Lightweight-based arc fault detection method, device and storage medium

Abstract

The application provides a lightweight arc fault detection method and device and a storage medium. The method comprises: obtaining an arc current signal of a power distribution network load in a preset time period, determining a signal-to-noise ratio parameter, determining a dynamic window length according to the signal-to-noise ratio parameter, performing Hilbert transform on the arc current signal according to the dynamic window length, obtaining m amplitude envelope data, extracting features from the m amplitude envelope data, obtaining p time domain statistical features, calculating the arc current signal using a preset Fourier transform method, obtaining arc current spectrum data, processing the arc current spectrum data, obtaining q frequency domain features, fusing the p time domain statistical features and the q frequency domain features based on a preset feature fusion algorithm, obtaining a target arc current feature, inputting the target arc current feature into a preset lightweight arc fault detection model, and obtaining an arc fault detection result. In this way, the accuracy of arc fault detection is improved.

Description

Technical Field

[0001] This application relates to the field of power safety detection technology, and in particular to a lightweight arc fault detection method, device and storage medium. Background Technology

[0002] Currently, with the acceleration of urbanization and the continuous increase in residential electricity consumption, the load on the power distribution network is also increasing. When the load on the power distribution network is high, especially when the voltage or current rises sharply, it is prone to arc faults. In low-voltage power distribution networks, current faults can easily cause electrical fires, resulting in economic or safety losses for users. For arc fault detection, current methods for detecting AC arc faults mainly include detection based on changes in electrical parameter characteristics, detection based on spectrum analysis, and detection based on machine learning or deep learning. These methods have achieved high accuracy in arc fault detection. However, due to the presence of various normal transient events in the user's power environment (e.g., motor starting, electrical switch operation), false positives or false negatives may occur. Furthermore, the large model size, slow inference speed, and high computational resource requirements of AC arc fault detection methods limit their application in embedded electronic devices.

[0003] Therefore, improving the accuracy of arc fault detection and reducing the computational overhead of arc fault detection are urgent issues to be addressed when conducting arc fault detection in power distribution networks. Summary of the Invention

[0004] This application provides a lightweight arc fault detection method, device, and storage medium, which improves the accuracy of arc fault detection and reduces the computational overhead of arc fault detection when performing arc fault detection in power distribution networks.

[0005] In a first aspect, embodiments of this application provide a lightweight arc fault detection method applied to electronic devices, the method comprising:

[0006] Acquire the arc current signal of the load in the target distribution network within a preset time period;

[0007] Determine the signal-to-noise ratio parameters of the arc current signal;

[0008] The dynamic window length is determined based on the signal-to-noise ratio parameter.

[0009] The arc current signal is subjected to Hilbert transform according to the dynamic window length to obtain m amplitude envelope data; m is a positive integer greater than 1.

[0010] Statistical features of the arc current signal in the time domain are extracted from the m amplitude envelope data to obtain p time-domain statistical features; p is greater than or equal to m.

[0011] The arc current signal is calculated using a preset Fourier transform method to obtain arc current spectrum data;

[0012] The arc current spectrum data is processed to obtain q frequency domain features; q is a positive integer greater than 1.

[0013] Based on a preset feature fusion algorithm, the p time-domain statistical features and the q frequency-domain features are fused to obtain the target arc current features;

[0014] The target arc current characteristics are input into a preset lightweight arc fault detection model to obtain arc fault detection results.

[0015] Secondly, embodiments of this application provide a lightweight arc fault detection device for use in electronic equipment, the device comprising:

[0016] The acquisition module is used to acquire the arc current signal of the load in the target distribution network within a preset time period;

[0017] The determination module is used to determine the signal-to-noise ratio (SNR) parameter of the arc current signal; determine the dynamic window length based on the SNR parameter; and calculate the arc current signal using a preset Fourier transform method to obtain arc current spectrum data.

[0018] The calculation module is used to perform a Hilbert transform on the arc current signal according to the dynamic window length to obtain m amplitude envelope data; m is a positive integer greater than 1; extract statistical features of the arc current signal in the time domain from the m amplitude envelope data to obtain p time-domain statistical features; p is greater than or equal to m; process the arc current spectrum data to obtain q frequency-domain features; q is a positive integer greater than 1; and fuse the p time-domain statistical features and the q frequency-domain features based on a preset feature fusion algorithm to obtain the target arc current features.

[0019] The control module is used to input the target arc current characteristics into a preset lightweight arc fault detection model to obtain the arc fault detection result.

[0020] Thirdly, embodiments of this application provide an electronic device, including a processor, a memory, a communication interface, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the processor, and the programs include instructions for performing steps in any method of the first aspect of this application.

[0021] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program for electronic data interchange, wherein the computer program causes a computer to perform some or all of the steps described in any method of the first aspect of this application.

[0022] Fifthly, embodiments of this application provide a computer program product, wherein the computer program product includes a non-transitory computer-readable storage medium storing a computer program, the computer program being operable to cause a computer to perform some or all of the steps described in any method of the first aspect of this application. The computer program product may be a software installation package.

[0023] By implementing the embodiments of this application, the following beneficial effects are achieved:

[0024] This application describes a lightweight arc fault detection method, device, and storage medium, applicable to electronic devices. The method involves acquiring the arc current signal of a load in a target distribution network within a preset time period, determining the signal-to-noise ratio (SNR) parameter of the arc current signal, determining the dynamic window length based on the SNR parameter, performing a Hilbert transform on the arc current signal according to the dynamic window length to obtain m amplitude envelope data (m is a positive integer greater than 1), extracting statistical features of the arc current signal in the time domain from the m amplitude envelope data to obtain p time-domain statistical features (p is greater than or equal to m), calculating the arc current signal using a preset Fourier transform method to obtain arc current spectrum data, processing the arc current spectrum data to obtain q frequency domain features (q is a positive integer greater than 1), fusing the p time-domain statistical features and q frequency domain features based on a preset feature fusion algorithm to obtain target arc current features, and inputting the target arc current features into a preset lightweight arc fault detection model to obtain the arc fault detection result. Thus, on the one hand, by using Hilbert transform to extract the current signal envelope, further calculating statistics such as envelope mean, peak factor, and kurtosis, and applying fast Fourier transform to extract indicators such as harmonic components and spectral energy ratio within a specific frequency band, and combining time-domain and frequency-domain features for prediction, the accuracy of arc fault detection is improved; on the other hand, a lightweight classification model is used to detect normal arc current and fault arc current based on the extracted features, thereby reducing the computational overhead of arc fault detection. Attached Figure Description

[0025] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0026] Figure 1 This is a system architecture diagram of a lightweight arc fault detection method provided in an embodiment of this application;

[0027] Figure 2 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application;

[0028] Figure 3 This is a schematic flowchart of a lightweight arc fault detection method provided in an embodiment of this application;

[0029] Figure 4 This is a schematic diagram of the structure of a feature extraction unit provided in an embodiment of this application;

[0030] Figure 5 This is a schematic diagram of the structure of an AC arc fault detection system provided in an embodiment of this application;

[0031] Figure 6 This is a schematic flowchart of the feature extraction and classification process of an AC arc fault detection method provided in an embodiment of this application;

[0032] Figure 7 This is a schematic diagram of an arc fault detection process based on LightGBM provided in an embodiment of this application;

[0033] Figure 8 This is a block diagram of the functional modules of a lightweight arc fault detection device provided in an embodiment of this application. Detailed Implementation

[0034] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present application.

[0035] The terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or apparatuses.

[0036] It should be understood that the term "and / or" in this document is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this document indicates that the preceding and following related objects are in an "or" relationship. In the embodiments of this application, "multiple" refers to two or more.

[0037] In the embodiments of this application, "at least one item" or its similar expression refers to any combination of these items, including any combination of a single item or a plurality of items. "One or more" means one or more, while "multiple" means two or more. For example, "at least one item" of a, b, or c can represent the following seven cases: a, b, c; a and b; a and c; b and c; a, b, and c. Each of a, b, and c can be an element or a set containing one or more elements.

[0038] In this application, the term "connection" refers to various connection methods, such as direct connection or indirect connection, to achieve communication between devices. This application does not impose any limitations on this.

[0039] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0040] The following is an explanation of the relevant terms used in this application:

[0041] Arc current signal: Arc current refers to the gas discharge phenomenon generated when current passes through an insulating medium (such as air) in an electrical system, and the arc current signal is the current signal that generates the arc of discharge.

[0042] Hilbert Transform: The Hilbert Transform is a linear operator that transforms a real-valued signal into a complex-valued signal. Mathematically, the Hilbert Transform is defined as convolving a real signal with another signal, allowing for more comprehensive analysis and processing of the signal.

[0043] For arc fault detection, current methods mainly include detection based on changes in electrical parameter characteristics, detection based on spectrum analysis, and detection based on machine learning or deep learning. These methods have achieved high accuracy in arc fault detection. However, due to the presence of various normal transient events in the user-side power environment (e.g., motor starting, electrical switch operation), false positives or false negatives may occur. Furthermore, the large model size, slow inference speed, and high computational resource requirements of AC arc fault detection methods limit their application in embedded electronic devices.

[0044] To address the aforementioned issues, this application provides a lightweight arc fault detection method, device, and storage medium applied to electronic devices. The method involves acquiring the arc current signal of a load in a target distribution network within a preset time period, determining the signal-to-noise ratio (SNR) parameter of the arc current signal, determining the dynamic window length based on the SNR parameter, performing a Hilbert transform on the arc current signal according to the dynamic window length to obtain m amplitude envelope data (m being a positive integer greater than 1), extracting statistical features of the arc current signal in the time domain from the m amplitude envelope data to obtain p time-domain statistical features (p being greater than or equal to m), calculating the arc current signal using a preset Fourier transform method to obtain arc current spectrum data, processing the arc current spectrum data to obtain q frequency domain features (q being a positive integer greater than 1), fusing the p time-domain statistical features and q frequency domain features based on a preset feature fusion algorithm to obtain the target arc current features, and inputting the target arc current features into a preset lightweight arc fault detection model to obtain the arc fault detection result. Thus, on the one hand, by using Hilbert transform to extract the current signal envelope, further calculating statistical quantities such as envelope mean, peak factor, and kurtosis, and applying fast Fourier transform to extract indicators such as harmonic components and spectral energy ratio within a specific frequency band, and combining time-domain and frequency-domain features for prediction, the accuracy of arc fault detection is improved; on the other hand, a lightweight classification model is used to detect normal arc current and fault arc current based on the extracted features, thereby reducing the computational overhead of arc fault detection.

[0045] The following is combined Figure 1 The system architecture of a lightweight arc fault detection method according to an embodiment of this application is described. Figure 1 This is a system architecture diagram of a lightweight arc fault detection method provided in an embodiment of this application. The lightweight arc fault detection system includes an AC arc current experimental platform 110 and an arc fault detection device 120.

