Series arc fault detection method and system based on Lissajous trajectory visualization

Abstract

The invention discloses a series arc fault detection method and system based on Lissajous trajectory visualization, which utilize the Lissajous trajectory visualization low-voltage user current, amplify weak fault features during series arc fault, introduce the image recognition field technology to simplify the calculation demand of the complex diagnosis process, and have the advantages of fast diagnosis speed, high accuracy and high reliability. The contradiction that the current low-voltage arc fault detection accuracy and the computing power requirement are restricted is effectively solved, and a good detection effect is achieved in the field of low-voltage series arc fault detection.

Description

Technical Field

[0001] The present invention relates to the technical field of low-voltage series arc fault detection, and in particular to a series arc fault detection method and system based on Lissajous locus visualization. Background Art

[0002] Low-voltage series arc faults, caused by factors such as aging line insulation and loose terminals, are currently a major cause of large-scale residential fires. Due to their hidden nature, random and unknown locations, and the lack of effective monitoring methods, these faults pose significant fire safety risks, garnering widespread public attention and becoming a current research hotspot.

[0003] The current amplitude of a series arc is weaker than the normal load current and does not affect the vector sum of the hot-neutral and ground currents, rendering existing protective devices ineffective. Therefore, existing research primarily focuses on general arc characteristics, encompassing three areas: arc physical characteristics, arc mathematical and physical models, and arc time-frequency characteristics. The arc physical characteristics and arc model methods are limited in their applicability due to limitations in sensor installation location and model applicability. Arc time-frequency characteristics are currently the mainstream detection method, enabling fault detection by monitoring mainline electrical quantities. These methods are categorized into three types: time-domain, frequency-domain, and time-frequency characteristics, depending on the characteristics of the selected electrical quantities. Time-domain characteristics are based on random variables of voltage and current, such as the current mutation rate and voltage mutation slope, to determine faults. Other approaches utilize current waveform similarity to construct feature vectors for diagnostic purposes. However, due to the high frequency of disturbances in power systems, the one-to-one mapping between time-domain random characteristics is unclear, and random disturbances with arc-like characteristics lead to a high rate of false positives. Frequency-domain features utilize the harmonic component characteristics of the fault signal, such as the probability distribution of the fifth harmonic and the content ratio of lower-order harmonics, to achieve fault diagnosis. Modal decomposition can also be used to obtain the intrinsic mode function of a specific frequency band and use its kurtosis, margin, and other characteristics as a basis for differentiation to complete the diagnosis task. However, frequency-domain algorithms rely on the characteristics of linear observable systems, which conflicts with the randomness of arcs and limits their accuracy. Time-frequency domain methods combine the characteristics of both methods to adaptively extract the fault time-frequency matrix. Given an appropriate wavelet basis, they can effectively identify fault characteristics such as random pulses and high-frequency energy. However, the computational requirements of this method are higher than those of the previous two methods, limiting its practicality.

[0004] In general, due to the wide variety of loads and complex arc conditions on the user load side, the existing series arc fault detection methods all have a contradiction between algorithm complexity and algorithm accuracy. How to grasp the intrinsic characteristics of the arc and propose a streamlined detection method is the focus of research. Summary of the Invention

[0005] The present invention provides a series arc fault detection method and system based on Lissajous locus visualization to solve the problem of algorithm complexity and detection accuracy being constrained in existing low-voltage series arc fault detection methods.

[0006] According to a first aspect, an embodiment provides a series arc fault detection method based on Lissajous locus visualization, the method comprising: Periodically collect current sequence signals and perform continuous abnormal pulse detection; If continuous abnormal pulses are detected in the current sequence signal of the current cycle, the load type is determined by calculating the ratio of the power of the first frequency band of the current to the power of the second frequency band of the current in the frequency domain power of the full frequency band, and a Lissajous locus diagram of the current sequence signal of the current cycle is generated, wherein the full frequency band of the current is divided into a first frequency band that does not exceed the preset boundary frequency threshold and a second frequency band that exceeds the preset boundary frequency threshold according to a preset boundary frequency threshold; According to the load type of the current sequence signal of the current cycle, the Lissajous locus diagram of the current sequence signal of the current cycle is input into a pre-trained series arc fault classification detection network model of the corresponding load type, and the series arc fault classification detection result is output.

