An event primary key driven multi-source power failure data sample standardization construction and spatiotemporal consistency feature extraction method, device and equipment and readable storage medium

Abstract

This invention relates to the field of power technology, specifically to a method, apparatus, device, and readable storage medium for the standardized construction of multi-source power outage data samples and the extraction of spatiotemporal consistency features based on event primary keys. The method includes: collecting data; determining the power outage start point and power restoration end point of candidate power outage events; constructing a dynamic event window based on the power outage start point and power restoration end point; extracting multi-source evidence fragments related to the candidate event within the event window; clustering and merging duplicate events corresponding to the multi-source evidence fragments based on time overlap, impact range overlap, and power supply path similarity, and generating a unified event primary key ID for the merged events; performing power supply path backtracking based on the topological snapshot at the time of the power outage to obtain the theoretical power outage impact range, and comparing it with the actual recorded range to complete topological consistency verification and inconsistency cause identification.

Description

Technical Field

[0001] This invention relates to a method, apparatus, device, and readable storage medium for the standardized construction of multi-source power outage data samples driven by event primary keys and the extraction of spatiotemporal consistency features, belonging to the field of power technology. Background Technology

[0002] The Power Outage Management System (OMS) primarily records power outage plans, the start time of power outage, and the end time of power restoration; the Distribution Automation System records fault alarms, switch actions, reclosing, and isolation recovery processes; the Advanced Measurement System (AMI) records voltage, current, and undervoltage and power restoration curves on the user side or transformer substation side; the Work Order System records the manual dispatch, emergency repair, power transfer, and work order cancellation processes; in addition, Geographic Information Systems (GIS), equipment ledgers, topology models, and meteorological data provide support for power outage cause analysis and impact range assessment.

[0003] However, most existing power outage data analysis solutions are designed for single systems or single application scenarios. They commonly suffer from the problem of the same physical power outage event being repeatedly reported, split into records, or identified using different methods in OMS, AMI, distribution automation systems, and work order systems. This results in the same event being fragmented into multiple scattered record segments, making it impossible to form a stable primary key for the sample. Furthermore, the different sampling frequencies and timing mechanisms of each system lead to both overall timestamp drift and local misordering, making it difficult to reconstruct the power outage chain on a unified timeline. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a method, apparatus, device, and readable storage medium for the standardized construction of multi-source power outage data samples and the extraction of spatiotemporal consistency features based on event primary keys, aiming to solve the above problems.

[0005] The technical solution of this invention to solve the above-mentioned technical problems is as follows: A method for standardized construction and spatiotemporal consistency feature extraction of multi-source power outage data samples driven by event primary keys, comprising the following steps: Data is collected from the Power Outage Management System (OMS), Distribution Automation System (AMI), Work Order System, and Topology Ledger to determine the potential power outage initiation points for power outage events. and the end point of power restoration A dynamic event window is constructed based on the power outage start point and power restoration end point, and multi-source evidence fragments related to the event candidate are extracted within the event window; Based on time overlap, impact range overlap and power supply path similarity, the duplicate events corresponding to multi-source evidence fragments are clustered and merged, and a unified event primary key ID is generated for the merged events. Using the unified event primary key as an index, a logical sequence constraint model for alarms, actions, power outages, work orders, and power restoration is established. Overall drift correction is performed on the timestamps of each evidence source, and missing data detection, anomaly detection, transient identification, and imputation annotation are performed on the corrected evidence data. Power supply path backtracking is performed based on the topology snapshot at the time of power outage to obtain the theoretical power outage impact range, and then compared with the actual recorded range to complete the topology consistency verification and inconsistency cause identification; Temporal features, spatial features, and recovery process features are extracted from the corrected and verified event samples to form a feature dictionary; The sample quality score Q is calculated based on the evidence sufficiency (E), topological consistency (C), and missing rate (M). The samples are then stratified into high-confidence, medium-confidence, or low-confidence databases according to the quality score Q. Version numbers and change logs are maintained for revised samples. The samples are stored in the database according to a four-table structure: event table, evidence table, feature table, and quality table.

[0006] In a preferred embodiment of this application, the dynamic event window Represented as: ; in, Forward extension duration of power loss The duration of backward extension after power restoration; and The quantiles are adaptively determined based on the historical latency distribution of different data sources or obtained through training with historical labeled samples.

[0007] In a preferred embodiment of this application, the comprehensive similarity S satisfies: ; in The proportion of time overlap. To determine the overlap ratio of the affected areas, For power supply path similarity, , , The weighting coefficients and When S is not less than the preset threshold When the threshold is reached, the corresponding records are merged into the same event cluster and assigned a unified event primary key ID; The threshold is determined by the optimal F1 value on the historical labeled samples.

[0008] In a preferred embodiment of this application, the cutoff voltage calculation step includes: The net load sequence corresponding to the optimized scheduling result is input into the established net load-gate voltage mapping model. The gate voltage for each time period is calculated. Based on the Gaussian process regression model, the predicted distribution of the gate voltage is obtained. The following formula is used for calculation: In the formula, for Predicted average voltage at time-bound thresholds To predict variance.

[0009] In a preferred embodiment of this application, the anomaly detection includes: detecting anomalies whose duration is less than a transient threshold. Trip alarms are identified as transient records and removed or downgraded in weight; values ​​exceeding preset engineering boundaries are identified as outliers; the missing data processing includes assigning missing labels to missing channels, using linear interpolation, nearest-neighbor event interpolation, or interpolation based on historical curves of similar equipment for interpolable data, and recording the interpolation identifier in the evidence table.