[0046] The AC arc current experimental platform 110 is used to simulate and generate arc current signals under different operating conditions, providing raw experimental data support for the arc fault detection device 120. The AC arc current experimental platform 110 may include various load units, such as load A111, load B112, and load C113. These load units can be resistive, inductive, or capacitive loads, or combinations of multiple loads, used to simulate the characteristics of arc fault occurrence under different electrical environments. Through this experimental platform, the complex current fluctuations during arc generation can be reproduced under controllable conditions, thus providing realistic and reliable signal samples for subsequent feature extraction and classification.

[0047] The arc fault detection device 120 is connected to the AC arc current experimental platform 110. It receives real-time acquired arc current signals, processes, extracts features, and classifies the acquired signals, ultimately outputting the arc fault detection result. The arc fault detection device 120 includes: a signal acquisition module 121, a preprocessing module 122, a feature extraction module 123, and a classification decision module 124. The signal acquisition module 121 is used to acquire arc current signals in real-time from the AC arc current experimental platform 110. This module may include hardware units such as a current sensor, a data acquisition card, and signal conditioning circuitry to ensure that the acquired arc current signal is complete, continuous, and meets sampling accuracy requirements. The preprocessing module 122 is used to preprocess the acquired arc current signal, mainly including noise reduction, normalization, and window truncation. The preprocessing module 122 effectively removes common power frequency interference, background noise, and measurement errors in power systems, ensuring the purity and analyzability of the input signal. The feature extraction module 123 is used to extract multi-dimensional feature parameters from the preprocessed arc current signal, including two main categories: time-domain statistical features and frequency-domain spectral features. Time-domain statistical features can be obtained by analyzing the amplitude envelope of the arc current signal, such as mean, variance, skewness, and kurtosis; frequency-domain features are obtained through methods such as Fourier transform or wavelet transform, including harmonic distribution, high-frequency pulse features, and adaptive frequency band energy features. The classification decision module 124 is used to identify and classify the extracted arc features, outputting the final arc fault detection result. This module can employ a lightweight machine learning model, such as LightGBM or an ensemble learning algorithm. In one possible embodiment, the classification decision module 124 can also combine an adaptive weighted fusion strategy, dynamically adjusting the feature weights based on the classification accuracy of time-domain and frequency-domain features in historical training data, thereby improving detection accuracy while maintaining a lightweight model. This module enables rapid arc fault identification with low computational overhead, making it suitable for embedded devices or edge computing terminals in actual power distribution network environments. Furthermore, the various modules within the arc fault detection device 120 have clearly defined input-output relationships: the signal acquisition module 121 outputs the raw arc current signal as input to the preprocessing module 122; the preprocessing module 122 outputs the denoised and truncated signal as input to the feature extraction module 123; the feature extraction module 123 outputs multi-dimensional feature parameters as input to the classification decision module 124; and the final classification decision module 124 outputs the arc fault detection result as the final output of the entire system. Therefore, this system possesses high modularity and scalability, allowing for flexible upgrades or replacements of functional modules according to actual needs.

[0048] As can be seen, the system architecture based on the lightweight arc fault detection method described above can accurately detect and classify complex arc current signals. The system provides rich and realistic arc current signal samples through the AC arc current experimental platform 110, and the arc fault detection device 120 performs signal processing and identification, effectively reducing the missed detection rate and false judgment rate of arc faults, and improving the operational safety and reliability of the distribution network. Simultaneously, the system adopts a lightweight feature processing and classification model, which has the advantages of low computational complexity, low resource consumption, and ease of deployment. It is particularly suitable for application in edge nodes or embedded terminals of the distribution network, thereby ensuring the safe operation of the power system while reducing the hardware cost and maintenance burden of the detection system.

[0049] The following is combined Figure 2 The electronic devices in the embodiments of this application will be described. Figure 2 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application, such as... Figure 2 As shown, the electronic device 200 includes one or more processors 210, a memory 220, a communication interface 230, and one or more programs 221. The processor 210 is communicatively connected to the memory 220 and the communication interface 230 via an internal communication bus.

[0050] The processor 210 can be a central processing unit (CPU), a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute the various exemplary logic blocks, units, and circuits described in conjunction with the disclosure of this application. The processor can also be a combination that implements computational functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, etc. The communication unit can be a communication interface, transceiver, transceiver circuit, etc., and the storage unit can be a memory.

[0051] The memory 220 can be volatile memory or non-volatile memory, or it can include both. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of random access memory (RAM) are available, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate synchronous DRAM (DDR SDRAM), enhanced synchronous DRAM (ESDRAM), synchronous linked DRAM (SLDRAM), and direct rambus RAM (DR RAM).

[0052] The one or more programs 221 are stored in the memory 220 and configured to be executed by the processor 210. The one or more programs 221 include instructions for performing any step in an embodiment of a lightweight arc fault detection method described below.

[0053] It is understood that the electronic device 200 may include more or fewer structural elements than those shown in the above block diagram, such as a power module, physical buttons, a Wi-Fi module, a speaker, a Bluetooth module, sensors, a display module, etc., without limitation. It is understood that the electronic device 200 may be equipped with... Figure 1 The system architecture of a lightweight arc fault detection method is described above.

[0054] After understanding the software and hardware architecture of this application, the following will be combined with... Figure 3 This application describes a lightweight arc fault detection method based on embodiments. Figure 3 This is a flowchart illustrating a lightweight arc fault detection method provided in an embodiment of this application, specifically including the following steps:

[0055] Step S310: Obtain the arc current signal of the load in the target distribution network within a preset time period.

[0056] Among them, the arc current signal refers to the current signal flowing through the load circuit when a series arc is generated in the load circuit due to poor contact, aging of wires or loose connection in a low-voltage AC distribution network. This current signal exhibits characteristics that are significantly different from the normal operating current. Its time-domain waveform shows intermittent jitter and distortion, rich high-frequency components, and randomness and non-stationarity.

[0057] Specifically, the arc current signal is acquired through a power distribution network signal acquisition device. This device includes a current transformer (CT), a high-frequency sampling device, an oscilloscope or equivalent data acquisition unit, and a signal preprocessing module, etc., which are not limited here. The current transformer, as the coupling unit between the primary current and the sampling device, enables non-contact measurement of the current signal, ensuring both measurement safety and maintaining the original waveform characteristics of the signal. The high-frequency sampling device performs high-resolution discretization processing on the acquired current signal, and the sampling frequency needs to be greater than 10kHz to ensure that it can cover the typical high-frequency harmonics and transient components in the arc signal. The oscilloscope or data acquisition unit is used to record and store the arc signal in real time, supporting subsequent time-domain and frequency-domain analysis.

[0058] In experiments and applications, to better simulate real-world household and industrial load scenarios, resistive loads, inductive loads, and combined resistive-inductive loads are typically selected as test objects. An experimental platform is built according to the Chinese national standard GB-31143 to collect AC arc current signals under different load types in real time. The arc current signal exhibits different characteristics under different load conditions. For example, under purely resistive loads, the arc current waveform distortion is more pronounced, with prominent high-frequency components; under inductive loads, the envelope characteristics of the arc signal are more complex, potentially exhibiting delay effects and additional harmonic components. By collecting arc signals under multiple load types, a more representative dataset can be constructed, improving the model's adaptability to different application environments.

[0059] It should be noted that arc current signals differ significantly from ordinary load current signals. Under normal operating conditions, the load current waveform is highly correlated with the power supply voltage, exhibiting stability and relatively limited harmonic components. However, during an arc fault, due to the nonlinear discharge phenomenon of an arc, the signal displays the following typical characteristics: the current waveform exhibits intermittent discontinuities within certain cycles, demonstrating significant non-stationarity; harmonic components are significantly enhanced, particularly with abnormal changes in energy distribution in the mid-to-high frequency range; the amplitude of the time-domain signal undergoes abrupt changes, manifesting as short-duration spike pulses; and the signal envelope exhibits irregular fluctuations over time. Because the frequency and time domain characteristics of arc current signals differ significantly from those of ordinary load current signals, analysis of these characteristics can help distinguish between arc faults and normal currents.

[0060] Step S320: Determine the signal-to-noise ratio parameter of the arc current signal.

[0061] The signal-to-noise ratio (SNR) is the ratio of the relative intensity of the effective arc information contained in the arc current signal of a load in a power distribution network to the background noise over a certain period. Arc current signals are non-stationary and random, easily affected by external interference and normal transient events in the power distribution system, such as motor starting, switch switching, and load surges, all of which introduce additional noise components. Calculating the SNR of the arc current signal quantifies the significance of the effective arc features in the signal, thus providing a foundation for subsequent Hilbert transform and feature extraction. In other words, determining the SNR is not only a necessary step in distinguishing arc signals from noise interference, but also a prerequisite for adaptive feature extraction and classification modeling.

[0062] Specifically, the arc current signal is first segmented. Since the arc signal exhibits strong transient characteristics in the time domain and its statistical properties change significantly with time, the long-term signal is divided into several window intervals for analysis. The length of each window is selected according to the sampling frequency and detection requirements, generally between 0.05s and 0.2s, which can ensure that it covers one or several power cycles while reflecting the short-term characteristics of the arc signal.

[0063] Secondly, the signal power and noise power are calculated within each window. The signal power can be estimated using the squared mean of the current signal within the window, i.e.:

[0064]

[0065] in, This represents the current value at the i-th sampling point, and N represents the number of sampling points within the window. Indicates signal power.

[0066] Noise power estimation requires filtering and separation techniques. Specifically, the original current signal is first subjected to high-pass filtering or wavelet denoising to remove components related to the power frequency and arc frequency; the remaining portion can be approximated as noise. Its power is calculated using the mean square method. Then, the signal-to-noise ratio (SNR) parameter is calculated based on the ratio of signal power to noise power.

[0067]

[0068] in, Indicates noise power.

[0069] Finally, the obtained SNR is expressed in decibels (dB). A higher value indicates a more prominent arc characteristic component in the signal and weaker noise interference; a lower value indicates that the signal is severely masked by noise, requiring enhancement processing. The signal-to-noise ratio (SNR) parameter of the arc current signal not only measures signal quality but also provides the basis for subsequent adaptive Hilbert transform adjustments. Furthermore, the SNR parameter can be used for quality control in the arc detection process. For example, when the SNR falls below a certain threshold, the system can automatically trigger secondary sampling or enhancement processing to ensure that the acquired data meets the requirements for feature extraction. In large-scale deployment scenarios, this mechanism helps improve the automation and intelligence level of the detection system and reduces manual intervention. Further, to improve the real-time performance and adaptability of the calculation, this invention also introduces a sliding window mechanism in the SNR parameter determination process. That is, overlapping sliding windows are used on the time axis to segment the signal, and the SNR parameter is calculated and updated in real time.

[0070] Step S330: Determine the dynamic window length based on the signal-to-noise ratio parameter.