[0007] Furthermore, the current sequence signal is periodically collected and continuous abnormal pulse detection is performed, specifically including: The current sequence waveform of the low-voltage 220V user power main circuit is collected, and the current sequence i of one cycle is stored with the positive zero-crossing point as the starting point and the negative zero-crossing point as the ending point.

[0008] Furthermore, the current sequence signal is periodically collected and continuous abnormal pulse detection is performed, specifically including: Calculate the current mean The calculation method is to remove the maximum and minimum values ​​of every 5 current sampling points and then calculate the average value. The calculation formula is as follows: ; in, n is the sampling point sequence number, Indicates the current sampling point i n The local current mean, max( i m ) represents the current sampling point i n The local maximum value, min( i m ) represents the current sampling point i n The local minimum of ; Calculate the pulse coefficient of the current , the calculation method is for each current sampling point i n and current mean The ratio is calculated as follows: ; When the product of the ratio of three consecutive pulse coefficients to the normal pulse coefficient is greater than or equal to 8, it is determined that there are continuous abnormal pulses in the current sequence signal of the current cycle. The formula is as follows: ; in, is the pulse coefficient during normal operation.

[0009] Furthermore, the load type is determined by calculating the ratio of the current power in the first frequency band to the current power in the second frequency band to the current power in the full frequency band. Specifically, the method includes: Perform fast Fourier transform on the current sequence signal i of the current cycle to obtain the frequency spectrum F[i]; Calculate the energy concentration distribution of the frequency band: ; Where, f Represents the frequency, P f is the full-band power of the current sequence, P low is the power of the first frequency band of the current sequence, P high is the second frequency band power of the current sequence, BandStop is the preset dividing frequency threshold between the first and second frequency bands, S is the sampling rate, S / 2>BandStop; Calculate the ratio of the power in the first frequency band to the power in the second frequency band to the frequency domain power of the full frequency band current using the following formula: ; Among them, R pl is the ratio of the first frequency band power to the full frequency band power, R ph It is the ratio of the second frequency band power to the full frequency band power.

[0010] Furthermore, the load type is determined by calculating the ratio of the current power in the first frequency band to the current power in the second frequency band to the current power in the full frequency band. Specifically, the method includes: First, determine whether the ratio of the first frequency band power to the total frequency band power exceeds 80%; If the ratio of the first frequency band power to the full frequency band power exceeds 80%, the load type is determined to be a linear RLC load; If the ratio of the first frequency band power to the total frequency band power does not exceed 80%, then further determine whether the ratio of the second frequency band power to the total frequency band power exceeds 40%; If the ratio of the second frequency band power to the full frequency band power exceeds 40%, the load type is determined to be a nonlinear SW load; If the ratio of the second frequency band power to the full frequency band power does not exceed 40%, the load type is determined to be a mixed load.

[0011] Furthermore, generating a Lissajous locus diagram of the current sequence signal of the current cycle specifically includes: Perform Hilbert transform on the current sequence signal of the current cycle to obtain the complex analytical signal of the current signal. The formula is as follows: ; Where s(t) is the transformed complex analytic signal, t is the time variable, and f(t) is the real part of the complex analytic signal. is the imaginary part of the negative analytic signal, j is the imaginary operator; The real and imaginary parts of the complex analytic signal are used as variables to draw the generalized Lissajous locus of the current sequence signal and obtain the grayscale image of the Lissajous locus.

[0012] Furthermore, the training method of the series arc fault classification detection network model includes: For each load type, a Lissajous trajectory diagram sample set is constructed and the classifier network is trained to obtain the series arc fault classification detection network model corresponding to different load types.

