A method for quickly locating branch faults of a cable branch box

Abstract

This invention discloses a method for rapid fault location in cable branch boxes, specifically relating to the field of power distribution fault location technology. It addresses the problem that existing cable branch boxes, in operating scenarios where multiple branches are simultaneously disturbed, struggle to effectively distinguish between shared disturbances and individual branch anomalies, thus hindering the rapid and accurate identification of specific faulty branches. By establishing a reference set of branches within the same box, forming a simultaneous event window group around the abnormal event, constructing background disturbance trajectories, extracting differential response trajectories, and creating branch fault fingerprints, and then executing an exclusive decision according to a progressive rule of first-to-start change, strong deviation, and persistent residual faults, the method achieves stable identification of specific faulty branches even in scenarios with shared disturbances across multiple branches, thus solving the problem of existing technologies' difficulty in rapidly and accurately locating specific faulty branches.

Description

Technical Field

[0001] This invention relates to the field of power distribution fault location technology, specifically a method for rapid location of branch faults in cable branch boxes. Background Technology

[0002] Cable branch boxes, as tapping nodes in power distribution networks, typically connect multiple outgoing branches and are widely used in industrial parks, commercial complexes, factory power distribution, and municipal power distribution scenarios. Current technologies for diagnosing abnormalities in cable branch boxes primarily employ fault indication, threshold exceeding alarms, single-branch current or zero-sequence quantity monitoring, and section-level fault diagnosis to detect and warn of abnormal states. While these technologies can reflect the presence of abnormalities in branch boxes or power supply sections to a certain extent, their judgments are mostly limited to the box level, section level, or single-measurement point level. In actual operation and maintenance, further confirmation of the specific fault location is still necessary through manual inspection, branch-by-branch verification, or power outage / restoration operations.

[0003] When multiple branches in a cable distribution box are operating simultaneously, each branch is often affected by fluctuations in the upstream power supply, bus-side disturbances, load switching within the box, and environmental changes. Different branches may simultaneously experience current fluctuations, zero-sequence offsets, or short-term abnormal responses. Current technologies in this scenario typically focus on judging based on single-branch measurement results or individual anomalies, rarely considering the interrelationships and shared disturbances among multiple branches within the same box under the same event. Therefore, it's easy to only detect anomalies but struggle to further distinguish between shared disturbance components and the branch's own anomaly components. Consequently, even if an anomaly in the target cable distribution box can be identified, it's often difficult to quickly determine which specific branch caused the actual fault.

[0004] In scenarios where multiple branches in a cable distribution box operate simultaneously and are affected by disturbances, existing technologies struggle to effectively distinguish between shared disturbances and individual branch anomalies without relying on repeated manual checks, and further hinder the rapid and stable identification of specific faulty branches. This issue directly impacts fault confirmation efficiency and the targeted nature of subsequent handling, and constitutes a key technical problem that future solutions need to address. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a method for rapid fault location in cable branch boxes, thereby resolving the problems mentioned in the background section.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for rapid location of branch faults in a cable branch box, comprising: S1. Collect the current response sequence, zero-sequence response sequence and load fluctuation sequence of each branch of the target cable branch box during normal operation, extract the steady-state offset boundary, zero-sequence response stable interval, fluctuation rhythm boundary and recovery time boundary of each branch, and establish a reference set of branches in the same box. S2. When any branch path experiences an abnormal response, the first abnormal moment is used as the event anchor point. The leading window, action window and reset window of all branches are simultaneously captured to form a synchronous event window group in the same box. S3. Based on the synchronous event window group of the same box, the main abnormal branch is screened out, and the background branch set is screened from the remaining branches. The common disturbance components of the background branches at each time are aggregated to generate the background disturbance trajectory of the same box. S4. Subtract the background disturbance trajectory in the same box from the event response sequence of each branch to be judged to generate the difference response trajectory of the corresponding branch; S5. Extract the initial partial sequence quantity, coupling deviation quantity and regression residual quantity based on the different response trajectories of each branch to form the branch fault fingerprint, and execute the exclusive decision according to the progressive rule of first change, strong deviation and residual persistence. S6. Identify the branch that meets the exclusive decision condition as the faulty branch, perform a response verification on the branch in the subsequent observation period, and update the reference set of branches in the same box based on the verification results.

[0007] Furthermore, S1 includes: When establishing a reference set for branches in the same box, the current response sequence, zero-sequence response sequence, and load fluctuation sequence of each branch are synchronously collected according to a unified time base. The collected results are sorted in order, missing time slices are filled in, and sampling spikes are removed. During normal operation, the steady-state offset boundary, zero-sequence response stable interval, fluctuation rhythm boundary, and recovery time boundary are extracted and written into the reference file.

[0008] Furthermore, S2 includes: When any branch experiences an abnormal response, the moment of the first abnormality is taken as the event anchor point for the synchronous windowing of all branches. The leading window, action window, and return window of all branches are synchronously captured to form a synchronous event window group in the same box. For branch time segments that meet the low confidence determination criteria within the candidate cutoff window range, they are recorded as low confidence segments of the corresponding branch and retained in the same box synchronous event window group.

[0009] Furthermore, S3 includes: Based on the synchronous event window group of the same box, the main abnormal branch is determined by exclusive comparison in the order of first boundary crossing time, number of consecutive boundary crossings, number of consecutive deviations from zero-sequence response, and number of consecutive non-return boundary within the recovery window. Then, from the remaining branches, branches corresponding to low-confidence segments, branches whose leader window stability is lower than the stability threshold, and branches whose number of persistent fragments at the non-return boundary in the return window exceeds the upper limit of their respective return duration boundaries are filtered out to form a background branch set containing at least two branches.

[0010] Furthermore, S3 also includes: Align all branches within the background branch set with the same time slice in the leading window, action window, and return window according to the event anchor point, and aggregate the common current disturbance value and common zero-sequence disturbance value at each moment to form the background disturbance trajectory in the same box. The background disturbance trajectory in the same box consists of a common current disturbance sub-trajectory and a common zero-sequence disturbance sub-trajectory.

[0011] Furthermore, S4 includes: After aligning the event response sequence of each branch to be judged within the leading window, action window and return window with the background disturbance trajectory of the same box using the same time slice, subtracting each slice one by one, a difference response trajectory composed of current difference sub-trajectory and zero sequence difference sub-trajectory is generated. When the length of the missing segment of the common disturbance does not exceed the number of missing time slices for the missing detection judgment, a missing detection supplementation mark is written into the difference response trajectory; When the length of the missing segment of the common disturbance exceeds the number of missing time slices for the missing detection judgment, a long missing detection degradation marker is written into the differential response trajectory.

[0012] Furthermore, S5 includes: Based on the differential response trajectories of each branch to be judged, the initial partial sequence quantity is extracted in the initial section of the action window, the coupling deviation quantity is extracted in the main section of the action window, and the regression residual quantity is extracted in the regression window. The initial partial sequence quantity, coupling deviation quantity and regression residual quantity constitute the branch fault fingerprint.

[0013] Furthermore, S5 also includes: Based on the fault fingerprint of each branch to be judged, exclusive decision is performed in the order of initial partial sequence, coupling deviation and regression residual, priority branch is determined and written into the event-level decision record.

[0014] Furthermore, S6 includes: When the verification passes, the abnormal sample segment in the event corresponding to the faulty branch is registered as an abnormal sample segment. If the backtest fails, the event segment that meets the condition of returning to the zero-difference stable interval will be registered as a candidate sample segment to be included again. If the verification is insufficient, the section to be verified in the corresponding event is registered as the section to be verified, and the reference set of the branch in the same box is updated accordingly.

[0015] Compared with the prior art, the present invention has the following beneficial effects: 1. By establishing a reference set of branches in the same box and forming a simultaneous event window group around the abnormal event, the background disturbance trajectory of the same box is constructed, the difference response trajectory is extracted, and the branch fault fingerprint is formed for the event response sequence of each branch. The exclusive decision is executed according to the progressive rule of first change, strong deviation, and residual persistence. In this way, even in the operation scenario where multiple branches of the cable branch box are affected by the common disturbance at the same time, the common disturbance component can still be distinguished from the abnormal component of the branch itself, and the specific fault branch can be stably locked. This solves the problem in the existing technology that can only judge the abnormality of the box or the fault section, but it is difficult to quickly and accurately locate the specific fault branch.

[0016] 2. By setting a subsequent observation period to perform response verification after the candidate fault branch is determined, and distinguishing and registering abnormal sample segments, re-included candidate sample segments and information to be verified based on the verification results, and then incrementally updating the reference set of branches in the same box, the system can avoid the direct impact of misjudgment of a single event on the long-term reference boundary, enhance the consistency of identification of similar events in the future and the adaptive capability of the reference set, and further improve the reliability of the fault branch confirmation results and the discrimination stability during continuous system operation. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the overall process of the rapid fault location method for cable branch boxes according to the present invention. Figure 2 A schematic diagram of the on-site deployment structure of a rapid fault location system for a target cable branch box; Figure 3 A timing diagram for constructing a synchronous event window group around an event anchor point; Figure 4 A schematic diagram for background branch set filtering and generation of background disturbance trajectory in the same box; Figure 5 A schematic diagram for generating differential response trajectories and extracting branch fault fingerprints; Figure 6 This diagram illustrates the process of exclusive decision-making, fault branch confirmation, and updating the reference set for branches within the same box. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Example: Combined with Appendix Figure 1-6 This embodiment provides a method for rapid location of branch faults in cable branch boxes, including: S1. Collect the current response sequence, zero-sequence response sequence, and load fluctuation sequence of each branch in the target cable branch box during normal operation. Extract the steady-state offset boundary, zero-sequence response stable interval, fluctuation rhythm boundary, and recovery time boundary of each branch. Establish a reference set for branches in the same box. The specific implementation is as follows: When collecting the current response sequence, zero-sequence response sequence, and load fluctuation sequence of each branch of the target cable branch box during normal operation, the field acquisition unit installed inside the target cable branch box synchronously acquires data from each branch according to a unified time reference. The field acquisition unit is fixedly connected to the monitoring terminal on the branch box side. The monitoring terminal has a built-in edge controller and edge recording module. The field acquisition unit is responsible for sampling the measurement points, the monitoring terminal is responsible for time synchronization verification, sequence organization, and uploading control, the edge controller is responsible for noise reduction, completion, and boundary generation, and the edge recording module is responsible for version storage, evidence chain recording, and downstream access. The target cable branch box is a power distribution node where the upstream incoming line is branched to multiple outgoing branches via the busbar or connecting busbar inside the box. Each branch is an independent power supply circuit drawn from the target cable branch box.

