Wire terminal fault arc multi-level feature selection and detection method and system

Abstract

The application discloses a kind of connection terminal fault arc multilevel feature selection and detection method and system, the method of the present application includes collecting the original current waveform data sample of connection terminal under a variety of load conditions, the feature of the feature domain of the sample in sample set is extracted to obtain original feature set;The quality of the feature in original feature set is evaluated and the feature that quality does not meet the requirement is eliminated;According to the quality of the feature in each feature domain, the feature domain weight of each feature domain is determined;The global importance evaluation score of each feature in combination feature set is obtained;The feature in combination feature set is screened based on target dimension hard constraint and information loss soft constraint Execution screening decision finally generates optimization feature subset;Based on the feature in optimization feature subset, the fault detection of connection terminal is carried out.The present application aims at realizing the independent evaluation of feature effectiveness, dynamic optimization and fault arc detection under the premise of not depending on specific identification model.

Description

Technical Field

[0001] This invention relates to the field of power, and specifically to a method and system for selecting and detecting multi-level characteristics of fault arcs in terminal blocks. Background Technology

[0002] Electric arc faults are discharge phenomena caused by gas gap breakdown in electrical circuits due to insulation aging, poor contact, mechanical damage, or environmental pollution. In low-voltage power distribution systems, new energy vehicles, photovoltaic energy storage, and data centers, electric arc faults have been proven to be a major cause of electrical fires. Therefore, timely and accurate detection of electric arc faults is a core technical aspect for ensuring personal and property safety and meeting electrical safety standards. Currently, the mainstream technical approach for electric arc fault detection relies on multi-dimensional feature extraction of line current signals, combined with machine learning or deep learning models to achieve pattern recognition. Specifically, existing technologies typically construct a feature system from the following four levels: extracting parameters such as zero-time and slope change rate at the time domain level; analyzing indicators such as harmonic component distribution and odd-even harmonic ratio at the frequency domain level; obtaining coefficient matrix features using wavelet transform and S-transform methods at the time-frequency domain level; and calculating complexity measures such as sample entropy at the information entropy level. Finally, the above multi-source heterogeneous features are concatenated into a high-dimensional feature vector, which is then input into classifiers such as random forests and long short-term memory networks for arc state discrimination. However, the aforementioned technical solutions face certain technical bottlenecks in engineering applications. The randomness and nonlinearity of fault arcs are complex, and single-dimensional features are often insufficient for stable identification. Existing technologies generally employ multi-dimensional feature combinations to cover various operating conditions, resulting in high-dimensional feature sets and significant computational resource consumption for high-dimensional features. High-dimensional features not only increase floating-point computation and storage overhead but may also introduce noise interference due to redundant features, increasing the risk of model overfitting and affecting generalization performance across load types and current levels. In existing fault arc detection technology architectures, a pre-emptive, model-decoupled feature quantification evaluation mechanism has not yet been established between the feature extraction stage and the identification model. Current mainstream solutions primarily rely on post-processing methods such as weight analysis and recursive elimination after model training. This process has the following characteristics: feature selection is performed after model training, leading to invalid or weakly correlated features already entering the early computation stages, resulting in irreversible waste of computational resources; the selection results are highly coupled with specific models, requiring repeated training and analysis after model replacement; and the evaluation process rarely considers the fundamental attributes of the features themselves, such as physical interpretability and operating condition stability, for multi-dimensional value judgment. This mechanism allows existing technologies to primarily optimize features in the later stages of algorithm design, leaving room for optimization in the active reconstruction of the feature space.

[0003] In summary, the main challenges faced by existing technologies or solutions in the field of terminal block fault arc detection technology at the feature processing level include: 1. Lack of a pre-emptive feature selection mechanism: Existing methods rarely embed an independent evaluation step between feature extraction and model training; feature selection is usually performed after model training is completed. In this processing mode, invalid or weakly relevant features have already entered the early calculation process, resulting in ineffective consumption of computing resources and unnecessary increase in model complexity. 2. Limited dimensionality and methodological innovation in importance assessment: Existing technologies rely heavily on weight analysis or contribution ranking after model training to assess feature importance, resulting in a high degree of coupling between the assessment process and the final identification model. This assessment method rarely makes a comprehensive value judgment from multiple dimensions such as the physical meaning, discriminability, and stability of the features themselves, leaving room for methodological innovation in the "selection of the best among the best" and "fusion and refinement" of features. 3. Insufficient synergy between feature processing flow and model optimization: In existing technologies, the three stages of feature extraction, selection, and identification are mostly linearly connected, with little closed-loop feedback and adaptation optimization between the selection process and the final model performance. In this structure, while the selected feature subset can reduce dimensionality, its optimal matching with the subsequent recognition model is difficult to guarantee, potentially leading to information loss. Strengthening the synergy between feature space reconstruction and model performance improvement is needed. Therefore, the core bottleneck of existing fault arc detection technologies based on high-dimensional feature sets lies in the lack of a lightweight, pre-processed, multi-dimensional feature value quantification, optimization, and detection mechanism in the feature processing flow. This, to some extent, restricts the deployment feasibility and performance optimization potential of the algorithm in resource-constrained environments. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a multi-level feature selection and detection method and system for terminal block fault arcs, which addresses the above-mentioned problems in the prior art. The present invention aims to build a multi-level evaluation and screening mechanism that is decoupled from the feature extraction and recognition model, embedded with physical constraints, and has cross-domain collaborative capabilities, so as to achieve independent evaluation, dynamic optimization and fault arc detection of feature effectiveness without relying on a specific recognition model.

[0005] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: A method for selecting and detecting multi-level characteristics of fault arcs in terminal blocks, comprising the following steps: S101, collect samples of the original current waveform data of the terminals under various load conditions, construct a sample set consisting of the terminal fault arc sample set and the normal operating condition sample set, extract features of multiple feature domains from the samples in the sample set, and obtain the original feature set consisting of feature subsets of multiple feature domains. S102, perform quality assessment on the features in the original feature set, remove features that do not meet the quality requirements, and obtain the optimized feature subsets for each feature domain; S103, determine the feature domain weights of each feature domain based on the quality of the features in each feature domain; S104. Based on the feature domain weights of each feature domain, perform cross-domain global importance evaluation on the combined feature set of the optimized feature subsets of each feature domain to obtain the global importance evaluation score of each feature in the combined feature set. S105, based on the global importance evaluation scores of each feature in the combined feature set, a screening decision is performed on the features in the combined feature set according to the hard constraint of target dimension and the soft constraint of information loss shown in the following formula, finally generating an optimized feature subset: ; ; in, To optimize feature subsets, For combined feature sets, Represents the set of candidate features; Indicates the number of features in the feature candidate set; Features within the combined feature set The global importance assessment score, The desired number of features to be selected. For information loss rate, This is the information loss rate threshold; S106, fault detection of terminal blocks is performed based on features in the optimized feature subset.

