A fault boundary positioning method for a low-voltage electric energy metering box

Abstract

The present application belongs to the technical field of low-voltage metering equipment fault event perception and intelligent operation and maintenance, and relates to a fault boundary positioning method for a low-voltage electric energy metering box, which collects electric parameters and power-on / off event information of each meter position and box-level bus; clusters the power-on / off event information according to a preset time alignment window to form a candidate fault event cluster; then constructs a fault event propagation graph and a candidate fault boundary based on incoming lines, buses, branch lines and meter position nodes; then calculates the score of the candidate boundary based on indexes of event synchronicity, topological coverage, recovery sequence consistency and zero sequence coupling, and performs fault boundary category identification using a lightweight fault boundary positioning model; finally, determines the affected meter position set according to the identified fault boundary category and the upstream and downstream topological relationship of the fault event propagation graph, and outputs the fault boundary position, risk level and repair priority after performing effectiveness review. The present application is conducive to reducing the risk of false positioning of low-voltage metering equipment.

Description

Technical Field

[0001] This invention belongs to the field of fault event perception and intelligent operation and maintenance technology of low-voltage metering equipment, and relates to a fault boundary location method for low-voltage power metering boxes. Background Technology

[0002] Low-voltage energy metering boxes may experience problems during operation, such as incoming line faults, busbar faults, single branch faults, single meter position faults, and neutral line anomalies. These anomalies typically manifest as power outages, voltage drops, current interruptions, zero-sequence current anomalies, or abnormal recovery sequences. Failure to promptly determine the fault boundary location will expand the scope of on-site troubleshooting and reduce maintenance efficiency.

[0003] Existing fault location methods typically rely on single meter-level events or simple rules for judgment, such as locating faults based solely on a power outage alarm at a specific meter-level location, a voltage loss in a single phase, or manual circuit-by-circuit verification. These methods struggle to fully utilize the synchronicity, topological coverage, recovery sequence, and zero-sequence coupling relationships between multiple meter-level events, making them prone to mislocation and missed location in complex operating conditions such as simultaneous anomalies at multiple meter-level locations, local branch anomalies, and neutral line anomalies.

[0004] Furthermore, fault boundaries in low-voltage energy metering boxes often manifest as differences in the propagation range of events within the topology. Incoming-side faults typically affect most or all meter positions simultaneously, while bus-side faults have a shared impact on meter positions within a specific phase or bus section. Single-branch faults and single-meter-position faults exhibit more localized event distributions. Without establishing a boundary identification mechanism that combines the phase to which the meter position belongs, the box-level bus status, and the propagation relationships of multiple meter position events, it is difficult to generate accurate, stable, and interpretable fault boundary location results. Summary of the Invention

[0005] To address the shortcomings of existing fault location methods, such as reliance on manual inspection, difficulty in distinguishing fault boundaries at different levels, and insufficient accuracy in complex multi-meter-position event scenarios, this invention proposes a fault boundary location method for low-voltage energy metering boxes. This method performs correlation analysis on power outage and power-on events at multiple meter positions to pinpoint the fault boundary location and obtain the risk level and maintenance priority. The specific technical solution is as follows:

[0006] A method for locating the fault boundary of a low-voltage power metering box includes the following steps:

[0007] S1. Collect electrical parameters and power outage / on / off event information for each meter position and the box-level busbar;

[0008] S2. Cluster the power outage and power on event information according to a preset time alignment window to form candidate fault event clusters;

[0009] S3. Construct a fault event propagation diagram based on the phase, branch affiliation, and meter box topology of each meter position, and establish candidate fault boundaries for incoming line side faults, busbar side faults, single branch faults, single meter position faults, and neutral line anomalies.

[0010] S4. For each candidate fault event cluster and candidate fault boundary, calculate the event synchronization index, topology coverage index, recovery order consistency index, and zero-order coupling index, and calculate the candidate boundary score based on the above indexes.

[0011] S5. Based on the candidate boundary scores, a lightweight fault boundary localization model is used to perform fault boundary category identification;

[0012] S6. Determine the set of affected tabletops based on the identified fault boundary categories and the upstream and downstream topological relationships of the fault event propagation graph, and output the fault boundary location, risk level, and maintenance priority after execution validity verification.

