A power distribution network weak fault fast starting method and system

Abstract

The present application belongs to the technical field of power distribution network relay protection and fault detection, and particularly relates to a weak fault fast starting method and system for a power distribution network. The method comprises: obtaining zero sequence voltage and / or zero sequence current continuous sampling data, performing Teager-Kaiser energy calculation, absolute value processing and median filtering on the effective sampling data to form short window TK energy and long window TK cumulative energy; using normal working condition samples to establish an adaptive baseline and a threshold value, freezing sample access and threshold value updating when the short window is pre-triggered, and using long window energy to confirm a continuous fault under the frozen threshold; and reconstructing the baseline with continuous static stable data after the fault is restored. The present application can reduce the pollution of suspected faults on the adaptive baseline, and take into account early detection and continuous confirmation of weak faults.

Description

Technical Field

[0001] This invention belongs to the field of power distribution network relay protection and fault detection technology, specifically relating to a method and system for rapid activation of weak faults in power distribution networks. Background Technology

[0002] When a distribution network experiences grounding faults such as high-resistance grounding, arcing grounding, intermittent grounding, or secondary grounding, engineering practices typically utilize zero-sequence electrical quantities such as bus zero-sequence voltage and feeder zero-sequence current to set fault initiation criteria. After initiation, fault recording, fault line selection, fault location, or subsequent protection judgments are performed. Existing initiation methods include setting fixed over-limit thresholds based on zero-sequence electrical quantities, and forming adaptive thresholds based on normal or historical operating data. To enhance weak transient characteristics, energy transformation, filtering, and time window accumulation processing can also be performed on the zero-sequence electrical quantities before threshold comparison.

[0003] Existing technologies still face the challenge of balancing sensitivity to weak grounding faults with reliability against disturbances. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for rapid initiation of weak faults in power distribution networks, so as to solve the problem that it is difficult to balance the initiation sensitivity and anti-disturbance reliability of weak grounding faults in the existing technology.

[0005] To achieve the above objectives, the following solution is adopted: In a first aspect, the present invention provides a method for rapid initiation of minor faults in a distribution network, comprising: Acquire continuous sampling data of at least one zero-sequence signal channel in the distribution network, wherein the zero-sequence signal channel is a zero-sequence voltage channel and / or a zero-sequence current channel; The valid sampled data in the continuous sampling data are sequentially subjected to Teager-Kaiser energy calculation, absolute value processing and median filtering to obtain the TK energy filtered sequence. Based on the TK energy filtering sequence, short-window TK energy and long-window TK cumulative energy are formed respectively. Under normal conditions, a normal operating condition baseline is established based on the normal operating condition samples that have been verified by data quality. Based on the normal operating condition baseline, the short-window pre-trigger threshold and long-window confirmation threshold are determined. The normal operating condition baseline and threshold are updated only using subsequent sampling data where the short-window TK energy has not reached the short-window pre-trigger threshold. When the short window TK energy of any zero-sequence signal channel reaches the short window pre-trigger threshold, the operating status will be switched to the suspected status, the normal operating condition sample admission will be stopped, and the normal operating condition baseline and threshold will be frozen. In the suspected state, the long window TK accumulated energy of the zero-sequence signal channel is judged by the frozen long window confirmation threshold. When the long window TK accumulated energy reaches the long window confirmation threshold and does not fall back to the background range determined by the normal operating condition baseline within the set duration, a fault start signal is output and kept frozen. After the fault is recovered, the baseline and thresholds of normal operating conditions are re-established using continuous and effective static and stable operating data, and the normal state is restored.

[0006] The above scheme forms short-window TK energy for pre-triggering and long-window TK cumulative energy for continuous confirmation on the same TK energy filtering sequence. The short-window pre-triggering result directly controls the normal operating condition sample admission and the freezing of the baseline and threshold. Under the frozen detection benchmark, the long-window continuous confirmation is continued. Then, the detection benchmark is reconstructed in the static and stable phase after the fault returns. This solves the problems that slow-developing weak faults may contaminate the normal baseline and that a single observation scale is difficult to take into account both early detection and continuous confirmation. Thus, it realizes the rapid and stable start-up of weak grounding faults and the closed-loop recovery of the adaptive detection benchmark.

[0007] Furthermore, the valid sampled data are sequentially subjected to Teager-Kaiser energy calculation, absolute value processing, and median filtering to form short-window TK energy and long-window TK cumulative energy, including: For zero-sequence signal channel The Sample values The Teager-Kaiser energy is determined according to the following formula:

[0008] in, Indicates the channel type. hour This is the zero-sequence voltage sample value. hour The zero-sequence current sample value; for The absolute value is taken to obtain the non-negative energy sequence, and median filtering is applied to the non-negative energy sequence to obtain the TK energy filtered sequence. Based on the TK energy filter sequence, respectively according to the short window length and long window length By performing sliding accumulation, the short-window TK energy and the long-window TK accumulated energy are obtained, where .

[0009] The above-mentioned further scheme first uses the Teager-Kaiser energy operator to extract the instantaneous energy change of the zero-sequence signal, then uses absolute value processing to avoid the cancellation of positive and negative energy, and uses median filtering to suppress isolated pulses. Finally, it establishes cumulative quantities at different time scales on the same filtering result, which solves the problems that instantaneous energy is easily affected by isolated noise and that a single window cannot simultaneously undertake rapid detection and continuous confirmation, thereby improving the feature stability and time scale adaptability used for subsequent threshold judgment.

[0010] Furthermore, based on the normal operating condition baseline, the short-window pre-trigger threshold and long-window confirmation threshold are determined, including: for any zero-sequence signal channel Establish short-window background sample sets composed of effective short-window TK energy under normal conditions and long-window background sample sets composed of effective long-window TK accumulated energy, respectively; Determine the median of the short-window background sample set and the long-window background sample set respectively. and median absolute deviation ,in Indicates a short window. Indicates a long window; according to

[0011] Determine the short window pre-trigger threshold, and follow...

[0012] Determine the long window confirmation threshold; where, and For the reliability coefficient, and , and These are the lower limits of the short window threshold and the lower limit of the long window threshold, respectively, corresponding to the noise floor of the device.

[0013] The above-mentioned further scheme determines the corresponding threshold by using the center level and dispersion of normal samples of short window and long window respectively, and sets a lower threshold corresponding to the noise floor of the device, so that the threshold at different time scales can change with the fluctuation of normal operating conditions, while avoiding the threshold being too low when the background energy is close to zero, thereby improving the stability of the adaptive threshold under different normal operating conditions.

[0014] Furthermore, when the short-window TK energy of any zero-sequence signal channel reaches the short-window pre-trigger threshold, the operating state is switched to the suspected state, including: when the short-window TK energy reaches the short-window pre-trigger threshold and continues to reach the preset minimum number of sampling points or the preset minimum time, the pre-trigger time of the zero-sequence signal channel is recorded, and the operating state is switched from the normal state to the suspected state; when switching to the suspected state, the normal operating condition background sample, the normal operating condition baseline, the short-window pre-trigger threshold, and the long-window confirmation threshold are frozen, and the original zero-sequence signal, TK energy, short-window TK energy, and long-window TK cumulative energy during the suspected event are recorded; when the short-window TK energy drops to the background range determined by the normal operating condition baseline and the long-window TK cumulative energy does not reach the long-window confirmation threshold, this suspected event is judged as a disturbance, and the data during the suspected event is prohibited from being added to the normal operating condition sample, and the normal state is restored after the static observation time ends.

[0015] The above-mentioned further scheme freezes the normal background and threshold during the pre-trigger stage before formal fault confirmation and isolates suspected event data from normal samples, so that slowly developing fault energy cannot continue to be written into the normal baseline before confirmation; at the same time, it sets a disturbance rejection and static observation process for short-term events that have not formed a long window of confirmation, thereby reducing the risk of the baseline following the fault changes and short-term disturbances contaminating normal samples.

