A low-voltage power distribution network fault detection method and system

Abstract

The application relates to the technical field of low-voltage power distribution network fault detection, and discloses a low-voltage power distribution network fault detection method and system.The method comprises the following steps: acquiring current signal data and preprocessing the current signal data to obtain an original data set containing time dimensions and space dimensions; extracting time transient fluctuation features based on the original data set and performing spatial comparative analysis to obtain spatial distribution feature data; constructing and correcting a long-term change trend sequence according to the spatial distribution feature data; determining an abnormal area according to the correction result; performing residual error analysis and spatial difference screening on the signals corresponding to the abnormal area to determine the specific position of a fault area; generating an alarm signal according to the specific position of the fault area and recording spatial distribution difference data to obtain an abnormal detection result of the micro branch current of the low-voltage power distribution network.The method can realize early high-sensitivity identification and accurate spatial positioning of weak and concealed current abnormalities in the micro branch.

Description

Technical Field

[0001] This invention relates to the field of low-voltage distribution network fault detection technology, and in particular to a low-voltage distribution network fault detection method and system. Background Technology

[0002] Current fault detection technology in low-voltage distribution networks is mainly based on fixed-threshold current monitoring schemes. These schemes use sensor chips to continuously collect the effective value or instantaneous amplitude signal of the power frequency current at each branch node of the distribution network. The system directly compares the real-time current data with a pre-set safety threshold based on the rated capacity of the line. When the current value at any monitoring point continuously exceeds the threshold, the system determines that a fault has occurred in the branch where that node is located (such as overload or short circuit) and triggers a generalized area alarm.

[0003] However, existing technologies rely solely on single-point judgment based on amplitude thresholds, neglecting the spatial correlation characteristics of faults. Secondly, they lack the separation and targeted analysis of multi-scale signal characteristics (especially instantaneous fluctuations and high-frequency noise), making it impossible to distinguish between normal load fluctuations and fault symptoms, which can easily lead to false alarms or missed alarms. Furthermore, they do not consider the relationship between long-term signal trends and short-term abrupt changes, and do not adequately handle boundary effects at data truncation points, which may distort the judgment of trends and affect the accuracy of subsequent analysis. Finally, their alarm mechanism is simplistic, lacking a multi-level, multi-criteria verification process from "preliminary anomaly" to "precise location," resulting in coarse location and low reliability.

[0004] Therefore, the main problem with existing technologies is that they cannot detect and locate weak current anomalies in small branches of low-voltage distribution networks with high sensitivity and high accuracy. Summary of the Invention

[0005] This invention provides a method and system for detecting faults in low-voltage distribution networks, enabling early, highly sensitive identification and precise spatial location of weak, concealed current anomalies in small branches.

[0006] Firstly, in order to solve the above-mentioned technical problems, the present invention provides a method for detecting faults in low-voltage distribution networks, comprising: Acquire current signal data and preprocess the current signal data to obtain a raw dataset containing time and spatial dimensions; Based on the original dataset, instantaneous fluctuation features in the time dimension are extracted, and the signal features corresponding to different spatial locations are compared and analyzed within the same time window to obtain spatial distribution feature data in the spatial dimension. Based on the spatial distribution characteristic data, a long-term trend sequence of the signal is constructed, and the long-term trend sequence is subjected to boundary correction processing in the endpoint region of the time series to obtain a corrected time series. Spatial distribution deviation data is constructed based on the corrected time series and the spatial distribution characteristic data, and preliminary abnormal regions are determined based on the spatial distribution deviation data; The original signal corresponding to the preliminary abnormal region is extracted from the original dataset, and residual analysis is performed between the original signal and the corrected time series to determine the concentrated abnormal region. Spatial difference calculation and temporal fluctuation screening are performed on the spatial units corresponding to the anomaly concentration area, and the specific location of the fault area is determined based on the screening results; An alarm signal is generated based on the specific location of the fault area, and the spatial distribution difference data corresponding to the fault area is recorded to obtain the abnormal detection result of the micro-branch current of the low-voltage distribution network.

[0007] Secondly, the present invention provides a low-voltage distribution network fault detection system, comprising: The data processing module is used to acquire current signal data and preprocess the current signal data to obtain a raw dataset containing time and spatial dimensions. The spatial distribution module is used to extract instantaneous fluctuation features in the time dimension based on the original dataset, and to compare and analyze the signal features corresponding to different spatial locations within the same time window to obtain spatial distribution feature data in the spatial dimension. The time correction module is used to construct a long-term trend sequence of the signal based on the spatial distribution feature data, and to perform boundary correction processing on the long-term trend sequence in the endpoint region of the time series to obtain a corrected time series. An anomaly extraction module is used to construct spatial distribution deviation data based on the corrected time series and the spatial distribution feature data, and to determine the preliminary anomaly region based on the spatial distribution deviation data; An anomaly focusing module is used to extract the original signal corresponding to the preliminary anomaly region from the original dataset, and perform residual analysis on the original signal and the corrected time series to determine the anomaly concentration region. The fault location module is used to calculate spatial differences and filter temporal fluctuations in spatial units corresponding to areas of concentrated anomalies, and determine the specific location of the fault area based on the filtering results. The anomaly detection module is used to generate an alarm signal based on the specific location of the fault area and record the spatial distribution difference data corresponding to the fault area to obtain the anomaly detection result of the micro-branch current of the low-voltage distribution network.

[0008] Compared with the prior art, the present invention has the following beneficial effects: (1) This invention proposes a fault detection method for low-voltage distribution networks. By preprocessing the current signal, a raw dataset containing both time and spatial dimensions is constructed simultaneously. Based on this, instantaneous fluctuation characteristics are extracted and spatial comparative analysis is performed to form spatial distribution characteristic data. Compared with traditional methods that only analyze based on a single time series, this technical solution introduces spatial distribution information at the signal processing stage, enabling subsequent analysis to simultaneously reflect the temporal variation characteristics of the signal and the distribution differences at different spatial locations. This is beneficial for identifying fault symptoms where the overall amplitude change is not significant but abnormal characteristics have appeared in local branches or regions.

[0009] (2) This invention constructs and corrects the long-term trend sequence of the signal, and combines it with spatial distribution characteristic data to form spatial distribution deviation data, thereby progressively filtering out preliminary abnormal regions and concentrated abnormal regions. This technical solution effectively distinguishes between discrete fluctuations caused by random noise and abnormal changes with spatial continuity by performing residual analysis and spatial density judgment on the measured signal and the corrected time series based on historical trends, thus enhancing the ability of the fault detection process to suppress random interference and improving the stability and reliability of abnormal region screening.

[0010] (3) After identifying the concentrated anomaly area, this invention further performs spatial difference calculation and temporal fluctuation screening on the corresponding spatial units, and determines the specific location of the fault area by combining spatial difference characteristics and temporal fluctuation characteristics. This spatiotemporal correlation analysis mechanism makes fault location not only dependent on single-dimensional anomaly judgment, but also based on the joint constraints of spatial distribution differences and temporal change characteristics for confirmation, thereby reducing the probability of false alarms and improving the credibility of the final output fault location information, providing more valuable technical support for the refined operation and maintenance and rapid fault handling of low-voltage distribution networks. Attached Figure Description

[0011] Figure 1 This is a schematic diagram of the low-voltage distribution network fault detection method provided in the first embodiment of the present invention; Figure 2 This is a schematic diagram of the low-voltage power distribution network fault detection system provided in the second embodiment of the present invention. Detailed Implementation

[0012] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0013] Reference Figure 1The first embodiment of the present invention provides a low-voltage distribution network fault detection method, including the following steps: S11, acquire current signal data, and preprocess the current signal data to obtain a raw dataset containing time and spatial dimensions; S12, Based on the original dataset, extract the instantaneous fluctuation features in the time dimension, and compare and analyze the signal features corresponding to different spatial locations within the same time window to obtain spatial distribution feature data in the spatial dimension; S13, construct a long-term trend sequence of the signal based on the spatial distribution feature data, and perform boundary correction processing on the long-term trend sequence in the endpoint region of the time series to obtain a corrected time series; S14, construct spatial distribution deviation data based on the corrected time series and the spatial distribution characteristic data, and determine the preliminary abnormal region based on the spatial distribution deviation data; S15, extract the original signal corresponding to the preliminary abnormal region from the original dataset, and perform residual analysis on the original signal and the corrected time series to determine the abnormal concentration region; S16, perform spatial difference calculation and time fluctuation screening on the spatial units corresponding to the anomaly concentration area, and determine the specific location of the fault area based on the screening results; S17, generate an alarm signal based on the specific location of the fault area, and record the spatial distribution difference data corresponding to the fault area to obtain the abnormal detection result of the micro-branch current of the low-voltage distribution network.