[0013] Furthermore, step S1 specifically involves: collecting the undervoltage event marker, revoltage event marker, current interruption event marker, event timestamp, event duration, and voltage and current of each meter position, wherein the event duration includes the power outage event timestamp and the power restoration event timestamp; and simultaneously collecting the voltage status, current status, phase undervoltage marker, and zero-sequence current information of each phase bus at the box level.

[0014] Furthermore, step S2 specifically involves: aligning the box-level events and each tabletop event occurring within the same time alignment window according to their timestamps, and clustering them into candidate fault event clusters; for undervoltage, overvoltage restoration, current interruption, box-level phase undervoltage, and zero-sequence abnormality events falling within the same time alignment window, merging them into the same candidate fault event cluster; wherein, based on the event type, the event timestamp used for synchronization calculation is selected, specifically: when a power outage event exists, the power outage event timestamp is taken first; when there is no power outage event but a power restoration event exists, the power restoration event timestamp is taken; for events with short-term sampling delays or uploading time differences, they are aligned and corrected using a preset time tolerance. This step achieves a process of uniformly grouping scattered tabletop events into the same candidate fault.

[0015] Furthermore, in step S3, a fault event propagation diagram is constructed from the incoming node, each phase bus node, branch node, and meter position node according to the meter box topology. For each meter position, an event propagation edge is established from the incoming side and bus side of the box to the meter position side, based on its phase and branch affiliation. The direction of the event propagation edge is uniformly defined as pointing from the upstream power supply node to the downstream power consumption node, that is, the incoming node is upstream of the bus node, the bus node is upstream of the branch node, and the branch node is upstream of the meter position node, which is used to represent the propagation of the fault impact downstream along the topology. For any directed edge (u,v), the expected order of the recovery event is that the timestamp of the power restoration event of the upstream node u is not later than the timestamp of the power restoration event of the downstream node v.

[0016] Furthermore, in step S4, let... Let b be the set of theoretically affected epitopes corresponding to the candidate fault boundary. For each candidate fault event cluster, there is a set of actual event table positions. Calculate the event synchronization index separately. Topology coverage index , recovery order consistency index Coupling index with zero sequence The Used to measure the degree of synchronization of power outage or restoration times for multiple meter positions within the same fault process, if If it is an empty set, then ; Used to measure the degree of matching between the set of epitopes that actually occur and the set of epitopes affected by the candidate boundary theory; Used to measure whether the timestamp of a power restoration event conforms to the topological constraint in the fault event propagation graph that the upstream node recovers no later than the downstream node; It is used to couple the effective value of the zero-sequence current at the box level and the three-phase voltage imbalance in a weighted manner to enhance the ability to identify neutral line anomalies and asymmetrical faults.

[0017] Furthermore, the candidate boundary score is calculated using the following formula:

[0018]

[0019] in to The weight coefficients are non-negative and satisfy the following conditions: .

[0020] Furthermore, in step S5, if the maximum candidate boundary score is not lower than a preset confidence threshold... If the candidate fault boundary score is the highest, the candidate fault boundary with the highest score is directly output as the initial judgment result. For low-confidence candidate fault event clusters, the candidate boundary scores are converted into fixed-dimensional statistical features, which are then input into the lightweight fault boundary localization model. These fixed-dimensional statistical features include the highest candidate boundary score. Second largest candidate boundary score The candidate boundary score mean, candidate boundary score variance, number of epitopes of events occurring within an event cluster, event duration, effective value of zero-sequence current at the box level, three-phase voltage imbalance, and number of comparable node pairs for recovery order.

[0021] Furthermore, in step S6, the method for determining the set of affected positions is as follows: for incoming line faults, all positions within the same candidate fault event cluster that experience power outages or restorations are identified as affected positions; for busbar faults, positions that experience events in the corresponding phase or busbar section are identified as affected positions; for single-branch faults, positions that experience events downstream of the corresponding branch are identified as affected positions; for single-position faults, the single position where the event occurs is identified as affected; for neutral line anomalies, positions exhibiting zero-sequence current anomalies, voltage imbalances, or power factor anomalies are identified as affected positions.

