A method and system for monitoring residual current of an alternating current power supply for a station

Abstract

This invention discloses a method and system for monitoring residual current of AC power supply in stations, relating to the field of electrical parameter measurement technology. The method includes: performing micro-amplitude reference self-check and acquisition deviation correction based on a time-scale aligned monitoring set; stitching together the spectral domains of low-frequency and high-frequency channels to generate a wideband calibrated residual current set; extracting frequency domain and transient features from the wideband calibrated residual current set; and using gated fusion to perform sparse inversion localization to obtain suspected abnormal feeders and target feature vectors; performing hierarchical cause discrimination based on the target feature vectors and suspected abnormal feeders; quantifying the deviation degree of the target feature vectors; and generating a handling judgment set. This invention generates self-check response records and forms acquisition deviation correction information through micro-amplitude reference self-check, achieving consistent calibration correction between low-frequency and high-frequency channel data to obtain a wideband calibrated residual current set, thereby stabilizing the wideband measurement reference.

Description

Technical Field

[0001] This invention relates to the field of electrical parameter measurement technology, and in particular to a method and system for monitoring the residual current of a station AC power supply. Background Technology

[0002] Station AC power supplies provide auxiliary power for secondary equipment, lighting, and DC system charging in substations and power plants. Leakage and insulation degradation are often reflected in the form of residual current. Existing projects mostly use zero-sequence current transformers or residual current sensors for online sampling, combined with power frequency bandwidth measurement and alarm threshold setting, supplemented by regular insulation testing and event logging, to achieve closed-loop monitoring and maintenance of the operating status of each feeder circuit.

[0003] When multiple feeder loads coexist, conventional power frequency monitoring is affected by data acquisition link drift and bandwidth differences, making it difficult to maintain consistent data over the long term. When harmonics and transients are intertwined, alarms struggle to locate abnormal feeders and differentiate between interference and insulation-related causes, requiring manual intervention for proper handling. Existing station residual current monitoring technology is prone to data drift and insufficient causation. Summary of the Invention

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

[0005] Therefore, this invention provides a method for monitoring residual current in station AC power supplies to solve the problems of inconsistent residual current monitoring standards and difficulty in accurately locating and determining the cause of alarms in multi-feeder scenarios.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides a method for monitoring the residual current of a station AC power supply, which includes collecting residual current data and feeder current monitoring data, and solidifying them through a monitoring topology table to generate a time-scale aligned monitoring set.

[0008] Based on the time-scale aligned monitoring set, a micro-amplitude reference self-test and acquisition deviation correction are performed, and the spectral domains of the low-frequency channel and the high-frequency channel are spliced ​​to generate a wideband calibration residual current set;

[0009] Frequency domain features and transient features are extracted from the broadband calibration residual current set, and sparse inversion localization is performed by gated fusion to obtain the feature vectors of suspected abnormal feeders and targets.

[0010] Based on the target feature vector and suspected abnormal feed lines, hierarchical cause discrimination is performed, and the deviation degree of the target feature vector is quantified to generate a treatment judgment set;

[0011] The residual current hazard level is obtained by mapping the risk classification based on the disposal judgment set, and the execution sequence of the control strategy is arranged to generate a power supply residual current report.

[0012] As a preferred embodiment of the residual current monitoring method for station AC power supply according to the present invention, the residual current data includes zero-sequence residual current measurement value, sampling channel identifier and monitoring point number;

[0013] The feeder current monitoring data includes the effective value of the feeder loop current, loop identification, and monitoring point number.

[0014] In a preferred embodiment of the method for monitoring residual current of station AC power supply according to the present invention, the step of solidifying the monitoring topology table to generate a time-scale aligned monitoring set includes the following steps:

[0015] Write the residual current data and feeder current monitoring data into a unified time stamp and monitoring point number to obtain a time-stamped sampling frame.

[0016] The sampled frames with timestamps are sorted and aligned according to their timestamps, and the sampled contents at the same time are merged to generate an aligned sampled sequence.

[0017] The aligned sampling sequence is used as input and matched with the monitoring topology table to determine the loop identifier and the relationship with the upstream power source corresponding to each sampling record, thereby generating a topology solidification record;

[0018] The topology-fixed records are encapsulated according to a unified time window and written with complete index information to generate a time-stamped aligned monitoring set.

[0019] As a preferred embodiment of the residual current monitoring method for station AC power supplies described in this invention, the steps for performing micro-amplitude reference self-check and acquisition deviation correction are as follows:

[0020] Trigger micro-amplitude benchmark self-test based on time-scale aligned monitoring set, extract benchmark response within corresponding time window and generate self-test response record;

[0021] The acquisition deviation is identified based on the self-test response record, and the gain compensation and zero-point compensation parameters are obtained according to the difference between the expected amplitude and phase of the reference sequence and the measured amplitude and phase, thereby generating acquisition deviation correction information.

[0022] The low-frequency channel data and high-frequency channel data in the time-scale alignment monitoring set are corrected for consistency based on the acquisition deviation correction information to obtain the calibration time-scale alignment monitoring set.

[0023] As a preferred embodiment of the residual current monitoring method for station AC power supplies described in this invention, the steps for splicing the spectral domains of the low-frequency channel and the high-frequency channel to generate a broadband calibration residual current set are as follows:

[0024] Denoising and outlier suppression are performed on the calibration time-scale alignment monitoring set while keeping the time window index unchanged to obtain the spectral domain preprocessed record;

[0025] Based on the spectral domain preprocessing record, the low-frequency channel and the high-frequency channel are spliced ​​in the overlapping frequency band, and the broadband calibration residual current set is obtained through amplitude and phase consistency verification.

[0026] As a preferred embodiment of the residual current monitoring method for station AC power supplies described in this invention, the steps of extracting frequency domain features and transient features from the broadband calibration residual current set, and performing sparse inversion localization using gated fusion to obtain the suspected abnormal feeder and target feature vectors are as follows.

[0027] Frequency domain analysis and transient analysis are performed on the wideband calibration residual current set to extract frequency domain features and transient features respectively, and then the dimensions are aligned and combined to generate a candidate feature set;

[0028] Based on the self-test response record, the detection peak value of the baseline sequence is calculated for low-frequency channel data and high-frequency channel data. When the detection peak value is lower than the baseline detection lower limit threshold, it is determined to be a low health channel.

[0029] Based on the monitoring topology table, the relationship between the circuit power supply object and the upstream power supply is extracted, and the failure impact range is obtained. When the failure impact range covers the security load, it is determined to be a high-importance circuit.

[0030] Gated fusion is performed on the candidate feature set to suppress candidate features corresponding to low health channels and enhance candidate features corresponding to high importance loops, thereby generating a gated fusion feature set;

[0031] Sparse inversion localization is performed on the gated fusion feature set to obtain and sort the feeder contribution of each feeder loop, and to locate suspected abnormal feeders.

