Excitation load grounding fault detection method, system and device based on EEMD and Hilbert spectrum analysis and medium

Abstract

The invention discloses an excitation load grounding fault detection method, system and device based on EEMD and Hilbert spectrum analysis and a medium, and belongs to the technical field of power system energy storage equipment, and the method comprises the steps: collecting and preprocessing an electric signal, and constructing a discretization time domain data sequence; performing iterative decomposition on the sequence to obtain multiple groups of components; time domain structure analysis is carried out, and target components containing fault mutation features are screened; frequency domain transformation is carried out on the target component to construct spectrum distribution, and normalization processing is carried out; and then time-frequency joint transformation is carried out, energy distribution and energy entropy indexes are calculated, and diagnosis is completed. According to the method, modal aliasing is suppressed by using EEMD adaptive noise injection and a set average strategy, accurate separation of high-frequency transient disturbance and low-frequency harmonic waves is realized, multi-dimensional feature combined diagnosis is formed by combining frequency domain screening and time-frequency analysis, and the fault detection precision and reliability are improved.

Description

Technical Field

[0001] This invention relates to the field of power system energy storage equipment technology, specifically to a method, system, equipment, and medium for detecting excitation load grounding faults based on EEMD and Hilbert spectrum analysis. Background Technology

[0002] In the field of short-circuit fault detection in excitation power generation systems, existing technologies are mainly based on steady-state signal processing and electromechanical protection mechanisms. The existing core technology framework comprises three layers:

[0003] First, electromechanical protection devices, with overcurrent protection of the excitation winding and differential protection as their core, use fixed threshold settings to trigger circuit breaker operation for the former and current transformers to collect the current difference between the excitation winding and the detection coil for fault identification for the latter. However, they are prone to false tripping under dynamic load fluctuations and inrush current impacts. Second, frequency domain detection technology, with excitation current harmonic analysis and shaft voltage spectrum analysis as its core, uses Fourier transform to extract power frequency harmonic characteristics, but is limited by harmonic superposition interference. Third, time-frequency analysis methods aimed at excitation current transient decomposition, such as empirical mode decomposition, are used to separate high-frequency transient components in grounding faults. However, mode aliasing causes high-frequency transients to overlap with the power frequency fundamental frequency, and it cannot analyze the coupling characteristics of subsynchronous oscillation and high-frequency switching noise.

[0004] In addition, a manual inspection system based on physical rules remains an important supplement, including visually inspecting for anomalies such as wear on the excitation carbon brushes and peeling of the slip ring oxide layer, as well as manually testing the insulation resistance of the rectifier cabinet. These methods are reliable under no-load or steady-state conditions, but their core drawback lies in:

[0005] 1. Mismatch between static excitation current threshold and dynamic operating conditions (threshold sensitivity during load changes);

[0006] 2. The single-frequency domain characteristics are susceptible to interference from high-frequency switching noise in the excitation system;

[0007] 3. The lack of multi-domain collaborative analysis capability for non-stationary characteristics of excitation current (such as subsynchronous oscillation amplitude modulation signal) leads to a relatively high rate of missed detection of complex faults. Summary of the Invention

[0008] In view of the above-mentioned problems, the present invention is proposed.

[0009] Therefore, the technical problem solved by this invention is: how to achieve dynamic power coordination between thermal power units and distributed energy storage, improve frequency regulation accuracy and system redundancy and fault tolerance through a segmented high-voltage converter architecture and an adaptive droop control strategy, while suppressing high-voltage DC bus voltage fluctuations.

[0010] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a method for detecting grounding faults in excitation loads based on EEMD and Hilbert spectral analysis, comprising,

[0011] The electrical signals acquired during the operation of the acquisition equipment are preprocessed to construct a discrete time-domain data sequence.

[0012] Perform iterative decomposition on the data sequence to obtain a sequence of several component quantities;

[0013] Temporal structure analysis was performed on the sequence of each component to identify the target component sequence containing fault mutation features;

[0014] The target component sequence is subjected to frequency domain transformation processing, a spectral distribution is constructed through frequency mapping operation, and the amplitude of each frequency component in the spectrum is normalized and quantized.

[0015] Perform time-frequency joint transformation on the target component sequence to generate the time-frequency matrix of the component sequence, and calculate the energy distribution data and energy entropy index based on the matrix;

[0016] Based on the component sequences, spectral distribution data, time-frequency matrix, energy distribution data, and energy entropy index generated by iterative decomposition, fault diagnosis is performed, and the diagnostic results, including fault type, fault channel, and fault initiation time, are output.

[0017] As a preferred embodiment of the excitation load grounding fault detection method based on EEMD and Hilbert spectral analysis described in this invention, the method involves: acquiring electrical signals during the operation of the acquisition device, preprocessing the acquired signals to construct a discretized time-domain data sequence, including...

[0018] Acquire electrical signal data reflecting the operating status of the excitation system;

[0019] Perform preprocessing operations on the acquired electrical signal data;

[0020] Data segments with time-series indexes are generated by sampling at fixed time intervals.

[0021] Perform time base alignment on the sampled data and construct a multi-channel discrete dataset with a unified time identifier.

[0022] As a preferred embodiment of the excitation load grounding fault detection method based on EEMD and Hilbert spectral analysis described in this invention, wherein: the iterative decomposition process performed on the data sequence yields a series of component quantities including,

[0023] A perturbation signal is introduced into the constructed discrete data sequence and several repeated decomposition operations are performed.

[0024] In each round of decomposition, the original data sequence is split into several components with different frequency characteristics;

[0025] The structural consistency of the decomposition results in each round is judged, and the components with the same index are integrated and averaged.

[0026] After completing a preset number of iterations, a set of components with a frequency hierarchy is output.

[0027] By performing steps such as introducing perturbations, multi-round decomposition, structural consistency judgment, and integrated averaging in the decomposition process, the mode mixing between different frequency components can be effectively suppressed, allowing characteristic components such as transient impacts, high-frequency noise, and low-frequency steady state to be presented independently in different components. At the same time, multi-round iteration and averaging significantly reduce the impact of random perturbations on the single decomposition result, ensuring the stability and comparability of components at the frequency level.

[0028] As a preferred embodiment of the excitation load grounding fault detection method based on EEMD and Hilbert spectral analysis described in this invention, wherein: the step of performing time-domain structure analysis on the sequence of each component to identify the target component sequence containing fault abrupt change characteristics includes,

[0029] Construct a corresponding time variation curve for each component sequence;

[0030] Compare the fluctuation trend, amplitude fluctuation and rate of change of each component within the target time interval;

[0031] Component sequences with mutation characteristics are selected based on preset judgment rules;

[0032] Components that meet the mutation characteristics are marked as target objects for subsequent frequency domain and time-frequency domain processing.

[0033] As a preferred embodiment of the excitation load grounding fault detection method based on EEMD and Hilbert spectral analysis described in this invention, the step of performing frequency domain transformation processing on the target component sequence, constructing a spectral distribution through frequency mapping operations, and normalizing and quantizing the amplitude of each frequency component in the spectrum includes,

[0034] The target component sequence is normalized for signal length, and a frequency index relationship corresponding to the time series is established.

[0035] Perform frequency domain transformation based on discrete signal structure to extract complex coefficients corresponding to each frequency point;

[0036] The amplitude information at each frequency point is used to calculate the magnitude and generate a one-to-one mapping table between frequency and amplitude.