[0022] Furthermore, in step S6, the validity verification specifically involves: removing meter positions that do not fall within the time alignment window and do not have abnormal electrical parameters; removing meter positions whose event timestamps do not fall within the time alignment window and do not have abnormal voltage, current interruption, zero-sequence current, or power factor from the affected meter position set; for meter positions located downstream of a known fault boundary but which have not formed a meter position event record due to communication delay or missing event transmission, if there is a corresponding box-level abnormality in their bus or branch, then adding the meter position to the affected meter position set; and outputting the verified affected meter position set according to the meter position number, phase, branch number, and fault boundary category.

[0023] Furthermore, in step S6, the risk level and maintenance priority are calculated and output using the following formula:

[0024]

[0025] in To ultimately output the model probability or fault confidence corresponding to the fault boundary category, For the duration of the event, Based on duration, , and The weights are non-negative and satisfy the following conditions: ; Score the maximum candidate boundary;

[0026] according to Output low-risk, medium-risk, and high-risk levels and maintenance priorities.

[0027] Beneficial Effects: This invention does not simply rely on a single power-on / off event or fixed rules for fault location. Instead, it integrates time-aligned event clusters, the theoretically affected set of tabletops at candidate boundaries, a directed fault event propagation graph, and power-on timestamp order constraints into a single scoring process. Compared to schemes that only utilize event clustering or only utilize topological range matching, this invention achieves this by processing the intersection of the theoretically affected tabletop set corresponding to the candidate fault boundary with the actual event tabletop set in the candidate fault event cluster to form an empty set, based on... The recovery sequence consistency calculation, weighted zero-order coupling calculation, and fixed-dimensional low-confidence model input ensure that each boundary feature has a clear data source and an executable calculation process. At the same time, without adding dedicated positioning hardware, the fault boundary category, the set of affected positions, and the maintenance priority are linked and output, which helps to narrow the scope of on-site investigation, reduce the risk of mislocation, and improve the stability of engineering deployment. Attached Figure Description

[0028] Figure 1 This is a main flowchart of a fault boundary location method for a low-voltage power metering box according to an embodiment of the present invention;

[0029] Figure 2 This is a schematic diagram of the fault event propagation in an embodiment of the present invention;

[0030] Figure 3 This is a schematic diagram of the alignment of power-on / off events for multiple positions according to an embodiment of the present invention;

[0031] Figure 4 This is a schematic diagram of the fault boundary location results according to an embodiment of the present invention;

[0032] Figure 5 This is a schematic diagram illustrating the identification of the affected epitope set according to an embodiment of the present invention;

[0033] Figure 6 This is a schematic diagram comparing the fault boundary location effect of the method of the present invention with other different methods in an embodiment of the present invention. Detailed Implementation

[0034] To make the objectives, technical solutions, and technical effects of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0035] like Figure 1As shown in this embodiment, a fault boundary location method for a low-voltage energy metering box can be applied to a low-voltage energy metering box or energy metering box monitoring terminal that has the functions of meter position electrical parameter acquisition, box-level bus status acquisition, and power outage / on-off event recording. The method includes the following steps:

[0036] S1. Multi-position event and electrical parameter acquisition: Collect voltage, current, zero-sequence current and power-on / off event information for each position and the box-level bus.

[0037] Specifically, single-phase multi-meter low-voltage energy metering boxes and three-phase multi-meter low-voltage energy metering boxes were selected as the implementation targets.

[0038] The system collects the undervoltage event markers, revoltage event markers, current interruption event markers, event timestamps, event durations, and voltage and current at each meter position. The event duration includes both the outage and re-energization event timestamps. Simultaneously, it collects the voltage and current states, phase undervoltage markers, and zero-sequence current information for the A, B, and C phase buses at the box level. This allows for the simultaneous acquisition of the event evolution process at both the meter position and box level, providing multi-source input for subsequent fault boundary location.

[0039] S2. Event alignment and fault event cluster generation: The power outage and power on event information is clustered according to a preset time alignment window to form candidate fault event clusters.

