Dynamic power supply fault monitoring system, method and equipment for multi-load branch distribution network, and medium

Abstract

The invention discloses a dynamic power supply fault monitoring system, method and device for a multi-load branch distribution network, and a medium, and the method comprises a first power supply signal collection module which is used for arranging a multi-branch sensor in the multi-load branch distribution network, and collecting a first power supply signal of a target power distribution branch; the second power supply signal acquisition module is used for performing disturbance desensitization processing on the first power supply signal through a signal disturbance desensitization mechanism to obtain a second power supply signal; the time domain mutation feature extraction module is used for extracting the time domain mutation feature of the second power supply signal; the frequency domain modal feature extraction module is used for extracting frequency domain modal features of the second power supply signal; and a fault monitoring result output module. According to the method, a signal disturbance desensitization mechanism is constructed, and a processing mode of combining frequency band weighting, sub-band filtering and time domain enhancement is adopted, so that interference signals generated by load switching, frequency conversion operation and the like can be effectively inhibited, and meanwhile, the characteristics of real fault signals are reserved and enhanced.

Description

Technical Field

[0001] This invention relates to the field of distribution network fault monitoring technology, and in particular to a dynamic power supply fault monitoring system, method, equipment and medium for multi-load distribution networks. Background Technology

[0002] With the continuous expansion of distribution network scale, the increasing diversification of load types, and the widespread access of distributed power sources, the operating environment of multi-load distributed distribution networks has become increasingly complex, leading to a continuous increase in the frequency of power supply faults and the scope of their external impact. However, in actual operation, frequent load switching, frequency conversion operations, and start-up and shutdown actions inevitably introduce certain current or voltage disturbances. The resulting signal timing waveforms, abrupt change slopes, and energy changes are highly similar to the waveforms of actual faults.

[0003] However, traditional fault detection methods struggle to delve into the complex characteristics of signals and effectively distinguish between the aforementioned disturbances and actual faults. When faced with interference signals from the parallel operation of multiple load branches, it is easy to misjudge the disturbances of normal loads as faults, thus causing malfunctions of protection devices. Or, when a real fault occurs, the corresponding signal characteristics are masked by the interference signals, resulting in a failure to operate. This not only seriously affects the reliability and stability of the power supply system but also wastes power resources, damages equipment, and even threatens electrical safety. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a dynamic power supply fault monitoring system, method, device, and medium for multi-load distributed power supply networks to solve the problems in the prior art, such as poor modal recognition capability and dynamic disturbance adaptability of complex signals, resulting in insufficient fault monitoring accuracy, high misjudgment rate, and unstable and unreliable power supply.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a dynamic power supply fault monitoring system for multi-load distributed power supply networks, comprising: The first power supply signal acquisition module is used to set up multi-branch sensors in the multi-load distribution network and acquire the first power supply signal of the target power distribution branch. The second power supply signal acquisition module is used to perform perturbation and desensitization processing on the first power supply signal through a signal perturbation and desensitization mechanism to obtain the second power supply signal; A time-domain abrupt change feature extraction module is used to extract the time-domain abrupt change features of the second power supply signal; A frequency domain modal feature extraction module is used to extract the frequency domain modal features of the second power supply signal; The fault monitoring result output module is used to input the time-domain abrupt change features and the frequency-domain modal features as a time-frequency dual-domain feature set into the dual-domain discrimination model for analysis, analyze the time-domain abrupt change features based on the discrimination results, and output the fault monitoring results.

[0007] As a preferred embodiment of the dynamic power supply fault monitoring system for multi-load distributed power grids described in this invention, the steps performed by the second power supply signal acquisition module include: constructing a sliding window disturbance sample library; performing waveform similarity matching between the first power supply signal and the sliding window disturbance sample library through a signal disturbance desensitization mechanism to obtain waveform similarity; when the waveform similarity is higher than a set threshold, activating the signal disturbance desensitization mechanism to perform disturbance desensitization processing on the first power supply signal to obtain the second power supply signal.

[0008] As a preferred embodiment of the dynamic power supply fault monitoring system for multi-load distributed distribution networks described in this invention, the signal disturbance desensitization mechanism performs the following steps: the disturbance desensitization processing includes frequency band weighting, sub-band filtering, and time-domain enhancement; wherein, the frequency band weighting involves identifying the disturbance frequency band of the first power supply signal and setting a weight factor to suppress the disturbance frequency band; the sub-band filtering involves identifying the disturbance sub-band of the first power supply signal and filtering the disturbance sub-band; and the time-domain enhancement involves identifying the time-domain abrupt change characteristics of the first power supply signal and enhancing the slope of the time-domain abrupt change characteristics.

[0009] The beneficial effects of this preferred technical solution are as follows: through the triple processing mechanism of frequency band weighting, sub-band filtering and time domain enhancement, it can suppress disturbance interference in a targeted manner, while retaining or even enhancing fault characteristic signals, thereby improving the distinguishability between fault signals and disturbance signals.

