A distribution box arc fault monitoring system and safety warning method

Abstract

The application relates to a low-voltage power distribution circuit fault monitoring technology and discloses a power distribution box arc fault monitoring system and a safety early warning method. The system uses high-frequency transients generated by the on-off operation of existing circuit breakers or contactors in the normal operation of the power distribution box, learns the first-arrival time sequence, polarity relationship and frequency band energy decay interval of the incoming line and branch monitoring points without injecting diagnostic signals, and forms a topological response reference table indexed by branches. For the to-be-judged transient that does not exist the edge of the corresponding operation instruction, the event source branch is located by matching the above relationship, and the to-be-judged event that the incoming line monitoring point is ahead of each branch monitoring point and each branch shows a common mode response is identified as an external disturbance and is excluded; when the repeated breakdown pulse clusters in multiple power supply half cycles continuously locate the same branch, an arc fault early warning is output. The application can consider normal switch transients, branch positioning and safety disposal.

Description

Technical Field

[0001] This invention relates to the field of low-voltage power distribution circuit fault monitoring technology, and in particular to branch location and safety early warning of arc faults in distribution boxes. Background Technology

[0002] Damaged wires, loose terminals, aging insulation, or burnt contact points in low-voltage distribution boxes can create series or parallel electric arcs. The arc current may not exceed the circuit breaker's overcurrent setting, but it will continuously generate high temperatures near the conductor. If the faulty branch cannot be identified early, maintenance personnel must shut down power to each branch to troubleshoot, affecting the continuity of power supply.

[0003] Electric arcs generate high-frequency current transients during ignition and extinguishing. Existing arc detection devices often analyze high-frequency energy, harmonics, or waveform distortion at the main inlet line, while other solutions deploy sensors in multiple branches and use classification models to diagnose faults. Detection at the main inlet line cannot directly identify the source branch; multi-sensor solutions, if only comparing energy magnitude, are easily affected by cable length, sensor installation direction, and coupling paths within the enclosure.

[0004] Normal switching operations in distribution boxes can also generate high-frequency transients. Contactor engagement, circuit breaker disconnection, switching power supply activation, and external grid disturbances can all manifest as one or more spikes. Simply using high-frequency energy exceeding limits as an arc criterion can easily misinterpret normal operations as faults. While using operation window shielding can reduce false alarms, it only answers whether an event occurred near the switching action and cannot utilize the distribution topology propagation information carried by the normal operation itself.

[0005] Some line location techniques use a dedicated pulse source to inject test signals into the circuit and then compare the responses of each node. This method is feasible during power outage maintenance or with dedicated diagnostic equipment, but adding an injection circuit to an operating distribution box introduces requirements for insulation, compatibility, and control timing. For distribution boxes that already have the capability to acquire switch status and monitor transients in multiple branches, a more suitable approach is to obtain references from normal switching operations rather than generating additional test pulses.

[0006] When normal switching transients propagate from the operated branch to the incoming line and other branches, in addition to amplitude attenuation, they also form a relationship of arrival time and polarity determined by the actual wiring, cable length, and sensor direction. This relationship is relatively stable within the same distribution box and naturally carries the label of the operated branch. How to accumulate these tagged passive transients into an updatable topology response reference and use it for branch location without command events and elimination of external disturbances remains a problem that current technology has not yet solved. Summary of the Invention

[0007] This invention sets up synchronous transient monitoring points on the incoming line and each branch, and connects them to the operating commands and position status of circuit breakers or contactors. When a switch is normally opened or closed, the system knows which branch is being operated on and that the transient is not an unknown fault event. The system then extracts the order of arrival times of the first pulses from multiple monitoring points, the polarity of the first main pulse, and the energy attenuation range in multiple frequency bands, forming a topology response reference table according to branch identification.

[0008] When there is no corresponding operation command edge for the event to be judged, the system extracts relationships of the same dimension and matches them with the topology response reference table. The first arrival time relationship adopts a partial order that allows parallel levels, without requiring all monitoring points to form an absolute sequence, which can accommodate synchronization errors and cases where the propagation distance is very short. Polarity relationship helps to offset differences in sensor installation orientation, while frequency band attenuation supplements the differentiation of propagation paths. The three types of relationships jointly participate in the localization, and the source branch is not directly determined by a single point of maximum energy.

[0009] Specifically, the synchronous transient acquisition module is used to acquire the high-frequency current transient sequences of the incoming line monitoring point and each branch monitoring point. The switch status access module is used to acquire the edge and position status of the operation command. The passive reference establishment module is used to extract the first arrival time sequence, polarity relationship, and frequency band energy attenuation range during normal switching and opening, and to establish a topology response reference table. The event location module is used to perform relationship matching and determine the event source branch when there is no corresponding operation command edge. The safety warning module is used to output an arc fault warning when repeated breakdown pulse clusters are continuously located to the same event source branch. All modules use the same time reference for acquiring or processing objects.

[0010] When the incoming line monitoring point responds first and all branch monitoring points show a common-mode response of the same polarity approximately simultaneously, the event is more likely to originate from outside the distribution box, and the system marks it as an external disturbance. Normal switching operations are confirmed by the edge of the operation command, the completion of the position, and changes in the power frequency current. Only if the remaining uncommanded events form recurring breakdown pulse clusters within multiple power supply half-cycles and continuously pinpoint the same branch will an arc fault warning be output.

[0011] This invention transforms transients generated during normal operation into passive references with actual branch labels, eliminating the need for an additional pulse injection source. The reference table contains a comprehensive response of the cabinet's actual cables, terminals, sensor orientation, and coupling conditions, and can be progressively updated during normal operation. Compared to simply shielding switching events within a time window, normal events are not only excluded but also used for subsequent positioning; compared to simply comparing energy magnitudes, first-arrival partial order and polarity relationships are more robust to amplitude drift.