[0016] Furthermore, when both the zero-sequence voltage channel and the zero-sequence current channel are valid, time-synchronized, and the device association relationship conforms to the preset electrical association relationship, a fault start signal is output, including: after the long window TK energy accumulation of a zero-sequence signal channel forms continuous fault evidence, the zero-sequence signal channel is determined as the main trigger channel, and it is determined whether the other zero-sequence signal channel has a short window TK energy change exceeding the background upper bound of its normal operating condition baseline within the allowable time difference relative to the pre-trigger time of the main trigger channel; when the other zero-sequence signal channel has a short window TK energy change, it is determined that the two zero-sequence signal channels form fault evidence with consistent time and a fault start signal is output; when the main trigger channel forms continuous fault evidence, and the other zero-sequence signal channel does not have a short window TK energy change exceeding the background upper bound of its normal operating condition baseline within the extended observation time, the operating state is switched to the pending verification state, and no fault start signal is output.

[0017] The above-mentioned further scheme uses the channel that has formed continuous fault evidence as the main trigger channel and requires another effective channel to provide supportive changes beyond its normal background within the allowable time difference, rather than requiring the two channels to reach the same strength of the confirmation condition at the same time. This solves the problem that simple "AND" logic is prone to missed detection and simple "OR" logic is prone to false triggering when the strength of the two channels is inconsistent, thereby improving the reliability of zero-sequence voltage and zero-sequence current evidence conflict identification.

[0018] Furthermore, data quality verification is performed on the continuously sampled data, including: determining the validity of the sampled data based on sampling continuity, sampling missingness, sampling saturation, timestamp continuity, and channel online status; when any of the three adjacent sampled values ​​involved in the Teager-Kaiser energy calculation is invalid, the corresponding Teager-Kaiser energy is marked as invalid, and the Teager-Kaiser energy is not calculated using the interpolated sampled value; when the valid TK energy sampling in a short or long window does not meet the window validity requirements, the corresponding short window TK energy or long window TK accumulated energy is marked as invalid, and the use of the corresponding window data to update the normal operating baseline or form a fault confirmation basis is prohibited; when the time difference between the zero-sequence voltage channel and the zero-sequence current channel exceeds the allowable time difference, the time consistency verification between the two zero-sequence signal channels is stopped, and fault judgment is performed separately according to the single-channel method.

[0019] The above-mentioned further scheme solves the problem that missing samples, saturation, or time misalignment may be incorrectly interpreted as energy changes by setting data validity constraints at different levels of TKEO calculation, sliding window accumulation, and dual-channel verification, and by excluding invalid data from baseline updates and fault confirmation. This improves the validity and traceability of fault judgment results under abnormal sampling conditions.

[0020] Furthermore, when only one zero-sequence signal channel is valid, or when the time consistency verification between two zero-sequence signal channels cannot be performed, a downgrade verification is performed on the valid zero-sequence signal channel. This includes: using a frozen long-window confirmation threshold to determine the long-window TK accumulated energy of the valid zero-sequence signal channel and increasing the long-window persistence requirement; and / or, determining the abrupt change between the current sampled value of the valid zero-sequence signal channel and the in-phase sampled value one power frequency cycle ago. When the abrupt change exceeds the abrupt change threshold determined by normal operating conditions at multiple consecutive sampling points, and the occurrence time of the abrupt change is adjacent to the occurrence time of the abnormal long-window TK accumulated energy, supplementary fault evidence is determined. When the downgrade verification forms supplementary fault evidence, a fault initiation signal is output. When the downgrade verification does not form supplementary fault evidence, the system remains in a pending verification state and fault recording is triggered.

[0021] The above-mentioned further scheme resolves the contradiction between detection capability and false start constraints when dual-channel evidence is unavailable by using stricter single-channel persistence conditions and / or in-phase mutation amounts as supplementary evidence, and by retaining the state to be verified without forcibly confirming the fault when the evidence is still insufficient. This improves the adaptability of the fault start path to data loss and channel degradation.

[0022] Furthermore, after fault recovery, the normal operating condition baseline and thresholds are re-established using continuous and effective static stable operating data. This includes: when the original zero-sequence electrical quantity of the zero-sequence signal channel that was started during the fault recovers to the allowable operating range, and the accumulated energy of the long window TK drops below the recovery threshold and continues to reach the recovery delay time, the operating state is switched to the recovery state; in the recovery state, the use of data from the fault period and the suspected event period to update the normal operating condition baseline is stopped, and continuous static stable operating data is collected; when the continuous static stable operating data continuously reaches the preset static stable period and the data quality is valid, the normal operating condition baseline, short window pre-trigger threshold, and long window confirmation threshold are re-established using the continuous static stable operating data, and the normal state is restored; when the energy of the short window TK reaches the short window pre-trigger threshold again during the process of re-establishing the normal operating condition baseline, the re-establishment of the normal operating condition baseline is stopped and the system switches to the suspected state.

[0023] The above-mentioned further scheme uses the recovery of the original zero-sequence electrical quantity, the recovery of the accumulated energy of the long window TK, the recovery delay and the continuous static stable data as the preconditions for the recovery baseline learning. When an anomaly occurs again during the reconstruction, the reconstruction is stopped, so that fault decay, arc reignition or recovery transient cannot be immediately written into the new normal baseline, thereby improving the stability and closed-loop controllability of adaptive detection baseline reconstruction after fault recovery.

[0024] In a second aspect, the present invention provides a rapid start-up system for minor faults in a power distribution network, comprising: The zero-sequence signal acquisition module is used to acquire continuous sampling data of at least one zero-sequence signal channel in the distribution network and output the sampling value, sampling time and data quality information. The zero-sequence signal channel is a zero-sequence voltage channel and / or a zero-sequence current channel. The data quality and time alignment module is used to receive continuous sampled data and data quality information, determine valid sampled data, and output the sampled sequence that has been quality verified. The TK energy processing module is used to receive the quality-verified sampling sequence, perform Teager-Kaiser energy calculation, absolute value processing and median filtering on the valid sampled data in sequence, and output the TK energy filtered sequence. The dual time-scale accumulation module is used to receive the TK energy filter sequence and form short-window TK energy and long-window TK accumulated energy respectively; The normal operating condition baseline module is used to establish a normal operating condition baseline based on normal operating condition samples that have been verified by data quality under normal conditions. It determines the short window pre-trigger threshold and the long window confirmation threshold based on the normal operating condition baseline, and updates the normal operating condition baseline and thresholds only using subsequent sampling data where the short window TK energy has not reached the short window pre-trigger threshold. The status control module is used to receive the short window TK energy, the long window TK accumulated energy, the short window pre-trigger threshold, and the long window confirmation threshold. When the short window TK energy reaches the short window pre-trigger threshold, it switches to the suspected state and controls the normal operating condition baseline module to stop sample admission and threshold update. In the suspected state, it uses the frozen long window confirmation threshold to determine the long window TK accumulated energy and outputs a fault start signal when continuous fault evidence is formed. The recovery and reconstruction module is used to obtain continuous and effective static and stable operating data after fault recovery, and to control the normal operating condition baseline module to re-establish the normal operating condition baseline, as well as the short window pre-trigger threshold and long window confirmation threshold, based on the static and stable operating data.

[0025] The second aspect mentioned above is based on the same inventive concept as the first aspect, and achieves the same or corresponding technical effects of rapid start-up of weak faults and closed-loop recovery of detection benchmark as the first aspect.

[0026] Furthermore, the system also includes a channel consistency verification module and a degradation detection module. The channel consistency verification module is used to receive the pre-trigger time, short-window TK energy, and long-window TK accumulated energy of the two zero-sequence signal channels when both the zero-sequence voltage channel and the zero-sequence current channel are valid, time-synchronized, and the equipment association relationship conforms to the preset electrical association relationship. After a continuous fault evidence is formed in one zero-sequence signal channel, it determines whether the other zero-sequence signal channel has a short-window TK energy change exceeding the background upper limit of its normal operating condition baseline within the allowable time difference, and outputs a consistent fault evidence result or a channel conflict result. The degradation detection module is used to perform long-window continuous verification and / or one power frequency cycle mutation amount verification on the valid zero-sequence signal channel when only one zero-sequence signal channel is valid, the two zero-sequence signal channels are time-synchronized, or the channel consistency verification module outputs a channel conflict result, and outputs supplementary fault evidence or insufficient evidence result to the state control module. The state control module is also used to output a fault start signal according to the consistent fault evidence result or supplementary fault evidence, and switch to the pending verification state according to the channel conflict result or insufficient evidence result.