[0040] Specifically, the metering box has a rated operating voltage of 220 / 380V and a frequency of 50Hz. The event sampling period is preferably 1 second to 1 minute. The time alignment window is determined based on the time resolution of the acquisition system and the propagation speed of on-site events, with 30 seconds being the preferred value.

[0041] like Figure 3 As shown, box-level events and table-level events occurring within the same time window are aligned by timestamp and clustered into candidate fault event clusters. Among them, undervoltage, overvoltage, current interruption, box-level phase undervoltage, and zero-sequence abnormality events falling into the same time alignment window are grouped into the same candidate fault event cluster.

[0042] The event timestamp used for synchronization calculation can be selected according to the event type: when there is a power outage event, the power outage event timestamp is taken first; when there is no power outage event but there is a power restoration event, the power restoration event timestamp is taken.

[0043] For events with short-term sampling delays or uploading time differences, a preset time tolerance is used. Perform alignment correction.

[0044] Therefore, scattered surface events can be grouped into the same candidate fault process, avoiding isolated judgments based on a single alarm point.

[0045] S3. Construction of topology propagation diagram: Construct a fault event propagation diagram based on the phase, branch affiliation, and meter box topology of each meter position, and establish candidate fault boundaries for incoming line side faults, busbar side faults, single branch faults, single meter position faults, and neutral line anomalies.

[0046] Specifically, such as Figure 2 As shown, based on the metering box topology, a fault event propagation diagram is constructed from the incoming node, A-phase bus node, B-phase bus node, C-phase bus node, branch node, and meter position node. For each meter position, an event propagation edge is established from the box-level incoming side, bus side, to the meter position side, based on its phase and branch affiliation. The direction of the event propagation edge is uniformly defined as from the upstream power supply node to the downstream power consumption node; that is, the incoming node is upstream of the bus node, the bus node is upstream of the branch node, and the branch node is upstream of the meter position node, used to represent the propagation of fault effects downstream along the topology. For any directed edge (u,v), the expected order of recovery events is that the timestamp of the power restoration event at the upstream node u is no later than the timestamp of the power restoration event at the downstream node v.

[0047] Therefore, the topological hierarchy can be used to describe the influence range, propagation path, and recovery sequence constraints corresponding to different fault boundaries.

[0048] S4. Boundary Feature Calculation: For each candidate fault event cluster and candidate fault boundary, calculate the event synchronization index, topology coverage index, recovery order consistency index, and zero-order coupling index, and calculate the candidate boundary score based on the above indexes.

[0049] Specifically, let's set Let b be the set of theoretically affected epitopes corresponding to the candidate fault boundary. For each candidate fault event cluster, there is a set of actual event table positions. Calculate the event synchronization index separately. Topology coverage index , recovery order consistency index Coupling index with zero sequence .in, Used to measure the degree of synchronization of power outage or restoration times at multiple meter positions within the same fault process; Used to measure the degree of matching between the set of epitopes that actually occur and the set of epitopes affected by the candidate boundary theory; Used to measure whether the timestamp of a power restoration event conforms to the topological constraint in the fault event propagation graph that the upstream node recovers no later than the downstream node; Used to convert the effective value of zero-sequence current at the box level and the three-phase voltage imbalance. Weighted coupling is used to enhance the ability to identify neutral line anomalies and asymmetric faults. The main calculation formula is as follows:

[0050]

[0051]

[0052]

[0053]

[0054]

[0055] in, Let be the event timestamp used for synchronization calculation at the i-th bit; if a power outage event occurs at the i-th bit, then Take the timestamp of the power outage event. If there is no power outage event but a power restoration event, then... Retrieve the timestamp of the power restoration event; if If it is an empty set, then Within the theoretical influence subgraph corresponding to candidate fault boundary b, select directed edge node pairs (u,v) that simultaneously have the timestamp of the power restoration event to form a set of comparable node pairs. , where u is the upstream node and v is the downstream node; , To meet The number of node pairs, For event acquisition time tolerance; if ,but . and These are the timestamps of the power restoration events for upstream node u and downstream node v, respectively. This represents the effective value of the zero-sequence current at the box level. The zero-sequence current reference value is determined by the rated permissible zero-sequence current of the normally operating sample or metering box; For time-aligned windows; and The weights are non-negative and satisfy the following conditions: ; This represents the average voltage of phases A, B, and C. To prevent tiny positive numbers with a denominator of zero; Indicates will Limited to between 0 and 1.