[0010] As a preferred embodiment of the dynamic power supply fault monitoring system for multi-load distributed power supply networks described in this invention, the frequency domain modal feature extraction module performs the following steps: decomposing the second power supply signal using a preset layer wavelet packet to obtain multiple sub-bands, wherein the preset layer is X; calculating the energy of the multiple sub-bands to form a frequency band energy feature vector; outputting the frequency band energy feature vector as wavelet packet decomposition frequency domain features; performing a Fourier transform on the second power supply signal to obtain spectral distribution data; and outputting the peak frequency, spectral centroid, and spectral amplitude range of the spectral distribution data as Fourier transform frequency domain features.

[0011] The beneficial effects of this preferred technical solution are as follows: By combining wavelet packet decomposition and Fourier transform, two complementary frequency domain analysis methods, wavelet packets provide multi-scale time-frequency localization features, while Fourier transform provides global spectral distribution features. The fusion of the two can comprehensively characterize the frequency domain modal characteristics of the power supply signal and improve the accuracy of fault identification.

[0012] As a preferred embodiment of the dynamic power supply fault monitoring system for multi-load distributed power supply networks described in this invention, the steps performed by the fault monitoring result output module include: initializing the model architecture; collecting disturbance sample power supply signals labeled with disturbance events and fault sample power supply signals labeled with fault events in the multi-load distributed power supply system at a preset sampling period; combining the disturbance sample power supply signals and the fault sample power supply signals to obtain time-frequency dual-domain feature training samples; training the model architecture based on the time-frequency dual-domain feature training samples to obtain a trained dual-domain discrimination model; and deploying the trained dual-domain discrimination model to the monitoring master station of the multi-load distributed power supply system.

[0013] As a preferred embodiment of the dynamic power supply fault monitoring system for multi-load distributed power grids described in this invention, the steps executed by the fault monitoring result output module further include: the dual-domain discrimination model calculating the modal distance index based on time-domain abrupt change characteristics and frequency-domain modal characteristics. Set the perturbation mode threshold; when the modal distance index When the disturbance mode threshold is greater than the specified threshold, a fault event discrimination result is output, and a fault index is calculated based on the temporal abrupt change characteristics. A fault alarm signal is generated based on the magnitude of the fault index. When the modal distance index... If the disturbance event threshold is not greater than the specified disturbance mode threshold, output the disturbance event discrimination result.

[0014] The beneficial effects of this preferred technical solution are as follows: by introducing the modal distance index as a quantitative discrimination criterion, the automatic and quantitative distinction between disturbance events and fault events is realized, avoiding the drawbacks of traditional methods that rely on human experience for judgment, and at the same time, the severity of faults is graded and alarmed through fault index calculation.

[0015] As a preferred embodiment of the dynamic power supply fault monitoring system for multi-load distributed power grids described in this invention, the following is included: calculating the modal distance index. The steps include: ; in, The wavelet packet decomposition frequency domain features of the current power supply signal, The wavelet packet decomposition frequency domain feature samples stored in the time-frequency dual-domain feature training samples, The Fourier transform frequency domain characteristics of the current power supply signal. These are the Fourier transform frequency domain feature samples stored in the time-frequency dual-domain feature training samples. is a norm used to characterize distance metrics in the feature space.

[0016] The beneficial effects of this preferred technical solution are as follows: by superimposing the wavelet packet decomposition feature distance and the Fourier transform feature distance with norms, a quantitative index that comprehensively considers the degree of time-frequency dual-domain offset is constructed, so that fault identification has a clear mathematical basis and reproducibility.

[0017] Secondly, the present invention provides a dynamic power supply fault monitoring method for multi-load distribution networks, comprising the following steps: setting up multi-branch sensors in the multi-load distribution network to collect the first power supply signal of the target distribution branch; performing perturbation desensitization processing on the first power supply signal through a signal perturbation desensitization mechanism to obtain a second power supply signal; extracting the time-domain abrupt change features of the second power supply signal; extracting the frequency-domain modal features of the second power supply signal; inputting the time-domain abrupt change features and the frequency-domain modal features as a time-frequency dual-domain feature set into a dual-domain discrimination model for analysis; analyzing the time-domain abrupt change features based on the discrimination result; and outputting the fault monitoring result.

[0018] Thirdly, the present invention provides an electronic device, comprising: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, they implement the steps of a dynamic power supply fault monitoring system for multi-load distributed power grids.

[0019] Fourthly, the present invention provides a computer-readable storage medium storing computer-executable instructions that, when executed by a processor, implement the steps of the dynamic power supply fault monitoring system for multi-load distributed power grids.

[0020] Compared with the prior art, the beneficial effects of the present invention are as follows: By constructing a signal disturbance desensitization mechanism, the present invention adopts a processing method that combines frequency band weighting, sub-band filtering and time domain enhancement, which can effectively suppress interference signals generated by load switching, frequency conversion operation, etc., while retaining and enhancing the characteristics of real fault signals. This solves the technical problem that traditional detection methods have difficulty distinguishing between disturbance signals and fault signals, significantly reduces the probability of protection device malfunction and failure to operate, and improves the accuracy and reliability of multi-load distributed power grid power supply fault monitoring.