[0027] The above-mentioned further solution sets up a channel consistency verification module and a degradation detection module at the system level, so that dual-channel consistency, dual-channel conflict, single-channel validity and time asynchrony are respectively entered into clear processing branches, and the verification results are fed back to the state control module, thereby improving the system's path adaptability and state control integrity in the case of channel degradation and evidence conflict. Attached Figure Description

[0028] The accompanying drawings, which form part of this application, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings: Figure 1This is a schematic diagram of the system composition and data flow in one embodiment of the present invention; Figure 2 This is a schematic diagram of the main fault start-up process in one embodiment of the present invention; Figure 3 This is a schematic diagram of the fault state machine and the normal operating condition baseline closed-loop control relationship in one embodiment of the present invention. Figure 4 Zero-sequence voltage waveforms before and after a fault in a high-resistance grounding example of an arc suppression coil grounding system; Figure 5 The zero-sequence current waveforms before and after a fault are shown in the embodiment of a high-resistance grounding system for an arc suppression coil grounding system. Figure 6 The diagram shows the normal operating zero-sequence voltage and TK cumulative energy of a high-resistance grounding embodiment of an arc suppression coil grounding system; Figure 7 An adaptive threshold diagram for a high-resistance grounding embodiment of an arc suppression coil grounding system; Figure 8 Normalized waveform and starting point diagram for a high-resistance grounding embodiment of an arc suppression coil grounding system; Figure 9 The TK cumulative energy over-limit start-up diagram is shown in the embodiment of a high-resistance grounding system for an arc suppression coil grounding system. Figure 10 Waveforms of zero-sequence voltage and zero-sequence current before and after a fault in an example of an FTU field arc grounding; Figure 11 A diagram showing the accumulated energy and threshold of zero-sequence voltage under normal background TK in an example of FTU field arc grounding; Figure 12 A diagram illustrating the zero-sequence voltage normalization startup process for an FTU field arc grounding embodiment; Figure 13 Diagram showing the starting point of the zero-sequence voltage TK cumulative energy in an example of FTU field arc grounding; Figure 14 A diagram showing the accumulated energy and threshold of zero-sequence current under normal background TK in an example of FTU field arc grounding; Figure 15 Diagram showing the starting point of zero-sequence current TK cumulative energy for an FTU field arc grounding embodiment. Detailed Implementation

[0029] The present invention will now be described in detail with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other.

[0030] The following detailed description is exemplary and intended to provide further detailed explanation of the invention. Unless otherwise specified, all technical terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. The terminology used in this invention is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention.

[0031] like Figure 2 As shown in the figure, this embodiment provides a method for rapid start-up of weak faults in a distribution network, including the following steps: S1. Acquire continuous sampling data from at least one zero-sequence signal channel in the distribution network. The zero-sequence signal channel can be a zero-sequence voltage channel and / or a zero-sequence current channel. Zero-sequence voltage can be derived from open delta voltage, zero-sequence voltage transformers, or internal calculations of protection devices. Zero-sequence current can be derived from zero-sequence current transformers, three-phase current composite values, or the zero-sequence sampling channel of FTUs or DTUs. Each sampling data item may include the sampled value, sampling time or sampling sequence number, channel type, equipment identifier, feeder or busbar association identifier, and data quality flag. The data quality flag is used to characterize whether sampling is continuous, whether there are missing samples, whether the range is exceeded or saturated, whether communication is abnormal, and whether time is synchronized. The acquired continuous sampling data is used for subsequent data quality verification and TK energy calculation.

[0032] Zero-sequence voltage and zero-sequence current sampled using synchronous analog-to-digital conversion within the same device can be aligned according to the same sampling sequence number; data from different sampling units can be aligned according to the sampling time. In dual-channel mode, the zero-sequence voltage channel and the zero-sequence current channel should also satisfy the equipment association relationship, that is, they belong to the same bus or belong to the electrical association range formed by the bus and the corresponding feeder. Data that does not satisfy the equipment association relationship does not constitute dual-channel fault evidence. When only one zero-sequence signal channel is available, that channel is retained and enters the single-channel processing path.

[0033] S2. For valid sampled data in the continuous sampling data, Teager-Kaiser energy calculation, absolute value processing, and median filtering are performed sequentially to obtain the TK energy filtered sequence. First, the validity of the sampled data is determined based on sampling continuity, missing samples, sampling saturation, timestamp continuity, and channel online status. For continuous sampled data that passes quality verification, Teager-Kaiser energy is calculated using adjacent sample values; then, the absolute value of the Teager-Kaiser energy is taken to avoid cancellation of positive and negative energies in subsequent accumulation processes; finally, median filtering is performed on the non-negative energy sequence to suppress isolated impulse noise. The resulting TK energy filtered sequence serves as the common input for both short-window TK energy and long-window TK accumulated energy, ensuring that the detection quantities at both time scales are based on the same instantaneous energy characteristics.

[0034] S3. Based on the TK energy filtering sequence, short-window TK energy and long-window TK cumulative energy are generated respectively, and a normal operating condition baseline and corresponding thresholds are established. Short-window TK energy is used to detect anomalies before the fault energy has accumulated significantly over a long period, while long-window TK cumulative energy is used to determine the persistence of the anomaly. Under normal conditions, only normal operating condition samples that have passed data quality verification and whose short-window TK energy has not reached the short-window pre-trigger threshold are added to the background sample set. A normal operating condition baseline is established based on the background samples, and then the short-window pre-trigger threshold and long-window confirmation threshold are determined based on the normal operating condition baseline. Subsequent sampling data is only allowed to continue updating the normal operating condition baseline and thresholds if the short-window TK energy has not reached the short-window pre-trigger threshold, thus creating a data admission constraint between baseline learning and fault detection status.

[0035] S4. When the short-window TK energy of any zero-sequence signal channel reaches the short-window pre-trigger threshold, the operating state is switched to the suspected state, and the normal operating condition baseline and threshold are frozen. When the short-window TK energy reaches the pre-trigger condition, the short-window result is not directly used as the final fault initiation conclusion. Instead, the pre-trigger time is recorded, the operating state is switched from the normal state to the suspected state, the normal operating condition sample admission is stopped, and the normal operating condition baseline, short-window pre-trigger threshold, and long-window confirmation threshold formed before the start of this suspected event are frozen. The original zero-sequence signal, TK energy, short-window TK energy, and long-window TK accumulated energy during the suspected event can continue to be recorded, but are not backfilled into the normal background sample set.

[0036] S5. In the suspected state, use the frozen long-window confirmation threshold to determine the long-window TK accumulated energy, and output a fault initiation signal or terminate the suspected event based on the judgment result. After entering the suspected state, the long-window TK accumulated energy continues to be updated, but the comparison threshold remains at the long-window confirmation threshold frozen when entering the suspected state. When the long-window TK accumulated energy reaches the long-window confirmation threshold and does not fall back to the background range determined by the normal operating condition baseline within the set duration, it indicates that the anomaly is persistent, outputs a fault initiation signal, and maintains the normal operating condition baseline and threshold frozen state. If the short-window TK energy falls back to the background range and the long-window TK accumulated energy does not reach the long-window confirmation threshold, the suspected event is judged as a disturbance, and the data during the suspected event is not added to the normal operating condition sample. After static observation, the normal state is restored.

[0037] S6. After fault recovery, the normal operating baseline and thresholds are re-established using continuous and effective statically stable operating data, and the normal state is restored. Thresholds are frozen after fault initiation. When the original zero-sequence electrical quantity of the zero-sequence signal channel that triggered the fault recovers to the allowable operating range, and the accumulated energy of the long-window TK drops below the recovery threshold and continues to reach the recovery delay time, the system enters the recovery state. In the recovery state, data from the fault period and suspected event period is still prohibited from entering the normal background, and continuous statically stable operating data is re-acquired. When the statically stable operating data continuously reaches the set statically stable period and the data quality is valid, the normal operating baseline, short-window pre-trigger threshold, and long-window confirmation threshold are re-established using this batch of data, and the normal state is then restored. If the short-window TK energy reaches the short-window pre-trigger threshold again during the reconstruction process, the baseline reconstruction is immediately stopped and the system re-enters the suspected state.