[0010] In a preferred embodiment of this application, the topology consistency index satisfy: ; in This is the theoretical set of power outage impacts obtained from topological snapshots and switch state backtracking at the moment of the power outage. For the actual set of recorded effects; when Less than the threshold At that time, output inconsistency reason labels and revision candidates, and feed the revision results back to the quality scoring and version evolution process.

[0011] This application provides a device for standardizing and constructing multi-source power outage data samples and extracting spatiotemporal consistency features based on event primary keys, including: The feature selection module is used to collect data from the power outage management system (OMS), distribution automation system, AMI, work order system, and topology ledger to determine the power outage initiation points of candidate power outage events. and the end point of power restoration A dynamic event window is constructed based on the power outage start point and power restoration end point, and multi-source evidence fragments related to the event candidate are extracted within the event window; The merging module is used to cluster and merge duplicate events corresponding to multi-source evidence fragments based on time overlap, impact range overlap, and power supply path similarity, and generate a unified event primary key ID for the merged events. The detection module is used to establish a logical sequence constraint model of alarms, actions, undervoltage, work orders and power restoration using a unified event primary key as an index, perform overall drift correction on the timestamps of each evidence source, and perform missing detection, anomaly detection, transient identification and imputation labeling on the corrected evidence data. The comparison module is used to perform power supply path backtracking based on the topology snapshot at the time of power outage, obtain the theoretical power outage impact range, and compare it with the actual recorded range to complete the topology consistency verification and the determination of the cause of inconsistency; The dictionary module is used to extract temporal features, spatial features, and recovery process features from the corrected and verified event samples to form a feature dictionary; The storage module is used to calculate the sample quality score Q based on the sufficiency of evidence E, topological consistency C, and missing rate M. The samples are then stratified into high-confidence, medium-confidence, or low-confidence databases according to the quality score Q. The module also maintains version numbers and change logs for revised samples and stores the samples in the database according to a four-table structure: event table, evidence table, feature table, and quality table.

[0012] In a preferred embodiment of this application, the dynamic event window Represented as: ; in, Forward extension duration of power loss The duration of backward extension after power restoration; and The quantiles are adaptively determined based on the historical latency distribution of different data sources or obtained through training with historical labeled samples.

[0013] Furthermore, to achieve the above objectives, the present invention also provides a device for standardizing and extracting spatiotemporal consistency features of multi-source power outage data samples driven by event primary keys. This device includes a processor, a memory, and a program for standardizing and extracting spatiotemporal consistency features of multi-source power outage data samples driven by event primary keys, stored in the memory and executable by the processor. When the program is executed by the processor, it implements the steps of the aforementioned method for standardizing and extracting spatiotemporal consistency features of multi-source power outage data samples driven by event primary keys.

[0014] Furthermore, to achieve the above objectives, the present invention also provides a computer-readable storage medium storing a program for the standardized construction and spatiotemporal consistency feature extraction of multi-source power outage data samples driven by event primary keys. When the program is executed by a processor, it implements the steps of the method described above for the standardized construction and spatiotemporal consistency feature extraction of multi-source power outage data samples driven by event primary keys.

[0015] The beneficial effects of this invention are: by collecting data from the power outage management system (OMS), distribution automation system, AMI, work order system, and topology ledger, the potential power outage initiation points of power outage events can be determined. and the end point of power restoration A dynamic event window is constructed based on the power outage start point and power restoration end point, and multi-source evidence fragments related to the event candidate are extracted within the event window; Based on temporal overlap, overlapping impact range, and power supply path similarity, repetitive events corresponding to multi-source evidence fragments are clustered and merged, and a unified event primary key ID is generated for the merged events. Using the unified event primary key as an index, a logical sequence constraint model for alarms, actions, power outages, work orders, and power restoration is established. Overall drift correction is performed on the timestamps of each evidence source, and the corrected evidence data is processed for missing data detection, anomaly detection, transient identification, and imputation annotation. Power supply path backtracking is performed based on the topological snapshot at the time of power outage to obtain the theoretical power outage impact range, which is compared with the actual recorded range to complete topological consistency verification and inconsistency cause identification. Temporal features, spatial features, and recovery process features are extracted from the corrected and verified event samples to form a feature dictionary. The sample quality score Q is calculated based on evidence sufficiency E, topological consistency C, and missing rate M. The samples are stratified into high-confidence, medium-confidence, or low-confidence databases according to the quality score Q, and the version number and change log are maintained for revised samples. The samples are stored in the database according to a four-table structure: event table, evidence table, feature table, and quality table. Compared with existing technologies, this invention has at least the following beneficial effects: It jointly identifies cross-source candidate records by considering time overlap, impact range overlap, and power supply path similarity, and uses a unified event primary key to connect the event table, evidence table, feature table, and quality table, achieving a unique cross-system representation of the same physical power outage event; by establishing logical order constraints, it performs overall drift correction on the timestamps of each evidence source, achieving reconfigurability of the power outage process on a unified timeline and improving the temporal consistency of cross-source evidence; by using a topological snapshot of the power outage moment to trace back the theoretical impact range and comparing it with the actual recorded range in a closed loop, it can identify inconsistencies such as delayed ledger updates, abnormal switch status, missing AMI data, and work order positioning deviations, reducing sample label contamination; by jointly extracting temporal features, spatial features, and recovery process features, and by jointly scoring evidence sufficiency, topological consistency, and missing rate, it achieves hierarchical management of sample quality, facilitating the direct use of high-confidence samples for model training, while medium- and low-confidence samples enter the review and revision process; by maintaining the revision history through version numbers and change logs, it ensures that power outage samples retain complete traceability after multiple rounds of automatic revision and manual review. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the hardware structure of a device for standardizing and extracting spatiotemporal consistency features of multi-source power outage data samples based on event primary key driven by the present invention. Figure 2 This is a flowchart illustrating a method for standardizing and extracting spatiotemporal consistency features of multi-source power outage data samples based on event primary keys, as described in this invention. Figure 3 This is a schematic diagram of the four-table structure and relationships of the event samples; Figure 4 This is a schematic diagram illustrating multi-source candidate merging and event primary key generation. Figure 5 This is a schematic diagram of the power supply path tracing. Figure 6 This is a schematic diagram for topology consistency verification; Figure 7 This is a schematic diagram illustrating sample quality scoring and stratified data entry. Figure 8 A schematic diagram illustrating the time alignment of multi-source evidence and the construction of event windows; Figure 9 A schematic diagram illustrating the merging of duplicate events and the generation of event primary keys; Figure 10 This diagram illustrates the identification of causes of topological inconsistency and the generation of revision candidates.