[0020] The current response sequence is taken from the sampling values ​​of the secondary side of the current transformer at the fixed measuring point at the outgoing end of each branch. The record name is "Branch Current Instantaneous Sampling Value", and the unit is Amperes. The sampling rhythm is set to 20 milliseconds or 40 milliseconds. For branches where the number of current changes within 30 minutes accounts for more than 5% of the total number of sampling pieces, 20 milliseconds is used, and for the remaining branches, 40 milliseconds is used. The default value for the first collection is 20 milliseconds. The overall acquisition error is controlled within ±1.0% of the rated range.

[0021] The zero-sequence response sequence is taken from the output value of the zero-sequence acquisition component installed on the outside of the three-phase cable of a single branch. The record name is the branch zero-sequence response sampling value, and the unit is ampere. The sampling rhythm is consistent with the current response sequence.

[0022] When the site only has the conditions for sampling three-phase branch current and the three-phase measuring points are located at the same installation section, the sampling time deviation does not exceed 2 milliseconds, and the three-phase sampling missing rate does not exceed 0.5% within 72 consecutive hours, the three-phase branch current sampling values ​​within the same time slice can be summarized to form an equivalent zero-sequence response value. This alternative approach is only used for three-phase complete sampling branches and is not used for sampling branches with missing phases, branches with non-coplanar measuring points, or branches with a long-term three-phase imbalance exceeding 15%.

[0023] The load fluctuation sequence is formed within a continuous time slice of the same branch. First, the current difference between adjacent time slices is calculated in the order of timestamps. Then, the direction of change is marked according to the positive, negative and zero directions of the difference. Then, the continuous time slices in the same direction are merged into a single change segment. The start time, end time, amplitude and duration of the segment are recorded. A segment with a duration of less than 3 time slices and an amplitude that does not exceed 1.5% of the rated current of the corresponding branch is recorded as a stable segment. A segment with a duration of 3 time slices and an amplitude that exceeds 1.5% of the rated current is recorded as a fluctuating segment. A segment with a duration of 10 time slices and an amplitude that exceeds 5% of the rated current is recorded as a switching segment.

[0024] To ensure comparability across different branches, the monitoring terminal performs clock calibration before data acquisition begins, prioritizing alignment of the local clock with the station's time source. If the station's time source fails, the distribution automation terminal clock is switched, and if that fails again, the BeiDou time module is switched. Alignment deviation is controlled within 2 milliseconds. If it exceeds 2 milliseconds, the current data acquisition task is paused and the time base mismatch event is recorded. Time slices that have been acquired but not yet entered into the database are invalidated and no re-acquisition is performed. Data is re-registered from the next complete time slice after all branches have completed recalibration. After the raw records enter the edge controller, they are first sorted in order by branch identifier, measurement point identifier, and timestamp. For duplicate records with completely identical timestamps, branch identifiers, and measurement point identifiers, only the first valid record is retained.

[0025] Missing time slices are filled in according to the filling boundary of no more than 3 consecutive time slices. If one consecutive time slice is missing, it is filled in with the median value of the next and next valid values. If two consecutive time slices are missing, it is filled in with the value at equal intervals from the previous value to the next value. If three consecutive time slices are missing, it is filled in with the previous value, the median value, and the next value. The filling records are all written to the filling mark and are not included in the calculation of the center value of the stable segment. Segments that are missing more than 3 consecutive time slices, have the beginning and end of the window missing at the same time, still have time sequence inversion after filling, or have the filled segment accounting for more than 20% of the total number of slices in the consecutive segment are marked as unusable segments and are not included in the boundary generation.

[0026] During denoising, the absolute value of the difference between adjacent time slices is used as the comparison metric. When the difference of a single time slice exceeds three times the average of the absolute values ​​of the differences of the five valid time slices before and after it, and the change only lasts for one or two time slices, and it recovers to within ±1.5% of the level before the change within the next three time slices, it is judged as a sampling glitch and replaced by a linear transition between adjacent valid values. If the change lasts for three time slices, or if it does not return to the aforementioned range after recovery, it is retained as a true change.

[0027] During normal operation, the edge controller, in conjunction with the switch position status, protection action records, manual maintenance interlock status, upstream power supply switching records, bus transfer records, substation voltage over-limit records, and branch current continuity, jointly confirms the operation. Only when the switch position in the box does not change, the protection does not operate, the maintenance interlock is not set, the upstream power supply is not switched, the bus is not transferred, the substation phase voltage is within 90% to 110% of the rated value for 10 consecutive time periods, the branch current is continuous, and no fault triggering mark appears, will the corresponding time period be included in the normal operation period. The first version of the reference set for branches in the same box must cover at least 2 complete day and night cycles and include no less than 20 fluctuation segments and no less than 5 switching segments. If this sample size is not met, it will continue to accumulate, up to a maximum of 168 hours.

[0028] During normal operation, segments that are unusable, switching segments, glitch replacement segments with a proportion exceeding 20%, and segments with time base mismatch markers are first removed. Then, segments with a length of no less than 300 time slices among the remaining continuous segments are selected as candidates for stable segments.

[0029] The steady-state offset boundary is generated separately for each branch. First, the median value of all current response samples in each stable section is taken as the center value of the section. Then, the deviation of each time slice relative to the center value is calculated. The upper and lower limits of the absolute value of the deviation within the 95% coverage range of the section are retained. Finally, the upper and lower limits of each stable section are weighted and merged according to the time length to obtain the steady-state offset boundary of the branch.

[0030] Zero-sequence response stable intervals are generated separately for each branch. First, the zero-sequence response sample values ​​corresponding to all stable sections during normal operation are collected in timestamp order. For each stable section, the median value of all zero-sequence response sample values ​​is taken as the zero-sequence center value of that section. Then, the deviation of each time slice from the zero-sequence center value is calculated. The upper and lower limits of the absolute value of the deviation within the 95% coverage range of that section are retained. Finally, the upper and lower limits of each stable section are weighted and merged according to the time length to obtain the zero-sequence response stable interval of that branch. When the deviation of the zero-sequence response center value of the same branch in different day and night periods exceeds 20% of the width of the aforementioned weighted merged interval, the zero-sequence response stable interval is generated separately for day shift, night shift, and night.

[0031] The boundaries of the fluctuation rhythm are collected by the same time period within the day. Only when the fluctuations repeat within the same time period for two consecutive days and nights, and the deviation of the start time does not exceed 120 seconds, the deviation of the amplitude does not exceed 20% of the previous amplitude, and the deviation of the duration does not exceed 20% of the previous duration, are such fluctuations included in the repeatable rhythm. Based on this, the range of start time, the range of duration, and the range of amplitude are generated to form the boundaries of the fluctuation rhythm. The boundaries of the non-periodic impact load branch are established separately for the three time periods of day shift, night shift, and night.

[0032] The recovery time boundary is timed starting from the beginning of a normal load change event. The starting point is the first time slice when the current response sequence first exceeds the steady-state offset boundary for three consecutive time slices, and the ending point is the first time slice when the current response sequence returns to the steady-state offset boundary and remains there for 10 consecutive time slices. If the current response sequence exceeds the boundary again within three time slices after returning to the boundary, the timing is reset. The recovery time boundary is formed by covering 95% of the recovery time obtained from no less than 20 normal load change events.

[0033] The aforementioned steady-state offset boundary, zero-sequence response stable interval, fluctuation rhythm boundary, and recovery duration boundary are all written into the same reference file. The reference file includes at least the task number, target branch box identifier, branch identifier, measurement point identifier, acquisition start and end time, time reference version number, denoising rule version number, completion rule version number, boundary generation version number, timekeeping mark, record integrity mark, original record summary fingerprint, and generation time. It is stored in read-only mode on the industrial storage medium of the edge recording module, and the summary is synchronously sent to the station host for use in the next step.

[0034] Collection creation tasks with the same target branch box, the same time interval, and the same version number are controlled by idempotent keys, and duplicate requests will directly return existing files.

[0035] The minimum feasible set of the interface includes the task number, target branch box identifier, branch identifier set, time interval, version number, timekeeping flag, record integrity flag, and response status field. The response status field returns five types of error codes: success, missing measurement point, time base mismatch, insufficient valid samples, and invalid file version. Among them, missing measurement point corresponds to the number of valid branches being less than 80% of the total number of branches; time base mismatch corresponds to any branch's time synchronization deviation exceeding 2 milliseconds for 30 seconds; insufficient valid samples corresponds to not meeting two complete day-night cycles or sample count requirements; and invalid file version corresponds to the downstream requested version being inconsistent with the currently locked version. The local reference set is generated only when the number of valid branches reaches more than 80% of the total number of branches and offline branches are offline for no more than 1 hour continuously. It is only used for daily status tracking of a single branch and screening before anomaly triggering, and is not used for exclusive adjudication of the entire box.