[0006] Optionally, the functional expression of the original feature set obtained in step S101 is: ; in, For the original feature set, ~ These are feature subsets of the 1st to kth feature domains, and feature subsets of any kth feature domain. Include Dimensional features, the original feature set has the following dimensions: ; Where N is the dimension of the original feature set, This represents the number of feature domains.

[0007] Optionally, when performing quality evaluation on the features in the original feature set in step S102, it includes performing quality evaluation on the features in the original feature set at the feature domain level: ; in, Features quality , and These are the weight values ​​for the quality assessment indicators; , and The quality assessment indicators are as follows: inter-class separability, operating condition robustness, and physical conformity. Specifically: the inter-class separability indicator quantifies the difference in feature distribution across terminal fault arc sample sets and normal operating condition sample sets; the operating condition robustness indicator quantifies the stability of features across different operating conditions; and the physical conformity indicator quantifies the degree of conformity between features and the physical process of arc discharge. For each feature domain, screening rules eliminate features that do not meet the quality requirements. ; in, The k-th optimized feature subset is obtained after removing features that do not meet the quality requirements by selecting rules. The quality threshold for the k-th feature domain. is the screening coefficient for the k-th feature domain, and max is the maximum value.

[0008] Optionally, the plurality of feature domains includes the time domain and the frequency domain; when performing quality evaluation on the features in the original feature set at the feature domain level, the evaluation is performed on the features in the time domain. The function expression for the quality assessment function used to evaluate the quality is: ; in, Features of the time domain quality , and As weight, , and Features of the time domain The indices include inter-class separability, operational condition robustness, and physical compliance, and include: ; ; ; in, and Features The sample mean on the terminal fault arc sample set and the normal operating condition sample set. and Features The sample standard deviation on the terminal fault arc sample set and the normal operating condition sample set. It is a constant used to prevent the denominator from being divided by zero; The load variation coefficient is a characteristic. The ratio of the standard deviation to the mean under different load types; The physical conformity of the time-domain feature represents the feature. The degree of conformity with the physical process of electric arc discharge; when evaluating the quality of features in the original feature set at the feature domain granularity, the focus is on the features in the frequency domain. The function expression for the quality assessment function used to evaluate the quality is: ; in, Features quality , and As weight, , and Features The indices are: inter-class separability, operating condition robustness, and physical compliance. The operating condition robustness index is a constant, and the following applies: ; ; in, frequency band The weight, frequency band The frequency band normalized energy, frequency band Frequency band resolution, The physical conformity of the frequency domain features represents the feature. The degree of conformity with the physical process of electric arc discharge.

[0009] Optionally, the functional expression for determining the feature domain weights of each feature domain based on the quality of the features in each feature domain in step S103 is as follows: ; ; in, The feature domain weights are the feature domain weights of the k-th feature domain. Let be the physical credibility gain coefficient for the k-th feature domain. The average quality of the features in the k-th feature domain. The optimized feature subset after removing features that do not meet the quality requirements from the k-th feature domain. The number of features, Features The quality.

[0010] Optionally, the function expression for performing cross-domain global importance evaluation on the combined feature set of the optimized feature subsets of each feature domain in step S104 is as follows: ; in, Features within the combined feature set The global importance assessment score, Features within the combined feature set Feature domain Feature domain weights, For the feature domain within the combined feature set Chinese characteristics quality The redundancy penalty strength coefficient, Features and characteristics The square of the Pearson correlation coefficient, Features within the combined feature set Quality, characteristics For the feature domain Chinese characteristics Other characteristics.

[0011] Optionally, step S105 includes: S201: Sort all features in the combined feature set in descending order of their global importance assessment scores. S202, from all features sorted in descending order, select the first K features that are expected to be selected to form a feature candidate set; S203, calculate the information loss rate of the feature candidate set; S204, determine whether the information loss rate of the feature candidate set is less than the preset information loss rate threshold. If it is true, the feature candidate set is taken as the final optimized feature subset and the process jumps to step S106; otherwise, the process jumps to step S205. S205, reduce the redundancy penalty strength coefficient used when performing cross-domain global importance assessment on the combined feature set of the optimized feature subsets of each feature domain, and re-perform cross-domain global importance assessment on the combined feature set of the optimized feature subsets of each feature domain to calculate the global importance assessment score of all features in the combined feature set, then jump to step S201.

[0012] The present invention also provides a multi-level feature selection and detection system for terminal fault arcs, including a microprocessor and a memory connected to each other, wherein the microprocessor is programmed or configured to execute the multi-level feature selection and detection method for terminal fault arcs.

[0013] The present invention also provides a computer-readable storage medium storing a computer program or instructions that are programmed or configured to execute the multi-level feature selection and detection method for terminal fault arcs by a processor.

[0014] The present invention also provides a computer program product, including a computer program or instructions, which are programmed or configured to execute the multi-level feature selection and detection method for terminal fault arcs via a processor.

[0015] Compared with existing technologies, the present invention mainly achieves the following beneficial effects: The multi-level feature selection and detection method for terminal fault arcs of the present invention first targets the original current waveform data at the terminal, and features from multiple feature domains; then, it enters the first level, independently performing quality assessments that conform to their physical characteristics within each feature subspace and eliminating low-quality features; subsequently, it enters the second level, merging the preferred features from each subspace into a combination set, dynamically calculating the subspace weights to reflect the differences in importance between domains, and attenuating the importance of highly correlated feature pairs through a cross-domain redundancy penalty mechanism; then, under the hard constraint of target dimension and the soft constraint of information loss, a refined feature subset is generated; finally, the optimized feature subset obtained based on this method is integrated into the classifier model to achieve fault arc detection and identification. The multi-level feature selection and detection method for terminal fault arcs of the present invention can achieve pre-emptive lightweight selection and fault detection without relying on downstream identification models, ensuring the effectiveness and adaptability of features to terminal arc detection, and can effectively prevent accidents such as fault arc fires. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the basic process of the method in an embodiment of the present invention. Detailed Implementation

[0017] To enable those skilled in the art to better understand the technical solutions of the present invention, the technical solutions of the present invention will be further described in detail below with reference to the accompanying drawings in the embodiments of the present invention.