[0017] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0018] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0019] The present invention relates to a method for standardizing and extracting spatiotemporal consistency features of multi-source power outage data samples driven by event primary keys. This method is mainly applied to a device for standardizing and extracting spatiotemporal consistency features of multi-source power outage data samples driven by event primary keys. This device can be a PC, a portable computer, a mobile terminal, or other device with display and processing functions.

[0020] Reference Figure 1 , Figure 1This is a schematic diagram of the hardware structure of a device for standardizing and extracting spatiotemporal consistency features of multi-source power outage data samples based on event primary keys, as described in an embodiment of the present invention. In this embodiment, the device may include a processor 1001 (e.g., CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. The communication bus 1002 is used to establish communication between these components; the user interface 1003 may include a display screen or an input unit such as a keyboard; the network interface 1004 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface); the memory 1005 may be a high-speed RAM or a stable, non-volatile memory, such as a disk drive, and optionally, the memory 1005 may be a storage device independent of the aforementioned processor 1001.

[0021] Those skilled in the art will understand that Figure 1 The hardware structure shown does not constitute a limitation on a device for standardizing the construction of multi-source power outage data samples and extracting spatiotemporal consistency features based on event primary key driven data. It may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0022] Continue to refer to Figure 1 , Figure 1 The memory 1005, which is a computer-readable storage medium, may include an operating system, a network communication module, and a program for standardizing and extracting spatiotemporal consistency features of multi-source power outage data samples based on event primary keys.

[0023] exist Figure 1 In this embodiment, the network communication module is mainly used to connect to the server and communicate with the server for data. The processor 1001 can call a program stored in the memory 1005 for the standardized construction and spatiotemporal consistency feature extraction of multi-source power outage data samples based on event primary key, and execute the method for standardized construction and spatiotemporal consistency feature extraction of multi-source power outage data samples based on event primary key provided by this invention.

[0024] This invention provides a method for standardized construction of multi-source power outage data samples and extraction of spatiotemporal consistency features based on event primary keys, referencing... Figure 2 The steps include: Step S1: Collect data from the Power Outage Management System (OMS), Distribution Automation System (AMI), Work Order System, and Topology Ledger to comprehensively determine the candidate power outage initiation points for power outage events. and the end point of power restoration ; Step S2: Construct a dynamic event window based on the power outage start point and power restoration end point, and extract multi-source evidence fragments related to the event candidate within the event window; Step S3: Cluster and merge the duplicate events corresponding to the multi-source evidence fragments based on time overlap, influence range overlap, and power supply path similarity, and generate a unified event primary key ID for the merged events; Step S4: Using the unified event primary key as the index, establish a logical sequence constraint model of alarm-action-loss-work order-power restoration, and perform overall drift correction on the timestamps of each evidence source; Step S5: Perform missing data detection, anomaly detection, transient identification, and imputation annotation on the corrected evidence data; Step S6: Based on the topology snapshot at the time of the power outage, perform power supply path backtracking to obtain the theoretical power outage impact range, and compare it with the actual recorded range to complete the topology consistency check and inconsistency cause identification; Step S7: Extract temporal features, spatial features, and recovery process features from the corrected and verified event samples to form a feature dictionary; Step S8: Calculate the sample quality score Q based on evidence sufficiency E, topological consistency C, and missing rate M; Step S9: Based on the quality score Q, stratify the samples into high confidence, medium confidence, or low confidence pools, and maintain version numbers and change logs for revised samples; Step S10: Store the samples in the database according to the four-table structure of event table, evidence table, feature table and quality table.