[0037] Perform a proportional conversion on the frequency amplitude results to construct a normalized amplitude sequence within a unified numerical range;

[0038] The normalized amplitude sequence is used as a frequency distribution feature vector for subsequent fault type identification processing.

[0039] By performing a time-frequency joint transformation on the target component sequence, the frequency and amplitude characteristics of the signal changing over time can be obtained simultaneously, and frequency energy distribution data and energy entropy index can be extracted on this basis. This process can accurately reflect the energy accumulation or diffusion state of a specific frequency band before and after a fault occurs, providing a basis for determining the time and location of the fault and analyzing the energy change trend of different frequency bands. It has obvious adaptability and stability, especially in distinguishing between subsynchronous resonance frequency band and power frequency component changes.

[0040] As a preferred embodiment of the excitation load grounding fault detection method based on EEMD and Hilbert spectral analysis described in this invention, the step of performing time-frequency joint transformation processing on the target component sequence to generate a time-frequency matrix of the component sequence, and calculating energy distribution data and energy entropy index based on the matrix includes...

[0041] A synchronous analytical operation is performed on the target component sequence to perform time-frequency joint transformation, and a two-dimensional data array describing the local variation characteristics of the signal is constructed.

[0042] In the data array, the amplitude distribution information corresponding to each time segment is extracted according to the frequency direction, and the cumulative energy value of each frequency point is calculated.

[0043] The cumulative energy value is expressed in a normalized distribution in the frequency domain to generate an energy distribution vector.

[0044] Based on the proportion of each frequency component in the energy distribution vector, a statistical index value representing the degree of energy concentration is calculated as the energy entropy result.

[0045] The time-frequency matrix, energy distribution vector, and energy entropy result are used as the input data set for subsequent judgment and processing.

[0046] As a preferred embodiment of the excitation load grounding fault detection method based on EEMD and Hilbert spectral analysis described in this invention, the method involves performing fault judgment processing based on the component sequences, spectral distribution data, time-frequency matrix, energy distribution data, and energy entropy index generated by iterative decomposition, and outputting diagnostic results including fault type, fault channel, and fault initiation time.

[0047] Analyze the changing trends and persistence characteristics of key waveforms in the component sequence to determine their corresponding operating state categories;

[0048] Comparing the amplitude differences of each characteristic frequency in the spectral distribution data is used to distinguish different types of electrical disturbance modes;

[0049] The specific time of the fault occurrence is determined based on the starting time node of energy accumulation in the time-frequency matrix;

[0050] Combining the frequency concentration of energy distribution with the trend of energy entropy changes helps to identify the duration and local impact characteristics of abnormal signals;

[0051] By logically integrating various judgment criteria, a comprehensive diagnostic result is generated, which includes the nature of the fault, the corresponding phase, and the time of occurrence.

[0052] This invention provides a grounding fault detection system for excitation load based on EEMD and Hilbert spectrum analysis.

[0053] To solve the above-mentioned technical problems, the present invention provides the following technical solution: an excitation load grounding fault detection system based on EEMD and Hilbert spectral analysis, comprising: a data collection module, a data decomposition module, a data identification module, a data mapping module, a data calculation module, and a result output module.

[0054] The data collection module collects electrical signals during the operation of the equipment, preprocesses the collected signals, and constructs a discrete time-domain data sequence.

[0055] The data decomposition module performs iterative decomposition processing on the data sequence to obtain several component quantity sequences;

[0056] The data identification module performs time-domain structural analysis on the component sequence to identify the target component sequence containing fault mutation features.

[0057] The data mapping module performs frequency domain transformation on the target component sequence, constructs a spectral distribution through frequency mapping operations, and normalizes and quantizes the amplitude of each frequency component in the spectrum.

[0058] The data calculation module performs time-frequency joint transformation processing on the target component sequence to generate the time-frequency matrix of the component sequence, and calculates the energy distribution data and energy entropy index based on the matrix.

[0059] The result output module performs fault judgment processing based on the component sequence, spectral distribution data, time-frequency matrix, energy distribution data, and energy entropy index generated by iterative decomposition processing, and outputs the diagnostic results of fault type, fault channel, and fault start time.

[0060] The present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, characterized in that the processor executes the computer program to implement the steps of the excitation load grounding fault detection method based on EEMD and Hilbert spectrum analysis.

[0061] The present invention provides a computer-readable storage medium having a computer program stored thereon, characterized in that, when the computer program is executed by a processor, it implements the steps of the excitation load grounding fault detection method based on EEMD and Hilbert spectrum analysis.

[0062] The beneficial effects of this invention are as follows: This diagnostic method effectively suppresses the mode aliasing problem of traditional EMD through the adaptive noise injection and ensemble averaging strategy of EEMD, enabling high-frequency transient disturbances (such as arc faults) and low-frequency harmonics to be accurately separated into different IMF components. Secondly, by integrating FFT frequency domain screening and HHT time-frequency analysis, multi-dimensional feature joint diagnosis is achieved. For example, multi-criterion cross-validation of IMF high-frequency energy mutations (time domain), FFT dominant frequency energy ratio (frequency domain), and HHT energy entropy (time-frequency domain) significantly improves detection accuracy. Overall, this combined technology, through the coordinated process of signal decomposition, frequency domain localization, and time-frequency verification, achieves highly robust diagnosis of non-stationary and nonlinear power grid faults, providing a solution with both theoretical innovation and engineering practicality for the safe operation of power systems. Attached Figure Description

[0063] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0064] Figure 1 The present invention provides an overall flowchart of a grounding fault detection method for excitation load based on EEMD and Hilbert spectrum analysis, which is an embodiment of the present invention.

[0065] Figure 2 The flowchart shows the EEMD algorithm decomposition of an excitation load grounding fault detection method based on EEMD and Hilbert spectrum analysis, as provided in one embodiment of the present invention.

[0066] Figure 3The following is a time-domain waveform diagram of the IMF component of an excitation load grounding fault detection system based on EEMD and Hilbert spectrum analysis, provided as an embodiment of the present invention. In the diagram, a is the time-domain waveform diagram of the component with IMD=4, b is the time-domain waveform diagram of the component with IMD=5, c is the time-domain waveform diagram of the component with IMD=6, and d is the time-domain waveform diagram of the component with IMD=7.

[0067] Figure 4 A time-domain waveform of the three-phase voltage IMF5 component of an excitation load grounding fault detection method based on EEMD and Hilbert spectrum analysis, provided as an embodiment of the present invention.

[0068] Figure 5 The image shows the FFT spectrum analysis of a fault signal in an excitation load grounding fault detection method based on EEMD and Hilbert spectrum analysis, provided as an embodiment of the present invention.

[0069] Figure 6 The fault signal HHT time-frequency energy probability distribution diagram is provided for an excitation load grounding fault detection method based on EEMD and Hilbert spectrum analysis in one embodiment of the present invention.

[0070] Figure 7 This invention provides an embodiment of an IMF (Intensive Function Fault) condition under phase A fault in an excitation load grounding fault detection method based on EEMD and Hilbert spectrum analysis.

[0071] Figure 8 This invention provides an embodiment of an IMF (Intensive Function Fault) detection method for excitation load grounding faults based on EEMD and Hilbert spectrum analysis, specifically for a phase B fault.

