Arc fault detection method and system based on dynamic fuzzy threshold, and storage medium

Abstract

The invention relates to an arc fault detection method and system based on a dynamic fuzzy threshold and a storage medium, and the method comprises the steps: collecting the current circuit data of to-be-detected electrical equipment, and carrying out the feature extraction of the current circuit data, and obtaining a current feature parameter; the method comprises the following steps: training historical circuit data by taking a decision tree algorithm as a model architecture, and in the training process, performing parameter updating through a double-sliding window mechanism and optimizing a fuzzy threshold band range through an information gain maximization principle to obtain a dynamic fuzzy threshold decision model; and performing arc fault analysis on the current characteristic parameters through the dynamic fuzzy threshold decision model to obtain a fault detection result, and outputting early warning information when the fault detection result is that an arc fault occurs. According to the method, the dynamic fuzzy threshold value band is constructed through the Gaussian mixture model, and the self-adaptive adjustment of the threshold value is realized in combination with a double-sliding-window online updating mechanism. According to the technical scheme, the accuracy of arc faults can be remarkably improved, and particularly the false alarm rate is reduced.

Description

Technical Field

[0001] The present application relates to the field of electrical safety detection technology, and in particular to an arc fault detection method, system, and storage medium based on a dynamic fuzzy threshold. Background Art

[0002] With the increasing electrification of modern society, the use of electrical equipment in production and daily life has become increasingly widespread, and the potential for electrical fires has become increasingly prominent. Arc faults, a key factor in electrical fires, pose a serious threat to people's lives and property. Arc fault detection is a core technology in the field of electrical safety. Existing methods mainly fall into two categories: fixed-threshold decision tree methods, such as those that use high-frequency component time-frequency features to set fixed thresholds, cannot adapt to load fluctuations and have a high false alarm rate; deep learning models, such as atrous convolutional neural networks (CNNs), improve detection accuracy, but the models are complex and difficult to deploy on embedded devices.

[0003] In summary, existing arc fault detection technology has the following shortcomings: the threshold setting is rigid, which will lead to an increased misjudgment rate in nonlinear load scenarios; the lack of a dynamic adjustment mechanism makes it difficult to adapt to real-time working condition changes; the detection model is complex and computationally intensive, making it unsuitable for deployment in embedded devices. Summary of the Invention

[0004] The embodiments of the present application provide an arc fault detection method, system, and storage medium based on a dynamic fuzzy threshold, so as to at least solve the problem that arc fault detection methods in related technologies cannot simultaneously take into account dynamic detection and a simple detection model structure.

[0005] In a first aspect, an embodiment of the present application provides an arc fault detection method based on a dynamic fuzzy threshold, characterized by comprising:

[0006] Collecting current circuit data of the electrical device to be tested, and performing feature extraction on the current circuit data to obtain current feature parameters;

[0007] The decision tree algorithm is used as the model architecture to train historical circuit data. During the training process, the parameters are updated through a dual sliding window mechanism and the fuzzy threshold band range is optimized based on the principle of maximizing information gain, thus obtaining a dynamic fuzzy threshold decision model.

[0008] The current characteristic parameters are analyzed for arc faults using the dynamic fuzzy threshold decision model, and when the result of the arc fault is that an arc fault occurs, warning information is output.

[0009] In one embodiment, extracting features from the current circuit data to obtain current feature parameters includes:

[0010] Decomposing the current signal in the current circuit data by using mother wavelet coif1 to obtain a preprocessed signal;

[0011] Extracting features from the preprocessed signal based on single-cycle and multi-cycle dimensions to obtain single-cycle feature parameters and multi-cycle feature parameters respectively;

[0012] The single-cycle characteristic parameters and the multi-cycle characteristic parameters are subjected to mutation characteristic addition and characteristic reconstruction processing to obtain the current characteristic parameters.

[0013] In one embodiment, the performing mutation feature addition and feature reconstruction processing on the single-cycle feature parameters and the multi-cycle feature parameters includes:

[0014] Using a Teager energy operator to enhance the mutation characteristics in the zero-sequence current coupling signal of the single-cycle characteristic parameter and the multi-cycle characteristic parameter to obtain a first characteristic parameter with enhanced characteristics;

[0015] The periodic symmetry characteristic of the AC power supply voltage is utilized to calculate the ratio of the first characteristic parameter after feature enhancement in two preset time windows before and after, and the first characteristic parameter is reconstructed according to the ratio to obtain the current characteristic parameter.

[0016] In one embodiment, the decision tree algorithm is used as a model architecture to train historical circuit data. During the training process, parameters are updated through a dual sliding window mechanism, and the fuzzy threshold band range is optimized according to the principle of maximizing information gain to obtain a dynamic fuzzy threshold decision model, including:

[0017] Constructing a machine learning model based on a decision tree algorithm and constructing a probability density function of characteristic parameters, initializing and calculating the parameters of the probability density function based on the historical circuit data to obtain initial distribution parameters;

[0018] Online updating of the parameters of the probability density function through a double sliding window mechanism to obtain updated distribution parameters;

[0019] Calculate the information gain ratio of different historical feature parameters and optimize the fuzzy threshold band range according to the principle of maximizing information gain;

[0020] The splitting threshold is determined according to the initial distribution parameters and the updated distribution parameters, and the dynamic fuzzy threshold decision model is constructed by using the optimized fuzzy threshold band range as the boundary condition for splitting; wherein, the dynamic fuzzy threshold decision model realizes threshold adaptive adjustment through the probability density function and the dual sliding window mechanism.