[0018] like Figure 1 As shown, the multi-level feature selection and detection method for terminal block fault arcs in this embodiment includes the following steps: S101, collect samples of the original current waveform data of the terminals under various load conditions, construct a sample set consisting of the terminal fault arc sample set and the normal operating condition sample set, extract features of multiple feature domains from the samples in the sample set, and obtain the original feature set consisting of feature subsets of multiple feature domains. S102, perform quality assessment on the features in the original feature set, remove features that do not meet the quality requirements, and obtain the optimized feature subsets for each feature domain; S103, determine the feature domain weights of each feature domain based on the quality of the features in each feature domain; S104. Based on the feature domain weights of each feature domain, perform cross-domain global importance evaluation on the combined feature set of the optimized feature subsets of each feature domain to obtain the global importance evaluation score of each feature in the combined feature set. S105, based on the global importance evaluation scores of each feature in the combined feature set, a screening decision is performed on the features in the combined feature set according to the hard constraint of target dimension and the soft constraint of information loss shown in the following formula, finally generating an optimized feature subset: ; ; in, To optimize feature subsets, For combined feature sets, Represents the set of candidate features; Indicates the number of features in the feature candidate set; Features within the combined feature set The global importance assessment score, The desired number of features to be selected. For information loss rate, The information loss rate threshold (e.g., a value of 15%). S106, fault detection of terminal blocks is performed based on features in the optimized feature subset.

[0019] Step S101 aims to convert the original physical signal into a calculable numerical feature space, providing an input basis for subsequent screening. In this embodiment, step S101 sets up three types of load conditions in a low-voltage distribution circuit with rated voltage and rated current: purely resistive, mixed resistive-inductive, and nonlinear rectifier. Terminal current data is collected at a fixed sampling frequency, with one sample extracted per power frequency cycle, accumulating a sample set of [data missing]. A sample set of terminal fault arcs was constructed respectively. Sample set under normal operating conditions Specifically, in this embodiment, a low-voltage power distribution environment with a rated voltage of 220V and a rated current of 20A is selected. The sampling frequency of the current data is 100kHz, and samples are extracted every 20ms of each power frequency cycle, accumulating 300 samples, including 150 samples of fault arc and 150 samples of normal operating conditions.

[0020] The functional expression of the original feature set obtained in step S101 of this embodiment is: ; in, For the original feature set, ~ These are feature subsets of the 1st to kth feature domains, and feature subsets of any kth feature domain. Indicates terminal fault arc sample set Sample set under normal operating conditions In the The feature set extracted from each feature subspace contains Dimensional features, the original feature set has the following dimensions: ; Where N is the dimension of the original feature set, To determine the number of feature domains, the feature domains in this embodiment can cover time domain, frequency domain, entropy domain, time-frequency domain, etc. For example, for each current data segment in the sample set, the time domain feature is extracted: current peak value. (A) Current rise slope (A / us), zero-day work week (us), waveform distortion factor and peak value (A); Extracting frequency domain features: energy proportion in the 1-5kHz frequency band Energy proportion in the 5-10kHz frequency band Energy proportion in the 10-20kHz frequency band odd harmonic ratio (Ratio of 3rd / 5th / 7th harmonic energy to total harmonic energy) and spectral entropy (Normalization, range of values) Zero-day work week The calculations include current sampling sequences within a single power frequency cycle. Threshold detection is performed, where , Set the current threshold for the number of sampling points per cycle. (Typically, 5% to 10% of the peak rated current is taken), here we take 5%. A comprehensive statistical analysis is performed to satisfy... Number of consecutive sampling points Then zero-day work week The expression for the computation function is: ; in, To meet The number of consecutive sampling points, Sampling frequency. Zero rest time. The duration of the arc current directly quantifies the duration of the arc-extinguished state after crossing zero due to insufficient recovery strength of the arc gap medium. It is a direct measure of the physical mechanism of the "zero-out-reignition" of series arcs.

[0021] Waveform distortion factor The calculation includes extracting the peak value of the current waveform within a single power frequency cycle. With the effective value of current ,in: ; The peak value factor is then calculated as the waveform distortion factor. The function expression is: .

[0022] Step S102 involves quality evaluation of the features in the original feature set, constituting the first-level screening. In this first-level screening, different features within different domains are evaluated and screened based on quality evaluation functions constructed within those domains. In this embodiment, a unified design framework for heterogeneous subspace evaluation operators is constructed for the first-level screening: targeting the original feature set... The differences in physical characterization properties between features can be assessed using independent evaluation functions within different feature domains. In this embodiment, step S102, when assessing the quality of features in the original feature set, includes performing quality assessments on the features in the original feature set at the feature domain level: ; in, Features quality , and These are the weight values ​​for the quality assessment indicators; , and The quality assessment indicators are as follows: inter-class separability indicator, operating condition robustness indicator, and physical compliance indicator. Among them, the inter-class separability indicator is used to quantify the characteristics in the terminal fault arc sample set. Sample set under normal operating conditions The distribution differences are considered; the robustness index is used to quantify the stability of features under various operating conditions; the physical conformity index is used to quantify the degree of conformity between features and the physical processes of arc discharge (such as ionization, thermal runaway, electromagnetic transients, etc.), and the value of the physical conformity index can be determined based on whether the feature calibration value of the feature and the physical process of arc discharge are equal or the error is less than a preset threshold; for each feature domain, the screening rules remove features that do not meet the quality requirements: ; in, The k-th optimized feature subset is obtained after removing features that do not meet the quality requirements by selecting rules. The quality threshold for the k-th feature domain. Let be the selection coefficient for the k-th feature domain, and max be the maximum value. The selection coefficient for the k-th feature domain is usually set to . By setting a filtering threshold, feature subsets from different feature domains are selected respectively. All features within the domain are scored using the evaluation function of the corresponding domain, and those with scores higher than [the specified domain] are retained. The features are used to form an optimized feature subset. Dimensions reduced to This step performs preliminary quality screening of features without introducing cross-domain interference, preventing invalid features from entering subsequent calculations. At this point, the quality within each subset has been optimized, but information overlap still exists between features in different domains, such as the strong correlation between current abrupt changes and high-frequency noise. Therefore, a second level of cross-domain collaborative optimization is required.