[0072] Figure 9 This invention provides an embodiment of an IMF (Intensive Function Fault) detection method for excitation load grounding faults based on EEMD and Hilbert spectrum analysis, specifically for a C-phase fault. Detailed Implementation

[0073] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0074] Example 1, referring to Figure 1 This is one embodiment of the present invention, which provides a method for detecting excitation load grounding faults based on EEMD and Hilbert spectrum analysis, including:

[0075] Existing methods for detecting short-circuit faults in excitation systems generally suffer from the following problems: First, traditional signal processing methods typically rely on single time-domain or frequency-domain analysis techniques, such as Empirical Mode Decomposition (EMD) or Fast Fourier Transform (FFT), which are insufficient to accurately process non-stationary signals generated during excitation system operation. This is especially true in the presence of high-frequency interference and subsynchronous oscillation coupling characteristics, which can easily lead to mode aliasing and prevent accurate extraction of fault features. Second, existing methods are insensitive to frequency changes at the moment of fault in frequency-domain energy extraction, often resulting in frequency offset or amplitude misjudgment, leading to unstable frequency identification. Furthermore, traditional methods generally lack the ability to track the joint evolution of time and frequency, making it difficult to pinpoint the exact moment of fault occurrence, especially when dealing with latent fault types such as high-resistance grounding, which can easily lead to false detections or missed detections. Finally, existing methods typically only output single-dimensional fault information, lacking the ability to comprehensively judge fault type, influencing channels, and start time, making it difficult to meet the requirements of diagnostic completeness and response speed in practical engineering applications.

[0076] S1. Acquire electrical signals during the operation of the equipment, preprocess the acquired signals, and construct a discrete time-domain data sequence;

[0077] S2. Perform iterative decomposition on the data sequence to obtain a sequence of several component quantities;

[0078] S3. Perform time-domain structural analysis on the sequence of each component to identify the target component sequence containing fault mutation features;

[0079] S4. Perform frequency domain transformation on the target component sequence, construct the spectrum distribution through frequency mapping operation, and normalize and quantize the amplitude of each frequency component in the spectrum.

[0080] S5. Perform time-frequency joint transformation on the target component sequence to generate the time-frequency matrix of the component sequence, and calculate the energy distribution data and energy entropy index based on the matrix.

[0081] S6. Based on the component sequences, spectral distribution data, time-frequency matrix, energy distribution data, and energy entropy index generated by iterative decomposition, perform fault judgment processing and output the fault type, fault channel, and diagnostic results of the fault start time.

[0082] This invention improves the decoupling capability of different frequency components in non-stationary signals and enhances the component separability of fault features by introducing a multi-round perturbation superposition and repeated decomposition mechanism during fault signal processing. By performing frequency domain feature extraction and amplitude normalization on the selected component sequences, a structurally stable frequency distribution information is constructed, providing a discriminative frequency feature foundation for fault identification. Furthermore, by combining time-series and frequency joint analysis operations, a complete time-frequency energy description structure is formed, and based on energy distribution and entropy index extraction operations, comprehensive identification of fault mutation behavior is supported. Finally, relying on the above-mentioned structured multidimensional feature data, this invention can output joint discrimination results regarding fault type, fault location, and start time, possessing good diagnostic integrity and processing stability, and is suitable for the automated detection needs of short-circuit grounding faults in excitation power generation systems.

[0083] Example 2, refer to Figures 2-6 As an embodiment of the present invention, based on the previous embodiment, a method for detecting grounding faults in excitation loads based on EEMD and Hilbert spectrum analysis is provided, comprising:

[0084] In step S1, electrical signals during equipment operation are collected, and the collected signals are preprocessed to construct a discretized time-domain data sequence, including the following steps A1 to A4:

[0085] A1. Acquire electrical signal data that reflects the operating status of the excitation system.

[0086] The collected electrical signals are the voltage dynamic signals of phases A, B, and C of the excitation system, covering the stable operation stage before the fault occurs, the transient stage at the moment of the fault, and the recovery stage after the fault, and fully recording the dynamic changes of electrical quantities in each phase.

[0087] A2. Perform preprocessing operations on the acquired electrical signal data.

[0088] A3. Sample data at fixed time intervals and generate data segments with time series indexes.

[0089] The fixed time interval is set according to the system power frequency and the subsynchronous resonance characteristic frequency band, and the sampling frequency satisfies the Nyquist sampling theorem. The continuous voltage signal is converted into discrete data points through this sampling frequency, and each data point corresponds to a unique timestamp, forming a time-domain sequence covering the entire fault process.

[0090] A4. Perform time base alignment on the sampled data and construct a multi-channel discrete dataset with a unified time identifier.

[0091] Time reference alignment eliminates phase deviations in phases A, B, and C caused by time differences in the acquisition devices by synchronizing the acquisition trigger times of each phase voltage signal; unified time identification uses the same timestamp to integrate the discrete sampling data of the three phases into a multi-channel dataset, enabling accurate comparison of signals of different phases in the time dimension.

[0092] In this application implementation, the preprocessing operation in step A2 involves filtering out high-frequency interference noise, retaining the 20-50Hz subsynchronous resonance frequency band and the 60Hz power frequency component, ensuring that fault-related transient characteristics and steady-state components are not lost; simultaneously, baseline correction is performed to eliminate DC drift in the signal, making the voltage signal reference of each phase uniform. Specific steps include:

[0093] High-frequency interference noise is filtered out by designing a 20-60Hz bandpass filter, while retaining the 20-50Hz subsynchronous resonance frequency band and the 60Hz power frequency component to ensure the integrity of fault transient characteristics and steady-state components. The baseline correction in the preprocessing step above eliminates DC drift in the signal, making the voltage signal reference of each phase uniform. This provides high-quality input for subsequent EEMD decomposition, frequency domain analysis, and multiphase signal comparison, avoiding the impact of noise interference and reference differences on fault feature identification.

[0094] In one alternative implementation, the preprocessing operation involves bandpass filtering the acquired phase voltage signals, setting the retained frequency band to cover the power frequency and subsynchronous resonance frequency range, and filtering out irrelevant frequency components that are below the set lower limit or above the set upper limit.

[0095] The filtered signal is normalized to ensure the comparability of different phase signals on the amplitude scale and to eliminate the impact of gain differences in the acquisition system.

[0096] In another alternative implementation, the preprocessing operation is to use wavelet denoising to decompose and reconstruct the original voltage signal, suppress non-periodic high-frequency impact interference, and at the same time retain as much transient information as possible during the fault.

[0097] The denoised signal is subjected to endpoint extension and interpolation completion processing to eliminate the discontinuity of the sampling endpoints and ensure the smoothness of the boundary conditions for subsequent frequency analysis.

[0098] The acquisition phase ensures that the acquired three-phase signals cover the complete operational process before, during, and after the fault, and achieves precise alignment of the data for each phase under a unified time reference, thus guaranteeing the comparability of signals from different channels in the time dimension. After preprocessing, high-frequency interference noise unrelated to the fault is effectively filtered out, the target frequency band relevant to the fault characteristics is retained, and DC drift is eliminated, ensuring consistency in amplitude across all phases. The sampling process satisfies the Nyquist sampling theorem, and the obtained discretized time-domain data sequence possesses both full coverage and high signal-to-noise ratio and amplitude consistency, providing stable and reliable input conditions for subsequent component decomposition, frequency feature extraction, and time-frequency analysis.

[0099] In step S2, iterative decomposition is performed on the data sequence to obtain several component quantity sequences, including the following steps B1 to B4:

[0100] B1. Introduce a perturbation signal into the constructed discrete data sequence and perform several repeated decomposition operations.

[0101] The collected electrical signals are the voltage dynamic signals of phases A, B, and C of the excitation system, covering the stable operation stage before the fault occurs, the transient stage at the moment of the fault, and the recovery stage after the fault, and fully recording the dynamic changes of electrical quantities in each phase.

