An electric metering fault intelligent diagnosis method and system

Abstract

The application discloses a kind of electric meter measurement fault intelligent diagnosis method and system, it is related to electric power technical field, the method is by obtaining electric meter end measurement data stream, concentrator communication log and main station receiving record, after data pre-processing, obtain path structure set, and by path structure set, identify potential retransmission data, and construct initial annotation matrix;According to initial annotation matrix, by similarity clustering analysis, identify data structure type, obtain ghost feature set;Based on ghost feature set, analyze the structural characteristics of measurement behavior under different parameter states, and identify abnormal window set in combination with historical data;Classify faults by abnormal window set, generate path propagation sequence in combination with data structure type, and lock the fault label of corresponding device identification according to the participation degree of each propagation path, to complete the fault intelligent diagnosis of electric meter measurement.

Description

Technical Field

[0001] This invention relates to the field of power technology, specifically to an intelligent diagnostic method and system for metering faults. Background Technology

[0002] In the field of electricity metering in smart grids, electricity meters not only perform basic electricity metering functions, but also communicate with upper-level systems through concentrators and master stations to achieve remote monitoring and data acquisition. Metering data from the meters, communication logs from the concentrators, and reception records from the master station constitute the core data source of electricity metering information and are an important foundation for smart grid metering reliability assessment and fault analysis.

[0003] In existing technologies, electricity metering data and communication data suffer from problems such as multi-source, asynchronous, and repetitive transmission, making it difficult for traditional fault diagnosis methods to accurately identify anomalies. First, data retransmission and repeated reporting of links lead to redundant information mixing, forming ghost data, which interferes with the determination of true metering fluctuations; second, the metering response after the remote parameters of the electricity meter are issued has time lag and status misalignment, which traditional methods find difficult to quantify, resulting in inaccurate fault location.

[0004] The aforementioned defects primarily stem from the multi-layered data transmission structure and communication link instability of the electricity metering system. On one hand, local clock drift, communication delays, and retransmission mechanisms exist between different devices, making time series alignment difficult and resulting in ghosting and redundant data. On the other hand, remote parameter changes take effect with a lag at the meter end, causing misalignment between metering data and parameter status, making it difficult to accurately identify abnormal windows. These problems can lead to various abnormal effects, such as misjudging meter fault types, delaying abnormal alarms, reducing grid operation and maintenance efficiency, and may even affect billing accuracy and the reliability of power quality assessment, thus restricting the smart grid's ability to manage and monitor electricity metering data with high precision in real time. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides an intelligent diagnostic method and system for electricity metering faults, solving the problems mentioned in the background section.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] Firstly, an intelligent diagnostic method for electricity meter metering faults includes:

[0008] The metering data stream, concentrator communication log, and main station reception record are acquired. After data preprocessing, a path structure set is obtained. Through the path structure set, potential retransmission data is identified, and an initial annotation matrix is ​​constructed.

[0009] Based on the initial annotation matrix, similarity clustering analysis is used to identify the data structure type and obtain the ghosting feature set;

[0010] Based on the ghosting feature set, the structural characteristics of the measurement behavior under different parameter states are analyzed, and combined with historical data, the abnormal window set is identified.

[0011] By classifying fault categories through anomaly window sets and combining them with data structure types, a path propagation sequence is generated. Based on the degree of participation of each propagation path, the fault label of the corresponding device is locked to complete the intelligent diagnosis of faults in electricity metering.

[0012] Secondly, an intelligent diagnostic system for electricity meter metering faults includes:

[0013] The data reconstruction module is used to acquire metering data streams from the electricity meter, communication logs from the concentrator, and receiving records from the main station. After data preprocessing, a path structure set is obtained. Through the path structure set, potential retransmission data is identified, and an initial annotation matrix is ​​constructed.

[0014] The structure recognition module is used to identify the data structure type based on the initial annotation matrix through similarity clustering analysis, and obtain the ghosting feature set.

[0015] The anomaly diagnosis module is used to analyze the structural characteristics of metrological behavior under different parameter states based on the ghosting feature set, and to identify the set of abnormal windows by combining historical data.

[0016] The positioning module is used to classify fault categories by abnormal window set, generate path propagation sequence by combining data structure type, and lock the fault label of corresponding device according to the degree of participation of each propagation path, so as to complete the intelligent diagnosis of faults in electricity metering.

[0017] The above-described solution of the present invention has at least the following beneficial effects:

[0018] By performing multi-dimensional unified time correction, path reconstruction, and initial annotation processing on the meter-side metering data stream, concentrator communication logs, and master station receiving records, the system achieves accurate identification of data retransmissions, link anomalies, and time misalignments. By constructing an initial metering-communication joint sequence, path structure set, and initial annotation matrix, the system can effectively extract potential retransmission data and utilize similarity clustering analysis, differential variance, and path distribution entropy to identify ghosting structures, distinguishing between duplicated ghosting and normally changing structures. This enables the compression of redundant data and the retention of valid metering data. This differentiated weighted fusion mechanism not only ensures the purity of the deduplication sequence but also quantifies the contribution of each metering data point on different transmission paths, providing a reliable foundation for subsequent coupled analysis of parameter state changes and metering behavior. Furthermore, by converting the remote parameter transmission records of electricity meters into continuous state functions and coupling them with deduplicated metering sequences, the time lag and state misalignment between parameter changes and metering responses can be accurately quantified. This provides an objective basis for identifying abnormal windows and determining fault types, enabling joint analysis and classification of parameter state misalignment and ghosting faults. It provides power operation and maintenance personnel with a highly operable intelligent diagnostic tool, greatly improving fault identification accuracy and response speed.

[0019] By constructing path propagation sequences and topology propagation graphs, the propagation trajectory of abnormal events in network transmission is visualized and path weights are quantitatively analyzed. By statistically analyzing the state misalignment intensity, ghosting intensity, and coordination coefficient of each interval within the abnormal window set, abnormal intervals can be classified into various fault labels, enabling refined identification of fault types. Based on this, combined with the path propagation sequence, the participation degree of each transmission path within the fault window is calculated, key paths and key nodes are screened, and a precise location set is formed, thereby achieving accurate spatial and temporal location of meter faults. This method can not only identify single-point anomalies but also effectively analyze the propagation effect of anomalies in network transmission, supporting rapid location and type determination of meter faults in complex multi-node, multi-path systems. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating an intelligent diagnostic method for metering faults in electricity meters according to the present invention.

[0021] Figure 2 This is a schematic diagram of the structure of an intelligent diagnostic system for metering faults of an electric meter according to the present invention;

[0022] In the attached diagram, the components represented by each number are as follows:

[0023] Data reconstruction module 11, structure recognition module 12, anomaly diagnosis module 13, and location module 14. Detailed Implementation

[0024] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] In the description of this invention, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0026] In the description of this invention, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this invention is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the invention. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that the invention can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid obscuring the description of the invention with unnecessary detail. Therefore, the invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed herein.