[0021] This invention extracts the time-domain abrupt change features and frequency-domain modal features of the power supply signal, constructs a time-frequency dual-domain feature set, and combines it with a dual-domain discrimination model for comprehensive analysis. It uses the modal distance index to quantitatively distinguish between disturbance events and fault events. Compared with detection methods based on a single feature dimension, it can more comprehensively characterize the abnormal characteristics of the power supply signal, effectively avoid the masking effect of interference signals on fault identification in complex operating environments, and ensure the stable operation of the power supply system and the safety of electricity use. Attached Figure Description

[0022] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the 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.

[0023] Figure 1 This is a schematic diagram of the overall process of a dynamic power supply fault monitoring system for multi-load distributed power grids according to an embodiment of the present invention. Detailed Implementation

[0024] 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.

[0025] Example 1, referring to Figure 1 As one embodiment of the present invention, a dynamic power supply fault monitoring system for multi-load distributed power grids is provided, comprising: The first power supply signal acquisition module is used to set up multi-branch sensors in a multi-load distribution network to acquire the first power supply signal of the target distribution branch.

[0026] The second power supply signal acquisition module is used to perform perturbation and desensitization processing on the first power supply signal through a signal perturbation and desensitization mechanism to obtain the second power supply signal.

[0027] The temporal abrupt change feature extraction module is used to extract the temporal abrupt change features of the second power supply signal.

[0028] The frequency domain modal feature extraction module is used to extract the frequency domain modal features of the second power supply signal.

[0029] The fault monitoring result output module is used to input the time-domain abrupt change features and the frequency-domain modal features as a time-frequency dual-domain feature set into the dual-domain discrimination model for analysis, analyze the time-domain abrupt change features based on the discrimination results, and output the fault monitoring results.

[0030] It should be noted that with the continuous expansion of distribution network scale, the increasing diversification of load types, and the widespread integration of distributed power sources, the operating environment of multi-load distributed distribution networks has become increasingly complex, and the frequency and scope of power supply faults have also increased. In actual operation, frequent load switching, frequency conversion operations, and start-up and shutdown actions inevitably introduce current or voltage disturbances, and the resulting signal timing waveforms, abrupt change slopes, and energy changes are highly similar to actual fault waveforms. Traditional fault detection methods are difficult to deeply explore the complex characteristics of signals, and it is difficult to effectively distinguish between disturbances and real faults. They are prone to misjudging disturbances of normal loads as faults, causing protection devices to malfunction, or failing to operate when a real fault occurs because the signal characteristics are masked by interference signals, seriously affecting the reliability and stability of the power supply system.

[0031] Therefore, to address the aforementioned issues of disturbance and fault identification, the following approach is adopted: First, the original power supply signals of the multi-load distribution network are collected. The disturbance signals are effectively suppressed through a signal disturbance desensitization mechanism, while preserving and enhancing the true fault characteristics. Then, the abrupt change features and modal features of the signals are extracted from both the time and frequency domains to construct a time-frequency dual-domain feature set. Finally, the dual-domain discrimination model is used to comprehensively analyze the features, enabling accurate differentiation between disturbance events and fault events, effectively reducing the probability of malfunctions and failures to operate, and ensuring the safe and stable operation of the multi-load distribution network.

[0032] Example 2, refer to Figure 1 As an embodiment of the present invention, based on the above embodiment, a dynamic power supply fault monitoring system for multi-load distributed power supply networks is provided.

[0033] In this embodiment of the application, the first power supply signal acquisition module is used to set up multi-branch sensors in the multi-load distribution network to acquire the first power supply signal of the target distribution branch.

[0034] Specifically, current transformers, voltage transformers, and other multi-branch sensors are deployed at each distribution branch node of the multi-load distribution network. The current and voltage signals of the target distribution branch are synchronously collected at a preset sampling frequency to obtain the first power supply signal containing features such as time-series waveforms, amplitude changes, and phase information. The sampling frequency is set according to the operation characteristics of the distribution network, and is usually selected as an integer multiple of the power frequency to ensure signal integrity. At the same time, the collected raw signals are converted from analog to digital and timestamped to facilitate subsequent signal processing and feature analysis.

[0035] In this embodiment of the application, the second power supply signal acquisition module is used to perform perturbation and desensitization processing on the first power supply signal through a signal perturbation and desensitization mechanism to obtain the second power supply signal.

[0036] The steps performed by the second power supply signal acquisition module include A1 to A3: A1. Construct a sliding window perturbation sample library.

[0037] Typical disturbance events in historical operating data are extracted using a preset time window, including signal waveforms of common disturbance types such as load switching, inverter start-stop, motor switching, and capacitor bank switching. A sliding window mechanism is used to segment and store the disturbance signals, with each sample containing annotation information such as the timing characteristics, amplitude variation range, and duration of the disturbance waveform. By continuously accumulating and updating new disturbance samples identified during operation, a dynamic sample library covering multiple disturbance types is formed, providing a comparison benchmark for subsequent waveform similarity matching.

[0038] A2. The waveform similarity of the first power supply signal is obtained by matching the waveform with the sliding window perturbation sample library through the signal perturbation desensitization mechanism.