[0014] In step S11, it is necessary to acquire current signal data and preprocess the current signal data to obtain a raw dataset containing time and spatial dimensions, including: The current signal data is subjected to analog-to-digital conversion to obtain a digital current signal; Multi-scale frequency decomposition is performed on the digital current signal to obtain layered signal sequences corresponding to different frequency levels; Wavelet transform filtering is performed on the layered signal sequence to extract high-frequency and low-frequency components, respectively, to obtain frequency-separated signals; Based on the digital current signals corresponding to the frequency separation signals at different sampling time points, time alignment and unified indexing are performed to construct a time series structure; Based on pre-stored sensor configuration data, spatial indexing and reorganization processing is performed on the frequency-separated signals to construct a spatial distribution structure, thereby obtaining the original dataset containing both time and spatial dimensions.

[0015] The system first acquires current signal data from multiple monitoring locations in the low-voltage distribution network. This current signal data is collected by current sensors installed at distribution lines, branch circuits, or key nodes. During the acquisition process, the corresponding sampling time information is recorded synchronously for each current signal. For parallel acquisition by multiple sensors, each current sensor employs a unified sampling sequence or synchronous triggering method to ensure the comparability of current signals at different monitoring locations in the time dimension.

[0016] Subsequently, analog-to-digital conversion (ADC) processing is performed on the current signal data, converting the continuous current signal output by the current sensor into a digital current signal. This ADC is performed by a data acquisition terminal located at the acquisition end, which includes an ADC unit and a data buffer unit. The ADC unit samples the current signal according to a pre-set sampling frequency and quantizes the sampling results into discrete digital current signals, while simultaneously recording and storing the corresponding sampling time information. In the case of parallel acquisition by multiple sensors, each acquisition terminal uses a unified clock source or synchronous triggering method to generate sampling time information to ensure the consistency of digital current signals from different spatial locations in the time dimension. During the ADC process, basic filtering and quantization processing are performed on the sampled signal according to the pre-configured acquisition parameters of the system to suppress sampling noise and quantization errors. The acquisition parameters can be calibrated and determined based on historical data under normal operating conditions during the system deployment phase. In one embodiment, the quantization accuracy can be set to no less than twelve bits. Through the above processing, a digital current signal containing the sampled value and its corresponding sampling time information is obtained.

[0017] After obtaining the digital current signal, a multi-scale frequency decomposition process is performed on it, dividing the current signal in the same time series into multiple hierarchical signal sequences according to different change scales. During the multi-scale frequency decomposition process, the number of decomposition levels is preset based on the sampling frequency of the digital current signal and the current variation characteristics of the low-voltage distribution network, so that different hierarchical signal sequences cover different frequency variation ranges. In one embodiment, three to six decomposition levels can be set, where higher levels correspond to the slow-changing components in the current signal, and lower levels correspond to the fast-fluctuating components. The decomposition process can be completed in the acquisition terminal or the upper-level processing module. Each hierarchical signal sequence is stored separately while maintaining its original time index, used to characterize the variation characteristics of the current signal at different time scales. Through the above multi-scale frequency decomposition process, hierarchical signal sequences are obtained for subsequent filtering, feature extraction, and time series analysis.

[0018] Next, wavelet transform filtering is performed on the layered signal sequence. Noise suppression and feature preservation operations are performed on signals at different levels to extract high-frequency and low-frequency components and form frequency-separated signals. During the filtering process, filtering parameters are set for each level of signal according to the frequency range corresponding to the layered signal sequence. In one embodiment, the high-frequency level signal mainly retains rapidly changing components and suppresses random noise, while the low-frequency level signal mainly retains slowly changing components to reflect the overall trend. The filtering threshold can be set after statistical analysis of historical current data collected under normal operating conditions of the low-voltage distribution network, or determined through experimental calibration during the system deployment phase. In one embodiment, the threshold setting can be determined based on the data distribution of no less than 1000 normal operating sampling points. Through the above wavelet transform filtering process, frequency-separated signals representing rapid fluctuation characteristics and slow changing trend characteristics are obtained and stored in consistency with the original time index.

[0019] After obtaining the frequency separation signals, time alignment and unified indexing are performed based on the sampling time information corresponding to the frequency separation signals. Specifically, frequency separation signals from different current sensors or with different sampling start times are mapped to a unified time axis, and the positions corresponding to the same time index represent signal data within the same sampling period. In one embodiment, a unified time index sequence can be constructed according to fixed time intervals, and the frequency separation signals of each sensor can be merged into the corresponding time index according to their sampling time. When there is a sampling gap or time offset at a certain time index, the missing data is processed by interpolation or padding. In one embodiment, linear interpolation or forward padding can be used to complete the alignment process. Through the above time alignment and unified indexing processes, a time series structure with the time index as the main order is formed, so that frequency separation signals from different spatial locations maintain a one-to-one correspondence in the same time dimension.

[0020] Finally, based on the pre-stored sensor configuration data, spatial index reorganization processing is performed on the frequency-separated signals to construct a spatial distribution structure. The sensor configuration data is a data table established and stored during the system deployment phase, used to record the correspondence between each current sensor and its installation location information in the low-voltage distribution network. In one embodiment, the configuration data includes at least the sensor's unique identifier and its corresponding line segment, node location, or physical installation location identifier. During spatial index reorganization, the frequency-separated signals, after time alignment processing, are reorganized according to the location information, enabling signal data from different spatial locations under the same time index to be arranged and stored by spatial units. In one embodiment, signal data corresponding to different sensors are mapped to a preset number of spatial unit sets according to their location information. Through the above spatial index reorganization processing, an original dataset containing both time and spatial dimensions is formed. This original dataset is used for subsequent spatial distribution feature extraction, abnormal area identification, and fault location analysis.

[0021] In step S12, it is necessary to extract instantaneous fluctuation features in the time dimension based on the original dataset, and to compare and analyze the signal features corresponding to different spatial locations within the same time window to obtain spatial distribution feature data in the spatial dimension, including: The original dataset is subjected to time-dimensional signal decomposition processing to extract instantaneous time fluctuation features and separate high-frequency noise components, forming a decomposed signal sequence. Based on the decomposed signal sequence, the instantaneous fluctuation characteristics of different spatial locations are compared within the same time window to obtain a spatially layered sequence that reflects spatial differences. The frequency domain energy distribution data is calculated based on the spatial layering sequence to obtain the energy distribution characteristics corresponding to each spatial layer; Identify the time points of energy changes from the energy distribution characteristics to obtain the location of the time changes; Based on the time-varying locations, a consistency analysis is performed on the energy changes at corresponding time points in different spatial levels, thereby obtaining spatial distribution characteristic data in the spatial dimension.

[0022] When performing time-dimensional signal decomposition processing on the original dataset, the time series corresponding to each spatial unit in the original dataset is segmented according to a time window of preset length. Specifically, continuous current sampling data is divided into several overlapping or connected time windows, and the data within each time window is processed sequentially according to time order. In one embodiment, the time window length can be set to 10 to 50 consecutive sampling points, and the sliding step size between windows can be set to 1 to 5 sampling points.

[0023] Within each time window, local variation analysis is performed on the current signal, calculating the amplitude and intensity of the change in the current signal during that time period to characterize the fluctuation of the current signal over a short period; the amplitude and intensity of the change are used as instantaneous fluctuation characteristics. Simultaneously, signal components with a frequency significantly higher than the average change level of adjacent time windows are identified and classified as random high-frequency noise components; these high-frequency noise components are suppressed or eliminated, retaining only the effective signal components reflecting the actual operating state changes of the current.

[0024] By sequentially performing the above local analysis, feature extraction, and noise separation processes on each time window, the instantaneous fluctuation features corresponding to each time window are arranged in chronological order to form a decomposed signal sequence that corresponds one-to-one with the original time axis.