[0027] Example 1, such as Figure 1 As shown, the present invention provides an intelligent diagnostic method for metering faults, including:

[0028] S100: Acquire metering data streams from the electricity meter, communication logs from the concentrator, and reception records from the main station. After data preprocessing, obtain a path structure set. Then, identify potential retransmission data using the path structure set and construct an initial annotation matrix.

[0029] S200: Based on the initial annotation matrix, similarity clustering analysis is used to identify the data structure type and obtain the ghosting feature set;

[0030] S300: Based on the ghosting feature set, analyze the structural characteristics of metrological behavior under different parameter states, and identify the abnormal window set by combining historical data;

[0031] S400: By classifying fault categories through anomaly window sets and combining data structure types, a path propagation sequence is generated. Based on the degree of participation of each propagation path, the fault label of the corresponding device is locked to complete the intelligent diagnosis of faults in electricity metering.

[0032] Through a systematic and multi-level data processing workflow, accurate identification and fault location of electricity metering anomalies are achieved, effectively improving the reliability of electricity metering data and the level of intelligent grid operation and maintenance.

[0033] In practical implementation, the metering data stream from the electricity meter, the concentrator communication log, and the main station's receiving records are first acquired through step S100. These data are then preprocessed and their path structures constructed, enabling a unified timeline mapping for complex data sources and preliminary identification of retransmitted data, thereby building an initial annotation matrix. This step not only ensures the accurate correspondence of data in time and transmission path but also provides a reliable foundation for subsequent structured analysis, enabling the system to distinguish between potential ghosting data and valid metering information.

[0034] Next, in step S200, by performing similarity clustering analysis on the initial annotation matrix, different data structure types can be identified and a ghost feature set can be generated. This allows the system to automatically distinguish between redundant data caused by repeated transmissions and valid measurement data with real fluctuations, solving the problems of link duplication, retransmission and time misalignment that are difficult to quantify in traditional methods, and further enhancing the accuracy of data purification and feature extraction.

[0035] In step S300, the method combines the ghosting feature set with historical data to identify anomaly window sets by analyzing metering behavior under different parameter states, thus achieving dynamic capture of anomaly patterns in metering data. This step not only reflects the time misalignment between the remote parameter issuance and metering response of the meter, but also quantifies the probability of anomalies by constructing a conditional probability model using historical data, making the determination of anomaly windows more scientific and interpretable.

[0036] Finally, in step S400, the method comprehensively analyzes the anomaly window and data structure type, generates a path propagation sequence, quantifies the participation degree of each propagation path, and can accurately pinpoint the fault label of a specific device, thereby achieving precise location of the fault type and location of the meter.

[0037] Through the above logical processing chain, this invention not only achieves the comprehensive application of data fusion, ghosting removal, anomaly identification and path tracing in terms of technology, but also significantly reduces fault location time and manual intervention in actual operation and maintenance, and can maintain stable diagnostic performance under various abnormal conditions, thereby greatly improving the intelligence level and reliability of the power grid metering system.

[0038] Specifically: S100 includes: based on the metering data stream at the meter end, the concentrator communication log and the main station receiving record, the timestamp is corrected by using a unified time axis and round-trip delay, and the communication behavior is parsed and encoded to obtain the initial quantity communication joint sequence;

[0039] In practice, the metering data stream from the electricity meter, the communication log from the concentrator, and the receiving record from the main station are first acquired, and the three types of data are mapped to the same time axis system. Among them, due to the local clock drift and communication link instability of different devices, a time correction function needs to be constructed to adjust the original timestamp.

[0040] Specifically, for each metering data record, its transmission and reception times are extracted, and a delay estimation model is established based on the round-trip time (RTT) in the communication log. The local average delay is calculated using a sliding window statistical method. The calculation method for the local average delay is as follows: select the i-th data record and the n data records before it, for a total of n+1 data records, as samples within the sliding window. Calculate the difference between the main station's reception time and the meter's transmission time in each sample. Sum these differences and then divide by the number of samples n+1 to obtain the local average delay of the i-th data record. This local average delay is used to correct the timestamp of the metering data and solve the time misalignment problem caused by local clock drift of different devices and unstable communication links.

[0041] By calculating the local average delay of each data point, the actual transmission sequence of the metering data can be deduced, and a corrected timestamp can be generated. This ensures that the metering data, concentrator communication logs, and main station reception records are mapped to the same timeline, providing an accurate time reference for subsequent steps such as binding metering data with communication behavior sequences and reconstructing transmission paths, thus ensuring the accuracy of subsequent analysis.

[0042] Subsequently, the local average delay is used for time correction to form a corrected timestamp, which is the reception time of each data point minus the corresponding local average delay; and the corrected data is reordered to construct a unified time series set.

[0043] After time alignment is completed, the concentrator communication log is parsed to encode sending, retransmission, acknowledgment and other behaviors into discrete event identifiers, and arranged in chronological order to form a communication behavior sequence.

[0044] The unified time series set and the communication behavior sequence are bound together by time window to generate an initial volume-communication joint sequence, which serves as the input basis for subsequent path identification and ghosting detection.

[0045] A unified time series set is a set of measurement data sorted after timestamp correction, used for unifying time bases;

[0046] Communication behavior sequences are encoded communication events arranged in time, used to correlate metering data;

[0047] The meter data stream is the electricity consumption data collected by the meter, which is generated in real time by the meter.

[0048] The concentrator communication log is a record of the interaction between the concentrator and the electricity meter or master station, and is recorded synchronously by the concentrator.

[0049] The master station receive record is the metering or communication data received by the master station, which is generated after the master station receives the data;

[0050] Based on the initial communication sequence, the transmission path is reconstructed by parsing the frame sequence number and device identifier in the concentrator communication log, the path redundancy and time dispersion are quantified, and a path attribute vector is constructed to obtain a set of attributed path structures.

[0051] In practice, based on the initial combined sequence of metering and communication after time alignment, the transmission path of each metering data is reconstructed. Specifically, the frame sequence number, device identifier and confirmation mechanism in the communication log are used to construct a set of data transmission paths, where each path represents the complete transmission trajectory from the meter to the master station, which is used for path modeling.

[0052] During the path reconstruction process, the path node sequence of each path is formed by matching the association identifiers between the sending record and the receiving record, and the duplicate transmission behavior in the path is marked to obtain the path duplication degree. The path duplication degree refers to the ratio of the total number of times the data received by the main station in the path to the number of unique data records after removing duplicates. It is used to quantify the degree of duplicate transmission in the path and assist in the identification of ghost data.

[0053] Each parameter in the path node sequence represents a transmission node on the corresponding path, used to depict the specific node trajectory of data transmission; transmission nodes include electricity meters, concentrators, etc.

[0054] Meanwhile, in order to characterize the data distribution characteristics within the path, all corrected timestamps within the same path are taken, their average value is calculated first, and then the square of the difference between each timestamp and the average value is calculated. The sum of all square values ​​is divided by the total number of timestamps to obtain the time dispersion of the path, which is used to reflect the degree of dispersion of data arrival time within the same path.