[0039] The waveform of the first power supply signal acquired in real time is compared with the disturbance sample in the sliding window disturbance sample library. The waveform similarity between the two is calculated by using the dynamic time warping algorithm or the Pearson correlation coefficient method. When an abnormal change is detected in the first power supply signal, the disturbance template in the sample library is traversed for matching, and the highest similarity value and the corresponding disturbance type label are output.

[0040] The signal perturbation desensitization mechanism performs perturbation desensitization processing including frequency band weighting, subband filtering, and time domain enhancement; The frequency band weighting involves identifying the disturbance frequency band of the first power supply signal and setting a weighting factor to suppress the disturbance frequency band.

[0041] Specifically, a spectrum analysis is performed on the first power supply signal to identify the frequency band range where the disturbance energy is concentrated; based on the characteristics of the disturbance type, different weighting factors are set for different frequency bands, with a suppression weight of less than 1 set for the disturbance frequency band and a retention or enhancement weight of greater than or equal to 1 set for the fault characteristic frequency band; the energy proportion of the disturbance frequency band is reduced through weighted processing to weaken the interference effect of the disturbance signal on fault identification.

[0042] The sub-band filtering is used to identify the disturbance sub-bands of the first power supply signal and to filter the disturbance sub-bands.

[0043] Specifically, the first power supply signal is decomposed into multiple frequency sub-bands based on wavelet decomposition or bandpass filter banks. Combined with the disturbance frequency characteristics marked in the disturbance sample library, the sub-band intervals where the disturbance energy is concentrated are located. The identified disturbance sub-bands are filtered out in a targeted manner using adaptive notch filtering or band-stop filtering. While suppressing the disturbance components, the effective signal components in other sub-bands are preserved, avoiding the loss of fault feature information.

[0044] The time-domain enhancement involves identifying the time-domain abrupt change features of the first power supply signal and performing slope enhancement processing on the time-domain abrupt change features.

[0045] Specifically, time-domain analysis is performed on the signal after frequency band weighting and subband filtering to detect the location and amplitude of abrupt changes in the signal waveform; differential operations or slope detection algorithms are used to extract the slope features of the abrupt change region, and the slope of the abrupt change that matches the fault characteristics is enhanced and amplified to improve the contrast between the fault signal and the background signal; the time-domain enhancement process makes the signal abrupt changes caused by the real fault more significant, which facilitates the subsequent extraction and identification of fault features.

[0046] A3. When the waveform similarity is higher than the set threshold, the signal perturbation desensitization mechanism is activated to perform perturbation desensitization processing on the first power supply signal to obtain the second power supply signal.

[0047] Specifically, the waveform similarity calculated in step A2 is compared with a preset similarity threshold. When the similarity is higher than the threshold, it is determined that there is a disturbance component in the first power supply signal. The signal disturbance desensitization mechanism is activated to sequentially perform frequency band weighting, sub-band filtering and time domain enhancement processing, and output the second power supply signal after disturbance suppression and fault enhancement. When the similarity is lower than the threshold, it is determined that there is no obvious disturbance interference in the first power supply signal. The first power supply signal is directly output as the second power supply signal to avoid unnecessary signal processing that causes feature distortion.

[0048] In one optional implementation, the sliding window disturbance sample library constructed in step A1 can also employ an online learning and incremental update mechanism. By deploying edge computing nodes to perform real-time analysis of distribution network operation data, when a new disturbance event is detected, its waveform features are automatically extracted and added to the sample library after manual confirmation. Simultaneously, a sample aging mechanism is set up to clean up outdated samples that have not been matched for a long time, maintaining the timeliness and relevance of the sample library and improving its adaptability to changes in the distribution network operation environment.

[0049] In another optional implementation, the waveform similarity matching in step A2 can also employ a feature matching method based on deep learning. By using convolutional neural networks or long short-term memory networks to encode the features of the disturbance samples, waveform similarity matching is transformed into a distance metric in the feature space. This method can automatically learn the deep feature representation of the disturbance signal, exhibiting stronger robustness to waveform distortion and noise interference, and is particularly suitable for distribution network operation scenarios with complex and varied disturbance types.

[0050] In this embodiment of the application, the time-domain abrupt change feature extraction module is used to extract the time-domain abrupt change features of the second power supply signal.

[0051] Specifically, time-domain analysis is performed on the second power supply signal. A sliding window is used to traverse the signal sequence, and the instantaneous rate of change of the signal amplitude is detected through first-order difference operations to identify the start and end points of abrupt changes in the signal waveform. Within the abrupt change interval, characteristic parameters such as abrupt change amplitude, abrupt change slope, abrupt change duration, and overshoot are extracted. The abrupt change amplitude characterizes the degree of signal deviation caused by the fault, the abrupt change slope reflects the severity of the fault development, and the abrupt change duration indicates the length of the transient process of the fault. These parameters are combined to form a time-domain abrupt change feature vector, which serves as an important basis for fault identification. In an optional implementation, the time-domain abrupt change feature extraction module can also employ an adaptive feature extraction method based on empirical mode decomposition (EMD). EMD decomposes the second power supply signal into multiple intrinsic mode function (IMF) components, each representing the signal's oscillation mode at different time scales. The instantaneous amplitude and instantaneous frequency of each IMF component are calculated, and the energy mutation rate and frequency drift of each component at the abrupt change moment are extracted. This method does not require preset basis functions and can adaptively capture the time-domain abrupt change characteristics of nonlinear and non-stationary signals, making it particularly suitable for fault feature extraction under complex disturbance backgrounds.