[0056] Furthermore, the boundary score of any candidate fault event cluster corresponding to the candidate fault boundary b is calculated using the following formula:

[0057]

[0058] in to The weight coefficients are non-negative and satisfy the following conditions: In one embodiment, it is preferable to The weighting coefficients are jointly determined by the power outage test data of the physical prototype, on-site fault records, and manually verified closed-loop data, and are calibrated by grid search, least squares fitting, or maximizing the boundary positioning accuracy on the validation set.

[0059] S5. Fault Boundary Localization and Identification: Based on the candidate boundary scores, a lightweight fault boundary localization model is used to perform fault boundary category identification. Figure 4 The diagram shows the results of fault boundary location.

[0060] Specifically, if the maximum candidate boundary score is not lower than the preset confidence threshold... If the candidate fault boundary with the highest score is selected, it will be directly output as the initial judgment result. A pre-set threshold value is used. The boundary scores are determined based on the statistical distribution of known boundary samples and non-corresponding boundary samples on the validation set. The value ranges from 0.65 to 0.85.

[0061] For low-confidence candidate fault event clusters, instead of directly using all score vectors whose dimensions change with the number of candidate boundaries, the candidate boundary scores are converted into fixed-dimensional statistical features, including the maximum candidate boundary score. Second largest candidate boundary score The candidate boundary score mean, candidate boundary score variance, number of epitopes of events occurring within an event cluster, event duration, effective value of zero-sequence current at the box level, three-phase voltage imbalance, and number of comparable node pairs for recovery order.

[0062] The aforementioned fixed-dimensional statistical features are input into the lightweight fault boundary localization model. The model preferably adopts a lightweight gradient boosting classification model, which can be composed of 60 to 100 CART classification trees. The maximum depth of each tree is 2 to 4, the learning rate is 0.02 to 0.10, the minimum number of leaf node samples is 3 to 10, and the model storage occupancy is no more than 256kB. In this embodiment, there are 80 CART classification trees, the maximum depth is 3, the learning rate is 0.05, and the minimum number of leaf node samples is 5.

[0063] The labels for the fault boundary categories are jointly determined by standard power supply samples, manually set incoming side power outage samples, manually set bus section fault samples, manually set single branch power outage samples, manually set single meter position abnormality samples, neutral line abnormality samples, and historical fault samples that have been manually verified on-site.

[0064] S6. Affected meter location maintenance suggestion output: Based on the fault boundary category and the upstream and downstream topology of the fault event propagation diagram, determine the set of affected meter locations, output the fault boundary location, risk level and maintenance priority, thereby realizing automatic location of fault boundaries of low-voltage power metering boxes and operation and maintenance assistance.

[0065] Specifically, for incoming line faults, all positions within the same candidate fault event cluster that experience power outages or restorations are identified as affected positions; for busbar faults, positions where events occur in the corresponding phase or busbar section are identified as affected positions; for single-branch faults, positions where events occur downstream of the corresponding branch are identified as affected positions; for single-position faults, the single position where the event occurs is identified as affected; and for neutral line anomalies, positions exhibiting zero-sequence current anomalies, voltage imbalances, or power factor anomalies are identified as affected positions.

[0066] After determining the set of affected meter positions, a validity check is then performed: meter positions that do not fall within the time alignment window and do not have abnormal electrical parameters are removed; meter positions whose event timestamps do not fall within the time alignment window and do not have abnormal voltage, current interruption, zero-sequence current, or power factor are removed from the set of affected meter positions; for meter positions located downstream of a determined fault boundary but which have not formed a meter position event record due to communication delay or missing event transmission, if there is a corresponding box-level abnormality in their bus or branch, the meter position is added to the set of affected meter positions; and the checked set of affected meter positions is output according to the meter position number, phase, branch number, and fault boundary category.