[0055] The path repetition and temporal dispersion are combined to form a path attribute vector, which is then bound to the data transmission path set to generate a set of attributed path structures. This set provides path-level constraint inputs for subsequent data annotation and ghosting recognition.

[0056] The path structure set organizes every data transmission and communication path between the meter, concentrator, and master station into a structured record, which serves as the underlying topology and feature foundation for subsequent ghosting identification, data path grouping, fault propagation topology construction, and critical path fault location.

[0057] When the path repetition exceeds the preset repetition threshold, the corresponding data is marked as potential retransmission data, and the data is initially labeled in combination with the time dispersion to obtain the initial labeling matrix.

[0058] In practice, based on the set of attributed path structures and the unified time series set after time alignment, path attribution mapping is performed on each measurement data point to bind the data point to the corresponding path and record its correction timestamp, thereby constructing a ternary relationship set; the data point refers to a single measurement data point in the unified time series set after time correction.

[0059] In this ternary relation set, each element simultaneously contains data value, transmission path, and time information, allowing subsequent analysis to be carried out in both path and time dimensions. Furthermore, based on the path attribute vector, initial state labeling is performed on each data in the ternary relation set, and a labeling function is constructed. This function is an indicator function. When the path repetition exceeds the preset repetition threshold, it indicates that there is obvious repeated transmission behavior on the path, and the corresponding data is likely to be potential retransmission data, which is not unique and valid measurement data. It needs to be marked to assist in the subsequent ghosting structure identification. In this case, the output is 1; otherwise, the output is 0.

[0060] Simultaneously, based on the time dispersion, a threshold for judging time deviation is generated for each data point; the difference between the timestamp of each data point and the average time of the group is calculated one by one, which is the time deviation magnitude; when the time deviation magnitude of a single data point is greater than the time deviation judgment threshold, a time anomaly auxiliary mark is added to the data to form a composite annotation vector; finally, the annotation results of all data are summarized to construct the initial annotation matrix.

[0061] The parameters of the composite annotation vector include path repetition labels and time discrete anomaly status indicators, which support accurate subsequent ghosting identification.

[0062] The initial labeling matrix corresponds to each meter reading data point in each row, and corresponds to the basic retransmission label and the time anomaly auxiliary label in sequence. It is used to provide prior labels and to constrain the clustering screening of ghost data.

[0063] In this embodiment of the invention, the time correction and path structure reconstruction method based on meter-side metering data stream, concentrator communication log, and master station receiving record proposed in step S100 achieves high-precision synchronization of metering data and ghosting data identification through systematic data preprocessing and path-level labeling mechanism, thereby improving the reliability of fault diagnosis and power grid data analysis.

[0064] Specifically, this invention first corrects for local clock drift and communication link instability between different devices by using a unified time axis and round-trip delay model. It then calculates the local average delay using a sliding window to correct the reception time of each metering data entry and generate a corrected timestamp, thereby achieving precise time mapping between meter data, concentrator communication logs, and master station reception records. This time alignment not only ensures the synchronization of data across devices and nodes but also provides a stable time reference for subsequent communication behavior analysis, transmission path reconstruction, and ghosting identification.

[0065] Subsequently, by parsing the concentrator communication logs, communication events such as sending, retransmission, and acknowledgment are encoded as discrete event identifiers and bound to the corrected unified time series set according to the time window to generate an initial quantitative-communication joint sequence. This ensures that each data point retains both measurement information and embeds path and communication behavior attributes, providing a basic input for subsequent path attribute quantification and ghosting recognition.

[0066] In the path reconstruction stage, by constructing a complete set of data transmission paths using frame sequence numbers and device identifiers, and calculating path redundancy and time dispersion to form path attribute vectors, this invention can accurately quantify the repetition of data in the transmission link and the distribution characteristics of arrival time, thus marking potential retransmission data. Combined with time deviation determination, a composite annotation vector is further formed, providing rich prior information for constructing the initial annotation matrix. This method binds each data point to path and time through a set of ternary relationships, enabling detailed analysis of data in both path and time dimensions. This provides constraints for subsequent ghosting data clustering screening and abnormal window identification, reducing the risk of misjudgment and improving the accuracy of ghosting recognition.

[0067] Logically, time correction ensures the synchronization of data sequences, communication event parsing provides behavioral context, path reconstruction and attribute quantification implement ghosting recognition constraints, and the composite annotation matrix integrates information from various dimensions, providing high-quality prior labels for intelligent diagnosis. This end-to-end data fusion and structured annotation method not only effectively solves the problem of difficult anomaly identification caused by clock drift, link jitter, and duplicate transmission in traditional electricity metering data, but also provides a reliable data foundation for power grid fault diagnosis, remote operation and maintenance, and intelligent monitoring, thereby improving the data quality, analysis accuracy, and operation and maintenance efficiency of the power system. It also enhances the system's ability to process complex, multi-source, and asynchronous metering data, ensuring the stability and accuracy of intelligent diagnosis.

[0068] Specifically: S200 includes: performing similarity clustering analysis on data within the same time neighborhood based on the initial annotation matrix to obtain a set of candidate ghost clusters for ghost structure difference analysis;

[0069] Perform temporal difference analysis and statistics on each candidate cluster in the candidate ghost cluster set to obtain the difference variance for cluster subdivision, and count the transmission path numbers corresponding to all data in a single candidate cluster, calculate the path distribution entropy, and determine whether the data is concentrated on a few paths or dispersed on multiple paths.

[0070] Therefore, it is identified as a replica-type ghosting structure, used to mark invalid and redundant samples to support subsequent data deduplication and purification.

[0071] If the differential variance does not exceed the preset variance threshold and the path distribution entropy does not exceed the preset entropy threshold, it indicates that the intra-cluster measurement data fluctuates gently, the values ​​are highly similar, and the data is concentrated in a small number of fixed transmission paths. Both of these characteristics indicate that the data is homogeneous copy data formed by repeated forwarding of links. Therefore, it is determined to be a replica type ghosting structure. Otherwise, it is determined to be a variable type normal structure, and a ghosting structure identification matrix is ​​constructed.

[0072] The replica ghosting structure refers to homogeneous copying redundant data formed by repeated transmission of links, secondary forwarding of data, and repeated reporting within the same time interval. It belongs to invalid ghosting noise data and is used to accurately locate pure repetitive redundant data and distinguish real business fluctuation data.

[0073] The ghost structure identifier matrix corresponds to each measurement data point in the candidate ghost cluster in each row, and the columns represent storage classification labels, including replication type and variation type; it is used to distinguish data structure types and provide a label basis for differentiated weight allocation;

[0074] Difference variance is used to measure the severity of temporal fluctuations in intra-cluster econometric data;

[0075] Path distribution entropy characterizes the centralized and discrete state of data transmission paths within a cluster;

[0076] The variable normal structure refers to the real and valid metering data within the cluster that fluctuates normally with actual operating conditions, has reasonable differences in values, and has a dispersed transmission path with no concentrated repeated transmission characteristics. It does not belong to the ghosting redundancy data caused by repeated forwarding of links.