[0102] B2. In each round of decomposition, the original data sequence is split into several components with different frequency characteristics.

[0103] B3. Perform structural consistency judgment on the decomposition results of each round, and perform integrated averaging on components with the same index.

[0104] B4. After completing the preset number of iterations, output a set of components with a frequency hierarchy.

[0105] In this implementation, the decomposition operation in step B2 employs the Ensemble Empirical Mode Decomposition (EEMD) method. Before each round of decomposition, amplitude-controlled Gaussian white noise is superimposed onto the original signal. Through multiple decompositions and averaging processes, mode aliasing is suppressed, improving the separation of high-frequency and low-frequency components. A detailed explanation of the algorithm includes...

[0106] EEMD is an adaptive signal decomposition method for processing nonlinear and non-stationary signals. It obtains multiple Intrinsic Mode Functions (IMFs) of the original signal by adding white noise to the signal and performing Empirical Mode Decomposition (EMD) multiple times. These IMFs can better capture the local features of the signal, especially for nonlinear and non-stationary signals. EEMD adds appropriate white noise to EMD to eliminate the "mode aliasing" phenomenon, making the extreme points of the signal evenly distributed throughout the time-frequency space. The decomposition results are averaged through multiple iterations; the more iterations, the less noise interferes with the decomposition results, and the more the useful signal is highlighted. The original signal X(t) is set to the number of averaging iterations N, with initial i = 1, 2, ..., N. Specific decomposition method:

[0107] (1) Add Gaussian white noise ρ to x(t) i (t) Obtain the composite signal X i (t):

[0108] X i (t)=x(t)+ρ i (t)

[0109] In the formula, X i ρ(t) represents the synthesized signal after adding noise in the i-th iteration, serving as the input to the EEMD decomposition and used to separate multi-scale features in the signal with the aid of noise. x(t) represents the original power grid fault signal (such as voltage and current dynamic signals), containing complex components such as fault transients, power frequency steady-state, and noise; it is a non-stationary and nonlinear original input. i (t) is Gaussian white noise injected in the i-th iteration, which makes the extreme points of the signal uniformly distributed in the time-frequency space and suppresses mode mixing.

[0110] (2) Using EMD to decompose X1(t) to obtain IMF components at different scales:

[0111]

[0112] Where n is the number of IMFs decomposed by EMD, r i,n X1(t) represents the residual component, and X1(t) represents the synthesized signal after the initial addition of noise, serving as the input for a single EMD decomposition. i,n (t) is the nth intrinsic mode function (IMF) obtained by the i-th iteration decomposition. It is a signal component with physical meaning and corresponds to the characteristics of a specific time scale.

[0113] (3) Add different Gaussian white noise to X(t), and repeat steps (1) and (2) to obtain the k-th component of x(t) after EEMD decomposition:

[0114]

[0115] Among them, X k (t) represents the synthesized signal after adding noise in the k-th iteration. The decomposition results under different noise conditions are obtained through multiple iterations. i,n (t) represents the nth IMF obtained from the kth iteration decomposition, which is used together with the components from other iterations for subsequent averaging to reduce noise interference. i,n (t) is the residual component of the k-th iteration, which, corresponding to the former, tends to stabilize as the number of iterations increases.

[0116] (4) Average the IMF components after EEMD decomposition to eliminate noise interference to the signal.

[0117]

[0118] Among them, IMF n (t) represents the nth IMF obtained after N iterations of averaging, which is the final effective component used for feature extraction. Averaging reduces the interference of noise on the decomposition results; N is the number of ensemble averaging iterations, which helps to offset the influence of random noise and improve the stability of the IMF. k,n (t) is the nth IMF in the kth iteration, which is the basic data for averaging. Its fluctuations are randomly distributed due to noise injection, and averaging can highlight the effective signal.

[0119] Use the IMF n The error between the reconstructed signal obtained by summing (t) and the original signal X(t) satisfies the following equation:

[0120]

[0121] Where N is the number of clusters in EEMD; ω is the standard deviation with added white noise; as can be seen from the above formula, ω n The value decreases as the number of polymerization cycles N increases. Let ω = 0.2 and N = 10.

[0122] The flowchart of the steps for EEMD signal decomposition is as follows: Figure 2 As shown.

[0123] The original signal is decomposed into multiple IMF components by EEMD, each IMF corresponding to physical characteristics at different time scales: high-frequency transients or low-frequency trends.

[0124] Waveform morphology: Based on the waveform morphology displayed at high frequencies (IMF1~IMF4), the high-frequency IMF exhibits dense oscillations, corresponding to the high-frequency transient impact at the moment of the fault; the low-frequency IMF4 ( Figure 3 The fluctuations are gentle, reflecting the steady-state current of the system. (Refer to...) Figure 3The time-domain waveforms of the four intermediate frequency components (IMF4 to IMF7) of the fault signal after EEMD decomposition are shown. These four components are all related to the coupling characteristics of subsynchronous oscillation (20-50Hz) and power frequency (60Hz), but they differ significantly in waveform morphology, abrupt change characteristics, and physical meaning. IMF4, as the high-frequency component in the intermediate frequency band, has a relatively dense waveform oscillation with a short period. It exhibits a significant abrupt change at the moment of fault occurrence, although the abrupt change amplitude is smaller than that of IMF5, and the oscillation decays faster after the abrupt change. It mainly reflects the energy transfer process from the high-frequency transient to the intermediate frequency in the early stage of the fault, including the initial coupling signal between the high-frequency electromagnetic oscillation generated by arc breakdown and the subsynchronous resonance. IMF5 is the core component with the highest fault differentiation among the four components. Its waveform oscillation period is between that of IMF4 and IMF6, and its amplitude dynamic range is the widest. When a fault occurs, the waveform abrupt change is the most significant. When phase A is metallically grounded, it exhibits a sudden drop followed by a sustained low position without recovery. When phase B is affected, the time of impact shows instantaneous fluctuation followed by gradual recovery. When phase C is affected, the time of impact shows a state of small fluctuation followed by rapid recovery. It concentrates the core energy of subsynchronous resonance and is the main carrier of fault transient energy and power frequency modulation, directly corresponding to the fault type and phase. IMF6 is the low-frequency component in the mid-frequency band. Its waveform oscillation is relatively smooth, and its period is longer than that of IMF5. Although there is a waveform abrupt change when a fault occurs, the amplitude is significantly smaller than that of IMF5, and the decay rate after the abrupt change is faster. It mainly reflects the energy diffusion process at lower frequencies in subsynchronous resonance and can help verify the fault characteristics of IMF5. IMF7 is the component closest to the power frequency in the mid-frequency band. Its waveform is the smoothest, its oscillation period is the longest, and it approximates quasi-steady-state fluctuations. When a fault occurs, it only exhibits weak fluctuations with short durations, quickly recovering to the pre-fault steady-state trend. It primarily carries the fundamental power frequency component and is least affected by fault transients, reflecting more of the system's steady-state power frequency characteristics. It is used to provide a system steady-state benchmark and assist in quantifying the abrupt changes in other components. These four components reflect the energy changes and characteristics during the fault process from different perspectives, collectively providing important time-domain information for fault diagnosis. Amplitude label: The values ​​on the vertical axis represent the energy amplitude range of different IMFs.