[0067] The risk level and maintenance priority are calculated and output using the following formula:

[0068]

[0069] in To ultimately output the model probability or fault confidence corresponding to the fault boundary category, For the duration of the event, Based on duration, , and The weights are non-negative and satisfy the following conditions: ;according to Output low-risk, medium-risk, and high-risk levels and maintenance priorities. Figure 5 A schematic diagram illustrating the identification of the affected epitope set is shown.

[0070] In this embodiment, the method of the present invention is trained and verified using physical prototype power outage and restoration test data, on-site fault records, and manually verified closed-loop data. Normal samples are collected under standard power supply and normal recovery conditions; abnormal samples are collected under manually set conditions such as incoming line power outage, busbar section fault, single branch power outage, single meter position anomaly, and neutral line anomaly. During the verification phase, indicators such as boundary location accuracy, affected meter position set recall rate, false location rate, average location time, secondary identification accuracy of low-confidence samples, and model storage usage are statistically analyzed to verify the comprehensive effectiveness of event alignment, boundary scoring, fixed-dimensional model input, and affected meter position set verification. This verification process is used to illustrate the implementation effect of the present invention and does not change the scope of protection of the aforementioned method steps. (Reference) Figure 6 As shown in the diagram, the fault boundary location effect of the method of the present invention is compared with that of other different methods. The method of the present invention has the highest boundary location accuracy and maintains the lowest mislocation rate.

[0071] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention in any way. Although the implementation process of the present invention has been described in detail above, those skilled in the art can still modify the technical solutions described in the foregoing examples or make equivalent substitutions for some of the technical features. All modifications and equivalent substitutions 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 locating fault boundaries in a low-voltage power metering box, characterized in that, Includes the following steps: S1. Collect electrical parameters and power outage / on / off event information for each meter position and the box-level busbar; S2. Cluster the power outage and power on event information according to a preset time alignment window to form candidate fault event clusters; S3. Construct a fault event propagation diagram based on the phase, branch affiliation, and meter box topology of each meter position, and establish candidate fault boundaries for incoming side faults, bus side faults, single branch faults, single meter position faults, and neutral line anomalies of the meter box. S4. For each candidate fault event cluster and candidate fault boundary, calculate the event synchronization index, topology coverage index, recovery order consistency index, and zero-order coupling index, and calculate the candidate boundary score based on the above indexes. S5. Based on the candidate boundary scores, a lightweight fault boundary localization model is used to perform fault boundary category identification; S6. Determine the set of affected tabletops based on the identified fault boundary categories and the upstream and downstream topological relationships of the fault event propagation graph, and output the fault boundary location, risk level, and maintenance priority after execution validity verification.

2. The fault boundary location method for a low-voltage power metering box as described in claim 1, characterized in that, Specifically, step S1 collects: the undervoltage event marker, the revoltage event marker, the current interruption event marker, the event timestamp, the event duration, and the voltage and current of each meter position, where the event duration includes the power outage event timestamp and the power restoration event timestamp; and collects the voltage status, current status, phase undervoltage marker, and zero-sequence current information of the A-phase, B-phase, and C-phase busbars at the box level.

3. The fault boundary location method for a low-voltage power metering box as described in claim 2, characterized in that, Step S2 specifically involves: aligning the box-level events and each tabletop event occurring within the same time alignment window according to their timestamps, and clustering them into candidate fault event clusters; for undervoltage, overvoltage restoration, current interruption, box-level phase undervoltage, and zero-sequence abnormality events falling within the same time alignment window, merging them into the same candidate fault event cluster; wherein, according to the event type, the event timestamp used for synchronization calculation is selected, specifically: when there is a power outage event, the power outage event timestamp is taken first, and when there is no power outage event but there is a power restoration event, the power restoration event timestamp is taken; for events with short-term sampling delays or uploading time differences, they are aligned and corrected using a preset time tolerance.