[0077] Based on the ghost structure identifier matrix, data types are distinguished. A differential weight fusion algorithm is used to obtain a clean deduplication sequence. During the fusion process, ghost structure features are obtained, including the number of repetitions, path source, and differential statistics, and a ghost feature set is constructed.

[0078] In practice, the ghosting structure identification matrix is ​​used to determine whether a single data point is a copy or a variation, and different structure coefficients are assigned according to the type.

[0079] The structure coefficient is an adaptive adjustment parameter used to distinguish between replicative ghosting structures and variable normal structures. Combined with path repetition, it dynamically calculates the fusion weight of each measurement data point, automatically reducing the weight of ghosting data and preserving the weight of normal, valid data, ensuring the purity and reliability of the subsequently generated deduplication sequence. Specifically, a smaller structure coefficient is automatically calculated for replicative ghosting data to lower its fusion weight and weaken the contribution of retransmitted redundant data, while a larger structure coefficient is automatically calculated for variable normal data to increase its fusion weight and retain the effective information of the true measurement data. Its calculation method is as follows: First, determine the structure type of the current data, and then obtain the difference variance and preset variance threshold of the candidate cluster to which the data belongs. If the data is a replica ghost structure, first calculate the ratio of the difference variance of the cluster to the preset variance threshold, take the negative of the ratio, and then perform an exponential operation. The result is the structure coefficient of the data. If the data is a variable normal structure, first calculate the ratio of the difference variance of the cluster to the preset difference variance threshold using the same method, take the negative of the ratio, and then perform an exponential operation. Finally, subtract the result of the exponential operation from 1. The final value is the structure coefficient of the data.

[0080] Next, the reciprocal of the sum of the path repetition degree plus one is taken as the denominator and multiplied by the structural coefficient. In this way, the fusion weight of each measurement data is calculated point by point to realize the differentiated weighting of ghosting and normal data.

[0081] Within the same time window, multiple sets of homogeneous ghost data are weighted and averaged according to the calculated weights to compress redundant and duplicate samples, generating a pure measurement sequence without ghost interference, i.e., a pure deduplication sequence; at the same time, the number of times each set of data is repeatedly reported, its transmission path, and the differential statistical characteristics within the cluster are counted, and each record is encapsulated and summarized to form a complete set of ghost features.

[0082] Same-source ghosting data refers to multiple metering data that originate from the same source and transmission path, have almost identical content, and are reported repeatedly within the same time window.

[0083] The clean deduplication sequence is used to provide clean, non-repeating benchmark time series data; the ghost feature set is used to retain all ghost quantization features for subsequent probabilistic modeling, fault classification, and path localization.

[0084] Based on the initial annotation matrix, similarity clustering analysis is performed on data within the same temporal neighborhood to obtain a set of candidate ghost clusters for ghost structure difference analysis, including:

[0085] Based on the correction timestamp, a sliding time window of fixed width is divided to delineate all measurement data in the same time neighborhood, limit the time range of ghost data, calculate the similarity between measurement data within the sliding time window, and obtain a similarity matrix for identifying time-series redundant retransmission data, which is used to characterize the similarity of data values.

[0086] The potential retransmission data in the initial annotation matrix is ​​retrieved to make a weighted correction to the similarity matrix, strengthen the association weight of suspected retransmission data, and use the path attribute vector as a constraint to construct a clustering priority. This priority is used to set the clustering order and priority for different transmission paths and different data samples, so that data on the same high retransmission risk path are preferentially clustered into one class, avoiding random clustering across paths and conforming to the actual transmission link pattern of the power grid.

[0087] Based on clustering priority, a density-based clustering algorithm is used to group and aggregate the weighted similarity matrix, obtaining candidate clusters and generating a candidate ghosting cluster set. This set is used to filter out all suspected ghosting data groups and serves as the raw material for subsequent ghosting feature extraction and fault analysis.

[0088] In practice, the ternary relation set and the initial annotation matrix are used as inputs to perform similarity clustering on data within the same time neighborhood. First, the data is divided into sliding windows according to the correction timestamp to construct a set of time windows, which is used to limit the temporal proximity range of candidate ghosting data. Within each time window, a pairwise similarity calculation function is constructed for the data values. Specifically, an exponential operation is used to first calculate the absolute value of the difference between the two sets of measurement data values, divide the absolute value by the preset fluctuation scale parameter and take the negative, and finally calculate the similarity value between the data.

[0089] The preset fluctuation scale parameter is a fixed empirical parameter preset by the system. It is pre-calibrated and configured in conjunction with the fluctuation characteristics of power grid metering data. It is used to control the rate of decay of the similarity of metering data and to define the allowable range of normal numerical fluctuations. The value range is from 0 to 1. The smaller the value, the faster the similarity decays; the larger the value, the wider the data fault tolerance matching range.

[0090] Based on this similarity function, a similarity matrix is ​​constructed. Combined with the potential retransmission markers in the initial annotation matrix, the retransmission markers are used as weighting coefficients to amplify and adjust the pairwise similarity values ​​of corresponding data in the similarity matrix. The similarity weight of potential retransmission data is increased to strengthen the correlation between similar ghosting data. The original similarity of normal data without abnormal markers is retained, and the aggregation interference of irrelevant data is weakened. After completing the global numerical correction calculation, the matrix elements are uniformly integrated to generate a weighted similarity matrix.

[0091] Subsequently, a weighted similarity matrix was used as the basis for determining data association, measuring the similarity of values ​​for each data point within the window to identify highly similar data. The path repetition rate corresponding to each data point was then superimposed, prioritizing data with frequent retransmissions. A density threshold and neighborhood range were set for density clustering, treating highly similar and highly repetitive data as core points. Around these core points, similar data meeting the similarity requirements were continuously absorbed and grouped into a candidate cluster. Path attribute constraints were used to eliminate disorderly and abnormal data, ensuring that each cluster consisted of suspected retransmissions. The clustering results from all time windows were traversed, summarized, and integrated to ultimately obtain a set of candidate ghosting clusters.

[0092] Finally, the data index, path attributes, and time distribution information within each cluster are recorded in a unified manner to form a set of candidate ghosting structures.

[0093] Clustering priority is used to prioritize the aggregation of data along highly repetitive paths, accurately pinpointing the source of ghosting;

[0094] In this embodiment of the invention, in step S200, similarity clustering and ghosting structure difference analysis are used to perform refined data type differentiation and weight assignment on the metering data in the initial annotation matrix, thereby improving the accuracy of metering data processing and the ability to suppress redundant data.

[0095] Specifically, this step first uses the corrected timestamps to divide data within the same time neighborhood into sliding time windows, ensuring that candidate ghosting data are adjacent and logically consistent in the time dimension, thus providing strict temporal constraints for subsequent similarity calculations. Within each window, the numerical differences between pairwise measurement data are calculated using exponential operations, and combined with a preset fluctuation scale parameter, the similarity function is precisely quantified as the degree of numerical convergence between data, providing a foundation for the identification of potential retransmission data.