[0021] In one embodiment, the initializing calculation of the parameters of the probability density function based on the historical circuit data to obtain the initial distribution parameters includes:

[0022] Extracting features from the historical circuit data to obtain historical feature parameters; wherein the historical circuit data includes data of the electrical equipment in a normal state and an arc fault state;

[0023] The historical characteristic parameters obey multimodal distribution, and corresponding Gaussian distribution functions are constructed for characteristic parameters of different modes;

[0024] The Gaussian distribution function is solved and calculated using a maximum likelihood estimation or expectation maximization algorithm to obtain the initial distribution parameters.

[0025] In one embodiment, the online updating of the parameters of the probability density function by a dual sliding window mechanism to obtain updated distribution parameters includes:

[0026] Setting a real-time detection window and a calibration window, wherein the real-time detection window is used to collect the latest current characteristic parameters in real time, and the calibration window is used to store the historical characteristic parameters and update the probability distribution parameters of the characteristic parameters; wherein the length of the real-time detection window is smaller than the length of the calibration window;

[0027] Obtain a preset smoothing factor range, and incrementally update the mean and equation of the current feature parameter using the exponentially weighted moving average method and the preset smoothing factor range to obtain the updated distribution parameter; wherein, each time a new sample is collected, the real-time detection window is controlled to slide forward one position, and the calibration window is controlled to be synchronously updated and outdated data is eliminated.

[0028] In one embodiment, calculating the information gain ratio of different historical feature parameters and optimizing the fuzzy threshold band range according to the information gain maximization principle includes:

[0029] A fuzzy threshold band is defined based on a bimodal distribution of a Gaussian mixture model, wherein the fuzzy threshold band represents an overlapping interval of normal state and fault state distributions;

[0030] Traverse the discrete points within the candidate threshold band, calculate the information gain under different candidate thresholds, and select the threshold that maximizes the information gain as the splitting threshold of the current node.

[0031] In one embodiment, performing arc fault analysis on the current characteristic parameters using the dynamic fuzzy threshold decision model includes:

[0032] Collecting current feature parameters in real time through a sliding window, and updating the distribution parameters of the probability density function according to the real-time probability distribution of the current feature parameters;

[0033] Each layer of nodes in the dynamic fuzzy threshold decision model calculates the information gain of candidate thresholds according to the real-time probability distribution, and selects the current feature parameter and threshold with the largest gain for splitting;

[0034] Every time a preset number of current feature parameters are detected, the model is triggered to calculate the information gain, re-optimize the threshold band range, and prune or re-split the local nodes of the decision tree according to the new threshold band range.

[0035] In a second aspect, an embodiment of the present application provides an arc fault detection system based on a dynamic fuzzy threshold, characterized in that the system is used to perform any of the above arc fault detection methods based on a dynamic fuzzy threshold, and includes:

[0036] A data analysis module is used to collect current circuit data of the electrical device under test and perform feature extraction on the current circuit data to obtain current feature parameters;

[0037] The model building module is used to train historical circuit data using the decision tree algorithm as the model architecture. During the training process, the dual sliding window mechanism is used to update parameters and the fuzzy threshold band range is optimized based on the principle of maximizing information gain to obtain a dynamic fuzzy threshold decision model.

[0038] The fault warning module is used to perform arc fault analysis on the current characteristic parameters through the dynamic fuzzy threshold decision model to obtain a fault detection result, and output warning information when the fault detection result is an arc fault.

[0039] In a third aspect, an embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the arc fault detection method based on dynamic fuzzy threshold as described in the first aspect above.

[0040] The arc fault detection method, system, and storage medium based on dynamic fuzzy threshold provided by the embodiments of the present application have at least the following technical effects:

[0041] By combining multimodal feature fusion with a dynamic fuzzy threshold decision tree, this approach achieves efficient and accurate detection of arc faults, demonstrating significant benefits. Specifically, the dynamic fuzzy threshold decision tree model dynamically adjusts the node splitting threshold based on the probability distribution of feature parameters, avoiding the misjudgment problem caused by fixed thresholds. This model can adaptively adjust the threshold, effectively reducing the false alarm rate, particularly in complex load scenarios. Furthermore, compared to deep learning models, this technical solution, built on a decision tree algorithm, has a relatively simple model structure and requires less computation, making it more suitable for deployment in embedded devices and reducing computing resource consumption.

[0042] The details of one or more embodiments of the present application are set forth in the following drawings and description to make other features, objects, and advantages of the present application more readily apparent. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0044] Figure 1 is a flow chart of an arc fault detection method based on a dynamic fuzzy threshold in an embodiment of the present application;

[0045] Figure 2 This is a flowchart of a process for constructing a dynamic fuzzy threshold decision model in one embodiment of the present application;

[0046] Figure 3 This is a flowchart of the online analysis process of the dynamic fuzzy threshold decision model in one embodiment of the present application. DETAILED DESCRIPTION

[0047] In order to make the purpose, technical solutions and advantages of this application more clearly understood, the present application is described and illustrated below in conjunction with the accompanying drawings and examples. It should be understood that the specific embodiments described herein are merely used to explain this application and are not intended to limit this application. Based on the embodiments provided in this application, all other embodiments obtained by those of ordinary skill in the art without making any creative efforts are within the scope of protection of this application.

[0048] Obviously, the drawings described below are merely examples or embodiments of the present application. Those skilled in the art can, without inventive effort, apply the present application to other similar scenarios based on these drawings. Furthermore, it is also understood that, although the effort involved in such a development process may be complex and lengthy, for those skilled in the art related to the content disclosed in this application, changes in design, manufacturing, or production based on the technical content disclosed in this application are merely conventional technical means and should not be construed as an insufficiency of the content disclosed in this application.