[0038] The above S1 to S6 constitute a closed loop consisting of normal operating condition baseline learning, short-timescale anomaly detection, early baseline freezing, long-timescale continuous confirmation, and baseline reconstruction after fault recovery. Short-window pre-triggering occurs before long-window formal confirmation, thus utilizing a faster timescale to prevent suspected fault data from entering the normal operating condition baseline. Long-window confirmation continues to use the comparison benchmark frozen before the fault to determine whether energy continues to accumulate, preventing the detection benchmark from rising synchronously with fault development. During the recovery phase, normal sample access is reopened through statically stable operating data, allowing normal operating condition adaptation to resume after the event ends. This maintains early sensitivity to weak faults while reducing the impact of isolated disturbances, slow baseline contamination, and recovery transients on the initiation judgment.

[0039] In one embodiment, for any zero-sequence signal channel The valid sampled sequence is used to perform Teager-Kaiser energy calculations. Indicates the zero-sequence signal channel type; when hour, Indicates the first One zero-sequence voltage sample value; when hour, Indicates the first Each zero-sequence current sample value. The Teager-Kaiser energy is determined by the following formula:

[0040] In the formula, Indicates channel In the The Teager-Kaiser energy corresponding to each sampling location; , and These are three consecutive valid sample points from the same zero-sequence signal channel. Since this calculation requires three adjacent sample values, if any of the sample values ​​used in the calculation is invalid, the corresponding... It is not used for subsequent cumulative and normal baseline statistics.

[0041] right Performing absolute value processing, we get:

[0042] In the formula, This represents the non-negative TK energy after absolute value processing. Absolute value processing is used to convert the bipolar output of the Teager-Kaiser energy into a unipolar energy quantity, avoiding positive and negative cancellation during accumulation in the sliding window.

[0043] In one specific implementation, for Using a median filter window of length 5, we obtain:

[0044] In the formula, Indicates the first TK energy filter value at each sampling location, This indicates that the median of the given samples is used. At sequence boundaries, the median can be taken using 3 or 4 currently available samples. The median filter window can also use 3 or 7 points, with 5 points being a preferred implementation. The output of the median filter... It serves as a cumulative object for both short-term and long-term time windows.

[0045] Furthermore, in the same Set short window length and long window length ,in We obtained the following respectively:

[0046]

[0047] In the formula, Indicates channel In the The energy of the short window TK at each sampling time. This represents the accumulated energy of the long window TK in the same channel; Indicates the sampling index within the sliding window; and These represent the number of sampling points contained in the short window and the long window, respectively. The window length can be configured according to the number of sampling points corresponding to the power frequency cycle to adapt to different sampling frequencies. In one embodiment, the long window uses 4 power frequency cycles, and the number of sampling points in each power frequency cycle is denoted as . ,but The short window is smaller than the long window, and can be 0.5 to 1 power frequency cycle. Through this process, the energy of the short window TK is used for pre-triggering, and the accumulated energy of the long window TK is used for continuous confirmation in the suspected state. The two are not weighted and summed.

[0048] The above implementation avoids inconsistencies in the baseline caused by using different front-end features at different time scales by sharing the same TK energy filtering sequence between short and long windows. The short window prioritizes reflecting the initial changes in energy, while the long window continuously accumulates data to differentiate short-term isolated disturbances over time, thus balancing real-time performance with stability assessment.

[0049] In one embodiment, the normal operating condition baseline corresponds to the short-window background sample set and the long-window background sample set, respectively. For channels... and window type ,in Indicates short window, Let the long window be denoted as and the background sample set be denoted as . First, determine the median based on this background sample set:

[0050] Then, the absolute bias of the median is determined based on the degree of deviation of the background samples from the median:

[0051] In the formula, Indicates channel In window type Background center value of TK energy under normal operating conditions; It represents the median absolute deviation of the corresponding background sample, used to characterize the dispersion of energy under normal operating conditions; This represents any valid TK energy sample in the background sample set. and All data were obtained from statistics based on data that had been verified for quality under normal conditions and had not experienced any short-window anomalies.

[0052] The short-window pre-trigger threshold can be determined according to the following formula:

[0053] The long window confirmation threshold can be determined using the following formula:

[0054] In the formula, Indicates channel The short window pre-trigger threshold, Indicates channel The long window confirmation threshold; and They are respectively from the short-window background sample set and the long-window background sample set; and These represent the median absolute deviations of the corresponding background samples; and The reliability coefficient is used to adjust the interval between normal fluctuations and the trigger threshold. This ensures that the statistical threshold for formal confirmation is no lower than the reliability requirements adopted for pre-triggering; and These represent the lower limits of the short-window and long-window thresholds, respectively, which match the device quantization noise, the transformer background noise, and the corresponding TK energy dimensions. The lower thresholds are used to prevent the thresholds from being too low when the background dispersion is very small. Under normal conditions, these thresholds can be updated with background samples; however, they remain frozen after entering a suspected state and are used for long-window confirmation.

[0055] As a basic threshold formation method, it can also be used to accumulate energy sample sets of long-window TK during normal operation. Calculate the median :

[0056] Based on median and reliability coefficient Determine the adaptive threshold:

[0057] And set an absolute lower limit:

[0058] In the formula, This represents the median energy accumulated during normal operation of the long window TK; The reliability coefficient can be 5 in one implementation, 3 to 4 when higher sensitivity is required, and 6 to 8 when higher resistance to false triggering is required. This represents the adaptive threshold obtained from statistics of the normal background. This indicates the absolute lower limit set based on the device's quantization noise and signal amplitude. This represents the threshold ultimately used for comparison. In one implementation, It can be configured according to the dimensions and normalization method of TK energy. to The corresponding magnitude. Regardless of whether the median multiple method or the median absolute deviation method is used, background samples are only allowed to enter under normal conditions, and updates are stopped after entering the suspected state.

[0059] The above implementation method ensures that the thresholds for short windows and long windows are derived from normal background samples at the same time scale. It also utilizes the median to reduce the impact of isolated outliers, the median absolute deviation to characterize the degree of normal fluctuation, and the absolute lower limit to constrain the noise level, thereby making the source of the threshold clear and directly comparable to the corresponding sliding window energy.

[0060] like Figure 3 As shown, in one embodiment, when the short-window TK energy of any valid zero-sequence signal channel reaches the short-window pre-trigger threshold and continues to reach a preset minimum number of sampling points or a preset minimum time, the pre-trigger time of that channel is recorded. The running status will be switched from normal to suspected status. Determined by sampling time or sampling sequence number, it is subsequently used in dual-channel mode to compare the order in which anomalies occur in different channels.

[0061] Upon entering a suspected state, a background freeze is executed: [The process of freezing background processes is stopped.] and Write new samples, freeze existing ones. , , , The corresponding short-window pre-trigger threshold and long-window confirmation threshold are also recorded. During suspected events, the original zero-sequence signal, Teager-Kaiser energy, short-window TK energy, and long-window TK cumulative energy are recorded to determine whether the event continues. However, these data are not used as normal samples for threshold learning.

[0062] If, during a suspected event, the energy of the short-window TK drops to the background range determined by the frozen normal operating condition baseline, while the accumulated energy of the long-window TK consistently fails to reach the frozen long-window confirmation threshold, then this event is rejected as a disturbance. Data during the suspected event period continues to be excluded from the normal operating condition samples, and normal status and background updates are resumed after a period of static observation. Through this sequence, the short-window pre-triggering serves a "stop learning" control function, rather than directly serving as the final fault confirmation function.