[0096] Potential retransmission data in the initial annotation matrix are assigned weighted enhancement coefficients, which prioritizes suspected redundant data in the clustering calculation of the similarity matrix, thus forming a weighted similarity matrix that provides accurate input for density clustering. Through path attribute vectors and clustering priority constraints, the algorithm can eliminate messy or abnormal data during the clustering process, ensuring that each candidate ghost cluster originates from a highly similar data set within the same time window, and prioritizing the aggregation of path data with high frequency of repeated transmissions, accurately locating the source of duplicate ghosting.

[0097] Subsequently, temporal difference analysis and statistical calculation of difference variance are performed on each candidate cluster. Simultaneously, the distribution of transmission paths within the cluster is statistically analyzed and path entropy is calculated. This enables the system to distinguish between homogeneous, replicated ghosting data concentrated in a small number of paths and dispersed, varied, normal structure data across multiple paths. Replicated ghosting structures, as invalid redundant samples formed by repeated forwarding, secondary reporting, or data retransmission, are marked and assigned lower structure coefficients and fusion weights during the fusion process, thus achieving automatic weight reduction. Conversely, varied normal data is assigned higher structure coefficients and fusion weights based on differential statistical characteristics, ensuring that accurate measurement information is preserved.

[0098] Finally, within the same time window, the homogeneous ghosting data are weighted and averaged according to the calculated weights to achieve redundant data compression and generate clean deduplication sequences. Simultaneously, the repetition count, path source, and differential characteristics of each data set are statistically analyzed to form a complete ghosting feature set. This method not only logically forms a closed-loop processing mechanism from data pre-labeling, similarity clustering, differential variance and path entropy calculation to adaptive adjustment of structural coefficients and weighted fusion, but also functionally accurately distinguishes between ghosting redundant data and valid metering data, automatically purifies redundant information, reduces the risk of misjudgment, and provides a reliable and clean data foundation for subsequent abnormal window identification and fault diagnosis. This significantly improves the accuracy, stability, and operational efficiency of intelligent metering analysis, achieving high-precision data purification and structured management in complex multi-source data environments.

[0099] Specifically: S300 includes: using a clean deduplication sequence as a time reference, combined with the remote parameter transmission record of the electricity meter, and converting discrete parameter events into continuous state trajectory curves to obtain a parameter state sequence;

[0100] In specific implementation, a set of remote parameter issuance records for the electricity meter is obtained. These records include parameter type identifiers, issuance times, effective times, and parameter change amplitudes. Then, time alignment processing is performed on the parameter records, mapping the parameter sequence to a unified time axis to form a discrete parameter event sequence. Based on this, a continuous state function is introduced to model parameter changes. This is achieved by constructing a step-superposition state function. Specifically, this state function is obtained as follows: First, the initial parameter state of the electricity meter is used as a baseline value. Each time a remote parameter change event occurs, the parameter change amplitude is multiplied by a unit step function. The trigger time of the unit step function is set as the time node of the corresponding parameter event. Then, the calculation results of all parameter change events are sequentially accumulated and finally superimposed onto the initial parameter state baseline value to obtain a parameter state value that changes continuously over time. This value is used to transform scattered and discrete remote parameter issuance events into a continuous and computable time-series state trajectory, achieving time alignment between parameter events and metering time sequences. Simultaneously, it provides continuous parameter state foundation data for subsequent metering sequence and parameter trajectory coupling segmentation, extraction of interval features, and identification of time state misalignment between parameters and metering.

[0101] The unit step function is a time-triggered logic switch function. Before the set trigger time, the function takes the value of 0; after the set trigger time, the function immediately jumps to 1 and remains unchanged at 1. It is used to precisely control the start time of each parameter change, thereby piecing together a complete and continuous parameter timing trajectory.

[0102] It should be noted that the parameters are kept in an initial fixed state under normal circumstances. At each time a parameter is changed, the parameter value will undergo a step jump, and the jump magnitude is equal to the parameter adjustment amount. After the jump, the parameter will maintain the new value and will not fall back. Multiple parameter changes will form multiple step jumps superimposed. The unit step function is used to precisely control when each parameter change starts and then takes effect permanently, which is in line with the actual operating law of the meter maintaining the new configuration for a long time after the remote parameters are issued.

[0103] Subsequently, the function is sampled at the measurement time point to obtain the parameter state sequence, and then bound to the clean deduplication sequence to form a preliminary measurement-parameter mapping pair; this mapping pair serves as the input basis for the next sub-step of segmented modeling.

[0104] Parameter type identifiers refer to the classification numbers or type labels of various remote configuration parameters of the electricity meter, which are used to distinguish specific parameter types, such as rate parameters and metering threshold parameters. They serve to classify and identify parameter categories and avoid confusion between different parameters.

[0105] The issuance time refers to the moment when the main station operation and maintenance system remotely issues parameter commands to the electricity meter;

[0106] The effective time refers to the moment when the meter receives the parameter instruction, completes the update, and officially activates the new parameters.

[0107] The parameter change magnitude refers to the difference in parameter values ​​before and after remote parameter modification. The difference between the old and new parameter values ​​represents the extent of the remote parameter adjustment and serves as the core variable for subsequent trajectory modeling and step function superposition.

[0108] The parameter state sequence is a discrete time series formed by sampling the continuous parameter state trajectory function one by one at each sampling time point of the deduplicated measurement sequence, resulting in a one-to-one correspondence between time and parameter state.

[0109] The records of remote parameter distribution for electricity meters are obtained directly from three sources: the power master station system, the marketing and maintenance platform, and the concentrator background logs.

[0110] The parameter state transitions within the parameter state sequence are identified to divide the sequence into several state intervals. Statistical features of each state interval are calculated to analyze the structural features of the measurement behavior under different parameter states, so as to construct a state segmentation feature matrix.

[0111] In practice, the parameter state sequence is coupled point by point with the clean, deduplicated sequence to construct a coupled sequence. Based on this coupled sequence, the parameter state change points are identified, and the absolute change of the parameter at adjacent time points is calculated for each point in the parameter state sequence. When the absolute change of the parameter exceeds a preset jump threshold, it indicates that a jump has occurred in the parameter state value, which is determined as the state segmentation boundary. Thus, the entire time series is divided into several state intervals. For each interval, its internal measurement data subsequence is extracted, and the corresponding statistical characteristics are calculated, including the interval mean, interval variance, and interval rate of change, to characterize the distribution characteristics of the measurement data under the parameter state.

[0112] Meanwhile, the parameter state values ​​of each interval are used as an additional dimension and combined with the above statistical features to construct a state segment feature vector. All interval features are then summarized to form a state segment feature matrix. This matrix is ​​used to describe the structural features of the measurement behavior under different parameter states and serves as the input for state misalignment identification in the next sub-step.

[0113] Based on the state segmentation feature matrix, the time relationship between parameter changes and measurement response is analyzed to identify state misalignment phenomena, obtain the state misalignment vector, and fuse it with the ghost feature set to obtain a composite feature set.