[0049] References to "embodiments" in this application mean that a particular feature, structure, or characteristic described in connection with the embodiment may be included in at least one embodiment of the application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor does it refer to independent or alternative embodiments that are mutually exclusive of other embodiments. It is understood, both explicitly and implicitly, by those skilled in the art that the embodiments described in this application may be combined with other embodiments unless there is a conflict.

[0050] Unless otherwise defined, the technical or scientific terms used in this application should have the ordinary meaning understood by a person of ordinary skill in the technical field to which this application belongs. The words "one", "a", "the" and the like used in this application do not indicate a limit on quantity and may indicate the singular or plural. The terms "include", "comprise", "have" and any variations thereof used in this application are intended to cover non-exclusive inclusions; for example, a process, method, system, product or device that includes a series of steps or modules (units) is not limited to the listed steps or units, but may also include steps or units that are not listed, or may also include other steps or units that are inherent to these processes, methods, products or devices. The words "connect", "connected", "coupled" and the like used in this application are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. The word "multiple" used in this application refers to two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships can exist. For example, "A and / or B" can mean: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the objects before and after are in an "or" relationship. The terms "first", "second", "third", etc. involved in this application are only used to distinguish similar objects and do not represent a specific order for the objects.

[0051] In the first aspect, the embodiment of the present application provides a first aspect, the embodiment of the present application provides an arc fault detection method based on dynamic fuzzy threshold, referring to Figure 1 , including the following steps.

[0052] Step S1 : collecting circuit data of the electrical device to be tested, and performing feature extraction on the circuit data to obtain feature parameters.

[0053] During data acquisition, in offline detection mode, a high-precision oscilloscope can be used for data acquisition. For example, setting the oscilloscope's sampling rate to 10 MHz and the time window to 20 ms ensures synchronous and accurate acquisition of current, arc voltage, and zero-sequence current coupling signals. During the acquisition process, a high-speed data transmission interface is used to transmit the collected data to a computer in real time and store the data in a categorized manner for subsequent in-depth analysis. In online detection mode, a high-performance microprocessor can be used for data acquisition. This microprocessor integrates a high-speed A / D conversion module, enabling rapid and accurate conversion of analog signals into digital signals.

[0054] In the process of feature extraction of the circuit data, the embodiment of the present application uses the mother wavelet coif1 to decompose the current signal in the circuit data to obtain a preprocessed signal; extracts feature parameters from the preprocessed signal based on the two dimensions of single cycle and multi-cycle to obtain single cycle feature parameters and multi-cycle feature parameters respectively; performs mutation feature addition and feature reconstruction processing on the single cycle feature parameters and multi-cycle feature parameters to obtain the feature parameters, so as to reduce the ambiguity of fault features caused by circuit noise interference.

[0055] More specifically, in the process of adding mutation features and reconstructing features of the single-cycle characteristic parameters and the multi-cycle characteristic parameters, the Teager energy operator is first used to enhance the mutation features in the zero-sequence current coupling signal of the single-cycle characteristic parameters and the multi-cycle characteristic parameters to obtain the first characteristic parameter after feature enhancement; then, the periodic symmetry characteristics of the AC power supply voltage are used to calculate the ratio of the fault arc characteristic parameters after feature enhancement in two preset time windows (such as 10ms) before and after, and the first characteristic parameter is reconstructed according to the ratio to obtain the current characteristic parameter. In this application, the single-cycle characteristic parameter reflects the arc mutation characteristics within a single cycle, and the multi-cycle characteristic parameter reflects the load fluctuation situation, which is longer in time, such as long-term change characteristics in units of several cycles. After the mutation features of the two are enhanced and reconstructed by the Teager energy operator, they are input into the dynamic fuzzy threshold decision tree as multimodal features for node splitting and fault classification.

[0056] It should be noted that, given that the waveform characteristics of the current signal are easily affected by the interference of the circuit load and circuit topology, the present invention uses the mother wavelet coif1 to perform a layer decomposition of the current signal. During the decomposition process, by optimizing the wavelet decomposition parameters (such as reasonably selecting the number of decomposition layers, accurately setting the threshold, etc.), it is ensured that the 50Hz background signal and low-frequency noise signal can be effectively eliminated, and the characteristic information related to the fault arc is retained to the greatest extent. The first layer detail coefficient is used as a preprocessing signal for subsequent fault arc feature extraction, and feature parameter extraction is performed from two dimensions: single cycle and multi-cycle.

[0057] It should be further explained that the single-cycle characteristic parameters include D1 energy, D1 Shannon entropy, D1 range, pulse factor, interval time, pulse peak standard deviation, zero-sequence current standard deviation, and zero-sequence current Shannon entropy.

[0058] Taking D1 energy as an example, the first layer detail coefficient d of the preprocessed signal j (j=1, 2, ..., N / 2) and traverse them one by one, according to the formula Accumulate and calculate to obtain the D1 energy value in the cycle, where N is the number of sampling points in one current cycle.

[0059] Taking the pulse factor as an example, first find the zero-sequence current coupling signal p within the unit current cycle. j The maximum value max(|p j |, and then calculate its absolute value with the average The ratio of This accurately characterizes the pulse characteristics of the current signal within one cycle.

[0060] In addition, D1 Shannon entropy D1 range C r =max(d j )-min(d j ), interval time (f s is the sampling frequency), the pulse peak standard deviation (x1 represents the sequence p j (j=1, 2, ..., N), x2 represents the maximum value of the sequence p j (j=1, 2, ..., N), and so on, m is the number of pulse peaks involved in the calculation), zero-sequence current standard deviation Zero-sequence current Shannon entropy The above single-cycle characteristic parameters are used to accurately characterize the pulse characteristics of the current signal within one cycle.