[0063] The above implementation links the sample admission status with the fault status, so that the comparison benchmark is stopped from being updated when the fault just appears but before the long window confirmation is met, preventing the fault energy from being absorbed as normal background over time; at the same time, perturbation rejection and static observation are performed on suspected events that disappear in a short time, so that the background recovery has a clear time boundary.

[0064] In one embodiment, when both the zero-sequence voltage channel and the zero-sequence current channel are valid, time-synchronized, and the device associations conform to a preset electrical association, a dual-channel time consistency verification is performed after one channel has generated persistent fault evidence. The channel that generates persistent fault evidence is designated as the primary triggering channel, and its pre-trigger time is denoted as... The other channel serves as a support channel. It determines whether the support channel is relative to... Within the allowable time difference, a short window TK energy change occurs that exceeds the upper limit of its normal operating condition baseline background.

[0065] The allowable time difference is used to limit the time correlation between the occurrence of abnormalities in relevant zero-sequence electrical quantities for the same grounding fault. For zero-sequence voltage and zero-sequence current sampled synchronously within the same terminal, the sampling sequence number can be directly compared; for data from different sampling units, alignment is first completed based on the timestamp. Support channels are not required to reach their own long-window confirmation threshold; they only need to generate supportive changes exceeding their normal background within the allowable time difference. This avoids the possibility of a weak amplitude in a certain channel being missed due to the double-strong over-limit condition during high-resistance grounding.

[0066] When the above changes occur in the support channel, it is determined that the fault evidence of the two channels supports each other in time, and a fault start signal is output. When the main trigger channel has formed continuous fault evidence, and the other effective support channel has not shown a short window TK energy change exceeding the background upper limit within the extended observation time, the running state is switched to the pending verification state, the fault start signal is not directly output, and the cause of the conflict is further judged by subsequent degradation detection.

[0067] The above implementation method uses the temporal correlation between zero-sequence voltage and zero-sequence current caused by the same ground fault to provide auxiliary constraints. At the same time, it adopts the asymmetric evidence relationship between the main trigger channel and the support channel, which can distinguish between the two states of "mutual support between the two channels" and "conflict between the effective channels" without using weighted comprehensive scoring.

[0068] In one embodiment, continuous sampling data undergoes data quality verification before being incorporated into the Teager-Kaiser energy calculation. Sampling continuity can be determined based on whether the sampling sequence number or timestamp increases according to a predetermined sampling interval; missing sampling can be determined based on sampling sequence number jumps, buffer gaps, or communication quality flags; sampling saturation can be determined based on whether the sampled value reaches the upper or lower limit of the acquisition range and the device saturation flag; the channel online status is determined by the status information of the acquisition interface or communication interface.

[0069] When used for calculation of , and If any sampled value in the dataset is invalid, the Teager-Kaiser energy is not generated by interpolating that sampled value; instead, the corresponding value is used. Marked as invalid. Subsequent TK energy calculations will only resume after consecutive valid sample values ​​are obtained again. If valid TK energy sampling within a short or long window does not meet the window validity requirement, the corresponding... or Invalid samples are marked as invalid and are not replaced with a value of zero. Invalid windows are neither added to the background sample set nor used as a basis for fault confirmation.

[0070] In one implementation, it can be required that there are no consecutive missing samples within the sliding window and that the effective TK sampling ratio is not less than 95%, with the specific effectiveness ratio configured according to the sampling frequency and hardware communication method. When the time difference between the zero-sequence voltage channel and the zero-sequence current channel exceeds the allowable time difference, this time mismatch is not directly interpreted as a conflict of evidence of faults in the two channels; instead, time consistency verification is stopped and the system switches to single-channel mode. This allows for the differentiation and handling of "data incomparability" and "inconsistent physical evidence."

[0071] The above implementation maintains validity at four levels: sampling point, TK energy, sliding window, and dual-channel evidence. This prevents invalid data from being converted into artificial energy peaks or artificial energy decreases through interpolation, zeroing, or misalignment comparisons, thereby improving the reliability of results and the traceability of event records under abnormal sampling conditions.

[0072] In one embodiment, if only one zero-sequence signal channel is valid, or if the time consistency check between the zero-sequence voltage channel and the zero-sequence current channel cannot be performed, a downgraded check is performed on the currently valid channel. The downgraded check continues to use the long-window confirmation threshold frozen at the start of the suspected state, but the long-window persistence requirement is increased, for example, requiring the long-window confirmation condition to be met consecutively at two adjacent judgment times, to compensate for the lack of supporting evidence from the dual channels.

[0073] Furthermore, a power frequency cycle mutation verification can be performed on the effective channel. This involves comparing the current sampled value with the in-phase sampled value one power frequency cycle ago, and using this difference as the mutation amount. When the mutation amount exceeds the mutation amount threshold determined by normal operating conditions at multiple consecutive sampling points, and the occurrence time of the mutation amount anomaly is adjacent to the occurrence time of the long-window TK accumulated energy anomaly, supplementary fault evidence is formed. This mutation amount criterion uses the periodic correspondence of the original zero-sequence signal to supplement the TK long-window evidence without changing the calculation method of TK energy and normal operating condition baseline.

[0074] When the improved persistence conditions and / or mutation amount verification form supplementary fault evidence, a fault initiation signal is output; when the supplementary evidence is still insufficient, the pending verification state is maintained and fault recording is triggered. The fault recording saves the zero-sequence signal and related TK energy information before and after the suspected event for subsequent fault line selection, location, or manual analysis, without automatically writing the pending verification event into the normal operating condition sample.

[0075] The above implementation method enables the system to retain single-channel detection capability even when dual-channel degradation occurs, while compensating for insufficient evidence by increasing the constraint of persistence or in-phase abrupt change amount; for events that still cannot be confirmed, they only enter the verification and waveform recording path, which can reduce the risk of false start caused by directly relaxing the start conditions under channel failure conditions.

[0076] In one embodiment, after a fault is initiated, all normal operating baseline and threshold updates are stopped and kept frozen. When the original zero-sequence electrical quantity of the zero-sequence signal channel that was initiated by the fault recovers to the allowable operating range, and the accumulated energy of the long window TK drops below the reset threshold and continues to reach the reset delay time, the operating state is switched from the fault confirmation state to the reset state. The recovery of the original zero-sequence electrical quantity is used to determine whether the fault electrical quantity has returned to the normal operating range, the recovery of the accumulated energy of the long window TK is used to determine whether the continuous transient energy has decayed, and the reset delay is used to avoid premature recovery learning caused by instantaneous drop.

[0077] Upon entering the recovery state, the system does not directly reuse the pre-fault background samples for online learning; instead, it re-collects continuous statically stable operating data. Only when the statically stable operating data consistently reaches the preset statically stable period and the sampling quality is valid will the short-window and long-window background sample sets be re-established using this batch of data, and the normal operating baseline, short-window pre-trigger threshold, and long-window confirmation threshold be redefined. If the short-window TK energy reaches the short-window pre-trigger threshold again during baseline reconstruction, reconstruction is immediately stopped, the current event data is retained, and the system switches back to the suspected state.

[0078] The above implementation method enables the lifecycle of the normal operating condition baseline to be jointly controlled by the normal, suspected, fault confirmed and recovery states, avoiding fault decay oscillation, arc reignition or the initial transient state of recovery being immediately learned as the new normal background, thereby maintaining the stability of the adaptive threshold under continuous event conditions.

[0079] like Figure 1 As shown, in one embodiment, the system can be deployed in a feeder terminal unit (FTU), a distribution terminal unit (DTU), a relay protection device, a station-end fault detection device, or an edge computing unit with synchronous sampling capability. The system includes a zero-sequence signal acquisition module, a data quality and time alignment module, a TK energy processing module, a dual-timescale accumulation module, a normal operating condition baseline module, a state control module, and a regression reconstruction module; in dual-channel verification or degradation processing scenarios, it also includes a channel consistency verification module and a degradation detection module. Figure 1 The arrows in the diagram indicate the transmission relationship between the sampling data, TK energy, baseline threshold, verification results, and status control information in each module.

