Method for extracting electromagnetic interference features from the acquisition ports of primary and secondary integrated power distribution switches

Abstract

This invention discloses a method for extracting electromagnetic interference features from the acquisition ports of a primary and secondary integrated power distribution switch, belonging to the field of electromagnetic interference prediction and simulation technology. The method includes: extracting time-domain feature vectors from the original time-domain signals synchronously acquired from multiple ports, and extracting frequency-domain feature vectors through Welch power spectrum estimation; extracting coupling path feature vectors from the voltage signals of the power supply port and signal port through cross-correlation analysis; extracting magnetic field feature vectors from the time-domain signal of the spatial magnetic field through oscillation mode analysis; and fusing the time-domain feature vectors, frequency-domain feature vectors, coupling feature vectors, and magnetic field feature vectors to obtain electromagnetic interference features. This invention solves the problem that existing technologies cannot accurately extract electromagnetic interference features from the acquisition ports of a primary and secondary integrated power distribution switch.

Description

Technical Field

[0001] This invention relates to the field of electromagnetic interference prediction and simulation technology, and in particular to a method for extracting electromagnetic interference features from the acquisition port of a primary and secondary integrated power distribution switch. Background Technology

[0002] In recent years, integrated primary and secondary distribution switches have become key equipment in smart distribution networks. The operation of high-voltage switches and high-power electronic devices on the primary side generates strong transient electromagnetic interference (EMI). This interference can disrupt the normal operation of sensitive electronic devices such as voltage and current sensors and communication modules at the secondary side acquisition ports through conduction, radiation, and coupling. In severe cases, it can lead to measurement distortion, protection malfunctions, or communication interruptions, directly affecting the reliability and intelligence level of the distribution network. Therefore, accurately extracting the characteristics of EMI at the acquisition ports is fundamental for locating interference sources, analyzing interference mechanisms, and subsequent electromagnetic compatibility (EMC) design and protection. Existing technologies primarily focus on analyzing the interference signals of single ports, such as power ports or signal ports, in the independent time or frequency domain. Common approaches include directly extracting statistical features such as peak value and rise time from the acquired time-domain waveform; or obtaining the spectrum through Fast Fourier Transform (FFT) and then analyzing its amplitude, dominant frequency, and other frequency domain characteristics. Some studies also employ time-frequency analysis methods such as wavelet transform to reveal the energy distribution of the interference signal in the joint time-frequency domain. These methods can reflect some of the time-frequency characteristics of electromagnetic interference to a certain extent.

[0003] However, the aforementioned existing methods have significant limitations and cannot meet the requirements for accurate and comprehensive feature extraction of electromagnetic interference (EMI) at ports of complex systems such as integrated primary and secondary power distribution switches. First, existing methods often analyze from a single physical dimension—time domain, frequency domain, or time-frequency domain—ignoring the differences in the manifestation of EMI across multiple ports, including power, signal, and spatial aspects, as well as the correlations between them. For example, analyzing only the voltage signal at a single port cannot reveal the coupling path characteristics of the EMI from the power port to each signal port. Second, the compact internal structure of integrated primary and secondary equipment makes the conducted coupling and near-field radiative coupling effects of EMI extremely significant. Existing technologies lack quantitative analysis of the cross-correlation between EMI signals at the power port and multiple signal ports, resulting in an inability to effectively characterize the EMI propagation path and coupling strength, leading to a disconnect between features and physical mechanisms. Third, the spatial magnetic field is an important characteristic of radiated EMI; existing methods either fail to collect this signal or only perform simple amplitude statistics, failing to delve into its underlying physical characteristics such as oscillation modes and damping properties. Therefore, the comprehensive shortcomings of existing technologies are that the extracted feature sets are one-sided and isolated, and cannot comprehensively and accurately construct feature models that can accurately reflect the essential attributes of electromagnetic interference at the primary and secondary integrated power distribution switch acquisition ports from a systematic perspective of multiple ports, multiple paths, and multiple physical fields. This limits the effectiveness of feature-based interference type identification, hazard assessment, and precise suppression strategies. Summary of the Invention

[0004] The technical problem solved by this invention is that existing technologies cannot accurately extract the electromagnetic interference characteristics of the primary and secondary fusion power distribution switch acquisition ports.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0006] As a preferred embodiment of the electromagnetic interference feature extraction method for the primary and secondary integrated power distribution switch acquisition ports described in this invention, wherein:

[0007] The time-domain feature vector is extracted from the original time-domain signal acquired synchronously from multiple ports, and the frequency-domain feature vector is extracted by Welch power spectrum estimation.

[0008] The coupling path feature vector is extracted by cross-correlation analysis based on the voltage signals of the power port and the signal port.

[0009] Magnetic field feature vectors are extracted from the time-domain signal of the spatial magnetic field through oscillation mode analysis.

[0010] Electromagnetic disturbance characteristics are obtained by fusing features from time-domain feature vectors, frequency-domain feature vectors, coupling feature vectors, and magnetic field feature vectors.

[0011] Furthermore, time-domain feature vectors are extracted from the raw time-domain signals acquired synchronously from multiple ports, and frequency-domain feature vectors are extracted through Welch power spectrum estimation, including:

[0012] The raw time-domain signal acquired synchronously from multiple ports is preprocessed, including denoising, normalization and alignment, to obtain the preprocessed time-domain signal;

[0013] The time-domain feature vector is extracted from the preprocessed time-domain signal. The time-domain feature vector includes statistical features such as mean, variance, peak value, impulse factor, waveform factor and margin factor, and is calculated using time-domain parameters.

[0014] Welch power spectrum estimation is performed on the preprocessed time-domain signal. The Hanning window function and the overlap segment with a preset percentage are used to calculate the power spectral density of each overlap segment and average them to obtain a smooth power spectral density.

[0015] Frequency domain feature vectors are extracted from the smooth power spectral density. These feature vectors include the peak frequency, total power, energy ratio of a specific frequency band, spectral centroid, and spectral flatness, and are calculated using frequency domain parameters.

[0016] Furthermore, based on the voltage signals of the power supply port and the signal port, the coupling path feature vector is extracted through cross-correlation analysis, including:

[0017] The voltage signals of the power supply port and the signal port are bandpass filtered respectively to obtain the filtered voltage signals of the power supply port and the voltage signals of each signal port.

[0018] The filtered power port voltage signal and the voltage signals of each signal port are sequentially aligned using a dynamic time warping algorithm to obtain the aligned power port voltage signal and the voltage signals of each signal port.