[0012] The present invention also provides a safety early warning method, a computer-readable storage medium, and an electronic device. The early warning output may include audible and visual prompts, sending alarms, or controlling the tripping of the faulty branch when safety interlocking conditions are met, and retains multi-point transient records before tripping for verification. Attached Figure Description

[0013] Figure 1 This is a flowchart of the arc fault safety early warning method of the present invention; Figure 2 This is a functional structure diagram of the arc fault monitoring system of the present invention; Figure 3 A schematic diagram of the on-site layout of the monitoring points for the incoming and branch lines inside the distribution box; Figure 4 A schematic diagram illustrating the state sequence for forming a passive reference during normal switching operation; Figure 5 This is a schematic diagram of the interactive timing of transient propagation between the incoming line, the target branch, and adjacent branches; Figure 6 This is a schematic diagram of the grid state for the branch-monitoring point relationship in the topology response reference table; Figure 7 This is a schematic diagram of the wiring topology of the distribution box's incoming line and multiple branch lines; Figure 8 This is a cross-sectional view of a high-frequency current sensor mounted on a conductor. Figure 9 A schematic diagram showing the planar location and propagation path of each branch within the cabinet; Figure 10 This is a diagram illustrating the transition of a safety warning status for an event pending judgment. Figure 11 This is a partially enlarged schematic diagram of the branch sensor installation location. Detailed Implementation

[0014] The embodiments of the present invention are described below with reference to the accompanying drawings. The numerical values ​​described are illustrative examples and do not constitute a limitation on the scope of protection of the present invention. The embodiments are used to illustrate the operational processes of system configuration, reference establishment, and event localization, and are not intended to limit the present invention. The frequencies, times, energy ratios, and confidence levels mentioned herein are all feasible configuration examples.

[0015] The terminology is explained below. Regarding the event localization module: This refers to the functional unit that matches the relationship to be determined with reference items and outputs candidate branches. Regarding the event source branch: This refers to the branch with the highest matching degree to the transient propagation relationship to be determined and that passes the discrimination value verification. Regarding the synchronous transient acquisition module: This refers to the functional unit that acquires transient sequences from multiple monitoring points using a unified trigger or a calibrated clock. Regarding the safety early warning module: This refers to the functional unit that performs repeated breakdown confirmation, alarm classification, and safety interlock requests.

[0016] Regarding the switch status access module: This refers to the functional unit that receives operation commands, position status, and power frequency current status. Regarding the topology response reference table: This refers to the data table that stores the first arrival partial sequence, polarity, and frequency band energy attenuation range according to branch index. Regarding the passive reference establishment module: This refers to the functional unit that uses high-frequency transients generated by normal switching operations to establish or update the topology response reference table. Regarding the high-frequency current sensor: This refers to a current inductor, Rogowski coil, or magnetic field detection device that responds to megahertz-level current transients.

[0017] In this embodiment, the edge of the operation command is the moment when the controller issues a tripping or closing command to the circuit breaker or contactor, or it can be the moment when the auxiliary contacts begin to change. Normal operation transient is when the switch position reaches the target state within a specified completion time, the power frequency current change matches the target state, and there is no transient that involves repeated breakdowns across multiple half-cycles. Passive reference refers to establishing the reference without injecting dedicated diagnostic voltage or current into the distribution circuit, but rather utilizing the normal switching operations that occur during the daily operation of the distribution box.

[0018] The first arrival time is the moment when the high-frequency transient envelope continuously reaches the first arrival judgment value above the noise baseline. The order of first arrival times is represented using a partial order: if the time difference between the first arrival times of two monitoring points is not greater than the time tolerance threshold, they belong to the same order level; only when the time difference exceeds the time tolerance threshold is the order clearly recorded. The polarity relationship is the sign of the first main pulse of each monitoring point relative to the pre-defined positive direction of the sensor. The frequency band energy attenuation range is the statistical range of the ratio of the energy of other monitoring points within a given frequency band to the energy of the monitored point in the operated branch.

[0019] like Figure 1 As shown, the safety early warning method includes steps 101 to 106. Step 101: During normal operation of the distribution box, collect synchronous high-frequency current transient sequences from the incoming monitoring point and each branch monitoring point, while simultaneously receiving switch operation commands and position status. Step 102: After confirming a normal switching operation, extract the first-arrival partial sequence, polarity relationship, and bandwidth energy ratio using the operated branch as the label. Step 103: Accumulate multiple normal operations for the same branch and establish or update the topology response reference table.

[0020] Step 104: Detect the transient to be judged when there is no corresponding operation command edge, and extract the to-be-judged relation isomorphic to the reference table. Step 105: Match the to-be-judged relation with the reference items of each branch to determine the candidate event source branch, and perform external disturbance elimination. Step 106: Perform cross-half-cycle confirmation on the recurring breakdown pulse cluster of the to-be-judged event, and output an early warning when the branch consistency and safety criteria are met. Each step does not need to be completed within a single central processing unit; the acquisition node can first calculate the local first arrival time and frequency band energy, and then send the features to the cabinet controller.

[0021] like Figure 2 As shown, the monitoring system includes a synchronous transient acquisition module, a switch status access module, a passive reference establishment module, an event location module, and a safety early warning module. The synchronous transient acquisition module stores a circular buffer for all channels before and after the trigger point. The switch status access module correlates operating commands, auxiliary contacts, and power frequency current states to the same time axis. The passive reference establishment module only receives samples confirmed as normal operation, preventing unknown arcs from being written into the reference table.

[0022] In a four-branch distribution box, monitoring point M0 can be set on the main incoming line, and monitoring points M1 to M4 can be set on the branches L1 to L4 respectively. High-frequency current sensors can be high-frequency current transformers, Rogowski coils, or magnetic field sensors with high-frequency response. The effective detection bandwidth can be 0.5MHz to 30MHz, with 1MHz to 10MHz typically used. The lower limit of the bandwidth should be higher than the main power frequency harmonics and general load gradient components, while the upper limit is determined by the sensor bandwidth, sampling rate, and electromagnetic compatibility conditions. The lower limit can be increased when the switching power supply noise inside the cabinet is strong, and the upper limit can be decreased when the cable is long and high-frequency attenuation is significant.

[0023] The sampling rate can range from 20MS / s to 100MS / s, with 50MS / s being a typical value. When all channels are sampled synchronously by the same analog-to-digital converter, the inter-channel error mainly comes from the analog front-end. When multiple acquisition nodes are sampled distributedly, a common trigger line, IEEE 1588 hardware timestamps, or delay correction for known normal switching transients can be used. The time tolerance threshold can range from 0.1μs to 5μs, with 0.8μs being a typical value, and is determined by the sum of clock synchronization error, analog link group delay deviation, and one sampling period. Increasing this value will result in more same-level judgments, while decreasing this value can improve positioning resolution but requires stricter synchronization accuracy.