[0080] The zero-sequence signal acquisition module connects to the zero-sequence voltage and / or zero-sequence current sampling interface to acquire continuous sampling data from at least one zero-sequence signal channel and outputs the sampled value, sampling time, and data quality information. The data quality and time alignment module receives the continuous sampling data and data quality information, and determines the valid sampling data based on sampling continuity, missing samples, sampling saturation, timestamp continuity, and channel online status. When dual channels exist, time alignment is also performed according to sampling time or sampling sequence number, and the quality-verified sampling sequence is output to the TK energy processing module.

[0081] The TK energy processing module receives the quality-verified sampled sequence, performs TK energy filtering according to the aforementioned Teager-Kaiser energy, absolute value, and median filtering to obtain the TK energy filtered sequence, and outputs it to the dual-timescale accumulation module. The dual-timescale accumulation module maintains a short window and a long window on the same TK energy filtered sequence, and outputs the short-window TK energy and the long-window TK accumulated energy, respectively.

[0082] The normal operating condition baseline module receives the short-window TK energy, the long-window TK accumulated energy, and the data validity status. Under normal conditions, it establishes a background sample set and a normal operating condition baseline using normal operating condition samples that have passed quality verification and do not have short-window anomalies. The baseline determines the short-window pre-trigger threshold and the long-window confirmation threshold. The status control module receives the short-window TK energy, the long-window TK accumulated energy, the short-window pre-trigger threshold, and the long-window confirmation threshold. When the short-window TK energy reaches the pre-trigger condition, the status control module sends a status control signal to the normal operating condition baseline module to stop sample admission and stop threshold updates, and switches to a suspected state. In the suspected state, the status control module uses the frozen long-window confirmation threshold to determine the long-window TK accumulated energy, and outputs a fault initiation signal when persistent fault evidence is formed.

[0083] The recovery and reconstruction module receives the fault initiation status, original zero-sequence electrical quantities, long-window TK accumulated energy, and data quality status. After fault recovery, it determines the static stability operating conditions and, when the conditions are met, controls the normal operating condition baseline module to re-establish the background sample set, normal operating condition baseline, short-window pre-trigger threshold, and long-window confirmation threshold using continuous and effective static stability operating data. Thus, all modules are sequentially connected around the same data chain, and the acquired data, intermediate energy, baseline thresholds, and state control results all have clear uses in subsequent modules.

[0084] In one embodiment, the system further includes a channel consistency verification module and a degradation detection module. The channel consistency verification module receives device association information, data validity status, pre-trigger time, short-window TK energy, and long-window TK accumulated energy for two zero-sequence signal channels. When both channels are valid, time-synchronized, and device association is correct, after one channel forms persistent fault evidence, the channel consistency verification module identifies it as the primary triggering channel and determines whether the other channel exhibits a short-window TK energy change exceeding the upper limit of its normal operating baseline within the allowed time difference, outputting a fault evidence consistency result or a channel conflict result.

[0085] The degradation detection module receives channel validity status and channel consistency verification results. When only one channel is valid, two channels are out of sync, or the channel consistency verification module outputs a channel conflict result, the degradation detection module performs a more stringent long-window continuous verification and / or a one-power-frequency cycle abrupt change verification on the valid channel, outputting supplementary fault evidence or insufficient evidence results. The status control module outputs a fault initiation signal based on the fault evidence consistency result or supplementary fault evidence, switches to the pending verification state based on the channel conflict result or insufficient evidence result, and can trigger fault recording. In this way, the verification module and degradation detection module in the system do not provide independent final conclusions in parallel, but provide different evidence states to the status control module, which then uniformly executes the state transition.

[0086] In a complete embodiment, the distribution network uses an arc suppression coil grounding method with an overcompensation degree of 8%. The lines in the simulation model include three-phase cable lines and overhead lines. A ground fault occurs via a 10000Ω circuit when the closing angle is zero, and the fault time is 0.61s. Figure 4 As shown, the zero-sequence voltage gradually changed from small fluctuations to obvious fault oscillations before and after the fault occurred; as Figure 5 As shown, after the fault occurs, the zero-sequence current synchronously exhibits fault transient changes.

[0087] In this embodiment, the first cycle after the fault occurs The primary value is approximately 0.198 kV, which is below a relatively high fixed zero-sequence voltage threshold, making timely triggering difficult. Zero-sequence voltage data from normal operation before the fault is calculated using a basic processing chain of Teager-Kaiser energy, absolute value, median filtering, and a 4-cycle long window accumulation. For example... Figure 6 As shown, the normal operating zero-sequence voltage and its accumulated energy TK reflect the normal background; as Figure 7 As shown, the corresponding adaptive threshold is used to compare with the subsequent real-time TK accumulated energy.

[0088] In this data, the example value of the normal TK accumulated energy after 4 power frequency cycles is approximately 13.156. With a reliability coefficient of 5, the corresponding adaptive startup threshold is approximately 65.7801. After a fault occurs, the long-window TK accumulated energy continues to increase and exceeds this threshold. For example... Figure 8 As shown, the normalized zero-sequence voltage and the accumulated energy TK exhibit corresponding changes near the fault initiation point; as Figure 9 As shown, the accumulated energy of TK continues to increase and exceeds the adaptive threshold, thus forming a startup result. In this embodiment, the long-window TK energy can form a startup result approximately 30ms after a fault occurs.

[0089] When using the dual-timescale method of this invention on the aforementioned data, a short window and a long window with 4 power frequency cycles are maintained simultaneously on the same TK energy filtering sequence. When the short window TK energy reaches the pre-trigger threshold, the baseline of the normal operating condition before the fault is frozen first, so that the subsequently increasing fault TK energy cannot continue to change the long window confirmation threshold; then, the accumulated energy of the long window TK is still used to determine whether the fault continues under the frozen threshold. Thus, the original long window energy detection results can be retained, while adding a processing chain to protect the normal detection baseline before formal confirmation.

[0090] In another complete embodiment, waveform data from a field-operated FTU was used, with the neutral point grounding method being an ungrounded system, and an arcing ground fault occurred. For example... Figure 10 As shown, the field waveform simultaneously contains both zero-sequence voltage and zero-sequence current before and after the fault, and both zero-sequence signals show significant changes after the fault occurs. Therefore, this field waveform can be processed separately using a single-channel TK sliding window, or it can be combined into dual-channel evidence based on equipment association and time synchronization relationships.

[0091] Performing a 4-cycle TK accumulation on the normal zero-sequence voltage data before the fault, the example value of the normal background TK accumulated energy is 16806.3. With a reliability coefficient of 5, the zero-sequence voltage adaptive threshold is approximately 84031.48. For example... Figure 11 As shown, the accumulated energy TK corresponding to the normal zero-sequence voltage before the fault remains below the adaptive threshold; Figure 12 As shown, after a fault occurs, the normalized zero-sequence voltage and the accumulated energy of TK form corresponding startup changes; as Figure 13 As shown, the zero-sequence voltage TK accumulates energy and crosses the adaptive threshold to form the starting point.

[0092] The same TK energy processing is applied to the zero-sequence current in the same field recording. The example value of the TK cumulative energy of the normal zero-sequence current over 4 power frequency cycles is 11.166, and an adaptive threshold for the zero-sequence current is formed based on the same normal background statistical method. Figure 14 As shown, the accumulated energy TK corresponding to the zero-sequence current before the fault remains below the adaptive threshold; as Figure 15As shown, after a fault occurs, the accumulated energy of the zero-sequence current TK increases rapidly and exceeds the corresponding threshold, forming the starting point of the zero-sequence current channel.

[0093] In the dual-channel implementation, after both zero-sequence voltage and zero-sequence current pass data quality verification, pre-trigger times for the two channels can be formed respectively. When one channel first forms long-window persistent fault evidence, it is used as the main trigger channel, and then it is determined whether the other channel exhibits a short-window TK energy change exceeding its own background upper limit within the allowable time difference; if a supporting change exists, then time-consistent dual-channel evidence is formed. This method utilizes the zero-sequence voltage and zero-sequence current information existing simultaneously in the same field recording for mutual verification, without requiring the two channels to have the same fault amplitude or simultaneously reach their respective long-window confirmation thresholds.