[0125] Time axis distribution: The horizontal axis covers the complete fault cycle. Abrupt changes in the IMF component within a specific time window correspond to fault events. The observed phenomenon is abrupt changes in the waveforms of IMF4-IMF7 in the mid-frequency band, and this band corresponds to the power grid operating frequency of 60Hz. Due to IMF5 ( Figure 3 Because IMF5 has higher discrimination and correlation, it was chosen for feature extraction and repeatability testing. Similarly, the EEMD algorithm was used to process the voltage data of phases B and C, resulting in a comparison chart of the IMF5 components of the three-phase voltage data for the same time period, as shown below. Figure 4 .

[0126] For phase A as the faulty phase, the voltage fluctuation of the IMF5 component after EEMD decomposition shows a significant and sharp drop, with the amplitude remaining at a low level and no rapid recovery observed. This indicates that the fault triggered an irreversible voltage collapse. For phase B, the IMF5 component exhibits instantaneous amplitude fluctuations at the moment of the fault, but quickly recovers to a stable state after the fluctuations. This characteristic suggests that the fault in phase B did not directly cause a sustained short circuit, but may have been indirectly affected by the fault in phase A. For phase C, the fault manifests as instantaneous fluctuations in the IMF5 component, but its fluctuation amplitude is the smallest and its recovery speed is the fastest, even approaching normal levels, possibly corresponding to a transient process caused by a momentary arc short circuit.

[0127] The core function of EEMD is to decompose the nonlinear, non-stationary raw power grid fault signal into 14 intrinsic mode functions (IMFs) and 1 residual component (output). These IMF components are ordered from high to low frequency and correspond to physical processes at different time scales.

[0128] The output is directly used as input for FFT and HHT: FFT analyzes the selected key IMFs (such as IMF5 with the highest discrimination), while HHT performs time-frequency domain transformation on all IMFs (with a focus on IMF5). Both rely on EEMD's "signal decoupling" capability, that is, by stripping away noise and multi-scale interference, the fault characteristics are highlighted in a single IMF, laying the foundation for subsequent feature extraction.

[0129] In one alternative implementation, the decomposition operation employs the EMD algorithm. Each round of decomposition uses the EMD algorithm, which splits the data sequence into components that satisfy the intrinsic mode function (IMF) condition and one residual component through an iterative screening process. Each IMF component is sorted from high to low frequency, with high-frequency components reflecting transient shocks and low-frequency components reflecting steady-state trends.

[0130] In another alternative implementation, the decomposition operation employs the Complete Set Empirical Mode Decomposition (CEEMDAN) method, which introduces noise while successively correcting the residual components, thereby further improving the stability of the component decomposition and reducing the interference of noise residue on low-frequency components during repeated decomposition.

[0131] In step S3, time-domain structural analysis is performed on the sequence of each component to identify the target component sequence containing fault mutation features, including the following steps C1 to C4:

[0132] C1. Construct the corresponding time variation curve for each component sequence;

[0133] C2. Compare the fluctuation trend, amplitude fluctuation and rate of change of each component within the target time interval;

[0134] The target time interval is 0.5 seconds before and after the fault occurs; the fluctuation trend is monitored for sudden drops, fluctuations, or stabilization; the amplitude fluctuation is calculated as the difference between the amplitude at the fault time and the steady state time.

[0135] C3. Select component sequences with mutation characteristics based on preset judgment rules;

[0136] C4. Mark the components that meet the mutation characteristics as the target objects for subsequent frequency domain and time-frequency domain processing.

[0137] The component that meets the requirements of mutation characteristics is IMF5. This component has the largest mutation amplitude and the most stable persistence characteristics at the time of the fault, and it embodies the core characteristics of subsynchronous resonance. Therefore, it is marked as the target component.

[0138] In step S4, frequency domain transformation is performed on the target component sequence. A spectral distribution is constructed through frequency mapping operations, and the amplitude of each frequency component in the spectrum is normalized and quantized. This includes the following steps D1 to D5:

[0139] D1. Perform signal length standardization on the target component sequence and establish a frequency index relationship with the time series.

[0140] D2. Perform frequency domain transformation based on the discrete signal structure to extract the complex coefficients corresponding to each frequency point.

[0141] D3. Calculate the magnitude of the amplitude information at each frequency point and generate a one-to-one mapping table between frequency and amplitude.

[0142] D4. Perform proportional conversion on the frequency amplitude results to construct a normalized amplitude sequence within a unified numerical range.

[0143] The scaling is based on the maximum amplitude in the mapping table. The amplitude of all frequency points is divided by this maximum value to obtain the normalized amplitude within the interval. The normalized amplitudes of 34Hz and 45Hz are retained as key feature values.

[0144] D5. The normalized amplitude sequence is used as the frequency distribution feature vector for subsequent fault type identification processing.

[0145] The frequency distribution feature vector contains the normalized amplitude of each frequency point in the 20-50Hz band, with a focus on the difference between the amplitude at 34Hz and 45Hz.

[0146] In this implementation, step D2, the frequency domain transformation processing, employs a Fast Fourier Transform (FFT) to calculate the standardized target component sequence. The output complex coefficients contain the amplitude and phase information of each frequency component, with the core retaining coefficient data in the 20-50Hz frequency band. The specific algorithm includes...

[0147] The Fast Fourier Transform (FFT) is an efficient algorithm for calculating the Discrete Fourier Transform and its inverse. In power grid fault detection, it can accurately separate signal frequency components and quantify the energy distribution at different frequencies, providing crucial frequency domain information for fault characteristic frequency location and type determination. It can calculate the signal's frequency domain representation in a short time, thus providing the signal's frequency characteristics. The implementation steps of FFT in a program are as follows:

[0148] 1. Input parameter processing: parameter initialization, obtaining signal length;

[0149] The input parameters are the IMF components after EEMD decomposition, which are discrete-time sequence x(n).

[0150] 2. Frequency vector generation: based on sampling frequency f s Calculate the frequencies across the entire range and extract the effective frequency bands;

[0151]

[0152] Where k is the frequency domain index, and f(k) corresponds to the frequency value at point k.

[0153] This allows us to retain the 20-50Hz subsynchronous resonance frequency band and the 60Hz power frequency, eliminate irrelevant high-frequency noise, and focus on the fault characteristic frequency.

[0154] 3. FFT core operations: Directly call the built-in FFT function and perform IMF amplitude spectrum calculation;

[0155]

[0156] Where X(k) is the frequency domain complex result, reflecting the amplitude and phase of the k-th frequency.

[0157] Calculate the amplitude spectrum to represent the energy intensity of each frequency component. The amplitude of characteristic frequencies such as 34Hz is a key indicator for fault diagnosis.

[0158] A(k)=|X(k)|

[0159] 4. Result Processing and Visualization: Combining fault characteristic frequencies (such as the 20-50Hz subsynchronous resonance band), the spectrum is analyzed and finally visualized. First, normalization is performed to eliminate the influence of signal amplitude differences, facilitating comparison of different IMF components. Second, spectrum analysis is conducted, combining the characteristics of the 20-50Hz band to analyze the presence or absence of the 34Hz amplitude and 45Hz secondary harmonic in IMF5. The amplitude differences of the three phases are used to help determine the fault type, and finally, a normalized spectrum diagram is output.

[0160] Reference Figure 5The Fast Fourier Transform (FFT) spectrum of the IMF components shows the frequency domain energy distribution of the IMF5 components, locating fault characteristic frequencies (such as high-frequency resonances and power frequency harmonics); according to Figure 4 After noise reduction, the EEMD decomposition IMF diagram is obtained. IMF5 shows a peak, so the characteristic frequency corresponding to the power grid circuit fault is about 30Hz, which is consistent with the frequency characteristics of the subsynchronous resonance of the SSR system and is lower than the original system's natural frequency of 60Hz.