[0061] It should be further explained that the multi-cycle characteristic parameters include D1 energy standard deviation, D1 Shannon entropy standard deviation, D1 range standard deviation, pulse factor standard deviation, and zero-sequence current Shannon entropy standard deviation. Taking D1 energy standard deviation as an example, we first calculate the D1 energy in multiple adjacent current cycles. q is the number of calculation cycles, and then according to the formula Calculate the standard deviation, where is the average value of D1 energy over multiple cycles. Other multi-cycle characteristic parameters are calculated similarly. These multi-cycle characteristic parameters can effectively reflect the fluctuation of D1 energy over multiple cycles, providing more comprehensive and accurate characteristic information for arc fault detection.

[0062] In this embodiment, to solve the problem that the boundary between normal state and fault state becomes unclear after some characteristic parameters are interfered by circuit noise, the present invention introduces the Teager energy operator to process the zero-sequence current coupling signal. j ,…,x N}, according to the formula Calculate the energy value of each sampling point. Perform point-by-point calculations on the zero-sequence current coupling signal, thereby effectively enhancing the mutation characteristics in the signal.

[0063] It should be further explained that the low voltage AC power supply voltage us The fault arc characteristics are reconstructed by taking into account the periodic symmetry of the first 10ms and the last 10ms of the electrical signal (t) (i.e., the electrical signals are symmetrical in the first 10ms and the last 10ms), and calculating the ratio of the fault arc characteristic parameters within the first and last 10ms time window. Under normal conditions, the fault arc characteristic parameters extracted from the first and last two electrical signals are of the same order of magnitude, and their ratio fluctuates within a small range around 1. However, when a series arc fault occurs, the unstable characteristics of the fault arc cause the fault arc characteristic parameters extracted from the first and last two electrical signals to differ significantly, and their ratio fluctuates significantly around 1. Based on this characteristic, the characteristic ratio is used to reconstruct the fault arc characteristic parameters (single-cycle characteristic parameters—pulse factor ratio, zero-sequence current Shannon entropy ratio, D1 energy ratio, D1 Shannon entropy ratio, D1 range ratio, etc., and the corresponding multi-cycle characteristic parameters—D1 energy ratio standard deviation, D1 Shannon entropy ratio standard deviation, D1 range ratio standard deviation, pulse factor ratio standard deviation, and zero-sequence current Shannon entropy ratio standard deviation). Through this feature ratio reconstruction, the distinguishability of features under different working conditions is significantly improved, laying a solid foundation for subsequent machine learning models to accurately detect fault arcs.

[0064] Step S2, obtain historical circuit data, train the historical circuit data using the decision tree algorithm as the model architecture, and obtain a dynamic fuzzy threshold decision model. In the training phase of this application, it is necessary to collect historical normal state data (no arc) and fault state data (including arc) to initialize the probability density function; in the real-time detection phase, real-time electrical data (including data with uncertain normal or fault status) is collected, and online judgment is performed through dynamic feature extraction and threshold decision tree. Therefore, two types of data are required for training and are marked separately, and real-time unknown state data is collected during detection.

[0065] Specific reference Figure 2The historical circuit data is obtained, and the historical circuit data is trained using a decision tree algorithm as a model architecture to obtain a dynamic fuzzy threshold decision model, including: S21, multimodal features are input into the basic model; S22, threshold band segmentation, which is achieved through a Gaussian distribution curve in this embodiment; S23, probability calculation and probability accumulation; S24, output of classification results. In more detail, the embodiment of the present application constructs a machine learning model based on a decision tree algorithm, and constructs a probability density function of characteristic parameters, and initializes and calculates the parameters of the probability density function based on the historical circuit data to obtain initial distribution parameters; then, the parameters of the probability density function are updated online through a double sliding window mechanism to obtain updated distribution parameters; the information gain ratio of different features is calculated, and the fuzzy threshold band range is optimized according to the principle of maximizing information gain; the splitting threshold is determined according to the initial distribution parameters and the updated distribution parameters, and the dynamic fuzzy threshold decision model is constructed using the optimized fuzzy threshold band range as the boundary condition for splitting; wherein, the dynamic fuzzy threshold decision model realizes threshold adaptive adjustment through the probability density function and the double sliding window mechanism; the circuit data and the reconstructed fault arc characteristic parameters are input into the machine learning model, the network weight of the model is determined, and a trained machine learning model is obtained.

[0066] Specifically, in the process of calculating the initial distribution parameters, the historical circuit data is first feature extracted to obtain historical feature parameters; wherein, the historical circuit data includes data of electrical equipment in normal state and arc fault state; the historical feature parameters obey multimodal distribution, and corresponding Gaussian distribution functions are constructed for the feature parameters of different modes; the Gaussian distribution function is solved and calculated using the maximum likelihood estimation or expectation maximization algorithm to obtain the initial distribution parameters.

[0067] In the process of obtaining updated distribution parameters, a real-time detection window and a calibration window are set, the real-time detection window is used to collect the latest current feature parameters in real time, and the calibration window is used to store the historical feature parameters and update the probability distribution parameters of the feature parameters; wherein, the length of the real-time detection window is smaller than the length of the calibration window; a preset smoothing factor range is obtained, and the mean and equation of the current feature parameters are incrementally updated by the exponentially weighted moving average method and the preset smoothing factor range to obtain the updated distribution parameters; wherein, each time a new sample is collected, the real-time detection window is controlled to slide forward one position, and the calibration window is controlled to be synchronously updated and outdated data is eliminated.