[0019] Calculate the normalized cross-correlation function between the aligned power port voltage signal and the voltage signals of each signal port;

[0020] The coupling path feature vector is extracted from the cross-correlation function. The coupling path feature vector includes the maximum cross-correlation coefficient, the lag time when the maximum correlation coefficient is reached, the coupling strength factor, and the main lobe width of the cross-correlation function. The coupling strength factor is obtained by weighted integration of the cross-correlation function values ​​within a preset window before and after the lag time.

[0021] Furthermore, magnetic field feature vectors are extracted from the spatial magnetic field time-domain signal through oscillation mode analysis, including:

[0022] Variational mode decomposition is performed on the time-domain signal of the spatial magnetic field, which is adaptively decomposed into finite-bandwidth eigenmode functions with different center frequencies.

[0023] Based on a preset electromagnetic interference frequency range, one or more finite bandwidth intrinsic mode functions with center frequencies located within the preset electromagnetic interference frequency range are selected from finite bandwidth intrinsic mode functions with different center frequencies as effective oscillation modes.

[0024] Calculate the modal energy, modal center frequency, modal damping ratio, and modal complexity index for each effective oscillation mode;

[0025] The vector consisting of the modal energy, modal center frequency, modal damping ratio, and modal complexity index of all effective oscillation modes is used as the magnetic field characteristic vector.

[0026] Furthermore, variational mode decomposition is performed on the time-domain signal of the spatial magnetic field, adaptively decomposing the signal into finite-bandwidth eigenmode functions with different center frequencies, including:

[0027] For time-domain signals of spatial magnetic fields, a constrained variational model is constructed with the constraint of minimizing the sum of the bandwidths of all finite-bandwidth intrinsic mode functions and the optimization objective of signal reconstruction accuracy within a preset accuracy range.

[0028] By introducing a quadratic penalty factor and Lagrange multipliers, the constrained variational model is transformed into an unconstrained variational model for solution.

[0029] The alternating direction multiplier method is used to iteratively update the time-domain representation of each finite-bandwidth eigenmode function and its corresponding center frequency, and the Lagrange multipliers are updated synchronously until the preset convergence condition is met, thus obtaining finite-bandwidth eigenmode functions with different center frequencies.

[0030] Furthermore, based on a preset electromagnetic interference frequency range, one or more finite-bandwidth eigenmode functions with center frequencies within the preset electromagnetic interference frequency range are selected from finite-bandwidth eigenmode functions with different center frequencies as effective oscillation modes, including:

[0031] Identify at least one target frequency band associated with typical electromagnetic interference events of primary and secondary integrated power distribution switches, and define the union of all target frequency bands as the preset electromagnetic interference frequency range;

[0032] Iterate through all the finite-bandwidth eigenmode functions obtained from variational mode decomposition, and calculate the center frequency and the normalized energy percentage at the center frequency for each finite-bandwidth eigenmode function.

[0033] If the center frequency of the finite bandwidth intrinsic mode function is within the preset electromagnetic interference frequency range and the normalized energy ratio of the finite bandwidth intrinsic mode function at the center frequency exceeds the preset energy threshold, then the finite bandwidth intrinsic mode function is regarded as an effective oscillation mode.

[0034] Furthermore, the modal complexity index includes at least one complexity quantification index for quantifying the irregularity or complexity of the effective oscillation mode and a spectral morphology index for describing the degree of energy concentration in the power spectrum of the effective oscillation mode.

[0035] Furthermore, based on the fusion of time-domain feature vectors, frequency-domain feature vectors, coupling feature vectors, and magnetic field feature vectors, electromagnetic interference characteristics are obtained, including:

[0036] The time-domain feature vector, the frequency-domain feature vector, the coupling feature vector, and the magnetic field feature vector are defined as four different feature subsets;

[0037] Based on predefined combination rules, at least two different candidate feature combinations are generated from four different feature subsets.

[0038] The feature subsets contained in each candidate feature combination are concatenated in a predetermined order to form the corresponding candidate feature vector;

[0039] Using a gradient boosting decision tree model, the importance of all feature dimensions in each candidate feature vector is evaluated to obtain an importance score for each feature dimension.

[0040] Based on the importance scores of each feature dimension, calculate the overall quality score of each candidate feature vector; select the candidate feature vector with the highest overall quality score as the preferred candidate feature vector.

[0041] From the preferred candidate feature vectors, feature dimensions with importance scores higher than a preset threshold are selected to form the electromagnetic interference features.

[0042] Furthermore, based on predefined combination rules, at least two different candidate feature combinations are generated from four different feature subsets, including:

[0043] Define the composition hierarchy, including first-level composition, second-level composition, and third-level composition;

[0044] The first-level combination includes all four feature subsets, the second-level combination includes all combinations formed by randomly selecting three from all four feature subsets, and the third-level combination includes all combinations formed by randomly selecting two from all four feature subsets.

[0045] From the first-level combination, the second-level combination, and the third-level combination, select at least two candidate feature combinations in total.

[0046] Furthermore, using a gradient boosting decision tree model, the importance of all feature dimensions in each candidate feature vector is evaluated to obtain an importance score for each feature dimension, including:

[0047] Construct a historical electromagnetic interference event sample set, wherein each historical electromagnetic interference event sample contains a full-dimensional feature vector composed of the time domain feature vector, the frequency domain feature vector, the coupling feature vector and the magnetic field feature vector, as well as a corresponding electromagnetic interference event type label;

[0048] Using the historical electromagnetic interference event sample set, a gradient boosting decision tree model is trained.

[0049] For each candidate feature vector, perform the following steps:

[0050] From the trained gradient boosting decision tree model, extract the global importance scores corresponding to each feature dimension contained in the current candidate feature vector, and use them as the importance scores of each feature dimension.

[0051] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0052] 1. This invention is based on the fusion of four-dimensional features: time-domain feature vector extracted from the original time-domain signal acquired by multi-port synchronous acquisition, frequency-domain feature vector extracted from Welch power spectrum estimation, coupling feature vector extracted from the cross-correlation analysis of power supply and signal port voltages, and magnetic field feature vector extracted from the oscillation mode analysis of the spatial magnetic field time-domain signal. This fusion can completely characterize the time-frequency coupling magnetic synergistic effect of electromagnetic interference at power distribution switch ports, effectively improve the accuracy of interference source location and anti-interference capability in complex electromagnetic environments, ensure the stability of power distribution system operation, and solve the problem that existing technologies cannot accurately extract the electromagnetic interference features of power distribution switch acquisition ports through primary and secondary fusion.