[0023] In the first-level screening, quality evaluation functions conforming to the physical properties of subspaces with different feature domains are constructed. The design framework of heterogeneous subspace evaluation operators can support any number of... The heterogeneous feature subspaces are not limited to the time and frequency domains. For newly added subspaces, such as the wavelet domain and mode decomposition domain, only the corresponding quality evaluation function needs to be defined and embedded in the unified framework, without modifying the overall architecture. The following sections will further illustrate the definition methods of quality evaluation functions in the time and frequency domains using multiple feature domains, including the time and frequency domains, as examples.

[0024] Features of the time domain The function expression for the quality assessment function used to evaluate the quality is: ; in, Features of the time domain quality , and As weight, , and Features of the time domain The indices include inter-class separability, operational condition robustness, and physical compliance, and include: ; ; ; in, and Features The sample mean on the terminal fault arc sample set and the normal operating condition sample set. and Features The sample standard deviation on the terminal fault arc sample set and the normal operating condition sample set. It is a constant used to prevent the denominator from being divided by zero; The load variation coefficient is a characteristic. The ratio of the standard deviation to the mean under different load types; The physical conformity of the time-domain feature represents the feature. The degree of conformity with the physical process of arc discharge can be determined by whether the time-domain characteristics are equal to or have an error less than a preset threshold, thus setting the value of the physical conformity index. Its value can be determined based on prior knowledge of the transient current characteristics of the arc discharge physical process. Based on the arc physical mechanism and expert experience, different time-domain characteristics characterize different arc dynamic processes: for zero-time... It reflects the duration of zero current near the zero-crossing point of the arc current. This is the most fundamental physical phenomenon of a series arc—when the current crosses zero, the arc channel is temporarily extinguished, and it needs to wait for the voltage to recover to the breakdown voltage before it can reignite. This characteristic directly corresponds to the "zero-crossing-reignition" physical mechanism of the arc, and has clear physical interpretability and high discrimination specificity; regarding the current rise slope It reflects the rate of change of current during the instantaneous breakdown of the arc. When the gas gap breaks down to form a conductive channel, the current rises sharply within microseconds. This characteristic can capture the instantaneous change in current during arcing; for the peak current... Peak-to-peak value While it reflects the amplitude variation range of the arc current and can demonstrate the current increase caused by the arc, it is easily affected by load power level and grid voltage fluctuations, resulting in relatively weak physical specificity. Regarding waveform distortion factors... This reflects the degree of deviation of the current waveform from a standard sine wave. The nonlinear resistive characteristics of the electric arc cause distortions such as clipping and burrs in the current waveform. This feature can characterize the waveform quality degradation caused by the electric arc. By mapping the above time-domain features to the physical process of "breakdown-burning-extinguishing-reignition" of the electric arc, experts assign corresponding physical conformity scores based on the closeness of the correlation between each feature and the essential mechanism of the electric arc, ensuring that the selected time-domain features can accurately reflect the transient physical nature of the electric arc. The subspace of the time-domain feature domain... When evaluating and filtering features, the evaluation function is used to quantify the features. The ability to distinguish between arc states and normal states, and the inter-class separability index. Operating condition robustness indicators Physical compliance indicators weight , and The values ​​can be 0.5, 0.3, and 0.2 respectively.

[0025] Inter-class separability index Fisher's discriminant variant can be used, with linear superposition of the denominators to avoid square operations, thus reducing computational load in hardware. The numerator is the distance between the centers of the two classes of samples, and the denominator is the sum of the cluster radii; the ratio directly reflects the inter-class overlap probability. In arc detection, the arc state dispersion of current abrupt changes is greater, and a linear denominator can balance the "center separation but morphological diffusion" characteristic. Inter-class separability index The definition is as follows: ; in, and The two parameters represent the features respectively. In the terminal fault arc sample set Sample set under normal operating conditions The mean of the above; and Representing features respectively In the terminal fault arc sample set Sample set under normal operating conditions The sample standard deviation is used to measure the dispersion of the data; It is a very small positive number. Pick The formulas for calculating the mean and standard deviation are shown below. Given the size of the sample set, the functions for calculating the mean and standard deviation are: .

[0026] in, Features Feature set, and Corresponding to this feature The mean and standard deviation are given.

[0027] Operating condition robustness index Selectable load variation coefficient Characterization, this parameter is used to quantify features Stability under different load types, such as purely resistive loads, mixed resistive-inductive loads, and nonlinear loads. Mathematically, the load variation coefficient is defined. The formula is shown below: ; in, It is a feature Standard deviation under different load types It is a feature Average value under different load types. This parameter reflects the characteristic's sensitivity to load changes. Smaller The value indicates that the characteristic changes little under different load conditions, demonstrating good robustness. Therefore, the operating condition robustness index... The definition is as follows: .

[0028] Physical compliance indicators Physical compliance can be selected To perform characterization for quantifying features The theoretical correlation strength with the physical mechanism of the electric arc core. Mathematically, This can be preset to a value between 0 and 1 based on expert knowledge or historical data, representing the degree to which the characteristic matches the physical process of arc discharge. For example, for a characteristic describing a sudden change in current, its... It might be close to 1 because it is directly related to the rapid change in current during arcing; while for features with a weaker correlation to arcing, its... The physical compliance index may be low. The definition is as follows: ; Combining the above parameters, the quality evaluation function for time-domain features This can be specifically expressed as: ; This quality assessment function comprehensively considers the consistency of time-domain characteristics in distinguishing between arc and normal states, cross-load stability, and the physical process of arc discharge, providing a quantitative basis for subsequent feature selection. In this embodiment, the process variables involved in the calculation of different features in the quality assessment function of time-domain characteristics are shown in Table 1.