[0032] The gated fusion feature set and the suspected abnormal feed line are encapsulated to form the target feature vector and maintain the same time window index as the suspected abnormal feed line to obtain the target feature vector.

[0033] As a preferred embodiment of the residual current monitoring method for station AC power supplies described in this invention, the step of determining the cause of hierarchical abnormality based on the target feature vector and suspected abnormal feeders includes the following steps:

[0034] Based on the suspected abnormal feeders, determine the handling feeder sequence and bind the handling feeder sequence to the target feature vector to generate the handling target feature vector;

[0035] Nonlinear load interference is determined by analyzing the feature vector of the treatment target to obtain a nonlinear load interference label; when the nonlinear load interference label is negative, insulation consistency is determined by comparing the feature vector of the treatment target with temperature monitoring data to obtain an insulation degradation label.

[0036] When the insulation degradation label is not specified, the target feature vector is input into the fault classification model to distinguish between grounding faults and equipment leakage, and the fault cause label is obtained.

[0037] As a preferred embodiment of the residual current monitoring method for station AC power supply described in this invention, the generation of the handling judgment set refers to quantifying the degree of deviation of the handling target feature vector according to the deviation quantization caliber to obtain the degree of abnormality, and combining the fault cause label, the degree of abnormality and the handling feeder sequence to generate the handling judgment set.

[0038] As a preferred embodiment of the method for monitoring the residual current of station AC power supply according to the present invention, the steps for arranging the execution timing of the control strategy and generating a power supply residual current report are as follows.

[0039] Select the classification criteria corresponding to the fault cause label according to the risk classification rules, and map the degree of abnormality to the risk classification to generate the residual current hazard level.

[0040] Based on the monitoring topology table, the sequence of feeder lines to be handled is mapped to a list of handling circuits, and the priority order of handling is determined in combination with the residual current hazard level;

[0041] According to the strategy arrangement rules, the residual current hazard level and the list of disposal circuits are matched with the residual current disposal strategy and the execution order is determined to generate the control strategy execution sequence.

[0042] The handling judgment set, residual current hazard level and control strategy execution sequence are summarized into structured content according to the loop relationship of the monitoring topology table to generate a power supply residual current report.

[0043] Secondly, the present invention provides a station AC power supply residual current monitoring system, including a data acquisition module, which collects residual current data and feeder current monitoring data, and solidifies them through a monitoring topology table to generate a time-scale aligned monitoring set;

[0044] The self-test correction module performs micro-amplitude reference self-test and acquisition deviation correction based on the time-scale aligned monitoring set, and splices the spectral domains of the low-frequency channel and the high-frequency channel to generate a wideband calibration residual current set.

[0045] The anomaly identification module extracts frequency domain features and transient features from the wideband calibration residual current set, and performs sparse inversion localization using gated fusion to obtain the feature vectors of suspected abnormal feeders and targets.

[0046] The hierarchical discrimination module performs hierarchical cause discrimination based on the target feature vector and the suspected abnormal feed line, quantifies the deviation degree of the target feature vector, and generates a treatment judgment set;

[0047] The report generation module performs risk classification mapping based on the handling judgment set to obtain the residual current hazard level, and arranges the execution sequence of control strategies to generate a power supply residual current report.

[0048] The beneficial effects of this invention are as follows: by generating a self-test response record through micro-amplitude benchmark self-test and forming acquisition deviation correction information, consistent aperture correction and spectral domain splicing of low-frequency channel data and high-frequency channel data are achieved to obtain a wideband calibration residual current set, thereby stabilizing the wideband measurement benchmark; based on the candidate feature set combined with the self-test response record to identify low-health channels and combined with the monitoring topology table to identify high-importance loops to complete the gating fusion feature set, sparse inversion positioning is performed to obtain suspected abnormal feeders and target feature vectors, thereby enhancing the directionality of abnormal positioning and the traceability of the basis for handling. Attached Figure Description

[0049] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0050] Figure 1 This is a flowchart of a method for monitoring the residual current of AC power supplies used in stations.

[0051] Figure 2 This is a schematic diagram of a residual current monitoring system for AC power supplies used in stations.

[0052] Figure 3 This is a graph showing the change in the proportion of low-health channels as a function of the center of the time window.

[0053] Figure 4 This is a graph showing the variation of low-frequency gain compensation parameters with the center of the time window. Detailed Implementation

[0054] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0055] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0056] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0057] Reference Figures 1-4 As one embodiment of the present invention, this embodiment provides a method for monitoring the residual current of a station AC power supply, comprising the following steps:

[0058] S1. Collect residual current data and feeder current monitoring data, and solidify them through the monitoring topology table to generate a time-scale aligned monitoring set.

[0059] The residual current data includes the zero-sequence residual current measurement value, sampling channel identifier, and monitoring point number.

[0060] Feeder current monitoring data includes the effective value of the feeder loop current, loop identification, and monitoring point number.

[0061] The residual current data and feeder current monitoring data are written with a unified time stamp and monitoring point number to obtain a time-stamped sampling frame.

[0062] Furthermore, the zero-sequence residual current measurement value, sampling channel identifier, and monitoring point number are extracted from the residual current data, and the effective value of the feeder loop current, loop identifier, and monitoring point number are extracted from the feeder current monitoring data. A unified time synchronization timestamp is generated based on the unified time synchronization, and the residual current data records and feeder current monitoring data records are associated according to the monitoring point number. The unified time synchronization timestamp and monitoring point number are written to the corresponding associated residual current data records, and the unified time synchronization timestamp and monitoring point number are written to the corresponding associated feeder current monitoring data records. The written residual current data records and written feeder current monitoring data records are encapsulated into sampling units according to the unified time synchronization timestamp and monitoring point number to obtain a time-stamped sampling frame.

[0063] The sampled frames with timestamps are sorted and aligned according to their timestamps, and the sampled contents at the same time are merged to generate an aligned sampled sequence.

[0064] Furthermore, using time-stamped sampling frames as input, the unified time synchronization timestamp field and monitoring point number field are extracted, and duplicate frames are removed from the time-stamped sampling frames to obtain deduplicated time-stamped sampling frames. The deduplicated time-stamped sampling frames are then sorted in ascending order based on the unified time synchronization timestamps to obtain a sorted time-stamped sampling frame sequence. The sorted time-stamped sampling frame sequence is then grouped according to the monitoring point number to obtain a grouped time-stamped sampling frame sequence. Within the grouped time-stamped sampling frame sequence, a timestamp index is established based on the unified time synchronization timestamp, and sampling units corresponding to the same unified time synchronization timestamp are aligned to obtain a timestamp-aligned sampling frame group. The sampling content of the same time moment is merged within the timestamp-aligned sampling frame group, and the zero-sequence residual current measurement value, sampling channel identifier, feeder loop current effective value, and loop identifier are retained according to the unified time synchronization timestamp and monitoring point number to obtain a merged sampling record. The merged sampling record is then encapsulated and output according to the unified time synchronization timestamp order to generate an aligned sampling sequence.