[0053] 2. This invention synchronously acquires raw time-domain signals through multiple ports, and after preprocessing, extracts time-domain feature vectors containing various statistical features such as mean and variance, and frequency-domain feature vectors containing various parameters such as peak frequency. Based on the voltage signals of the power and signal ports, it extracts coupling path feature vectors with indicators such as the maximum cross-correlation coefficient. It also extracts magnetic field feature vectors containing modal energy and other indicators from the spatial magnetic field time-domain signal through operations such as variational mode decomposition. This multi-dimensional and comprehensive feature extraction method can accurately characterize various properties of electromagnetic interference, providing rich and precise feature information for subsequent accurate identification of electromagnetic interference events, and greatly improving the accuracy of electromagnetic interference identification.

[0054] 3. When performing variational mode decomposition on time-domain signals of spatial magnetic fields, this invention constructs a constrained variational model with the minimum sum of bandwidths as a constraint and signal reconstruction accuracy within a preset accuracy range as the optimization objective. This model is then transformed into an unconstrained variational model by introducing a quadratic penalty factor and Lagrange multipliers. The finite-bandwidth eigenmode functions are obtained through iterative updates using the alternating direction multiplier method. Simultaneously, effective oscillation modes are adaptively selected based on a preset electromagnetic disturbance frequency range. This adaptive mode decomposition and selection method can flexibly and accurately extract effective features according to different electromagnetic environments and signal characteristics, enhancing the adaptability of the feature extraction method to various complex situations.

[0055] 4. This invention defines time-domain, frequency-domain, coupling, and magnetic field feature vectors as four feature subsets. Based on predefined combination rules, it generates multiple candidate feature combinations, concatenates them to form candidate feature vectors, and then uses a gradient boosting decision tree model to evaluate the importance of feature dimensions in each candidate feature vector. It calculates the overall quality score, selects the preferred candidate feature vectors, and chooses feature dimensions with high importance scores to form electromagnetic interference features. This scientific feature fusion and screening mechanism can select the most representative and crucial features from numerous features, remove redundant information, optimize the quality of electromagnetic interference features, and improve the efficiency and effectiveness of subsequent analysis and processing. Attached Figure Description

[0056] Figure 1 This is a schematic diagram of the basic process of an electromagnetic interference feature extraction method for a primary and secondary integrated power distribution switch acquisition port provided in an embodiment of the present invention. Detailed Implementation

[0057] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0058] Example 1

[0059] like Figure 1 As shown in the figure, this embodiment introduces a method for extracting electromagnetic interference features from the acquisition port of a primary and secondary integrated power distribution switch, including:

[0060] Step 1: Extract time-domain feature vectors from the original time-domain signals acquired synchronously from multiple ports, and extract frequency-domain feature vectors through Welch power spectrum estimation.

[0061] This invention is based on multi-port synchronous acquisition of raw time-domain signals, enabling comprehensive and synchronous acquisition of raw data from different ports of a power distribution switch. This avoids data deviations caused by asynchronous acquisition, ensuring data integrity and consistency. Time-domain feature vectors are extracted from the acquired raw time-domain signals, encompassing rich statistical features such as mean, variance, and peak value. These features reflect the signal's variation patterns and characteristics in the time domain from different perspectives, helping to provide a preliminary understanding of the manifestations of electromagnetic interference in the time domain. Simultaneously, Welch power spectrum estimation is used to extract frequency-domain feature vectors. Through Hanning window function and overlapping segmentation with a preset percentage, spectral leakage is effectively reduced, resulting in a smoother and more accurate calculated power spectral density. The extracted frequency-domain feature vectors, such as peak frequency and total power, accurately characterize the distribution and energy features of electromagnetic interference in the frequency domain, providing crucial information in both the time and frequency domains for comprehensive analysis of electromagnetic interference.

[0062] Step 2: Extract the coupling path feature vector by cross-correlation analysis based on the voltage signals of the power port and signal port.

[0063] This invention performs bandpass filtering on the voltage signals of the power supply port and the signal port respectively, which can accurately filter out the effective signals within a specific frequency band, remove noise and interference from other irrelevant frequency bands, and improve the signal quality and purity. A dynamic time warping algorithm is used to align the filtered voltage signals, effectively solving the time misalignment problem caused by factors such as signal transmission delay, ensuring the accuracy of subsequent analysis. The normalized cross-correlation function between the aligned signals is calculated, and the coupling path feature vectors extracted from this function, such as the maximum cross-correlation coefficient and the lag time at which the maximum correlation coefficient is reached, clearly reveal the coupling relationship and energy transfer path between the power supply port and the signal port. This provides important evidence for a deeper understanding of the propagation mechanism of electromagnetic interference between ports and helps to take targeted measures to suppress the propagation of electromagnetic interference.

[0064] Step 3: Extract the magnetic field feature vector based on the spatial magnetic field time domain signal through oscillation mode analysis.

[0065] This invention performs variational mode decomposition on time-domain signals of spatial magnetic fields, adaptively decomposing the signal into finite-bandwidth eigenmode functions with different center frequencies. This adaptive decomposition method can accurately segment the signal based on its inherent characteristics, avoiding mode aliasing problems that may occur in traditional decomposition methods. This allows each eigenmode function to more accurately reflect the local characteristics of the magnetic field signal. Effective oscillation modes are selected based on a preset electromagnetic interference frequency range, focusing on modes closely related to electromagnetic interference and eliminating interference from irrelevant modes, thus improving the targeting and effectiveness of feature extraction. The modal energy, modal center frequency, and other indicators of the effective oscillation modes are calculated and used to construct magnetic field feature vectors. These feature vectors comprehensively characterize the spatial magnetic field from multiple aspects, including energy distribution and frequency characteristics, providing crucial information for accurately identifying and assessing the impact of electromagnetic interference on the magnetic field.

[0066] Step 4: Perform feature fusion based on time-domain feature vectors, frequency-domain feature vectors, coupling feature vectors, and magnetic field feature vectors to obtain electromagnetic interference features.