[0052] In this embodiment of the application, the frequency domain modal feature extraction module is used to extract the frequency domain modal features of the second power supply signal.

[0053] The frequency domain modal feature extraction module performs steps B1 to B4: B1. The second power supply signal is decomposed by wavelet packet decomposition of a preset layer to obtain multiple sub-bands, wherein the preset layer is X.

[0054] Specifically, wavelet basis functions with good time-frequency localization characteristics are selected to perform X-level wavelet packet decomposition on the second power supply signal, decomposing the signal into... Each frequency sub-band; compared with traditional wavelet decomposition, wavelet packet decomposition can further subdivide the high-frequency part and obtain a finer frequency resolution; the preset number of layers X is determined according to the frequency characteristics of the signal and the distribution range of fault characteristics. In this embodiment, three layers are selected to ensure both frequency resolution and computational efficiency; the sub-band signals obtained by decomposition retain the time-domain waveform information of the original signal in the corresponding frequency band.

[0055] B2. Calculate the energy of the multiple sub-bands, form a frequency band energy feature vector, and output the frequency band energy feature vector as wavelet packet decomposition frequency domain features.

[0056] The energy value of each sub-band signal obtained from wavelet packet decomposition is calculated using the sum of squares of the sub-band signal amplitudes. The energy of each sub-band is then normalized to eliminate the influence of signal amplitude differences on the features. The normalized sub-band energies are then arranged in frequency order, forming a dimensional array. The frequency band energy feature vector reflects the distribution of signal energy in different frequency bands. Different types of faults and disturbances have different energy distribution patterns, which can be used as effective features for fault identification.

[0057] B3. Perform a Fourier transform on the second power supply signal to obtain the spectral distribution data.

[0058] Specifically, the Fast Fourier Transform algorithm is applied to the second power supply signal to convert the time-domain signal into a frequency-domain representation and obtain the signal's spectral distribution data. The spectral data contains the amplitude and phase information of each frequency component. By calculating the amplitude spectrum, the frequency components of the signal and their intensity distribution can be displayed intuitively. In order to improve the spectral resolution and reduce spectral leakage, the signal is preprocessed by adding a Hanning window or a Hamming window before the Fourier transform, and the zero-filling technique is used to increase the number of interpolation points in the spectrum.

[0059] B4. Output the peak frequency, spectral centroid, and spectral amplitude range of the spectral distribution data as Fourier transform frequency domain features.

[0060] Specifically, three key feature parameters are extracted from the spectral distribution data: peak frequency, which is the frequency value corresponding to the maximum peak in the amplitude spectrum, representing the dominant frequency component of the signal. When a fault occurs, the dominant frequency often shifts or a new peak appears; centroid, which is the frequency-weighted average value calculated with the amplitude of each frequency component as the weight, reflects the overall distribution position of the signal's spectral energy. Spectral changes caused by a fault will cause the centroid to drift; and amplitude range, which is the difference between the maximum and minimum values ​​in the amplitude spectrum, representing the dynamic range and energy concentration of the spectrum. The above three parameters are combined to form a Fourier transform frequency domain feature vector, which together with the wavelet packet decomposition frequency domain features constitutes a complete frequency domain modal feature.

[0061] In an optional implementation, the wavelet packet decomposition in step B1 can also employ an adaptive wavelet basis selection method. Based on the time-frequency characteristics of the second power supply signal, the wavelet basis that best matches the signal characteristics is automatically selected from the wavelet basis function library. The selection criteria can adopt the principle of minimizing entropy or the principle of maximizing energy concentration. Adaptive wavelet basis selection can improve the adaptability of wavelet packet decomposition to different types of fault signals and obtain better time-frequency decomposition results.

[0062] In another optional implementation, the Fourier transform frequency domain feature extraction in steps B3 and B4 can also employ the short-time Fourier transform method. By sliding a fixed-length analysis window along the time axis, a Fourier transform is performed on the signal within each window to obtain the joint time-frequency distribution map of the signal. Instantaneous spectral features at each moment, including parameters such as instantaneous dominant frequency, instantaneous bandwidth, and spectral entropy, are extracted from the time-frequency map. This method can capture the dynamic evolution of frequency domain features during fault development and is particularly suitable for feature analysis of non-stationary fault signals.

[0063] In this embodiment of the application, the fault monitoring result output module is used to input the time-domain abrupt change feature and the frequency-domain modal feature as a time-frequency dual-domain feature set into the dual-domain discrimination model for analysis, analyze the time-domain abrupt change feature according to the discrimination result, and output the fault monitoring result.

[0064] The steps executed by the fault monitoring result output module include C1 to C5: C1. Initialize the model architecture.