[0071] The dynamic window length refers to the length of the analysis time interval that is adaptively adjusted according to the signal-to-noise ratio (SNR) when performing Hilbert transform and feature extraction on the arc current signal. The appropriate selection of the window length is crucial for capturing arc signal features: if the window is too short, the frequency domain resolution may be insufficient, making it difficult to fully reflect the statistical characteristics of the arc signal; if the window is too long, the transient characteristics of the arc signal may be smoothed out, reducing sensitivity to the suddenness of arc faults. Therefore, by dynamically determining the window length based on the SNR parameter, a balance can be achieved between the noisy environment and signal clarity, thereby improving the accuracy and robustness of arc feature extraction.

[0072] In one possible embodiment, determining the dynamic window length based on the signal-to-noise ratio parameter specifically includes the following steps:

[0073] 331. Obtain the preset low signal-to-noise ratio threshold, high signal-to-noise ratio threshold, minimum window length, and maximum window length;

[0074] 332. Extract the arc current signal corresponding to the minimum window length from the arc current signal to obtain the first arc current signal set;

[0075] 333. Extract the arc current signal corresponding to the maximum window length from the arc current signal to obtain a second arc current signal set;

[0076] 334. Calculate the first signal power and the first noise power by calculating all the first arc current signals in the first arc current signal set based on the preset root mean square estimation method.

[0077] 335. Calculate the second signal power and the second noise power for all second arc current signals in the second arc current signal set based on the root mean square estimation method.

[0078] 336. Determine the first signal-to-noise ratio parameter based on the first signal power and the first noise power;

[0079] 337. Determine the second signal-to-noise ratio parameter based on the second signal power and the second noise power;

[0080] 338. Determine a reference signal-to-noise ratio parameter based on the first signal-to-noise ratio parameter and the second signal-to-noise ratio parameter;

[0081] 339. If the signal-to-noise ratio parameter is less than or equal to the low signal-to-noise ratio threshold, then the dynamic window length is determined based on the maximum window length, the reference signal-to-noise ratio parameter, and the signal-to-noise ratio parameter.

[0082] 3310. If the signal-to-noise ratio parameter is greater than or equal to the high signal-to-noise ratio threshold, then the dynamic window length is determined based on the minimum window length, the high signal-to-noise ratio threshold, and the signal-to-noise ratio parameter.

[0083] 3311. If the signal-to-noise ratio parameter is greater than the low signal-to-noise ratio threshold and less than the high signal-to-noise ratio threshold, then the dynamic window length is determined based on the preset interpolation formula, the low signal-to-noise ratio threshold, the high signal-to-noise ratio threshold, the minimum window length, the maximum window length, and the signal-to-noise ratio parameter.

[0084] The low signal-to-noise ratio (SNR) threshold and the high SNR threshold are used to distinguish different ranges of SNR levels and serve as boundary parameters for adaptive adjustment. The low SNR threshold corresponds to situations where the arc signal is subject to strong noise interference, while the high SNR threshold corresponds to situations where the arc signal has high clarity. By setting thresholds for different ranges, the dynamic window length can be adjusted differentially according to the actual signal environment, thereby ensuring the accuracy and robustness of feature extraction. The minimum and maximum window lengths are time range boundaries determined based on the sampling frequency and the characteristics of the arc signal. The minimum window length ensures sensitivity to the burst characteristics of the arc signal, while the maximum window length ensures the ability to smooth and suppress noise under low SNR conditions. By dynamically selecting the window length, a balance can be achieved between time-domain resolution and frequency-domain resolution.

[0085] Specifically, by extracting signal segments of corresponding lengths from the original arc current signal, two signal sets at different scales are constructed. The first arc current signal set retains short-term arc characteristics, suitable for analyzing arc bursts, while the second arc current signal set covers a longer time interval, helping to smooth noise interference and extract overall features. Next, the root mean square (RMS) estimation method is used to calculate the power of both signal sets. The RMS estimation method can accurately reflect the energy levels of signal and noise within a limited number of sampling points, and its results are applicable to windows of different lengths. By comparing the power distribution under different windows, the signal-to-noise ratio (SNR) characteristics of the arc signal at different scales can be more accurately reflected. The first and second SNR parameters correspond to the SNR estimates for the minimum and maximum window lengths, respectively, providing comparative information on arc signals and noise at different scales. Then, a reference SNR parameter is determined using the first and second SNR parameters. This reference SNR can be obtained through weighted averaging or other fusion methods and is used to provide a benchmark value during window adjustment. By introducing a reference signal-to-noise ratio (SNR), the impact of instantaneous fluctuations caused by signal abrupt changes on window length selection can be reduced, improving the stability of the adaptive mechanism. Different window adjustment strategies are adopted based on the SNR level range: when the SNR is below a low threshold, the arc signal is severely interfered with by noise, requiring a longer window to enhance signal smoothing. In this case, the dynamic window length is determined within a range close to the maximum window and adjusted in conjunction with the reference SNR to ensure sufficient statistical reliability. When the SNR is above a high threshold, indicating high signal clarity, the window can be shortened to improve the ability to capture arc details. The dynamic window length is determined within a range close to the minimum window and optimized in conjunction with a high SNR threshold to achieve high-resolution analysis of signal details. When the SNR is between two thresholds, the noise level is moderate, so an interpolation formula is used to dynamically adjust the window length between the minimum and maximum window lengths. This interpolation method ensures that the window length changes continuously and smoothly with the SNR, avoiding instability caused by parameter abrupt changes. Specifically, the window length can be expressed by the following formula:

[0086]

[0087] in, L Indicates the window length. Indicates the maximum window length. Indicates the minimum window length. This indicates a low signal-to-noise ratio threshold. This indicates the high signal-to-noise ratio threshold. represents the reference signal-to-noise ratio, and k represents the adjustment coefficient.

[0088] It should be noted that determining the dynamic window length depends not only on the signal-to-noise ratio parameter but also on the actual sampling conditions of the arc signal. At different sampling frequencies, the time scale of the window length has a conversion relationship with the number of sampling points. For example, when the sampling frequency is 10kHz, the minimum window length can be set to 0.05 seconds (corresponding to 500 sampling points), and the maximum window length can be set to 0.2 seconds (corresponding to 2000 sampling points). Through reasonable parameter configuration, it can be ensured that the extracted arc features can reflect transient changes while possessing strong noise resistance.

[0089] Step S340: Perform Hilbert transform on the arc current signal according to the dynamic window length to obtain m amplitude envelope data; m is a positive integer greater than 1.

[0090] The Hilbert transform maps the original real signal into a complex analytic signal, thereby extracting the instantaneous amplitude and phase information. In arc current signal processing, the Hilbert transform can effectively obtain the amplitude envelope curve of the signal to characterize the dynamic characteristics of current signal energy changes during arc discharge. Compared with traditional time-domain or frequency-domain statistical methods, the Hilbert transform has significant advantages in capturing the transient characteristics of non-stationary signals, and is particularly suitable for electrical phenomena such as arc faults, which are highly random and intermittent.

[0091] Among them, the amplitude envelope data is the instantaneous amplitude sequence of the signal obtained by performing a Hilbert transform on the arc current signal, which can intuitively reflect the fluctuation of the arc signal energy over time. For the load current under normal operating conditions, its amplitude envelope is relatively stable, showing a smooth change with an approximate periodicity; while under arc fault conditions, the amplitude envelope exhibits violent fluctuations and irregular ups and downs, often accompanied by spikes or abrupt changes. These amplitude envelope characteristics provide an important basis for subsequent statistical extraction and machine learning modeling.

[0092] In one possible embodiment, performing a Hilbert transform on the arc current signal according to the dynamic window length to obtain m amplitude envelope data specifically includes the following steps:

[0093] 341. Extract the arc current signal corresponding to the dynamic window length from the arc current signal to obtain m arc current signal segments; the time interval between each of the m arc current signal segments is... t ;

[0094] 342. Based on a preset Hilbert transform formula, calculate each of the m arc current signal segments to obtain m arc current data; wherein, the Hilbert transform formula is:

[0095]

[0096] in, x_hat(t) For time points t Arc current data, Let P be the integral variable, and let P be the Cauchy principal value integral. This is a segment of the arc current signal. t It is a time variable;

[0097] 343. Determine the amplitude envelope of each of the m arc current data to obtain the m amplitude envelope data.

[0098] The arc current signal segment is a time-series data segment extracted according to a dynamic window length. The dynamic window length is adaptively adjusted based on the signal-to-noise ratio (SNR) parameter. Therefore, in low SNR scenarios, the segment time interval t is relatively long to ensure noise smoothing and enhance statistical stability; in high SNR scenarios, the segment time interval t is relatively short to ensure that the transient characteristics of the arc are not smoothed out. A Hilbert transform is performed on the signal segment x(t) with a window length of L, and the result is a complex signal. The amplitude envelope can be calculated using the magnitude of this complex signal.

[0099]

[0100]

[0101] in, Indicates the amplitude envelope. The original signal is represented by x_hat(t), which represents the arc current data at time point t. Let P be the integral variable, and let P be the Cauchy principal value integral. This is a segment of the arc current signal. t It is a time variable;

[0102] Specifically, amplitude envelope data exhibits the following characteristics: under normal load current, the amplitude envelope curve shows regularity and smoothness with relatively small fluctuations; under arc fault conditions, the amplitude envelope curve displays spikes, abrupt changes, irregular fluctuations, and high-frequency oscillations. These significant differences provide a foundation for subsequent statistical feature extraction. For example, based on m amplitude envelope data points, indicators such as envelope mean, variance, kurtosis, and climax factor can be further calculated to distinguish arc fault signals from normal load signals.

[0103] It should be noted that amplitude envelope data not only reflects the characteristics of the arc signal in the time domain, but can also complement the frequency domain features. For example, performing a Fast Fourier Transform on the envelope curve can yield the energy distribution of the arc signal at different frequencies. By combining amplitude envelope features with frequency domain energy features, a more comprehensive time-frequency domain feature vector can be formed, thereby improving the performance of the LightGBM-based classifier in arc fault identification.

[0104] Step S350: Extract the statistical features of the arc current signal in the time domain from the m amplitude envelope data to obtain p time domain statistical features; p is greater than or equal to m.

[0105] Among them, time-domain statistical features refer to a set of numerical quantification indicators calculated based on the variation law of the arc current signal amplitude envelope in the time domain. These statistical features can reflect the characteristics of the arc current signal in different time periods, such as mean, volatility, peak anomaly degree, and non-Gaussianity. Compared with the original current signal, amplitude envelope data can more intuitively reveal the fluctuation law under arc fault conditions. Therefore, extracting time-domain features based on this can effectively enhance the robustness and accuracy of arc fault identification.

[0106] The time-domain statistical features include: Mean, which characterizes the average energy level of the arc current signal over a period of time and reflects the stability of the overall power of the arc signal; Variance, which reflects the magnitude of signal fluctuations, with arc signals typically exhibiting a larger variance; Kurtosis, which measures the peaking of the signal distribution, with higher kurtosis values ​​in arc signals indicating abnormal peaks in the waveform; Skewness, which characterizes the symmetry of the signal distribution, as arc signals may exhibit waveform shifts, resulting in differences in skewness characteristics compared to normal signals; Crest Factor, the ratio of peak value to root mean square value, an important characteristic for measuring instantaneous peak energy in arc signals; Form Factor, the ratio of root mean square value to average value, used to characterize the overall waveform characteristics of the signal; and Impulse Factor, the ratio of signal peak value to average absolute value, used to reflect the intensity of abnormal pulses in the arc signal. These statistical characteristics can reflect the non-stationarity and randomness of arc current signals from different dimensions, enabling subsequent classification models to better distinguish between normal current and arc current.