[0067] This invention defines time-domain, frequency-domain, coupling, and magnetic field feature vectors as four distinct feature subsets. This classification method systematically organizes the characteristics of electromagnetic interference (EMI) from multiple dimensions, covering its characteristics in the time domain, frequency domain, port coupling, and spatial magnetic field, providing a rich feature foundation for comprehensive EMI analysis. Multiple candidate feature combinations are generated based on predefined combination rules, fully considering the combined effects of different features and uncovering potentially valuable feature combinations. A gradient boosting decision tree model is used to evaluate the importance of each candidate feature vector and calculate its overall quality score. Machine learning methods are used to objectively evaluate the merits of different feature combinations, selecting the candidate feature vector with the highest overall quality score and filtering out feature dimensions with high importance scores to form the EMI features. This scientific feature fusion and filtering mechanism extracts the most representative and crucial features from numerous features, removes redundant information, and greatly improves the accuracy and effectiveness of EMI features, providing high-quality feature support for subsequent EMI identification, analysis, and processing.

[0068] Example 2

[0069] This embodiment describes the implementation steps of the electromagnetic interference feature extraction method for the primary and secondary integrated power distribution switch acquisition ports, including:

[0070] Step 1: Extract time-domain feature vectors from the original time-domain signals acquired synchronously from multiple ports, and extract frequency-domain feature vectors through Welch power spectrum estimation.

[0071] Step 1.1: Preprocess the raw time-domain signal acquired by multi-port synchronous acquisition, including denoising, normalization and alignment, to obtain the preprocessed time-domain signal.

[0072] Raw time-domain signals acquired synchronously from multiple ports are often subject to various noise interferences, such as environmental noise and equipment noise. These noises can mask the true characteristics of the signal and affect the accuracy of subsequent analysis. Noise denoising effectively removes noise components from the signal, retains its main characteristics, and improves signal purity. Normalization unifies signals acquired from different ports to the same scale range, avoiding analytical biases caused by excessive differences in signal amplitude and making signals from different ports comparable. Alignment resolves time misalignment issues caused by asynchronous acquisition times or signal transmission delays, ensuring consistency of signals from each port on the time axis. After this series of preprocessing operations, the preprocessed time-domain signal is more accurate and reliable, laying a solid foundation for subsequent feature extraction and analysis, and contributing to improving the performance and accuracy of the entire electromagnetic interference analysis system.

[0073] Step 1.2: Extract time-domain feature vectors from the preprocessed time-domain signal.

[0074] In this embodiment, the time-domain feature vector includes statistical features such as mean, variance, peak value, impulse factor, waveform factor, and margin factor, and is calculated using time-domain parameters.

[0075] The preprocessed time-domain signal contains rich information. By extracting time-domain feature vectors, the signal's characteristics in the time domain can be characterized from different perspectives. The mean reflects the average level of the signal over a period of time, helping to understand the overall energy level of the signal; the variance reflects the dispersion of the signal, allowing for the assessment of signal fluctuations; and the peak value represents the maximum amplitude of the signal, intuitively reflecting extreme cases. Statistical features such as the impulse factor, waveform factor, and margin factor further describe the shape and characteristics of the signal from different aspects. For example, the impulse factor reflects the presence of impulse components in the signal, the waveform factor helps distinguish signals of different shapes, and the margin factor is related to signal wear or faults. These time-domain feature vectors complement each other, comprehensively presenting various characteristics of the signal in the time domain, providing crucial information for accurately identifying and analyzing electromagnetic interference, and contributing to a deeper understanding of the manifestations and patterns of electromagnetic interference in the time domain.

[0076] Step 1.3: Perform Welch power spectrum estimation on the preprocessed time-domain signal. Use the Hanning window function and a preset percentage of overlap segments to calculate the power spectral density of each overlap segment and average them to obtain a smooth power spectral density.

[0077] Welch power spectral density estimation is a commonly used spectral analysis method. Applying this method to preprocessed time-domain signals effectively analyzes the energy distribution of the signal in the frequency domain. Using the Hanning window function reduces spectral leakage because it has good sidelobe attenuation characteristics, concentrating the signal energy in the frequency domain and preventing energy leakage to other frequency bands, thus improving the accuracy of spectral estimation. Preset percentage overlap segmentation increases data utilization, ensuring some overlap between segments. This reduces information loss due to segmentation, resulting in a more continuous and smooth power spectral density. Averaging the power spectral density of each overlapping segment further reduces the impact of random errors. The resulting smooth power spectral density more accurately reflects the true energy distribution of the signal in the frequency domain, providing reliable data support for subsequent extraction of frequency domain feature vectors and contributing to a deeper understanding of the characteristics and patterns of electromagnetic interference in the frequency domain.

[0078] Step 1.4: Extract frequency domain feature vectors from the smoothed power spectral density.

[0079] In this embodiment, the frequency domain feature vector includes the peak frequency of the spectrum, total power, energy ratio of a specific frequency band, spectral centroid, and spectral flatness, and is calculated using frequency domain parameters.

[0080] A smooth power spectral density contains rich information about the signal in the frequency domain. Extracting frequency domain feature vectors can characterize the signal's frequency domain properties from different perspectives. The peak frequency represents the frequency point where the signal has the highest energy in the frequency domain, directly reflecting the main frequency components of electromagnetic interference (EMI) and helping to determine the source and type of EMI. Total power reflects the total energy of the signal across the entire frequency domain and is an important indicator for measuring the intensity of EMI. The specific frequency band energy ratio focuses on the energy proportion within a specific frequency band. By analyzing the energy distribution in different frequency bands, the activity level of EMI in different bands can be understood, providing a basis for targeted suppression measures. The spectral centroid describes the location of the signal energy distribution centroid in the frequency domain, reflecting the signal's spectral concentration trend. Spectral flatness reflects the uniformity of the signal spectrum; higher flatness indicates a more even distribution of signal energy across frequency bands, while lower flatness indicates that energy is concentrated in certain specific frequency bands. These frequency domain feature vectors work together to comprehensively present various characteristics of the signal in the frequency domain, providing crucial frequency domain information for accurately identifying, analyzing, and evaluating EMI, and helping to improve the accuracy and effectiveness of EMI detection and processing.

[0081] Step 2: Extract the coupling path feature vector by cross-correlation analysis based on the voltage signals of the power port and signal port.

[0082] Step 2.1: Perform bandpass filtering on the voltage signals of the power port and signal port respectively to obtain the filtered power port voltage signal and the voltage signals of each signal port.