[0013] Furthermore, the training method of the series arc fault classification detection network model includes: The classifier network adopts the ResNet network.

[0014] Furthermore, the training method of the series arc fault classification detection network model includes: The sample set is divided into a training set and a validation set, the classifier network is hyper-tuned, and the accuracy is selected as the evaluation indicator of the classifier network model.

[0015] According to a second aspect, an embodiment provides a series arc fault detection system based on Lissajous locus visualization, the system comprising: Signal acquisition and detection module, used to periodically collect current sequence signals and perform continuous abnormal pulse detection; a Lissajous locus diagram generating module, for determining the load type of the current sequence signal of the current cycle by calculating the ratio of the power of the current in the first frequency band to the power of the current in the second frequency band to the power in the frequency domain of the current in the full frequency band, if continuous abnormal pulses are detected in the current sequence signal of the current cycle, and generating a Lissajous locus diagram of the current sequence signal of the current cycle, wherein the full frequency band of the current is divided into a first frequency band that does not exceed the preset boundary frequency threshold and a second frequency band that exceeds the preset boundary frequency threshold according to a preset boundary frequency threshold; The fault detection module is used to input the Lissajous locus diagram of the current sequence signal of the current cycle into a pre-trained series arc fault classification detection network model of the corresponding load type according to the load type of the current sequence signal of the current cycle, and output the series arc fault classification detection result.

[0016] The present invention provides a method and system for detecting series arc faults based on Lissajous loci visualization. The method utilizes Lissajous loci to visualize low-voltage user currents, amplifies the weak fault characteristics of series arc faults, and introduces image recognition technology to simplify the computational requirements of complex diagnostic processes. The method has a fast diagnostic speed and effectively resolves the contradiction between the accuracy of low-voltage arc fault detection and the computing power requirements today, achieving better detection results in the field of low-voltage series arc fault detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 A flowchart of a series arc fault detection method based on Lissajous locus visualization provided by one embodiment of the present invention; Figure 2 A series arc fault detection method based on Lissajous locus visualization provided by one embodiment of the present invention provides a load current and Lissajous locus diagram when a series arc fault occurs and under normal conditions; Figure 3 A schematic diagram of the logical structure of a series arc fault detection system based on Lissajous locus visualization is provided in accordance with an embodiment of the present invention. DETAILED DESCRIPTION

[0018] The present invention will be further described in detail below by means of specific embodiments in conjunction with the accompanying drawings. Similar elements in different embodiments are numbered with associated similar elements. In the following embodiments, many detailed descriptions are provided to enable the present invention to be better understood. However, those skilled in the art will readily appreciate that some of the features may be omitted under different circumstances, or may be replaced by other elements, materials, or methods. In some cases, some operations related to the present invention are not shown or described in the specification. This is to avoid the core of the present invention being overwhelmed by excessive descriptions, and for those skilled in the art, it is not necessary to describe these related operations in detail. They can fully understand the related operations based on the description in the specification and the general technical knowledge in the art.

[0019] In addition, the features, operations, or characteristics described in the specification may be combined in any appropriate manner to form various embodiments. Furthermore, the steps or actions in the method description may be reordered or adjusted in a manner readily apparent to those skilled in the art. Therefore, the various sequences in the specification and drawings are provided solely for the purpose of clearly describing a particular embodiment and are not intended to be mandatory, unless otherwise specified.

[0020] The first embodiment of the present invention provides a series arc fault detection method based on Lissajous locus visualization, which uses Lissajous locus to visualize current signals, captures the characteristics of series arc faults under different load conditions, determines the power mode type based on the observed current harmonic components, relies on abnormal current pulses to start, and finally completes series arc fault detection based on image recognition technology. Figure 1 Provide detailed explanation.

[0021] like Figure 1 As shown, in step S100, the current sequence signal is periodically collected and continuous abnormal pulse detection is performed.