[0107] Step S360: The arc current signal is calculated using a preset Fourier transform method to obtain arc current spectrum data.

[0108] The Fourier transform decomposes a time-domain signal into a superposition of sine and cosine components of different frequencies, thus characterizing the energy distribution of the signal in the frequency domain. Because arc current signals exhibit strong non-stationarity and randomness in the time domain, relying solely on time-domain statistical characteristics is insufficient to fully characterize arc properties. In the frequency domain, however, arc signals typically display rich high-frequency components and harmonic characteristics. Therefore, using the Fourier transform method to calculate the arc current signal yields spectral data that intuitively reflects the frequency distribution pattern of the arc signal, providing an important basis for the detection and classification of arc faults. Arc current spectral data refers to the energy distribution sequence of the arc current signal at different frequencies. Compared to normal load current, the spectral characteristics of arc current show significant differences: under normal operating conditions, the spectral energy is mainly concentrated near the power frequency and its lower harmonics, with relatively limited high-frequency energy; while under arc fault conditions, the spectrum shows abnormal energy enhancement in the mid-to-high frequency region, accompanied by irregular harmonics and broadband noise.

[0109] Specifically, the pre-defined Fourier transform calculation process is as follows: First, the arc current signal is preprocessed. Preprocessing includes removing the DC component to eliminate the impact of baseline drift on spectral analysis and normalization to ensure comparability of signals with different amplitude levels in the frequency domain. Additionally, windowing methods such as Hanning and Hamming windows can be used to reduce spectral leakage caused by window function effects during the Fourier transform. Next, the preprocessed arc current signal is calculated using a Fast Fourier Transform. Power Spectral Density (PSD) analysis further quantifies the energy distribution of the signal at different frequencies. The formula for calculating the power spectral density is:

[0110]

[0111] Where N is the signal strength, f For frequency, The signal is after Fourier transform. This represents the power spectral density.

[0112] Finally, the spectral data is processed and output to obtain the arc current spectral data. This data can be represented as a two-dimensional sequence of frequency-amplitude or frequency-energy, which can be used to directly observe the frequency domain characteristics of the arc signal, or as input for subsequent feature extraction and classification modeling.

[0113] Step S370: Process the arc current spectrum data to obtain q frequency domain features; q is a positive integer greater than 1.

[0114] Frequency domain features are a class of characteristic quantities extracted based on the distribution patterns of arc current spectrum data across different frequency ranges. They are used to characterize key properties of arc current signals in the frequency domain, such as energy distribution, harmonic distortion, and spectral energy concentration. Frequency domain features can characterize the essential features of arc signals from a frequency domain perspective. Because arc fault signals exhibit significant high-frequency components and irregular harmonics in the frequency domain, extracting frequency domain features can significantly improve the distinguishability between arc faults and normal load conditions. The frequency domain features include: frequency band energy distribution characteristics, which divide the spectrum into several sub-bands and calculate the energy proportion of each band to reflect the energy distribution of the arc signal in different frequency ranges; the ratio of the dominant frequency amplitude to the fundamental frequency amplitude, which quantifies the degree of harmonic distortion of the arc signal by calculating the ratio of the amplitude of the dominant frequency point or harmonic point to the amplitude of the power frequency fundamental wave; spectral entropy, which measures the uniformity of the spectral energy distribution using the definition of information entropy; the more dispersed the spectrum, the higher the spectral entropy value, reflecting the complexity of the arc signal; center frequency and spectral shift, which reflects the overall shift of the spectral distribution by calculating the weighted average frequency of the spectral energy; and bandwidth characteristics, which calculate the effective bandwidth of the energy distribution in the spectrum to characterize the frequency domain expansion of the arc signal. These frequency domain features not only reflect the differences between the arc signal and normal current, but also provide highly discriminative feature inputs for subsequent machine learning models.

[0115] In one possible embodiment, processing the arc current spectrum data to obtain q frequency domain features specifically includes the following steps:

[0116] 371. Obtain historical arc current fault signals;

[0117] 372. High-frequency pulse features are extracted from the historical arc current fault signal to obtain high-frequency pulse features;

[0118] 373. The historical arc current fault signal is transformed using the Fourier transform method, and the frequency domain harmonic distribution characteristics are extracted to obtain the harmonic distribution characteristics.

[0119] 374. Determine the historical arc current energy distribution characteristics corresponding to the historical arc current fault signal based on the high-frequency pulse characteristics and the harmonic distribution characteristics;

[0120] 375. Determine the historical power spectral density corresponding to the historical arc current energy distribution characteristics;

[0121] 376. Based on a preset significant frequency band detection method, a preset energy threshold, and the historical power spectral density, determine the adaptive frequency band parameters;

[0122] 377. Divide the arc current spectrum data based on the adaptive frequency band parameters to obtain q first arc current spectrum data;

[0123] 378. Extract frequency domain features from the q first arc current data to obtain the q frequency domain features.

[0124] Historical arc current fault signals refer to multiple representative arc fault current signals collected and stored by monitoring devices during the long-term operation of the power system. These signals exhibit typical high-frequency pulse and harmonic distortion characteristics, thus serving as prior samples for frequency domain feature extraction. High-frequency pulse characteristics are transient high-frequency components generated during the intermittent arc discharge. Specifically, high-frequency pulse characteristics can be extracted using methods such as bandpass filters, short-time Fourier transforms, or wavelet packet decomposition. Their main components include pulse amplitude, frequency of occurrence, and duration; these parameters reflect the suddenness and instability of high-frequency energy during arc faults. Harmonic distribution characteristics refer to the amplitude distribution and proportion of each order of harmonics in the current spectrum obtained through Fourier transform. Arc fault signals typically exhibit significant fundamental distortion and a substantial increase in higher-order harmonic content; therefore, harmonic distribution characteristics can directly reflect the difference between arc signals and normal current signals. Specifically, harmonic distribution characteristics include total harmonic distortion (THD), the proportion of odd harmonics, the proportion of even harmonics, and the energy proportion of the 3rd / 5th harmonics.

[0125] Energy distribution characteristics refer to the regularity of energy distribution with frequency within the frequency spectrum. The transient energy corresponding to high-frequency pulse characteristics and the steady-state energy corresponding to harmonic distribution characteristics are weighted and fused to obtain the energy distribution curve of historical arc signals. This curve reflects the energy proportions of the arc signal in the low-frequency region (fundamental and low-order harmonics), mid-frequency region (middle-order harmonics), and high-frequency region (high-frequency pulses). Power spectral density (PSD) is used to characterize the power distribution of the signal within a unit frequency bandwidth. By calculating the spectral power of the historical arc current energy distribution characteristics, its power spectral density function can be obtained. Power spectral density quantifies the concentration and range of signal energy at different frequencies, typically manifested as energy peaks in a specific frequency band.

[0126] Specifically, after obtaining the above features, they are combined to form a q-dimensional frequency domain feature vector, which serves as the input for subsequent arc fault detection. It should be noted that the value of q can be adjusted according to actual needs; either fewer frequency domain features can be selected to reduce computational complexity, or more frequency domain features can be selected to improve detection accuracy and robustness.

[0127] In one possible embodiment, dividing the arc current spectrum data based on the adaptive frequency band parameters to obtain q first arc current spectrum data specifically includes the following steps:

[0128] 3771. Determine the frequency range in the adaptive frequency band parameters;

[0129] 3772. Extract the spectrum data corresponding to the frequency range from the arc current spectrum data to obtain the candidate arc current dataset;

[0130] 3773. Perform energy distribution calculation on each candidate arc current data in the candidate arc current dataset to obtain the candidate arc current energy set;

[0131] 3774. Statistically analyze the energy parameters of all candidate arc currents in the candidate arc current energy set to obtain the candidate arc current energy parameters;

[0132] 3775. Determine the energy parameters in the arc current spectrum data to obtain the first arc current energy parameter;

[0133] 3776. Determine the energy percentage parameter based on the candidate arc current energy parameter and the first arc current energy parameter;

[0134] 3777. Determine the spectral entropy parameter based on the candidate arc current energy parameter;

[0135] 3778. Adjust the frequency range according to the energy ratio parameter and the spectral entropy parameter to obtain the target frequency range;

[0136] 3779. Extract the arc current spectrum data of the target frequency range from the arc current spectrum data to obtain the q first arc current spectrum data.

[0137] The adaptive frequency band parameters are obtained through frequency domain energy analysis and significant frequency band detection methods of historical arc current signals, and mainly include information such as the start frequency, end frequency, and center frequency. The adaptive frequency band parameters can dynamically characterize the energy concentration region of the arc signal in the frequency domain.

[0138] Specifically, the candidate arc current dataset refers to a subset of spectral data falling within the adaptive frequency band of the original spectrum. This is achieved through bandpass filtering, index filtering, or frequency range pruning. The extracted candidate arc current dataset is significantly smaller than the original spectral data, thus reducing data redundancy and computational complexity while retaining core information containing arc characteristics. Energy distribution calculation is performed by integrating or squaring the power spectral density at each frequency point, used to characterize the energy distribution characteristics of the arc signal in a specific frequency range. The candidate arc current energy set contains multiple energy values, reflecting the energy concentration in different frequency ranges. For example, when an arc occurs, significant energy peaks often appear in certain high-frequency ranges; these peaks are crucial for subsequent feature identification. Candidate arc current energy parameters include total energy, mean, variance, energy peaks, and their corresponding frequency positions. These statistics reflect the overall shape and fluctuation of the energy distribution within the candidate frequency band. If the energy is concentrated at a few frequency points, it indicates a strong correlation between this range and arc characteristics; if the energy is relatively dispersed, it indicates a smaller contribution from this range. The first arc current energy parameter is an overall energy index statistically obtained across the entire frequency spectrum, used as a benchmark for comparing the energy of candidate intervals. This parameter can be calculated by integrating the power spectral density across the entire frequency range. The principle for adjusting the frequency range is as follows: when the energy proportion is too low, it indicates insufficient energy in the frequency band, and the band can be eliminated; when the spectral entropy is too high, it indicates that the energy is too dispersed, and the frequency band can be narrowed to concentrate on the main energy region; when the energy proportion is high and the spectral entropy is low, the frequency band is retained and appropriately expanded to enhance the expressive power of the arc features. Through the joint constraint of energy proportion and spectral entropy, the frequency band boundary can be dynamically corrected, thereby obtaining a more accurate target frequency range. The resulting q first arc current spectrum data correspond to several optimized sub-bands, each retaining the frequency domain information most relevant to the arc features. Compared with fixed frequency band division, the q first arc current spectrum data generated by this method have higher information density and stronger feature correlation, which can significantly improve the accuracy and robustness of subsequent frequency domain feature extraction and classification model training.