[0161] In one alternative implementation, the frequency domain transformation process employs the Discrete Fourier Transform (DFT). In scenarios where the number of time-domain sample points is small or where precise control of computational accuracy is required, a complex exponential weighted summation operation is performed point by point on the target component sequence to obtain the amplitude and phase results at each discrete frequency point, and the data of the target frequency band is retained for constructing the spectral distribution.

[0162] In another alternative implementation, the frequency domain transformation processing employs the Welch method for power spectrum estimation. By segmenting the target component sequence, windowing, and performing frequency domain transformation and averaging on the segmented results, the variance fluctuation of the spectrum estimation is reduced, thereby obtaining stable frequency amplitude distribution characteristics under low signal-to-noise ratio conditions.

[0163] It can accurately map the target component sequence in the time domain to the frequency domain and form a normalized spectral distribution, enabling the quantification of the relative energy differences between different frequency components. Signal length standardization and frequency indexing ensure the consistency of each component on the frequency scale; amplitude magnitude calculation and normalization make components of different amplitude orders comparable; the finally constructed frequency distribution feature vector highlights the amplitude differences between key frequency points (such as 34Hz and 45Hz), providing a clear frequency domain basis for subsequent fault type identification.

[0164] In step S5, a time-frequency joint transform is performed on the target component sequence to generate the time-frequency matrix of the component sequence. Based on the matrix, the energy distribution data and energy entropy index are calculated, including the following steps E1 to E5:

[0165] E1. A synchronous analytical operation is performed on the target component sequence to conduct time-frequency joint transformation, and a two-dimensional data array describing the local variation characteristics of the signal is constructed.

[0166] E2. In the data array, extract the amplitude distribution information corresponding to each time segment according to the frequency direction, and calculate the cumulative energy value of each frequency point.

[0167] E3. The cumulative energy value is expressed in a normalized distribution in the frequency domain to generate an energy distribution vector.

[0168] E4. Based on the proportion of each frequency component in the energy distribution vector, calculate the statistical index value representing the degree of energy concentration, which is used as the energy entropy result.

[0169] E5. Use the time-frequency matrix, energy distribution vector, and energy entropy result as the input data set for subsequent judgment and processing.

[0170] In this implementation, step E1, the time-frequency joint transformation processing, employs Hilbert transform for synchronization analysis. By constructing an analytical signal, the instantaneous frequency and amplitude at each time point (millisecond level) are extracted, constructing a two-dimensional time-frequency matrix. The matrix elements are the amplitude values ​​at the corresponding time-frequency points, highlighting local features, particularly the energy clusters in the 20-50Hz frequency band at the time of the fault. The specific algorithm includes...

[0171] HHT is a signal analysis method based on Hilbert spectrum, capable of providing the time-frequency characteristics of a signal. It decomposes the signal into a series of Intrinsic Mode Functions (IMFs), selects the IMF components that best represent the characteristics, and calculates the instantaneous frequencies of the IMFs, thereby better capturing local features and nonlinear behavior in the signal. Further marginal spectral analysis, specifically performing HHT transform on IMF5, reveals a surge in marginal spectral energy of IMF5, such as... Figure 6 The marginal spectrum is the integral of the time-frequency spectrum in the HHT on the time axis, reflecting the total energy distribution at different frequencies. IMF5, as a mid-frequency component, has strong coupling characteristics with the power frequency (60Hz) and subsynchronous oscillations (20-50Hz). Under normal operating conditions, its energy is mainly concentrated at the power frequency, and the marginal spectrum exhibits a single main peak. When a ground fault occurs, the arc discharge and voltage drop at the fault point cause energy to shift from the power frequency to the subsynchronous frequency band, resulting in a significant increase in the energy of IMF5 in the 20-50Hz band, manifested as a surge in marginal spectrum energy, which can be used to identify a ground fault. Its marginal spectrum energy is mainly concentrated near the low-frequency band. Comparing this with the fault characteristic frequency distribution obtained in the previous section, the two are consistent, and it can be determined that the fault is a short-circuit fault caused by a sudden change in the detection signal (voltage signal) to 0. Furthermore, the HHT energy entropy, by quantifying the disorder of the time-frequency energy distribution, provides a cross-scale quantitative indicator for feature extraction and state identification of complex signals. Empirical Mode Decomposition (EMD) is performed on the signal using HHT to obtain a set of Intrinsic Mode Functions (IMFs). Then, a Hilbert Transform is applied to each IMF to generate a time-frequency spectrum matrix. The energy distribution of each frequency component in the time-frequency spectrum is considered as a probability density function, and its information entropy value is calculated using the following formula:

[0172]

[0173] Where H is the HHT energy entropy, the value of which is related to the energy distribution, and f i Let P(f) be the i-th frequency point in the time spectrum.i ) represents a specific frequency f i The time-frequency energy ratio is the probability density function of the entropy value.

[0174] The meaning of entropy: The higher the entropy, the more uniform the signal energy is distributed in the time and frequency domain (such as white noise); the lower the entropy, the more concentrated the energy is in a specific frequency band or time point (such as fault impact signal). Therefore, when a fault occurs in the power grid, the entropy at the fault point will surge or drop sharply. In the simulation process, this invention calculated that the entropy of IMF5 is higher than the entropy before and after, and determined that this characteristic frequency is the subsynchronous frequency.

[0175] In one alternative implementation, the time-frequency joint transformation process employs a short-time Fourier transform (STFT). By applying a sliding time window to the target component sequence and performing a frequency domain transformation within each time window, the spectral distribution that varies with time is obtained. This leads to the construction of a time-frequency two-dimensional matrix, and the calculation of the energy distribution and energy entropy index for the corresponding frequency range.

[0176] In another alternative implementation, the time-frequency joint transformation process employs wavelet transform, using multi-scale analysis to decompose the target component sequence, obtain a time-scale two-dimensional matrix, convert it into a time-frequency matrix, extract the energy distribution characteristics of the target frequency band, and calculate the energy entropy index.

[0177] It can simultaneously retain both temporal and frequency information of the signal, visually presenting the frequency energy changes of the target component before and after a fault in the form of a two-dimensional matrix. Using the energy distribution vector, the proportion of different frequency components in the overall energy can be accurately quantified, while the energy entropy index reflects the concentration and uniformity of energy distribution in the frequency domain, thus revealing the energy accumulation or diffusion characteristics of key frequency bands during a fault. The Hilbert transform in the underlying implementation can capture instantaneous frequency and amplitude changes at millisecond-level resolution.

[0178] In step S6, based on the component sequences, spectral distribution data, time-frequency matrix, energy distribution data, and energy entropy index generated by the iterative decomposition process, fault judgment processing is performed, and the diagnostic results of fault type, fault channel, and fault start time are output, including the following steps F1 to F5:

[0179] F1. Analyze the changing trends and persistence characteristics of key waveforms in the component sequence to determine their corresponding operating state categories.

[0180] The key waveforms are the time-domain waveforms of IMF5. Phase A shows "sudden drop followed by sustained low position without recovery" (fault phase characteristics), Phase B shows "instantaneous fluctuation followed by partial recovery" (affected characteristics), and Phase C shows "small fluctuation followed by rapid recovery" (slightly affected characteristics).

[0181] F2. Compare the amplitude differences of each characteristic frequency in the spectral distribution data to distinguish different types of electrical disturbance modes.