[0068] In the process of optimizing the fuzzy threshold band range, it is necessary to define the fuzzy threshold band based on the bimodal distribution of the Gaussian mixture model. The fuzzy threshold band represents the overlapping interval of the normal state and fault state distribution; traverse the discrete points in the candidate threshold band, calculate the information gain under different candidate thresholds, and select the threshold that maximizes the information gain as the splitting threshold of the current node.

[0069] In the process of constructing a machine learning model based on a decision tree algorithm, the machine learning model of this embodiment uses information gain ratio as a metric for feature selection. By calculating the information gain ratios of different features, the feature with the largest information gain ratio is selected for decision tree node splitting; the circuit data and the reconstructed fault arc characteristic parameters are input into the machine learning model, the network weight of the model is determined, and a trained machine learning model is obtained.

[0070] It should be noted that during the training and application phases, the input parameters of the dynamic fuzzy threshold decision model include single-cycle features and multi-cycle features that have been enhanced and reconstructed into multimodal features. The output parameters are probabilistic. That is, through the dynamic fuzzy threshold band and GMM model, the decision tree outputs fault probabilities or classification results. This supports probabilistic graded warnings, such as determining whether an arc fault has occurred, or outputting probabilistic classification results such as "high probability fault," "low probability fault," and "normal."

[0071] Step S3: performing arc fault analysis on the characteristic parameters through the dynamic fuzzy threshold decision model to obtain a fault detection result, and outputting a warning message when the fault detection result is "arc fault occurs".

[0072] Specific reference Figure 3 The arc fault analysis of the characteristic parameters using the dynamic fuzzy threshold decision model to obtain a fault detection result includes: S31, sliding window data; S32, parameter update, namely, mean μ and variance σ; S33, threshold band recalibration. More specifically, the current characteristic data is collected in real time through the sliding window, and the distribution parameters of the probability density function are updated based on the real-time probability distribution of the current characteristic parameters; each layer of nodes in the dynamic fuzzy threshold decision model calculates the information gain of the candidate thresholds based on the real-time probability distribution, and selects the feature and threshold with the largest gain for splitting; each time a preset number of new samples are detected, the model is triggered to calculate the information gain, re-optimize the threshold band range, and prune or re-split the local nodes of the decision tree based on the new threshold band range.

[0073] If the machine learning model detects an arc fault in the circuit, it outputs a warning message and triggers the protective switch, which then disconnects the circuit. It should be noted that the detection results of the detection device serve as the trigger signal for the arc fault protective switch to operate. When an arc fault is detected, the protective switch should disconnect the circuit within a set time. Based on the arc fault characteristics extracted from the detection method and the set detection algorithm, the protective switch's operation logic is optimized, improving its accuracy and reliability. This allows the protective switch to operate promptly and accurately based on the detection results, effectively disconnecting the faulty circuit and ensuring the safety of the electrical system.

[0074] In summary, the arc fault detection method based on dynamic fuzzy thresholds provided in the embodiments of the present application achieves efficient and accurate detection of arc faults by combining multimodal feature fusion with a dynamic fuzzy threshold decision tree, and has significant beneficial effects, specifically manifested in the following aspects:

[0075] (1) Improve detection accuracy: The mother wavelet coif1 is used to decompose the current signal of the circuit data, combined with single-cycle and multi-cycle feature extraction, and the Teager energy operator to enhance the mutation feature, which effectively improves the characterization ability of the fault arc characteristic parameters, thereby significantly improving the detection accuracy;

[0076] (2) Reduce false alarm rate: Through the dynamic fuzzy threshold decision tree model, the node splitting threshold is dynamically adjusted according to the probability distribution of the feature parameters, avoiding the misjudgment problem caused by fixed thresholds. Especially in complex load scenarios, the model can adaptively adjust the threshold and effectively reduce the false alarm rate;

[0077] (3) Enhanced robustness: The periodic symmetry of the AC power supply voltage is used for feature reconstruction, which suppresses harmonic interference and improves the stability and reliability of feature parameters. At the same time, the combination of the Gaussian mixture model (GMM) and the sliding window mechanism realizes adaptive adjustment of the threshold, thus enhancing the robustness of the system.

[0078] (4) Reduce computing resource consumption: Compared with deep learning models, this technical solution is based on the decision tree algorithm. The model structure is relatively simple and the computational complexity is small. It is more suitable for deployment in embedded devices and reduces computing resource consumption.

[0079] (5) Achieve precise protection: Through the optimization of the dynamic fuzzy threshold band range, a probabilistic graded early warning is achieved, which provides strong support for the precise protection of electrical equipment and further ensures the safety of people’s lives and property.

[0080] In another specific embodiment, the present invention provides an arc detection method based on wavelet decomposition and dynamic fuzzy threshold, which includes the following steps.

[0081] a) Extracting arc fault characteristic parameters, including:

[0082] a1) Decomposing the current signal of the circuit data using the mother wavelet coif1 to obtain a preprocessed signal; wherein the specific process of signal preprocessing and wavelet decomposition is as follows:

[0083] First, the number of decomposition layers is optimized, and the optimal number of decomposition layers is dynamically determined by the particle swarm algorithm. The formula is:

[0084]

[0085] Among them E fault 、E normal are fault signal capability and normal signal energy respectively;

[0086] Then the noise is suppressed and the high frequency component of D1 is subjected to wavelet packet threshold denoising. The formula for threshold setting is: Where σ is the noise standard deviation and N is the signal length.