[0139] In one possible embodiment, the step of extracting frequency domain features from the q first arc current data to obtain the q frequency domain features specifically includes the following steps:

[0140] 3781. Normalize the q first arc current spectrum data to obtain q second arc current spectrum data;

[0141] 3782. Extract the primary harmonic amplitude and secondary harmonic amplitude from each of the q secondary arc current spectrum data to obtain q primary harmonic amplitudes and q secondary harmonic amplitudes;

[0142] 3783. Determine the proportion of each of the q harmonic amplitudes in each of the q second arc current data to obtain the q harmonic proportions;

[0143] 3784. Determine the q frequency domain characteristics based on the amplitudes of the q primary harmonics and the proportions of the q secondary harmonics.

[0144] Normalization is used to eliminate differences in sampling amplitude and spectral energy among different arc current signals, ensuring comparability between different samples. The normalization method can employ amplitude normalization, energy normalization, or maximum / minimum value standardization, etc., and is not limited here. The primary harmonic amplitude refers to the amplitude of the frequency component closest to the fundamental frequency and its integer multiples; it is the component with the highest energy proportion and strongest stability in the arc current spectrum. The secondary harmonic amplitude refers to the amplitude of significant secondary frequency components other than the primary harmonics. These components often appear during arc discharge and exhibit nonlinear and high-frequency pulse characteristics. Extracting the primary and secondary harmonic amplitudes helps construct a unique spectral fingerprint of the arc. The secondary harmonic ratio is obtained by comparing the secondary harmonic amplitude with the primary harmonic amplitude or the total amplitude. The secondary harmonic ratio reflects the energy proportion of the arc current signal in higher frequency components. When an electric arc occurs, the subharmonic components are significantly enhanced compared to the signal under normal operating conditions, resulting in an increased subharmonic proportion. The subharmonic proportion is a crucial parameter for characterizing arc faults. After extracting the primary harmonic amplitude and the subharmonic proportion, they are further combined using vectorization to form a complete set of frequency domain features. These q frequency domain features encompass both the low-frequency stable components (primary harmonics) and the high-frequency disturbance components (subharmonic proportion), enabling a complete characterization of the energy distribution characteristics of the arc signal within the frequency domain.

[0145] Specifically, to ensure the robustness and real-time performance of frequency domain feature extraction, the primary harmonic detection method is first optimized by employing a fast Fourier transform combined with a peak search algorithm to quickly locate the amplitude corresponding to the primary harmonic frequency. When grid frequency drift exists, an adaptive filtering method can be introduced to dynamically track the primary harmonic frequency. Next, the secondary harmonic identification method is optimized by combining threshold detection with energy clustering to accurately identify secondary harmonic components among multiple frequency components, avoiding interference from normal background noise. Then, the ratio calculation is standardized. To eliminate the influence of different sampling window sizes on the secondary harmonic ratio, energy normalization is introduced to ensure consistency of the calculation results across operating conditions. Finally, feature vectors are constructed, resulting in q frequency domain feature vectors. These vectors reflect both the main energy distribution of the arc current spectrum and the high-frequency disturbance characteristics during arc generation, and are expressed as follows:

[0146]

[0147] in, This indicates the corresponding principal harmonic ratio. This indicates the corresponding subharmonic ratio. F It represents the frequency domain characteristics.

[0148] Step S380: Based on a preset feature fusion algorithm, the p time-domain statistical features and the q frequency-domain features are fused to obtain the target arc current features.

[0149] The purpose of feature fusion processing is to unify the multi-dimensional information from the time and frequency domains into a single feature vector with higher discriminative power. Arc current signals exhibit irregular pulsations, mean shifts, and enhanced fluctuations in the time domain, while in the frequency domain they show prominent primary and secondary harmonic characteristics and uneven energy distribution. Therefore, jointly representing the time and frequency domain features helps to comprehensively characterize the dynamic and spectral characteristics of arc fault signals, thus providing high-quality input for subsequent detection and identification based on machine learning models. Feature fusion algorithms can be simple feature-level concatenation methods, weighted strategy-based fusion methods, or fusion methods implemented through dimensionality reduction algorithms (such as Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA)) or deep learning feature encoding methods (such as autoencoders and attention mechanisms). No specific limitations are imposed here. The target arc current feature is a fused comprehensive feature vector, which includes p time-domain statistical features (such as mean, variance, kurtosis, skewness, short-time energy, etc.) and q frequency-domain features (such as main harmonic amplitude, subharmonic ratio, spectral energy ratio, spectral entropy, etc.), and can comprehensively reflect the performance of the arc signal in different domains.

[0150] In one possible embodiment, the step of fusing the p time-domain statistical features and the q frequency-domain features based on a preset feature fusion algorithm to obtain the target arc current features specifically includes the following steps:

[0151] 381. Obtain historical arc current data;

[0152] 382. Extract time-domain statistical features from the historical arc current data to obtain multiple historical time-domain statistical features;

[0153] 383. Extract the characteristic frequency domain from the historical arc current data to obtain multiple historical frequency domain features;

[0154] 384. Construct a historical arc current dataset based on the multiple historical time-domain statistical features, the multiple historical frequency-domain features, and the historical arc current data;

[0155] 385. Input the historical arc current dataset into a preset dimensionality reduction model to obtain the dimensionality reduction projection matrix;

[0156] 386. Input the plurality of historical time-domain statistical features and the plurality of historical frequency-domain features into a preset lightweight machine learning classification model to obtain a first classification accuracy and a second classification accuracy; the first classification accuracy is the classification accuracy output by the lightweight machine learning classification model based on the plurality of historical time-domain statistical features; the second classification accuracy is the classification accuracy output by the lightweight machine learning classification model based on the plurality of historical frequency-domain features.

[0157] 387. Determine the time-domain feature weights and frequency-domain feature weights based on the first classification accuracy and the second classification accuracy;

[0158] 388. Determine the first arc current characteristic based on the p time-domain statistical characteristics, the q frequency-domain characteristics, the time-domain characteristic weights, and the frequency-domain characteristic weights;

[0159] 389. Project the first arc current feature based on the dimensionality reduction projection matrix to obtain the target arc current feature.

[0160] Historical arc current data is a dataset obtained by long-term collection of current waveforms from power systems experiencing arc faults or similar load disturbances. This data contains a large number of current time-series signals under different operating conditions, including both currents from normal load fluctuations and abnormal current signals generated by arc discharge. Arc fault currents typically exhibit characteristics such as short-duration bursts, irregular amplitudes, and spectral diffusion, making them highly suitable as foundational samples for training and validation. Dimensionality reduction models are mathematical models capable of extracting the most representative information from a high-dimensional feature space. These can be statistical methods based on Principal Component Analysis (PCA) or Linear Discriminant Analysis (LDA), or deep learning methods based on autoencoders, convolutional neural networks, etc., without limitation. Dimensionality reduction models generate projection matrices to map high-dimensional time-domain and frequency-domain features to a low-dimensional subspace, thereby reducing feature redundancy and improving the compactness of feature representation and the generalization performance of the classifier. Lightweight machine learning classification models can be models such as LightGBM, XGBoost, Random Forest, or Support Vector Machines, which maintain high efficiency with limited computing resources. In arc fault detection scenarios, LightGBM, with its efficient decision tree construction mechanism based on gradient boosting, can reduce computational complexity while maintaining classification accuracy, making it particularly suitable for embedded or edge computing devices. The temporal and frequency domain feature weights are quantized coefficients obtained by comparing the classification accuracy of classification models using a single feature source.

[0161] Specifically, firstly, historical arc current data is generated through on-site acquisition or a simulation platform and preprocessed, such as through denoising, normalization, and slicing, to ensure data consistency and usability. Then, time-domain statistical features are extracted from the preprocessed arc current data. Next, frequency-domain analysis is performed on the same batch of historical arc current data, using methods such as Fast Fourier Transform (FFT) and wavelet packet decomposition to extract frequency-domain features. Then, the extracted historical time-domain statistical features, historical frequency-domain features, and the corresponding original arc current data are combined to construct a complete historical arc current dataset. This dataset contains input features and corresponding labels (normal state or arc fault state) for subsequent model training and evaluation. The historical arc current dataset is then input into a dimensionality reduction model to obtain a dimensionality-reduced projection matrix. For example, if the PCA method is used, the projection matrix consists of the first few eigenvectors of the data covariance matrix.

[0162] Then, a lightweight machine learning classification model is trained using historical time-domain statistical features and historical frequency-domain features as inputs, and its classification accuracy on the validation set is calculated. The obtained first and second classification accuracies reflect the contributions of time-domain and frequency-domain features to arc fault classification, respectively. Based on the first and second classification accuracies, a weighted average method is used to determine the weights of the time-domain features and the frequency-domain features. These weights are then used to weight p time-domain statistical features and q frequency-domain features to obtain the first arc current feature. This feature vector numerically integrates time-domain and frequency-domain information and adjusts their importance through weights. Its weight allocation formula can be expressed as:

[0163]

[0164] in, This represents the accuracy obtained by using only temporal features as input to a lightweight classification model. This represents the accuracy obtained by using only frequency domain features as input to a lightweight classification model.

[0165] Finally, the first arc current feature is input into the dimension-reduced projection matrix to complete the feature space mapping and obtain the low-dimensional target arc current feature. This feature retains the discriminative information of the original feature while reducing redundancy, thus improving detection efficiency and the model's generalization performance.

[0166] Step S390: Input the target arc current characteristics into a preset lightweight arc fault detection model to obtain the arc fault detection result.

[0167] The lightweight arc fault detection model refers to a machine learning or deep learning model that can run with low computational complexity, low storage overhead, and high real-time performance while ensuring arc fault detection accuracy. This model can be a LightGBM model based on gradient boosting decision trees, an XGBoost model, an improved support vector machine model, or a lightweight convolutional neural network model suitable for embedded environments. Preferably, the LightGBM model is used as the lightweight arc fault detection model. This model has a fast training speed and low resource consumption, which can meet the application requirements of rapid arc fault detection in power distribution systems.

[0168] Specifically, firstly, the obtained target arc current features are used as input vectors and fed into the lightweight arc fault detection model. These features integrate weighted fusion information from time-domain statistical features and frequency-domain features, and undergo dimensionality reduction to ensure the discriminativeness and compactness of the feature vectors, thereby improving the effectiveness of the model input. During the model inference phase, when new target arc current features are input into the model, the model performs feature matching and classification calculations based on its internal decision tree structure or classification network structure, ultimately outputting the corresponding arc fault detection result. The detection result can be a binary classification result (i.e., arc fault / non-arc fault) or a multi-class classification result (e.g., series arc fault, parallel arc fault, transient disturbance, etc.). Furthermore, the arc fault detection result can include the probability value of arc fault occurrence. The model can not only output whether an arc fault exists in the current signal at a certain moment, but also provide a confidence score for the detection result, thus providing a reference for subsequent protection strategies or scheduling decisions.