[0022] The above steps specifically include: S110: Collect the waveform of the low-voltage 220V user power main current sequence, starting from the positive zero-crossing point and ending at the negative zero-crossing point, and store a current sequence i of one cycle. The length of the current sequence i is one power frequency cycle.

[0023] S120, calculate the current average The calculation method is to remove the maximum and minimum values ​​of every 5 current sampling points and then calculate the average value. The calculation formula is as follows: ; in, n is the sampling point sequence number, Indicates the current sampling point i n The local current mean, max( i m ) represents the current sampling point i n The local maximum value, min( i m ) represents the current sampling point i n The local minimum of ; S130, calculate the pulse coefficient of the current , the calculation method is for each current sampling point i n and current mean The ratio is calculated as follows: ; S140: When it is detected that the product of the ratios of three consecutive pulse coefficients to the normal pulse coefficient is greater than or equal to 8, it is determined that there are consecutive abnormal pulses in the current sequence signal of the current cycle. The formula is as follows: ; in, It is the pulse coefficient during normal operation (such as one hour or one day).

[0024] like Figure 1 As shown, in step S200, if it is detected that there are continuous abnormal pulses in the current sequence signal of the current cycle, the load type is determined by calculating the ratio of the current first frequency band power to the current second frequency band power in the full frequency band current frequency domain power of the current sequence signal of the current cycle, and a Lissajous locus diagram of the current sequence signal of the current cycle is generated, wherein the full frequency band of the current is divided into a first frequency band that does not exceed the preset boundary frequency threshold and a second frequency band that exceeds the preset boundary frequency threshold according to the preset boundary frequency threshold.

[0025] The above steps specifically include: When continuous abnormal pulses are detected, the following algorithm steps are executed; if no continuous abnormal pulses are detected, current signal sampling continues.

[0026] S210, performing fast Fourier transform on the current sequence signal i of the current cycle to obtain a frequency spectrum F[i]; S220, calculate the frequency band energy concentration distribution: ; Where, f Represents the frequency, P f is the full-band power of the current sequence, P low is the power of the first frequency band of the current sequence, P high is the power of the second frequency band of the current sequence, BandStop is the preset dividing frequency threshold between the first and second frequency bands, which is set to 0.8 kHz in this embodiment, and S is the sampling rate. S / 2>BandStop, that is, the sampling rate should be at least greater than 1.6 kHz, with at least 32 points per cycle; S230, calculating the ratio of the first frequency band power to the second frequency band power to the full frequency band current frequency domain power, using the following formula: ; Among them, R pl is the ratio of the first frequency band power to the full frequency band power, R ph It is the ratio of the second frequency band power to the full frequency band power.

[0027] S240: Determine the load type and perform modal identification, specifically including: First, determine whether the ratio of the first frequency band power to the total frequency band power exceeds 80%, that is: ; If the ratio of the first frequency band power to the full frequency band power exceeds 80%, the load type is determined to be a linear RLC load, and the modal identifier is set as a linear RLC load; If the ratio of the first frequency band power to the total frequency band power does not exceed 80%, then further determine whether the ratio of the second frequency band power to the total frequency band power exceeds 40%, that is: ; If the ratio of the second frequency band power to the full frequency band power exceeds 40%, the load type is determined to be a nonlinear SW load, and the modal identifier is set as a nonlinear SW load; If the ratio of the second frequency band power to the full frequency band power does not exceed 40%, the load type is determined to be a mixed load, and the modal identifier is set to mixed load.

[0028] S250, perform Hilbert transform on the current sequence signal of the current cycle to obtain a complex analysis signal of the current signal, which is expressed as follows: ; Where s(t) is the transformed complex analytic signal, t is the time variable, and f(t) is the real part of the complex analytic signal. is the imaginary part of the negative analytical signal, j is the imaginary operator, the real part f(t) is the original signal, and the imaginary part It is the H-transformed signal after Hilbert transform of the original signal f(t); S260, using the real part and the imaginary part of the complex analytical signal as variables, draws the generalized Lissajous trajectory of the current sequence signal, and outputs a Lissajous trajectory grayscale image with a pixel size of 224×224.