[0114] In practice, for each parameter change event, the corresponding response time point is searched in the clean, deduplicated sequence. This response time point is determined by abrupt changes in interval features, i.e., it is determined to be a response point when adjacent interval features meet the following conditions:

[0115] Condition 1: The difference between the interval means of two adjacent intervals is greater than the preset threshold for judging a sudden change in the mean; Condition 2: The difference between the interval change rates of two adjacent intervals is greater than the preset threshold for judging a sudden change in the rate of change.

[0116] After determining the response time point, the offset between the response time point and the effective time is calculated. This represents how long the metering data lags behind after the parameters have taken effect. In other words, it represents the duration of time delay and misalignment in metering behavior after the parameters take effect. This is used to distinguish between normal delay, minor misalignment, and severe misalignment faults, providing core quantitative basis for identifying the delay in remote parameter issuance and the misalignment of metering status.

[0117] The response time point is the actual moment when the metering data changes significantly after the remote parameters of the meter officially take effect. Essentially, changing parameters does not immediately change the metering; there is a lag. The response time point is the moment when the metering data truly senses the parameter change and a characteristic jump occurs.

[0118] Next, all offsets are grouped into an offset sequence, and the offsets are bound to the corresponding interval features to construct a state misalignment vector, which is used to describe the temporal misalignment relationship between parameter changes and measurement behavior. Finally, the state misalignment vector is fused with the ghost feature set to construct a composite feature set for joint analysis of ghosting and state misalignment.

[0119] A conditional probability model is constructed based on historical data, and the current composite feature set is input into the conditional probability model to obtain the measurement anomaly probability at the corresponding time and generate a continuous anomaly probability sequence.

[0120] In this embodiment, the path repetition, differential variance, and structural identifier in the ghost feature set are uniformly encoded with the time offset and segmentation features in the state misalignment vector to form a unified feature representation. Subsequently, a conditional probability model is introduced to estimate the probability of an abnormal event occurring under given feature conditions. The model is automatically calculated according to Bayes' theorem. The features at the current moment are substituted into the model for calculation to generate the corresponding quantitative anomaly probability. This probability is used to quantify the anomaly risk at each moment. This calculation process is performed on the entire time series to form a continuous anomaly probability sequence.

[0121] To eliminate the interference of single-point fluctuations on the overall judgment, a sliding window aggregation process is performed on the continuous anomaly probability sequence;

[0122] The average anomaly probability of each sliding time window within a continuous anomaly probability sequence is calculated. If the average anomaly probability exceeds a preset probability threshold, the corresponding window is marked as an anomaly interval. After filtering and merging, an anomaly window set is formed. Each window contains a start and end time and its corresponding set of composite features. The average anomaly probability refers to the average value of the measured anomaly probability within the corresponding window.

[0123] Extract the state misalignment vector and ghosting feature set corresponding to all time intervals within the abnormal window set, and perform collaborative relationship analysis on the composite features within each interval to calculate the collaborative coefficient.

[0124] In this embodiment, for each interval within the abnormal window set, the state misalignment vector and ghost feature set corresponding to all time points are extracted, and the covariance between the two is calculated. At the same time, their respective standard deviations are calculated, and the standard deviations of the state misalignment vector and the ghost feature set are multiplied together as the denominator, and the covariance is used as the numerator to obtain the synergy coefficient. This coefficient is used to characterize the degree of linkage between the two types of features in the same interval.

[0125] In this embodiment of the invention, in step S300, by deeply coupling the clean deduplication sequence with the remote parameter transmission record of the electricity meter, a refined mapping and continuous modeling between metering data and parameter status is achieved, thereby effectively improving the accuracy and sensitivity of electricity metering anomaly detection and fault identification.

[0126] Specifically, this step first aligns the remote parameter transmission records of the electricity meter with time, mapping discrete transmission events to a unified time axis. Then, a step superposition function is used to transform each parameter change into a continuous state trajectory, ensuring that the effective time and duration of each parameter state are accurately reflected. This allows scattered, discrete remote parameter events to form a computable continuous parameter sequence. This continuous processing not only guarantees the time alignment of parameter changes with metering data but also provides stable foundational data for subsequent segmented modeling.

[0127] Subsequently, the parameter state sequence is coupled point-by-point with the clean, deduplicated sequence. By identifying state transition points, the time series is divided into several parameter state intervals. Within each interval, statistical features such as the interval mean, variance, and rate of change are extracted, and the corresponding parameter state values ​​are added to the feature vector to construct a state segmentation feature matrix. This matrix can characterize the structural features of metering behavior under different parameter states. The construction of this matrix not only supports dynamic analysis of metering data but also quantifies response lag and state misalignment. By comparing the parameter effective time with the response time to sudden changes in metering data, a state misalignment vector is generated, reflecting the lag time of metering behavior after remote parameter adjustments. This distinguishes between normal delays, minor misalignments, and severe misalignment faults, providing accurate quantitative basis for operation and maintenance.

[0128] Furthermore, the state misalignment vector is fused with the previously obtained ghosting feature set to form a composite feature set, providing multi-dimensional feature input for anomaly detection. This enables the system to accurately assess the anomaly probability of measurement data at each moment, considering ghosting data, duplicate paths, and parameter lag. A continuous anomaly probability sequence is calculated using a conditional probability model, and sliding window aggregation is employed to eliminate single-point fluctuation interference, achieving anomaly window identification and obtaining a set of anomaly intervals with clear start and end times and complete features. State misalignment and ghosting features are further extracted from these anomaly intervals, and a coordination coefficient is calculated to characterize the degree of linkage between parameter state misalignment and redundant data, thereby achieving precise localization of anomaly events under the interaction of multiple factors.

[0129] Overall, this step not only transforms discrete parameter events into continuous state trajectories, ensuring the temporal consistency between metering data and parameter states, but also achieves comprehensive quantification of metering behavior under parameter changes through state segmentation feature matrices, state misalignment vectors, composite feature sets, and synergy coefficient analysis. This provides high-precision, multi-dimensional data support for subsequent anomaly window segmentation, fault category identification, and intelligent diagnosis, enabling the intelligent metering monitoring system to maintain high accuracy, reliability, and responsiveness in complex operating environments.

[0130] Specifically: S400 includes: extracting the state misalignment vector and ghosting feature set corresponding to all time intervals within the abnormal window set, and obtaining the ghosting intensity and misalignment intensity of each interval within the abnormal window set;

[0131] Based on the average intensity of ghosting features, the average intensity of state misalignment features, and the coordination coefficient, the intervals within the abnormal window set are classified to generate a fault label set.

[0132] In this embodiment, firstly, by using the state misalignment vector and ghost feature set corresponding to all times in each interval within the abnormal window set, the ghost intensity is obtained by averaging all ghost features within the interval through average calculation, and the misalignment intensity is obtained by averaging all state misalignment features within the interval.

[0133] If the ghost intensity exceeds the preset ghost threshold and the misalignment intensity does not exceed the misalignment threshold, it indicates that the ghost feature is strong and the parameter state misalignment is weak, and it is judged as a ghost fault.