[0169] For easier understanding, please refer to Figure 4 , Figure 4 This is a schematic diagram of the structure of a feature extraction unit provided in an embodiment of this application. As can be seen, Figure 4The internal composition of the feature extraction unit and the hierarchical relationship of its modules are presented, demonstrating the complete processing flow from current signal input to time-frequency domain feature extraction, and clarifying the division of labor and cooperation mechanism between time-domain and frequency-domain feature extraction. Specifically, the feature extraction unit 400, as the core processing unit, is responsible for coordinating the extraction process of time-domain and frequency-domain features, and it contains two parallel processing sub-units: the Hilbert transform unit 410 and the fast Fourier transform unit 420. The Hilbert transform unit 410 is a time-domain feature extraction module consisting of three functional modules: a signal-to-noise ratio (SNR) estimation module 411, which segments the input current signal and calculates the signal-to-noise power ratio (SNR) to provide a basis for subsequent window length adjustment; a dynamic window adjustment module 412, which dynamically selects the window length based on the SNR estimation result (a longer window length is used to smooth noise when the SNR is low, a shorter window length is used to preserve local features when the SNR is high, and the window length in the middle region is determined by linear interpolation); and a Hilbert transform calculation module 413, which performs a Hilbert transform on the signal based on the adjusted window length, constructs the analytical signal through a fast Fourier transform (FFT), and finally extracts the amplitude envelope of the current signal. The Fast Fourier Transform (FFT) unit 420 is used for frequency domain feature extraction and includes four functional modules: a signal preprocessing module 421, which performs mean removal and normalization on the input current signal to reduce the impact of baseline drift and amplitude differences on subsequent analysis; a Fourier calculation module 422, which performs a Fast Fourier Transform (FFT) on the preprocessed signal to obtain the signal's spectral distribution; a dynamic frequency band segmentation module 423, which adaptively divides significant frequency bands based on the power spectral density (PSD) calculation results and an energy threshold (determined by an empirical coefficient based on the power spectral mean and standard deviation); and a frequency domain feature extraction module 424, which calculates the normalized energy, main harmonic amplitude, and subharmonic proportion of each frequency band based on the segmented frequency bands, providing key inputs for subsequent classification models. This structure, through the collaborative work of the Hilbert Transform (HFT) unit and the Fast Fourier Transform (FFT) unit, achieves efficient extraction of the time-frequency domain features of the current signal, providing a comprehensive feature input foundation for the accurate discrimination of the LightGBM classification model.

[0170] For easier understanding, please refer to Figure 5 , Figure 5 This is a schematic diagram of the structure of an AC arc fault detection system provided in an embodiment of this application. As can be seen, Figure 5The complete circuit structure for arc generation and detection, from AC power supply to the load, is clearly presented, including the connection methods of the signal source, load, measurement module, and arc generation unit, demonstrating the process of arc signal acquisition, processing, and analysis. Specifically, the AC arc fault detection system includes: AC power supply (220V / 50Hz), controllable switches (S1 and S2), arc generation device (point contact arc generator), detection device (current transformer), display device (RIGOL MSO5000 oscilloscope), and load. The AC power supply serves as the system's signal input, providing a stable AC current to simulate the power supply process in an actual power distribution network. An arc generation device is connected in series between the power supply and the load. This device can intentionally induce an arc phenomenon by controlling the on / off state of the switch or by gap breakdown, thus creating a typical arc fault environment for detection and analysis. At the output of the arc generation device, a detection module is connected. This module can be a current transformer, voltage transformer, or high-precision sampling circuit, and its function is to acquire and preprocess the arc current signal in real time. The acquired raw current signal, after being filtered and amplified, is transmitted to the display device and subsequent data processing unit to ensure signal quality and the accuracy of subsequent feature extraction. Furthermore, the display device employs a digital oscilloscope or power analyzer to display the arc current waveform, spectral distribution, and amplitude characteristics in real time, facilitating observation of the arc signal's variation patterns at different times. By monitoring the displayed results, researchers can intuitively determine whether an arc has occurred, its duration, and its characteristic intensity. In addition, the display device can output digitized arc current data, providing experimental data support for subsequent arc fault detection methods based on Hilbert transform and the LightGBM algorithm. At the load end, a variable impedance load or a typical household appliance model is connected to simulate the performance characteristics of arc faults under different operating scenarios. For example, when the load impedance is low, the high-frequency components of the arc signal are significantly enhanced; while when the load impedance is high, the arc signal exhibits obvious intermittent characteristics. By adjusting the load parameters, arc faults under different operating conditions in the actual power grid can be simulated, providing diverse data samples for algorithm training and model verification. In addition, the arc generating device is equipped with a controllable switch or mechanical contact device to generate an arc discharge process periodically or randomly. By adjusting the switching frequency and contact resistance, strong arc, weak arc or intermittent arc signals can be obtained.

[0171] For easier understanding, please refer to Figure 6 , Figure 6This application provides a flowchart illustrating the feature extraction and classification process for an AC arc fault detection method. Specifically, the arc current signal, as input data, first enters the signal processing stage. During this process, the arc current features are processed through two feature channels: first, the Hilbert transform is used to extract the envelope features of the arc signal, obtaining the instantaneous amplitude and instantaneous energy distribution of the arc current; second, Fourier transform is used to perform spectral analysis on the arc current, obtaining the amplitude characteristics and harmonic components of the arc current in different frequency bands. The Hilbert transform effectively reveals the instability and intermittent nature of the arc signal in the time domain, while the Fourier transform can characterize its harmonic distortion characteristics in the frequency domain. The combination of the two can comprehensively characterize the multidimensional features of the arc fault. After the transformation is completed, all time-domain and frequency-domain features are input to the feature extraction module for feature extraction. Feature extraction normalizes, statistically calculates, and filters the above data to ensure that the obtained feature vector has high discriminative power and robustness. For example, it can extract various time-frequency domain statistical features of arc current, such as the amplitude of the main harmonic, the proportion of the second harmonic, the spectral entropy parameter, the root mean square value, kurtosis, and skewness. Through the comprehensive expression of these features, it can effectively distinguish between arc fault states and normal operating states, overcoming the limitation that a single feature is insufficient to fully reflect the characteristics of the arc. The extracted feature vectors are then input into the lightweight machine learning model LightGBM (LGBM). LightGBM, based on the gradient boosting decision tree concept, has advantages such as efficient training, low memory consumption, and strong generalization ability, making it particularly suitable for rapid arc fault detection in scenarios with limited hardware resources. During the model inference phase, LGBM traverses the decision tree layer by layer according to the input feature vectors, ultimately outputting the detection result. If the model determines that the signal features are highly consistent with the features of normal operating conditions, it outputs "Normal"; if the features have a high degree of matching with the features of arc operating conditions, it outputs "Arc," thus achieving accurate identification of arc faults. This detection result not only provides real-time early warning for power grid operation but also provides a basis for decision-making in subsequent protection and control actions.

[0172] For easier understanding, please refer to Figure 7 , Figure 7This application provides a schematic diagram of an arc fault detection process based on LightGBM. As can be seen, firstly, AC current signals are acquired in real time and input into a preprocessing module for processing. The acquired AC arc current signal is a waveform sequence, which is then processed by the preprocessing module. The main function of preprocessing is to remove common background noise, power frequency interference, and random disturbances in power systems, thereby ensuring the accuracy and stability of subsequent feature extraction. Next, an adaptive Hilbert transform method is used to optimize the extraction of the amplitude envelope by dynamically adjusting the window based on the signal-to-noise ratio. Specifically, the arc current signal is divided based on adaptive frequency band parameters, extracting energy features and spectral entropy features at different frequency bands, and constructing a preliminary multidimensional feature vector. This adaptive division method can reduce redundant features while highlighting high-frequency feature information related to the arc, helping to improve the discriminative power of the feature representation. Then, Fast Fourier Transform (FFT) is used to perform spectral analysis on the real-time acquired AC current signal. Dynamic frequency band segmentation technology is employed to adaptively select the analysis frequency band based on the characteristics of the arc fault signal to extract frequency domain features. The statistical features of the time and frequency domains are then combined to extract a feature vector. Specifically, after initial feature extraction, a feature fusion and optimization mechanism is introduced. First, a weighted fusion strategy is used to combine time-domain statistical features and frequency-domain features, achieving balance and complementarity among different features through reasonable allocation of weight parameters. Then, Principal Component Analysis (PCA) is used to reduce the dimensionality of the fused high-dimensional features, retaining the main component features to reduce computational overhead and avoid interference from feature correlations. Through these steps, the optimized target arc current features are finally obtained. Finally, the LightGBM lightweight gradient boosting decision tree model is used for classification training and inference on the feature vector, outputting the discrimination result of arc fault or normal state. Specifically, the optimized features are input into the classification module for modeling and recognition. By using the Lightweight Gradient Boosting Tree (LightGBM) model as the classifier, it can significantly reduce computational complexity while maintaining high classification accuracy. This makes it suitable for deployment scenarios of embedded or edge computing devices and improves detection accuracy.

[0173] As can be seen, the lightweight arc fault detection method described in this application is applied to electronic equipment. The method acquires the arc current signal of the load in the target distribution network within a preset time period, determines the signal-to-noise ratio (SNR) parameter of the arc current signal, determines the dynamic window length based on the SNR parameter, performs a Hilbert transform on the arc current signal based on the dynamic window length to obtain m amplitude envelope data, where m is a positive integer greater than 1, extracts the statistical features of the arc current signal in the time domain from the m amplitude envelope data to obtain p time-domain statistical features, where p is greater than or equal to m, calculates the arc current signal using a preset Fourier transform method to obtain arc current spectrum data, processes the arc current spectrum data to obtain q frequency domain features, where q is a positive integer greater than 1, fuses the p time-domain statistical features and q frequency domain features based on a preset feature fusion algorithm to obtain the target arc current features, and inputs the target arc current features into a preset lightweight arc fault detection model to obtain the arc fault detection result. Thus, on the one hand, by using Hilbert transform to extract the current signal envelope, further calculating statistics such as envelope mean, peak factor, and kurtosis, and applying fast Fourier transform to extract indicators such as harmonic components and spectral energy ratio within a specific frequency band, and combining time-domain and frequency-domain features for prediction, the accuracy of arc fault detection is improved; on the other hand, a lightweight classification model is used to detect normal arc current and fault arc current based on the extracted features, thereby reducing the computational overhead of arc fault detection.

[0174] The above primarily describes the solutions of the embodiments of this application from the perspective of the method execution process. It is understood that, in order to achieve the above functions, the electronic device includes corresponding hardware structures and / or software modules for executing each function. Those skilled in the art should readily recognize that, in conjunction with the units and algorithm steps of the various examples described in the embodiments provided herein, this application can be implemented in hardware or a combination of hardware and computer software. Whether a function is executed by hardware or by computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0175] This application embodiment can divide the electronic device into functional units according to the above method example. For example, each function can be divided into a separate functional unit, or two or more functions can be integrated into one processing unit. The integrated unit can be implemented in hardware or as a software functional unit. It should be noted that the unit division in this application embodiment is illustrative and only represents one logical functional division. In actual implementation, there may be other division methods.