[0029] Figure 2 Figure 2 is the Lissajous locus diagram of the load current (current analysis signal) and the (normal load current) under series arc fault and normal conditions.

[0030] In addition, when visualizing current through Lissajous loci, possible methods include not only obtaining obvious fault images through Hilbert transform, but also distinguishing currents in different frequency bands by designing digital filters and drawing Lissajous loci to visualize currents in different frequency bands.

[0031] like Figure 1 As shown, in step S300, according to the load type of the current sequence signal of the current cycle, the Lissajous locus diagram of the current sequence signal of the current cycle is input into the pre-trained series arc fault classification detection network model of the corresponding load type, and the series arc fault classification detection result is output.

[0032] In this embodiment, the training method of the series arc fault classification detection network model includes: constructing a Lissajous locus diagram sample set for each load type and training a classifier network to obtain a series arc fault classification detection network model corresponding to different load types.

[0033] In this embodiment, the classifier network uses a ResNet network. Regarding classifier model selection, this embodiment specifically uses the ResNet50 network for recognition, but this does not mean that only the ResNet50 network can perform efficient recognition. For example, other AlexNet, ordinary CNN, and even the lightweight network SqueezeNet may be potential excellent classifiers. The specific selection of the best classifier should be considered based on the actual hardware conditions and sample data characteristics.

[0034] The training steps of the pre-trained ResNet network include the following steps: a. Lissajous trajectory sample generation and cleaning and screening. First, consider different load types, load power, load access volume, fault occurrence time, and fault severity (such as zero-off time and power level of the fault). Then, generate training sample data for different signal-to-noise ratios and clean and screen out abnormal samples such as damaged images or incorrect file types.

[0035] b. Model training. The sample data should have a 1:1 balance between normal and abnormal samples. The data should be divided into a training set and a validation set at an 8:2 ratio for hyperparameter tuning. This paper uses a batch size of 64, an initial learning rate of 0.01, an Adam optimizer, a learning descent rate of 0.9, and 100 iterations.

[0036] c. Evaluation and Verification: Accuracy is used as the basis for model evaluation. The calculation formula is as follows: ; Where T P 、T N 、F P 、F N The numbers represent the number of correctly predicted positive and negative samples, and the number of incorrectly predicted positive and negative samples, respectively. Model pre-training ends when the training results show a validation accuracy rate above 95%. Otherwise, continue fine-tuning and retraining.

[0037] After obtaining the Lissajous locus grayscale image of the current sequence signal in step S260, the 224×224 Lissajous locus image is input into the pre-trained ResNet network corresponding to the load type based on the modal identification result. Offline diagnosis is then performed, outputting a normal or abnormal arc fault. If the diagnosis is an arc fault, the fault indicator is set to true, and a time-limit threshold is set. If the fault indicator remains true for longer than the set time threshold, a fault signal is output to issue an alarm.

[0038] This embodiment of the present invention uses Lissajous loci to visualize low-voltage user currents, amplifying the subtle fault characteristics of series arc faults. It also incorporates image recognition technology to simplify the computational requirements of complex diagnostic processes. This method offers rapid diagnostic speed and effectively addresses the current conflict between arc fault detection accuracy and computing power requirements. It achieves excellent detection results in the field of series arc fault detection. The experimental results are shown in Tables 1-3 below.