[0036] In a preferred embodiment, the target cable branch box includes four outgoing branches with a rated current of 250 amps, a sampling rhythm of 20 milliseconds, continuous sampling for 72 hours, a steady-state offset boundary of branch A of 198 amps to 214 amps, a zero-sequence response stable range of 0.12 amps to 0.46 amps, a typical fluctuation duration corresponding to the fluctuation rhythm boundary of 4 seconds to 18 seconds, and a return time boundary of 6 seconds to 22 seconds.

[0037] Alternatively, the edge controller and the edge recording module can be integrated into the same hardware. As long as the sampling, timing, noise reduction, completion, boundary generation, version locking, and calling caliber remain consistent, they are equivalent implementations. The entire process is limited to the scope of power distribution operation monitoring and fault location assistance, without changing the protection settings of primary equipment, without replacing the legally mandated protection action chain, and without exceeding the maintenance isolation permission boundary.

[0038] S2. When any branch experiences an abnormal response, the moment of the first abnormality is used as the event anchor point. The leading window, action window, and recovery window are simultaneously captured for all branches to form a synchronous event window group within the same box. The specific implementation is as follows: When any branch exhibits an abnormal response, the monitoring terminal, which is fixedly connected to the target cable branch box, continuously reads the reference set of branches in the same box established in the previous step, and compares the current response sequence, zero-sequence response sequence, and load fluctuation sequence sent by each branch in real time according to a unified time base. When any branch simultaneously meets any of the following conditions for three consecutive time slices, it is confirmed that the branch has an abnormal response. The conditions include the current response sequence exceeding the steady-state offset boundary, the load fluctuation sequence exceeding the fluctuation rhythm boundary, and the zero-sequence response sequence exceeding the zero-sequence response stable range.

[0039] Among them, the abnormal response is the abnormal deviation state of the branch relative to the reference set of branches in the same box during the current operating period. It is obtained by continuously comparing the instantaneous sampled value of the branch current, the sampled value of the branch zero-sequence response, and the load fluctuation sequence segment through the monitoring terminal on site.

[0040] The first abnormal moment is the first valid time slice moment that meets the abnormal response judgment conditions. In one embodiment, the first time slice is recorded as the first abnormal moment only when the first two time slices out of three consecutive time slices are valid slices and at least two of the three time slices do not have a padding mark and do not belong to the glitch replacement segment, so as to avoid misrecording the sampling jitter as the event start point.

[0041] The event anchor point is taken from the moment of the first anomaly and serves as a unified benchmark for the synchronous windowing of all branches. The event anchor point is written into the event registration file by the edge controller. The event registration file includes at least the task number, target branch box identifier, triggering branch identifier, event anchor point time, trigger basis field, time base version number, and boundary generation version number. It is stored in the edge record module in an increment-only manner and the summary is synchronously sent to the station host for direct use in the next step.

[0042] When capturing the leading window, action window and return window simultaneously for all branches, the leading window is a continuous segment of the operation of each branch in the same box before the event anchor point, used to preserve the stable background and load change background before the anomaly occurs. It is taken from the circular buffer of the edge recording module on site. The circular buffer saves the most recent continuous sampling record of each branch in ascending time order. Preferably, it can be set to cover 30 seconds to 300 seconds before the event anchor point.

[0043] The action window is an abnormal evolution segment captured from the event anchor point. It is formed by the monitoring terminal continuously collecting real-time sampling values ​​of each branch on-site. The end time of the action window is the time corresponding to the event anchor point plus the length of the action window. Preferably, it can be set to cover 2 seconds to 60 seconds after the event anchor point. The recovery window is a recovery segment captured after the action window ends. The first piece of the recovery window is the next time piece after the last piece of the action window. It is used to record whether each branch returns to the steady-state offset boundary and the fluctuation rhythm boundary. Preferably, it can be set to cover 10 seconds to 300 seconds after the action window ends.

[0044] The lengths of the preamble window, action window, and return window are determined by the edge controller based on the branch type of the trigger branch, the fluctuation rhythm boundary formed in the previous step, and the return duration boundary. For branches where the load fluctuation sequence changes more than 5% of the total number of samples within 30 consecutive minutes, a default combination of a 60-second preamble window, a 20-second action window, and a 120-second return window is used. For the remaining branches, a default combination of a 120-second preamble window, a 10-second action window, and a 60-second return window is used.

[0045] To ensure comparability of windowing results, the edge controller checks the time base consistency, the proportion of completion markers, and the record integrity markers before and after the event anchor point before windowing. If any branch has a time base mismatch marker, more than 3 consecutive missing time slices, or a completion segment accounts for more than 20% of the total number of slices in the window within the candidate window range, the corresponding time slice of that branch is marked as a low-confidence segment. However, it is still retained in the same box synchronous event window group along with the other branches, and a low-confidence marker is written in the field of the window. The low-confidence segment is not used as the basis for background branch selection in the next step, but is only used to maintain the integrity of branch timing and manual review.

[0046] Synchronous capture involves retrieving the leading window from the circular buffer and continuously writing the action window and return window from the real-time acquisition stream for all branches at the same start and end time under a unified event anchor point. The next time slice after the last slice of the leading window is used as the first slice of the action window, and the next time slice after the last slice of the action window is used as the first slice of the return window. No repeated slices are retrieved at the boundaries of each window, thus forming a synchronous event window group in the same box.

[0047] The synchronous event window group is a collection of all branch leading windows, action windows, and regression windows obtained around the same event anchor point. It is stored on-site in the form of event files. The event file includes at least the target branch box identifier, event number, all branch identifiers, start and end times of each branch and window, integrity flag, completion flag ratio, glitch replacement flag ratio, low confidence flag, triggering branch identifier, and generation time. It is written to the industrial storage medium of the edge recording module in the form of binary master file with text index file. The station host reads the summary through polling or passive upload.

[0048] To prevent the same event from being repeatedly windowed during jitter, the monitoring terminal uses the target branch box identifier, trigger branch identifier, event anchor point time, and boundary generation version number to form an idempotent key. Within the suppression interval and when the difference between event anchor points does not exceed the length of the recovery window, no new event number is created. Instead, the end time of the recovery window of the original event file is extended to the end time of the corresponding recovery window after the new trigger point, and the newly added trigger record is appended to the original event file as a sub-record. In one implementation, the suppression interval can be set to 30 seconds.

[0049] If multiple branches meet the abnormal response judgment condition for the first time in the same time slice, the degree of continuous deviation of the zero-sequence response is compared first. If the degree of continuous deviation of the zero-sequence response is the same, the number of consecutive slices of current response exceeding the limit is compared. If the number of consecutive slices of current response exceeding the limit is the same, the stability of the pre-trigger window before triggering is compared. If the stability of the pre-trigger window is still the same, the triggering branch identifier is determined according to the branch identifier order, but the simultaneous windowing result of all branches is not changed.

[0050] The entire window capture and group creation process is completed locally on the monitoring terminal. The maximum delay from the confirmation of the first abnormality of a single event to the end of the event file writing to disk is preferably no more than 500 milliseconds. When the number of concurrent events reaches the maximum allowed by the monitoring terminal, they are queued and executed in the order of the event anchor points. The original sampling stream is still retained during the waiting period of the later events. If the event file writing to disk fails, the edge recording module retryes twice with the same idempotent key. If it still fails, the recording file is written with a failure code and the event summary is saved in the queue to be supplemented on the station host.

[0051] The minimum implementable set of the interface includes at least the target branch box identifier, event number, trigger branch identifier, event anchor time, window length configuration, record integrity flag, and response status field. The response status field returns five types of error codes: success, time base mismatch, insufficient cache, duplicate event, and file write failure. Among them, insufficient cache corresponds to the circular buffer not covering the entire leading window length, and duplicate event corresponds to the idempotent key already existing and within the suppression interval.

[0052] The evaluation criteria for on-site performance can be set as follows: based on the number of events, in no less than 30 samples of manually labeled events and naturally occurring events, the event anchor point deviation should not exceed 3 time slices, the complete capture rate should not be less than 95%, the duplicate windowing rate should not be higher than 2%, and the false rejection rate of low confidence segments based on the number of manually labeled window segments should not be higher than 3%.

[0053] In a preferred embodiment, the target cable branch box contains 4 outgoing branches with a sampling rhythm of 20 milliseconds. Branch B satisfies the current response sequence exceeding the steady-state offset boundary and the zero-sequence response sequence exceeding the zero-sequence response stable interval for three consecutive time slices at 14:23:16.480. The monitoring terminal records 14:23:16.440 as the event anchor point, and synchronously captures the 4 branches according to the pre-window of 60 seconds, the action window of 20 seconds, and the return window of 120 seconds, generating a synchronous event window group of the same box containing 4 groups of branch window segments, and completing the disking within 220 milliseconds.

[0054] Alternatively, the leading window can be stored in a circular buffer for a long time without relying on it. Instead, the monitoring terminal can continuously write the sampling stream of the most recent time period into an overlay sequential file and retrieve it by looking back at the event anchor point. As long as the time base, idempotent key rules, window retrieval order, boundary splicing rules, and event file field caliber are consistent, it is considered an equivalent implementation. The entire process is limited to the scope of power distribution operation monitoring and fault location assistance. It does not change the protection settings of primary equipment, does not replace the statutory protection action chain, and does not cross the maintenance isolation permission boundary.