[0134] If the ghost intensity does not exceed the preset ghost threshold and the misalignment intensity exceeds the misalignment threshold, it indicates that the state misalignment feature is strong and the ghost is weak, and it is judged as a parameter state misalignment fault.

[0135] If the ghost intensity does not exceed the preset ghost threshold and the misalignment intensity does not exceed the misalignment threshold, it indicates that the state misalignment feature is weak and the ghost is weak, and it is judged as a minor disturbance and not a serious fault.

[0136] If the ghost intensity exceeds the preset ghost threshold, the misalignment intensity exceeds the misalignment threshold, and the coordination coefficient exceeds the threshold coordination threshold, it indicates that the state misalignment feature is strong, the ghost is strong, and there is a high degree of linkage. Therefore, it is judged as a ghost and state misalignment coupled concurrent fault type.

[0137] If the ghost intensity exceeds the preset ghost threshold, the misalignment intensity exceeds the misalignment threshold, and the coordination coefficient does not exceed the threshold coordination threshold, it indicates that the state misalignment feature is strong, the ghost is strong, and they are highly uncoordinated. Therefore, it is determined to be a composite abnormal interval where the ghost and state misalignment are independent and concurrent and have no coupling relationship.

[0138] The fault label set includes ghosting faults, parameter state misalignment faults, minor disturbance non-serious faults, ghosting and state misalignment coupled concurrent faults, and ghosting and state misalignment independent concurrent faults without coupling.

[0139] Extract the data values, transmission paths, and time information of each interval within the fault label set from the path structure set to obtain a ternary subset. Based on the transmission path number, classify the data within the ternary subset according to the same transmission path to obtain a data subset.

[0140] Using the device identifiers of the corresponding transmission paths within the data subset as graph nodes and the transmission flow direction of the data subset from the previous node to the next node as edges, a topology propagation graph is constructed and a path propagation sequence is generated.

[0141] In this embodiment, for each interval within the fault label set, all triples whose time falls within the start and end range of the interval are selected to form a triple subset exclusive to the interval. Based on the path identifier, the data in the triple subset are classified according to the same transmission path, and each group corresponds to a data subset of a complete communication transmission path.

[0142] Subsequently, a topology propagation graph is constructed, and the fault label corresponding to the current interval is assigned and bound to each node of the propagation graph. Path attribute vectors are introduced to quantitatively characterize the data forwarding and abnormal fluctuation behavior of each transmission node in the graph.

[0143] Following the actual data transmission sequence, the occurrence and spread of anomalies from upstream nodes to downstream nodes are recorded. Nodes, tags, transmission time sequence, and attribute features are arranged into a sequence in order to generate the path propagation sequence corresponding to the interval. This sequence provides temporal and topological basis for subsequent calculation of path influence weights.

[0144] Based on the path propagation sequence, the degree of participation of each propagation path in all intervals within the fault label set is analyzed to obtain the path influence weight;

[0145] The set of paths corresponding to the largest path influence weight is selected as the critical path set, and the set of nodes with the largest path influence weight in each critical path in the critical path set is selected as the critical node set.

[0146] Based on the fault label set, the fault labels corresponding to each key node in the key node set are identified to obtain the location set; this location set is used to accurately locate the specific fault location and determine the fault type.

[0147] In this embodiment of the invention, in step S400, the invention achieves intelligent classification and precise location of metering faults by finely extracting and quantifying the state misalignment vectors and ghosting feature sets of each interval within the abnormal window set, thereby improving the power system's ability to identify and process complex anomalies.

[0148] Specifically, this step first calculates the average value of the state misalignment vector and ghosting feature set within each abnormal window interval to obtain the misalignment intensity and ghosting intensity. This processing not only quantifies the degree of misalignment between parameter state lag and measurement behavior, but also reflects the degree of ghosting caused by data redundancy or repeated transmission of links, providing an intuitive feature basis for distinguishing different types of faults. Subsequently, by combining the ghosting intensity, misalignment intensity, and the synergy coefficient of the two types of features, the system can accurately divide the abnormal interval into five categories: ghosting faults, parameter state misalignment faults, minor disturbance non-serious faults, ghosting and state misalignment coupled concurrent faults, and ghosting and state misalignment independent concurrent composite anomalies. This multi-dimensional and conditional classification logic not only takes into account the independence of a single abnormal feature, but also fully considers the linkage between the two types of features, thereby avoiding misjudgment and missed judgment, and enhancing the reliability of diagnosis.

[0149] Furthermore, this step utilizes a path structure set to categorize data within each fault interval into ternary subsets based on transmission paths. Using path numbers as nodes, a topology propagation graph is constructed. By recording the transmission direction and temporal order of data between network nodes, a path propagation sequence is generated. This operation not only visualizes the propagation trajectory of anomalies from upstream to downstream nodes but also provides clear temporal and topological basis for subsequent path influence weight calculations. Based on the path propagation sequence, the system further quantifies the participation degree of each transmission path within the fault interval. By identifying critical paths and key nodes through path influence weights, the system ultimately locks down the key nodes with fault labels, forming a location set and achieving a closed-loop mapping from anomaly characteristics to specific devices.

[0150] Overall, the S400 process, through multi-level and multi-dimensional feature extraction and analysis, not only accurately classifies abnormal windows and assigns fault labels, but also achieves precise location of fault sources and key nodes by combining topology propagation analysis. This upgrades the detection and diagnosis process of metering faults from traditional experience-based judgment to a systematic, quantitative, and traceable intelligent decision-making model, thereby effectively improving the safety, reliability, and operation and maintenance response speed of power operation. It also provides a scientific basis and operational guidance for subsequent fault handling, enabling the entire power metering system to achieve high-precision monitoring and proactive protection when facing complex transmission anomalies and parameter state misalignments.

[0151] Example 2, as Figure 2 As shown, based on the same inventive concept as the intelligent diagnostic method for metering faults provided in Embodiment 1, this embodiment of the invention also provides an intelligent diagnostic system for metering faults, including:

[0152] The data reconstruction module 11 is used to acquire metering data streams from the electricity meter, communication logs from the concentrator, and receiving records from the main station. After data preprocessing, a path structure set is obtained, and potential retransmission data is identified through the path structure set, and an initial annotation matrix is ​​constructed.

[0153] The structure recognition module 12 is used to identify the data structure type based on the initial annotation matrix through similarity clustering analysis, and obtain the ghosting feature set.

[0154] The anomaly diagnosis module 13 is used to analyze the structural characteristics of measurement behavior under different parameter states based on the ghost feature set, and to identify the set of abnormal windows in combination with historical data.

[0155] The positioning module 14 is used to classify fault categories by abnormal window set, generate path propagation sequence by combining data structure type, and lock the fault label of corresponding device according to the degree of participation of each propagation path, so as to complete the intelligent diagnosis of faults in electricity metering.

[0156] Furthermore, the data reconstruction module 11 is also used for:

[0157] Based on the metering data stream, concentrator communication log and main station reception record, the timestamp is corrected by unifying the time axis and using round-trip delay, while the communication behavior is parsed and encoded to obtain the initial quantity communication joint sequence.