[0176] When dividing each function into modules according to its corresponding function. Figure 8 This is a functional module block diagram of a lightweight arc fault detection device 800 provided in an embodiment of this application. The lightweight arc fault detection device 800 includes:

[0177] The acquisition module 810 is used to acquire the arc current signal of the load in the target distribution network within a preset time period;

[0178] The determination module 820 is used to determine the signal-to-noise ratio parameter of the arc current signal; determine the dynamic window length based on the signal-to-noise ratio parameter; and calculate the arc current signal using a preset Fourier transform method to obtain arc current spectrum data.

[0179] The calculation module 830 is used to perform a Hilbert transform on the arc current signal according to the dynamic window length to obtain m amplitude envelope data; m is a positive integer greater than 1; extract statistical features of the arc current signal in the time domain from the m amplitude envelope data to obtain p time-domain statistical features; p is greater than or equal to m; process the arc current spectrum data to obtain q frequency-domain features; q is a positive integer greater than 1; and fuse the p time-domain statistical features and the q frequency-domain features based on a preset feature fusion algorithm to obtain the target arc current features.

[0180] The control module 840 is used to input the target arc current characteristics into a preset lightweight arc fault detection model to obtain the arc fault detection result.

[0181] In one possible embodiment, the determining module 820, in determining the dynamic window length based on the signal-to-noise ratio parameter, is specifically configured to:

[0182] Obtain the preset low signal-to-noise ratio threshold, high signal-to-noise ratio threshold, minimum window length, and maximum window length;

[0183] The arc current signal corresponding to the minimum window length is extracted from the arc current signal to obtain the first arc current signal set;

[0184] The arc current signal corresponding to the maximum window length is extracted from the arc current signal to obtain the second arc current signal set;

[0185] The first signal power and the first noise power are obtained by calculating all the first arc current signals in the first arc current signal set based on the preset root mean square estimation method.

[0186] The second signal power and the second noise power are calculated based on the root mean square estimation method for all the second arc current signals in the second arc current signal set.

[0187] The first signal-to-noise ratio parameter is determined based on the first signal power and the first noise power;

[0188] The second signal-to-noise ratio parameter is determined based on the second signal power and the second noise power;

[0189] A reference signal-to-noise ratio parameter is determined based on the first signal-to-noise ratio parameter and the second signal-to-noise ratio parameter;

[0190] If the signal-to-noise ratio parameter is less than or equal to the low signal-to-noise ratio threshold, then the dynamic window length is determined based on the maximum window length, the reference signal-to-noise ratio parameter, and the signal-to-noise ratio parameter.

[0191] If the signal-to-noise ratio parameter is greater than or equal to the high signal-to-noise ratio threshold, then the dynamic window length is determined based on the minimum window length, the high signal-to-noise ratio threshold, and the signal-to-noise ratio parameter.

[0192] If the signal-to-noise ratio parameter is greater than the low signal-to-noise ratio threshold and less than the high signal-to-noise ratio threshold, then the dynamic window length is determined based on the preset interpolation formula, the low signal-to-noise ratio threshold, the high signal-to-noise ratio threshold, the minimum window length, the maximum window length, and the signal-to-noise ratio parameter.

[0193] In one possible embodiment, the calculation module 830 is specifically used for performing a Hilbert transform on the arc current signal according to the dynamic window length to obtain m amplitude envelope data:

[0194] The arc current signal corresponding to the dynamic window length is extracted from the arc current signal to obtain m arc current signal segments; the time interval of each of the m arc current signal segments is... t ;

[0195] Based on a preset Hilbert transform formula, each of the m arc current signal segments is calculated to obtain m arc current data; wherein, the Hilbert transform formula is:

[0196]

[0197] in, x_hat(t) For time points t Arc current data, Let P be the integral variable, and let P be the Cauchy principal value integral. This is a segment of the arc current signal. t It is a time variable;

[0198] The amplitude envelope of each of the m arc current data is determined to obtain the m amplitude envelope data.

[0199] In one possible embodiment, the calculation module 830, in processing the arc current spectrum data to obtain q frequency domain features, is specifically used for:

[0200] Acquire historical arc current fault signals;

[0201] High-frequency pulse features are extracted from the historical arc current fault signals to obtain high-frequency pulse features;

[0202] The historical arc current fault signal is transformed using the Fourier transform method, and the frequency domain harmonic distribution characteristics are extracted to obtain the harmonic distribution characteristics.

[0203] The energy distribution characteristics of the historical arc current corresponding to the historical arc current fault signal are determined based on the high-frequency pulse characteristics and the harmonic distribution characteristics.

[0204] Determine the historical power spectral density corresponding to the historical arc current energy distribution characteristics;

[0205] The adaptive frequency band parameters are determined based on a preset significant frequency band detection method, a preset energy threshold, and the historical power spectral density.

[0206] The arc current spectrum data is divided based on the adaptive frequency band parameters to obtain q first arc current spectrum data.

[0207] Frequency domain features are extracted from the q first arc current data to obtain the q frequency domain features.

[0208] In one possible embodiment, the calculation module 830, in dividing the arc current spectrum data based on the adaptive frequency band parameters to obtain q first arc current spectrum data, is specifically used for:

[0209] Determine the frequency range in the adaptive frequency band parameters;

[0210] Extract the spectrum data corresponding to the frequency range from the arc current spectrum data to obtain the candidate arc current dataset;

[0211] Energy distribution calculation is performed on each candidate arc current data in the candidate arc current dataset to obtain the candidate arc current energy set.

[0212] The energy parameters of all candidate arc currents in the candidate arc current energy set are statistically analyzed to obtain the candidate arc current energy parameters;

[0213] Determine the energy parameters in the arc current spectrum data to obtain the first arc current energy parameter;

[0214] The energy percentage parameter is determined based on the candidate arc current energy parameter and the first arc current energy parameter.

[0215] The spectral entropy parameter is determined based on the candidate arc current energy parameter;

[0216] The frequency range is adjusted according to the energy ratio parameter and the spectral entropy parameter to obtain the target frequency range;

[0217] The arc current spectrum data of the target frequency range is extracted from the arc current spectrum data to obtain the q first arc current spectrum data.

[0218] In one possible embodiment, the calculation module 830, in the process of fusing the p time-domain statistical features and the q frequency-domain features based on a preset feature fusion algorithm to obtain the target arc current features, is specifically used for:

[0219] Acquire historical arc current data;

[0220] Time-domain statistical features are extracted from the historical arc current data to obtain multiple historical time-domain statistical features;

[0221] The historical arc current data is subjected to feature frequency domain extraction to obtain multiple historical frequency domain features;

[0222] A historical arc current dataset is constructed based on the multiple historical time-domain statistical features, the multiple historical frequency-domain features, and the historical arc current data.

[0223] The historical arc current dataset is input into a preset dimensionality reduction model to obtain a dimensionality reduction projection matrix;

[0224] The multiple historical time-domain statistical features and the multiple historical frequency-domain features are respectively input into a preset lightweight machine learning classification model to obtain a first classification accuracy and a second classification accuracy; the first classification accuracy is the classification accuracy output by the lightweight machine learning classification model based on the multiple historical time-domain statistical features; the second classification accuracy is the classification accuracy output by the lightweight machine learning classification model based on the multiple historical frequency-domain features.

[0225] The time-domain feature weights and frequency-domain feature weights are determined based on the first classification accuracy and the second classification accuracy.

[0226] The first arc current characteristic is determined based on the p time-domain statistical features, the q frequency-domain features, the time-domain feature weights, and the frequency-domain feature weights.

[0227] The target arc current feature is obtained by projecting the first arc current feature onto the dimensionality-reduced projection matrix.

[0228] As can be seen, the embodiments described in this application describe a lightweight arc fault detection device. By using Hilbert transform to extract the current signal envelope, further calculating statistical quantities such as envelope mean, peak factor, and kurtosis, and applying fast Fourier transform to extract indicators such as harmonic components and spectral energy ratio within a specific frequency band, and comprehensively inputting the time-domain and frequency-domain features into a lightweight classification and prediction model, the accuracy of arc fault detection is improved and the computational overhead of arc fault detection is reduced.

[0229] This application also provides a computer-readable storage medium storing a computer program for electronic data interchange, which causes a computer to perform some or all of the steps of any of the methods described in the above method embodiments, wherein the computer includes an electronic device.

[0230] This application also provides a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program operable to cause a computer to perform some or all of the steps of any of the methods described in the above method embodiments. The computer program product may be a software installation package, and the computer may include an electronic device.

[0231] It should be noted that, for the sake of simplicity, the above embodiments are all described as a series of actions. Those skilled in the art should understand that this application is not limited to the described order of actions, as some steps in the embodiments of this application can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions, steps, modules, or units involved are not necessarily essential to the embodiments of this application.

[0232] In the above embodiments, the descriptions of each embodiment in this application have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0233] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. This program can be stored in a computer-readable storage medium, and when executed, it can include the processes described in the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as ROM or random access memory (RAM), magnetic disks, or optical disks.

[0234] The steps of the methods or algorithms described in the embodiments of this application can be implemented in hardware or by a processor executing software instructions. The software instructions can consist of corresponding software modules, which can be stored in RAM, flash memory, ROM, EPROM, electrically erasable programmable read-only memory (EEPROM), registers, hard disk, portable hard disk, read-only optical disk (CD-ROM), or any other form of storage medium well known in the art. An exemplary storage medium is coupled to a processor, enabling the processor to read information from and write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and storage medium can reside in an ASIC. Furthermore, the ASIC can reside in a terminal device or management device. Alternatively, the processor and storage medium can exist as discrete components in the terminal device or management device.

[0235] Those skilled in the art will recognize that, in one or more of the examples above, the functions described in the embodiments of this application can be implemented, in whole or in part, by software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. This computer program product includes one or more computer instructions. When these computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, electronic device, or data center to another via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium accessible to a computer or a data storage device such as an electronic device or data center that integrates one or more available media.

[0236] The modules / units included in the various devices and products described in the above embodiments can be software modules / units, hardware modules / units, or a combination of both. For example, for devices and products applied to or integrated into a chip, all modules / units can be implemented using hardware methods such as circuits, or at least some modules / units can be implemented using software programs that run on a processor integrated within the chip, while the remaining (if any) modules / units can be implemented using hardware methods such as circuits. For devices and products applied to or integrated into a chip module, all modules / units can be implemented using hardware methods such as circuits. Different modules / units can be located in the same component (e.g., chip, circuit module, etc.) or different components of the chip module, or at least some modules / units can be implemented using hardware methods such as circuits. The implementation is achieved through a software program that runs on the processor integrated within the chip module. The remaining modules / units (if any) can be implemented using hardware methods such as circuits. For various devices and products applied to or integrated into terminal equipment, each of their modules / units can be implemented using hardware methods such as circuits. Different modules / units can be located in the same component (e.g., chip, circuit module, etc.) or different components within the terminal equipment. Alternatively, at least some modules / units can be implemented through a software program that runs on the processor integrated within the terminal equipment, while the remaining modules / units (if any) can be implemented using hardware methods such as circuits.

