analyze current signals, and then extract subtle features and identify them using neural networks. Although this method can achieve good results in simulation experiments, the signal after decomposition shows detailed information while increasing the data dimension, which makes most of them stay in simulation experiments. To solve this problem, some scholars use dimension reduction algorithms such as singular value decomposition (SVD), principal component analysis (PCA), factor analysis (FA), and independent component analysis (ICA) to reduce the dimension of the data for analysis. However, decomposition algorithms and dimension reduction algorithms themselves have high complexity, which makes it difficult to experiment during software transplantation and requires high performance of hardware. Therefore, there are still many difficulties in the experiment of current fault arc detection equipment.

[0006] The application divides the detection mode into a starting detection mode and a process detection mode, the starting detection mode algorithm is relatively complex, but the detection time breaks the process detection mode, so that the long-term load stability of the equipment can be ensured, the process detection threshold is divided into a fault area, a sensitive area and a normal area, three characteristic quantities and a deep convolutional neural network are used for progressive joint judgment, and finally the algorithm of the application is applied to an STM32F743, and a 4G communication module and other peripheral circuits are combined to realize an online cloud real-time fault detection device. SUMMARY

[0007] The application aims to solve the problems of the prior art. A dual-mode multi-load circuit arc fault detection system is provided, which mainly solves the problems that the arc fault characteristics of multi-load are complex and difficult to distinguish, the single-load fault arc is very similar to the normal current of part of multi-load, and the conventional algorithm is prone to misjudgment. The technical scheme of the application is as follows: in order to achieve the above purpose, the technical scheme adopted by the application is as follows:

[0008] A dual-mode multi-load circuit arc fault detection system, the detection algorithm scheme is as follows:

[0009] A dual-mode multi-load circuit arc fault detection system, characterized in that it comprises normal and fault current collection, multi-mode fault arc detection algorithm principle and verification, detection system module design and implementation;

[0010] The normal and fault current collection is to collect current signals of different types of typical loads and different combinations of multi-loads in the circuit under the series alternating current system by using a current transformer and an oscilloscope, thereby forming different types of single-load and multi-load current sample sets.

[0011] The multi-mode fault arc detection algorithm principle is: firstly, whether there is normal characteristic data in the system is determined to determine the detection mode, the detection mode is divided into a starting detection mode and a process detection mode, four characteristic quantities are introduced into a trained deep convolutional neural network for judgment in the starting detection mode, and the data judged by the starting mode is further used to calculate each characteristic quantity in the process detection mode, the increase ratios of the three characteristic quantities are calculated, and whether an arc fault is generated is determined by partition.

[0012] Further, the starting detection mode is to calculate wavelet energy entropy, kurtosis, skewness and rectified mean value by using collected current data, and then the four characteristic quantities are introduced into a trained neural network for judgment, wherein the wavelet energy entropy is to calculate the energy of each coefficient array after wavelet transform, then the relative energy is calculated, and then the entropy is calculated according to the definition, and the calculation formula is as follows:

[0013]

[0014] Kurtosis is a random distribution characteristic of a set of data, which is a fourth-order accumulation, and the calculation formula is as follows:

[0015]

[0016] Skewness describes the skewness of some non-normal or asymmetric distribution function, which is a measure of the skewness direction and the degree of asymmetry of the function distribution. The definition is as follows:

[0017]

[0018] Where E i and E are the energy of the i-th wavelet coefficient and the total energy, respectively; wherein the standard deviation of the signal is sigma, N, x i The number of discrete signals, the amplitude of the i-th discrete signal.