[0055] S3. Based on the synchronous event window group of the same box, the main abnormal branches are screened out, and the background branch set is selected from the remaining branches. The common disturbance components of the background branches at each time are aggregated to generate the background disturbance trajectory of the same box. The specific implementation is as follows: When filtering out the main abnormal branches based on the synchronous event window group in the same box, the monitoring terminal calls the event file generated in the previous step. In the edge controller, the leading window, action window and return window segments of all branches are read in the order of event number. Combined with the steady-state offset boundary, fluctuation rhythm boundary, return duration boundary and zero-sequence response stability interval in the reference set of branches in the same box, the first time of the branch's first boundary crossing, the number of consecutive boundary crossing segments, the number of consecutive zero-sequence response deviation segments, the number of persistent segments of the non-return boundary in the return window and the stability of the leading window are compared item by item. The main abnormal branches are determined according to a fixed exclusive order, and then the background branch set is screened from the remaining branches.

[0056] The main abnormal branch is determined in the following exclusive order: first, compare the first time of the boundary crossing within the action window; if the first time of the boundary crossing is the same, then compare the number of consecutive boundary crossings; if the number of consecutive boundary crossings is the same, then compare the number of consecutive deviations from the zero-sequence response; if they are still the same, then compare the number of consecutive non-return boundary fragments within the recovery window; if they are still the same, then determine the branch according to the branch identifier order.

[0057] The first boundary crossing time is taken from the first valid time slice when the branch crosses the steady-state offset boundary, fluctuation rhythm boundary, or zero-sequence response stable interval for the first three consecutive time slices within the action window; the number of consecutive boundary crossing slices is accumulated according to the time slices within the action window that meet the abnormal response conditions and are not interrupted by more than two normal time slices in between; the number of continuous slices that do not return to the boundary within the return window is obtained by counting the number of time slices in which the branch current response sequence is still outside the steady-state offset boundary from the first slice of the return window; the stability of the leading window is taken as the lower of the proportion of time slices in the leading window where the current response sequence is within the steady-state offset boundary and the proportion of stable segments in the load fluctuation sequence; all the above quantities are obtained by the edge controller by comparing slices one by one in time stamp order under a unified time base, without relying on external manual selection.

[0058] When filtering out the main abnormal branch, its original fragment is not physically deleted. Instead, the branch is written into the main abnormal branch field in the event file and excluded from the subsequent background branch candidate list to maintain the integrity of the event file. When selecting the remaining branches to form the background branch set, the edge controller continues to filter out the following branches from the remaining branches other than the main abnormal branch: branches corresponding to low confidence fragments, branches whose number of consecutive out-of-bounds fragments in the action window reaches or exceeds 50% of the number of consecutive out-of-bounds fragments in the main abnormal branch and is rounded up to the nearest integer, branches whose number of consecutive deviations in zero-order response reaches or exceeds 50% of the number of consecutive deviations in zero-order response in the main abnormal branch and is rounded up to the nearest integer, branches whose stability in the leading window is less than 90%, and branches whose number of consecutive fragments at the non-return boundary in the return window exceeds the upper limit of their respective return duration boundaries. The remaining branches are retained as the background branch set.

[0059] The background branch set must contain at least 2 branches; when the number of background branches after filtering is less than 2, the edge controller records the background branch insufficiency code and marks the event as a background missing event. It does not generate a formal in-box background disturbance trajectory, but only retains the candidate background record for the next step to enter the degradation path.

[0060] To ensure the comparability of the background branch set, the edge controller checks the time base consistency, completion mark ratio, glitch replacement mark ratio, and integrity mark of each window segment of the candidate branch before filtering. If any candidate branch has more than 3 consecutive missing time slices, completion segments account for more than 20% of the total number of window segments, glitch replacement segments account for more than 10% of the total number of window segments, or a time base mismatch mark in the action window or regression window, the branch will not be included in the background branch set, but the original record will still be retained in the event file and written to the removal reason field.

[0061] When aggregating the common disturbance components of the background branches at each time step, the edge controller aligns all branches in the background branch set with the same time slice in the leading window, action window, and return window according to the event anchor point. It reads the current response sample value and zero-sequence response sample value of each background branch at each time step. First, it removes the records with padding marks, glitch replacement marks, or low confidence marks at that time step. Then, it sorts the remaining records by value and takes the median value as the initial intermediate level at that time step.

[0062] Then, the deviation of each background branch record relative to the initial median level is calculated. Records with deviations exceeding twice the median value of all deviations at that time are removed. The remaining records after removal are then re-sorted by value and the median value is taken. The recalculated median value is recorded as the common current disturbance value and the common zero-sequence disturbance value at that time.

[0063] The common disturbance component is the synchronous change component presented by multiple background branches in the same time slice. In the field, it corresponds to the common fluctuations of each branch caused by the fluctuation of the upstream power supply, the voltage swing of the bus side, the common disturbance of the enclosure environment, or the synchronous change of the normal load in the same enclosure. It is obtained by the horizontal comparison of the sampled values ​​of multiple background branches in the same time slice.

[0064] When there are fewer than 2 valid background branch records remaining after elimination at a certain moment, the common disturbance component at that moment is maintained for 1 time slice according to the common disturbance value of the most recent valid moment. If it is maintained for more than 3 consecutive time slices, the segment is recorded as a common disturbance missing segment and written into the missing segment mark. The common disturbance missing segment does not participate in the difference trajectory deduction of the branch to be judged in the next step, but is only used to maintain the continuity of the time axis and for manual review.

[0065] When generating the background disturbance trajectory in the same box, the common current disturbance value and common zero-sequence disturbance value obtained at each time are arranged continuously in the order of timestamps to form the background disturbance trajectory in the same box, which consists of the common current disturbance sub-trajectory and the common zero-sequence disturbance sub-trajectory, corresponds to the event anchor point and covers the leading window, the action window and the return window. This trajectory, together with the event number, the set of background branch identifiers, the number of valid background branches at each time, the missing measurement mark, the version number and the generation time, are written into the background trajectory file. The background trajectory file is stored in the industrial storage medium of the edge recording module in the form of a binary master file with a text index file, and the summary is synchronously sent to the station host for the next step to call.

[0066] Background trajectory generation tasks with the same event number, the same boundary generation version number, and the same set of background branch identifiers are controlled by idempotent keys. Repeated calls will directly return the existing background trajectory file and will not generate it again.

[0067] The entire process of screening, selection, aggregation, and generation is completed locally by the edge controller on the monitoring terminal. The upper limit of the delay from the start of the event file call to the end of the background trajectory file writing to disk is preferably no more than 800 milliseconds. When the concurrent events reach the upper limit, they are queued and executed in the order of the event anchor points. During the waiting period of the later events, the existing event files are not overwritten. If the background trajectory file writing to disk fails, the edge recording module retryes twice with the same idempotent key. If it still fails, the background trajectory writing failure code is recorded and the event number is written to the queue to be supplemented.

[0068] The minimum implementable set of the interface should include at least the event number, target branch box identifier, main abnormal branch identifier, background branch identifier set, trajectory coverage period, version number, and response status field.

[0069] The response status field returns five types of error codes: success, insufficient background branches, time base mismatch, missing trajectory, and file write failure. Among them, insufficient background branches correspond to fewer than 2 background branches after filtering, and missing trajectory corresponds to more than 3 consecutive time slices of missing common disturbances.

[0070] The verification criteria for on-site performance can be set as follows: based on a sample of no less than 30 manually labeled events and naturally occurring events, the consistency rate of main abnormal branch identification should be no less than 95%, and the false inclusion rate of background branch set should be no higher than 5%.

[0071] In a preferred embodiment, the target cable branch box contains four outgoing branches. After the event anchor point, branch B first crosses the steady-state offset boundary in three consecutive time slices. The number of consecutive boundary-crossing segments within the action window is 286, the number of consecutive deviation segments in the zero-sequence response is 251, and the number of segments that do not return to the boundary within the recovery window is 318. It is identified as the main abnormal branch. Among the remaining branches, branches A and C meet the background branch conditions, while branch D is removed because the supplementary segment accounts for 22% of the total number of segments in the action window. The edge controller aggregates the common disturbance components of branches A and C piece by piece according to a 20-millisecond sampling rhythm, generates a background disturbance trajectory of the same box covering a 60-second pre-window, a 20-second action window, and a 120-second recovery window, and completes the disk placement within 460 milliseconds.

[0072] Alternatively, the common disturbance component can also be generated based on the truncated mean of the effective background branch records at each time. As long as the main abnormal branch screening rules, background branch screening rules, missing measurement judgment rules, idempotent key rules, and background trajectory file field caliber are consistent, it is considered an equivalent implementation. The entire process is limited to the scope of power distribution operation monitoring and fault location assistance, without changing the protection settings of primary equipment, without replacing the statutory protection action chain, and without exceeding the maintenance isolation permission boundary.