[0024] Normal operation confirmation requires the simultaneous fulfillment of three conditions. First, there must be a clearly defined edge of the operation command. Second, the auxiliary contact of the switch must reach the target position within a completion time of 20ms to 500ms, with the typical completion time determined by adding a 20% margin to the upper limit of the switch product specifications. Third, the post-operation power frequency current state must match the switching target, and the event must not form multiple adjacent half-cycles of repetitive breakdown pulse clusters. If any condition is not met, the sample will not be included in the passive reference table but will be moved to the pending judgment queue.

[0025] For a normal operational transient, the system first calculates the noise baseline for each channel. The first-reach judgment value can be taken as the mean of the background envelope from 1 ms to 20 ms before the event plus 4 to 8 times the standard deviation, typically the mean plus 6 times the standard deviation. The factor can be increased when the background noise increases, and the factor can be decreased and the continuous sampling point requirement can be increased for weak signal branches. The first-reach time requires that the envelope exceeds the judgment value for 3 to 12 consecutive sampling points, typically 5 sampling points, to avoid mistaking isolated electromagnetic glitches for propagation first-reach.

[0026] Polarity extraction does not use the positive and negative areas of the entire oscillation waveform, but instead takes the polarity sign of the first main pulse after the first arrival time. During sensor installation, a positive direction is registered for each channel. If a sensor's direction is opposite to the preset direction, simply flip the polarity in the channel configuration without changing the physical installation. This allows relative polarity patterns such as "target branch is positive, incoming line is negative, adjacent branch is positive" to be stably written into the reference table.

[0027] The frequency band energy can be calculated in three frequency bands: 1MHz to 3MHz, 3MHz to 6MHz, and 6MHz to 10MHz. The specific frequency band is not limited to these bands and can be set based on sensor response and noise avoidance results within the cabinet. To avoid the denominator approaching zero, the larger value between the frequency band energy of the monitored point of the operated branch and the lower energy limit is used as the denominator. The lower energy limit can be taken as 10 to 50 times the median background energy of the same frequency band, typically 20 times. This factor should be increased when the background changes significantly, and decreased for weakly loaded branches, but should still be higher than the upper noise limit.

[0028] An initial reference is established after 5 to 30 normal operations on the same branch, typically 10. The first arrival time difference and band energy ratio are represented by the median, and the dispersion is represented by the median absolute deviation. The reference interval can be ±2 to 5 times the median absolute deviation, typically ±3 times. For branches with large dispersion, the interval can be expanded, but if there is significant overlap with the reference of adjacent branches after expansion, it should be suggested to add sensors or adjust monitoring points, rather than expanding it indefinitely.

[0029] The reference update adopts a gradual update method. Each new normal operating sample is first checked against the current reference; only samples whose dimensions all fall within the current range or whose only low-weight dimension exceeds the range are included in the update. If three or more consecutive normal samples exceed the range, a prompt message indicating a change in wiring topology, sensor relocation, or switch replacement is output, automatic updates are paused, and a reconfirmation of the cabinet topology is required. This process prevents the reference table from gradually becoming mismatched after equipment topology changes.

[0030] like Figure 3As shown, M0 is located at the main incoming line conductor, and M1 to M4 are located after each branch circuit breaker. Branch sensors should be close to the branch start point so that the arc within the branch first reaches the monitoring point of that branch before propagating to the common bus. Incoming line sensors should be located between the common bus and the upstream power supply. The data acquisition unit and sensor cables are laid separately along the cabinet sidewall, with the same length used at the analog front end or recording the inherent propagation delay of each channel.

[0031] like Figure 4 As shown, a normal closing operation consists of three states: "receiving the closing command, contact movement, and contact stabilization." A single switching transient generated during contact movement is marked as the reference event for that branch. If a high-frequency pulse recurs in the adjacent power supply half-cycle after contact stabilization, this operation is no longer considered a normal reference. A normal opening operation is handled in the same way, but opening and closing can have separate reference entries because the polarity of their first main pulses may be opposite.

[0032] like Figure 5 As shown, when branch L2 is normally closed, M2 usually responds first, followed by M0, and then M1, M3, and M4 coupled through the common bus. M1 and M3 have similar propagation distances and can be at the same partial order level. The partial order can be recorded as M2 earlier than M0, M0 earlier than the parallel levels of M1 and M3, and this parallel level earlier than M4. If M1 and M3 are swapped in another sampling, as long as they still fall at the same parallel level, the matching result is unaffected.

[0033] like Figure 6 As shown, the topology response reference table can store branch identifiers by row and monitoring points by column. Each unit records the first arrival level, polarity, and multiple frequency band energy ratio intervals. For monitoring points that do not participate in the location of a particular branch, the unit can be marked as default. The reference table is not a replacement for the electrical primary diagram, but rather an observation table of actual high-frequency propagation relationships; it can be relearned after changes in the position of conductors within the cabinet.

[0034] like Figure 7 As shown, the incoming monitoring point and each branch monitoring point form a connection topology with the common bus as the upstream. During event localization, branches are first screened based on the first arrival partial order, then reference items with inconsistent installation directions are excluded based on polarity, and finally, the matching degree is calculated based on the frequency band energy attenuation range. The matching degree for the three dimensions is either 0 or 1, or mapped to a value between 0 and 1 according to the interval distance. The final matching degree can be taken as a weighted average of the three. The weights are determined by the discriminative power of the health references, and no arbitrary fixed weights are used.

[0035] For example, a candidate event might have a partial order where M3 precedes M0, M0 precedes M2 and M4, and this partial order precedes M1, perfectly matching the partial order of the L3 reference. Four of the five monitoring points show consistent polarity relationships. Eight energy ratios across the three frequency bands fall within the specified intervals, and two fall outside the intervals within a median absolute deviation. If the matching degrees for first arrival, polarity, and energy dimensions are 1.00, 0.80, and 0.90 respectively, and the weights for distinguishing healthy samples are 0.45, 0.20, and 0.35, then the total matching degree is 1.00 × 0.45 + 0.80 × 0.20 + 0.90 × 0.35 = 0.925. This value is higher than the localization determination value of 0.82, therefore the candidate event source is L3. This example is for illustrative purposes only and does not represent actual measured accuracy.

[0036] The positioning judgment value can be between 0.65 and 0.95, with a typical value of 0.82. It is determined by performing leave-one-out verification during the reference establishment phase to maintain a gap between the lower bound of the correct branch matching degree and the upper bound of the incorrect branch matching degree. This value is increased when the number of branches increases or reference overlap increases; it can be decreased when candidate branches need to be provided first and then manually reviewed. The difference between the highest and second-highest matching degree should also reach a discrimination value of 0.05 to 0.30, typically 0.12, which is also determined by leave-one-out verification.