[0029] Table 1: Calculation parameters of the quality assessment function for time-domain features

[0030] In Table 1, and Five time-domain indicators ~ The sample mean on the terminal fault arc sample set and the normal operating condition sample set. and Five time-domain indicators ~ The sample standard deviation on the terminal fault arc sample set and the normal operating condition sample set. Five time-domain indicators ~ In the load variability coefficient, Five time-domain indicators ~ Physical conformity of time-domain features Five time-domain indicators ~ The quality.

[0031] In this embodiment, when performing quality evaluation on features in the original feature set at the feature domain level, the focus is on features in the frequency domain. The function expression for the quality assessment function used to evaluate the quality is: ; in, Features quality , and As weight, , and Features The indices are: inter-class separability, operating condition robustness, and physical compliance. The operating condition robustness index is a constant, and the following applies: ; ; in, frequency band The weight, frequency band The frequency band normalized energy, frequency band Frequency band resolution, The physical conformity of the frequency domain features represents the feature. The degree of conformity with the physical process of arc discharge can be determined by whether the time-domain characteristics are equal to or have an error less than a preset threshold, based on whether the characteristic calibration values ​​of the time-domain characteristics and the physical process of arc discharge are equal. The physical conformity of the frequency-domain characteristics... This indicates the degree of conformity between the characteristics and the physical process of arc discharge, and its value is determined based on prior knowledge of the spectral characteristics of arc discharge (depending on expert experience). Specifically, when a fault arc occurs, the gas gap breaks down to generate a plasma channel, accompanied by broadband electromagnetic radiation and high-frequency oscillations of the current waveform. This physical mechanism manifests in the frequency domain as an abnormal distribution of energy in a specific frequency band and a significant change in the spectral structure. Based on the physical mechanism of electric arcs and expert experience, different frequency domain characteristics characterize different physical states of electric arcs: For the energy proportion in the 1-5kHz frequency band, this band mainly covers the fundamental power frequency and its low-order harmonics, reflecting the overall distortion of the power frequency current waveform caused by the electric arc; for the energy proportion in the 5-10kHz frequency band, this band corresponds to the mid-frequency noise in the unstable combustion stage of the electric arc, reflecting the random fluctuation characteristics of the arc resistance; for the energy proportion in the 10-20kHz frequency band, this band corresponds to the high-frequency oscillations and electromagnetic pulses generated at the moment of arc breakdown, which is one of the most significant spectral characteristics of the electric arc; for the odd harmonic ratio, the nonlinear volt-ampere characteristics of the electric arc cause the current waveform to show odd harmonic enrichment, and this ratio reflects the degree of nonlinear distortion of the electric arc; for the spectral entropy, the randomness and chaotic characteristics of arc discharge lead to an increase in the uncertainty of the spectral energy distribution, and the increase in entropy reflects the complexity and disorder of the arc state. By matching and mapping these frequency domain features with the physical mechanism of electric arcs and assigning corresponding physical conformity scores, it is ensured that the selected frequency domain features can accurately reflect the essential physical characteristics of electric arcs from the perspective of spectral energy distribution.

[0032] When constructing a quality evaluation function for features in the frequency domain feature domain, a subset of features in the frequency domain feature domain is used. The features in the data are evaluated and filtered. Quality evaluation function. Used for quantifying features The ability to distinguish between arc states and normal states. This function consists of three parts: an inter-class separability index. Operating condition robustness indicators Physical compliance indicators Among them, the inter-class separability index Selectable frequency band normalized energy To characterize, to represent features In total frequency bands In the middle of The energy of a frequency band is expressed by the following formula: ; in, It is a feature The Fourier transform of the frequency represents the frequency. Spectral density at that location; This indicates the total number of frequency bands into which the entire frequency range is divided, as specified. and It is the Nyquist frequency.

[0033] In this embodiment, frequency band resolution is introduced. Further measure the terminal fault arc sample set Sample set under normal operating conditions The difference in energy distribution within a specific frequency band is defined by the following expression: ; in, This indicates that all samples in the sample set are calculated. Calculate the average of the values; This is a constant (a minimal constant) used to prevent the denominator from being zero. Then, for each frequency band... Assign a weight This reflects the importance of different frequency bands in arc detection. For example, arc faults are usually accompanied by high-frequency noise, and high-frequency bands are often given higher weight. Meet the conditions .

[0034] Therefore, the inter-class separability index It can be defined as follows: ; Since frequency domain energy calculation is based on FFT, the influence of the window function has been smoothed through integration, and the fluctuation error under different operating conditions is extremely small and within the tolerance range of numerical error. Therefore, no independent robustness term is established. Thus, robustness indicators for operating conditions can be used. Set as In addition, it can be set to other constants as needed.

[0035] Physical compliance indicators Physical compliance can be selected To perform characterization for quantifying features The theoretical correlation strength with the core physical mechanism of electric arcs. Different characteristics can be quantified into a value between 0 and 1 based on prior knowledge or expert experience, representing the degree of agreement between the characteristic and the physical process of electric arc discharge. If the characteristic... It primarily characterizes high-frequency energy, because arc discharge is accompanied by strong high-frequency noise. It is close to 1.

[0036] Combining the above parameters, the quality evaluation function of the frequency domain features This can be specifically expressed as: .

[0037] In this embodiment, take , and The values ​​are 0.7, 0, and 0.3 respectively. The frequency band is divided into high frequency, mid frequency, and low frequency. Equals 3; take the frequency band statistical weights for high frequency, mid frequency, and low frequency. (Right now , and The values ​​are 0.2, 0.3, and 0.5 respectively.

[0038] In this embodiment, the process variables involved in the calculation of different features in the quality evaluation function of frequency domain features are shown in Table 2.

[0039] Table 2: Calculation parameters of the quality assessment function for frequency domain features

[0040] In Table 2, , and Five frequency domain indicators ~ Energy in the high-frequency, mid-frequency, and low-frequency bands , Five frequency domain indicators ~ The energy mean and normalized energy mean. Five frequency domain indicators ~ Frequency band resolution, Five frequency domain indicators ~ Inter-class separability index; Five frequency domain indicators ~ The quality.

[0041] In this embodiment, the first-level filtering settings ,but ,reserve The characteristics obtained Dimension Removed and .set up ,but ,reserve The characteristics obtained Dimension Removed , and .

[0042] The first-level screening performs preliminary quality screening of features without introducing cross-domain interference, preventing invalid features from entering subsequent calculations. At this point, the quality within each subset has been optimized, but information overlap still exists between features in different domains, such as the strong correlation between current abrupt changes and high-frequency noise. Therefore, a second-level cross-domain collaborative optimization is required.