4. The fault boundary location method for a low-voltage power metering box as described in claim 2, characterized in that, In step S3, a fault event propagation diagram is constructed based on the metering box topology, consisting of the incoming node, each phase bus node, branch node, and meter position node. For each meter position, an event propagation edge is established from the box-level incoming side, bus side, to the meter position side, based on its phase and branch affiliation. The direction of the event propagation edge is uniformly defined as pointing from the upstream power supply node to the downstream power consumption node, i.e., the incoming node is upstream of the bus node, the bus node is upstream of the branch node, and the branch node is upstream of the meter position node, which is used to represent the propagation of fault effects downstream along the topology. For any directed edge (u,v), the expected order of recovery events is that the timestamp of the power restoration event of the upstream node u is no later than the timestamp of the power restoration event of the downstream node v.

5. The fault boundary location method for a low-voltage power metering box as described in claim 2, characterized in that, In step S4, let Let b be the set of theoretically affected epitopes corresponding to the candidate fault boundary. Given the set of actual event table positions in the candidate fault event cluster, obtain the set of intersections between them. Calculate the event synchronization index separately. Topology coverage index , recovery order consistency index Coupling index with zero sequence ; The Used to measure the degree of synchronization of power outage or restoration times for multiple meter positions within the same fault process, if If it is an empty set, then ; Used to measure the degree of matching between the set of epitopes that actually occur and the set of epitopes that are theoretically affected; Used to measure whether the timestamp of a power restoration event conforms to the topological constraint in the fault event propagation graph that the upstream node recovers no later than the downstream node; It is used to couple the effective value of the zero-sequence current at the tank level and the three-phase voltage imbalance at the tank level in a weighted manner to enhance the ability to identify neutral line anomalies and asymmetrical faults.

6. The fault boundary location method for a low-voltage power metering box as described in claim 5, characterized in that, The candidate boundary score is calculated using the following formula: ， in to The weight coefficients are non-negative and satisfy the following conditions: .

7. The fault boundary location method for a low-voltage power metering box as described in claim 1, characterized in that, In step S5, if the maximum candidate boundary score is not lower than the preset confidence threshold... If the candidate fault boundary score is the highest, the candidate fault boundary with the highest score is directly output as the initial judgment result. For candidate fault event clusters with low confidence, the candidate boundary scores are converted into fixed-dimensional statistical features, and these fixed-dimensional statistical features are input into the lightweight fault boundary localization model. The fixed-dimensional statistical features include the highest candidate boundary score. Second largest candidate boundary score The candidate boundary score mean, candidate boundary score variance, number of epitopes of events occurring within an event cluster, event duration, effective value of zero-sequence current at the box level, three-phase voltage imbalance, and number of comparable node pairs for recovery order.

8. The fault boundary location method for a low-voltage power metering box as described in claim 1, characterized in that, In step S6, the method for determining the set of affected positions is as follows: For incoming line faults, all positions within the same candidate fault event cluster that experience power outages or restorations are identified as affected positions; for busbar faults, positions that experience events in the corresponding phase or busbar section are identified as affected positions; for single-branch faults, positions that experience events downstream of the corresponding branch are identified as affected positions; for single-position faults, the single position where the event occurs is identified as affected; for neutral line anomalies, positions exhibiting zero-sequence current anomalies, voltage imbalances, or power factor anomalies are identified as affected positions.

9. The fault boundary location method for a low-voltage power metering box as described in claim 1, characterized in that, In step S6, the validity verification specifically involves: removing meter positions that do not fall within the time alignment window and do not have abnormal electrical parameters; removing meter positions whose event timestamps do not fall within the time alignment window and do not have abnormal voltage, current interruption, zero-sequence current, or power factor from the affected meter position set; for meter positions located downstream of a known fault boundary but which have not formed a meter position event record due to communication delay or missing event transmission, if there is a corresponding box-level abnormality in their bus or branch, then adding the meter position to the affected meter position set; and outputting the verified affected meter position set according to the meter position number, phase, branch number, and fault boundary category.

10. The fault boundary location method for a low-voltage power metering box as described in claim 1, characterized in that, In step S6, the risk level and maintenance priority are calculated and output using the following formula: ， in To ultimately output the model probability or fault confidence corresponding to the fault boundary category, For the duration of the event, Based on duration, , and The weights are non-negative and satisfy the following conditions: ; Score the maximum candidate boundary; according to Output low-risk, medium-risk, and high-risk levels and maintenance priorities.

Images