[0182] F3. Based on the starting time node of energy accumulation in the time-frequency matrix, determine the specific time when the fault occurs.

[0183] The starting time point for energy accumulation is the timestamp of the first appearance of the energy cluster in the 20-50Hz frequency band. This time coincides with the abrupt change in the IMF5 time-domain waveform and serves as the fault initiation time.

[0184] F4. Combining the frequency concentration of energy distribution with the changing trend of energy entropy helps to identify the duration of abnormal signals and local impact characteristics.

[0185] The concentration of energy distribution is judged by the energy proportion of 34Hz. The energy entropy trend of phase A is "sudden increase followed by decline and sustained low entropy", reflecting the continuous and dispersed characteristics of fault energy.

[0186] F5. Logically integrate various judgment criteria to generate a comprehensive diagnostic result that includes the nature of the fault, the corresponding phase, and the time of occurrence.

[0187] Logical integration requires consistency of multi-dimensional characteristics: when the IMF5 time domain characteristics, 34Hz amplitude, and energy entropy trend all point to phase A, it is determined to be "phase A metallic ground fault", that is, the fault channel is phase A; similarly, the affected states of phase B and phase C are distinguished, and finally a unique diagnostic result is formed.

[0188] Example 3, referring to Figures 7-9 As one embodiment of the present invention, a method for detecting grounding faults in excitation loads based on EEMD and Hilbert spectrum analysis is provided. To verify the beneficial effects of the present invention, scientific demonstration is carried out through experiments.

[0189] When a short circuit occurs in phase A, from the time domain perspective, the high-frequency IMF component exhibits specific pulse characteristics, while the mid-frequency IMF5 component shows a voltage drop followed by a sustained low level with no recovery trend. In the frequency domain, after FFT analysis, IMF5 shows a significant energy peak at a specific frequency without the appearance of secondary harmonics at another specific frequency. From the time-frequency domain perspective, the HHT energy entropy changes abruptly, and the HHT marginal spectrum energy of IMF5 increases significantly in a specific frequency band.

[0190] For phase B, the mid-frequency IMF5 component in the time domain will have instantaneous fluctuations when a fault occurs, and can recover to a stable state relatively quickly afterward. The low-frequency IMF component, on the other hand, exhibits continuous fluctuation characteristics. In the frequency domain, the energy amplitude corresponding to a specific frequency in the FFT spectrum of IMF5 is relatively lower than that of phase A when it is short-circuited, and there is no obvious secondary harmonic at another specific frequency.

[0191] For phase C, the fluctuation amplitude of the IMF5 component in the intermediate frequency in the time domain is small and can quickly recover to the normal level; in the frequency domain, the energy amplitude of the IMF5 FFT spectrum corresponding to a specific frequency is lower than that of phase B when short-circuited, and it is accompanied by a secondary harmonic at another specific frequency.

[0192] If it is a multi-phase short circuit, taking phases A and B as an example, its characteristics will simultaneously reflect the core characteristics of phases A and B when they are short-circuited, but the corresponding characteristic intensity will change to some extent.

[0193] In terms of fault time location, the core is to base it on the moment when the time domain waveform of the intermediate frequency IMF5 changes abruptly, combined with the time point when the energy cluster of a specific frequency band first appears in the HHT time spectrum, and then use the pulse start time of the high frequency IMF component for auxiliary verification. The time determined after cross-verification is the fault start time, which can achieve high location accuracy.

[0194] By decomposing the power grid fault signal using EEMD, different fault types are determined based on the IMF component:

[0195] A-phase single-phase short-circuit ground fault: When a metallic short-circuit fault occurs in phase A, the voltage fluctuation of the IMF5 component after EEMD decomposition shows a significant and sharp drop ( Figure 7 Furthermore, the amplitude remained low, with no rapid recovery observed. This indicates that the fault triggered an irreversible voltage collapse, with transient energy concentrated in the IMF5 component through low-frequency subsynchronous resonance (20-50Hz). FFT analysis further confirmed the existence of fault characteristic frequencies; the 34Hz harmonic had the highest amplitude in the IMF5 spectrum, while the system fundamental frequency of 60Hz still showed a faint presence due to filter residue, but the fault characteristics remained clearly discernible.

[0196] Phase B affected: When a phase B fault occurs, the IMF5 component exhibits instantaneous amplitude fluctuations at the moment of the fault. Figure 8 However, the system quickly recovered to a stable state after the initial fluctuation. This characteristic indicates that the phase B fault did not directly cause a sustained short circuit, but was likely caused by indirect disturbances transmitted from the phase A fault. Although the dominant frequency of the IMF5 component is still concentrated in the subsynchronous range (20-50Hz), the amplitude of the 34Hz harmonic in its FFT spectrum is lower than that of phase A, reflecting the difference in the fault energy transmission path. The system restored the voltage through the excitation regulation section, indicating that the fault severity was lower than that of the phase A condition.

[0197] C-phase impact: The C-phase fault manifests as instantaneous fluctuations in the IMF5 component. Figure 9 However, its fluctuation amplitude is the smallest and its recovery speed is the fastest, even approaching normal levels, which may correspond to the transient process caused by a momentary arc short circuit. The amplitude of the 34Hz harmonic in the FFT spectrum further decreases, accompanied by a 45Hz secondary harmonic ( Figure 8 This indicates that the fault energy distribution is more dispersed. The system completely suppressed the fault's impact through rapid adjustment, verifying the instantaneous characteristics of the C-phase fault.

[0198] Although all three single-phase short-circuit faults reflect the subsynchronous resonance characteristic frequency (20-50Hz) through the intermediate frequency component of the IMF, their transient processes and recovery characteristics differ significantly: Phase A faults cause a sudden and irreversible drop in IMF5, corresponding to a metallic short circuit; Phase B fluctuates and partially recovers, reflecting indirect fault propagation; Phase C fluctuates the least and recovers the fastest, indicating instantaneous short-circuit characteristics. The amplitude differences of the intermediate frequency harmonics in the FFT spectrum (A>B>C) further quantify the fault energy intensity, while the dynamic response of the IMF components (such as sudden drop and recovery speed) combined with frequency domain characteristics can accurately distinguish the fault type and locate the fault phase, providing an effective criterion for the diagnosis of subsynchronous resonance faults in low-frequency systems.

[0199] Short-circuit faults in the excitation load power grid generate harmonic components in the grid signal. Since the grid current signal is non-stationary, traditional Fourier transforms are insufficient for spectral analysis. Leveraging the HHT algorithm's ability to analyze nonlinear and non-stationary signals, this algorithm addresses the issue of spurious components and avoids interference from irrelevant frequencies. Finally, simulation experiments verify that the algorithm can extract the characteristic frequencies of short-circuit faults, thereby diagnosing the fault occurrence and determining the fault type through the diagnostic waveform.

[0200] Example 4 is an embodiment of the present invention. This embodiment provides an excitation load grounding fault detection system based on EEMD and Hilbert spectrum analysis, including a data collection module, a data decomposition module, a data identification module, a data mapping module, a data calculation module, and a result output module.

[0201] The data collection module collects electrical signals during the operation of the equipment, preprocesses the collected signals, and constructs a discrete time-domain data sequence.

[0202] The data decomposition module performs iterative decomposition processing on the data sequence to obtain several component quantity sequences;

[0203] The data identification module performs time-domain structural analysis on the component sequence to identify the target component sequence containing fault mutation features.

[0204] The data mapping module performs frequency domain transformation on the target component sequence, constructs a spectral distribution through frequency mapping operations, and normalizes and quantizes the amplitude of each frequency component in the spectrum.