[0043] This embodiment includes dynamically calculating the weights of each feature subset in the second-level filtering to reflect the differences in importance between domains. The combined feature set is constructed as follows: ; Each optimized feature subset Include Each feature; each optimized feature subset Include One feature; original feature set Total dimensions ,in and The same meaning, indicating The number of features within. Specifically, in this embodiment, the combined feature set is: Dimension .

[0044] In step S103 of this embodiment, the functional expression for determining the feature domain weights of each feature domain based on the quality of the features in each feature domain is as follows: ; ; in, The feature domain weights are the feature domain weights of the k-th feature domain. Let be the physical credibility gain coefficient for the k-th feature domain. The average quality of the features in the k-th feature domain. The optimized feature subset after removing features that do not meet the quality requirements from the k-th feature domain. The number of features, Features The quality of features in the k-th feature domain is the average quality of the feature subset. All features are retained. The average evaluation score reflects the overall quality level of the feature subset; this value can quantify the overall discriminative ability of feature subsets in the time domain, frequency domain, etc., after independent screening, for example... This demonstrates that under current operating conditions, time-domain features are generally more effective than frequency-domain features. The calculation formula uses the mean rather than the maximum value to avoid individual high-scoring features masking the overall quality of the feature subset, ensuring that the weight allocation reflects the collective performance of features within the domain. In the terminal block arc scenario, time-domain features are typically more sensitive to transient changes. The value is often higher. The physical reliability gain coefficient of the k-th feature domain. Used to adjust the influence of feature subset quality scores on the final weights, requiring , This is a previously set, extremely small constant greater than 0. The value can be determined based on the correlation between different feature subsets and the occurrence of terminal arcing. For example, for feature subsets in the time domain, the sudden change in current is the most direct and earliest physical symptom of terminal arcing, and has the highest theoretical reliability; therefore, it can be assigned... For the feature subset in the frequency domain, high-frequency noise only becomes significant after the arc has stabilized, resulting in information lag and susceptibility to power grid harmonic interference. Therefore, the gain can be set relatively low. Other feature subsets can be evaluated for magnitude based on prior knowledge or expert knowledge.

[0045] In step S103 of this embodiment, when determining the feature domain weights of each feature domain based on the quality of features in each feature domain, the feature domain weight of the k-th feature domain is... Representing a feature subset Weight values ​​in global filtering. For example, in the case of only two feature subsets, time domain and frequency domain, if and Therefore, the importance of time-domain features is assigned a weight of 67%, while the frequency domain accounts for 33%. The Softmax function can be used to convert the linear score... Normalization to a probability distribution makes And satisfy Compared to linear normalization, the Softmax function can more significantly differentiate between two feature subsets and prevent invalid features from lowering the overall screening quality. This completes the feature subset weighting. After calculation, each feature obtained an initial value for cross-domain comparability, but this value did not consider cross-domain redundancy. For example, the "current mutation index" in the time domain and the "high-frequency energy integral" in the frequency domain may characterize the same physical phenomenon; if directly calculated... Retaining the sorted order would result in information waste. Therefore, a further step is needed to eliminate cross-domain feature overlap using a redundancy penalty term. In this embodiment, during the weight calculation of the feature subset, the following is calculated: , ; Credibility coefficient of a given feature subset and The feature subset weights can be calculated. and .

[0046] In step S104 of this embodiment, the function expression for performing cross-domain global importance evaluation on the combined feature set of optimized feature subsets of each feature domain is as follows: ; in, Features within the combined feature set The global importance assessment score, Features within the combined feature set Feature domain Feature domain weights, For the feature domain within the combined feature set Chinese characteristics quality Redundancy penalty strength coefficient (value range is) (This can be set based on experience). Features and characteristics The square of the Pearson correlation coefficient, Features within the combined feature set Quality, characteristics For the feature domain Chinese characteristics Other features. and characteristics The Pearson correlation coefficient is calculated using the following function: ; in, Features and characteristics covariance, and Features and characteristics The standard deviation. In step S104 of this embodiment, the function expression for performing cross-domain global importance evaluation on the combined feature set of the optimized feature subsets of each feature domain, Representation of features The weighted source scores can be converted into a globally unified scale, achieving preliminary comparability of features from different domains; As a redundancy penalty term, it is calculated by characteristic Pearson correlation coefficient with all other features within or outside the domain To represent the information overlap rate between different features, and combined with The term represents the weight of score differences. For example, if two features are highly correlated but have large score differences, then... The calculated value is small but If the calculated result is large, the penalty term is significant, and high-scoring features are retained while low-scoring redundant features are eliminated, thus selecting the best from the best. If the scores of two features are similar, i.e. When the value is approximately zero, even if the two types of features are correlated, the penalty term approaches zero, allowing both to be retained to capture subtle discriminative information. When the two feature information highly overlaps, the marginal contribution of the suboptimal feature decreases sharply and should be suppressed. Specifically, in this embodiment, a redundancy penalty strength coefficient is given. The calculation results are shown in Table 3.

[0047] Table 3: Process Parameters in the Calculation of the Global Importance Evaluation Function

[0048] Table 3 lists the features within the combined feature set. , , , and The process parameters are obtained, including those in their respective feature domains. Feature domain weights Feature domain within the combined feature set Quality of features Weighted source score, redundancy penalty, and global importance assessment score .

[0049] Step S105 evaluates the global importance scores of each feature in the combined feature set, and then performs a screening decision on the features in the combined feature set based on the hard constraint of target dimension and the soft constraint of information loss shown in the following formula, ultimately generating an optimized feature subset: ; ; in, To optimize feature subsets, For combined feature sets, Represents the set of candidate features; Indicates the number of features in the feature candidate set; Features within the combined feature set The global importance assessment score, The desired number of features to be selected. For information loss rate, The information loss rate threshold is set (e.g., a value of 15%); step S105 in this embodiment includes: S201: Sort all features in the combined feature set in descending order of their global importance assessment scores. S202, from all features sorted in descending order, select the first K features that are expected to be selected to form a feature candidate set; S203, calculate the information loss rate of the feature candidate set; S204, determine whether the information loss rate of the feature candidate set is less than the preset information loss rate threshold. If it is true, the feature candidate set is taken as the final optimized feature subset and the process jumps to step S106; otherwise, the process jumps to step S205. S205, reduce the redundancy penalty strength coefficient used when performing cross-domain global importance assessment on the combined feature set of the optimized feature subsets of each feature domain, and re-perform cross-domain global importance assessment on the combined feature set of the optimized feature subsets of each feature domain to calculate the global importance assessment score of all features in the combined feature set, then jump to step S201.