[0073] S4. Subtract the background disturbance trajectory from the same box from the event response sequence of each branch to be judged, and generate the difference response trajectory of the corresponding branch. The specific implementation is as follows: When subtracting the background disturbance trajectory from the same box from the event response sequence of each branch to be judged, the monitoring terminal calls the event file and background trajectory file generated in the previous step. A one-to-one correspondence is established within the edge controller according to the event number, target branch box identifier, and version number. The event response sequence of each branch to be judged within the leading window, action window, and return window is aligned with the background disturbance trajectory of the same box using the same time slice, and then subtracted slice by slice to generate the difference response trajectory of the corresponding branch. A branch to be judged is a branch that has not been identified as a main anomaly branch, has not been included in the background branch set, and has not been completely removed from the event file. If there are no branches to be judged, this step ends and a message indicating "no branches to be judged" is written. Road marker; the time slice with the same name is the time slice with the same timestamp and window type as the event response sequence and the background disturbance trajectory in the same box; the event response sequence is the current response sampling value sequence and zero-sequence response sampling value sequence of the branch to be judged under the current event number, covering the leading window, the action window and the return window, with units of amperes and amperes respectively, and the sampling rhythm follows the upstream 20 milliseconds or 40 milliseconds, without changing the sampling interval in this step; the background disturbance trajectory in the same box is a continuous time-series trajectory corresponding to the event number, consisting of a common current disturbance sub-trajectory and a common zero-sequence disturbance sub-trajectory, with the same coverage period as the event response sequence, and the time base version number is consistent with the event file.

[0074] To ensure comparability of deduction results, the edge controller checks the continuity of timestamps, record integrity markers, supplementation markers, glitch replacement markers, and missing measurement markers of the event response sequence of the branch to be judged and the background disturbance trajectory of the same box before deducting each piece. When any branch to be judged has time base mismatch in the corresponding time slice, more than 3 consecutive time slices are missing, the supplementation segment accounts for more than 20% of the total number of slices in the corresponding window, or the background disturbance trajectory of the same box has a common disturbance missing measurement segment in the corresponding time period, the record segment is not directly deleted. Instead, a deduction restriction marker is written into the result trajectory. The time slices with continuous deduction restriction markers, missing measurement supplementation markers, or long missing measurement degradation markers are used as restricted segments, and the remaining continuous time slices are used as normal segments. The difference response trajectory is generated by splicing the restricted segments and normal segments in the order of restricted segments to maintain the continuity of the entire time axis.

[0075] During the deduction process, the edge controller reads the current response sample value and zero-sequence response sample value of a branch to be judged in the current time slice according to the timestamp sequence. Then, it reads the common current disturbance value and common zero-sequence disturbance value of the same time slice. The common current disturbance value is subtracted from the current response sample value, and the common zero-sequence disturbance value is subtracted from the zero-sequence response sample value to obtain the current difference value and zero-sequence difference value of that time slice. When the time slice has a completion mark, glitch replacement mark, or missing measurement mark, the deduction does not stop, but the deduction result is retained and synchronously written to the source mark for the next step to determine whether to include it in the valid statistics. The difference response trajectory consists only of the current difference sub-trajectory and the zero-sequence difference sub-trajectory. The current difference value is continuously arranged according to the timestamp to form the current difference sub-trajectory, and the zero-sequence difference value is continuously arranged according to the timestamp to form the zero-sequence difference sub-trajectory. The two together constitute the difference response trajectory of the corresponding branch.

[0076] To prevent non-physical jumps in the deduction results due to local missing measurements in the background trajectory, when there is a common disturbance missing measurement segment in a continuous section of the background disturbance trajectory within the same box, and the length does not exceed 3 time slots, the edge controller fills in the missing measurement segment with the progressive value between adjacent valid common disturbance values. Specifically, when there is a missing measurement for 2 time slots, the 1 / 3 and 2 / 3 position values ​​between the consecutive valid values ​​are taken respectively; when there is a missing measurement for 3 time slots, the 1 / 4, 2 / 4, and 3 / 4 position values ​​are taken respectively, and a missing measurement filling mark is written into the difference response trajectory; when the common disturbance missing measurement segment... When the length exceeds 3 time slices, the difference response trajectory segment retains only the deduction result of the original event response sample value of the branch to be judged and the most recent effective common disturbance value, and writes it into the long missing measurement degradation mark. This long missing measurement degradation mark does not participate in the priority value selection in the next step, but is only used to maintain the continuity of the preceding and following time sequences and manual review. The missing measurement of the trajectory corresponds to the missing measurement segment of the common disturbance that exceeds 3 consecutive time slices and covers the key section of the action window. The key section of the action window is the action window segment corresponding to the first 10 seconds after the event anchor point. If the total length of the action window is less than 10 seconds, the entire action window is taken.

[0077] To prevent individual branches from being judged from being raised or lowered as a whole after deduction due to baseline drift of the leading window, the edge controller checks the center offset of the current difference sub-trajectory and the zero-sequence difference sub-trajectory in the leading window of the branch to be judged after completing the full window piece-by-piece deduction. The center offset is obtained by the median value after sorting all valid difference values ​​in the leading window. When the center offset of the leading window of the current difference sub-trajectory or the zero-sequence difference sub-trajectory exceeds 20% of the width of the corresponding steady-state offset boundary, the center offset of the leading window is used as the baseline correction value of the sub-trajectory and one whole-track translation correction is performed. The width of the steady-state offset boundary is the difference between the upper and lower limits of the steady-state offset boundary of the branch. The correction starting point is the first piece of the leading window of the difference response trajectory. After correction, the correction value, correction starting point and correction version number are written into the trajectory header field.

[0078] After the deduction is completed, the edge controller generates a difference trajectory record for each branch to be judged. The difference trajectory record includes at least the event number, target branch box identifier, branch to be judged identifier, trajectory coverage period, current difference sub-trajectory, zero-sequence difference sub-trajectory, deduction restriction mark, shortage measurement supplementation mark, long shortage measurement degradation mark, baseline correction value, version number, and generation time. It is written to the industrial storage medium of the edge recording module in the form of binary master file with text index file, and the summary is synchronously sent to the station host for the next step. The difference trajectory generation task corresponding to the same event number, the same branch to be judged identifier, and the same background trajectory version number is controlled by idempotent key. Repeated calls directly return the existing difference trajectory record and do not generate it again. If the version number of the event file is inconsistent with the version number of the background trajectory file, the version inconsistency code is returned first. If the versions are consistent but the branch to be judged identifier has already appeared in the main abnormal branch field or the background branch identifier set field, the branch type conflict code is returned.

[0079] The entire deduction and recording process is completed locally on the monitoring terminal by the edge controller. The maximum delay from the start of calling the event file and background trajectory file to the end of the differential trajectory recording on disk is preferably no more than 800 milliseconds. When the concurrent events reach the limit, they are queued and executed in the order of event anchor points. If the differential trajectory recording fails to write to disk, the edge recording module retryes twice using the same idempotent key. If it still fails, the differential trajectory writing failure code is recorded and the event number and the branch identifier to be judged are written to the queue to be supplemented. The minimum implementable set of the interface includes at least the event number, target branch box identifier, branch identifier to be judged, background trajectory version number, trajectory coverage period, and response status field. The response status field returns five types of error codes: success, version inconsistency, branch type conflict, trajectory missing test, and file writing failure.

[0080] The verification criteria for on-site performance can be set as follows: based on a sample of no less than 30 manually labeled events and naturally occurring events, the residual fluctuation amplitude of the difference response trajectory within the leading window shall not exceed 15% of the width of the steady-state offset boundary of the corresponding branch to be judged.

[0081] In a preferred embodiment, the target cable branch box includes four outgoing branches. Branch A is the branch to be judged, with a sampling rhythm of 20 milliseconds, a preamble window of 60 seconds, an action window of 20 seconds, and a return window of 120 seconds. The current response sampling value of branch A in a certain action window time slice is 212.4 amps, the common current disturbance value in the same time slice is 205.8 amps, and the current difference value after deduction is 6.6 amps. The zero-sequence response sampling value in the same time slice is 0.74 amps, the common zero-sequence disturbance value is 0.31 amps, and the zero-sequence difference value after deduction is 0.43 amps. After the full window deduction is completed, the center offset of the current difference sub-trajectory of branch A in the preamble window is 1.2 amps, which does not exceed the correction threshold of 20% of the corresponding steady-state offset boundary width. Without performing translation correction, the edge controller completes the differential trajectory recording and disking within 420 milliseconds; alternatively, when the current window center offset exceeds the correction threshold, the median value of the continuous effective differential values ​​at the end of the leading window can be used as the baseline correction value, where the end of the leading window takes 20 consecutive effective time slices before the end of the leading window. As long as the branch selection rules, time slice alignment rules, common disturbance missing measurement section handling rules, idempotent key rules, and differential trajectory recording field caliber are consistent, it is considered equivalent implementation; the entire process is limited to the scope of power distribution operation monitoring and fault location assistance, does not change the primary equipment protection settings, does not replace the statutory protection action chain, and does not exceed the maintenance isolation permission boundary.

[0082] S5. Based on the differential response trajectories of each branch, extract the initial partial sequence quantity, coupling deviation quantity, and regression residual quantity to form a branch fault fingerprint, and execute an exclusive decision according to the progressive rule of first-initiated change, strong deviation, and persistent residual. The specific implementation is as follows: When extracting the initial offset sequence, coupling deviation, and regression residue based on the differential response trajectory of each branch, the monitoring terminal calls the differential trajectory record generated in the previous step. In the edge controller, a one-to-one correspondence is established according to the event number, target branch box identifier, branch identifier to be judged, and version number. The current difference sub-trajectory and zero-sequence difference sub-trajectory of each branch to be judged covering the leading window, action window, and regression window are read and compared in sequence under a unified time reference. The initial offset sequence, coupling deviation, and regression residue are extracted from them to form a branch fault fingerprint. The exclusive decision is executed according to the progressive rule of first change, strong deviation, and persistent residue. The branch to be judged is the branch object that has not been identified as the main abnormal branch, has not been included in the background branch set, and has not been completely removed from the event file. The priority branch is the branch that ranks first after the exclusive decision of each branch to be judged in this step. It is not directly used as the final fault branch conclusion, but as the priority confirmation object for the next step.