[0087] a2) extracting fault arc characteristic parameters from the preprocessed signal based on single-cycle and multi-cycle dimensions to obtain single-cycle characteristic parameters and multi-cycle characteristic parameters of the fault arc; wherein the single-cycle characteristic parameters include D1 energy (formula: ED1 = ∑|D1(n)| 2 ), pulse factor (formula: Among them, I peak is the peak current, I rms is the effective value of the current); the multi-cycle characteristic parameters include the D1 energy standard deviation (formula: Where i is the cycle index in the multi-cycle feature), reflecting the load fluctuation characteristics;

[0088] a3) using a Teager energy operator to enhance the mutation characteristics in the zero-sequence current coupling signal of the circuit data to obtain the fault arc characteristic parameters after the characteristics are enhanced;

[0089] During the feature reconstruction process, the Teager energy operator (formula: TEO(x[n])=x 2 [n]-x[n-1]*x[n+1], where n is the sample index in the single-cycle feature) enhances the mutation feature, combines the voltage cycle symmetry to align the current waveform, and suppresses harmonic interference.

[0090] a4) using the periodic symmetry of the AC power supply voltage, calculating the ratio of the fault arc characteristic parameters after the feature enhancement in the 10 ms time window before and after, so as to obtain the reconstructed fault arc characteristic parameters.

[0091] b) Train the machine learning model based on the fault arc characteristic parameters and build a decision tree model based on dynamic fuzzy thresholds, including:

[0092] b1) Construct a machine learning model based on the decision tree algorithm; b2) Select the feature with the largest information gain ratio to split the decision tree node, and the node splitting threshold is dynamically determined according to the probability distribution of the feature parameters; b3) Initialize the probability density function of the feature parameters based on historical data; b4) Update the probability distribution parameters online through a sliding window mechanism; b5) Optimize the fuzzy threshold band range according to the principle of maximizing information gain; b6) Construct a dynamic fuzzy threshold decision tree and achieve adaptive threshold adjustment through the GMM model and sliding window mechanism.

[0093] Specifically, in the process of building a dynamic fuzzy decision tree, historical data is modeled based on the GMM threshold band, and the threshold range is defined as:

[0094] [μ normal -kσ normal , μ arc +kσ arc ](k=1.5-2.5)

[0095] where μ normal 、μ arc are the means of normal and arc states respectively, k is the adjustment factor;

[0096] For the online update mechanism, this embodiment adopts a dual sliding window (long window calibration parameters, short window real-time detection), update period N = [50, 200], smoothing factor α = [0.2, 0.5];

[0097] For information gain optimization, calculate the information gain of different threshold bands (formula: Select the threshold split node with the largest gain.

[0098] In one embodiment, initializing the probability density function of the characteristic parameters for historical data specifically includes the following steps:

[0099] (1) Data preprocessing

[0100] Input data: historically collected normal state data (no arc) and fault state data (including arc).

[0101] Feature extraction: Single-cycle features: D1 energy, pulse factor; multi-cycle features: D1 energy standard deviation.

[0102] (2) Gaussian mixture model (GMM) initialization

[0103] Modeling assumption: The characteristic parameters obey a multimodal distribution, and each mode corresponds to a Gaussian distribution. The formula is:

[0104]

[0105] Where, C = 2 (normal state c = 1, fault state c = 2), π c is the modal weight, μ c and are the mean and variance respectively.

[0106] (3) Parameter estimation

[0107] Algorithm selection: Maximum Likelihood Estimation (MLE) or Expectation Maximization (EM) algorithm;

[0108] Output parameters: Calculate the initial parameter set using historical data

[0109] In one embodiment, the online updating of probability distribution parameters through a sliding window mechanism specifically includes the following steps:

[0110] (1)Dual sliding window design

[0111] Short window (real-time detection window): Window size: N short =50 (number of samples); Function: real-time collection of current feature data x t (i.e. the reconstructed single / multi-cycle characteristic parameters).

[0112] Long window (calibration window): Window size: N long ; Function: Store historical feature parameters and update probability distribution parameters for calibrating probability distribution parameters (mean, variance, etc.).

[0113] (2) Parameter update algorithm

[0114] Exponentially Weighted Moving Average (EWMA): Introduce a smoothing factor range: α∈[0.2,0.5]; perform incremental updates on the mean and variance:

[0115]

[0116] Among them, μ t and are the mean and variance at time t.

[0117] Sliding window shift: Every time a new sample is collected, the short window slides forward one position, and the long window is updated synchronously and outdated data is eliminated to ensure that the parameters reflect the latest working conditions.

[0118] In one embodiment, the threshold optimization algorithm based on information gain specifically includes the following steps:

[0119] Definition of threshold band: Based on the bimodal distribution of GMM, the fuzzy threshold band is defined as the overlapping interval of normal state and fault state distribution: threshold band = [μ normal -kσ normal , μ arc +kσ arc ], where k∈[1.5, 2.5] is the adjustment factor that controls the threshold band width.

[0120] Information gain calculation: For the candidate threshold T, the feature space is divided into two subsets, {x≤T} and {x>T}, and the information gain is calculated. The candidate threshold T is a discrete value point within the threshold band (for example, a point is taken every 0.1 times the standard deviation). It is used to traverse and calculate the information gain. The T that maximizes the information gain is selected as the splitting threshold for the current node.

[0121] Parent node entropy:

[0122] H parent =-∑C=1 C p(c)log2p(c)

[0123] H parent =-∑p(c)log2p(c)

[0124] Where p(c) is the category prior probability;

[0125] Child node entropy:

[0126] Gain formula: N child is the number of child node samples, N parent is the number of parent node samples.