[0065] The aligned sampling sequence is used as input and matched with the monitoring topology table to determine the loop identifier and the relationship with the upstream power supply corresponding to each sampling record, thereby generating a topology solidification record.

[0066] Furthermore, the alignment sampling sequence is used as input to read the monitoring point number, loop identifier, unified time synchronization timestamp, and sampling content from each merged sampling record. The monitoring point number is used as the matching key to retrieve the corresponding topology entry in the monitoring topology table. When the monitoring topology table search finds a match, the loop identifier bound to the topology entry and its relationship with the upper-level power source are extracted from the monitoring topology table and written into the merged sampling record to form a topology mapping sampling record. When the monitoring topology table search does not find a match, a missing marker is written to the topology mapping sampling record, and the monitoring point number, unified time synchronization timestamp, and sampling content are retained to ensure subsequent alignment consistency. Loop identifier consistency verification and upper-level power source relationship integrity verification are performed on the topology mapping sampling record to obtain the verified topology mapping sampling record. The verified topology mapping sampling record is encapsulated and output in the order of the unified time synchronization timestamp to generate a topology solidification record.

[0067] It should be noted that the loop identifier refers to the identifier field in the monitoring topology table used to uniquely identify the feeder loop to which the monitoring point number belongs. The loop identifier is used to establish a correspondence between the merged sampling records in the aligned sampling sequence and the specific feeder loop. The loop identifier and the loop identifier in the feeder current monitoring data maintain the same naming standard and are used to summarize the residual current and the effective value of the feeder loop current according to the feeder loop dimension in subsequent steps.

[0068] The upstream power supply relationship refers to the topology field in the monitoring topology table used to describe the power supply connection relationship between the feeder circuit corresponding to the circuit identifier and the upstream power supply. The upstream power supply relationship includes at least the upstream power supply identifier and connection direction information. The upstream power supply relationship is used to determine the circuit power supply object and the upstream power supply relationship in subsequent steps and to deduce the failure impact range, thereby supporting the identification of high-importance circuits and the mapping of the circuit list for handling.

[0069] The sampling content is the measurement information carried in each sampling record, such as the residual current measurement value at that moment, the timestamp, and the monitoring point number.

[0070] The topology-fixed records are encapsulated according to a unified time window and written with complete index information to generate a time-stamped aligned monitoring set.

[0071] Furthermore, the topology-fixed records are segmented and grouped according to a unified time window, and the time window index corresponding to each unified time window is determined based on the unified time synchronization timestamp. Complete index information is written to each unified time window, which includes the time window index, the unified time window start timestamp, the unified time window end timestamp, the monitoring point number index, the loop identifier index, and the upper-level power supply relationship index. The complete index information and the topology-fixed records in the corresponding unified time window are encapsulated together to generate a time-aligned monitoring set organized according to the unified time window and maintaining the consistency of the time window index.

[0072] S2. Perform micro-amplitude reference self-check and acquisition deviation correction based on the time-scale aligned monitoring set, and stitch the spectral domains of the low-frequency channel and the high-frequency channel together to generate a wideband calibration residual current set;

[0073] The micro-amplitude baseline self-test is triggered based on the time-scale aligned monitoring set, and the baseline response is extracted within the corresponding time window to generate a self-test response record.

[0074] Furthermore, based on the time-aligned monitoring set, the self-test configuration parameters are read and the micro-amplitude reference self-test trigger cycle is determined. The micro-amplitude reference self-test trigger time is written into the unified time synchronization timestamp and the corresponding time window index is located to complete the micro-amplitude reference self-test trigger. Using the time window index as the positioning condition, the residual current waveform data within the corresponding time window is extracted from the time-aligned monitoring set. Based on the self-test configuration parameters, the reference sequence frequency or code and the upper limit of the self-test amplitude and phase reference are limited to generate a reference response sampling segment. The reference response sampling segment is encapsulated with the unified time synchronization timestamp, time window index, monitoring point number, and loop identifier to generate a self-test response record.

[0075] It should be noted that the residual current waveform data refers to the zero-sequence residual current measurement value sequence in the time-aligned monitoring set, which is distinguished by the sampling channel identifier and continuously arranged according to the unified time stamp. It is obtained by extracting the sampling content from the time-aligned monitoring set according to the time window index and the sampling channel identifier.

[0076] The acquisition deviation is identified based on the self-test response record, and the gain compensation and zero-point compensation parameters are obtained according to the difference between the expected amplitude and phase of the reference sequence and the measured amplitude and phase, thereby generating acquisition deviation correction information.

[0077] Furthermore, the reference response sampling segment and self-test configuration parameters are extracted based on the self-test response record. The expected amplitude and phase of the reference sequence are determined based on the self-test configuration parameters. For example, the self-test configuration parameters provide the upper limit of the self-test amplitude and the phase reference, and limit the frequency point or code pattern of the reference sequence. The expected amplitude and phase of the reference sequence are determined by the upper limit of the self-test amplitude, and the expected phase position is determined by the phase reference and the zero phase of the reference clock. Together with the frequency point or code pattern of the reference sequence, they constitute the expected amplitude and phase. Synchronous detection is performed on the reference response sampling segment to obtain the measured amplitude and phase of the reference sequence. The difference between the expected amplitude and the measured amplitude and phase of the reference sequence is calculated, and the direction and magnitude of the difference deviation are determined to identify the acquisition deviation. The gain compensation parameter is obtained based on the ratio of the amplitude difference to the expected amplitude, and the zero-point compensation parameter is obtained based on the phase reference and the zero-point reference. The gain compensation parameter and the zero-point compensation parameter are encapsulated with the unified time stamp, time window index, monitoring point number, and loop identifier to generate acquisition deviation correction information.

[0078] It should be noted that the reference sequence refers to the preset known waveform signal sequence output by the injection terminal during micro-amplitude reference self-test. It is generated according to the self-test configuration parameters and injected into the corresponding circuit by the injection terminal. The expected amplitude and phase refer to the amplitude and phase reference that the reference sequence should present at the injection point according to the self-test configuration parameters (such as self-test trigger period, self-test sequence frequency or code pattern, self-test amplitude upper limit and phase reference) set by the injection terminal during micro-amplitude reference self-test (determined by the set amplitude of the injection circuit, the zero phase of the reference clock and the definition of the reference sequence). The "measured amplitude and phase" refer to the amplitude and phase obtained by synchronous detection method for the known frequency or code pattern of the reference sequence after the corresponding waveform is intercepted from the time-aligned monitoring set within the same injection time window (the method is to perform discrete Fourier estimation of the target frequency point or perform correlation demodulation with the reference sequence to obtain in-phase and quadrature components and then convert them into amplitude and phase). The difference between the two is the gain deviation and phase deviation introduced by the acquisition link, which is used to generate gain compensation and zero-point compensation parameters and write the acquisition deviation correction information.