[0083] The zero-difference stable intervals are generated separately for the current difference sub-trajectory and the zero-sequence difference sub-trajectory, respectively. The current zero-difference stable interval and the zero-sequence zero-difference stable interval serve as unified comparison boundaries for the initial partial order quantity extraction, coupling deviation quantity extraction, regression residual quantity extraction, and subsequent observation period response backtesting, respectively. After sorting the absolute values ​​of all effective current difference values ​​within the leading window, the top 5% of samples are removed. The maximum absolute value of the remaining samples is used as the single-sided boundary value of the current, and its negative number and positive value are taken to form the current zero-difference stable interval. The zero-sequence zero-difference stable interval is generated for the absolute values ​​of all effective zero-sequence difference values ​​within the leading window according to the same rule. The starting segment of the action window is the segment corresponding to the first 20% of the action window duration after the event anchor point. When the total length of the action window is less than 10 seconds, the entire action window is taken. The main segment of the action window is the remaining action window segments excluding the starting segment of the action window.

[0084] The initial partial order value consists of the first consecutive boundary crossing time sequence field and the average overflow amount field of the first consecutive boundary crossing three effective time slices. On-site, the starting segment of the action window is first defined, and then the position of each branch to be judged that crosses the corresponding zero-difference stable interval for the first three consecutive effective time slices is found. The order of this position among all branches to be judged is recorded as the first consecutive boundary crossing time sequence. The average overflow amount is obtained by averaging the overflow amount of the three consecutive effective time slices relative to the boundary of the corresponding zero-difference stable interval. When comparing the initial partial order value, the first consecutive boundary crossing time sequence field is compared first, and then the average overflow amount field is compared. The branch with the earlier sequence and the larger the average overflow amount has the larger initial partial order value.

[0085] The coupling deviation is derived from the degree of continuous deviation of the difference response trajectory within the main segment of the action window from the zero-difference stable interval. On-site, the number of consecutive time slices in which the current difference value and zero-sequence difference value exceed their respective zero-difference stable intervals, the cumulative deviation, and the maximum continuous deviation length are recorded item by item in all effective time slices within the main segment of the action window. The cumulative deviation is the sum of the absolute values ​​of the excesses of all effective time slices within the main segment of the action window relative to the nearest boundary of the corresponding zero-difference stable interval. The current difference sub-trajectory and the zero-sequence difference sub-trajectory are calculated separately, and the larger of the two is taken as the cumulative deviation of the branch. Only time slices without long-missing measurement degradation markers and whose deviation segments are interrupted by more than two normal time slices are counted and accumulated in segments. When comparing the coupling deviation, the number of consecutive time slices is compared first, then the cumulative deviation, and then the maximum continuous deviation length. Branches with more consecutive time slices, larger cumulative deviations, and longer maximum continuous deviation lengths have larger coupling deviations.

[0086] The residual value is determined by the duration of the difference response trajectory within the regression window failing to return to the corresponding zero-difference stable interval. On-site, starting from the first piece in the regression window, the current difference sub-trajectory and the zero-sequence difference sub-trajectory are continuously checked to see if they simultaneously return to the corresponding zero-difference stable interval. When either sub-trajectory is still outside the interval, the number of residual pieces continues to be accumulated, and the average deviation of the last segment is recorded simultaneously. When the current difference sub-trajectory and the zero-sequence difference sub-trajectory simultaneously return to the corresponding zero-difference stable interval for three consecutive valid time pieces, the accumulation of the residual value is stopped. When comparing the residual value, the number of residual pieces is compared first, and then the average deviation of the last segment is compared. The branch with more residual pieces and a larger average deviation of the last segment has a larger residual value.

[0087] To ensure consistent extraction criteria, the edge controller checks the deduction restriction marker, missing measurement supplementation marker, long missing measurement degradation marker, and baseline correction value in the difference trajectory record before extraction. Segments with long missing measurement degradation markers that cover key sections of the action window do not participate in the extraction of initial partial order and coupling deviation, but only participate in the continuous registration of the reverted residual. Segments with missing measurement supplementation markers can participate in extraction, but supplementation participation markers are written into the branch fault fingerprint. Segments with deduction restriction markers are numbered separately according to the restricted segment. If the length of a single segment exceeds 20% of the total length of the action window, the fingerprint of that branch is written with a restricted calculation marker. Supplementation participation markers and restricted calculation markers are only used to record extraction conditions and do not directly change the exclusive decision priority.

[0088] When generating a branch fault fingerprint, the edge controller writes the initial partial sequence value, coupling deviation value, regression residual value, supplementary participation flag, restricted calculation flag, event number, branch identifier, version number, and generation time of each branch to be judged into the fingerprint record.

[0089] When executing the exclusive decision according to the progressive rule of first change, strong deviation, and residual persistence, the initial bias order of each branch to be decided is compared first. If the first consecutive out-of-bounds time order field of the initial bias order is the same or the difference does not exceed 5% of the total number of time slices in the leading window and is rounded up to the nearest integer, the coupling deviation is compared next. If the number of persistent slices of the coupling deviation is the same or the cumulative out-of-bounds difference does not exceed 10% of the median of the cumulative out-of-bounds set of all branches to be decided in the current event, the regression residual is compared next. The branch with the larger regression residual is given priority. If they are still the same, the exclusive decision result is determined according to the order of the branches to be decided. After the exclusive decision result is determined, the edge controller writes the priority branch identifier, comparison order, difference range and decision version number into the event-level decision record and synchronizes the summary to the station host for the next step.

[0090] The entire extraction, formation, and adjudication process is completed locally on the monitoring terminal by the edge controller. The upper limit of the delay from the start of calling the difference trajectory record to the end of the adjudication record being written to disk is preferably no more than 900 milliseconds. When the concurrent events reach the upper limit, they are queued and executed in the order of the event anchor points. The fingerprint generation task corresponding to the same event number, the same branch identifier to be judged, and the same difference trajectory version number is controlled by idempotent keys. Repeated calls directly return the existing fingerprint record. When the number of effective time slices for any branch to be judged to participate in the extraction of the initial partial order amount and coupling deviation amount is less than 60% of the total number of slices in the branch's action window, the branch does not participate in this exclusive adjudication. When all branches to be judged do not participate, the adjudication is stopped and an insufficient effective slice code is returned. If the difference trajectory version numbers are inconsistent, the version inconsistency code is returned first. If the versions are consistent but all branches to be judged have restricted calculation marks, the overall degradation code is returned. If the adjudication record writing to disk fails, the file writing failure code is returned.

[0091] The minimum implementable set of the interface includes at least the event number, the branch identifier to be judged, the difference trajectory version number, the fingerprint record version number, and the response status field. The response status field returns five types of error codes: success, version inconsistency, insufficient valid fragments, overall degradation, and file write failure.

[0092] The inspection criteria for on-site performance can be set as follows: based on a sample of no less than 30 manually labeled events and naturally occurring events, the consistency rate between the priority branch obtained by the exclusion decision and the manual review result is no less than 95%, the initial partial order quantity extraction deviation does not exceed 3 time slices, and the initial partial order quantity extraction deviation is calculated as the difference between the time slice corresponding to the first consecutive cross-boundary time sequence extracted automatically and the time slice labeled by manual review.

[0093] In a preferred embodiment, the target cable branch box includes four outgoing branches. Branch A and Branch C are the branches to be judged. The sampling rhythm is 20 milliseconds. Branch A crosses the current zero difference stable interval for the first time in the 12th segment of the action window, with an average crossover amount of 2.8 amps. Branch C crosses the current for the first time in the 19th segment, with an average crossover amount of 1.6 amps. Therefore, the initial deviation of Branch A takes priority. After continuing to count the main segment of the action window, Branch A crosses the current for 214 segments, with a cumulative crossover amount of 486 amps and a maximum continuous crossover length of 173 segments. Branch C crosses the current for 137 segments, with a cumulative crossover amount of 251 amps and a maximum continuous crossover length of 96 segments. Furthermore, Branch A has 127 segments remaining in the return window, and the average deviation of the last segment is... The quantity is 1.9 amps, with 42 residual pieces in branch C and an average deviation of 0.8 amps in the final segment. The edge controller ultimately determines branch A as the exclusive priority branch and completes the fingerprint recording and decision recording on disk within 510 milliseconds. Alternatively, the initial deviation quantity can also be determined by the first time slice priority and the average deviation quantity of the next 10 effective time slices. As long as the zero-difference stable interval determination rule, the action window and return window segmentation rule, the progressive comparison order, the idempotent key rule, and the fingerprint record field caliber are consistent, it is considered an equivalent implementation. The entire process is limited to the scope of power distribution operation monitoring and fault location assistance, without changing the primary equipment protection settings, without replacing the statutory protection action chain, and without exceeding the maintenance isolation permission boundary.

[0094] S6. Branches that meet the exclusive decision criteria are identified as faulty branches. During subsequent observation periods, a response verification is performed on these branches, and the reference set of branches within the same box is updated based on the verification results. Specifically, this is implemented as follows: When a branch that meets the exclusive decision criteria is identified as a faulty branch, the monitoring terminal calls the event-level decision record, branch fault fingerprint record, event file, and reference set of branches in the same box generated in the previous step. In the edge controller, a one-to-one correspondence is established according to the event number, target branch box identifier, priority branch identifier, and version number. The priority branch in the exclusive decision result is taken as the candidate object of the faulty branch. Then, the faulty branch is determined by whether the starting partial order, coupling deviation, and regression residual of the branch in the same event simultaneously meet the confirmation threshold under the current event version. The confirmation threshold is that the branch's starting partial order is ranked first among all branches to be judged, the coupling deviation is not lower than the median value of the set of coupling deviations of all branches to be judged, the regression residual is not lower than the median value of the set of regression residuals of all branches to be judged, and the branch does not have a version inconsistency code or an overall downgrade code. The faulty branch is the branch object with the highest priority after the exclusive decision of the current event and confirmed by the response back verification in the subsequent observation cycle. The single ranking result does not directly replace the final confirmation.