[0037] External disturbances are excluded based on propagation direction rather than energy magnitude. If M0 is in the earliest level, and the monitoring points of each branch are located one or two levels later, and at least 80% of the first main pulses of the branches maintain the common-mode polarity referenced with M0, then it is marked as an external disturbance. The common-mode ratio can be taken from 60% to 100%, typically 80%, determined according to the branch online rate; the required ratio is increased when there are fewer online branches, and the ratio can be reduced when the sensor occasionally fails to report, but there must be at least three effective branches.

[0038] like Figure 8 As shown, the high-frequency current sensor can be mounted on an insulated wire or clamped to the outside of the branch copper busbar. Electrical insulation is maintained between the sensor and the current-carrying conductor, and the magnetic core opening is locked by a clip after closure. The center of the conductor should be located as close as possible to the center of the magnetic core window, and the sensor leads should be kept away from the switch contacts. For a three-phase branch, each phase can be monitored separately, or the sensor can be installed on a single-phase branch provided that the magnetic flux does not cancel each other out.

[0039] like Figure 9 As shown, the planar position within the cabinet affects the coupling of adjacent branches. The system does not directly calculate attenuation based on geometric distance; instead, it allows normal operation to write the actual coupling results into a reference table. If branches L1 and L2 are adjacent within the cabinet, their energy ratios may be similar, and they can still be distinguished using first-arrival partial order and polarity relationships. If they cannot be distinguished in any of the three dimensions, the system outputs "branch group candidate" without forcibly specifying a single path, and prompts for the addition of monitoring points at the branch points.

[0040] Repeating breakdown pulse clusters are grouped according to their power frequency phase. For a 50Hz power supply, one half-cycle is 10ms. Within each half-cycle, the system groups multiple high-frequency pulses with intervals of less than 20μs to 500μs into a pulse cluster, with a typical interval of 100μs. The pulse cluster needs to repeat across 2 to 12 half-cycles, typically 3 half-cycles, to form evidence of arc repetition. The upper limit of the interval can be determined based on the maximum duration of a normal pulse in the switching power supply; increasing the number of repeated half-cycles can reduce false alarms of occasional glitches, while decreasing it can accelerate the response to high-risk arcs.

[0041] like Figure 10 As shown, event handling includes monitoring, pending judgment, location, early warning, and handling states. Events with operational instructions that are completed normally return to monitoring after reference update; events with M0 leading and branch common mode are returned to monitoring as external disturbances; events without instructions first enter the location state. If the location branches of multiple half-cycle pulse clusters are inconsistent, the record is retained but no arc warning is generated; if the same branch is continuously located and the matching degree reaches the location judgment value, the early warning state is entered.

[0042] Safety warnings can be divided into alert level and tripping suggestion level. The alert level requires the number of repetition half-cycles to reach a set value; the tripping suggestion level also requires a total matching degree of 0.90 or higher, or a frequency band energy level of 2 to 8 times the upper limit of the health reference, typically 4 times. This multiple is determined by the circuit breaker withstand capability, load importance, and fire-fighting linkage requirements. Important fire-fighting circuits may only send alarms without automatic tripping, while regular branches can be controlled to trip when interlocking permits.

[0043] like Figure 11 As shown, the branch sensor mounting components include an open magnetic core, a latch, a conductor window, and shielded leads. A partial enlarged view illustrates the sensor's positive orientation marking and lead fixing method. After replacing the sensor, channel delay and polarity correction should be re-performed; whenever the sensor's position or orientation changes, the corresponding channel in the original reference table should be marked as invalid to avoid using old parameters.

[0044] If a channel fails, event localization can be performed on the remaining channels, but the matching degree must be recalculated after deleting the column data corresponding to the same monitoring point from the reference table. When there are fewer than two branches plus the incoming line, single branch localization is not output; only distribution box-level arcing anomaly is output. If all branches have not undergone normal switching operations for an extended period, initial references can be formed by normal switching operations during factory commissioning or planned maintenance, but no additional diagnostic signals are injected.

[0045] This method does not rely on machine learning models. First-order partial matching, polarity relationships, statistical intervals, and weighted matching can all be performed by an embedded processor. If a classifier is used to further fuse the matching scores, the classifier input should still include the aforementioned physically meaningful relational features, and retain hard criteria for external disturbances and switch states to avoid the opaque model output directly controlling the circuit breaker's tripping.

[0046] The computer-readable storage medium can store procedures for synchronous acquisition, normal operation confirmation, reference table update, pending event matching, repeated breakdown confirmation, and safety handling. The electronic device includes a processor, memory, acquisition interface, and switch status interface. When the processor executes the program, it completes steps 101 to 106 and saves the warning record along with the waveforms before and after the trigger to non-volatile memory.

[0047] Before the system is put into operation, a channel list should be established. The channel list should at least record the monitoring point identifier, corresponding branch, sensor model, installation direction, analog front-end number, inherent propagation delay, and effective frequency band. The operation status list should record the switch identifier, the source of the open / close command, the auxiliary contact logic, and the rated completion time. Event handling should be associated with a unique identifier in both lists to avoid interpreting events as occurring on the old channels after branch rerouting.

[0048] The sensor's positive direction can be defined as outputting a positive pulse when the power supply flows to the load side. After installation, verify the symbol of the first main pulse when the branch is normally closed. If the actual symbol is opposite to the specified one, a polarity reversal bit can be set in the configuration. Polarity reversal only changes the symbol interpretation and does not change the original waveform for subsequent auditing.

[0049] The analog front end may include limiting, bandpass filtering, and differential amplifier circuitry. Limiting circuitry prevents prolonged saturation of the analog-to-digital converter due to large switching transients; the bandpass filter's passband covers a selected high-frequency band; multiple channels should use components from the same batch or have their group delays measured separately. If the group delay difference exceeds half the time tolerance threshold, it must be corrected in the timestamp.

[0050] Triggering can occur when the envelope of any high-frequency channel crosses the trigger value. After triggering, each channel extracts data from the circular buffer for 0.5ms to 10ms before triggering and 2ms to 50ms after triggering, typically 2ms before triggering and 10ms after triggering. The pre-trigger segment is used to calculate the noise baseline, and the post-trigger segment is used to extract the first arrival, polarity, and band energy.