[0065] Specifically, the basic architecture of the dual-domain discrimination model is constructed. The model includes a feature input layer, a feature fusion layer, a distance calculation layer, and a discrimination output layer. The feature input layer receives temporal abrupt change feature vectors and frequency domain modal feature vectors, respectively. The feature fusion layer concatenates or weights the two types of features to form a time-frequency dual-domain feature set. The distance calculation layer is used to calculate the modal distance between the current signal features and the reference sample features. The discrimination output layer outputs the discrimination result of disturbance or fault based on the distance threshold. The model parameters are initialized, and hyperparameters such as feature dimension, distance measurement method, and initial threshold value are set to prepare for subsequent training.

[0066] C2. Collect the disturbance sample power supply signal marked with disturbance events and the fault sample power supply signal marked with fault events in the multi-load distributed power supply system at a preset sampling period.

[0067] Specifically, during the actual operation of a multi-load distributed power supply system, power supply signal data is continuously collected at a preset sampling period. The collected signals are labeled with events by maintenance personnel or the automation system. Signal changes caused by normal operations such as load switching, frequency converter start-up and shutdown, and capacitor switching are labeled as disturbance events, while signal changes caused by abnormal conditions such as short circuits, grounding, open circuits, and overloads are labeled as fault events. Disturbance sample datasets and fault sample datasets are established separately. Each sample contains the original power supply signal waveform and corresponding event type label, occurrence time, duration, and other metadata.

[0068] C3. Combine the power supply signal of the disturbance sample and the power supply signal of the fault sample to obtain the time-frequency dual-domain feature training sample.

[0069] Specifically, feature extraction is performed on all sample signals in the disturbance sample dataset and the fault sample dataset, extracting the temporal abrupt change features and frequency domain modal features for each sample respectively; the temporal abrupt change features and frequency domain modal features are combined to form a time-frequency dual-domain feature vector for each sample, and the corresponding event type label is retained; the disturbance samples and fault samples are balanced to ensure that the number of samples of the two classes is equal, avoiding the impact of class imbalance on model training; finally, a time-frequency dual-domain feature training sample set containing feature vectors and labels is formed.

[0070] C4. Train the model architecture based on the time-frequency dual-domain feature training samples to obtain the trained dual-domain discrimination model.

[0071] Specifically, the time-frequency dual-domain feature training sample set is divided into a training set and a validation set according to a certain ratio. The training set is used to optimize the model parameters, and the validation set is used to evaluate the model performance and prevent overfitting. During the training process, the model parameters are adjusted by minimizing the classification loss function so that the model can accurately distinguish the feature patterns of perturbed samples and faulty samples. After the training is completed, the model performance is evaluated, and indicators such as accuracy, recall, and false positive rate are calculated to ensure that the model meets the requirements of practical applications.

[0072] The discrimination process of the domain discriminant model includes C4.1 to C4.4: C4.1 The dual-domain discriminant model calculates the modal distance index based on temporal abrupt change characteristics and frequency domain modal characteristics. Calculate the modal distance index. The steps include: ; in, The wavelet packet decomposition frequency domain features of the current power supply signal, The wavelet packet decomposition frequency domain feature samples stored in the time-frequency dual-domain feature training samples, The Fourier transform frequency domain characteristics of the current power supply signal. These are the Fourier transform frequency domain feature samples stored in the time-frequency dual-domain feature training samples. is a norm used to characterize distance metrics in the feature space.

[0073] C4.2 Set the perturbation mode threshold.

[0074] Specifically, based on the modal distance distribution of perturbed and faulty samples in the training sample set, a statistical analysis method is used to determine the perturbed modal threshold. The modal distance index distribution of all perturbed samples is calculated, and its mean plus a certain number of standard deviations is taken as the upper limit of the threshold. At the same time, the modal distance index distribution of all faulty samples is calculated, and its mean minus a certain number of standard deviations is taken as the lower limit of the threshold. The value that minimizes the classification error rate between the upper and lower limits is selected as the final perturbed modal threshold. The threshold setting needs to achieve a balance between the false alarm rate and the false negative rate, and can be adjusted according to the security requirements of the actual application scenario.

[0075] C4.3, when the modal distance index When the value exceeds the perturbation mode threshold, the fault event discrimination result is output, and the fault index is calculated for the time-domain abrupt change feature. A fault alarm signal is generated based on the magnitude of the fault index.

[0076] Specifically, when the calculated modal distance index exceeds the disturbance mode threshold, the current signal is determined to be a fault event, and the fault analysis process begins. Fault indexes are calculated for the time-domain abrupt change characteristics. The fault indexes comprehensively consider parameters such as the abrupt change amplitude, abrupt change slope, and abrupt change duration. A weighted summation or fuzzy comprehensive evaluation method is used to obtain a quantitative value of the fault severity. Fault levels are classified according to the magnitude of the fault indexes, and multi-level alarm thresholds are set. When the fault index is small, a general alarm signal is generated, and when the fault index is large, a severe alarm signal or an emergency alarm signal is generated. The alarm signal contains information such as fault type, fault location, fault level, and occurrence time, and is pushed to the monitoring master station and the operation and maintenance terminal.

[0077] C4.4, when the modal distance index If the disturbance event threshold is not greater than the specified disturbance mode threshold, output the disturbance event discrimination result.