[0025] After obtaining the decomposed signal sequence, a comparative analysis between spatial locations is performed based on the instantaneous fluctuation characteristics of each spatial unit within the same time window. Specifically, for the decomposed signal sequence within the same time window, the instantaneous fluctuation characteristics from different spatial units are arranged side by side, so that each spatial unit corresponds to a set of instantaneous fluctuation characteristic data within that time window.

[0026] Subsequently, the instantaneous fluctuation characteristic data are compared spatially, including differences in instantaneous fluctuation amplitude, fluctuation intensity, or change magnitude between different spatial units. Based on the comparison results, it is determined whether there are differences in the fluctuation characteristics of each spatial unit within the time window. When the instantaneous fluctuation characteristics of different spatial units show significant differences in numerical level or change trend, the corresponding spatial units are classified into different spatial levels; when the instantaneous fluctuation characteristics are similar, the corresponding spatial units are classified into the same spatial level.

[0027] By sequentially performing the above spatial comparison and hierarchical division processing on all spatial units within the same time window, a spatially hierarchical sequence is formed. This spatially hierarchical sequence describes the relative spatial distribution of instantaneous current fluctuation characteristics at each spatial location within the same time window, and serves as input data for subsequent energy distribution analysis and consistency judgment.

[0028] After obtaining the spatial layering sequence, frequency domain energy distribution calculation is performed on the decomposed signal sequences corresponding to each spatial layer based on the spatial layering sequence. Specifically, for spatial units divided into the same spatial layer within the same time window, their corresponding decomposed signal sequences are collected, and statistical analysis is performed on the decomposed signal sequences in the frequency domain to characterize the energy distribution of the current signal at different frequency components within that spatial layer.

[0029] In the process of frequency domain energy distribution calculation, the decomposed signal sequence can be divided into preset frequency intervals, the signal amplitude changes in each frequency interval can be statistically analyzed, and the statistical results can be used as the energy distribution data corresponding to that frequency interval. In one embodiment, the signal intensity changes in the low-frequency interval and the high-frequency interval can be statistically analyzed separately to reflect the differences in energy distribution of slow-changing components and fast-fluctuating components at different spatial levels.

[0030] By performing the aforementioned frequency domain statistical processing on the decomposed signal sequences corresponding to each spatial level, the energy distribution characteristics corresponding to each spatial level are obtained. These energy distribution characteristics describe the distribution of current fluctuation intensity in the frequency dimension within the same time window for different spatial levels, and serve as input data for subsequent time-varying location identification and spatial consistency analysis.

[0031] After obtaining the energy distribution characteristics corresponding to each spatial level, the energy distribution characteristics within a continuous time window are compared and analyzed to identify the time points of energy changes. Specifically, the energy distribution characteristics corresponding to the same spatial level within adjacent time windows are compared. When the energy change amplitude exceeds a preset change threshold, the time index of the corresponding time window is marked as the time change position. The change threshold can be statistically set based on historical energy distribution data under normal operation of the low-voltage distribution network, and remains fixed during system operation or is updated according to the maintenance cycle.

[0032] The energy change threshold is set during the system deployment phase based on historical current data collected under normal operating conditions of the low-voltage distribution network. Specifically, under fault-free or stable operating conditions, the energy distribution characteristics within multiple consecutive time windows are statistically analyzed to obtain the normal distribution range of energy change amplitude, and a value higher than the normal fluctuation level is selected as the energy change threshold based on this distribution range. In one embodiment, the energy change threshold can be set as the statistical upper limit of the energy change amplitude under normal operating conditions, or as an empirical threshold covering the vast majority of normal fluctuation samples.

[0033] During system operation, when the energy change amplitude within adjacent time windows exceeds the energy change threshold, the corresponding time point is determined as the time change location of the energy change. To adapt to long-term changes in operating conditions, the energy change threshold can be updated based on the latest collected normal operation data during system maintenance or periodic calibration, but it remains fixed during a single detection process to ensure consistency in the determination of the time change location.

[0034] The threshold for spatial consistency analysis can also be set based on historical normal operation data. Specifically, under normal operating conditions, the range of energy change differences between different spatial levels within the same time window is statistically analyzed, and a spatial consistency threshold is set accordingly. When the direction and magnitude of energy change at the same location of multiple spatial levels at the same time meet the consistency threshold, the change is determined to have spatial consistency characteristics, and the corresponding results are included in the spatial distribution characteristic data.

[0035] After determining the location of the time change, a consistency analysis is performed on the energy changes of different spatial levels within the corresponding time window. Specifically, a fixed-length time window is selected, centered on the time change location, and the energy change values ​​of each spatial level within this time window are statistically analyzed. These energy change values ​​characterize the energy increase or decrease of the spatial level during this time period.

[0036] Subsequently, the energy change values ​​of different spatial levels are compared layer by layer to determine whether their change directions are consistent, and further compare whether their change amplitudes fall within the preset change tolerance range; when at least two or more spatial levels simultaneously satisfy the same change direction at the same time change position, and the difference in change amplitude is less than the preset change tolerance, it is determined that the energy change corresponding to the time change position has spatial consistency characteristics.

[0037] In one embodiment, the consistent direction of change means that the energy changes at each spatial level are either increasing or decreasing. The tolerance for change can be set based on the statistical results of the differences in energy changes at different spatial levels under normal operating conditions, and is used to exclude non-consistent changes caused by measurement errors or local random disturbances. When the consistency determination is met, the spatial level identification information that meets the consistency condition and its corresponding energy change characteristics are recorded as a set of spatial distribution characteristic data. The spatial distribution characteristic data is used to characterize the spatial distribution state of the current signal at the time change location and serves as input data for subsequent abnormal area identification and spatial deviation analysis.

[0038] In step S13, it is necessary to construct a long-term trend sequence of the signal based on the spatial distribution feature data, and perform boundary correction processing on the long-term trend sequence in the endpoint region of the time series to obtain a corrected time series, including: The spatial distribution feature data is processed to retain low-frequency components, and a moving average operation is performed on the processed data to obtain a long-term trend sequence. The starting and ending segments of the long-term trend sequence are scanned, and the differences in the change amplitude between the starting and ending segments and the non-endpoint segments are compared to identify distortion points in the endpoint regions of the sequence, thus obtaining the distortion boundary sequence. Based on the distortion boundary sequence, the long-term trend sequence is expanded and the corresponding sequence value is calculated to obtain a preliminary correction sequence; The overall trend of the preliminary correction sequence is compared with that of the long-term trend sequence. When the comparison result meets the preset consistency threshold, wavelet reconstruction processing is performed on the preliminary correction sequence to obtain the corrected time series.

[0039] When performing low-frequency component retention processing on the spatial distribution feature data, the processing object of the spatial distribution feature data in the time dimension is first determined. Specifically, for each spatial unit, the spatial distribution feature data sequence formed under a continuous time window is taken and arranged into a time series as input according to the time order. Then, low-frequency retention processing is performed on the time series, which is implemented by "removing short-period fluctuations and retaining slow changes". In one embodiment, the difference between adjacent time points can be calculated first, and the segment where the absolute value of the difference is consistently lower than a preset fluctuation threshold is regarded as the low-frequency dominant segment, and the segment where the absolute value of the difference frequently exceeds the preset fluctuation threshold is regarded as the short-period fluctuation segment. After performing suppression processing on the short-period fluctuation segment, the low-frequency retained sequence is obtained. The preset fluctuation threshold can be set during the deployment phase based on normal operation historical data and remains fixed during a single detection process.

[0040] After obtaining the low-frequency retained sequence, a moving average operation is performed on the low-frequency retained sequence to smooth residual fluctuations and obtain a long-term trend sequence. Specifically, a fixed-length sliding window is used to calculate the average value within the window point by point, and the obtained average value is used as the trend value at the corresponding time point. In one embodiment, the moving average window length can be set to 5 to 30 data points corresponding to consecutive time windows, and the sliding step size can be set to 1. After the moving average operation is completed, the trend values ​​corresponding to each time point are arranged in chronological order to obtain the long-term trend sequence. The long-term trend sequence is consistent with the original time index and is used for subsequent endpoint segment scanning and boundary correction processing.