[0079] Acquisition deviation correction information refers to the set of compensation parameters used to correct acquisition link errors, such as gain compensation parameters and zero-point compensation parameters.

[0080] The low-frequency channel data and high-frequency channel data in the time-scale alignment monitoring set are corrected for consistency based on the acquisition deviation correction information to obtain the calibration time-scale alignment monitoring set.

[0081] Furthermore, gain compensation parameters and zero-point compensation parameters are extracted based on the acquisition deviation correction information, and the time window index and monitoring point number bound to the acquisition deviation correction information are read to locate the corresponding time window record in the time-aligned monitoring set. Zero-point compensation parameter correction is performed point by point on the low-frequency channel data and high-frequency channel data in the corresponding time window record, and then gain compensation parameter correction is performed to obtain low-frequency channel calibration data and high-frequency channel calibration data. The low-frequency channel calibration data and high-frequency channel calibration data are encapsulated back into the time-aligned monitoring set with the unified time stamp, monitoring point number, loop identifier and the relationship with the upper-level power supply to obtain the calibration time-aligned monitoring set.

[0082] It should be noted that the low-frequency channel data is the residual current waveform data collected by the low-frequency DC sensitive channel in the time-scaled alignment monitoring set, while the high-frequency channel data is the residual current waveform data collected by the high-frequency sensitive channel in the time-scaled alignment monitoring set.

[0083] like Figure 3 The curves of the low-frequency gain compensation parameter (mean) in the acquisition deviation correction information as a function of the time window center are given. Under the three levels of low / medium / high acquisition deviation, the parameter adaptively adjusts stepwise with the window center and fluctuates narrowly around 1.0 for a long time. This shows that the consistent aperture correction can still stabilize the aperture under different deviation intensities, support the amplitude and phase consistency verification of overlapping frequency bands and spectral domain splicing, and obtain a stable broadband calibration residual current set.

[0084] Denoising and outlier suppression are performed on the calibration time-scale alignment monitoring set while keeping the time window index unchanged to obtain the spectral domain preprocessed record.

[0085] Furthermore, the time window index, low-frequency channel data, and high-frequency channel data of the calibration time scale alignment monitoring set are read one by one, and the processing range is locked by the time window index. Denoising processing is performed on the low-frequency channel data and high-frequency channel data to suppress out-of-band noise and random jitter in the low-frequency channel data and high-frequency noise and spike interference in the high-frequency channel data, resulting in low-frequency denoised data and high-frequency denoised data. Based on the low-frequency denoised data and high-frequency denoised data, the criteria for anomaly point judgment are calculated and anomaly point suppression is performed. Anomalies are replaced with neighborhood consistent values ​​or interpolation results within the time window, resulting in low-frequency anomaly point suppressed data and high-frequency anomaly point suppressed data, respectively. The low-frequency anomaly point suppressed data and high-frequency anomaly point suppressed data are encapsulated and output with the time window index to obtain the spectral domain preprocessed record.

[0086] Based on the spectral domain preprocessing record, the low-frequency channel and the high-frequency channel are spliced ​​in the overlapping frequency band, and the broadband calibration residual current set is obtained through amplitude and phase consistency verification.

[0087] Furthermore, based on the spectral domain preprocessing records, low-frequency and high-frequency outlier suppression data corresponding to the time window index are extracted, and frequency domain transformation is performed on the low-frequency and high-frequency outlier suppression data to obtain the low-frequency channel spectrum and high-frequency channel spectrum. The common frequency range of the low-frequency and high-frequency channel spectra within the overlapping frequency band is determined according to the overlapping frequency band, and the amplitude-phase consistency verification result is calculated within the common frequency range. When the amplitude-phase consistency verification result meets the consistency condition, the low-frequency and high-frequency channel spectra are aligned according to the overlapping frequency band and spliced ​​in the spectral domain to obtain a wideband spliced ​​spectrum covering the frequency range of the low-frequency and high-frequency channels. The wideband spliced ​​spectrum is encapsulated with the time window index and inversely transformed to obtain the wideband calibration residual current waveform, generating a wideband calibration residual current set.

[0088] It should be noted that the overlapping frequency band is a common frequency range in which both the low-frequency channel and the high-frequency channel have effective measurement capabilities in the frequency response range. It is used for amplitude and phase consistency verification and spectral domain splicing alignment.

[0089] S3. Extract frequency domain features and transient features from the broadband calibration residual current set, and use gated fusion to perform sparse inversion localization to obtain the feature vectors of suspected abnormal feeders and targets.

[0090] Frequency domain analysis and transient analysis are performed on the wideband calibration residual current set to extract frequency domain features and transient features respectively. The features are then aligned and combined to generate a candidate feature set.

[0091] Furthermore, based on the time window index, the broadband calibration residual current waveform within each time window is extracted from the broadband calibration residual current set, and frequency domain analysis is performed on the broadband calibration residual current waveform to obtain a spectral representation. From the spectral representation, the power frequency component intensity, harmonic proportion characteristics, and DC bias characteristics are extracted to form frequency domain features. Simultaneously, transient analysis is performed on the broadband calibration residual current waveform to locate abrupt change segments and calculate the transient change intensity, abrupt change duration, and change rate characteristics to form transient features. The frequency domain features and transient features are dimensionally aligned according to a preset feature order. Dimensional alignment includes fixing the feature dimensions, unifying the dimensions, and filling missing features. The dimensionally aligned frequency domain features and dimensionally aligned transient features are combined while keeping the time window index unchanged to generate a candidate feature set.

[0092] It should be noted that the feature order refers to the arrangement order of frequency domain features and transient features when aligning and combining dimensions, and is set according to the fixed field positions of frequency domain features and transient features in the candidate feature set.

[0093] Frequency domain characteristics include power frequency component intensity, harmonic proportion characteristics, and DC bias characteristics.

[0094] Transient characteristics include the intensity of transient changes, the duration of mutations, and the rate of change.

[0095] Based on the self-test response record, the detection peak value of the baseline sequence is calculated for low-frequency channel data and high-frequency channel data. When the detection peak value is lower than the baseline detection lower limit threshold, it is determined to be a low health channel.