[0051] To prevent a physical event from being repeatedly registered due to multiple channels triggering sequentially, an event merging time can be set. The event merging time can be set from 20μs to 2ms, with a typical value of 200μs. Its lower limit is greater than the sum of the maximum propagation difference between channels and the synchronization error, and its upper limit is less than the minimum separable time of two independent switching operations. Triggers within the merging window share the same event identifier.

[0052] In applications where the power frequency current is small but high-frequency noise is persistent, a dual-condition triggering approach can be used: the envelope exceeds the trigger value, and at least two monitoring points respond within the event merging time. This condition is used to suppress localized electromagnetic interference near a single sensor. For branches with long terminations and significant attenuation in other channels, single-channel triggering can be retained, but reference matching must be required during the positioning phase.

[0053] The switch completion time does not use a fixed value applicable to all devices. The system can obtain an initial upper limit from the mechanical action time range given by the manufacturer, and measure a stable distribution from the command edge to the auxiliary contact over ten normal operations. The completion time is taken as the measured 99th percentile plus a 20% margin, but must not exceed the switch's allowable action time. A mechanical fault should be reported separately if the action time continues to increase.

[0054] Verification of normal operation samples can also check for contact bounce. If the auxiliary contact flips multiple times within a short period after reaching the target position, or if a high-frequency pulse cluster spans multiple power frequency half-cycles, this event is not considered. Contact bounce itself may be a signal of equipment degradation and can generate a switch maintenance prompt, but it should not be confused with the results of the line arc branch location.

[0055] Closing and opening references should be stored separately. The closing transient is determined by the load input capacitance, contactor contacts, and line impedance; the opening transient is also related to the inductive component of the load. When matching the event to be judged, it can be compared with both types of references simultaneously, but the reference type is an independent field. If both types match the same branch, the branch positioning confidence can be improved.

[0056] The first-pass envelope can be obtained by averaging the absolute value of the high-frequency signal over a period of 4 to 32 sampling points, typically 8 points. A window that is too short is susceptible to single-point ringing, while a window that is too long will smooth out the microsecond-level sequence between channels. The window length can be selected based on the repeatability of the first-pass time in normal operating samples to ensure that the first-pass level remains consistent across ten operations on the same branch.

[0057] When establishing the first-arrival partial order, all monitoring points are first sorted according to the corrected first-arrival time, and then layered starting from the earliest time. Monitoring points whose difference from the earliest time of the current layer is not greater than the time tolerance threshold are grouped into the same layer, and the remaining monitoring points move to the next layer. This algorithm is applicable to any number of branches and will not produce an uncertain order due to two channels having the same sampling point exceeding the limit.

[0058] Channels that do not reach the initial arrival threshold are not forcibly assigned a late arrival time; instead, they are registered as "not arrived." During matching, if a channel is considered to have arrived but is actually not, the matching score will be lowered; however, if a channel is actually not arrived, no points will be deducted. This distinguishes between genuine non-arrival caused by propagation attenuation and the collector mistaking missing values ​​for final arrival.

[0059] The first main pulse can be searched within the range of 0.2μs to 20μs after its initial arrival, with a typical window of 5μs. If the signal saturates within this window, its polarity is recorded as uncertain and is not used for polarity matching of the current event. The system does not infer the polarity of the first pulse based on the subsequent bounce of the saturated waveform to avoid amplifying the influence of the front-end recovery process.

[0060] Bandwidth energy can be calculated using discrete Fourier transform, digital bandpass filtering followed by squared integral, or wavelet decomposition. Regardless of the calculation method used, all channels must use the same window function and time slice. If Fourier transform is used, a Hanning window can be added to the 20μs to 500μs segment after the first arrival, with a typical segment length of 100μs, so that the energy of each channel corresponds to the same propagation stage.

[0061] The denominator for the energy ratio must not be zero. Let the energy of a target branch channel in a certain frequency band be E_s, and the lower limit of the background energy be E_min. Then the denominator is max(E_s, E_min). The ratio of the energy E_i of other channels to this denominator forms the attenuation. If E_s is lower than E_min, it indicates that the frequency band does not have a sufficient signal-to-noise ratio, and the entire frequency band is marked as invalid. No false effective ratio will be generated due to the protection factor.

[0062] The median absolute deviation is the median of the absolute values ​​of the differences between each sample and the median. If the median of a certain energy ratio for ten normal operations of the same branch is 0.28 and the median absolute deviation is 0.04, the reference interval is 0.16 to 0.40 when using a range of plus or minus 3. A new sample with a value of 0.35 is included in the update; a new sample with a value of 0.62 is retained as an out-of-bounds sample but the reference is not changed immediately.

[0063] When the median absolute deviation is zero, the minimum interval width can be determined using measurement resolution and background fluctuations. For example, the minimum width for energy ratio can be 0.02 to 0.10, typically 0.05; the minimum width for first-arrival time difference can be taken as the time tolerance threshold. This protection prevents a small number of identical discrete readings from forming a zero-width reference.

[0064] The reference table can save versions. A new version is generated after a wiring modification, and the old version is only used for historical event playback. Each version is associated with a wiring diagram check code, a channel list check code, and an effective time. Event logs reference the reference version that was in effect at that time to prevent subsequent updates from changing the already output positioning conclusions.

[0065] When the number of normal operations is insufficient, the system can generate a temporary reference. The temporary reference must contain at least the initial arrival partial sequence and polarity given by one switching operation, but uses a wider initial range for the energy decay interval. The temporary reference can only output candidate branches and does not trigger automatic tripping. Only after reaching the specified number of samples and passing leave-one-out verification will the reference status change to confirmed.

[0066] The system uses a fail-safe approach, treating one normal sample as the event to be judged and establishing references for the remaining samples. It checks whether the failed sample returns to the correct branch. If the correct branch's matching degree is lower than the positioning judgment value, or the difference between the incorrect and correct branches is less than the discrimination value, then the current monitoring point layout of the cabinet is insufficient for stable discrimination. The system should prompt for adjusting sensor positions or adding monitoring points, rather than simply lowering the positioning judgment value to force the process through.

[0067] Common-mode detection of external disturbances can also be combined with polarity reversal configuration. Common-mode does not require all original voltage signs to be identical, but rather that after normalization according to the installation orientation, they present a common propagation direction defined in the reference with the incoming line monitoring point. If the orientation of a branch sensor is incorrectly configured, it will be exposed during normal operation verification, and the configuration should be corrected first.