[0039]

[0040]

[0041]

[0042]

[0043]

[0044]

[0045] Corresponding to the above-disclosed method for detecting series arc faults based on visualization of Lissajous loci, an embodiment of the present invention further discloses a system for detecting series arc faults based on visualization of Lissajous loci, such as Figure 3 As shown, it specifically includes: Signal acquisition and detection module, used to periodically collect current sequence signals and perform continuous abnormal pulse detection; a Lissajous locus diagram generating module, for determining the load type of the current sequence signal of the current cycle by calculating the ratio of the power of the current in the first frequency band to the power of the current in the second frequency band to the power in the frequency domain of the current in the full frequency band, if continuous abnormal pulses are detected in the current sequence signal of the current cycle, and generating a Lissajous locus diagram of the current sequence signal of the current cycle, wherein the full frequency band of the current is divided into a first frequency band that does not exceed the preset boundary frequency threshold and a second frequency band that exceeds the preset boundary frequency threshold according to a preset boundary frequency threshold; The fault detection module is used to input the Lissajous locus diagram of the current sequence signal of the current cycle into a pre-trained series arc fault classification detection network model of the corresponding load type according to the load type of the current sequence signal of the current cycle, and output the series arc fault classification detection result.

[0046] It should be noted that for a detailed description of a series arc fault detection system based on Lissajous locus visualization provided in an embodiment of the present invention, reference can be made to the relevant description of a series arc fault detection method based on Lissajous locus visualization provided in an embodiment of the present invention, which will not be repeated here.

[0047] The above examples are used to illustrate the present invention, which are only used to help understand the present invention and are not intended to limit the present invention. Those skilled in the art can make several simple deductions, modifications or substitutions based on the concept of the present invention.

Claims

1. A series arc fault detection method based on Lissajous locus visualization, characterized in that: The method comprises: Periodically collect current sequence signals and perform continuous abnormal pulse detection; If continuous abnormal pulses are detected in the current sequence signal of the current cycle, the load type is determined by calculating the ratio of the power of the first frequency band of the current to the power of the second frequency band of the current in the frequency domain power of the full frequency band, and a Lissajous locus diagram of the current sequence signal of the current cycle is generated, wherein the full frequency band of the current is divided into a first frequency band that does not exceed the preset boundary frequency threshold and a second frequency band that exceeds the preset boundary frequency threshold according to a preset boundary frequency threshold; According to the load type of the current sequence signal of the current cycle, the Lissajous locus diagram of the current sequence signal of the current cycle is input into a pre-trained series arc fault classification detection network model of the corresponding load type, and the series arc fault classification detection result is output.

2. The method for detecting series arc faults based on Lissajous locus visualization according to claim 1, characterized in that: Periodically collect current sequence signals and perform continuous abnormal pulse detection, specifically including: Collect the current sequence waveform of the low-voltage 220V user power main circuit, and store the current sequence of one cycle with the positive zero crossing point as the starting point and the negative zero crossing point as the end point .

3. The method for detecting series arc faults based on Lissajous locus visualization according to claim 2, wherein: Periodically collect current sequence signals and perform continuous abnormal pulse detection, specifically including: Calculate the current mean The calculation method is to remove the maximum and minimum values ​​of every 5 current sampling points and then calculate the average value. The calculation formula is as follows: ； in, n is the sampling point sequence number, Indicates the current sampling point i n The local current mean, max( i m ) represents the current sampling point i n The local maximum value, min( i m ) represents the current sampling point i n The local minimum of Calculate the pulse coefficient of the current , the calculation method is for each current sampling point i n and current mean The ratio is calculated as follows: ； When the product of the ratio of three consecutive pulse coefficients to the normal pulse coefficient is greater than or equal to 8, it is determined that there are continuous abnormal pulses in the current sequence signal of the current cycle. The formula is as follows: ； in, is the pulse coefficient during normal operation.