[0078] When the calculated modal distance index does not exceed the disturbance mode threshold, the current signal change is determined to be a normal disturbance event, and the disturbance event discrimination result is output. The system records the characteristic information of the disturbance event, including disturbance type, occurrence time, duration, waveform characteristics, etc., for updating the sliding window disturbance sample library. Disturbance events do not trigger fault alarms, but disturbance logs can be generated for operation and maintenance personnel to view and analyze, which is convenient for understanding the distribution network operation status and load change patterns.

[0079] C5. Deploy the trained dual-domain discrimination model to the monitoring master station of the multi-load distributed power supply system.

[0080] Specifically, the trained and validated dual-domain discriminant model is solidified and compressed for optimization, and exported as a deployable model file. The model runtime environment is configured on the monitoring master station of the multi-load distributed power supply system, and the model file is loaded into the master station's computing unit. An interface is established between the model and the data acquisition system to achieve automatic input of real-time power supply signals and automatic output of discrimination results. An alarm push channel is configured to send fault monitoring results to the dispatch center, operation and maintenance terminals, and related protection devices via the network. Simultaneously, a model update interface is reserved to support online or offline iterative optimization of the model based on newly added sample data.

[0081] In an optional implementation, the dual-domain discrimination model in step C4 can also employ an end-to-end classification architecture based on deep learning. A deep neural network comprising convolutional layers, pooling layers, and fully connected layers is constructed. Temporal abrupt changes and frequency modal features are used as dual-channel inputs. The convolutional layers automatically learn deep representations of the features, and the fully connected layers achieve binary classification outputs for perturbations and faults. This method eliminates the need for manually designing distance metric formulas and thresholds, automatically learns the optimal discrimination boundary from the data, and has a stronger expressive power for complex nonlinear feature patterns.

[0082] In summary, this invention constructs a signal disturbance desensitization mechanism and adopts a processing method that combines frequency band weighting, sub-band filtering, and time domain enhancement. This effectively suppresses interference signals generated by load switching, frequency conversion operations, etc., while retaining and enhancing the characteristics of real fault signals. It solves the technical problem that traditional detection methods cannot distinguish between disturbance signals and fault signals, significantly reduces the probability of protection device malfunctions and failures to operate, and improves the accuracy and reliability of multi-load distributed power grid power supply fault monitoring.

[0083] This invention extracts the time-domain abrupt change features and frequency-domain modal features of the power supply signal, constructs a time-frequency dual-domain feature set, and combines it with a dual-domain discrimination model for comprehensive analysis. It uses the modal distance index to quantitatively distinguish between disturbance events and fault events. Compared with detection methods based on a single feature dimension, it can more comprehensively characterize the abnormal characteristics of the power supply signal, effectively avoid the masking effect of interference signals on fault identification in complex operating environments, and ensure the stable operation of the power supply system and the safety of electricity use.

[0084] Example 3 illustrates a schematic scheme for a dynamic power supply fault monitoring system for multi-load distributed power grids. It should be noted that the technical solution of this method for dynamic power supply fault monitoring for multi-load distributed power grids is based on the same concept as the technical solution of the aforementioned dynamic power supply fault monitoring system for multi-load distributed power grids. Details not described in detail in the technical solution of the method for dynamic power supply fault monitoring for multi-load distributed power grids in this embodiment can be found in the description of the technical solution of the aforementioned dynamic power supply fault monitoring system for multi-load distributed power grids.

[0085] This embodiment also provides a dynamic power supply fault monitoring method for multi-load distributed power supply networks, including the following steps: Multiple branch sensors are set up in the multi-load distribution network to collect the first power supply signal of the target power distribution branch; The first power supply signal is perturbed and desensitized using a signal perturbation and desensitization mechanism to obtain the second power supply signal; Extract the temporal abrupt change features of the second power supply signal; Extract the frequency domain modal features of the second power supply signal; The time-domain abrupt change features and the frequency-domain modal features are input into the dual-domain discrimination model as a time-frequency dual-domain feature set for analysis. The time-domain abrupt change features are analyzed based on the discrimination results, and the fault monitoring results are output.

[0086] This embodiment also provides an electronic device suitable for dynamic power supply fault monitoring in multi-load distributed power supply networks, including: 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 realize the dynamic power supply fault monitoring system for multi-load distributed power supply networks as proposed in the above embodiment.

[0087] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements a dynamic power supply fault monitoring system for multi-load distributed power grids as proposed in the above embodiments.

[0088] The storage medium proposed in this embodiment and the method for dynamic power supply fault monitoring for multi-load distributed control networks proposed in the above embodiments belong to the same inventive concept. 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.

[0089] 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. 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.