[0041] When scanning the starting and ending segments of the long-term trend sequence, the endpoint segment range is first defined on the time index. Specifically, the first few data points of the long-term trend sequence are defined as the starting segment, the last few data points of the long-term trend sequence are defined as the ending segment, and the intermediate part excluding the starting and ending segments is defined as the non-endpoint segment. In one embodiment, the length of the starting and ending segments can be set to 10 to 30 consecutive data points, and the length of the non-endpoint segment is the remaining data points.

[0042] Subsequently, change magnitude indices are calculated for the starting segment, ending segment, and non-endpoint segments. These change magnitude indices describe the degree of drastic change in the trend value within a segment over time. In practice, the changes in adjacent data points within a segment can be statistically analyzed to obtain the average and maximum change magnitude of the segment, with the average change magnitude corresponding to the non-endpoint segments serving as the baseline change level.

[0043] After obtaining the baseline change level, distortion points are identified point by point in the starting and ending segments. Specifically, starting from the time index near the beginning of the sequence in the starting segment, it is determined whether the change amplitude near the data point is higher than the baseline change level. When the change amplitude of a data point and several adjacent data points continuously exceeds a preset multiple threshold or a preset difference threshold, the data point is marked as a distortion point. Similarly, scanning backward from the time index near the end of the sequence in the ending segment, distortion points in the ending segment are marked in the same way.

[0044] In one embodiment, the preset multiplier threshold can be set to 2 to 5 to determine the abnormal amplification of the change amplitude of the endpoint segment relative to the non-endpoint segment; the preset difference threshold can be set according to the distribution of change amplitude in the historical data of normal operation. After the scan is completed, the time index of the distortion point closest to the non-endpoint segment in the starting segment is recorded as the starting distortion boundary, and the time index of the distortion point closest to the non-endpoint segment in the ending segment is recorded as the ending distortion boundary; the starting distortion boundary and the ending distortion boundary are combined in chronological order to obtain a distortion boundary sequence, which is used to limit the processing range of subsequent sequence expansion and endpoint correction. When scanning the starting and ending segments of the long-term trend sequence, the endpoint segment range is first defined on the time index. Specifically, the first few data points of the long-term trend sequence are defined as the starting segment, the last few data points of the long-term trend sequence are defined as the ending segment, and the intermediate part excluding the starting and ending segments is defined as the non-endpoint segment. In one embodiment, the length of the starting and ending segments can be set to 10 to 30 consecutive data points, and the length of the non-endpoint segment is the remaining data points.

[0045] Subsequently, change magnitude indices are calculated for the starting segment, ending segment, and non-endpoint segments. These change magnitude indices describe the degree of drastic change in the trend value within a segment over time. In practice, the changes in adjacent data points within a segment can be statistically analyzed to obtain the average and maximum change magnitude of the segment, with the average change magnitude corresponding to the non-endpoint segments serving as the baseline change level.

[0046] After obtaining the baseline change level, distortion points are identified point by point in the starting and ending segments. Specifically, starting from the time index near the beginning of the sequence in the starting segment, it is determined whether the change amplitude near the data point is higher than the baseline change level. When the change amplitude of a data point and several adjacent data points continuously exceeds a preset multiple threshold or a preset difference threshold, the data point is marked as a distortion point. Similarly, scanning backward from the time index near the end of the sequence in the ending segment, distortion points in the ending segment are marked in the same way.

[0047] In one embodiment, the preset multiplier threshold can be set to 2 to 5 to determine the abnormal amplification of the change amplitude of the endpoint segment relative to the non-endpoint segment; the preset difference threshold can be set according to the distribution of change amplitude in the historical data of normal operation. After the scan is completed, the time index of the distortion point closest to the non-endpoint segment in the starting segment is recorded as the starting distortion boundary, and the time index of the distortion point closest to the non-endpoint segment in the ending segment is recorded as the ending distortion boundary; the starting distortion boundary and the ending distortion boundary are combined in chronological order to obtain a distortion boundary sequence, which is used to limit the processing range of subsequent sequence expansion and endpoint correction. After obtaining the preliminary corrected sequence, an overall trend comparison is performed between the preliminary corrected sequence and the long-term trend sequence. Specifically, the preliminary corrected sequence and the long-term trend sequence are mapped point-by-point under the same time index, and the direction and magnitude of change of the two sequences over the overall time range are statistically compared to determine whether the preliminary corrected sequence maintains an overall trend consistent with the long-term trend sequence. In one embodiment, the average of the proportion of consistent change direction and the difference in change magnitude between the two sequences at adjacent time points can be calculated as the basis for judging the consistency of the overall trend.

[0048] When the proportion of consistent change directions is higher than a preset proportion threshold and the difference in change magnitude is lower than a preset magnitude threshold, the preliminary correction sequence is determined to meet the overall trend consistency condition. The proportion threshold and magnitude threshold are determined based on historical data of normal operation of the low-voltage distribution network during the system deployment phase. Specifically, before system commissioning, historical current data in continuous normal operation is selected as sample data. The sample data is processed using the same procedure as in the detection phase to construct a long-term trend sequence and generate a corresponding preliminary correction sequence. Subsequently, statistical analysis is performed on the change direction and magnitude of the two sequences under the same time index. In one embodiment, the proportion threshold is determined by the proportion of the preliminary correction sequence and the long-term trend sequence in the statistical sample data where their change directions are consistent at adjacent time points. Specifically, the proportion of time points with consistent change directions is counted relative to the total number of time points, and the lower limit of this proportion in multiple sample segments is taken as the proportion threshold. In engineering implementation, the proportion threshold can be set to a fixed value between 80% and 95%. The magnitude threshold is determined by the difference in the magnitude of change between the preliminary correction sequence and the long-term trend sequence at corresponding time points in the statistical sample data. Specifically, the average level and fluctuation range of the variation amplitude in the sample data are calculated, and a value higher than the average level but lower than the abnormal fluctuation level is selected as the amplitude threshold. In engineering implementation, the amplitude threshold can be set to 1.2 to 2 times the average variation amplitude of the long-term trend sequence.

[0049] After the system enters actual detection operation, the ratio threshold and amplitude threshold remain unchanged during a single detection process and are not dynamically adjusted with changes in instantaneous data, so as to avoid threshold drift affecting the trend consistency judgment result.

[0050] After satisfying the condition of overall trend consistency, the preliminary corrected sequence is reconstructed to further smooth local discontinuities. Specifically, the preliminary corrected sequence is decomposed into multiple scales in the time dimension, retaining components reflecting the overall trend while suppressing local abrupt changes. The retained components are then reconstructed to obtain the reconstructed time series. For example, a Daubechies wavelet family (such as db4) can be used for 3-5 level decomposition. During reconstruction, approximation coefficients (low-frequency components) are retained to maintain the overall trend, while soft thresholding is applied to detail coefficients (high-frequency components) to suppress local high-frequency oscillations that may be introduced during expansion and calculation. Finally, an inverse wavelet transform is performed to obtain a smoothed corrected time series. The original time index order is not changed during reconstruction; only the values ​​at corresponding time points are adjusted.

[0051] After the reconstruction process is completed, the reconstructed time series is output as the corrected time series. The corrected time series maintains the same overall trend as the long-term trend series, while eliminating the discontinuous changes introduced by distortion in the end regions of the series. It is used for subsequent spatial distribution deviation calculation and abnormal region determination steps.

[0052] In step S14, it is necessary to construct spatial distribution deviation data based on the corrected time series and the spatial distribution feature data, and determine the preliminary abnormal region based on the spatial distribution deviation data, including: Based on the spatial distribution characteristic data, the change weight of each spatial level is calculated to obtain the set of level weights; Based on the hierarchical weight set, the data corresponding to the spatial level in the spatial distribution feature data are weighted and fused to obtain a fused spatial sequence. Based on the difference between the fused spatial sequence and the preset normal state reference sequence, the spatial distribution deviation value is calculated to form spatial distribution deviation data; Based on the comparison between the deviation value corresponding to each spatial unit in the spatial distribution deviation data and the preset deviation threshold, spatial units whose deviation values ​​exceed the preset deviation threshold are selected to obtain abnormal candidate regions. The local density is calculated based on the adjacency relationship of spatial units in the abnormal candidate region, and an initial abnormal region is determined when the local density exceeds a preset density threshold.