[0127] Threshold optimization process: Traverse the discrete points within the candidate threshold band (for example, select a point every 0.1 standard deviation), select the threshold that maximizes Gain(t) as the split threshold of the current node, and use the split threshold as the basis for the decision tree to determine whether an arc fault has occurred. By dynamically adjusting the threshold band, the probability of the characteristic value falling into the normal state or fault state distribution interval is finally used to determine whether an arc fault has occurred.

[0128] For example, the characteristic value falls into the fault state distribution interval (μ arc +kσ arc When it falls into the normal state range (μ normal -kσ normal If the system is within the overlapping threshold bands, the fault probability is calculated based on the probability distribution, triggering a graded warning. The dynamic fuzzy threshold decision tree outputs at least one of the following: "Normal," "Suspected Fault (with probability grading, such as low / medium / high probability of failure)," and "Confirmed Fault."

[0129] In one embodiment, a dynamic fuzzy threshold decision tree is constructed, and threshold adaptive adjustment is achieved through a GMM model and a sliding window mechanism, specifically including the following steps:

[0130] (1) Decision tree node splitting mechanism

[0131] Each layer of nodes is based on the real-time probability distribution of the current feature parameters (μ updated by step b4) t and ), calculate the information gain of the candidate thresholds, and select the feature and threshold with the largest gain for splitting.

[0132] The splitting threshold is no longer a fixed value, but a dynamic interval based on GMM [μ normal -kσ normal , μ arc +kσ arc ], allowing for fuzzy decision making within a certain range.

[0133] (2) Online adaptive process

[0134] Initialization phase: Use historical data to build the initial GMM model through step b3 and generate the initial structure of the decision tree;

[0135] Real-time detection stage: feature data is collected in real time through a sliding window (step b4) to update GMM parameters;

[0136] Threshold adjustment trigger conditions: Every time a certain number of new samples (e.g., 50) are detected, the information gain calculation in step b5 is triggered, the threshold band range is re-optimized, and local nodes of the decision tree are pruned or re-split if necessary. For example, if real-time detection reveals a significant decrease in information gain (deterioration in splitting effectiveness), an increase in false alarm rate, or a significant shift in the probability distribution of feature parameters due to load changes, it is considered "necessary" and local nodes of the decision tree need to be pruned (removing invalid branches) or re-split (adjusting the threshold) to maintain model adaptability.

[0137] (3) Hardware deployment plan

[0138] Due to the simple structure and low computational complexity of the decision tree, it can be deployed on embedded devices (such as MCU) and parameter updates can be performed periodically through a timer to meet real-time requirements.

[0139] In a second aspect, an embodiment of the present application provides an arc fault detection system based on a dynamic fuzzy threshold, characterized in that the system is used to perform any of the above arc fault detection methods based on a dynamic fuzzy threshold, and includes:

[0140] A data analysis module is used to collect circuit data of the electrical equipment to be tested and perform feature extraction on the circuit data to obtain feature parameters;

[0141] A model building module is used to obtain historical circuit data, train the historical circuit data using a decision tree algorithm as a model architecture, and obtain a dynamic fuzzy threshold decision model;

[0142] The fault warning module is used to perform arc fault analysis on the characteristic parameters through the dynamic fuzzy threshold decision model to obtain a fault detection result, and output warning information when the fault detection result is "arc fault occurs".

[0143] It should be noted that the arc fault detection system provided in this embodiment is used to implement the aforementioned embodiments, and details already described will not be repeated. As used above, the terms "module," "unit," "subunit," etc. may refer to a combination of software and / or hardware that implements a predetermined function. Although the apparatus described in the above embodiments is preferably implemented in software, implementation in hardware, or a combination of software and hardware, is also possible and contemplated.

[0144] In a third aspect, an embodiment of the present application provides a computer-readable storage medium having a program stored thereon. When the program is executed by a processor, the arc fault detection method based on dynamic fuzzy threshold provided in the first aspect is implemented.

[0145] The readable storage medium may include, but is not limited to, a portable disk, a hard disk, a random access memory, a read-only memory, an erasable programmable read-only memory, an optical storage device, a magnetic storage device, or any suitable combination thereof.

[0146] In a possible embodiment, the present invention can also be implemented in the form of a program product, which includes program code. When the program product is run on a terminal device, the program code is used to enable the terminal device to execute the steps of the arc fault detection method based on dynamic fuzzy threshold provided in the first aspect.

[0147] The program code for executing the present invention may be written in any combination of one or more programming languages, and may be executed entirely on the user device, partially on the user device, as an independent software package, partially on the user device and partially on a remote device, or entirely on the remote device.

[0148] The technical features of the above-mentioned embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above-mentioned embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0149] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art could make various modifications and improvements without departing from the spirit of the present application, all of which fall within the scope of protection of the present application. Therefore, the scope of protection of the present patent application shall be determined by the appended claims.

Claims

1. A method for arc fault detection based on dynamic fuzzy threshold, characterized in that: include: Collecting current circuit data of the electrical device to be tested, and performing feature extraction on the current circuit data to obtain current feature parameters; The decision tree algorithm is used as the model architecture to train historical circuit data. During the training process, the parameters are updated through a dual sliding window mechanism and the fuzzy threshold band range is optimized based on the principle of maximizing information gain, thus obtaining a dynamic fuzzy threshold decision model. An arc fault analysis is performed on the current characteristic parameters using the dynamic fuzzy threshold decision model, and when the result of the arc fault analysis is that an arc fault occurs, an early warning message is output.