4. The method for detecting series arc faults based on Lissajous locus visualization according to claim 1, wherein: The load type is determined by calculating the ratio of the current power in the first frequency band and the current power in the second frequency band to the current power in the full frequency band for the current sequence signal of the current cycle, specifically including: Current sequence signal of the current cycle i Perform fast Fourier transform to obtain frequency spectrum ; Calculate the energy concentration distribution of the frequency band: ； Where, f Represents the frequency, P f is the full-band power of the current sequence, P low is the power of the first frequency band of the current sequence, P high is the second frequency band power of the current sequence, BandStop is the preset dividing frequency threshold between the first frequency band and the second frequency band, S is the sampling rate, S / 2> BandStop ; Calculate the ratio of the power in the first frequency band to the power in the second frequency band to the frequency domain power of the full frequency band current using the following formula: ； in, R pl is the ratio of the first frequency band power to the full frequency band power, R ph It is the ratio of the second frequency band power to the full frequency band power.

5. The method for detecting series arc faults based on Lissajous locus visualization according to claim 1, wherein: The load type is determined by calculating the ratio of the current power in the first frequency band and the current power in the second frequency band to the current power in the full frequency band for the current sequence signal of the current cycle, specifically including: First, determine whether the ratio of the first frequency band power to the total frequency band power exceeds 80%; If the ratio of the first frequency band power to the full frequency band power exceeds 80%, the load type is determined to be a linear RLC load; If the ratio of the first frequency band power to the total frequency band power does not exceed 80%, then further determine whether the ratio of the second frequency band power to the total frequency band power exceeds 40%; If the ratio of the second frequency band power to the full frequency band power exceeds 40%, the load type is determined to be a nonlinear SW load; If the ratio of the second frequency band power to the full frequency band power does not exceed 40%, the load type is determined to be a mixed load.

6. The method for detecting series arc faults based on Lissajous locus visualization according to claim 1, wherein: Generate the Lissajous locus diagram of the current sequence signal of the current cycle, including: Perform Hilbert transform on the current sequence signal of the current cycle to obtain the complex analytical signal of the current signal. The formula is as follows: ； in, s ( t ) is the transformed complex analytic signal, t is the time variable, f ( t ) is the real part of the complex analytic signal, is the imaginary part of the negative analytic signal, j is the imaginary operator; The real and imaginary parts of the complex analytic signal are used as variables to draw the generalized Lissajous locus of the current sequence signal and obtain the grayscale image of the Lissajous locus.

7. The method for detecting series arc faults based on Lissajous locus visualization according to claim 1, wherein: The training method of the series arc fault classification detection network model includes: For each load type, a Lissajous trajectory diagram sample set is constructed and the classifier network is trained to obtain the series arc fault classification detection network model corresponding to different load types.

8. The method for detecting series arc faults based on Lissajous locus visualization according to claim 7, characterized in that: The training method of the series arc fault classification detection network model includes: The classifier network adopts the ResNet network.

9. The method for detecting series arc faults based on Lissajous locus visualization according to claim 7, wherein: The training method of the series arc fault classification detection network model includes: The sample set is divided into a training set and a validation set, the classifier network is hyper-tuned, and the accuracy is selected as the evaluation indicator of the classifier network model.

10. A series arc fault detection system based on Lissajous locus visualization, characterized in that: The system comprises: Signal acquisition and detection module, used to periodically collect current sequence signals and perform continuous abnormal pulse detection; a Lissajous locus diagram generating module, for determining the load type of the current sequence signal of the current cycle by calculating the ratio of the power of the current in the first frequency band to the power of the current in the second frequency band to the power in the frequency domain of the current in the full frequency band, if continuous abnormal pulses are detected in the current sequence signal of the current cycle, and generating a Lissajous locus diagram of the current sequence signal of the current cycle, wherein the full frequency band of the current is divided into a first frequency band that does not exceed the preset boundary frequency threshold and a second frequency band that exceeds the preset boundary frequency threshold according to a preset boundary frequency threshold; The fault detection module is used to input the Lissajous locus diagram of the current sequence signal of the current cycle into a pre-trained series arc fault classification detection network model of the corresponding load type according to the load type of the current sequence signal of the current cycle, and output the series arc fault classification detection result.

Images