[0083] During the acquisition process, the voltage signals of power and signal ports are often mixed with various noise and interference signals. These interferences may originate from multiple factors, including the external environment and the equipment's own operation. Bandpass filtering can accurately filter out effective signals within a specific frequency band. This is because signals in different frequency bands often carry different information, and signals related to electromagnetic interference are usually concentrated within a specific frequency range. Bandpass filters can effectively block signals outside the target frequency band, allowing only signals from the target band to pass through, thereby removing noise and interference from irrelevant frequency bands and improving signal quality and purity. The resulting filtered power port voltage signal and the voltage signals of each signal port provide reliable basic data for subsequent accurate analysis of the coupling relationship between ports, helping to more clearly reveal the propagation characteristics of electromagnetic interference between ports.

[0084] Step 2.2: The filtered power port voltage signal and the voltage signals of each signal port are sequentially aligned using a dynamic time warping algorithm to obtain the aligned power port voltage signal and the voltage signals of each signal port.

[0085] In actual signal transmission, due to factors such as different signal transmission paths and differences in device response time, the voltage signals of the power port and signal port may be misaligned on the time axis. This time misalignment can seriously affect subsequent analysis of the correlation and coupling relationships between signals, leading to inaccurate analysis results. Dynamic time warping is an effective method for handling signal alignment problems of different lengths and time axes. It uses dynamic programming to find an optimal path that achieves the best alignment of two signal sequences on the time axis. After processing by this algorithm, the aligned power port voltage signal and the voltage signals of each signal port maintain temporal consistency, more accurately reflecting their true relationship. This provides an accurate time reference for subsequent calculation of cross-correlation functions and extraction of coupling path feature vectors, improving the accuracy and reliability of the analysis.

[0086] Step 2.3: Calculate the normalized cross-correlation function between the aligned power port voltage signal and the voltage signals of each signal port.

[0087] There is a certain correlation between the aligned power port voltage signal and the voltage signals of each signal port, reflecting the coupling relationship between the ports. The normalized cross-correlation function (RCF) is a mathematical tool used to measure the similarity between two signals. It performs cross-correlation on the two signals and then normalizes the result to a certain range, typically [-1, 1]. Calculating the RCF can quantitatively describe the similarity and phase relationship between the power port and signal port voltage signals. When the RCF value is close to 1, it indicates that the two signals have strong similarity; when the value is close to -1, it indicates that the two signals have opposite trends; when the value is close to 0, it indicates that there is almost no correlation between the two signals. By calculating the RCF, the coupling characteristics of the voltage signals between ports can be accurately characterized mathematically, providing a crucial quantitative basis for subsequent extraction of coupling path feature vectors.

[0088] Step 2.4: Extract the coupling path feature vector from the cross-correlation function.

[0089] In this embodiment, the coupling path feature vector includes the maximum cross-correlation coefficient, the lag time when the maximum correlation coefficient is reached, the coupling strength factor, and the main lobe width of the cross-correlation function. The coupling strength factor is obtained by weighted integration of the cross-correlation function values ​​within a preset window before and after the lag time.

[0090] The coupling path feature vectors extracted from the normalized cross-correlation function can comprehensively describe the coupling relationship between power supply ports and signal ports from multiple perspectives. The maximum cross-correlation coefficient directly reflects the maximum similarity between the voltage signals of the two ports and is an important indicator of coupling strength; a larger value indicates a stronger coupling relationship. The lag time at which the maximum correlation coefficient is reached represents the time delay of signal transmission between ports. This parameter provides information such as the signal propagation path and speed. The coupling strength factor is obtained by weighted integration of the cross-correlation function values ​​within a preset window before and after the lag time. This calculation method comprehensively considers the changes in the cross-correlation function within a certain time range, enabling a more accurate assessment of the coupling strength and stability. The main lobe width of the cross-correlation function reflects the frequency range characteristics of signal coupling; a wider main lobe indicates a wider frequency range involved in the coupling. These coupling path feature vectors complement each other, forming a complete feature system describing the coupling relationship between ports. This provides an important basis for in-depth analysis of the propagation mechanism of electromagnetic interference between ports, assessment of the impact of electromagnetic interference, and implementation of targeted suppression measures.

[0091] Step 3: Extract the magnetic field feature vector based on the spatial magnetic field time domain signal through oscillation mode analysis.

[0092] Step 3.1: Perform variational mode decomposition on the time-domain signal of the spatial magnetic field, and adaptively decompose the time-domain signal of the spatial magnetic field into finite-bandwidth eigenmode functions with different center frequencies.

[0093] In the operating environment of integrated primary and secondary distribution switches, the time-domain signal of the spatial magnetic field contains rich and complex information, which may be composed of the superposition of magnetic field oscillation components with different frequencies and characteristics. Variational Mode Decomposition (VMD), as an advanced signal processing method, has advantages such as strong adaptability and effective handling of non-stationary signals. It can adaptively decompose the time-domain signal of the spatial magnetic field into multiple finite-bandwidth intrinsic mode functions (IMFs) with different center frequencies based on the characteristics of the signal itself. Each IMF represents an oscillation mode within a specific frequency range in the signal. Through this decomposition method, complex magnetic field signals can be decomposed into a series of relatively simple sub-signals, facilitating subsequent in-depth analysis and processing of the magnetic field characteristics of different frequency components, and laying the foundation for accurate extraction of magnetic field features.

[0094] Step 3.1.1: For the time-domain signal of the spatial magnetic field, construct a constrained variational model with the constraint of minimizing the sum of the bandwidths of all finite-bandwidth intrinsic mode functions and the optimization objective of the signal reconstruction accuracy within a preset accuracy range.

[0095] Constructing a constrained variational model is one of the core steps of the VMD algorithm. Minimizing the sum of the bandwidths of all IMFs is used as a constraint to ensure that the decomposed IMFs are as compact as possible, avoiding over-decomposition or unnecessary frequency components, and ensuring that each IMF accurately reflects the oscillation characteristics of a specific frequency range in the signal. Simultaneously, optimizing the signal reconstruction accuracy within a preset accuracy range ensures that the recombined IMFs can accurately reconstruct the original spatial magnetic field time-domain signal. Through this constraint and optimization objective, the decomposition process effectively separates different frequency components while maintaining the integrity of signal information, thus improving the accuracy and reliability of the decomposition.

[0096] Step 3.1.2: By introducing a quadratic penalty factor and Lagrange multipliers, the constrained variational model is transformed into an unconstrained variational model for solution.