[0068] External lightning strikes, grid switching, or upstream switch operations typically reach M0 first, then enter the branch circuit via the common bus. If M0 arrives first, but only one branch responds clearly while the others do not, it cannot be judged as an external disturbance solely based on M0's early arrival; matching with that branch should continue. An external disturbance requires both the incoming line arriving first and a sufficient proportion of branches to share the same mode.

[0069] The elimination of normal switching events cannot solely rely on time proximity. If the switch fails to engage after the operation command is issued, and the high-frequency pulse repeats across multiple half-cycles, contact arcing may occur. In this case, the event should enter the pending judgment process and be associated with the switch identifier. The system may consider the branch containing the operated switch as a priori candidate, but ultimately confirmation is based on propagation reference and evidence of repeated breakdown.

[0070] The half-cycle grouping of the arc pulse cluster can be obtained from the zero-crossing time of the acquired power frequency current or the voltage phase. When there is no voltage path, the zero-crossing can be estimated using the fundamental phase of the incoming current. When the phase estimation error exceeds 1ms, only a fixed 10ms window grouping is used, and the weight of phase-related evidence is reduced without affecting the branch propagation relationship matching.

[0071] Multiple closely spaced reignition pulses may exist within a half-cycle. Pulse deduplication uses a cooling time: a new pulse is only registered on the same channel after 5 to 100 μs following the previous pulse and the envelope has fallen back below the first arrival threshold; a typical cooling time is 20 μs. This rule prevents a ringing wave train from being counted as multiple reignitions.

[0072] Repeated breakdown requires not only a certain number of pulses, but also consistent branch positioning in adjacent half-cycles. If the first half-cycle positions L2, the second L3, and the third L2 again, a single-branch warning is not directly generated. Instead, L2 and L3 can be output as branch group candidates, their waveforms saved, and subsequent events awaited. This approach is easier to maintain than simply counting high-frequency pulses on the input line.

[0073] The safety warning module can set the minimum evidence integrity. Single-branch localization is allowed to be output only when at least two of the three dimensions of first arrival, polarity and energy are valid, and the first arrival partial order must be included therein. When only the energy dimension is valid, only the high-frequency abnormality at the distribution box level is output. The first arrival partial order is the main evidence for branch propagation localization in the present case, and cannot be replaced by a single maximum amplitude rule.

[0074] The matching weight comes from the discrimination capability of the health reference. For example, for a certain dimension, the average interval between the matching degrees of the correct branch and the next closest branch is calculated, and the larger the interval, the higher the weight. The sum of the three normalized weights is 1. If the polarity is completely the same in all branches, its discrimination capability is zero, and the weight can be set to zero; the polarity field is still retained in the specification for use by other cabinet types.

[0075] The matching degree can be calculated by a method of assigning 1 within the interval and decreasing according to the distance outside the interval. For a certain energy ratio x, the reference interval is [a,b], and 1 is obtained when x is within the interval; when x<a, it decreases according to 1-(a-x) / d, when x>b, it decreases according to 1-(x-b) / d, and the result is truncated to 0 to 1. d can take the interval width or three times the median absolute deviation, and all symbols are determined by the reference samples.

[0076] After the equipment load is replaced, the band energy of normal switching transients may change, but the first arrival partial order usually still reflects the wiring position inside the cabinet. The system can divide the references into structural fields and load fields: the first arrival partial order and channel polarity belong to the structural fields, and the energy interval belongs to the load field. When the load is replaced, the energy field is frozen first, the structural fields are continuously used to output candidate branches, and the complete matching is resumed after sufficient new normal samples are obtained.

[0077] If one branch includes multiple parallel loads, the normal operation label can also be subdivided to the lower-level contactor. The row key of the topology response reference table may be a combination of the branch identifier and the operating equipment identifier. If the arc to be determined can only be localized to the upper-level branch, the upper-level branch is output; if the lower-level operation reference has a distinguishable propagation relationship, the localization can also be continued to the lower-level branch.

[0078] In a ring network or dual-power distribution box, the event propagation path may change with the state of the tie switch. The reference table should take the power supply mode and the combined state of the tie switch as the topology state key. Direct matching is performed only when the current topology state is consistent with that at the time the reference is established; after the state changes, the corresponding version can be switched to or a temporary reference mode can be entered.

[0079] Dual-power switching itself will generate a strong transient. The switching controller can provide the command edges of the two switches at the same time. The system regards this event as a combined normal operation, and records the topology states before and after the switching respectively. If the switching fails or repeated breakdown across half a cycle occurs, the reference is not updated, and the event is processed according to the to-be-determined event associated with the switch branch.

[0080] Figure 6 The grid status shown can also be used to display reference integrity. Solid nodes indicate that the branch-monitoring point combination has a stable first-reach level, hollow nodes indicate only the energy field, and triangular nodes indicate that the current channel is faulty. This display is for operation and maintenance purposes and does not participate in the mathematical processing described in the claims. The status is updated after each channel self-test, allowing maintenance personnel to know which location results have complete evidence.

[0081] The channel self-test calculates the background root mean square (RMS), DC bias, and saturation ratio when there are no events. An abnormal noise level is reported when the background RMS exceeds the upper limit of health by two to ten times, typically four times; this multiple is determined by the background distribution after installation. The corresponding feature is discontinued if the channel is continuously saturated or the DC bias exceeds 10% of the range.

[0082] An open sensor opening will reduce high-frequency response. During each normal operation, check the energy level of this branch channel against historical ranges. If the energy level is below 50% of the historical lower limit for three consecutive normal operations, it indicates a core opening or lead wire fault. This warning does not directly interpret low energy as the absence of an electric arc; instead, it signals a sensor link maintenance issue.

[0083] When data collectors synchronize via a network, secondary correction can be performed using the transients during normal operation. The two channels with the highest signal-to-noise ratio in the same event are selected, and the relative delay is obtained by calculating the cross-correlation peak value; the median delay from multiple normal operations is used as a fixed correction value. Secondary correction only compensates for the inherent propagation delay and cannot arbitrarily shift each event to the reference point, so as not to eliminate the true propagation sequence.

[0084] Event waveforms can be saved using lossless compression. The saved content includes the original sampled values ​​for each channel, sampling rate, range, filter configuration, timestamp, trigger channel, switch status, and reference version. When transmitting externally, only the characteristic and abbreviated waveforms can be sent, while the original waveforms are retained in the cabinet, facilitating auditing and reducing communication overhead.