[0025] In this embodiment, cross-source candidate records are jointly identified by time overlap, impact range overlap, and power supply path similarity. A unified event primary key is used to connect the event table, evidence table, feature table, and quality table to achieve a unique cross-system representation of the same physical power outage event. By establishing logical order constraints, overall drift correction is applied to the timestamps of each evidence source to achieve the reconstructability of the power outage process on a unified time axis and improve the temporal consistency of cross-source evidence. By backtracking the theoretical impact range through the topological snapshot at the time of the power outage and comparing it with the actual recorded range in a closed loop, inconsistencies such as delayed ledger updates, abnormal switch status, missing AMI data, and work order positioning deviations can be identified, reducing sample label contamination. By jointly extracting time features, spatial features, and recovery process features, and jointly scoring evidence sufficiency, topological consistency, and missing rate, sample quality hierarchical management is achieved, making it easy for high-confidence samples to be directly used for model training, while medium- and low-confidence samples enter the review and revision process. The revision history is maintained through version numbers and change logs, ensuring that power outage samples still have complete traceability after multiple rounds of automatic revision and manual review.

[0026] In this embodiment, the power loss starting point mentioned in step S1 The following priority is used to determine the time of power loss start time: OMS record is preferred; when OMS record is missing, the earliest voltage loss time detected by AMI is used; when both OMS and AMI are missing, the earliest fault alarm time recorded by the distribution automation system is used; the power restoration endpoint is determined according to the following priority: The following priorities apply: the power restoration endpoint time recorded by the OMS is used first; when the OMS power restoration record is missing, the power restoration stabilization time monitored by the AMI is used; when the above information is still missing, the work order processing completion time or the reset time of the distribution automation system is used.

[0027] Dynamic event window in step S2 Represented as: .

[0028] in, Forward extension duration of power loss The duration of backward extension after power restoration; and The quantiles are adaptively determined based on the historical latency distribution of different data sources, or obtained through training with historical labeled samples.

[0029] The comprehensive similarity S in step S3 satisfies: .

[0030] in The proportion of time overlap. To determine the overlap ratio of the affected areas, For power supply path similarity, , , The weighting coefficients and When S is not less than the preset threshold When the threshold is reached, the corresponding records are merged into the same event cluster and assigned a unified event primary key ID; The threshold is determined by the optimal F1 value on the historical labeled samples.

[0031] The time drift correction in step S4 is based on the overall offset of each evidence source. For the decision variables, an optimization model is established to minimize the sum of the absolute values ​​of the offsets of each evidence source, with the corrected timestamps satisfying the logical order that alarms precede actions, actions precede power outages, power outages precede work orders, and work orders precede power restoration as constraints; the corrected timestamps satisfy... .

[0032] The anomaly detection in step S5 includes: detecting anomalies with a duration less than a transient threshold. Trip alarms are identified as transient records and removed or downgraded in weight; values ​​exceeding preset engineering boundaries are identified as outliers; the missing data processing includes assigning missing labels to missing channels, using linear interpolation, nearest-neighbor event interpolation, or interpolation based on historical curves of similar equipment for interpolable data, and recording the interpolation identifier in the evidence table.

[0033] Topology consistency metrics in step S6 satisfy: .

[0034] in This is the theoretical set of power outage impacts obtained from topological snapshots and switch state backtracking at the moment of the power outage. For the actual set of recorded effects; when Less than the threshold At that time, output inconsistency reason labels and revision candidates, and feed the revision results back to the quality scoring and version evolution process.

[0035] The time characteristics in step S7 include at least the power outage duration, alarm-to-action delay, action-to-voltage loss delay, voltage loss-to-work order delay, work order-to-power restoration delay, and alarm frequency; the spatial characteristics include at least the number of users affected by the power outage, the number of transformers affected by the power outage, the length of the power supply line, the depth of the power supply path hierarchy, the reachability of the interconnection and transfer, the load density, and the proportion of important users; the restoration process characteristics include at least the restoration rate curve, segment boundary time, duration of each segment, slope, long-tailed restoration index, and secondary power outage indicator.

[0036] The sample quality score in step S8 satisfies: .

[0037] in As the scoring weight and The evidence sufficiency E is a weighted result of the completeness of each key evidence channel, and the missing rate M is the ratio of the number of missing fields to the total number of fields. The data is stratified and stored according to the Q value, where Q≥0.80 enters the high confidence database, 0.50≤Q<0.80 enters the medium confidence database, and Q<0.50 enters the low confidence database.

[0038] In steps S9 and S10, the four-table structure includes at least an event table, an evidence table, a feature table, and a quality table. The event table stores event primary keys, time information, and spatial information. The evidence table stores original evidence fragments and their correction results. The feature table stores extracted multidimensional features. The quality table stores quality scores, missing labels, consistency labels, inconsistency reason labels, and version logs. Each revised sample generates a new version number and retains the historical version chain for traceability.

[0039] Furthermore, to improve the feasibility and quantifiability of the method, the present invention adopts the following formula.

[0040] Event window construction formula: .

[0041] in, For the event The corresponding dynamic event window; This marks the start of the power outage. This marks the end point of power restoration. This refers to the forward propagation time of the power outage; This refers to the extended duration after power restoration. The event window is used to uniformly capture evidence fragments related to the same event from OMS, distribution automation system, AMI, and work order system.

[0042] Formula for calculating the overall similarity of event clustering and merging: ; ; ; .

[0043] Where S is the candidate record and candidate records Overall similarity between them; The percentage of overlap between time intervals; The percentage of overlap in the affected area; For power supply path similarity; and These represent the time intervals corresponding to the two records; and These are the sets of influence ranges corresponding to the two records; and These are the sets of power supply path nodes corresponding to the two records; , , The weighting coefficients and .