[0053] Based on spatial distribution characteristic data, the dynamic change weights of each spatial level under historical normal conditions are calculated. In specific implementation, for each predefined spatial level, according to the range of spatial units it contains, the feature value sequences of all units within each time window of the historical normal period are directly indexed and aggregated from the historical storage area of ​​the spatial distribution characteristic data, forming a hierarchical time-series data set. Next, the core statistical indicators for each level are calculated: first, the average change amplitude, which is the mean of the absolute values ​​of feature value changes of all spatial units within that level between adjacent time windows; and second, the change stability coefficient, characterized by the reciprocal of the coefficient of variation of the time-series changes of all units within that level. Then, a preset weight allocation function is applied, designed to assign higher weights to levels with larger average change amplitudes and higher stability coefficients. For example, a weighted scoring method is used, performing maximum-min normalization on both indicators, synthesizing a comprehensive score according to empirical proportions (e.g., change amplitude accounts for 40%, stability accounts for 60%), and finally using the Softmax function to convert the comprehensive scores of all levels into a weight set summing to 1. This set of weights is calibrated once using offline historical data during system deployment and is used consistently in subsequent online detection.

[0054] Using the aforementioned hierarchical weight set, the original spatial distribution feature data is weighted and fused to generate a standardized fused spatial sequence. Specifically, for each processing time window, all spatial units are traversed. For any spatial unit, its spatial hierarchy is identified, and the weight of that hierarchy, calculated previously, is obtained. The original feature value of the spatial unit in the current time window is multiplied by the weight of its hierarchy to obtain the weighted feature value of that unit. This process is executed in parallel for all spatial units. Finally, each time window outputs a data slice with a completely consistent spatial structure but with feature values ​​weighted according to hierarchical importance. Stacking the slices of all windows in chronological order constitutes the fused spatial sequence. This step unifies the contribution of different levels to the overall spatial pattern.

[0055] By comparing the fused spatial sequence with a pre-defined normal-state baseline sequence, spatial deviation is quantified, and spatial distribution deviation data is constructed. In practice, the normal-state baseline sequence is a reference template pre-established based on long-term historical anomaly-free data, and its data structure (spatial unit dimension, time window dimension) is identical to the fused spatial sequence. For each data point in the fused spatial sequence (i.e., the value of a specific spatial unit within a specific time window), its difference is calculated with the baseline value in the baseline sequence corresponding to the same spatial unit and time window index (or a time phase with the same business meaning, such as "weekday morning rush hour"), and the absolute value is taken to obtain the deviation value for that point. All deviation values ​​are reorganized into a new data cube according to the original spatial and time indices, which is the spatial distribution deviation data. This data directly reflects the degree to which each spatial location deviates from its normal baseline at each moment.

[0056] By comparing thresholds, anomalous spatial units are initially screened from the spatial distribution deviation data, forming anomaly candidate regions. Specifically, a deviation threshold is preset, determined based on the statistical distribution of deviation data during historical normal periods, for example, the 99th percentile of historical deviation values. Then, the spatial distribution deviation data is scanned window by window, and the deviation value of each spatial unit is compared with this threshold. If the deviation value of a spatial unit exceeds the threshold, the unit is marked as an "abnormal spatial unit" in the current time window. When all marked anomalous spatial units within a time window are aggregated, they constitute the "anomaly candidate region" corresponding to that window. This step completes the initial binarization judgment from continuous deviation values ​​to discrete anomaly suspects.

[0057] Spatial clustering analysis is performed on candidate anomaly regions to identify preliminary anomaly regions based on local anomaly density. First, based on the topological relationships of spatial units (such as adjacency relationships with shared boundaries), the "local anomaly density" is calculated for each anomalous spatial unit within the current candidate anomaly region. This density is the proportion of units belonging to the current candidate anomaly region among all its directly adjacent (e.g., first-order neighborhood) spatial units centered on the target unit. Then, a density threshold (e.g., 50%) is set. If the local anomaly density of an anomalous spatial unit exceeds this threshold, it is identified as a "core anomaly." Finally, based on spatial adjacency relationships, all interconnected core anomalies are aggregated into a connected component. Each such connected component is ultimately determined as a "preliminary anomaly region." This step, through spatial density filtering, effectively removes isolated noise anomalies and retains true anomaly regions that form spatial clustering patterns.

[0058] In step S15, it is necessary to extract the original signal corresponding to the preliminary abnormal region from the original dataset, and perform residual analysis on the original signal and the corrected time series to determine the concentrated abnormal region, including: Based on the spatial index corresponding to the preliminary abnormal region, the original signal of the corresponding spatial unit is extracted from the original dataset, and the corrected time series corresponding to the spatial unit is selected. The absolute difference between the original signal and the corrected time series is calculated at the same time point and the same spatial unit to obtain a set of residual values. When the deviation of the residual value set from the preset reference distribution data exceeds a preset deviation threshold, an inconsistent distribution sub-region is determined. Statistical analysis is performed on the adjacency relationships of spatial units in the inconsistent distribution sub-regions to calculate the local residual density, and the set of spatial units whose local residual density exceeds a preset density threshold is identified as an abnormally concentrated region.

[0059] Based on the preliminary abnormal region determined in step S14, a spatial index set corresponding to the preliminary abnormal region is obtained. The spatial index is used to identify the storage location of each spatial unit within the abnormal region in the original dataset. In this step, firstly, the current signal data stored in the original dataset is indexed and queried according to the spatial index set to locate the original current signal record of the corresponding spatial unit, and the original current signal sequence of the spatial unit within the continuous sampling time is read in chronological order. The original current signal is the digital current signal after analog-to-digital conversion in step S11, before trend correction processing. Subsequently, using the same spatial index and time index, the correction time series data of the corresponding spatial unit is retrieved from the correction time series generated in step S13, so that the original current signal sequence and the correction time series correspond one-to-one in sampling time point and spatial unit identifier, thereby forming a set of signal pairs for subsequent residual analysis.

[0060] After obtaining the original current signal sequence and the corresponding corrected time sequence within the same spatial unit, residual calculation is performed on both at each time point. Specifically, the original current signal sequence and the corrected time sequence are aligned one-to-one according to the time index. For each sampling time point, the absolute difference between the original current signal value and the corrected time sequence value at that time point is calculated as the residual value for that time point. The above calculation process is repeated for all sampling time points of each spatial unit within the initial anomaly region. The obtained residual values ​​are stored according to the correspondence between "spatial unit - time index" to form a residual value set. This residual value set is used to characterize the degree of deviation of the actual observed signal from the trend correction signal and serves as input data for subsequent residual distribution consistency analysis.

[0061] After obtaining the set of residual values, for each spatial unit within the initial abnormal region, the residual distribution characteristics of that spatial unit within the current detection time window are constructed and compared with preset reference distribution data to calculate the degree of deviation. Specifically, the residual values ​​of the spatial unit within the current time window are first divided into several intervals according to their numerical values, and the frequency of occurrence of each interval is counted to obtain the residual distribution statistics of the spatial unit. Subsequently, reference residual distribution statistics with the same time window length are read from the preset reference distribution data, and the statistical difference between the two in the same interval is used as the basis for evaluating the degree of deviation. If the degree of deviation exceeds the preset deviation threshold, it is determined that the residual distribution of the spatial unit is inconsistent with the normal operation distribution, and the spatial unit and its adjacent spatial units that also meet the deviation conditions are merged and marked as a distribution inconsistency sub-region. In this way, the distribution inconsistency sub-region consists of a group of spatial units that show "abnormal deviation of residual distribution" within the same time window, which is different from occasional fluctuation points where only a few moments show large residuals but the overall distribution is still close to normal.

[0062] It should be noted that, during the initialization or training phase, the preset reference distribution data is obtained by collecting long-term (e.g., several weeks) fault-free operation data. For each spatial unit, the set of residual values ​​within each time window is calculated according to the method described above. Then, the distribution characteristics of the set are statistically analyzed. A smooth probability density function can be generated using the kernel density estimation (KDE) method as the reference distribution for the unit, or it can be fitted to a Gaussian distribution and its mean and variance can be recorded. This process establishes an independent reference distribution model for each spatial unit.