[0050] Specifically, in step S105 of this embodiment, the combined feature set is first... All features within Sort in descending order, take the first few. The features constitute the candidate set Next, calculate. ,like The final selection set Equal to feature candidate set The process ends; if Then, the redundancy penalty is relaxed, reducing the redundancy penalty strength coefficient in step S06. Recalculate Then sort them until the information loss constraint is satisfied. The final output is... This refers to the final optimized feature subset with fixed dimensions and effective reduction of cross-domain redundancy. Specifically, in this embodiment, the given parameters... ,according to Arranged in descending order, the first four features can be taken as follows: , , and It is possible to calculate at this time Therefore, the optimized feature subset is: ; This optimized feature subset has a fixed dimension and effectively reduces cross-domain redundancy.

[0051] Finally, fault detection of the terminal block can be performed based on the features in the optimized feature subset in step S106. Fault detection of the terminal block based on the features in the optimized feature subset can be performed using methods as needed, such as thresholding, machine learning classification methods, clustering methods, etc. For example, the optimized feature subset... As a standardized input feature vector, a detection sample is constructed using the real-time calculated values ​​of each preferred feature within a fixed time window. This sample is then input into a classifier model. The input vector, constructed using an optimized feature subset, is mapped by the model to output a discrimination result. By setting fault determination conditions, when the discrimination result meets the preset fault determination conditions, the occurrence of a fault arc is confirmed. The classifier model includes, but is not limited to, machine learning or deep learning models such as support vector machines, random forests, neural networks, and long short-term memory networks. The specific selection can be flexibly configured according to hardware resources and accuracy requirements.

[0052] In summary, this embodiment addresses the bottlenecks of existing terminal block fault arc detection technologies, such as lagging feature selection, single evaluation dimensions, and insufficient cross-domain collaboration. It proposes a multi-level feature selection and detection method for terminal block fault arcs. To address the resource waste caused by the lack of a proactive feature selection mechanism for terminal block fault arcs, this embodiment provides a heterogeneous feature subset autonomous evaluation and two-level selection architecture. This architecture embeds a decoupled quantization evaluator between the feature extraction layer and the recognition model layer, evaluating inter-class separation in the time domain and frequency band discrimination in the frequency domain. and physical conformity By employing multiple metrics, pre-processing quality assessment and low-quality feature removal are performed before model training, preventing invalid features from entering the computation process and reducing the time complexity of feature selection. Addressing the technical issues of a single importance assessment dimension and uncontrollable cross-domain redundancy, this embodiment proposes a feature subset weighting-based approach. A collaborative refinement mechanism with redundant penalty terms. This is achieved through a reliability coefficient. It dynamically reflects the differences in physical importance of feature subsets in the time and frequency domains, and utilizes the squared correlation coefficient. Penalizing highly correlated cross-domain feature pairs proactively eliminates information overlap and reduces the risk of model overfitting. To address the issue of insufficient synergy between feature processing and model optimization, this embodiment designs a target dimension. Information loss rate A hard decision-making mechanism with joint constraints. Among them... It can be flexibly configured according to needs. The method mandates that information loss not exceed 15% to ensure controllable accuracy loss during dimensionality compression, achieving a synergistic optimization of the dual objectives of feature space reconstruction and model performance improvement. To verify the effectiveness of the method in this embodiment, a linear support vector machine is used as the recognition model, with a linear kernel function and a penalty factor C set to 1.0. The input is the optimized feature subset generated in step S107. The constructed 5-dimensional feature vector corresponds to the real-time calculated values ​​of zero-downtime, current rise slope, peak-to-peak value, energy proportion in the 10-20kHz frequency band, and energy proportion in the 5-10kHz frequency band, respectively. The output label is a binary classification result of 0, where 0 represents the normal state and 1 represents the fault arc state. Five-fold cross-validation was performed on the same 300 samples, and the results are shown in Table 4.

[0053] Table 4: Experimental results of using linear support vector machines as the recognition model in this embodiment.

[0054] In Table 4, the unit for time consumption is milliseconds (ms), and the unit for memory usage is bytes. Table 4 shows the optimized feature subset used in this embodiment. Relative to the original feature set In contrast, it has streamlined five feature dimensions and achieved better results in five value tables: fault arc detection accuracy, fault arc detection false alarm rate, fault arc detection false alarm rate, time consumption, and memory usage.

[0055] This embodiment also provides a multi-level feature selection and detection system for terminal fault arcs, including a microprocessor and a memory connected to each other, wherein the microprocessor is programmed or configured to execute the multi-level feature selection and detection method for terminal fault arcs.

[0056] This embodiment also provides a computer-readable storage medium storing a computer program or instructions that are programmed or configured to execute the multi-level feature selection and detection method for terminal fault arcs via a processor.

[0057] This embodiment also provides a computer program product, including a computer program or instructions, which are programmed or configured to execute the multi-level feature selection and detection method for terminal fault arcs via a processor.

[0058] Those skilled in the art will understand that the technical solutions provided by this invention may take the form of a method, system, or computer program product. Therefore, this invention may take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this invention may take the form of a computer program product embodied on one or more computer-readable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, produce an implementation of the flowchart. Figure 1 One or more processes and / or boxes Figure 1 The computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The functions specified in one or more boxes. These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable apparatus for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0059] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A method for selecting and detecting multi-level characteristics of fault arcs in terminal blocks, characterized in that, Includes the following steps: S101, collect samples of the original current waveform data of the terminals under various load conditions, construct a sample set consisting of the terminal fault arc sample set and the normal operating condition sample set, extract features of multiple feature domains from the samples in the sample set, and obtain the original feature set consisting of feature subsets of multiple feature domains. S102, perform quality assessment on the features in the original feature set, remove features that do not meet the quality requirements, and obtain the optimized feature subsets for each feature domain; S103, determine the feature domain weights of each feature domain based on the quality of the features in each feature domain; S104. Based on the feature domain weights of each feature domain, perform cross-domain global importance evaluation on the combined feature set of the optimized feature subsets of each feature domain to obtain the global importance evaluation score of each feature in the combined feature set. S105, based on the global importance evaluation scores of each feature in the combined feature set, a screening decision is performed on the features in the combined feature set according to the hard constraint of target dimension and the soft constraint of information loss shown in the following formula, finally generating an optimized feature subset: ； ； in, To optimize feature subsets, For combined feature sets, Represents the set of candidate features; Indicates the number of features in the feature candidate set; Features within the combined feature set The global importance assessment score, The desired number of features to be selected. For information loss rate, This is the information loss rate threshold; S106, fault detection of terminal blocks is performed based on features in the optimized feature subset.