[0205] The data calculation module performs time-frequency joint transformation processing on the target component sequence to generate the time-frequency matrix of the component sequence, and calculates the energy distribution data and energy entropy index based on the matrix.

[0206] The result output module performs fault judgment processing based on the component sequence, spectral distribution data, time-frequency matrix, energy distribution data, and energy entropy index generated by iterative decomposition processing, and outputs the diagnostic results of fault type, fault channel, and fault start time.

[0207] This embodiment also provides an electronic device applicable to a method for detecting excitation load grounding faults based on EEMD and Hilbert spectrum analysis, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the method for detecting excitation load grounding faults based on EEMD and Hilbert spectrum analysis as proposed in the above embodiment.

[0208] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements a ground fault detection method for excitation load based on EEMD and Hilbert spectrum analysis as proposed in the above embodiment.

[0209] The storage medium proposed in this embodiment belongs to the same inventive concept as the method for detecting excitation load grounding faults based on EEMD and Hilbert spectrum analysis proposed in the above embodiments. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0210] Based on the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0211] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not 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 detecting grounding faults in excitation loads based on EEMD and Hilbert spectral analysis, characterized in that: include, The electrical signals acquired during the operation of the acquisition equipment are preprocessed to construct a discrete time-domain data sequence. Perform iterative decomposition on the data sequence to obtain a sequence of several component quantities; Temporal structure analysis was performed on the sequence of each component to identify the target component sequence containing fault mutation features; The target component sequence is subjected to frequency domain transformation processing, a spectral distribution is constructed through frequency mapping operation, and the amplitude of each frequency component in the spectrum is normalized and quantized. Perform time-frequency joint transformation on the target component sequence to generate the time-frequency matrix of the component sequence, and calculate the energy distribution data and energy entropy index based on the matrix; Based on the component sequences, spectral distribution data, time-frequency matrix, energy distribution data, and energy entropy index generated by iterative decomposition, fault diagnosis is performed, and the diagnostic results, including fault type, fault channel, and fault initiation time, are output.

2. The excitation load grounding fault detection method based on EEMD and Hilbert spectrum analysis as described in claim 1, characterized in that: The electrical signals acquired during the operation of the acquisition device are preprocessed to construct a discretized time-domain data sequence, including... Acquire electrical signal data reflecting the operating status of the excitation system; Perform preprocessing operations on the acquired electrical signal data; Data segments with time-series indexes are generated by sampling at fixed time intervals. Perform time base alignment on the sampled data and construct a multi-channel discrete dataset with a unified time identifier.

3. The excitation load grounding fault detection method based on EEMD and Hilbert spectrum analysis as described in claim 2, characterized in that: The iterative decomposition process performed on the data sequence yields several component quantity sequences, including: A perturbation signal is introduced into the constructed discrete data sequence and several repeated decomposition operations are performed. In each round of decomposition, the original data sequence is split into several components with different frequency characteristics; The structural consistency of the decomposition results in each round is judged, and the components with the same index are integrated and averaged. After completing a preset number of iterations, a set of components with a frequency hierarchy is output.

4. The excitation load grounding fault detection method based on EEMD and Hilbert spectrum analysis as described in claim 3, characterized in that: The temporal structure analysis of each component sequence identifies target component sequences containing fault mutation features, including... Construct a corresponding time variation curve for each component sequence; Compare the fluctuation trend, amplitude fluctuation and rate of change of each component within the target time interval; Component sequences with mutation characteristics are selected based on preset judgment rules; Components that meet the mutation characteristics are marked as target objects for subsequent frequency domain and time-frequency domain processing.

5. The excitation load grounding fault detection method based on EEMD and Hilbert spectrum analysis as described in claim 4, characterized in that: The step of performing frequency domain transformation on the target component sequence, constructing a spectral distribution through frequency mapping operations, and normalizing and quantizing the amplitude of each frequency component in the spectrum includes, The target component sequence is normalized for signal length, and a frequency index relationship corresponding to the time series is established. The amplitude information at each frequency point is used to calculate the magnitude and generate a one-to-one mapping table between frequency and amplitude. Perform a proportional conversion on the frequency amplitude results to construct a normalized amplitude sequence within a unified numerical range; The normalized amplitude sequence is used as a frequency distribution feature vector for subsequent fault type identification processing.

6. The excitation load grounding fault detection method based on EEMD and Hilbert spectrum analysis as described in claim 5, characterized in that: The step of performing a joint time-frequency transformation on the target component sequence to generate a time-frequency matrix of the component sequence, and calculating energy distribution data and energy entropy index based on the matrix, includes... A synchronous analytical operation is performed on the target component sequence to perform time-frequency joint transformation, and a two-dimensional data array describing the local variation characteristics of the signal is constructed. In the data array, the amplitude distribution information corresponding to each time segment is extracted according to the frequency direction, and the cumulative energy value of each frequency point is calculated. The cumulative energy value is expressed in a normalized distribution in the frequency domain to generate an energy distribution vector. Based on the proportion of each frequency component in the energy distribution vector, a statistical index value representing the degree of energy concentration is calculated as the energy entropy result. The time-frequency matrix, energy distribution vector, and energy entropy result are used as the input data set for subsequent judgment and processing.

7. The excitation load grounding fault detection method based on EEMD and Hilbert spectrum analysis as described in claim 6, characterized in that: The component sequences, spectral distribution data, time-frequency matrix, energy distribution data, and energy entropy index generated by the iterative decomposition process are used to perform fault diagnosis processing, and the diagnostic results, including fault type, fault channel, and fault start time, are output. Analyze the changing trends and persistence characteristics of key waveforms in the component sequence to determine their corresponding operating state categories; Comparing the amplitude differences of each characteristic frequency in the spectral distribution data is used to distinguish different types of electrical disturbance modes; Combining the frequency concentration of energy distribution with the trend of energy entropy changes helps to identify the duration and local impact characteristics of abnormal signals; By logically integrating various judgment criteria, a comprehensive diagnostic result is generated, which includes the nature of the fault, the corresponding phase, and the time of occurrence.

8. A ground fault detection system for excitation loads based on EEMD and Hilbert spectral analysis, employing the ground fault detection method for excitation loads based on EEMD and Hilbert spectral analysis as described in any one of claims 1 to 7, characterized in that, include: Data collection module, data decomposition module, data identification module, data mapping module, data calculation module, and result output module. The data collection module collects electrical signals during the operation of the equipment, preprocesses the collected signals, and constructs a discrete time-domain data sequence. The data decomposition module performs iterative decomposition processing on the data sequence to obtain several component quantity sequences; The data identification module performs time-domain structural analysis on the component sequence to identify the target component sequence containing fault mutation features. The data mapping module performs frequency domain transformation on the target component sequence, constructs a spectral distribution through frequency mapping operations, and normalizes and quantizes the amplitude of each frequency component in the spectrum. The data calculation module performs time-frequency joint transformation processing on the target component sequence to generate the time-frequency matrix of the component sequence, and calculates the energy distribution data and energy entropy index based on the matrix. The result output module performs fault judgment processing based on the component sequence, spectral distribution data, time-frequency matrix, energy distribution data, and energy entropy index generated by iterative decomposition processing, and outputs the diagnostic results of fault type, fault channel, and fault start time.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the excitation load grounding fault detection method based on EEMD and Hilbert spectrum analysis as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the excitation load grounding fault detection method based on EEMD and Hilbert spectrum analysis as described in any one of claims 1 to 7.

Images