[0019] Further, the process detection mode is to determine whether a fault arc occurs by using the eigenvalue increase ratio, and the eigenvalue increase ratio formula is as follows:

[0020]

[0021] x is the eigenvalue of the current period, x nor is the eigenvalue of the normal period, the application introduces the eigenvalue increase ratio of the current detection period and the normal state by using the starting mode judgment result, which can greatly improve the correctness of the judgment, in order to ensure the stability and complexity of the algorithm, the application divides the three eigenvalue increase ratios into normal zone, sensitive zone and fault zone, calculates the partition according to the order of wavelet energy entropy increase ratio, skewness increase ratio and kurtosis increase ratio, and selects whether to perform the next eigenvalue increase ratio calculation according to the current partition, if the three eigenvalue partitions are in the sensitive zone, the convolutional neural network is used to determine whether a fault arc occurs. The application uses distributed calculation of eigenvalue partition, so that in most cases the system only needs to calculate one or two eigenvalues to detect whether an arc fault has occurred, greatly reducing the complexity of real-time detection, and in a particularly complex scene, multiple eigenvalues can be used together with a deep convolutional neural network to jointly determine, ensuring the accuracy of recognition.

[0022] The design and implementation of each module of the detection system are that the application uses STM32H743 as a microcontroller, combines a power supply circuit, a main control circuit, a current data acquisition circuit, a debugging interface circuit, a key test circuit, a 4G communication circuit, and realizes a dual-mode multi-load loop arc fault detection system, ensuring the integrity and feasibility of the system.

[0023] Compared with the prior art, the application has the following beneficial effects:

[0024] (1) The present application is aimed at single load circuit, double load circuit, four load circuit, five load circuit arc fault detection, greatly meet the arc fault situation in real life applicability is very strong.

[0025] (2) The present application divides arc fault into double mode multi-region detection, uses starting mode deep convolutional neural network to identify data and introduces data relative to normal data characteristic value increase ratio for partition judgment, reduces detection equipment load under the condition of ensuring high accuracy, and is easy to realize algorithm.

[0026] (3) The present application monitors the current of each node in real time, immediately alarms and disconnects the circuit once arc fault occurs, prevents the occurrence of electrical fire, improves safety; 4G communication module is added in the device design process, the current and the change of each feature can be monitored in real time, arc fault can be immediately received remotely, which more effectively reduces the loss, and the present application strictly plans the circuit and the device volume, ensures the convenience of the device. BRIEF DESCRIPTION OF DRAWINGS

[0027] Figure 1 The algorithm flowchart of the present application.

[0028] Figure 2 The neural network model diagram used in the present application

[0029] Figure 3 、 4 , 5 is the comparison diagram of three characteristic quantities fault and normal of 30 group experiment tests of the present application.

[0030] Figure 6 The device software design flowchart of the present application.

[0031] Figure 7 The hardware design module diagram of the present application. DETAILED DESCRIPTION

[0032] The present application will be further described below in combination with the drawings and examples, and the mode of the present application includes but is not limited to the following examples.

[0033] As Figure OneThe application divides the arc fault detection into two modes, namely a starting mode and a process mode. In the starting mode, the system uses a CNN neural network to judge the current data. If the data is normal, the process mode is judged. Otherwise, a fault alarm is given. In the process mode, the extreme point increase ratio, kurtosis increase ratio and wavelet energy entropy increase ratio are calculated by comparing the normal data of the first sampling period in the starting mode. If the data is normal, the current sampling period data is compared with the previous two sampling periods. The current interval of the current is determined according to the increase ratio data. In the process mode, in order to improve the detection reliability, the skewness increase ratio and the wavelet energy entropy increase ratio are used as the first level judgment, the kurtosis increase ratio is used as the second level judgment, and the CNN neural network is used as the third level judgment. When the detection feature data is in the sensitive area, the next stage is judged. According to the analysis and data statistics in the second chapter, the feature quantity threshold interval is set as shown in the following table. In order to ensure that the threshold interval can be applied to more scenes as much as possible, the normal interval threshold and the fault interval threshold are set very strictly. In the judgment of the wavelet energy entropy increase ratio, the normal data increase ratio is mainly concentrated in 0-20%. However, since the increase ratio of some intervals of the fault waveform is close to the normal waveform, the wavelet energy entropy increase ratio is only set to the sensitive area and the fault area. The wavelet energy entropy increase ratio and the skewness increase ratio are used as the first level judgment condition, which greatly reduces the false positive rate of the algorithm.