2. The method for selecting and detecting multi-level characteristics of fault arcs in terminal blocks according to claim 1, characterized in that, The functional expression of the original feature set obtained in step S101 is: ； in, For the original feature set, ~ These are feature subsets of the 1st to kth feature domains, and feature subsets of any kth feature domain. Include Dimensional features, the original feature set has the following dimensions: ； Where N is the dimension of the original feature set, This represents the number of feature domains.

3. The method for selecting and detecting multi-level characteristics of terminal fault arcs according to claim 1, characterized in that, Step S102, when evaluating the quality of features in the original feature set, includes evaluating the quality of features in the original feature set at the feature domain level: ； in, Features quality , and These are the weight values ​​for the quality assessment indicators; , and The quality assessment indicators are as follows: inter-class separability, operating condition robustness, and physical conformity. Specifically: the inter-class separability indicator quantifies the difference in feature distribution across terminal fault arc sample sets and normal operating condition sample sets; the operating condition robustness indicator quantifies the stability of features across different operating conditions; and the physical conformity indicator quantifies the degree of conformity between features and the physical process of arc discharge. For each feature domain, screening rules eliminate features that do not meet the quality requirements. ； in, The k-th optimized feature subset is obtained after removing features that do not meet the quality requirements by selecting rules. The quality threshold for the k-th feature domain. is the screening coefficient for the k-th feature domain, and max is the maximum value.

4. The method for selecting and detecting multi-level characteristics of terminal fault arcs according to claim 3, characterized in that, The multiple feature domains include the time domain and the frequency domain; when performing quality evaluation on the features in the original feature set at the feature domain level, the time domain features are considered. The function expression for the quality assessment function used to evaluate the quality is: ； in, Features of the time domain quality , and As weight, , and Features of the time domain The indices include inter-class separability, operational condition robustness, and physical compliance, and include: ； ； ； in, and Features The sample mean on the terminal fault arc sample set and the normal operating condition sample set. and Features The sample standard deviation on the terminal fault arc sample set and the normal operating condition sample set. It is a constant used to prevent the denominator from being divided by zero; The load variation coefficient is a characteristic. The ratio of the standard deviation to the mean under different load types; The physical conformity of the time-domain feature represents the feature. The degree of conformity with the physical process of electric arc discharge; when evaluating the quality of features in the original feature set at the feature domain granularity, the focus is on the features in the frequency domain. The function expression for the quality assessment function used to evaluate the quality is: ； in, Features quality , and As weight, , and Features The indices are: inter-class separability, operating condition robustness, and physical compliance. The operating condition robustness index is a constant, and the following applies: ； ； in, frequency band The weight, frequency band The frequency band normalized energy, frequency band Frequency band resolution, The physical conformity of the frequency domain features represents the feature. The degree of conformity with the physical process of electric arc discharge.

5. The method for selecting and detecting multi-level characteristics of fault arcs in terminal blocks according to claim 1, characterized in that, In step S103, the functional expression for determining the feature domain weights of each feature domain based on the quality of the features in each feature domain is as follows: ； ； in, The feature domain weights are the feature domain weights of the k-th feature domain. Let be the physical credibility gain coefficient for the k-th feature domain. The average quality of the features in the k-th feature domain. The optimized feature subset after removing features that do not meet the quality requirements from the k-th feature domain. The number of features, Features The quality.

6. The method for selecting and detecting multi-level characteristics of terminal fault arcs according to claim 1, characterized in that, The function expression for performing cross-domain global importance evaluation on the combined feature set of optimized feature subsets of each feature domain in step S104 is as follows: ； in, Features within the combined feature set The global importance assessment score, Features within the combined feature set Feature domain Feature domain weights, For the feature domain within the combined feature set Chinese characteristics quality The redundancy penalty strength coefficient, Features and characteristics The square of the Pearson correlation coefficient, Features within the combined feature set Quality, characteristics For the feature domain Chinese characteristics Other characteristics.

7. The method for selecting and detecting multi-level characteristics of terminal fault arcs according to claim 6, characterized in that, Step S105 includes: S201: Sort all features in the combined feature set in descending order of their global importance assessment scores. S202, from all features sorted in descending order, select the first K features that are expected to be selected to form a feature candidate set; S203, calculate the information loss rate of the feature candidate set; S204, determine whether the information loss rate of the feature candidate set is less than the preset information loss rate threshold. If it is true, the feature candidate set is taken as the final optimized feature subset and the process jumps to step S106; otherwise, the process jumps to step S205. S205, reduce the redundancy penalty strength coefficient used when performing cross-domain global importance assessment on the combined feature set of the optimized feature subsets of each feature domain, and re-perform cross-domain global importance assessment on the combined feature set of the optimized feature subsets of each feature domain to calculate the global importance assessment score of all features in the combined feature set, then jump to step S201.

8. A multi-level feature selection and detection system for terminal block fault arcs, comprising a microprocessor and a memory interconnected thereon, characterized in that, The microprocessor is programmed or configured to execute the multi-level feature selection and detection method for terminal fault arcs as described in any one of claims 1 to 7.

9. A computer-readable storage medium storing a computer program or instructions, characterized in that, The computer program or instructions are programmed or configured to execute, via a processor, the multi-level feature selection and detection method for terminal fault arcs as described in any one of claims 1 to 7.

10. A computer program product, comprising a computer program or instructions, characterized in that, The computer program or instructions are programmed or configured to execute, via a processor, the multi-level feature selection and detection method for terminal fault arcs as described in any one of claims 1 to 7.

Images