[0085] Before taking any safety precautions, check the controllability of the switch and the importance level of the load. If the branch is marked as not allowed to automatically trip, the system will only output audible and visual alarms and remote alarms. If tripping is permitted, a tripping suggestion can be issued first, and confirmation from the upper controller can be awaited. After tripping, continue recording the transient state for no less than 50ms to confirm whether the high-frequency arc pulse has disappeared.

[0086] If, after the circuit breaker trips, the high-frequency pulses still arrive first from the incoming line and persist in a common-mode manner along the branches, it indicates that the interference may originate from the upstream power grid. If, after the circuit breaker trips, only the target branch channel stops while other branches remain unchanged, the original positioning result is supported. This post-verification record is used for maintenance review and is not a condition that must be waited for before the circuit breaker trips.

[0087] The system can count maintenance events for each branch, including the number of normal operations, the number of external disturbances, the number of unidentified anomalies, the number of arc candidates, and the number of warnings issued. The count is only for operational statistics. The evidence requirements for the current event should not be lowered because a branch has few historical warnings, nor should normal operations and external disturbance elimination be skipped because of a large number of historical warnings.

[0088] Firmware upgrades should not clear the reference table. Before upgrading, perform a verification backup of the reference version and channel list. After upgrading, run an offline playback once, using saved normal events to verify that the matching results are consistent with those before the upgrade. When algorithm parameters change, a new parameter version is generated; old events can still be replayed using the old version.

[0089] For a simple distribution box with only two branches, three monitoring points, M0, M1, and M2, can still be set. When the number of partial sequence levels is small, the importance of polarity and bandwidth attenuation increases. For large cabinets with more than ten branches, the data acquisition unit can be grouped, with each group first forming a local first arrival level, and then the entire cabinet's partial sequence is spliced ​​together using the common M0 timestamp.

[0090] When multiple data acquisition units are grouped, each group must share at least one reference monitoring point. The reference monitoring point can be a branch output of M0 or a common magnetic field sensor near the bus. The shared point is used to correct inter-group time, ensuring that inter-group delay errors remain within the time tolerance threshold. If there is no shared point and hardware synchronization is not possible, the microsecond-level sequence of different data acquisition units should not be used for the same partial order.

[0091] This invention can also be used in DC distribution boxes. Since the DC system lacks a power frequency half-cycle, the repetitive breakdown condition can be changed to the appearance of multiple pulse clusters separated by cooling time within a specified observation period, continuously pinpointing the same branch. In this case, the implementation method for the AC half-cycle in the claims can be replaced by a time window repetition condition, but the passive reference establishment and propagation relationship matching remain unchanged.

[0092] In an AC system, an electric arc may only occur stably during a single polarity half-cycle. The system separately counts pulse clusters during the positive and negative half-cycles, without requiring them to occur in both polarity half-cycles. As long as multiple consecutive event source branches of the same polarity half-cycle are consistent, evidence of repetition can be formed. The polarity half-cycle is determined by the power frequency phase and is a different concept from the polarity field of the high-frequency main pulse.

[0093] The latches and lead wire fixings in the enlarged view of the sensor mounting are not part of the additional detection algorithm. Their function is to maintain core closure and channel delay stability. This invention can also be achieved using other mounting structures that maintain the sensor's position, orientation, and insulation. The reference table will account for actual propagation differences between different cabinet types.

[0094] The technical chain of this invention can be replayed step by step: operation instructions and position status prove whether the sample is normal; first arrival partial order and polarity prove the propagation direction; frequency band energy range supplements path differences; and cross-half-cycle repetition proves it is not a single-time switching glitch. When the evidence at any stage is insufficient, the system degrades its output, rather than being directly controlled by an opaque classification result.

[0095] In an embodiment with six branches, the system collects ten normal closing and ten normal opening data for each branch. If the stable offset sequence of L4 closing is, in order, a parallel layer of M4, M0, M3, and M5, or a parallel layer of M1, M2, and M6, then the four layers are directly saved in the reference table, instead of saving the accidental full sequence where M3 is 0.1 μs earlier than M5 in a certain sampling. This reduces the impact of synchronization jitter on subsequent matching.

[0096] If events M4 and M0 fall within the same priority level, but the health reference requires M4 to be explicitly earlier than M0, then the first-reach dimension is not a complete match. The system can record misalignment between adjacent levels as a partial match, and cross more than two levels as a non-match. The score for a partial match is determined by leave-one-out validation and cannot be temporarily set after the event is seen.

[0097] After the energy ratio is normalized according to the branch's own channels, the absolute intensity of different events can be different. For example, in two normal closing events of the same L4 branch, one event has a peak value of 1.2V and the other has 2.0V. As long as the frequency band energy ratio of other channels relative to M4 falls within the same range, they can still be written to the same reference. This normalization avoids directly treating load power changes as topology changes.

[0098] If the target branch channel becomes saturated, the energy ratio may be distorted. The system can select an unsaturated lower frequency band, or use the incoming channel as a temporary normalization reference, but the reference type must be written into the event characteristics. When all frequency bands are saturated, only the first arrival and polarity positioning are used, and the attenuation range cannot be calculated from the clipped waveform.

[0099] Electromagnetic compatibility (EMC) measures within the cabinet may include sensor lead shielding, single-point analog grounding, and isolation of the data acquisition unit's metal casing and digital interface. Changing the shielding grounding point may affect high-frequency propagation; therefore, after overhauling the shielding grounding, normal operation should be performed again for verification. The EMC structure is designed to maintain measurement stability without altering the fundamental method of passive reference.

[0100] When the host system only provides the final switch position without providing the command edge, the first change in the auxiliary contact can be used as the operation edge, combined with the change in power frequency current for confirmation. If the difference between the final position change and the high-frequency event exceeds the allowable operation window, then the event cannot be used as a tagged reference. The allowable operation window is determined by the sum of the upper limit of the switch action time and the communication delay.

[0101] When the system communicates with the fire protection or energy management platform, the early warning message can be divided into an event summary and a waveform index. The event summary includes candidate branches, total matching degree, second highest branch, distinguishing value, pulse cluster half-cycle number, and handling status. The waveform index points to a file stored in the cabinet. After receiving the summary, the platform can request the waveform, but communication failure does not affect the audible and visual prompts in the cabinet.

[0102] The control circuit for automatic tripping should be isolated from the monitoring and calculation circuit. The safety early warning module only sends requests to the execution controller with interlock verification. The execution controller checks the maintenance bypass, critical load indicators, and switch availability status before deciding whether to take action. This hierarchical approach prevents a single point of failure in the data acquisition unit from directly driving the switch.