[0097] Because constrained variational models are subject to constraints during the solution process, direct solutions are difficult. Introducing a quadratic penalty factor and Lagrange multipliers is a common method to transform constrained problems into unconstrained ones. The quadratic penalty factor penalizes signal reconstruction errors, causing the algorithm to tend to minimize these errors during the solution process to meet the accuracy requirements of signal reconstruction. Lagrange multipliers handle the constraint of minimizing the sum of bandwidths, integrating the constraints into the objective function, thus transforming the constrained variational model into an unconstrained variational model. This transformation simplifies the solution process, enabling efficient solutions using existing optimization algorithms such as the alternating direction multiplier method, and providing a feasible computational approach for obtaining accurate finite-bandwidth eigenmode functions.

[0098] Step 3.1.3: Iteratively update the time-domain representation of each finite-bandwidth eigenmode function and its corresponding center frequency using the alternating direction multiplier method, and simultaneously update the Lagrange multipliers until the preset convergence condition is met, thus obtaining finite-bandwidth eigenmode functions with different center frequencies.

[0099] The Alternating Direction Multiplier Method (ADMM) is an efficient algorithm suitable for solving large-scale optimization problems. When solving unconstrained variational models, ADMM updates the values ​​of each variable through alternating iterations. Specifically, in the VMD algorithm, ADMM sequentially updates the time-domain representation, corresponding center frequency, and Lagrange multipliers of each finite-bandwidth eigenmode function. In each iteration, new update values ​​are calculated based on the current variable values, causing the objective function to gradually decrease until a preset convergence condition is met. Through this iterative update process, a set of stable finite-bandwidth eigenmode functions with different center frequencies is ultimately obtained. These functions can accurately describe the oscillation characteristics of different frequency components in the time-domain signal of a spatial magnetic field.

[0100] Step 3.2: Based on the preset electromagnetic interference frequency range, select one or more finite bandwidth intrinsic mode functions with center frequencies located within the preset electromagnetic interference frequency range from finite bandwidth intrinsic mode functions with different center frequencies as effective oscillation modes.

[0101] Step 3.2.1: Determine at least one target frequency band associated with typical electromagnetic interference events of primary and secondary integrated power distribution switches, and define the union of all target frequency bands as the preset electromagnetic interference frequency range.

[0102] During operation, integrated primary and secondary power distribution switches generate electromagnetic interference (EMI) signals within a specific frequency range, which may interfere with surrounding equipment and systems. To accurately extract the magnetic field characteristics associated with EMI, it is necessary to first determine the target frequency bands associated with typical EMI events. These target frequency bands can be determined through analysis of actual operating conditions, experimental measurements, and relevant standards and specifications. By taking the union of all target frequency bands, a preset EMI frequency range is obtained. This range covers all EMI frequency components that may affect the system, providing a clear frequency range basis for selecting effective oscillation modes.

[0103] Step 3.2.2: Traverse all finite-bandwidth eigenmode functions obtained from variational mode decomposition, and calculate the center frequency and normalized energy percentage at the center frequency for each finite-bandwidth eigenmode function.

[0104] After obtaining all finite-bandwidth intrinsic mode functions (IMFs) from the variational mode decomposition, each needs to be analyzed individually. The center frequency of each IMF is calculated, reflecting the dominant frequency component of the oscillation mode it represents. Simultaneously, the normalized energy percentage at the center frequency of each IMF is calculated. This normalized energy percentage represents the proportion of that frequency component in the total signal energy, reflecting its energy intensity. By calculating these two parameters, a comprehensive understanding of the frequency characteristics and energy distribution of each IMF can be obtained, providing quantitative indicators for subsequent screening of effective oscillation modes.

[0105] Step 3.2.3: If the center frequency of the finite bandwidth intrinsic mode function is within the preset electromagnetic interference frequency range and the normalized energy ratio of the finite bandwidth intrinsic mode function at the center frequency exceeds the preset energy threshold, then the finite bandwidth intrinsic mode function is regarded as an effective oscillation mode.

[0106] Screening for effective oscillation modes requires considering both the center frequency and the normalized energy percentage. Only when the center frequency of an intrinsic mode function (IMF) falls within a preset electromagnetic interference (EMI) frequency range is the IMF considered potentially related to EMI. Furthermore, to ensure the IMF possesses sufficient energy to have a significant impact in practice, its normalized energy percentage at the center frequency must exceed a preset energy threshold. Only finite-bandwidth IMFs that simultaneously meet both conditions are considered effective oscillation modes. These effective oscillation modes contain magnetic field information closely related to EMI and are the focus of subsequent magnetic field characteristic analysis.

[0107] Step 3.3: Calculate the modal energy, modal center frequency, modal damping ratio, and modal complexity index for each effective oscillation mode.

[0108] In this embodiment, the modal complexity index includes at least a complexity quantification index for quantifying the irregularity or complexity of the effective oscillation mode and a spectral morphology index for describing the degree of energy concentration in the power spectrum of the effective oscillation mode.

[0109] For each selected effective oscillation mode, multiple characteristic indices need to be calculated to comprehensively describe its properties. Modal energy reflects the amount of energy carried by the oscillation mode and is an important indicator of its impact on the surrounding environment. The modal center frequency further clarifies the main frequency components of the oscillation mode. The modal damping ratio describes the rate attenuation of energy during oscillation; a larger damping ratio indicates faster oscillation decay and a shorter impact time on the system. Modal complexity indices characterize the complexity of the oscillation mode from different perspectives. Complexity quantification indices can quantify the irregularity of the oscillation mode, such as signal fluctuations and abrupt changes. Spectral morphology indices describe the concentration of energy in the power spectrum; the more concentrated the energy, the more singular the frequency components of the oscillation mode, and vice versa. These indices complement each other and together constitute a complete index system describing the characteristics of effective oscillation modes.

[0110] Step 3.4: The vector consisting of the modal energy, modal center frequency, modal damping ratio, and modal complexity index of all effective oscillation modes is used as the magnetic field characteristic vector.

[0111] The characteristic indices of all effective oscillation modes are combined into a single vector, which is the magnetic field characteristic vector. This vector integrates information from different effective oscillation modes in terms of energy, frequency, damping, and complexity, providing a comprehensive and accurate description of the electromagnetic interference-related characteristics of the spatial magnetic field time-domain signal. This magnetic field characteristic vector serves as a crucial basis for subsequent analysis, identification, and assessment of electromagnetic interference, such as determining its intensity, type, and source, thus providing strong technical support for ensuring the normal operation of integrated primary and secondary power distribution switches and the safe and stable operation of surrounding equipment.