[0096] Furthermore, based on the self-test response record, the self-test time window index is located, and the low-frequency channel data and high-frequency channel data corresponding to the self-test time window index are extracted. At the same time, the benchmark sequence is extracted from the self-test response record. For the low-frequency channel data, the benchmark sequence correlation calculation is performed by traversing the self-test time window index by time shift and the maximum correlation amplitude is extracted to obtain the low-frequency channel benchmark sequence detection peak value. For the high-frequency channel data, the benchmark sequence correlation calculation is performed by traversing the self-test time window index by time shift and the maximum correlation amplitude is taken to obtain the high-frequency channel benchmark sequence detection peak value. The low-frequency channel benchmark sequence detection peak value and the high-frequency channel benchmark sequence detection peak value are compared with the benchmark detection lower limit threshold respectively. When the low-frequency channel benchmark sequence detection peak value is lower than the benchmark detection lower limit threshold, the low-frequency channel is determined to be a low-health channel. When the high-frequency channel benchmark sequence detection peak value is lower than the benchmark detection lower limit threshold, the high-frequency channel is determined to be a low-health channel. Otherwise, they are determined to be high-health channels.

[0097] It should be noted that the lower limit threshold of the benchmark detection is set based on the statistical distribution of the detection peak of the benchmark sequence in the self-test response record under the channel health state, the upper limit of the self-test amplitude corresponding to the micro-amplitude benchmark self-test, and the noise floor within the self-test time window. An exemplary value range is 0.3–0.7 times the average detection peak of the benchmark sequence under the channel health state.

[0098] The peak detection value of the baseline sequence is calculated using the following expression:

[0099] ;

[0100] in, It represents a channel. In all time shifts The peak value of the baseline sequence obtained below, Indicates the channel index. This indicates the time shift used when performing alignment search between the reference sequence and the channel data. This indicates the sampling point number within the self-test time window. This indicates the total number of sampling points included in the calculation within the self-test time window. Indicates the channel within the self-test time window. The Each sample value, Indicates the reference sequence By time shift After aligning and translating, at the first The sequence values ​​corresponding to each position. Indicates the low-frequency channel. Indicates a high-frequency channel.

[0101] Based on the monitoring topology table, the relationship between the circuit power supply object and the upstream power supply is extracted, and the failure impact range is obtained. When the failure impact range covers the security load, it is determined to be a high-importance circuit.

[0102] Furthermore, based on the monitoring topology table, the relationship between the circuit power supply object corresponding to the circuit identifier and the upstream power source is read, and the downstream connection list of the circuit power supply object is determined using the relationship between the circuit identifier and the upstream power source as the search condition. The downstream connection list is expanded along the power supply connection direction based on the upstream power source relationship and summarized to obtain the failure impact range. The failure impact range consists of the set of circuit power supply objects corresponding to the circuit identifier and the set of downstream connection lists. The monitoring point number and circuit identifier list corresponding to the security load are read from the monitoring topology table, and the coverage judgment is performed between the circuit identifier list corresponding to the security load and the failure impact range. When the failure impact range covers the circuit identifier list corresponding to the security load, the circuit power supply object corresponding to the circuit identifier is determined to be a high-importance circuit and marked as a high-importance circuit.

[0103] It should be noted that security load refers to the set of key power users marked in the monitoring topology table as requiring priority to ensure continuous power supply even in the event of power supply anomalies.

[0104] Gated fusion is performed on the candidate feature set to suppress candidate features corresponding to low health channels and enhance candidate features corresponding to high importance loops, thereby generating a gated fusion feature set.

[0105] Furthermore, the time window index, monitoring point number, loop identifier, frequency domain features, and transient features of each candidate feature set are read one by one. A low-health channel mapping table is established based on the low-health channel determination results, and a high-importance loop mapping table is established based on the high-importance loop determination results. Gating suppression is performed on the candidate features corresponding to the low-health channels in the candidate feature set, adjusting or setting the candidate features corresponding to the low-health channels to zero according to the gating weight to reduce their contribution to subsequent positioning. Gating enhancement is performed on the candidate features corresponding to the high-importance loops in the candidate feature set, adjusting the candidate features corresponding to the high-importance loops to increase their contribution to subsequent positioning. The candidate features that have completed gating suppression and gating enhancement are encapsulated and output in a manner consistent with the time window index, monitoring point number, and loop identifier to generate a gating fusion feature set.

[0106] It should be noted that the gating weights are set based on the low health channel determination results and the high importance loop determination results. For example, when the low health channel determination result is true, the candidate feature corresponding to the low health channel is subject to a suppression gating weight; when the high importance loop determination result is true, the candidate feature corresponding to the high importance loop is subject to an enhancement gating weight; when both the low health channel determination result and the high importance loop determination result are false, the corresponding candidate feature is subject to a baseline gating weight.

[0107] like Figure 4 The curve showing the proportion of low-health channels as a function of the time window center is presented. As the magnitude of the acquisition deviation increases, the curve shifts upward and exhibits stronger fluctuations. The difference between the moment of maximum difference and the vertical axis value is magnified and annotated to determine the gating threshold and weighting. After gating the low-health channels and combining it with the high-importance loop constraints of the monitoring topology table, the gating fusion feature set is more concentrated on reliable channels, improving the directional accuracy and traceability of sparse inversion locating suspected abnormal feeders.

[0108] Sparse inversion localization is performed on the gated fusion feature set to obtain and sort the feeder contribution of each feeder loop, and to locate suspected abnormal feeders.

[0109] Furthermore, the gated fusion feature sets are grouped according to time window indices, and the loop identifier set and gated fusion feature vector corresponding to each group are extracted. A mapping relationship between loop identifiers and monitoring point numbers is established based on the monitoring topology table, forming sparse inversion positioning constraints. Using the gated fusion feature vectors and sparse inversion positioning constraints as input, a loop contribution relationship table is constructed based on the connection relationship between loop identifiers and monitoring point numbers in the monitoring topology table, and the gated fusion feature vectors are fitted to the loop contribution relationship table. During the fitting process, sparsity constraints are applied to the contribution parameters corresponding to the loop identifiers, ensuring that the loop identifiers... The corresponding contribution parameters are preferentially concentrated on a few loop identifiers; the contribution residuals are repeatedly calculated and the contribution parameters are updated using an iterative update method until the contribution residuals no longer decrease or the number of iterations reaches the upper limit, and the sparse coefficients corresponding to each loop identifier in the loop identifier set are obtained. The amplitude of the sparse coefficients is normalized to the feeder contribution degree; the feeder contribution degree is summarized by loop identifier and sorted in descending order of feeder contribution degree to obtain the feeder contribution degree ranking table; the loop identifiers with the highest feeder contribution degree ranking are selected as suspected abnormal feeders according to the feeder contribution degree ranking table, and the suspected abnormal feeders are output in a manner consistent with the time window index.