[0044] Time drift correction model: ; ; ; ; ; ; .

[0045] in, For the first The overall offset of each source of evidence; This is the offset penalty coefficient; These are slack variables used to handle records of local contradictions; The relaxation penalty coefficient; This is the original timestamp; The corrected timestamp; Represents event log The data source to which it belongs. The goal of this optimization model is to restore the event sequence that satisfies business logic from multiple sources of evidence, while minimizing overall offset.

[0046] Formula for topology consistency index: .

[0047] in, This is a topology consistency indicator; This is the theoretical set of power outage impacts obtained by backtracking the topology snapshot, switch status, and power supply path at the time of the power outage; This is the set of power outage impacts obtained from actual records. The closer it is to 1, the more consistent the theoretical range is with the actual records.

[0048] Formulas for sufficiency of evidence, missing rate, and quality score: ; ; .

[0049] in, For the sufficiency of evidence; Let be the weight of the u-th key evidence pathway; This is an indicator of whether the channel is complete; it is set to 1 when complete and 0 or a decay value lower than 1 when incomplete; M is the missing rate. This represents the number of missing fields. _t ... As the scoring weight and .

[0050] Segmented characteristic formula for the recovery process: ; .

[0051] in, To recover the curve The slope of the recovery rate of the segment; For the first The cumulative recovery ratio at each segment boundary moment; For the first Each segment boundary moment; For the long-tail recovery index; This marks the boundary moment for entering a stable recovery phase. and These represent the recovery start point and the recovery end point, respectively. The segmented features described above can characterize the differences between the fast recovery segment, the slow recovery segment, and the stable segment.

[0052] This example uses the construction of a 10kV feeder outage sample in a certain region. The data to be processed includes: OMS outage plans and accident records, SOE alarm and switch action records from distribution automation systems, AMI user power outage / restoration records, emergency repair work order records, GIS ledgers, feeder topology models, and meteorological data. The goal is to merge these multi-source records into a uniquely identifiable, traceable outage sample that can be used for model training.

[0053] Step 1: Generation of power outage event candidates.

[0054] First, extract the planned outage time, outage start time, power restoration end time, and planned outage range from the OMS; identify the earliest time and last stable power restoration time of the user's voltage loss cluster from the AMI; extract fault alarms, switch trips, reclosing, isolation, and power transfer actions from the distribution automation system; and extract the dispatch time, arrival time, processing completion time, and location description from the work order system.

[0055] For the original records of the same feeder or adjacent feeders that are close in time, a coarse screening is performed. If the OMS power outage record exists, the OMS time is used as the candidate event boundary. If the OMS record is missing, the earliest AMI voltage loss time and the last stable power restoration time are used as the candidate boundary. If the above data is insufficient, the candidate boundary is determined by the combination of the earliest alarm time of the distribution automation system and the work order processing completion time.

[0056] For example, in a candidate event, the OMS records the power outage start time as 09:12:00 and the power restoration end time as 10:05:00; the earliest time of the AMI user's voltage loss cluster is 09:13:24; the distribution automation system trips at 09:11:46; and the work order is cancelled at 10:07:10. Therefore, the initial power outage start time is preferentially set to 09:12:00, and the power restoration end time is preferentially set to 10:05:00.

[0057] Step 2: Building the dynamic event window.

[0058] Based on historical statistics, the typical lead time of the regional distribution automation system relative to the OMS is 0–3 minutes, the typical lag time of the AMI relative to the OMS is 0–8 minutes, and the post-recording delay of work orders is 2–10 minutes. Therefore, the power outage duration is extended forward. Set to 5 minutes to extend the time after power restoration. Set it to 10 minutes.

[0059] According to the formula An event window was constructed, resulting in the event window for the candidate event being [09:07:00, 10:15:00]. Subsequently, OMS records, SOE sequences, AMI depressurization / restoration curves, work order records, and relevant meteorological evidence were extracted from the same feeder, the same substation area, and adjacent connection areas within this event window.

[0060] Step 3: Clustering and merging duplicate events and generating event primary keys.

[0061] For candidate records acquired within the event window, calculate the time overlap ratio between each pair of records. Overlapping ratio of the scope of influence Similarity to power supply path Let the time intervals of the two candidate records be respectively... and Then it can be based on the formula Calculate the degree of time overlap between the two.

[0062] Then, based on the set of power outage users, the set of power outage areas, and the set of theoretical power supply path nodes, calculations are performed. and Let the weights be... , , If two records correspond , , Then the overall similarity =0.40×0.86+0.35×0.79+0.25×0.92=0.8525.

[0063] When S is not less than the threshold When the value is 0.80, two records are identified as cross-source representations of the same event and merged into the same event cluster. A unified event primary key is generated for this event cluster, for example, OUT_20250318_091200_003. This event primary key is simultaneously written to the event table, evidence table, feature table, and quality table, serving as a unified index for all subsequent processing flows.

[0064] For scenarios with parent-child event relationships, such as a main line failure causing a secondary power outage on a branch line, parent event ID and child event ID fields can be added to the event primary key. For cases of power loss after multi-stage recovery, the relationship is maintained using a unified event primary key, and the event stage number is recorded as follows. , wait.

[0065] Step 4: Overall time drift correction.