[0094] In another embodiment, the system normally operates in dual-channel mode, with the zero-sequence voltage channel continuously active, while the zero-sequence current channel experiences continuous sample loss due to sampling or communication anomalies. The data quality and time alignment module marks the zero-sequence current-related TK energy during the sample loss period as invalid and stops normal background updates for the zero-sequence current; the zero-sequence voltage channel continues to calculate the short-window and long-window TK energy. If the zero-sequence voltage short window does not reach the pre-trigger threshold, the operating state remains normal, but the data during the invalid zero-sequence current period is not included in the corresponding background sample set.

[0095] If the zero-sequence voltage short window reaches the pre-trigger threshold, the state switches to a suspected state and freezes all current normal operating condition baselines and thresholds. Since the zero-sequence current is invalid, dual-channel time consistency verification is not performed; instead, a downgraded verification is performed on the zero-sequence voltage: the frozen long window confirmation threshold is used to determine the accumulated energy of the long window TK, and stricter continuity requirements are adopted; if necessary, a supplementary judgment is made using a phase-change abrupt change over a power frequency cycle. A fault initiation signal is output when supplementary fault evidence is formed; if the evidence is still insufficient, the system remains in a pending verification state and fault recording is triggered. This implementation ensures that a temporary failure of one zero-sequence signal channel does not lead to the failure of all detection functions, while avoiding directly considering a single-channel anomaly as a sufficient condition equivalent to dual-channel consistency evidence.

[0096] In another embodiment, the control process for the normal operating condition baseline under a slowly developing weak ground fault is illustrated. When the system is in normal condition, the normal background sample is updated with new valid operating data. When the weak ground fault begins to develop slowly, the energy of the short-window TK first deviates from the normal background and reaches the pre-trigger threshold, but the accumulated energy of the long-window TK has not yet reached the long-window confirmation threshold. At this time, the operating state is immediately switched to the suspected state, normal operating condition sample admission is stopped, and the last normal operating condition baseline and threshold formed before the fault are frozen.

[0097] Subsequently, even as the original zero-sequence signal and TK energy continue to increase slowly, samples during the suspected event period no longer participate in normal background statistics. Therefore, the long-window confirmation threshold remains at the value before the pre-triggering moment. If the accumulated energy of the long-window TK eventually reaches the frozen long-window confirmation threshold and continues not to fall back to the background range, persistent fault evidence is formed, and a fault initiation signal is output after the applicable dual-channel consistency or single-channel degradation conditions are met. If the energy decays on its own and the long window never reaches the confirmation condition, the suspected event is rejected, and samples during the suspected event period still do not enter the normal background. Normal sample access is restored after the static steady-state observation period ends. This process demonstrates the direct control relationship between the short-window pre-triggering results and the subsequent baseline learning path.

[0098] In alternative implementations, the input zero-sequence signal can be zero-sequence voltage only, zero-sequence current only, or both zero-sequence voltage and zero-sequence current. When using a single channel, verification is performed according to the long window persistence and optional abrupt change supplementation conditions; when using a dual channel, time consistency and device association verification are added. Both methods share the same TK energy processing, short window pre-triggered, baseline freezing, and regression reconstruction mechanisms.

[0099] In alternative implementations, the median filtering window can use 3, 5, or 7 points, with 5 points being the preferred implementation. A long window can use 4 power frequency cycles, or 2 to 4 power frequency cycles depending on the transient duration characteristics of the fault to be detected. A short window is smaller than a long window and can use 0.5 to 1 power frequency cycle. The window length can be converted into the number of sampling points per power frequency cycle to adapt to different sampling frequencies.

[0100] In alternative implementations, the normal operating condition threshold can be calculated by multiplying the median of the normal sample by a reliability coefficient and combining it with the absolute lower limit, or by using the median of the normal sample, the median absolute deviation, and the absolute lower limit to form a short-window pre-trigger threshold and a long-window confirmation threshold. Regardless of the threshold statistical method used, the state constraint is maintained so that normal samples are only allowed to enter under normal conditions, frozen in advance under suspected conditions, and re-established after returning to static stability.

[0101] In alternative implementations, dual-channel consistency is determined by the pre-trigger time difference and whether the supporting channel exceeds its normal background range, without using subjective weights for comprehensive scoring; if the dual-channel conditions cannot be met due to time mismatch, invalid channels, or conflicting evidence, the process switches to a degradation path of single-channel persistence and / or one power frequency cycle mutation.

[0102] In alternative implementations, the fault initiation signal can be used to trigger fault recording, fault line selection, fault section location, or subsequent protection judgment. As the initiation quantity for subsequent functions, the fault initiation signal does not require the method itself to directly generate a tripping action; in implementation scenarios linked with protection actions, the existing downstream protection logic can continue to perform the verification of protection range, direction, time limit, and corresponding blocking conditions. In the description of this specification, the terms "an embodiment," "example," "specific example," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described can be combined in any suitable manner in one or more embodiments or examples.

[0103] As is known from common technical knowledge, this invention can be implemented through other embodiments that do not depart from its spirit or essential characteristics. Therefore, the disclosed embodiments described above are merely illustrative in all respects and are not the only ones. All modifications within the scope of this invention or its equivalents are included in this invention.

Claims

1. A method for rapid start-up of a distribution network under minor faults, characterized in that, include: Acquire continuous sampling data of at least one zero-sequence signal channel in the distribution network, wherein the zero-sequence signal channel is a zero-sequence voltage channel and / or a zero-sequence current channel; The valid sampled data in the continuous sampled data are sequentially subjected to Teager-Kaiser energy calculation, absolute value processing and median filtering to obtain the TK energy filtered sequence. Based on the TK energy filtering sequence, short-window TK energy and long-window TK cumulative energy are respectively generated; Under normal conditions, a normal operating condition baseline is established based on normal operating condition samples that have been verified by data quality. Based on the normal operating condition baseline, a short window pre-trigger threshold and a long window confirmation threshold are determined. The normal operating condition baseline and the threshold are updated only using subsequent sampling data where the short window TK energy has not reached the short window pre-trigger threshold. When the short window TK energy of any zero-sequence signal channel reaches the short window pre-trigger threshold, the operating state is switched to the suspected state, the normal operating condition sample admission is stopped, and the normal operating condition baseline and the threshold are frozen. In the suspected state, the long window TK accumulated energy of the zero-sequence signal channel is judged by using the frozen long window confirmation threshold. When the long window TK accumulated energy reaches the long window confirmation threshold and does not fall back to the background range determined by the normal operating condition baseline within a set duration, a fault start signal is output and kept frozen. After the fault is recovered, the normal operating condition baseline and the threshold are re-established using continuous and effective static and stable operating data, and the normal state is restored.

2. The method for rapid start-up of weak faults in a distribution network according to claim 1, characterized in that, The valid sampled data are sequentially subjected to Teager-Kaiser energy calculation, absolute value processing, and median filtering to form the short-window TK energy and the long-window TK cumulative energy, including: For zero-sequence signal channel The Sample values The Teager-Kaiser energy is determined according to the following formula: in, Indicates the channel type. hour This is the zero-sequence voltage sample value. hour The zero-sequence current sample value; for The absolute value is taken to obtain the non-negative energy sequence, and the median filtering is applied to the non-negative energy sequence to obtain the TK energy filtered sequence. Based on the TK energy filtering sequence, respectively according to the short window length and long window length By performing sliding accumulation, the short-window TK energy and the long-window TK accumulated energy are obtained, where .

3. The method for rapid start-up of weak faults in a distribution network according to claim 1, characterized in that, Determining the short-window pre-trigger threshold and the long-window confirmation threshold based on the normal operating condition baseline includes: For any zero-sequence signal channel q, establish a short-window background sample set composed of the energy of the effective short window TK under normal conditions and a long-window background sample set composed of the accumulated energy of the effective long window TK. Determine the median of the short-window background sample set and the long-window background sample set, respectively. and median absolute deviation ,in Indicates a short window. Indicates a long window; according to Determine the short window pre-trigger threshold, and follow the instructions. Determine the long window confirmation threshold; wherein, and For the reliability coefficient, and , and These are the lower limits of the short window threshold and the lower limit of the long window threshold, respectively, corresponding to the noise floor of the device.