[0063] The deviation threshold is preset based on historical data from normal operation during the system deployment phase and remains fixed during a single detection process. Specifically, continuously operating historical data is selected, and a set of residual values ​​is generated according to the same process as online detection. For each spatial unit, its "degree of deviation of residual distribution" is calculated within multiple time windows to form a deviation sample set. Subsequently, the upper limit range of the deviation sample set is statistically analyzed, and a threshold that can cover the vast majority of normal samples is selected as the deviation threshold. In one embodiment, the deviation threshold can be set to the high quantile level of the deviation of normal samples, such as the value corresponding to the 95th or 99th percentile, so that only a very small number of windows will trigger the distribution inconsistency judgment under normal operation, thereby controlling the probability of false alarms. For scenarios with different feeders or different load fluctuation intensities, the deviation threshold can be calibrated separately for each feeder or region, and a threshold configuration table can be established. During detection, the corresponding threshold is automatically called according to the region to which the sensor belongs.

[0064] After identifying the inconsistent distribution sub-regions, the spatial clustering characteristics of the spatial units within these sub-regions are further analyzed to distinguish between continuous anomalous regions and discrete anomalous points. Specifically, based on a pre-established spatial unit adjacency table in the system, for each inconsistently distributed spatial unit, the number or proportion of its neighboring spatial units that are also identified as inconsistently distributed is counted, and this statistical result is used as the local residual density of that spatial unit. The local residual density reflects the degree of spatial clustering of anomalous residuals. When the local residual density corresponding to a spatial unit exceeds a preset density threshold, it indicates that there are multiple neighboring spatial units with consistent anomalous characteristics around that spatial unit. Thus, that spatial unit and its neighboring spatial units that meet the density condition are merged and identified as an anomalous concentration region.

[0065] The density threshold is set during the system deployment phase based on historical data from normal operation. Specifically, the frequency and spatial clustering of spatial units judged as having inconsistent distributions in residual distribution consistency analysis under normal operating conditions are statistically analyzed. The distribution of the number of adjacent spatial units simultaneously showing anomaly judgments is recorded, and based on this, an upper limit value of density that can cover the vast majority of normal operating conditions is determined as the density threshold. In one embodiment, the density threshold can be set to the condition that at least two or more adjacent spatial units are judged as having inconsistent distributions, or that at least 50% of them are judged as having inconsistent distributions. This ensures that the judgment of anomaly concentration areas is triggered only when anomalies exhibit obvious spatial clustering characteristics, thereby reducing the probability of misjudgments caused by isolated anomalies.

[0066] In step S16, spatial difference calculation and temporal fluctuation screening are performed on the spatial units corresponding to the anomaly concentration area, and the specific location of the fault area is determined based on the screening results, including: Based on the spatial units corresponding to the anomaly concentration areas, the characteristic difference values ​​between adjacent spatial units are calculated to obtain spatial difference data. Select the spatial unit corresponding to the spatial difference data exceeding the preset difference threshold to obtain the spatial difference region; Based on the time series corresponding to the spatial difference region, calculate the time fluctuation parameter to obtain the time fluctuation value; When the time fluctuation value exceeds a preset time fluctuation threshold, the corresponding spatial unit is determined as a spatiotemporal related region; Select a set of spatial units that are adjacent to each other in the spatiotemporal correlation region and whose distribution density exceeds a preset density threshold, and determine the spatial location corresponding to the set of spatial units as the specific location of the fault area.

[0067] First, for the anomaly concentration area identified in step S15, spatial difference calculations are performed on the spatial units within this area. Specifically, based on the pre-established spatial unit adjacency table in the system, for each spatial unit within the anomaly concentration area, its directly adjacent spatial units are selected, and their corresponding spatial distribution characteristic data or fused characteristic values ​​are compared. By calculating the differences in characteristic values ​​between adjacent spatial units pairwise, spatial difference data reflecting the degree of unevenness in spatial location is obtained. This spatial difference data is used to characterize the gradient change of the anomaly in space, providing a basis for subsequent anomaly source location.

[0068] After obtaining the spatial difference data, a threshold filtering process is performed on the spatial difference values ​​corresponding to each spatial unit. Specifically, the difference value between each spatial unit and its neighboring spatial units is compared with a preset difference threshold. When the difference value exceeds the threshold, the corresponding spatial unit is marked as a spatial difference spatial unit. All marked spatial difference spatial units within the same anomaly concentration area are grouped into a spatial difference region. This filtering process can exclude areas with gentle or uniform spatial changes, allowing subsequent analysis to focus on key locations with prominent spatial changes.

[0069] After obtaining the spatial difference data, a threshold filtering process is performed on the spatial difference values ​​corresponding to each spatial unit. Specifically, the difference value between each spatial unit and its neighboring spatial units is compared with a preset difference threshold. When the difference value exceeds the threshold, the corresponding spatial unit is marked as a spatial difference spatial unit. All marked spatial difference spatial units within the same anomaly concentration area are grouped into a spatial difference region. This filtering process can exclude areas with gentle or uniform spatial changes, allowing subsequent analysis to focus on key locations with prominent spatial changes.

[0070] For the identified spatial difference regions, further temporal fluctuation analysis is performed on their corresponding time series. Specifically, the original signal or corrected time series of the spatial difference units within a continuous time window is extracted, and temporal fluctuation parameters reflecting the severity of time changes are calculated. These temporal fluctuation parameters describe the instability characteristics of the signal in the time dimension, such as statistics reflecting the amplitude, frequency, or duration of changes. By calculating the temporal fluctuation parameters, it can be determined whether spatial differences are accompanied by significant temporal dynamic changes.

[0071] After obtaining the time fluctuation parameters, the time fluctuation values ​​corresponding to each spatial unit are compared with a preset time fluctuation threshold. When the time fluctuation value of a spatial unit exceeds the time fluctuation threshold, it indicates that the spatial unit not only differs in the spatial dimension but also exhibits obvious fluctuation characteristics in the time dimension, thus identifying the spatial unit as a spatiotemporally correlated spatial unit. All spatial units within the anomaly concentration area that meet this condition are grouped into a spatiotemporally correlated region to further narrow down the fault location range.

[0072] The time fluctuation threshold is also set based on historical data from normal operation during the system deployment phase. Specifically, the distribution of time fluctuation parameters of each spatial unit under normal operating conditions is statistically analyzed, and a threshold that can cover most of the normal fluctuation range is selected as the time fluctuation threshold. In one embodiment, the time fluctuation threshold can be set to the high quantile level of the distribution of time fluctuation parameters under normal operating conditions, so that spatiotemporal correlation determination is triggered only when the time fluctuation is higher than the normal load change, thereby improving the reliability of fault location results.

[0073] After obtaining the spatiotemporally correlated region, its spatial clustering characteristics are finally determined. Specifically, based on the adjacency relationships between spatial units, the distribution density of spatial units within the spatiotemporally correlated region is statistically analyzed. When multiple spatiotemporally correlated spatial units are spatially adjacent and their distribution density exceeds a preset density threshold, the physical spatial location corresponding to this set of spatial units is determined as the specific location of the fault area. By combining spatial differences, temporal fluctuation characteristics, and spatial clustering characteristics, the finally determined fault area can possess consistent abnormal characteristics in both time and space dimensions, thereby improving the accuracy and engineering usability of the fault location results.

[0074] In step S17, an alarm signal needs to be generated based on the specific location of the fault area, and the spatial distribution difference data corresponding to the fault area needs to be recorded to obtain the abnormal detection result of the micro-branch current of the low-voltage distribution network, including: Based on the current signal corresponding to the specific location of the fault area, the current deviation value is calculated, and an initial alarm signal is generated when the current deviation value exceeds the preset current deviation threshold. Based on the spatial distribution difference data corresponding to the specific location of the fault area, extract the difference feature vector; The initial alarm signal and the difference feature vector are compared for consistency. When the feature intensity of the difference feature vector exceeds a preset energy threshold, an enhanced alarm signal is generated. The specific location of the fault area, the differential feature vector, and the alarm status of the enhanced alarm signal are recorded to form an anomaly detection result for the micro-branch current of the low-voltage distribution network.