[0110] It should be noted that sparsity constraint refers to the constraint rule in sparse inversion localization that limits the number of non-zero contribution parameters by adding sparsity penalty to contribution parameters and performing threshold compression to zero on small contribution parameters. For example, the part of the contribution parameter below the threshold is compressed to zero so that only the effective contribution of a few loop identifiers is retained.

[0111] The gated fusion feature set and the suspected abnormal feed line are encapsulated to form the target feature vector and maintain the same time window index as the suspected abnormal feed line to obtain the target feature vector.

[0112] Furthermore, the gated fusion feature set is grouped according to the time window index, and the corresponding gated fusion feature vector and loop identifier set are extracted. At the same time, the time window index and loop identifier corresponding to the suspected abnormal feeder are read. The gated fusion feature vector and the suspected abnormal feeder are matched for consistency based on the time window index to obtain the matched gated fusion feature vector. The loop identifier position corresponding to the suspected abnormal feeder is marked in the matched gated fusion feature vector, and the suspected abnormal feeder and the matched gated fusion feature vector are encapsulated into a target feature vector according to a unified data structure. The target feature vector is written into the time window index and kept consistent with the suspected abnormal feeder to obtain the target feature vector.

[0113] S4. Based on the target feature vector and the suspected abnormal feed line, perform hierarchical cause discrimination, quantify the deviation degree of the target feature vector, and generate a treatment judgment set.

[0114] Based on the suspected abnormal feeder, determine the handling feeder sequence and bind the handling feeder sequence to the target feature vector to generate the handling target feature vector.

[0115] Furthermore, based on the suspected abnormal feeder, the loop identifier and time window index are read, and the set of loop identifiers ranked first according to the feeder contribution is selected to form a treatment feeder sequence; the treatment feeder sequence and the target feature vector are matched for consistency according to the time window index, and the treatment feeder sequence is written into the loop identifier field of the target feature vector to complete the binding; the treatment feeder sequence index information is added to the bound target feature vector while keeping the time window index unchanged to generate the treatment target feature vector.

[0116] Nonlinear load interference is discriminated against the feature vector of the target to obtain a nonlinear load interference label.

[0117] Furthermore, the frequency domain features and transient features in the target feature vector are read, and the coupling relationship between the harmonic proportion feature and the power frequency component intensity is calculated based on the nonlinear load interference discrimination caliber. At the same time, the load switching disturbance mode is identified by combining the transient change intensity and change rate features. When the target feature vector satisfies the nonlinear load harmonic interference feature, a positive label is written; when the target feature vector does not satisfy the nonlinear load harmonic interference feature, a negative label is written, thus obtaining the nonlinear load interference label and maintaining a binding relationship with the treatment feeder sequence.

[0118] When the nonlinear load interference tag is not present, insulation consistency is judged by using the target feature vector and temperature monitoring data to obtain the insulation degradation tag.

[0119] Furthermore, under the condition that the nonlinear load interference label is negative, the time window index and the treatment feeder sequence of the treatment target feature vector are read, and the corresponding sampling record of the temperature monitoring data is located by mapping the monitoring point number according to the treatment feeder sequence; the temperature monitoring data is aligned according to the time window index and the temperature change intensity and temperature change rate index are extracted, and a consistency comparison is performed with the DC bias feature and transient change intensity in the treatment target feature vector; when the consistency comparison meets the insulation degradation consistency judgment condition, a positive label is written, and when the consistency comparison does not meet the insulation degradation consistency judgment condition, a negative label is written, thus obtaining the insulation degradation label and maintaining a binding relationship with the treatment feeder sequence.

[0120] When the insulation degradation label is not specified, the target feature vector is input into the fault classification model to distinguish between grounding faults and equipment leakage, and the fault cause label is obtained.

[0121] Furthermore, under the condition that the insulation degradation label is not present, the frequency domain features and transient features of the treatment target feature vector are read while keeping the time window index and treatment feeder sequence unchanged. The frequency domain features and transient features are concatenated according to the input format of the fault classification model to form the model input features. The model input features are input into the fault classification model to obtain the ground fault discrimination result and the equipment leakage discrimination result, and the ground fault label or equipment leakage label is written according to the discrimination result. The ground fault label or equipment leakage label is bound to the treatment target feature vector while keeping the time window index consistent to obtain the fault cause label.

[0122] It should be noted that the fault classification model uses an XGBoost classifier to classify grounding faults and equipment leakage into two categories. The parameter settings are as follows: tree depth of 4–6 to limit model complexity and reduce overfitting risk; learning rate of 0.05–0.1 for stable convergence; number of weak learners of 200–600 combined with early stopping to balance fitting and generalization capabilities; subsampling ratio of 0.7–0.9 and feature sampling ratio of 0.7–0.9 to reduce variance; minimum leaf sample weight of 1–5 to suppress the influence of noisy samples; L2 regularization coefficient of 1–10 and L1 regularization coefficient of 0–1 to improve robustness. The training method involves constructing a training feature set using frequency domain features and transient features from the target feature vector; using grounding fault conclusions and equipment leakage conclusions confirmed in maintenance records or fault handling records as labels; aligning with time window indices to obtain supervised training samples; using training / validation set separation and stratified sampling to maintain class proportions; using cross-validation to select the optimal parameter combination; and using validation set AUC or F1 as the early stopping metric to complete training.

[0123] Nonlinear load interference tags are used to characterize whether the target feature vector conforms to the characteristics of nonlinear load harmonic interference, including positive and negative tags. Insulation degradation tags are used to characterize whether the target feature vector and temperature monitoring data show consistent insulation degradation characteristics, including positive and negative tags. Fault cause tags are used to characterize the type of abnormal cause corresponding to the target feature vector, including ground fault tags and equipment leakage tags.

[0124] The deviation degree of the target feature vector is quantified according to the deviation quantization caliber to obtain the degree of anomaly. The fault cause label, the degree of anomaly, and the handling feeder sequence are combined to generate the handling judgment set.

[0125] Furthermore, based on the deviation quantization caliber, the target feature vector reference caliber required for deviation degree quantization is read, and frequency domain features and transient features are extracted from the target feature vector while maintaining consistency with the time window index. The deviation values ​​relative to the reference caliber for the frequency domain features and transient features are calculated according to the deviation quantization caliber and normalized to obtain the feature deviation degree. The feature deviation degrees are weighted and summarized according to the deviation quantization caliber to obtain the deviation degree quantization result and output the anomaly degree. The fault cause label and the handling feeder sequence are read and consistently bound with the anomaly degree according to the time window index. The fault cause label, anomaly degree, and handling feeder sequence are encapsulated into the same structured record while keeping the time window index unchanged to generate the handling judgment set.

[0126] It should be noted that the deviation quantization caliber refers to the set of quantization rules used to specify the calculation method, normalization method, and weighted summarization method for the deviation values ​​of frequency domain features and transient features in the feature vector of the disposal target relative to the benchmark caliber.