[0066] After the event primary key is determined, the multi-source records in this event cluster are mapped to a unified time axis. Let OMS be the reference time axis, and the offset of the distribution automation system be... AMI offset is The offset of the work order system is .

[0067] A constrained optimization model is established based on the logical sequence of alarms preceding actions, actions preceding power outages, power outages preceding work orders, and work orders preceding power restoration. If contradictory situations arise in the original records, such as SOE actions occurring later than AMI power outages or work order completion preceding power restoration, slack variables are introduced. The size of the offset is controlled by a penalty term to minimize the overall offset and make the event sequence as reasonable as possible.

[0068] For example, solving a certain event yields =-74s, =+96s, =+182s. Therefore, the original SOE trip time 09:11:46 is corrected to 09:10:32, the earliest AMI power failure time 09:13:24 is corrected to 09:15:00, and the work order cancellation time 10:07:10 is corrected to 10:10:12. After correction, the event chain satisfies the business sequence on a unified timeline and can be used for latency characteristic calculation.

[0069] Step 5: Handling missing, abnormal, and transient issues.

[0070] Perform field-level and channel-level integrity checks on the corrected data. For cases where OMS outage range descriptions are missing, AMI local user voltage curves are missing, or work order processing times are missing, mark the missing data with a missing tag in the evidence table.

[0071] If the duration of an alarm is less than the transient threshold For example, if the time interval is less than 5 seconds, the alarm is identified as a transient disturbance record and can be either removed or retained but given a low weight. If the voltage, current, load, or recovery ratio exceeds the acceptable engineering boundary, it is marked as an outlier. For recoverable fields, linear interpolation, adjacent similar event interpolation, or equipment-specific historical curve interpolation can be used, and the interpolation method should be recorded.

[0072] Step 6: Topology consistency closed-loop verification during power outage.

[0073] Obtain a snapshot of the feeder topology at the moment the event occurs. Starting from the power source, and combining the location of the trip switch, the status of the tie switch, and the open / closed status of the switch, traverse downstream to the affected nodes to obtain the theoretical set of power outage impacts. .

[0074] Simultaneously, an actual impact record set is generated based on the OMS outage range, the AMI undervoltage user set, the work order location description, and the repair scope. And through the formula Calculate the topology consistency index.

[0075] For example, the theoretical impact set includes 520 users and 8 distribution transformers, while the actual recorded impact set includes 486 users and 7 distribution transformers. The intersection of these two sets covers 472 users and 7 distribution transformers, so C≈0.86 can be calculated. If the threshold... If set to 0.90, the event is considered to be due to insufficient topology consistency.

[0076] Furthermore, the reasons for inconsistencies are categorized and identified: if the theoretical range is significantly larger than the actual recorded range and AMI data is severely missing, it is marked as an AMI missing report type; if the work order location differs significantly from the theoretical trip point, it is marked as a work order location offset type; if the status of the tie switch conflicts with the ledger record, it is marked as a switch status abnormal type; if the path topology does not match the GIS ledger, it is marked as a ledger lag type. The system generates revision candidates and enters the subsequent version update process.

[0077] Step 7: Multidimensional feature extraction.

[0078] Regarding the time characteristics, the duration of the power outage was calculated based on the corrected unified time axis. Features include alarm-to-action delay, action-to-power-loss delay, power-loss-to-work-order delay, work-order-to-power-restoration delay, and alarm frequency per unit time.

[0079] In terms of spatial characteristics, statistics include the number of users affected by power outages, the number of affected distribution areas, the number of affected transformers, the length of affected power lines, the depth of power supply path hierarchy, the accessibility of interconnection and transfer, the proportion of important users, and the load density. For scenarios involving multiple voltage levels, additional records can be made of the upper-level switch hierarchy, branch line depth, and ring network interconnection relationships.

[0080] Regarding the characteristics of the recovery process, based on the AMI recovery ratio curve or the transformer area power restoration ratio curve, the recovery process is segmented according to the points of change in the slope of the recovery curve. The duration, segment slope, percentage of restored power, and long-tail recovery index of the fast recovery segment, slow recovery segment, and stable segment are extracted. If a second power outage occurs during the recovery process, the secondary power outage indicator and its occurrence time are recorded.

[0081] Step 8: Sample quality scoring and stratified data entry.

[0082] For each event sample, calculate the sufficiency of evidence (E), topological consistency (C), and missing rate (M). Assume the key evidence channels include six categories: OMS, SOE, AMI, work orders, topological ledgers, and meteorological data, with weights of 0.20, 0.20, 0.25, 0.15, 0.15, and 0.05, respectively. If the completeness indicators for these six channels are 1, 1, 0.7, 1, 1, and 1, respectively, then E = (0.20 + 0.20 + 0.175 + 0.15 + 0.15 + 0.05) / 1 = 0.925.

[0083] Continuing with C=0.86, missing rate M=0.12, and rating weights a=0.40, b=0.40, c=0.20, we have Q=0.40×0.925+0.40×0.86+0.20×(1-0.12)=0.89. This sample enters the high-confidence database and can be directly used to train the sample library.

[0084] If a sample score is between 0.50 and 0.80, it is moved to the medium confidence pool for verification or pending review; if Q is less than 0.50, it is moved to the low confidence pool, awaiting manual review or automatic revision. After revision, Q is recalculated, and a new version number is generated.