2. The arc fault detection method according to claim 1, wherein: The extracting features of the current circuit data to obtain current feature parameters includes: Decomposing the current signal in the current circuit data by using mother wavelet coif1 to obtain a preprocessed signal; Extracting features from the preprocessed signal based on single-cycle and multi-cycle dimensions to obtain single-cycle feature parameters and multi-cycle feature parameters respectively; The single-cycle characteristic parameters and the multi-cycle characteristic parameters are subjected to mutation characteristic addition and characteristic reconstruction processing to obtain the current characteristic parameters.

3. The arc fault detection method according to claim 2, characterized in that: The performing mutation feature addition and feature reconstruction processing on the single-cycle feature parameters and the multi-cycle feature parameters includes: Using a Teager energy operator to enhance the mutation characteristics in the zero-sequence current coupling signal of the single-cycle characteristic parameter and the multi-cycle characteristic parameter to obtain a first characteristic parameter with enhanced characteristics; The periodic symmetry characteristic of the AC power supply voltage is utilized to calculate the ratio of the first characteristic parameter after feature enhancement in two preset time windows before and after, and the first characteristic parameter is reconstructed according to the ratio to obtain the current characteristic parameter.

4. The arc fault detection method according to claim 1, wherein: The decision tree algorithm is used as the model architecture to train historical circuit data. During the training process, the parameters are updated through a dual sliding window mechanism, and the fuzzy threshold band range is optimized according to the principle of maximizing information gain to obtain a dynamic fuzzy threshold decision model, including: Constructing a machine learning model based on a decision tree algorithm and constructing a probability density function of characteristic parameters, initializing and calculating the parameters of the probability density function based on the historical circuit data to obtain initial distribution parameters; Online updating of the parameters of the probability density function through a double sliding window mechanism to obtain updated distribution parameters; Calculate the information gain ratio of different historical feature parameters and optimize the fuzzy threshold band range according to the principle of maximizing information gain; The splitting threshold is determined according to the initial distribution parameters and the updated distribution parameters, and the dynamic fuzzy threshold decision model is constructed by using the optimized fuzzy threshold band range as the boundary condition for splitting; wherein, the dynamic fuzzy threshold decision model realizes threshold adaptive adjustment through the probability density function and the dual sliding window mechanism.

5. The arc fault detection method according to claim 4, characterized in that: The initializing calculation of the parameters of the probability density function based on the historical circuit data to obtain initial distribution parameters includes: Extracting features from the historical circuit data to obtain historical feature parameters; wherein the historical circuit data includes data of the electrical equipment in a normal state and an arc fault state; The historical characteristic parameters obey multimodal distribution, and corresponding Gaussian distribution functions are constructed for characteristic parameters of different modes; The Gaussian distribution function is solved and calculated using a maximum likelihood estimation or expectation maximization algorithm to obtain the initial distribution parameters.

6. The arc fault detection method according to claim 4, characterized in that: The online updating of the parameters of the probability density function through the dual sliding window mechanism to obtain updated distribution parameters includes: Setting a real-time detection window and a calibration window, wherein the real-time detection window is used to collect the latest current characteristic parameters in real time, and the calibration window is used to store the historical characteristic parameters and update the probability distribution parameters of the characteristic parameters; wherein the length of the real-time detection window is smaller than the length of the calibration window; Obtain a preset smoothing factor range, and incrementally update the mean and equation of the current feature parameter using the exponentially weighted moving average method and the preset smoothing factor range to obtain the updated distribution parameter; wherein, each time a new sample is collected, the real-time detection window is controlled to slide forward one position, and the calibration window is controlled to be synchronously updated and outdated data is eliminated.

7. The arc fault detection method according to claim 4, characterized in that: The information gain ratio of different historical characteristic parameters is calculated, and the fuzzy threshold band range is optimized according to the principle of maximizing the information gain, including: A fuzzy threshold band is defined based on a bimodal distribution of a Gaussian mixture model, wherein the fuzzy threshold band represents an overlapping interval of normal state and fault state distributions; Traverse the discrete points within the candidate threshold band, calculate the information gain under different candidate thresholds, and select the threshold that maximizes the information gain as the splitting threshold of the current node.

8. The arc fault detection method according to claim 1, wherein: The performing arc fault analysis on the current characteristic parameters by using the dynamic fuzzy threshold decision model includes: Collecting current feature parameters in real time through a sliding window, and updating the distribution parameters of the probability density function according to the real-time probability distribution of the current feature parameters; Each layer of nodes in the dynamic fuzzy threshold decision model calculates the information gain of candidate thresholds according to the real-time probability distribution, and selects the current feature parameter and threshold with the largest gain for splitting; Every time a preset number of current feature parameters are detected, the model is triggered to calculate the information gain, re-optimize the threshold band range, and prune or re-split the local nodes of the decision tree according to the new threshold band range.

9. An arc fault detection system based on dynamic fuzzy threshold, characterized in that: The system for executing the arc fault detection method based on dynamic fuzzy threshold according to any one of claims 1 to 8 comprises: A data analysis module is used to collect current circuit data of the electrical device under test and perform feature extraction on the current circuit data to obtain current feature parameters; The model building module is used to train historical circuit data using the decision tree algorithm as the model architecture. During the training process, the dual sliding window mechanism is used to update parameters and the fuzzy threshold band range is optimized based on the principle of maximizing information gain to obtain a dynamic fuzzy threshold decision model. The fault warning module is used to perform arc fault analysis on the current characteristic parameters through the dynamic fuzzy threshold decision model, and output warning information when the result of the arc fault analysis is that an arc fault occurs.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the arc fault detection method based on a dynamic fuzzy threshold is implemented according to any one of claims 1 to 8.

Images