[0127] The handling judgment set includes fault cause label, abnormality level and handling feeder sequence.

[0128] S5. Based on the handling judgment set, perform risk classification mapping to obtain the residual current hazard level, arrange the execution sequence of control strategy, and generate a power supply residual current report.

[0129] Select the classification criteria corresponding to the fault cause label according to the risk classification rules, and map the degree of abnormality to the risk classification to generate the residual current hazard level.

[0130] Furthermore, the fault cause label, anomaly degree, and handling feeder sequence in the handling judgment set are read and kept consistent with the time window index; according to the risk classification rules, the classification caliber matching the fault cause label is retrieved from the classification caliber set, and the classification caliber gives the segment boundary from the anomaly degree to the risk level; the anomaly degree is input into the classification caliber to perform risk classification mapping and output the risk level result to form the residual current hazard level; the residual current hazard level is bound to the fault cause label, anomaly degree, and handling feeder sequence according to the time window index to provide input for subsequent handling loop list mapping and control strategy execution timing generation.

[0131] It should be noted that risk classification rules refer to a set of rules used to define the correspondence between fault cause labels and classification criteria, as well as the segmentation boundaries and judgment methods from the degree of abnormality to the residual current hazard level. Risk classification rules are set by operation and maintenance management requirements and security load protection requirements.

[0132] The classification criteria refer to the quantitative rules that define the mapping boundary and judgment method from the degree of abnormality to the residual current hazard level for a certain fault cause label.

[0133] Based on the monitoring topology table, the sequence of feeder lines to be handled is mapped to a list of handling circuits, and the priority order of handling is determined in combination with the residual current hazard level.

[0134] Furthermore, the loop identifiers in the feeder sequence are read, and the relationship between the loop power supply object associated with each loop identifier and the upstream power source is retrieved in the monitoring topology table. These are then compiled into a list of loops to be handled. The residual current hazard level is bound to the loop identifier entries in the list of loops to be handled. The handling priority order is generated based on three rules: the order of residual current hazard level, the proximity of the upstream power source relationship, and whether the loop power supply object covers the security load. The handling priority order is written into the list of loops to be handled. The list of loops to be handled maintains consistency with the time window index and is used for subsequent strategy orchestration.

[0135] According to the strategy arrangement rules, the residual current hazard level is matched with the list of disposal circuits, the residual current disposal strategy is matched and the execution order is determined, and the control strategy execution sequence is generated.

[0136] Furthermore, the system reads the list of handling loops with residual current hazard levels and handling priorities, selects a set of residual current handling strategies matching the residual current hazard levels according to the strategy arrangement rules, and matches the residual current handling strategy actions to each loop in the handling loop list; generates an execution queue according to the handling priority, and sets mutual exclusion and minimum interval for loop identifiers under the same upper-level power supply relationship based on the upper-level power supply relationship to avoid concurrent actions under the same upper-level power supply relationship; encapsulates the loop identifier, strategy action, planned execution time, and constraint information to generate a control strategy execution sequence and maintains consistency in the time window index.

[0137] It should be noted that the strategy orchestration rules refer to the set of rules used to specify the matching relationship between the residual current hazard level and the list of disposal circuits to the residual current disposal strategy, as well as the sequential constraints and mutual exclusion conditions of the execution sequence of the control strategy. The strategy orchestration rules are set by the power distribution operation procedures and the requirements for continuous power supply to the security load.

[0138] The residual current handling strategy refers to the set of handling actions and execution constraints determined and executed based on the residual current hazard level for the circuit identifier in the handling circuit list. For example, it includes issuing alarm prompts, time-sharing retest confirmation, graded load limiting, graded circuit breaker isolation, and dispatching maintenance work orders to the power supply object corresponding to the circuit identifier.

[0139] The handling judgment set, residual current hazard level and control strategy execution sequence are summarized into structured content according to the loop relationship of the monitoring topology table to generate a power supply residual current report.

[0140] Furthermore, after aligning the judgment set, residual current hazard level, and time window index of the control strategy execution sequence, the loop identifier, the loop power supply object, and the relationship with the superior power source are extracted based on the monitoring topology table, and a loop relationship index is established. The fault cause label, abnormality level, handling feeder sequence, and residual current hazard level are written into the loop relationship index, and the strategy action and execution time corresponding to the loop identifier in the control strategy execution sequence are added. The loop relationship index is summarized according to the superior power source relationship level and a structured record is output to generate a power source residual current report.

[0141] This embodiment also provides a station AC power supply residual current monitoring system, including: a data acquisition module, which collects residual current data and feeder current monitoring data, and solidifies them through a monitoring topology table to generate a time-scale aligned monitoring set;

[0142] The self-test correction module performs micro-amplitude reference self-test and acquisition deviation correction based on the time-scale aligned monitoring set, and splices the spectral domains of the low-frequency channel and the high-frequency channel to generate a wideband calibration residual current set.

[0143] The anomaly identification module extracts frequency domain features and transient features from the wideband calibration residual current set, and performs sparse inversion localization using gated fusion to obtain the feature vectors of suspected abnormal feeders and targets.

[0144] The hierarchical discrimination module performs hierarchical cause discrimination based on the target feature vector and the suspected abnormal feed line, quantifies the deviation degree of the target feature vector, and generates a treatment judgment set;

[0145] The report generation module performs risk classification mapping based on the handling judgment set to obtain the residual current hazard level, and arranges the execution sequence of control strategies to generate a power supply residual current report.

[0146] In summary, this invention achieves consistent aperture correction and spectral domain splicing of low-frequency channel data and high-frequency channel data by generating a self-test response record and forming acquisition deviation correction information through micro-amplitude benchmark self-test, thereby stabilizing the wideband measurement benchmark. Then, based on the candidate feature set and the self-test response record, it identifies low-health channels and identifies high-importance loops by combining the monitoring topology table. After completing the gated fusion feature set, it performs sparse inversion positioning to obtain suspected abnormal feeders and target feature vectors, thereby enhancing the directional accuracy of abnormal positioning and the traceability of the basis for handling.

[0147] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for monitoring the residual current of a station AC power supply, characterized in that: include, Collect residual current data and feeder current monitoring data, and solidify them through a monitoring topology table to generate a time-scale aligned monitoring set; Based on the time-scale aligned monitoring set, a micro-amplitude reference self-test and acquisition deviation correction are performed, and the spectral domains of the low-frequency channel and the high-frequency channel are spliced ​​to generate a wideband calibration residual current set; Frequency domain features and transient features are extracted from the broadband calibration residual current set, and sparse inversion localization is performed by gated fusion to obtain the feature vectors of suspected abnormal feeders and targets. Based on the target feature vector and suspected abnormal feed lines, hierarchical cause discrimination is performed, and the deviation degree of the target feature vector is quantified to generate a treatment judgment set; The residual current hazard level is obtained by mapping the risk classification based on the disposal judgment set, and the execution sequence of the control strategy is arranged to generate a power supply residual current report.