[0237] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the embodiments of this application. It should be understood that the above descriptions are merely specific embodiments of the embodiments of this application and are not intended to limit the protection scope of the embodiments of this application. Any modifications, equivalent substitutions, improvements, etc., made on the basis of the technical solutions of the embodiments of this application should be included within the protection scope of the embodiments of this application.

Claims

1. A lightweight arc fault detection method, characterized in that, Applied to electronic devices, the method includes: Acquire the arc current signal of the load in the target distribution network within a preset time period; Determine the signal-to-noise ratio parameters of the arc current signal; The dynamic window length is determined based on the signal-to-noise ratio parameter. The arc current signal is subjected to Hilbert transform according to the dynamic window length to obtain m amplitude envelope data; m is a positive integer greater than 1. Statistical features of the arc current signal in the time domain are extracted from the m amplitude envelope data to obtain p time-domain statistical features; p is greater than or equal to m. The arc current signal is calculated using a preset Fourier transform method to obtain arc current spectrum data; The arc current spectrum data is processed to obtain q frequency domain features; q is a positive integer greater than 1. Based on a preset feature fusion algorithm, the p time-domain statistical features and the q frequency-domain features are fused to obtain the target arc current features; The target arc current characteristics are input into a preset lightweight arc fault detection model to obtain arc fault detection results. The step of determining the dynamic window length based on the signal-to-noise ratio parameter includes: Obtain the preset low signal-to-noise ratio threshold, high signal-to-noise ratio threshold, minimum window length, and maximum window length; The arc current signal corresponding to the minimum window length is extracted from the arc current signal to obtain the first arc current signal set; The arc current signal corresponding to the maximum window length is extracted from the arc current signal to obtain the second arc current signal set; The first signal power and the first noise power are obtained by calculating all the first arc current signals in the first arc current signal set based on the preset root mean square estimation method. The second signal power and the second noise power are calculated based on the root mean square estimation method for all the second arc current signals in the second arc current signal set. The first signal-to-noise ratio parameter is determined based on the first signal power and the first noise power; The second signal-to-noise ratio parameter is determined based on the second signal power and the second noise power; A reference signal-to-noise ratio parameter is determined based on the first signal-to-noise ratio parameter and the second signal-to-noise ratio parameter; If the signal-to-noise ratio parameter is less than or equal to the low signal-to-noise ratio threshold, then the dynamic window length is determined based on the maximum window length, the reference signal-to-noise ratio parameter, and the signal-to-noise ratio parameter. If the signal-to-noise ratio parameter is greater than or equal to the high signal-to-noise ratio threshold, then the dynamic window length is determined based on the minimum window length, the high signal-to-noise ratio threshold, and the signal-to-noise ratio parameter. If the signal-to-noise ratio parameter is greater than the low signal-to-noise ratio threshold and less than the high signal-to-noise ratio threshold, then the dynamic window length is determined based on the preset interpolation formula, the low signal-to-noise ratio threshold, the high signal-to-noise ratio threshold, the minimum window length, the maximum window length, and the signal-to-noise ratio parameter.

2. The method as described in claim 1, characterized in that, The step of performing a Hilbert transform on the arc current signal according to the dynamic window length to obtain m amplitude envelope data includes: The arc current signal corresponding to the dynamic window length is extracted from the arc current signal to obtain m arc current signal segments; the time interval of each of the m arc current signal segments is... t ; Based on a preset Hilbert transform formula, each of the m arc current signal segments is calculated to obtain m arc current data; wherein, the Hilbert transform formula is: in, x_hat(t) For time points t Arc current data, Let P be the integral variable, and let P be the Cauchy principal value integral. This is a segment of the arc current signal. t It is a time variable; The amplitude envelope of each of the m arc current data is determined to obtain the m amplitude envelope data.

3. The method as described in claim 1 or 2, characterized in that, The process of processing the arc current spectrum data yields q frequency domain features, including: Acquire historical arc current fault signals; High-frequency pulse features are extracted from the historical arc current fault signals to obtain high-frequency pulse features; The historical arc current fault signal is transformed using the Fourier transform method, and the frequency domain harmonic distribution characteristics are extracted to obtain the harmonic distribution characteristics. The energy distribution characteristics of the historical arc current corresponding to the historical arc current fault signal are determined based on the high-frequency pulse characteristics and the harmonic distribution characteristics. Determine the historical power spectral density corresponding to the historical arc current energy distribution characteristics; The adaptive frequency band parameters are determined based on a preset significant frequency band detection method, a preset energy threshold, and the historical power spectral density. The arc current spectrum data is divided based on the adaptive frequency band parameters to obtain q first arc current spectrum data. Frequency domain features are extracted from the q first arc current spectrum data to obtain the q frequency domain features.

4. The method as described in claim 3, characterized in that, The arc current spectrum data is divided based on the adaptive frequency band parameters to obtain q first arc current spectrum data, including: Determine the frequency range in the adaptive frequency band parameters; Extract the spectrum data corresponding to the frequency range from the arc current spectrum data to obtain the candidate arc current dataset; Energy distribution calculation is performed on each candidate arc current data in the candidate arc current dataset to obtain the candidate arc current energy set. The energy parameters of all candidate arc currents in the candidate arc current energy set are statistically analyzed to obtain the candidate arc current energy parameters; Determine the energy parameters in the arc current spectrum data to obtain the first arc current energy parameter; The energy percentage parameter is determined based on the candidate arc current energy parameter and the first arc current energy parameter. The spectral entropy parameter is determined based on the candidate arc current energy parameter; The frequency range is adjusted according to the energy ratio parameter and the spectral entropy parameter to obtain the target frequency range; The arc current spectrum data of the target frequency range is extracted from the arc current spectrum data to obtain the q first arc current spectrum data.

5. The method as described in claim 3, characterized in that, The step of extracting frequency domain features from the q first arc current data to obtain the q frequency domain features includes: The q first arc current spectrum data are normalized to obtain q second arc current spectrum data. The main harmonic amplitude and the second harmonic amplitude are extracted from each of the q second arc current spectrum data to obtain q main harmonic amplitudes and q second harmonic amplitudes; Determine the proportion of each of the q harmonic amplitudes in each of the q second arc current spectrum data to obtain the q harmonic proportions; The q frequency domain features are determined based on the amplitudes of the q primary harmonics and the proportions of the q secondary harmonics.

6. The method as described in claim 1 or 2, characterized in that, The preset feature fusion algorithm fuses the p time-domain statistical features and the q frequency-domain features to obtain the target arc current features, including: Acquire historical arc current data; Time-domain statistical features are extracted from the historical arc current data to obtain multiple historical time-domain statistical features; The historical arc current data is subjected to feature frequency domain extraction to obtain multiple historical frequency domain features; A historical arc current dataset is constructed based on the multiple historical time-domain statistical features, the multiple historical frequency-domain features, and the historical arc current data. The historical arc current dataset is input into a preset dimensionality reduction model to obtain a dimensionality reduction projection matrix; The multiple historical time-domain statistical features and the multiple historical frequency-domain features are respectively input into a preset lightweight machine learning classification model to obtain a first classification accuracy and a second classification accuracy; the first classification accuracy is the classification accuracy output by the lightweight machine learning classification model based on the multiple historical time-domain statistical features; the second classification accuracy is the classification accuracy output by the lightweight machine learning classification model based on the multiple historical frequency-domain features. The time-domain feature weights and frequency-domain feature weights are determined based on the first classification accuracy and the second classification accuracy. The first arc current characteristic is determined based on the p time-domain statistical features, the q frequency-domain features, the time-domain feature weights, and the frequency-domain feature weights. The target arc current feature is obtained by projecting the first arc current feature onto the dimensionality-reduced projection matrix.

7. A lightweight arc fault detection device, characterized in that, Applied to electronic devices, the device includes: The acquisition module is used to acquire the arc current signal of the load in the target distribution network within a preset time period; A determining module is used to determine the signal-to-noise ratio (SNR) parameter of the arc current signal; determine the dynamic window length based on the SNR parameter; and calculate the arc current signal using a preset Fourier transform method to obtain arc current spectrum data; wherein, determining the dynamic window length based on the SNR parameter includes: Obtain the preset low signal-to-noise ratio threshold, high signal-to-noise ratio threshold, minimum window length, and maximum window length; The arc current signal corresponding to the minimum window length is extracted from the arc current signal to obtain the first arc current signal set; The arc current signal corresponding to the maximum window length is extracted from the arc current signal to obtain the second arc current signal set; The first signal power and the first noise power are obtained by calculating all the first arc current signals in the first arc current signal set based on the preset root mean square estimation method. The second signal power and the second noise power are calculated based on the root mean square estimation method for all the second arc current signals in the second arc current signal set. The first signal-to-noise ratio parameter is determined based on the first signal power and the first noise power; The second signal-to-noise ratio parameter is determined based on the second signal power and the second noise power; A reference signal-to-noise ratio parameter is determined based on the first signal-to-noise ratio parameter and the second signal-to-noise ratio parameter; If the signal-to-noise ratio parameter is less than or equal to the low signal-to-noise ratio threshold, then the dynamic window length is determined based on the maximum window length, the reference signal-to-noise ratio parameter, and the signal-to-noise ratio parameter. If the signal-to-noise ratio parameter is greater than or equal to the high signal-to-noise ratio threshold, then the dynamic window length is determined based on the minimum window length, the high signal-to-noise ratio threshold, and the signal-to-noise ratio parameter. If the signal-to-noise ratio parameter is greater than the low signal-to-noise ratio threshold and less than the high signal-to-noise ratio threshold, then the dynamic window length is determined based on the preset interpolation formula, the low signal-to-noise ratio threshold, the high signal-to-noise ratio threshold, the minimum window length, the maximum window length, and the signal-to-noise ratio parameter. The calculation module is used to perform a Hilbert transform on the arc current signal according to the dynamic window length to obtain m amplitude envelope data; m is a positive integer greater than 1; extract statistical features of the arc current signal in the time domain from the m amplitude envelope data to obtain p time-domain statistical features; p is greater than or equal to m; process the arc current spectrum data to obtain q frequency-domain features; q is a positive integer greater than 1; and fuse the p time-domain statistical features and the q frequency-domain features based on a preset feature fusion algorithm to obtain the target arc current features. The control module is used to input the target arc current characteristics into a preset lightweight arc fault detection model to obtain the arc fault detection result.

8. An electronic device, characterized in that, include: Processor, memory, communication interface, and one or more programs; The one or more programs are stored in the memory and configured to be executed by the processor, the programs including instructions for performing the steps of the method as described in any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, the computer program including program instructions that, when executed by a processor, cause the processor to perform the method as described in any one of claims 1-6.

Images