[0034] The CNN has good fault tolerance, parallel processing and self-learning ability. Compared with the traditional full-link neural network, the detection algorithm accuracy can be improved. The application uses a convolutional neural network including a convolution layer, a pooling layer, a normalization layer, a full connection layer and an activation layer. The deep convolutional neural network model of the application is shown in Figure Two When detecting arc faults, the calculated feature quantity data is input, a one-dimensional CNN network is used for binary classification, and the probability of binary classification is used to realize the detection of fault data and normal data. During training, 30 groups of experimental data are input, each group of experimental data includes 400 feature data, including 200 normal state data and 200 fault state data. The trained neural network model is transplanted to an stm32H7 development board, and the test result is correct.

[0035] As shown in Figure Three , FourAs shown in Figure 5, this invention analyzes the wavelet energy entropy, kurtosis, and skewness of 30 groups of experimental fault signals and normal signals, and plots a comparison graph. The kurtosis comparison graph is cropped at 300° on the vertical axis to make the effect more obvious. Each group contains 400 samples, totaling 120,000 samples. The first half of each experimental group represents the fault characteristic values, and the second half represents the normal characteristic values. The graph shows that the three characteristic quantities clearly distinguish between fault and normal signals. In particular, the absolute values ​​of kurtosis and skewness are extremely sensitive to fault signals, while the characteristic values ​​of normal signals are relatively concentrated. The comparison graph shows that the characteristic values ​​of fault signals are uneven, and most characteristic values ​​are greater than those of normal signals. However, the graph also shows that the three characteristic quantities are not as clearly distinguishable for groups 14, 15, and 24 compared to other experimental groups. Therefore, using only these three characteristic quantities to determine the threshold for whether a fault arc has occurred may lead to misjudgment. Furthermore, it can be seen from the figure that the characteristic values ​​of the fault signals and normal signals in each group of experiments are very different. Therefore, the increase ratio of the characteristic values ​​can be used to determine whether an arc fault has occurred. The process detection mode of this invention takes advantage of this feature to detect fault arcs.

[0036] A dual-mode multi-load circuit arc fault detection system is described below, including the system design and module scheme:

[0037] like Figure Six As shown, the system software design of the dual-mode multi-load circuit arc fault detection device mainly includes hardware and software program initialization, detection mode selection, data acquisition, data preprocessing, algorithm fault detection, fault alarm, and data cloud platform upload. Hardware program initialization includes clock configuration, AD acquisition initialization, timer initialization, IO port configuration, EEPROM initialization, and 4G module initialization. Data preprocessing mainly includes data acquisition, data detection, and median filtering. Since the data acquired by the current transformer from an unenergized cable is entirely interference data, directly using feature values ​​for judgment will lead to serious misjudgments. To address this problem, this invention employs data preprocessing to detect whether the acquired data is usable, and then performs arc fault detection based on the usable data. Median filtering is used to eliminate the influence of minor interference from the current transformer during data acquisition. Median filtering ensures that large changes in the fault current waveform are not destroyed, while also making the normal current waveform smoother, which is more conducive to the detection and judgment of arc faults. To avoid the arc fault detection device failing to send fault information to the cloud platform due to power failure in the circuit, this invention uses EEPROM for storage, allowing fault information to be retained for easy analysis of the fault cause.

[0038] like Figure SevenAs shown, the hardware design circuit mainly includes a power supply circuit, a main control circuit, a current data acquisition circuit, a debugging interface circuit, a key test circuit, a 4G communication circuit and the like. In order to consider the portability of the arc fault detection, the power supply circuit interface uses two ways of household power plug direct access and terminal access, and the N pole of the power supply passes through the current transformer, and then passes through the conditioning circuit to access the ADC acquisition circuit, and the data is transmitted into the main control. The alarm circuit uses a buzzer for alarm; the main function of the debugging interface circuit is code debugging and serial communication.

[0039] In the embodiment, a single-chip microcomputer with a model of STM32H743 is used as a system main control chip, and the circuit structure belongs to the prior art, and those skilled in the art can configure it based on the basic knowledge of electronic circuits and the content described in the embodiment, which is not described here.

[0040] The present application has the characteristics of high monitoring precision and fast speed, and has strong popularization value and use value.