[0112] Step 4: Perform feature fusion based on time-domain feature vectors, frequency-domain feature vectors, coupling feature vectors, and magnetic field feature vectors to obtain electromagnetic interference features.

[0113] Step 4.1: Define the time-domain feature vector, the frequency-domain feature vector, the coupling feature vector, and the magnetic field feature vector as four different feature subsets.

[0114] By clearly dividing the feature into four independent feature subsets, a basic framework for multi-dimensional feature analysis is constructed, ensuring the independent expression of time domain, frequency domain, coupling path and magnetic field features, avoiding information loss caused by feature aliasing, providing structured data support for subsequent feature combination and screening, and improving the systematicness and scalability of feature processing.

[0115] Step 4.2: Generate at least two different candidate feature combinations based on four different feature subsets according to predefined combination rules.

[0116] Step 4.2.1: Define the combination hierarchy, including first-level combination, second-level combination, and third-level combination;

[0117] In this embodiment, the first-level combination includes all four feature subsets, the second-level combination includes all combinations formed by randomly selecting three from all four feature subsets, and the third-level combination includes all combinations formed by randomly selecting two from all four feature subsets.

[0118] Step 4.2.2: Select at least two candidate feature combinations from the first-level combination, the second-level combination, and the third-level combination.

[0119] By defining a combination hierarchy of full feature subsets, randomly selected three feature subsets, and randomly selected two feature subsets, and selecting no fewer than two candidate combinations, hierarchical and diversified coverage of feature combinations is achieved. This not only ensures the integrity of full-dimensional features, but also explores the synergistic effect between features through feature subset combinations, providing a rich candidate space for subsequent selection of the optimal feature combination.

[0120] Step 4.3: Concatenate the feature subsets contained in each candidate feature combination in a predetermined order to form the corresponding candidate feature vector.

[0121] By concatenating candidate feature vectors in a predetermined order, features of different dimensions are unified into a processable vector form. This not only preserves the original information of each subset of features but also achieves spatial integration of features, enabling subsequent machine learning models to process multi-dimensional features simultaneously and improving the efficiency and consistency of feature analysis.

[0122] Step 4.4: Using the gradient boosting decision tree model, evaluate the importance of all feature dimensions in each candidate feature vector to obtain the importance score of each feature dimension.

[0123] Step 4.4.1: Construct a historical electromagnetic interference event sample set, wherein each historical electromagnetic interference event sample contains a full-dimensional feature vector composed of the time-domain feature vector, the frequency-domain feature vector, the coupling feature vector and the magnetic field feature vector, as well as the corresponding electromagnetic interference event type label.

[0124] Step 4.4.2: Use the historical electromagnetic interference event sample set to train a gradient boosting decision tree model.

[0125] Step 4.4.3: For each candidate feature vector, perform the following steps:

[0126] From the trained gradient boosting decision tree model, extract the global importance scores corresponding to each feature dimension contained in the current candidate feature vector, and use them as the importance scores of each feature dimension.

[0127] By constructing a historical electromagnetic interference event sample set and training a gradient boosting decision tree model, an objective quantitative evaluation of feature dimensions was achieved. This model can automatically learn the correlation between features and electromagnetic interference event types, extract the global importance score of each feature dimension, provide a data-driven decision-making basis for feature selection, and enhance the scientific rigor and reliability of feature evaluation.

[0128] Step 4.5: Calculate the overall quality score of each candidate feature vector based on the importance score of each feature dimension; select the candidate feature vector with the highest overall quality score as the preferred candidate feature vector.

[0129] By calculating an overall quality score by combining the importance scores of each feature dimension, quantitative optimization of candidate feature vectors is achieved. This process considers both the contribution of individual features and the overall effectiveness of feature combinations, ensuring that the selected candidate feature vectors are optimal in terms of feature representativeness and information richness, thus improving the effect of feature fusion.

[0130] Step 4.6: Select feature dimensions with importance scores higher than a preset threshold from the preferred candidate feature vectors to form the electromagnetic interference features.

[0131] By removing redundant features through threshold filtering, the most representative key feature dimensions are retained. This reduces the number of feature dimensions while ensuring the efficiency and effectiveness of the final electromagnetic interference (EMI) features. This step achieves feature deredundancy and refinement, improving the efficiency and accuracy of subsequent analysis and providing high-quality feature input for the accurate identification and assessment of EMI.

[0132] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. These 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 function specified in one or more boxes.

[0133] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for extracting electromagnetic interference features from the acquisition port of a primary and secondary integrated power distribution switch, characterized in that, include: The time-domain feature vector is extracted from the original time-domain signal acquired synchronously from multiple ports, and the frequency-domain feature vector is extracted by Welch power spectrum estimation. The coupling path feature vector is extracted by cross-correlation analysis based on the voltage signals of the power port and the signal port. Magnetic field feature vectors are extracted from the time-domain signal of the spatial magnetic field through oscillation mode analysis. Electromagnetic disturbance characteristics are obtained by fusing features from time-domain feature vectors, frequency-domain feature vectors, coupling feature vectors, and magnetic field feature vectors.

2. The method for extracting electromagnetic interference features from the acquisition port of a primary and secondary integrated power distribution switch as described in claim 1, characterized in that, Time-domain feature vectors are extracted from the raw time-domain signals acquired synchronously from multiple ports, and frequency-domain feature vectors are extracted through Welch power spectrum estimation, including: The raw time-domain signal acquired synchronously from multiple ports is preprocessed, including denoising, normalization and alignment, to obtain the preprocessed time-domain signal; The time-domain feature vector is extracted from the preprocessed time-domain signal. The time-domain feature vector includes statistical features such as mean, variance, peak value, impulse factor, waveform factor and margin factor, and is calculated using time-domain parameters. Welch power spectrum estimation is performed on the preprocessed time-domain signal. The Hanning window function and the overlap segment with a preset percentage are used to calculate the power spectral density of each overlap segment and average them to obtain a smooth power spectral density. Frequency domain feature vectors are extracted from the smooth power spectral density. These feature vectors include the peak frequency, total power, energy ratio of a specific frequency band, spectral centroid, and spectral flatness, and are calculated using frequency domain parameters.