[0095] The subsequent observation period is initiated immediately by the edge controller after the candidate faulty branch is determined. The subsequent observation period is the time interval after the end of the recovery window, during which continuous sampling and status verification are performed on the branch. The current response sequence, zero-sequence response sequence, and load fluctuation sequence are continuously collected from the monitoring terminal on the branch box side in the field, and are used in conjunction with the current difference sub-trajectory, zero-sequence difference sub-trajectory, and zero-difference stable interval corresponding to the branch. The length of the subsequent observation period is determined according to the branch type and the aforementioned recovery residual amount. Branches with a recovery residual amount of more than 30% of the total number of recovery window segments adopt a 300-second observation period, while branches with a recovery residual amount below this threshold adopt a 120-second observation period. The default value for the initial configuration is 120 seconds.

[0096] When performing response verification on this branch, the edge controller reads the current response sampling value and zero-sequence response sampling value of this branch piece by piece according to a unified time base in the subsequent observation period. It then forms the current difference value and zero-sequence difference value for the corresponding period according to the same time slice rule of the previous step, and checks whether the branch continues to exhibit residual deviation characteristics consistent with the current ruling. The response verification is completed on-site using three continuous criteria. The first criterion is the number of segments that continuously exceed the corresponding zero-difference stable interval in the subsequent observation period. Continuously exceeding segments are constituted by valid time slices that are continuously outside the corresponding zero-difference stable interval in the subsequent observation period. If the segment is interrupted by two or more consecutive valid normal time slices, it is recounted as the next segment. The second criterion is the number of valid time slices. The percentage of films located outside the zero-difference stable interval in the inter-film data is considered. The third item is the average deviation of 20 consecutive valid time films at the end of the observation period. The retest is considered successful when the number of consecutive segments exceeding the interval is not less than 20%, the percentage of films located outside the interval is not less than 20%, and the average deviation at the end of the period does not fall back to within 50% of the single-sided boundary value of the corresponding zero-difference stable interval. The current difference sub-trajectory is calculated based on the absolute value of the positive boundary value of the current zero-difference stable interval, and the zero-sequence difference sub-trajectory is calculated based on the absolute value of the positive boundary value of the zero-sequence zero-difference stable interval. If any sub-trajectory does not meet the requirements, it is considered as not falling back. The retest is considered unsuccessful when two of the three items are not met. The retest is considered insufficient when the number of valid time films is less than 60% of the total number of films in the subsequent observation period.

[0097] To ensure comparability of the backtest results, the edge controller checks the completion markers, glitch replacement markers, long-missing-measurement-degradation markers, and time-base mismatch markers in the subsequent observation period of the branch before performing the response backtest. Segments with more than 3 consecutive missing time slices are not included in the backtest statistics; only the time axis position is retained and a backtest restricted marker is written. Segments with missing-measurement-filling markers can be included in the slice percentage statistics, but are not included in the end-segment average deviation statistics. When a time-base mismatch marker exists at the end of a subsequent observation period, the backtest is not directly determined to be a failure. Instead, a time-base restricted marker is written and the compensation segment is extended by 1. The compensation segment is fixed at 20 seconds, and the same event is only allowed to be extended once.

[0098] When updating the reference set for branches within the same box based on the backtest results, the edge controller distinguishes between three types of results: pass, fail, and insufficient. If the backtest is pass, the leading window, action window, regression window, and subsequent observation period involved in the current event for that branch are all registered as abnormal sample segments and excluded from the normal operation sample pool of the reference set for branches within the same box. These segments do not participate in the subsequent reconstruction of steady-state offset boundaries, fluctuation rhythm boundaries, and regression duration boundaries. If the backtest is fail, the latter half of the leading window in the current event for that branch, as well as segments that have returned to the zero-difference stable interval and have maintained this for more than 10 consecutive effective time slices within the subsequent observation period, are registered as candidate sample segments for re-inclusion. These segments are written into the queue of samples to be updated without changing the current effective version of the reference set master file, for inclusion in the reference set for branches within the same box during the next offline or timed resetting. If the backtest is insufficient, the current reference set master file for branches within the same box is not updated; only insufficient backtest flags and pending review flags are registered and retained until the next joint review when an event triggered by the same target branch box, the same branch, and with consistent triggering criteria fields arrives.

[0099] To avoid a single event directly eroding the existing reference set, updating the reference set of branches within the same box is performed using a dual-file approach. The currently effective version remains read-only, while newly generated incremental updates are written to the reference update file. The reference update file includes at least the event number, target branch box identifier, faulty branch identifier, verification result, abnormal sample segment, re-inclusion candidate sample segment, pending verification mark, update version number, generation time, and original record summary fingerprint. It is stored on the industrial storage medium of the edge recording module using a binary master file with a text index file, and the summary is synchronized to the station host for use in the next cycle of reference set reorganization. Verification and update tasks corresponding to the same event number, the same faulty branch identifier, and the same verification version number are controlled using idempotent keys. Repeated calls directly return the existing verification record and reference update file, without repeated execution.

[0100] The entire determination, verification, and update process is completed locally by the edge controller on the monitoring terminal. The maximum delay from the start of calling the adjudication record to the end of writing the reference update file to disk is preferably no more than 1200 milliseconds. When the concurrent events reach the maximum limit, they are queued and executed in the order of the event anchor points. If the reference update file fails to write to disk, the edge recording module retryes twice using the same idempotent key. If it still fails, the update write failure code is recorded and the event number and fault branch identifier are written to the queue to be filled.

[0101] The minimum implementable set of the interface includes at least the event number, target branch box identifier, fault branch identifier, verification version number, reference update version number, and response status field. The response status field returns five types of error codes: success, insufficient verification, version inconsistency, overall degradation, and file write failure. Among them, the version inconsistency code takes precedence over the overall degradation code. Although the candidate branch corresponding to the overall degradation is ranked first in the decision, it cannot enter the confirmation because it has an overall degradation code.

[0102] The inspection criteria for on-site performance can be set as follows: based on a sample of no less than 30 manually labeled events and naturally occurring events, the consistency rate between the final confirmed result of the faulty branch and the manual review result should be no less than 95%, and the proportion of samples that are mistakenly collected as normal samples after passing the retest should be no more than 3%.

[0103] In a preferred embodiment, the target cable branch box includes four outgoing branches. Branch A was identified as the priority branch in the previous stage. The subsequent observation period is 120 seconds, the sampling rhythm is 20 milliseconds, the number of consecutive segments exceeding the current threshold is 3, and the proportion of segments outside the zero-difference stable interval is 28%. The average deviation of the last 20 consecutive effective time segments at the end of the observation period is 1.7 amps, which does not fall back to within 50% of the single-sided boundary value of the current zero-difference stable interval. Therefore, the retest is passed, and the edge controller sets the leading window, action window, return window, and subsequent observation window for the event corresponding to branch A. All periods are registered as abnormal sample segments, and the reference update file is written to disk within 640 milliseconds. Alternatively, the response verification can also be divided into the first half and the second half of the subsequent observation period for separate statistics. As long as the fault branch determination rules, subsequent observation period setting rules, verification pass and fail judgment rules, incremental update caliber and idempotent key rules are consistent, it is considered equivalent implementation. The entire process is limited to the scope of power distribution operation monitoring and fault location assistance, does not change the protection settings of primary equipment, does not replace the statutory protection action chain, and does not exceed the maintenance isolation permission boundary.

[0104] In the operating scenario shown in this embodiment: a 10 kV target cable branch box in an industrial park is equipped with 4 outgoing branches, of which branch A supplies the CNC machining workshop, branch B supplies the surface treatment workshop, branch C supplies the air compressor station, and branch D supplies office and lighting loads.

[0105] A field acquisition unit is installed inside the target cable branch box. The field acquisition unit is fixedly connected to the monitoring terminal on the side of the branch box. The monitoring terminal has a built-in edge controller and edge recording module. Current transformers are installed at the outgoing ends of each branch. Zero-sequence acquisition components are installed on the outside of the three-phase cable of each branch. The monitoring terminal synchronously acquires the current response sequence, zero-sequence response sequence and load fluctuation sequence of the four branches according to a unified time reference. Among them, branches A, B and C adopt a 20-millisecond sampling rhythm because the number of current changes within 30 minutes accounts for more than 5% of the total number of sampling pieces. Branch D adopts a 40-millisecond sampling rhythm. However, in the initial collection phase, all branches uniformly adopt a 20-millisecond sampling rhythm. After 72 hours of continuous collection, the edge controller generates the steady-state offset boundary, zero-sequence response stable interval, fluctuation rhythm boundary and recovery time boundary of each branch based on the effective samples during normal operation, establishes the reference set of branches in the same box and writes it into the reference file.

[0106] Subsequently, during normal operation, the monitoring terminal continuously compares the real-time sampled values ​​with the reference set of branches in the same box piece by piece. When at 14:23:16.480 on a certain day, branch B simultaneously satisfies the current response sequence exceeding the steady-state offset boundary and the zero-sequence response sequence exceeding the zero-sequence response stable range for three consecutive time slices, the edge controller confirms that branch B has an abnormal response and records 14:23:16.440 as the event anchor point. Around this event anchor point, the leading window, action window and return window are synchronously captured for the four branches to form a synchronous event window group in the same box and generate an event file.