[0090] 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 dynamic power supply fault monitoring system for multi-load distributed power grids, characterized in that, include: The first power supply signal acquisition module is used to set up multi-branch sensors in the multi-load distribution network and acquire the first power supply signal of the target power distribution branch. The second power supply signal acquisition module is used to perform perturbation and desensitization processing on the first power supply signal through a signal perturbation and desensitization mechanism to obtain the second power supply signal; A time-domain abrupt change feature extraction module is used to extract the time-domain abrupt change features of the second power supply signal; A frequency domain modal feature extraction module is used to extract the frequency domain modal features of the second power supply signal; The fault monitoring result output module is used to input the time-domain abrupt change features and the frequency-domain modal features as a time-frequency dual-domain feature set into the dual-domain discrimination model for analysis, analyze the time-domain abrupt change features based on the discrimination results, and output the fault monitoring results.

2. The dynamic power supply fault monitoring system for multi-load distributed power grids as described in claim 1, characterized in that, The steps performed by the second power supply signal acquisition module include: Construct a sliding window perturbation sample library; The waveform similarity of the first power supply signal is obtained by matching the waveform similarity with the sliding window perturbation sample library through a signal perturbation desensitization mechanism. When the waveform similarity is higher than a set threshold, the signal perturbation desensitization mechanism is activated to perform perturbation desensitization processing on the first power supply signal to obtain the second power supply signal.

3. The dynamic power supply fault monitoring system for multi-load distributed power grids as described in claim 2, characterized in that, The steps performed by the signal perturbation desensitization mechanism include: The signal perturbation desensitization mechanism performs perturbation desensitization processing including frequency band weighting, subband filtering, and time domain enhancement; The frequency band weighting involves identifying the disturbance frequency band of the first power supply signal and setting a weighting factor to suppress the disturbance frequency band. The sub-band filtering is used to identify the disturbance sub-bands of the first power supply signal and to filter the disturbance sub-bands. The time-domain enhancement involves identifying the time-domain abrupt change features of the first power supply signal and performing slope enhancement processing on the time-domain abrupt change features.

4. The dynamic power supply fault monitoring system for multi-load distributed power grids as described in claim 3, characterized in that, The steps performed by the frequency domain modal feature extraction module include: The second power supply signal is decomposed by wavelet packet decomposition of a preset layer to obtain multiple sub-bands, wherein the preset layer is X; Calculate the energy of the multiple sub-bands, form a frequency band energy feature vector, and output the frequency band energy feature vector as wavelet packet decomposition frequency domain features; Perform a Fourier transform on the second power supply signal to obtain the spectral distribution data; The peak frequency, spectral centroid, and spectral amplitude range of the spectral distribution data are output as Fourier transform frequency domain features.

5. The dynamic power supply fault monitoring system for multi-load distributed power grids as described in claim 4, characterized in that, The steps performed by the fault monitoring result output module include: Initialize the model architecture; The power supply signals of disturbance samples marked with disturbance events and fault samples marked with fault events in a multi-load distributed power supply system are collected at a preset sampling period. By combining the power supply signals of the disturbed samples and the power supply signals of the fault samples, time-frequency dual-domain feature training samples are obtained. The model architecture is trained based on the time-frequency dual-domain feature training samples to obtain a trained dual-domain discrimination model; The trained dual-domain discrimination model is deployed to the monitoring master station of the multi-load distributed power supply system.

6. The dynamic power supply fault monitoring system for multi-load distributed power grids as described in claim 5, characterized in that, The steps performed by the fault monitoring result output module also include: The dual-domain discriminant model calculates the modal distance index based on temporal abrupt change characteristics and frequency domain modal characteristics. ; Set the perturbation mode threshold; When the modal distance index When the value exceeds the perturbation mode threshold, the fault event discrimination result is output, and the fault index is calculated for the time-domain abrupt change feature. A fault alarm signal is generated based on the magnitude of the fault index. When the modal distance index If the disturbance event threshold is not greater than the specified disturbance mode threshold, output the disturbance event discrimination result.

7. The dynamic power supply fault monitoring system for multi-load distributed power grids as described in claim 6, characterized in that, Calculate the modal distance index The steps include: ； in, The wavelet packet decomposition frequency domain features of the current power supply signal, The wavelet packet decomposition frequency domain feature samples stored in the time-frequency dual-domain feature training samples, The Fourier transform frequency domain characteristics of the current power supply signal. These are the Fourier transform frequency domain feature samples stored in the time-frequency dual-domain feature training samples. is a norm used to characterize distance metrics in the feature space.

8. A dynamic power supply fault monitoring method for multi-load distributed power grids, using the system described in any one of claims 1-7, characterized in that, Includes the following steps: Multiple branch sensors are set up in the multi-load distribution network to collect the first power supply signal of the target power distribution branch; The first power supply signal is perturbed and desensitized using a signal perturbation and desensitization mechanism to obtain the second power supply signal; Extract the temporal abrupt change features of the second power supply signal; Extract the frequency domain modal features of the second power supply signal; The time-domain abrupt change features and the frequency-domain modal features are input into the dual-domain discrimination model as a time-frequency dual-domain feature set for analysis. The time-domain abrupt change features are analyzed based on the discrimination results, and the fault monitoring results are output.

9. An electronic device, comprising: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, they implement the steps of the dynamic power supply fault monitoring system for multi-load distributed power grids as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing computer-executable instructions that, when executed by a processor, implement the steps of the dynamic power supply fault monitoring system for multi-load distributed power grids as described in any one of claims 1 to 7.

Images