3. The method for extracting electromagnetic interference features from the acquisition port of a primary and secondary integrated power distribution switch as described in claim 2, characterized in that, Based on the voltage signals at the power port and signal port, the coupling path feature vector is extracted through cross-correlation analysis, including: The voltage signals of the power supply port and the signal port are bandpass filtered respectively to obtain the filtered voltage signals of the power supply port and the voltage signals of each signal port. The filtered power port voltage signal and the voltage signals of each signal port are sequentially aligned using a dynamic time warping algorithm to obtain the aligned power port voltage signal and the voltage signals of each signal port. Calculate the normalized cross-correlation function between the aligned power port voltage signal and the voltage signals of each signal port; The coupling path feature vector is extracted from the cross-correlation function. The coupling path feature vector includes the maximum cross-correlation coefficient, the lag time when the maximum correlation coefficient is reached, the coupling strength factor, and the main lobe width of the cross-correlation function. The coupling strength factor is obtained by weighted integration of the cross-correlation function values ​​within a preset window before and after the lag time.

4. The method for extracting electromagnetic interference features from the acquisition port of a primary and secondary integrated power distribution switch as described in claim 3, characterized in that, Based on the time-domain signal of the spatial magnetic field, magnetic field feature vectors are extracted through oscillation mode analysis, including: Variational mode decomposition is performed on the time-domain signal of the spatial magnetic field, which is adaptively decomposed into finite-bandwidth eigenmode functions with different center frequencies. Based on a preset electromagnetic interference frequency range, one or more finite bandwidth intrinsic mode functions with center frequencies located within the preset electromagnetic interference frequency range are selected from finite bandwidth intrinsic mode functions with different center frequencies as effective oscillation modes. Calculate the modal energy, modal center frequency, modal damping ratio, and modal complexity index for each effective oscillation mode; The vector consisting of the modal energy, modal center frequency, modal damping ratio, and modal complexity index of all effective oscillation modes is used as the magnetic field characteristic vector.

5. The method for extracting electromagnetic interference features from the acquisition port of a primary and secondary integrated power distribution switch as described in claim 4, characterized in that, Variational mode decomposition is performed on the time-domain signal of the spatial magnetic field, adaptively decomposing the signal into finite-bandwidth eigenmode functions with different center frequencies, including: For time-domain signals of spatial magnetic fields, a constrained variational model is constructed with the constraint of minimizing the sum of the bandwidths of all finite-bandwidth intrinsic mode functions and the optimization objective of signal reconstruction accuracy within a preset accuracy range. By introducing a quadratic penalty factor and Lagrange multipliers, the constrained variational model is transformed into an unconstrained variational model for solution. The alternating direction multiplier method is used to iteratively update the time-domain representation of each finite-bandwidth eigenmode function and its corresponding center frequency, and the Lagrange multipliers are updated synchronously until the preset convergence condition is met, thus obtaining finite-bandwidth eigenmode functions with different center frequencies.

6. The method for extracting electromagnetic interference features from the acquisition port of a primary and secondary integrated power distribution switch as described in claim 5, characterized in that, Based on a preset electromagnetic interference frequency range, one or more finite-bandwidth eigenmode functions with center frequencies within the preset electromagnetic interference frequency range are selected from finite-bandwidth eigenmode functions with different center frequencies as effective oscillation modes, including: Identify at least one target frequency band associated with typical electromagnetic interference events of primary and secondary integrated power distribution switches, and define the union of all target frequency bands as the preset electromagnetic interference frequency range; Iterate through all the finite-bandwidth eigenmode functions obtained from variational mode decomposition, and calculate the center frequency and normalized energy percentage at the center frequency for each finite-bandwidth eigenmode function. If the center frequency of the finite bandwidth intrinsic mode function is within the preset electromagnetic interference frequency range and the normalized energy ratio of the finite bandwidth intrinsic mode function at the center frequency exceeds the preset energy threshold, then the finite bandwidth intrinsic mode function is regarded as an effective oscillation mode.

7. The method for extracting electromagnetic interference features from the acquisition port of a primary and secondary integrated power distribution switch as described in claim 6, characterized in that, The modal complexity index includes at least one complexity quantification index for quantifying the irregularity or complexity of the effective oscillation mode and a spectral morphology index for describing the degree of energy concentration in the power spectrum of the effective oscillation mode.

8. The method for extracting electromagnetic interference features from the acquisition port of a primary and secondary integrated power distribution switch as described in claim 7, characterized in that, Electromagnetic interference characteristics are obtained by fusing features from time-domain feature vectors, frequency-domain feature vectors, coupling feature vectors, and magnetic field feature vectors, including: The time-domain feature vector, the frequency-domain feature vector, the coupling feature vector, and the magnetic field feature vector are defined as four different feature subsets; Based on predefined combination rules, at least two different candidate feature combinations are generated from four different feature subsets. The feature subsets contained in each candidate feature combination are concatenated in a predetermined order to form the corresponding candidate feature vector; Using a gradient boosting decision tree model, the importance of all feature dimensions in each candidate feature vector is evaluated to obtain an importance score for each feature dimension. Based on the importance scores of each feature dimension, calculate the overall quality score of each candidate feature vector; select the candidate feature vector with the highest overall quality score as the preferred candidate feature vector. From the preferred candidate feature vectors, feature dimensions with importance scores higher than a preset threshold are selected to form the electromagnetic interference features.

9. The method for extracting electromagnetic interference features from the acquisition port of a primary and secondary integrated power distribution switch as described in claim 8, characterized in that, Based on predefined combination rules, at least two different candidate feature combinations are generated from four different feature subsets, including: Define the composition hierarchy, including first-level composition, second-level composition, and third-level composition; The first-level combination includes all four feature subsets, the second-level combination includes all combinations formed by randomly selecting three from all four feature subsets, and the third-level combination includes all combinations formed by randomly selecting two from all four feature subsets. From the first-level combination, the second-level combination, and the third-level combination, select at least two candidate feature combinations in total.

10. The method for extracting electromagnetic interference features from the acquisition port of a primary and secondary integrated power distribution switch as described in claim 9, characterized in that, Using a gradient boosting decision tree model, the importance of all feature dimensions in each candidate feature vector is evaluated to obtain an importance score for each feature dimension, including: Construct a historical electromagnetic interference event sample set, wherein each historical electromagnetic interference event sample contains a full-dimensional feature vector composed of the time domain feature vector, the frequency domain feature vector, the coupling feature vector and the magnetic field feature vector, as well as a corresponding electromagnetic interference event type label; Using the historical electromagnetic interference event sample set, a gradient boosting decision tree model is trained. For each candidate feature vector, perform the following steps: From the trained gradient boosting decision tree model, extract the global importance scores corresponding to each feature dimension contained in the current candidate feature vector, and use them as the importance scores of each feature dimension.

Images