[0107] Next, the edge controller compares the four branches in the same window based on the synchronous event window group in the same box. It finds that branch B has the earliest first boundary crossing time in the action window, the most consecutive boundary crossings, the largest number of consecutive zero-sequence response deviations, and the longest number of consecutive non-return boundary fragments in the return window. Therefore, branch B is identified as the main abnormal branch. Then, low-confidence fragment branches and branches with anomalies close to the main abnormal branch are screened out from the other branches. Branch A and branch C are retained to form the background branch set. The current response sampling value and zero-sequence response sampling value of branch A and branch C at each time are aligned according to the event anchor point. The common current disturbance value and common zero-sequence disturbance value are extracted time by time to form the same box background disturbance trajectory covering the leading window, action window and return window.

[0108] Then, the edge controller calls the event file and background trajectory file to perform time-slice subtraction on the branches to be judged. For example, in a certain action window time slice, the current response sampling value of branch A is 212.4 A, the corresponding common current disturbance value is 205.8 A, and the current difference value after subtraction is 6.6 A. In the same time slice, the zero-sequence response sampling value is 0.74 A, the corresponding common zero-sequence disturbance value is 0.31 A, and the zero-sequence difference value after subtraction is 0.43 A. This forms the current difference sub-trajectory and zero-sequence difference sub-trajectory corresponding to branch A. The same subtraction is performed on the other branches to be judged to generate the difference response trajectory of each branch.

[0109] Next, the edge controller extracts the initial bias quantity in the initial segment of the action window, the coupling deviation quantity in the main segment of the action window, and the regression residue quantity in the regression window based on the differential response trajectory of each branch to be judged, forming a branch fault fingerprint, and executes exclusive decision according to the progressive rule of first change, strong deviation, and persistent residue. For example, branch A crosses the current zero difference stable interval for the first three consecutive effective time segments in the 12th segment of the action window, with an average crossover amount of 2.8 amps. Branch C crosses the interval for the first time in the 19th segment, with an average crossover amount of 1.6 amps. In addition, branch A crosses the interval 214 segments continuously in the main segment of the action window, with a cumulative crossover amount of 486 amps and a maximum continuous crossover length of 173 segments. It has 127 residual segments in the regression window and an average deviation amount of 1.9 amps in the last segment, all of which are higher than branch C. Therefore, branch A ranks first in the exclusive decision and is written into the event-level decision record as a priority branch.

[0110] Finally, the edge controller does not directly regard the sorting result of branch A as the final faulty branch. Instead, it immediately starts a 120-second follow-up observation period after the reset window ends to perform a response verification on branch A. During this observation period, it is found that the number of consecutive segments that exceed the current range is 3, and the proportion of segments outside the zero-difference stable range is 28%. The average deviation of the last 20 consecutive effective time segments at the end of the observation period is 1.7 amps, which has not fallen back to within 50% of the single-sided boundary value of the current zero-difference stable range. Therefore, the verification is deemed successful, and branch A is identified as the faulty branch of this event. Based on this, the edge controller registers all the leading windows, action windows, reset windows, and follow-up observation periods related to the event involving branch A as abnormal sample segments and excludes them from the normal operation sample pool of the reference set of branches in the same box. At the same time, it generates a reference update file and writes it to the edge record module for reference set reorganization in the next cycle.

[0111] Through the above continuous operation process, the target cable branch box achieves a complete closed loop from normal sample set construction, abnormal triggering, simultaneous windowing in the same box, background disturbance stripping, differential response extraction, fault fingerprint adjudication to subsequent observation cycle backtesting and reference set update, without changing the protection settings of the primary equipment or replacing the statutory protection action chain. This enables stable identification of specific fault branches in the scenario of multiple branches in the same box experiencing common disturbances.

[0112] All calculations involved in the embodiments are dimensionless numerical calculations, and the preset parameters and thresholds in the calculations are set by those skilled in the art according to the actual situation.

[0113] It should be noted that this invention can be deployed on the device itself to realize embedded applications, or it can run on a PC or other terminal with a user interface, thereby meeting various hardware environments and usage requirements.

[0114] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wireless or wired transmission; wired transmission methods include optical fiber, twisted pair, coaxial cable, etc.; wireless transmission includes infrared, microwave, etc. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center containing one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0115] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and modules described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0116] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or modules may be electrical, mechanical, or other forms.

[0117] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0118] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0119] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0120] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0121] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for rapid location of a branch fault in a cable branch box, characterized in that, include: S1. Collect the current response sequence, zero-sequence response sequence and load fluctuation sequence of each branch of the target cable branch box during normal operation, extract the steady-state offset boundary, zero-sequence response stable interval, fluctuation rhythm boundary and recovery time boundary of each branch, and establish a reference set of branches in the same box. S2. When any branch path experiences an abnormal response, the first abnormal moment is used as the event anchor point. The leading window, action window and reset window of all branches are simultaneously captured to form a synchronous event window group in the same box. S3. Based on the synchronous event window group of the same box, the main abnormal branch is screened out, and the background branch set is screened from the remaining branches. The common disturbance components of the background branches at each time are aggregated to generate the background disturbance trajectory of the same box. S4. Subtract the background disturbance trajectory in the same box from the event response sequence of each branch to be judged to generate the difference response trajectory of the corresponding branch; S5. Extract the initial partial sequence quantity, coupling deviation quantity and regression residual quantity based on the different response trajectories of each branch to form the branch fault fingerprint, and execute the exclusive decision according to the progressive rule of first change, strong deviation and residual persistence. S6. Identify the branch that meets the exclusive decision condition as the faulty branch, perform a response verification on the branch in the subsequent observation period, and update the reference set of branches in the same box based on the verification results.

2. The method for rapid location of branch faults in a cable branch box according to claim 1, characterized in that, S1 includes: When establishing a reference set for branches in the same box, the current response sequence, zero-sequence response sequence, and load fluctuation sequence of each branch are synchronously collected according to a unified time base. The collected results are sorted in order, missing time slices are filled in, and sampling spikes are removed. During normal operation, the steady-state offset boundary, zero-sequence response stable interval, fluctuation rhythm boundary, and recovery time boundary are extracted and written into the reference file.

3. The method for rapid location of branch faults in a cable branch box according to claim 1, characterized in that, S2 include: When any branch experiences an abnormal response, the moment of the first abnormality is taken as the event anchor point for the synchronous windowing of all branches. The leading window, action window, and return window of all branches are synchronously captured to form a synchronous event window group in the same box. For branch time segments that meet the low confidence determination criteria within the candidate cutoff window range, they are recorded as low confidence segments of the corresponding branch and retained in the same box synchronous event window group.

4. The method for rapid location of branch faults in a cable branch box according to claim 1, characterized in that, S3 includes: Based on the synchronous event window group of the same box, the main abnormal branch is determined by exclusive comparison in the order of first boundary crossing time, number of consecutive boundary crossings, number of consecutive deviations from zero-sequence response, and number of consecutive non-return boundary within the recovery window. Then, from the remaining branches, branches corresponding to low-confidence segments, branches whose leader window stability is lower than the stability threshold, and branches whose number of persistent fragments at the non-return boundary in the return window exceeds the upper limit of their respective return duration boundaries are filtered out to form a background branch set containing at least two branches.

5. The method for rapid location of branch faults in a cable branch box according to claim 4, characterized in that, S3 also includes: Align all branches within the background branch set with the same time slice in the leading window, action window, and return window according to the event anchor point, and aggregate the common current disturbance value and common zero-sequence disturbance value at each moment to form the background disturbance trajectory in the same box. The background disturbance trajectory in the same box consists of a common current disturbance sub-trajectory and a common zero-sequence disturbance sub-trajectory.

6. The method for rapid location of branch faults in a cable branch box according to claim 1, characterized in that, S4 include: After aligning the event response sequence of each branch to be judged within the leading window, action window and return window with the background disturbance trajectory of the same box using the same time slice, subtracting each slice one by one, a difference response trajectory composed of current difference sub-trajectory and zero sequence difference sub-trajectory is generated. When the length of the missing segment of the common disturbance does not exceed the number of missing time slices for the missing detection judgment, a missing detection supplementation mark is written into the difference response trajectory; When the length of the missing segment of the common disturbance exceeds the number of missing time slices for the missing detection judgment, a long missing detection degradation marker is written into the differential response trajectory.

7. The method for rapid location of branch faults in a cable branch box according to claim 1, characterized in that, S5 includes: Based on the differential response trajectories of each branch to be judged, the initial partial sequence quantity is extracted in the initial section of the action window, the coupling deviation quantity is extracted in the main section of the action window, and the regression residual quantity is extracted in the regression window. The initial partial sequence quantity, coupling deviation quantity and regression residual quantity constitute the branch fault fingerprint.

8. The method for rapid location of branch faults in a cable branch box according to claim 7, characterized in that, S5 also includes: Based on the fault fingerprint of each branch to be judged, exclusive decision is performed in the order of initial partial sequence, coupling deviation and regression residual, priority branch is determined and written into the event-level decision record.

9. A method for rapid location of branch faults in a cable branch box according to claim 1, characterized in that, S6 include: When the verification passes, the abnormal sample segment in the event corresponding to the faulty branch is registered as an abnormal sample segment. If the backtest fails, the event segment that meets the condition of returning to the zero-difference stable interval will be registered as a candidate sample segment to be included again. If the verification is insufficient, the section to be verified in the corresponding event is registered as the section to be verified, and the reference set of the branch in the same box is updated accordingly.

Images