[0075] First, based on the specific location of the fault area determined in step S16, deviation determination processing is performed on the current signal corresponding to that area. Specifically, according to the spatial cell index corresponding to the fault area, the current signal of that spatial cell within the current detection time window is extracted from the original dataset and compared with its corresponding correction time series to calculate the current deviation value, which reflects the degree of deviation between the current current state and the normal trend. When the current deviation value exceeds a preset current deviation threshold, it is determined that there is a current anomaly in the spatial cell within the current time window, and a corresponding initial alarm signal is generated to identify the fault triggering state.

[0076] The current deviation threshold is set based on historical data from normal operation during the system deployment phase. Specifically, the deviation distribution of the current signal of each spatial unit relative to the correction time series is statistically analyzed under long-term stable operation conditions of the low-voltage distribution network, and a threshold that can cover the vast majority of normal fluctuation ranges is selected as the current deviation threshold. In one embodiment, the current deviation threshold can be set to the high quantile level of the normal deviation distribution to ensure that the initial alarm signal is triggered only when the current deviation is significantly higher than the normal load change, thereby reducing the probability of false alarms.

[0077] The current deviation threshold is set based on historical data from normal operation during the system deployment phase. Specifically, the deviation distribution of the current signal of each spatial unit relative to the correction time series is statistically analyzed under long-term stable operation conditions of the low-voltage distribution network, and a threshold that can cover the vast majority of normal fluctuation ranges is selected as the current deviation threshold. In one embodiment, the current deviation threshold can be set to the high quantile level of the normal deviation distribution to ensure that the initial alarm signal is triggered only when the current deviation is significantly higher than the normal load change, thereby reducing the probability of false alarms.

[0078] After obtaining the initial alarm signal and the differential feature vector, a consistency determination process is performed on both. Specifically, it is determined whether the spatial anomaly intensity reflected by the differential feature vector within the time window corresponding to the current initial alarm signal reaches a preset energy threshold. When the feature intensity of the differential feature vector exceeds the energy threshold, it indicates that the current deviation anomaly and the spatial distribution anomaly occur simultaneously within the same time period, thereby upgrading the initial alarm signal to an enhanced alarm signal; if the energy threshold is not exceeded, the initial alarm state is maintained or the processing is delayed. This consistency determination mechanism can avoid false alarms caused solely by instantaneous current fluctuations.

[0079] After obtaining the initial alarm signal and the differential feature vector, a consistency determination process is performed on both. Specifically, it is determined whether the spatial anomaly intensity reflected by the differential feature vector within the time window corresponding to the current initial alarm signal reaches a preset energy threshold. When the feature intensity of the differential feature vector exceeds the energy threshold, it indicates that the current deviation anomaly and the spatial distribution anomaly occur simultaneously within the same time period, thereby upgrading the initial alarm signal to an enhanced alarm signal; if the energy threshold is not exceeded, the initial alarm state is maintained or the processing is delayed. This consistency determination mechanism can avoid false alarms caused solely by instantaneous current fluctuations.

[0080] The energy threshold is also preset during the system deployment phase using historical data from normal operation. Specifically, statistical analysis is performed on the spatial difference feature vector under fault-free operating conditions to obtain the distribution range of its feature intensity under normal conditions, and a threshold that can cover most normal samples is selected as the energy threshold. In one embodiment, the energy threshold can be set to the high quantile of the spatial difference feature intensity distribution under normal operation, so that the enhanced alarm signal is triggered only when the spatial anomaly has a certain persistence and intensity, thereby improving the reliability of the final alarm result.

[0081] After generating the enhanced alarm signal, the detection results are uniformly recorded and output. Specifically, the specific location of the fault area, the corresponding differential feature vector, and the alarm status of the enhanced alarm signal are associated and stored to form a complete record of anomaly detection results. The anomaly detection results can be used for real-time alarm display, historical fault tracing, or operation and maintenance decision support, thereby completing the final detection and output of anomalies in the current of small branches of the low-voltage distribution network.

[0082] It should be noted that the preset parameters, thresholds and ratios involved in this invention are first set based on historical monitoring data under long-term stable operation of low-voltage distribution networks, and statistical analysis is performed to obtain a benchmark range; secondly, combined with fault simulation data or engineering experience, a safety margin is set on this benchmark range to ensure sensitivity to abnormal states and robustness against normal fluctuations. The spatial distribution deviation data described in this invention refers to the quantitative difference between the current feature value of each spatial unit (or the fused spatial sequence) and the preset normal state benchmark value under a specific time window, reflecting the deviation from the absolute normal benchmark; spatial difference data refers to the relative difference of feature values ​​between adjacent spatial units within an abnormal region, reflecting the spatial local gradient or non-smoothness; spatial distribution difference data is a higher-level concept, generally referring to a data set that can characterize the uneven distribution of signal features in the spatial dimension. In this invention, spatial distribution deviation data and the difference feature vector extracted from the fault area are both specific manifestations of spatial distribution difference data; spatial unit adjacency relationship refers to a relationship table predefined and stored in the system according to the electrical wiring topology diagram or sensor physical installation location diagram of the low-voltage distribution network, used to identify which sensor monitoring points (spatial units) are directly adjacent in electrical or physical location.

[0083] In summary, this invention provides a fault detection method for low-voltage distribution networks, enabling early, highly sensitive identification and precise spatial location of weak, concealed current anomalies in small branches.

[0084] Reference Figure 2 The second embodiment of the present invention provides a low-voltage distribution network fault detection system, comprising: The data processing module is used to acquire current signal data and preprocess the current signal data to obtain a raw dataset containing time and spatial dimensions. The spatial distribution module is used to extract instantaneous fluctuation features in the time dimension based on the original dataset, and to compare and analyze the signal features corresponding to different spatial locations within the same time window to obtain spatial distribution feature data in the spatial dimension. The time correction module is used to construct a long-term trend sequence of the signal based on the spatial distribution feature data, and to perform boundary correction processing on the long-term trend sequence in the endpoint region of the time series to obtain a corrected time series. An anomaly extraction module is used to construct spatial distribution deviation data based on the corrected time series and the spatial distribution feature data, and to determine the preliminary anomaly region based on the spatial distribution deviation data; An anomaly focusing module is used to extract the original signal corresponding to the preliminary anomaly region from the original dataset, and perform residual analysis on the original signal and the corrected time series to determine the anomaly concentration region. The fault location module is used to calculate spatial differences and filter temporal fluctuations in spatial units corresponding to areas of concentrated anomalies, and determine the specific location of the fault area based on the filtering results. The anomaly detection module is used to generate an alarm signal based on the specific location of the fault area and record the spatial distribution difference data corresponding to the fault area to obtain the anomaly detection result of the micro-branch current of the low-voltage distribution network.

[0085] It should be noted that the low-voltage distribution network fault detection system provided in this embodiment of the invention is used to execute all the process steps of the low-voltage distribution network fault detection method in the above embodiment. The working principles and beneficial effects of the two are one-to-one, so they will not be described again.

[0086] It should be noted that the system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the system embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0087] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A method for detecting faults in a low-voltage distribution network, characterized in that... include: Acquire current signal data and preprocess the current signal data to obtain a raw dataset containing time and spatial dimensions; Based on the original dataset, instantaneous fluctuation features in the time dimension are extracted, and the signal features corresponding to different spatial locations are compared and analyzed within the same time window to obtain spatial distribution feature data in the spatial dimension. A long-term trend sequence of the signal is constructed based on the spatial distribution feature data, and boundary correction processing is performed on the long-term trend sequence at the endpoint regions of the time series to obtain a corrected time series. This includes: preserving low-frequency components in the spatial distribution feature data and performing a moving average operation on the processed data to obtain the long-term trend sequence; scanning the starting and ending segments of the long-term trend sequence, comparing the differences in the change amplitudes of the starting and ending segments with those of the non-endpoint segments to identify distortion points in the endpoint regions of the sequence, and obtaining a distortion boundary sequence; expanding the long-term trend sequence based on the distortion boundary sequence and calculating the corresponding sequence values ​​to obtain a preliminary corrected sequence; comparing the overall trend of the preliminary corrected sequence with that of the long-term trend sequence, and when the comparison result meets a preset consistency threshold, performing wavelet reconstruction processing on the preliminary corrected sequence to obtain the corrected time series. Spatial distribution deviation data is constructed based on the corrected time series and the spatial distribution characteristic data, and preliminary abnormal regions are determined based on the spatial distribution deviation data; The original signal corresponding to the preliminary abnormal region is extracted from the original dataset, and residual analysis is performed between the original signal and the corrected time series to determine the concentrated abnormal region. Spatial difference calculation and temporal fluctuation screening are performed on the spatial units corresponding to the anomaly concentration area, and the specific location of the fault area is determined based on the screening results; An alarm signal is generated based on the specific location of the fault area, and the spatial distribution difference data corresponding to the fault area is recorded to obtain the abnormal detection result of the micro-branch current of the low-voltage distribution network.