4. The method for rapid start-up of weak faults in a distribution network according to claim 1, characterized in that, When the short-window TK energy of any zero-sequence signal channel reaches the short-window pre-trigger threshold, the operating state is switched to a suspected state, including: When the energy of the short window TK reaches the short window pre-trigger threshold and continues to reach the preset minimum number of sampling points or the preset minimum time, the pre-trigger time of the zero-sequence signal channel is recorded, and the operating state is switched from normal state to suspected state. When switching to the suspected state, freeze the normal operating condition background sample, normal operating condition baseline, short window pre-trigger threshold and long window confirmation threshold, and record the original zero-sequence signal, TK energy, short window TK energy and long window TK cumulative energy during the suspected event. When the energy of the short window TK drops to the background range determined by the normal operating condition baseline and the accumulated energy of the long window TK does not reach the long window confirmation threshold, the suspected event is judged as a disturbance, and the data during the suspected event is prohibited from being added to the normal operating condition sample. The normal state is restored after the static observation time ends.

5. The method for rapid start-up of weak faults in a distribution network according to claim 1, characterized in that, When both the zero-sequence voltage channel and the zero-sequence current channel are valid, time-synchronized, and the device association relationship conforms to the preset electrical association relationship, the output fault start signal includes: After the long window TK energy accumulation of a zero-sequence signal channel forms evidence of a persistent fault, the zero-sequence signal channel is determined as the main trigger channel, and it is determined whether the other zero-sequence signal channel has a short window TK energy change exceeding the background upper bound of its normal operating condition baseline within the allowable time difference relative to the pre-trigger time of the main trigger channel. When the energy change of the short window TK occurs in another zero-sequence signal channel, fault evidence that the formation time of the two zero-sequence signal channels is consistent is determined and the fault start signal is output. When the main trigger channel generates continuous fault evidence, and another zero-sequence signal channel does not show a short-window TK energy change exceeding the background upper limit of its normal operating condition baseline within the extended observation time, the operating state will be switched to the pending verification state, and the fault start signal will not be output.

6. The method for rapid start-up of weak faults in a distribution network according to claim 1, characterized in that, Performing data quality verification on the continuously sampled data includes: The validity of the sampled data is determined based on sampling continuity, missing samples, sampling saturation, timestamp continuity, and channel online status. When any of the three adjacent sample values ​​involved in the Teager-Kaiser energy calculation is invalid, the corresponding Teager-Kaiser energy is marked as invalid, and the Teager-Kaiser energy is not calculated using the interpolated sample value; When the valid TK energy sampling in a short window or long window does not meet the window validity requirements, the corresponding short window TK energy or long window TK accumulated energy is marked as invalid, and the data of the corresponding window is prohibited from being used to update the normal operating condition baseline or form a fault confirmation basis. When the time difference between the zero-sequence voltage channel and the zero-sequence current channel exceeds the allowable time difference, stop the time consistency verification between the two zero-sequence signal channels, and perform fault judgment according to the single-channel method respectively.

7. The method for rapid start-up of weak faults in a distribution network according to claim 1, characterized in that, When only one zero-sequence signal channel is valid, or when time consistency verification between two zero-sequence signal channels cannot be performed, a downgrade verification is performed on the valid zero-sequence signal channel, including: The accumulated energy of the effective zero-sequence signal channel within the long window TK is determined using a frozen long window confirmation threshold, and the long window duration requirement is increased; and / or, Determine the abrupt change between the current sampled value of the effective zero-sequence signal channel and the in-phase sampled value one power frequency cycle ago. When the abrupt change exceeds the abrupt change threshold determined by normal operating conditions at multiple consecutive sampling points, and the occurrence time of the abrupt change is adjacent to the occurrence time of the long window TK accumulated energy anomaly, determine supplementary fault evidence. When the downgrade verification generates supplementary fault evidence, the fault initiation signal is output; when the downgrade verification does not generate supplementary fault evidence, the verification status remains pending and fault recording is triggered.

8. The method for rapid start-up of weak faults in a distribution network according to claim 1, characterized in that, After fault recovery, the normal operating condition baseline and the threshold are re-established using continuous and effective static stable operating data, including: When the original zero-sequence electrical quantity of the zero-sequence signal channel that was started due to a fault recovers to the allowable operating range, and the accumulated energy of the long window TK drops below the reset threshold and continues to reach the reset delay time, the operating state is switched to the reset state. In the restored state, the data from the fault period and the suspected event period are no longer used to update the normal operating baseline, and continuous static and stable operating data are collected. When the continuous static and stable operation data reaches the preset static and stable period and the data quality is valid, the normal operating condition baseline, short window pre-trigger threshold and long window confirmation threshold are re-established using the continuous static and stable operation data, and the normal state is restored. If the short window TK energy reaches the short window pre-trigger threshold again during the process of re-establishing the normal operating condition baseline, the re-establishment of the normal operating condition baseline is stopped and the system switches to the suspected state.

9. A rapid start-up system for minor faults in a power distribution network, characterized in that, include: The zero-sequence signal acquisition module is used to acquire continuous sampling data of at least one zero-sequence signal channel in the power distribution network, and output the sampling value, sampling time and data quality information. The zero-sequence signal channel is a zero-sequence voltage channel and / or a zero-sequence current channel. The data quality and time alignment module is used to receive the continuous sampled data and the data quality information, determine the valid sampled data, and output the sampled sequence that has been quality verified. The TK energy processing module is used to receive the quality-verified sampling sequence, perform Teager-Kaiser energy calculation, absolute value processing and median filtering on the valid sampled data in sequence, and output the TK energy filtered sequence. A dual-timescale accumulation module is used to receive the TK energy filter sequence and form short-window TK energy and long-window TK accumulated energy, respectively. The normal operating condition baseline module is used to establish a normal operating condition baseline based on normal operating condition samples that have been verified by data quality under normal conditions, determine a short window pre-trigger threshold and a long window confirmation threshold based on the normal operating condition baseline, and update the normal operating condition baseline and the threshold only using subsequent sampling data where the short window TK energy has not reached the short window pre-trigger threshold. The status control module is used to receive the short window TK energy, the long window TK accumulated energy, the short window pre-trigger threshold, and the long window confirmation threshold. When the short window TK energy reaches the short window pre-trigger threshold, it switches to the suspected state and controls the normal operating condition baseline module to stop sample admission and threshold update. In the suspected state, it uses the frozen long window confirmation threshold to determine the long window TK accumulated energy and outputs a fault start signal when continuous fault evidence is formed. The recovery and reconstruction module is used to acquire continuous and effective static and stable operating data after fault recovery, and to control the normal operating condition baseline module to re-establish the normal operating condition baseline, short window pre-trigger threshold and long window confirmation threshold based on the static and stable operating data.

10. The rapid start-up system for weak faults in the power distribution network according to claim 9, characterized in that, It also includes a channel consistency verification module and a degradation detection module; The channel consistency verification module is used to receive the pre-trigger time, short window TK energy and long window TK accumulated energy of two zero-sequence signal channels when both zero-sequence voltage channel and zero-sequence current channel are valid, time-synchronized and the device association relationship conforms to the preset electrical association relationship. After a continuous fault evidence is formed in one zero-sequence signal channel, it determines whether the other zero-sequence signal channel has a short window TK energy change that exceeds the background upper limit of its normal operating condition baseline within the allowable time difference, and outputs the fault evidence consistency result or channel conflict result. The degradation detection module is used to perform long-window continuous verification and / or one power frequency cycle mutation verification on the valid zero-sequence signal channel when only one zero-sequence signal channel is valid, the two zero-sequence signal channels are out of time, or the channel consistency verification module outputs channel conflict results, and outputs supplementary fault evidence or evidence insufficient results to the status control module. The state control module is also used to output a fault start signal based on the consistency result of the fault evidence or the supplementary fault evidence, and switch to the pending verification state based on the channel conflict result or the insufficient evidence result.

Images