[0085] Step Nine: Importing the four-table structure into the database and version evolution.

[0086] The event table stores the event primary key, power outage start point, power restoration end point, associated feeder, main impact area, parent-child event relationship, stage number, and version number; the evidence table stores OMS records, SOE sequences, AMI segments, work order segments, meteorological segments, and their original time, correction time, missing information, and anomaly information; the feature table stores temporal features, spatial features, and recovery process features; and the quality table stores E, C, M, Q, inconsistency reason labels, revision candidates, review status, and change logs.

[0087] For example, after revising the work order location and supplementing the AMI for the aforementioned inconsistency event, the quality score of the original version improved from 0.72 to 0.84. The system automatically generated a new version and recorded the reason for the revision in the change log as work order location offset; the revision actions were location correction and AMI supplementation.

[0088] Step 10: Evaluation of application effectiveness.

[0089] In scenarios with numerous cross-source duplicate records and drifting original timestamps, this invention can converge multiple cross-source records into a single sample entity by using a unified event primary key, avoiding problems such as repeated training and inconsistent statistical standards; it can significantly reduce event chain disorder through overall drift correction; it can reduce sample label contamination through topological consistency verification and revision candidate generation; and it can ensure that the training sample library continuously converges to a high-quality state through quality scoring and version evolution management.

[0090] Therefore, this invention is not a single data cleaning method, but a closed-loop governance method for the standardized construction of power outage event samples. It is applicable to various applications such as power outage event analysis in distribution networks, intelligent fault assessment, emergency repair efficiency evaluation, experience rule extraction, and knowledge graph construction.

[0091] This application also provides a device for standardizing and constructing multi-source power outage data samples and extracting spatiotemporal consistency features based on event primary keys, including: The feature selection module is used to collect data from the power outage management system (OMS), distribution automation system, AMI, work order system, and topology ledger to determine the power outage initiation points of candidate power outage events. and the end point of power restoration A dynamic event window is constructed based on the power outage start point and power restoration end point, and multi-source evidence fragments related to the event candidate are extracted within the event window; The merging module is used to cluster and merge duplicate events corresponding to multi-source evidence fragments based on time overlap, impact range overlap, and power supply path similarity, and generate a unified event primary key ID for the merged events. The detection module is used to establish a logical sequence constraint model of alarms, actions, undervoltage, work orders and power restoration using a unified event primary key as an index, perform overall drift correction on the timestamps of each evidence source, and perform missing detection, anomaly detection, transient identification and imputation labeling on the corrected evidence data. The comparison module is used to perform power supply path backtracking based on the topology snapshot at the time of power outage, obtain the theoretical power outage impact range, and compare it with the actual recorded range to complete the topology consistency verification and the determination of the cause of inconsistency; The dictionary module is used to extract temporal features, spatial features, and recovery process features from the corrected and verified event samples to form a feature dictionary; The storage module is used to calculate the sample quality score Q based on the sufficiency of evidence E, topological consistency C, and missing rate M. The samples are then stratified into high-confidence, medium-confidence, or low-confidence databases according to the quality score Q. The module also maintains version numbers and change logs for revised samples and stores the samples in the database according to a four-table structure: event table, evidence table, feature table, and quality table.

[0092] In addition, embodiments of the present invention also provide a computer-readable storage medium.

[0093] The present invention stores on a computer-readable storage medium a program for the standardized construction and spatiotemporal consistency feature extraction of multi-source power outage data samples driven by event primary keys. When the program is executed by a processor, it implements the steps of the method described above for the standardized construction and spatiotemporal consistency feature extraction of multi-source power outage data samples driven by event primary keys.

[0094] The method implemented when the event-key-driven multi-source power outage data sample standardization construction and spatiotemporal consistency feature extraction program is executed can be referred to in various embodiments of the present invention, which will not be repeated here.

[0095] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0096] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0097] This application can be used in a wide variety of general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices. This application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform specific tasks or implement specific abstract data types. This application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0098] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0099] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are exhaustively listed. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0100] For those skilled in the art, various modifications and improvements can be made without departing from the concept of the present invention, and these modifications and improvements are all within the scope of protection of the present invention. The scope of protection of the present invention is defined by the appended claims.

Claims

1. A method for standardized construction and spatiotemporal consistency feature extraction of multi-source power outage data samples driven by event primary keys, comprising the following steps: Data is collected from the Power Outage Management System (OMS), Distribution Automation System (AMI), Work Order System, and Topology Ledger to determine the potential power outage initiation points for power outage events. and the end point of power restoration A dynamic event window is constructed based on the power outage start point and power restoration end point, and multi-source evidence fragments related to the event candidate are extracted within the event window; Based on time overlap, impact range overlap and power supply path similarity, the duplicate events corresponding to multi-source evidence fragments are clustered and merged, and a unified event primary key ID is generated for the merged events. Using the unified event primary key as an index, a logical sequence constraint model for alarms, actions, power outages, work orders, and power restoration is established. Overall drift correction is performed on the timestamps of each evidence source, and missing data detection, anomaly detection, transient identification, and imputation annotation are performed on the corrected evidence data. Power supply path backtracking is performed based on the topology snapshot at the time of power outage to obtain the theoretical power outage impact range, and then compared with the actual recorded range to complete the topology consistency verification and inconsistency cause identification; Temporal features, spatial features, and recovery process features are extracted from the corrected and verified event samples to form a feature dictionary; The sample quality score Q is calculated based on the evidence sufficiency (E), topological consistency (C), and missing rate (M). The samples are then stratified into high-confidence, medium-confidence, or low-confidence databases according to the quality score Q. Version numbers and change logs are maintained for revised samples. The samples are stored in the database according to a four-table structure: event table, evidence table, feature table, and quality table.