2. The low-voltage distribution network fault detection method according to claim 1, characterized in that, The process of acquiring current signal data and preprocessing the current signal data to obtain a raw dataset containing time and spatial dimensions includes: The current signal data is subjected to analog-to-digital conversion to obtain a digital current signal; Multi-scale frequency decomposition is performed on the digital current signal to obtain layered signal sequences corresponding to different frequency levels; Wavelet transform filtering is performed on the layered signal sequence to extract high-frequency and low-frequency components, respectively, to obtain frequency-separated signals; Based on the digital current signals corresponding to the frequency separation signals at different sampling time points, time alignment and unified indexing are performed to construct a time series structure; Based on pre-stored sensor configuration data, spatial indexing and reorganization processing is performed on the frequency-separated signals to construct a spatial distribution structure, thereby obtaining the original dataset containing both time and spatial dimensions.

3. The low-voltage distribution network fault detection method according to claim 1, characterized in that, The process involves extracting instantaneous fluctuation features in the time dimension based on the original dataset, and comparing and analyzing the signal features corresponding to different spatial locations within the same time window to obtain spatial distribution feature data in the spatial dimension, including: The original dataset is subjected to time-dimensional signal decomposition processing to extract instantaneous time fluctuation features and separate high-frequency noise components, forming a decomposed signal sequence. Based on the decomposed signal sequence, the instantaneous time fluctuation features corresponding to different spatial locations are compared within the same time window to obtain a spatially layered sequence that reflects spatial differences. The frequency domain energy distribution data is calculated based on the spatial layering sequence to obtain the energy distribution characteristics corresponding to each spatial layer; Identify the time points of energy changes from the energy distribution characteristics to obtain the location of the time changes; Based on the time-varying locations, a consistency analysis is performed on the energy changes at corresponding time points in different spatial levels, thereby obtaining spatial distribution characteristic data in the spatial dimension.

4. The low-voltage distribution network fault detection method according to claim 1, characterized in that, The step of constructing spatial distribution deviation data based on the corrected time series and the spatial distribution feature data, and determining preliminary abnormal regions based on the spatial distribution deviation data, includes: Based on the spatial distribution characteristic data, the change weight of each spatial level is calculated to obtain the set of level weights; Based on the hierarchical weight set, the data corresponding to the spatial level in the spatial distribution feature data are weighted and fused to obtain a fused spatial sequence. Based on the difference between the fused spatial sequence and the preset normal state reference sequence, the spatial distribution deviation value is calculated to form spatial distribution deviation data; Based on the comparison between the deviation value corresponding to each spatial unit in the spatial distribution deviation data and the preset deviation threshold, spatial units whose deviation values ​​exceed the preset deviation threshold are selected to obtain abnormal candidate regions. The local density is calculated based on the adjacency relationship of spatial units in the abnormal candidate region, and an initial abnormal region is determined when the local density exceeds a preset density threshold.

5. The low-voltage distribution network fault detection method according to claim 1, characterized in that, The step of extracting the original signal corresponding to the preliminary abnormal region from the original dataset and performing residual analysis on the original signal and the corrected time series to determine the concentrated abnormal region includes: Based on the spatial index corresponding to the preliminary abnormal region, the original signal of the corresponding spatial unit is extracted from the original dataset, and the corrected time series corresponding to the spatial unit is selected. The absolute difference between the original signal and the corrected time series is calculated at the same time point and the same spatial unit to obtain a set of residual values. When the deviation of the residual value set from the preset reference distribution data exceeds a preset deviation threshold, an inconsistent distribution sub-region is determined. Statistical analysis is performed on the adjacency relationships of spatial units in the inconsistent distribution sub-regions to calculate the local residual density, and the set of spatial units whose local residual density exceeds a preset density threshold is identified as an abnormally concentrated region.

6. The low-voltage distribution network fault detection method according to claim 1, characterized in that, The step of calculating spatial differences and filtering temporal fluctuations in spatial units corresponding to areas of concentrated anomalies, and determining the specific location of the fault area based on the filtering results, includes: Based on the spatial units corresponding to the anomaly concentration areas, the characteristic difference values ​​between adjacent spatial units are calculated to obtain spatial difference data. Select the spatial unit corresponding to the spatial difference data exceeding the preset difference threshold to obtain the spatial difference region; Based on the time series corresponding to the spatial difference region, calculate the time fluctuation parameter to obtain the time fluctuation value; When the time fluctuation value exceeds a preset time fluctuation threshold, the corresponding spatial unit is determined as a spatiotemporal related region; Select a set of spatial units that are adjacent to each other in the spatiotemporal correlation region and whose distribution density exceeds a preset density threshold, and determine the spatial location corresponding to the set of spatial units as the specific location of the fault area.

7. The low-voltage distribution network fault detection method according to claim 1, characterized in that, The process of generating an alarm signal based on the specific location of the fault area and recording the spatial distribution difference data corresponding to the fault area to obtain the abnormal detection result of the micro-branch current in the low-voltage distribution network includes: Based on the current signal corresponding to the specific location of the fault area, the current deviation value is calculated, and an initial alarm signal is generated when the current deviation value exceeds the preset current deviation threshold. Based on the spatial distribution difference data corresponding to the specific location of the fault area, extract the difference feature vector; The initial alarm signal and the difference feature vector are compared for consistency. When the feature intensity of the difference feature vector exceeds a preset energy threshold, an enhanced alarm signal is generated. The specific location of the fault area, the differential feature vector, and the alarm status of the enhanced alarm signal are recorded to form an anomaly detection result for the micro-branch current of the low-voltage distribution network.

8. A low-voltage distribution network fault detection system, characterized in that, include: The data processing module is used to acquire current signal data and preprocess the current signal data to obtain a raw dataset containing time and spatial dimensions. The spatial distribution module is used to extract instantaneous fluctuation features in the time dimension based on the original dataset, and to compare and analyze the signal features corresponding to different spatial locations within the same time window to obtain spatial distribution feature data in the spatial dimension. The time correction module is used to construct a long-term trend sequence of the signal based on the spatial distribution feature data, and to perform boundary correction processing on the long-term trend sequence in the endpoint region of the time series to obtain a corrected time series. This includes: preserving low-frequency components in the spatial distribution feature data and performing a moving average operation on the processed data to obtain the long-term trend sequence; scanning the start and end segments of the long-term trend sequence, comparing the difference in change amplitude between the start and end segments and non-endpoint segments to identify distortion points in the endpoint region of the sequence, and obtaining a distortion boundary sequence; expanding the long-term trend sequence based on the distortion boundary sequence and calculating the corresponding sequence value to obtain a preliminary corrected sequence; comparing the overall trend of the preliminary corrected sequence with the long-term trend sequence, and when the comparison result meets a preset consistency threshold, performing wavelet reconstruction processing on the preliminary corrected sequence to obtain the corrected time series. An anomaly extraction module is used to construct spatial distribution deviation data based on the corrected time series and the spatial distribution feature data, and to determine the preliminary anomaly region based on the spatial distribution deviation data; An anomaly focusing module is used to extract the original signal corresponding to the preliminary anomaly region from the original dataset, and perform residual analysis on the original signal and the corrected time series to determine the anomaly concentration region. The fault location module is used to calculate spatial differences and filter temporal fluctuations in spatial units corresponding to areas of concentrated anomalies, and determine the specific location of the fault area based on the filtering results. The anomaly detection module is used to generate an alarm signal based on the specific location of the fault area and record the spatial distribution difference data corresponding to the fault area to obtain the anomaly detection result of the micro-branch current of the low-voltage distribution network.

Images