[0103] The system can also lock the reference update after an early warning. During the lockout period, any switching operations occurring in the target branch are only archived and not written to the health reference until maintenance personnel confirm that the fault has been resolved and the baseline has been re-established. Otherwise, abnormal transients from a continuously arcing switch may gradually contaminate the reference range.

[0104] After maintenance and re-entry into operation, first clear the arc candidate count of the target branch, then collect data from at least five normal operations to form a temporary reference. The old reference and the waveform before the fault are retained for comparative analysis. If the maintenance only replaced the load without changing the sensor and wiring, the initial structure field can be inherited, but the energy field will still be relearned.

[0105] Through the above implementation method, normal switching transients have gained new technical applications: they are not merely interferences to be avoided, but propagation test events tagged with the operated branch. The reference table is generated by the actual equipment itself, and the arc to be judged is compared with the reference under the same set of sensors, the same wiring, and the same frequency band, thereby reducing deviations when a general model is migrated to different cabinet types.

[0106] Regardless of whether the monitoring system uses a single-board data acquisition unit, distributed data acquisition nodes, or intelligent circuit breakers with communication functions, as long as the normal operation sample and the event to be judged are under the same monitoring topology, the first arrival, polarity, and frequency band attenuation relationship can be extracted, and the repeated breakdown confirmation can be completed according to the consistency of the event source branch, the above-mentioned branch location and safety early warning can be achieved.

[0107] Comparative Example 1: If only a switch operation shielding window is set, since the window only provides the event time and does not save the branch propagation relationship, events without instructions outside the window can still only be judged as high-frequency anomalies, and the source branch cannot be determined; if only the monitoring point with the highest energy is taken, changes in sensor direction and branch attenuation will cause the maximum point to drift, thereby changing the positioning result. Through multiple experiments, the inventors found that using isomorphic matching of first-arrival partial order, polarity, and attenuation interval can transform normal events from simple interference into verifiable branch references.

Claims

1. A distribution box arc fault monitoring system, characterized in that, It includes a synchronous transient acquisition module, a switch status access module, a passive reference establishment module, an event location module, and a safety early warning module; The synchronous transient acquisition module is used to acquire the high-frequency current transient sequence of the incoming monitoring point and each branch monitoring point of the distribution box. The switch status access module is used to acquire the edge and position status of the operation command of the branch circuit breaker or contactor. The passive reference establishment module is used to extract the first arrival time, polarity, and frequency band energy attenuation range of each monitoring point from the high-frequency current transient sequence when the operation command is received at the edge and the corresponding switch completes normal opening and closing. It also establishes or updates the topology response reference table according to the identifier of the operated branch, wherein no diagnostic signal is injected into the power distribution circuit. The event localization module is used to extract the order of arrival time, polarity, and frequency band energy attenuation from pending events that do not have corresponding operation command edges. It is then matched with the topology response reference table to determine the event source branch. When the incoming line monitoring point leads the monitoring points of each branch and each branch exhibits a common-mode response, the pending event is marked as an external disturbance. The safety early warning module is used to output an arc fault warning for the event source branch when repeated breakdown pulse clusters occur within multiple power supply half cycles, each repeated breakdown pulse cluster is located to the same event source branch, the corresponding pending event is not marked as an external disturbance and there is no corresponding operation command edge.

2. The system according to claim 1, characterized in that, High-frequency current sensors are installed at the incoming line monitoring point and each branch monitoring point. The effective detection frequency band of the high-frequency current sensor is 0.5 MHz to 30 MHz, and transient sequences are acquired using a uniform sampling trigger signal or a calibrated synchronous clock.

3. The system according to claim 1, characterized in that, When the passive reference establishment module reaches the target position within the completion time after the switch position state is at the edge of the operation command, the power frequency current after the operation matches the target position, and no repeated breakdown pulse clusters occur, the corresponding switch transient is identified as a normal operation transient.

4. The system according to claim 1, characterized in that, The topology response reference table stores for each operated branch: the first arrival order of the incoming monitoring point and the partial order of each branch monitoring point, the polarity sign of the first main pulse of each monitoring point, and the attenuation range of the energy of each monitoring point in multiple preset frequency bands relative to the energy of the operated branch monitoring point.

5. The system according to claim 1, characterized in that, The passive reference establishment module extracts the first arrival time difference and frequency band energy ratio for multiple normal operation transients of the same branch, updates the topology response reference table using the median and median absolute deviation, and discards samples whose absolute value of the difference from the corresponding median exceeds three times the corresponding median absolute deviation.

6. The system according to claim 1, characterized in that, When the time difference between the first arrivals of adjacent monitoring points is less than the sum of the clock synchronization error and the sensor link delay, the event localization module registers the two corresponding monitoring points as being at the same priority level and uses partial order matching instead of total order matching to determine the event source branch.

7. A method for early warning of arc faults in a distribution box using the system described in any one of claims 1 to 6, characterized in that, include: During normal operation of the distribution box, in response to the normal opening and closing operation of the branch switch, the high-frequency current transient sequence of the incoming monitoring point and each branch monitoring point is collected, and a topological response reference table including the first arrival time relationship, polarity relationship and frequency band energy attenuation range is established according to the operated branch. When a transient event to be judged is detected that does not have a corresponding operation command edge, the order of arrival time of the transient event to be judged, the polarity relationship, and the energy attenuation relationship of the frequency band to be judged are extracted at each monitoring point; The extracted relationships are matched with the topology response reference table to determine the event source branch and exclude external disturbances where the incoming monitoring point is ahead of the monitoring points of each branch and each branch exhibits a common-mode response. Statistical analysis of recurring breakdown pulse clusters of transient events to be judged within multiple power supply half-cycles; When each repeated breakdown pulse cluster is located to the same event source branch, an arc fault warning for that event source branch is output.

8. The method according to claim 7, characterized in that, After outputting an arc fault warning, the system executes audible and visual prompts or sends an alarm based on the warning confidence level, the number of repeated breakdown pulse clusters, and the frequency band energy level of the event source branch. The system only controls the branch switch to open when the corresponding transient event to be judged is not marked as an external disturbance, the switch of the event source branch is in a controllable state, and the event source branch is not marked as not allowed to automatically open. The system also saves the transient sequence of each monitoring point for a preset time length before the branch switch opens.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the method of claim 7.

10. An electronic device, characterized in that, It includes a processor, a memory, and a computer program stored in the memory and capable of running on the processor, wherein the processor executes the computer program to implement the method of claim 7.

Images