2. The method for monitoring residual current of station AC power supply as described in claim 1, characterized in that: The residual current data includes the zero-sequence residual current measurement value, sampling channel identifier, and monitoring point number; The feeder current monitoring data includes the effective value of the feeder loop current, loop identification, and monitoring point number.

3. The method for monitoring residual current of station AC power supply as described in claim 2, characterized in that: The process of solidifying the monitoring topology table and generating a time-scale aligned monitoring set involves the following steps: Write the residual current data and feeder current monitoring data into a unified time stamp and monitoring point number to obtain a time-stamped sampling frame. The sampled frames with timestamps are sorted and aligned according to their timestamps, and the sampled contents at the same time are merged to generate an aligned sampled sequence. The aligned sampling sequence is used as input and matched with the monitoring topology table to determine the loop identifier and the relationship with the upstream power source corresponding to each sampling record, thereby generating a topology solidification record; The topology-fixed records are encapsulated according to a unified time window and written with complete index information to generate a time-stamped aligned monitoring set.

4. The method for monitoring residual current of station AC power supply as described in claim 1, characterized in that: The steps for performing micro-amplitude benchmark self-check and acquisition deviation correction are as follows: Trigger micro-amplitude benchmark self-test based on time-scale aligned monitoring set, extract benchmark response within corresponding time window and generate self-test response record; The acquisition deviation is identified based on the self-test response record, and the gain compensation and zero-point compensation parameters are obtained according to the difference between the expected amplitude and phase of the reference sequence and the measured amplitude and phase, thereby generating acquisition deviation correction information. The low-frequency channel data and high-frequency channel data in the time-scale alignment monitoring set are corrected for consistency based on the acquisition deviation correction information to obtain the calibration time-scale alignment monitoring set.

5. The method for monitoring residual current of station AC power supply as described in claim 4, characterized in that: The steps for splicing the spectral domains of the low-frequency and high-frequency channels to generate a broadband calibration residual current set are as follows. Denoising and outlier suppression are performed on the calibration time-scale alignment monitoring set while keeping the time window index unchanged to obtain the spectral domain preprocessed record; Based on the spectral domain preprocessing record, the low-frequency channel and the high-frequency channel are spliced ​​in the overlapping frequency band, and the broadband calibration residual current set is obtained through amplitude and phase consistency verification.

6. The method for monitoring residual current of station AC power supply as described in claim 1, characterized in that: The steps for extracting frequency domain and transient features from the broadband calibration residual current set, and then using gated fusion to perform sparse inversion localization to obtain the feature vectors of suspected abnormal feeders and targets are as follows. Frequency domain analysis and transient analysis are performed on the wideband calibration residual current set to extract frequency domain features and transient features respectively, and then the dimensions are aligned and combined to generate a candidate feature set; Based on the self-test response record, the detection peak value of the baseline sequence is calculated for low-frequency channel data and high-frequency channel data. When the detection peak value is lower than the baseline detection lower limit threshold, it is determined to be a low health channel. Based on the monitoring topology table, the relationship between the circuit power supply object and the upstream power supply is extracted, and the failure impact range is obtained. When the failure impact range covers the security load, it is determined to be a high-importance circuit. Gated fusion is performed on the candidate feature set to suppress candidate features corresponding to low health channels and enhance candidate features corresponding to high importance loops, thereby generating a gated fusion feature set; Sparse inversion localization is performed on the gated fusion feature set to obtain and sort the feeder contribution of each feeder loop, and to locate suspected abnormal feeders. The gated fusion feature set and the suspected abnormal feed line are encapsulated to form the target feature vector and maintain the same time window index as the suspected abnormal feed line to obtain the target feature vector.

7. The method for monitoring residual current of station AC power supply as described in claim 1, characterized in that: The steps for determining the cause of hierarchical anomalies based on the target feature vector and suspected abnormal feedlines are as follows: Based on the suspected abnormal feeders, determine the handling feeder sequence and bind the handling feeder sequence to the target feature vector to generate the handling target feature vector; Nonlinear load interference is determined by analyzing the feature vector of the treatment target to obtain a nonlinear load interference label; when the nonlinear load interference label is negative, insulation consistency is determined by comparing the feature vector of the treatment target with temperature monitoring data to obtain an insulation degradation label. When the insulation degradation label is not specified, the target feature vector is input into the fault classification model to distinguish between grounding faults and equipment leakage, and the fault cause label is obtained.

8. The method for monitoring residual current of station AC power supply as described in claim 7, characterized in that: The generation of the disposal judgment set refers to quantifying the degree of deviation of the disposal target feature vector according to the deviation quantization caliber to obtain the degree of abnormality, and combining the fault cause label, the degree of abnormality and the disposal feeder sequence to generate the disposal judgment set.

9. The method for monitoring residual current of station AC power supply as described in claim 8, characterized in that: The timing of the orchestration control strategy execution and the generation of a power supply residual current report are as follows: Select the classification criteria corresponding to the fault cause label according to the risk classification rules, and map the degree of abnormality to the risk classification to generate the residual current hazard level. Based on the monitoring topology table, the sequence of feeder lines to be handled is mapped to a list of handling circuits, and the priority order of handling is determined in combination with the residual current hazard level; According to the strategy arrangement rules, the residual current hazard level and the list of disposal circuits are matched with the residual current disposal strategy and the execution order is determined to generate the control strategy execution sequence. The handling judgment set, residual current hazard level and control strategy execution sequence are summarized into structured content according to the loop relationship of the monitoring topology table to generate a power supply residual current report.

10. A residual current monitoring system for station AC power supplies, based on the residual current monitoring method for station AC power supplies according to any one of claims 1 to 9, characterized in that: include, The data acquisition module collects residual current data and feeder current monitoring data, and solidifies them through a monitoring topology table to generate a time-stamped aligned monitoring set. The self-test correction module performs micro-amplitude reference self-test and acquisition deviation correction based on the time-scale aligned monitoring set, and splices the spectral domains of the low-frequency channel and the high-frequency channel to generate a wideband calibration residual current set. The anomaly identification module extracts frequency domain features and transient features from the wideband calibration residual current set, and performs sparse inversion localization using gated fusion to obtain the feature vectors of suspected abnormal feeders and targets. The hierarchical discrimination module performs hierarchical cause discrimination based on the target feature vector and the suspected abnormal feed line, quantifies the deviation degree of the target feature vector, and generates a treatment judgment set; The report generation module performs risk classification mapping based on the handling judgment set to obtain the residual current hazard level, and arranges the execution sequence of control strategies to generate a power supply residual current report.

Images