[0041] The above embodiment is only one of the preferred embodiments of the present application, and should not be used to limit the protection scope of the present application, but any modification or polishing without substantial meaning made within the main design idea and spirit of the present application, and the technical problems solved are still consistent with the present application, should be included in the protection scope of the present application.

Claims

1. A dual-mode multi-load circuit arc fault detection method, which involves collecting normal and fault currents and detecting them using a dual-mode fault arc detection algorithm; The dual-mode fault arc detection algorithm features include: firstly, determining the detection mode based on whether normal feature data exists in the system, dividing the detection mode into an initial detection mode and a process detection mode; in the initial detection mode, using the acquired current signal, calculating each feature quantity, and importing each feature quantity into a trained deep convolutional neural network for judgment; in the process detection mode, using the data judged in the initial mode, further calculating each feature quantity, calculating the increase ratio of each feature quantity, determining the calculation order of the feature quantity increase ratio based on the feature quantity increase ratio data, partitioning each feature quantity increase ratio, and using the partition where the feature quantity increase ratio is located to determine whether an arc fault has occurred. The initial detection mode uses the collected current data to calculate wavelet energy entropy, kurtosis, skewness, and rectified mean. These four features are then input into a trained neural network for judgment. Wavelet energy entropy is calculated by taking the energy of each set of coefficients after wavelet transform, then calculating its relative energy, and finally using the definition of entropy. The calculation formula is as follows: Where E i E and E are the energy and total energy of the i-th group of wavelet coefficients, respectively; Kurtosis reflects the random distribution characteristics of a set of data. It is a fourth-order cumulant, and its calculation formula is as follows: Where σ is the standard deviation of the signal, μ is the mean of the discrete signal, and N, x i Let i be the number of discrete signals and the amplitude of the i-th discrete signal. Skewness describes the degree of skewness of some non-normal or asymmetric distribution functions. It is a measure of the direction of skewness and the degree of asymmetry in the distribution. The definition is as follows: The process detection mode uses the eigenvalue increment ratio to determine whether a fault arc has occurred. The eigenvalue increment ratio formula is as follows: x is a characteristic quantity of the current period, x nor These are characteristic quantities of a normal cycle; The process detection mode is characterized in that it divides the three feature quantity increment ratios into normal zone, sensitive zone and fault zone, calculates the partitioning situation according to the detection order of wavelet energy entropy increment ratio, skewness increment ratio and kurtosis increment ratio, selects whether to perform the next feature quantity increment ratio calculation based on the current partitioning situation, and if all three feature quantity partitions are in the sensitive zone, a convolutional neural network is used to determine whether a fault arc has occurred.

2. The dual-mode multi-load circuit arc fault detection method according to claim 1, wherein the step of acquiring normal and fault currents is characterized in that, In a series AC system, current transformers and oscilloscopes are used to collect current signals from different types of typical loads and different combinations of multiple loads in the circuit. Current signals are collected from different states of different loads to form different types of single-load and multi-load current sample sets.

3. A detection system based on a dual-mode multi-load circuit arc fault detection method, comprising a power supply circuit, a main control circuit, a current data acquisition circuit, a debugging interface circuit, a button testing circuit, a 4G communication circuit, and an alarm circuit, realizing a dual-mode multi-load circuit arc fault detection system. The main control circuit is an STM32H743 microcontroller, and adopts the dual-mode arc fault detection algorithm in the dual-mode multi-load circuit arc fault detection method of claim 1. The power supply circuit interface uses both direct connection via a household power plug and terminal block connection, with the N-pole of the power supply circuit carrying the current. The current transformer, after passing through a conditioning circuit, is connected to the current data acquisition circuit, which then transmits the data to the main control circuit. The main control circuit uses a buzzer to sound an alarm via an alarm circuit. The 4G communication circuit is connected to the main control circuit via a UART serial port, and the button testing circuit is connected to the main control circuit via a standard I / O interface. The system functions include arc data acquisition, data analysis and processing, arc fault alarm, and remote cloud control. Arc data acquisition is implemented by the current data acquisition circuit; data analysis and processing is implemented by the main control circuit; arc fault alarm is implemented by the alarm circuit; and remote cloud control is implemented by the 4G communication circuit.

Images