2. The method for standardized construction and spatiotemporal consistency feature extraction of multi-source power outage data samples based on event primary key driven according to claim 1, characterized in that, The dynamic event window Represented as: ； in, Forward extension duration of power loss The duration of backward extension after power restoration; and The quantiles are adaptively determined based on the historical latency distribution of different data sources or obtained through training with historical labeled samples.

3. The method for standardized construction and spatiotemporal consistency feature extraction of multi-source power outage data samples based on event primary key driven according to claim 1, characterized in that, The comprehensive similarity S satisfies: ； in The proportion of time overlap. To determine the overlap ratio of the affected areas, For power supply path similarity, , , The weighting coefficients and ; When S is not less than the preset threshold When the event occurs, the corresponding records will be merged into the same event cluster and assigned a unified event primary key ID; The threshold The threshold is determined by the optimal F1 value on the historical labeled samples.

4. The method for standardized construction and spatiotemporal consistency feature extraction of multi-source power outage data samples based on event primary key driven according to claim 1, characterized in that, The steps for calculating the threshold voltage include: The net load sequence corresponding to the optimized scheduling result is input into the established net load-gate voltage mapping model. The gate voltage for each time period is calculated. Based on the Gaussian process regression model, the predicted distribution of the gate voltage is obtained. The following formula is used for calculation: In the formula, for Predicted average voltage at time-bound thresholds To predict variance.

5. The method for standardized construction and spatiotemporal consistency feature extraction of multi-source power outage data samples based on event primary key driven according to claim 1, characterized in that, The anomaly detection includes: detecting anomalies whose duration is less than a transient threshold. Trip alarms are identified as transient records and removed or downgraded in weight; values ​​exceeding preset engineering boundaries are identified as outliers; the missing data processing includes assigning missing labels to missing channels, using linear interpolation, nearest-neighbor event interpolation, or interpolation based on historical curves of similar equipment for interpolable data, and recording the interpolation identifier in the evidence table.

6. The method for standardized construction and spatiotemporal consistency feature extraction of multi-source power outage data samples based on event primary key driven according to claim 5, characterized in that, The topology consistency index satisfy: ； in This is the theoretical set of power outage impacts obtained from topological snapshots and switch state backtracking at the moment of the power outage. For the actual set of recorded effects; when Less than the threshold At that time, output inconsistency reason labels and revision candidates, and feed the revision results back to the quality scoring and version evolution process.

7. A device for standardized construction and spatiotemporal consistency feature extraction of multi-source power outage data samples driven by event primary keys, characterized in that, include: The feature selection module is used to collect data from the power outage management system (OMS), distribution automation system, AMI, work order system, and topology ledger to determine the power outage initiation points of candidate power outage events. and the end point of power restoration A dynamic event window is constructed based on the power outage start point and power restoration end point, and multi-source evidence fragments related to the event candidate are extracted within the event window; The merging module is used to cluster and merge duplicate events corresponding to multi-source evidence fragments based on time overlap, impact range overlap, and power supply path similarity, and generate a unified event primary key ID for the merged events. The detection module is used to establish a logical sequence constraint model of alarms, actions, undervoltage, work orders and power restoration using a unified event primary key as an index, perform overall drift correction on the timestamps of each evidence source, and perform missing detection, anomaly detection, transient identification and imputation labeling on the corrected evidence data. The comparison module is used to perform power supply path backtracking based on the topology snapshot at the time of power outage, obtain the theoretical power outage impact range, and compare it with the actual recorded range to complete the topology consistency verification and the determination of the cause of inconsistency; The dictionary module is used to extract temporal features, spatial features, and recovery process features from the corrected and verified event samples to form a feature dictionary; The storage module is used to calculate the sample quality score Q based on the sufficiency of evidence E, topological consistency C, and missing rate M. The samples are then stratified into high-confidence, medium-confidence, or low-confidence databases according to the quality score Q. The module also maintains version numbers and change logs for revised samples and stores the samples in the database according to a four-table structure: event table, evidence table, feature table, and quality table.

8. The device for standardized construction and spatiotemporal consistency feature extraction of multi-source power outage data samples driven by event primary keys according to claim 7, characterized in that, The dynamic event window Represented as: ； in, Forward extension duration of power loss The duration of backward extension after power restoration; and The quantiles are adaptively determined based on the historical latency distribution of different data sources or obtained through training with historical labeled samples.

9. A device for standardized construction and spatiotemporal consistency feature extraction of multi-source power outage data samples driven by event primary keys, characterized in that, The method includes a processor, a memory, and a computer program stored in the memory and executable by the processor, wherein when the computer program is executed by the processor, it implements the steps of a method for standardizing and extracting spatiotemporal consistency features of multi-source power outage data samples based on event primary key driven as described in any one of claims 1 to 6.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, wherein when the computer program is executed by a processor, it implements the steps of the method for standardizing the construction of multi-source power outage data samples and extracting spatiotemporal consistency features based on event primary key driven as described in any one of claims 1 to 6.

Images