[0158] Based on the initial communication sequence, the transmission path is reconstructed by parsing the frame sequence number and device identifier in the concentrator communication log, the path redundancy and time dispersion are quantified, and a path attribute vector is constructed to obtain a set of attributed path structures.

[0159] When the path repetition exceeds the preset repetition threshold, the corresponding data is marked as potential retransmission data, and the data is initially labeled in combination with the time dispersion to obtain the initial labeling matrix.

[0160] Furthermore, the structure recognition module 12 is also used for:

[0161] Based on the initial annotation matrix, similarity clustering analysis is performed on data within the same temporal neighborhood to obtain a set of candidate ghost clusters for ghost structure difference analysis, including:

[0162] Divide a fixed-width sliding time window based on the correction timestamp, calculate the similarity between measurement data within the sliding time window, and obtain a similarity matrix for identifying time-series redundant retransmission data;

[0163] The similarity matrix is ​​weighted and corrected by retrieving potential retransmission data from the initial annotation matrix, and the path attribute vector is used as a constraint to construct the clustering priority.

[0164] Based on clustering priority, density clustering algorithm is used to group and aggregate the weighted similarity matrix to obtain candidate clusters and generate a set of candidate ghost clusters.

[0165] Perform temporal difference analysis and statistics on each candidate cluster in the candidate ghosting cluster set to obtain the difference variance for cluster subdivision, and count the transmission path number corresponding to all data in a single candidate cluster to calculate the path distribution entropy.

[0166] If the difference variance does not exceed the preset variance threshold and the path distribution entropy does not exceed the preset entropy threshold, it is determined to be a replication ghost structure; otherwise, it is determined to be a variation normal structure, and a ghost structure identification matrix is ​​constructed.

[0167] Data types are distinguished based on the ghost structure identifier matrix. A differentiated weight fusion algorithm is used to obtain a pure deduplicated sequence. During the fusion process, ghost structure features are obtained and a ghost feature set is constructed.

[0168] Furthermore, the anomaly diagnosis module 13 is also used for:

[0169] Based on the pure deduplication sequence as the time reference, combined with the remote parameter transmission record of the electricity meter, the parameter state sequence is obtained by transforming discrete parameter events into continuous state trajectory curves.

[0170] The parameter state transitions within the parameter state sequence are identified to divide the sequence into several state intervals. Statistical features of each state interval are calculated to analyze the structural features of the measurement behavior under different parameter states, so as to construct a state segmentation feature matrix.

[0171] Based on the state segmentation feature matrix, the time relationship between parameter changes and measurement response is analyzed to identify state misalignment phenomena, obtain the state misalignment vector, and fuse it with the ghost feature set to obtain a composite feature set.

[0172] A conditional probability model is constructed based on historical data, and the current composite feature set is input into the conditional probability model to obtain the measurement anomaly probability at the corresponding time and generate a continuous anomaly probability sequence.

[0173] The average abnormal probability of each sliding time window within the continuous abnormal probability sequence is used. When the average abnormal probability exceeds a preset probability threshold, the corresponding window is marked as an abnormal interval. After filtering and merging, an abnormal window set is formed.

[0174] Extract the state misalignment vector and ghosting feature set corresponding to all time intervals within the abnormal window set, and perform collaborative relationship analysis on the composite features within each interval to calculate the collaborative coefficient.

[0175] Furthermore, the positioning module 14 is also used for:

[0176] Extract the state misalignment vector and ghosting feature set corresponding to all time intervals within the abnormal window set, and obtain the ghosting intensity and misalignment intensity of each interval within the abnormal window set;

[0177] Based on the average intensity of ghosting features, the average intensity of state misalignment features, and the coordination coefficient, the intervals within the abnormal window set are classified to generate a fault label set.

[0178] Extract the data values, transmission paths, and time information of each interval within the fault label set from the path structure set to obtain a ternary subset. Based on the transmission path number, classify the data within the ternary subset according to the same transmission path to obtain a data subset.

[0179] Using the device identifiers of the corresponding transmission paths within the data subset as graph nodes and the transmission flow direction of the data subset from the previous node to the next node as edges, a topology propagation graph is constructed and a path propagation sequence is generated.

[0180] Based on the path propagation sequence, the degree of participation of each propagation path in all intervals within the fault label set is analyzed to obtain the path influence weight;

[0181] The set of paths corresponding to the largest path influence weight is selected as the critical path set, and the set of nodes with the largest path influence weight in each critical path in the critical path set is selected as the critical node set.

[0182] Based on the fault label set, the fault labels corresponding to each key node in the key node set are identified to obtain the location set.

[0183] It should be noted that the descriptions of each embodiment in the above embodiments have different focuses. For parts that are not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0184] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0185] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0186] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0187] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0188] Although preferred embodiments of the invention have been described, those skilled in the art, once they have learned the basic inventive concept, can make other changes and modifications to these embodiments.

[0189] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of this invention and its equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for intelligent diagnosis of metering faults in electricity meters, characterized in that, The method includes: The metering data stream, concentrator communication log, and main station reception record are acquired. After data preprocessing, a path structure set is obtained. Through the path structure set, potential retransmission data is identified, and an initial annotation matrix is ​​constructed. Based on the initial annotation matrix, similarity clustering analysis is used to identify the data structure type and obtain the ghosting feature set; Based on the ghosting feature set, the structural characteristics of the measurement behavior under different parameter states are analyzed, and combined with historical data, the abnormal window set is identified. By classifying fault categories through anomaly window sets and combining them with data structure types, a path propagation sequence is generated. Based on the degree of participation of each propagation path, the fault label of the corresponding device is locked to complete the intelligent diagnosis of faults in electricity metering.

2. The intelligent diagnostic method for metering faults in electricity meters according to claim 1, characterized in that, The system acquires metering data streams from the electricity meter, concentrator communication logs, and master station reception records. After data preprocessing, it obtains a path structure set. Based on this path structure set, it identifies potential retransmissions and constructs an initial annotation matrix, including: Based on the metering data stream, concentrator communication log and main station reception record, the timestamp is corrected by unifying the time axis and using round-trip delay, while the communication behavior is parsed and encoded to obtain the initial quantity communication joint sequence. Based on the initial communication sequence, the transmission path is reconstructed by parsing the frame sequence number and device identifier in the concentrator communication log, the path redundancy and time dispersion are quantified, and a path attribute vector is constructed to obtain a set of attributed path structures. When the path repetition exceeds the preset repetition threshold, the corresponding data is marked as potential retransmission data, and the data is initially labeled in combination with the time dispersion to obtain the initial labeling matrix.

3. The intelligent diagnostic method for metering faults in electricity meters according to claim 2, characterized in that, Based on the initial annotation matrix, similarity clustering analysis is used to identify the data structure type, resulting in a set of ghosting features, including: Based on the initial annotation matrix, similarity clustering analysis is performed on the data in the same time neighborhood to obtain a set of candidate ghost clusters for ghost structure difference analysis; Perform temporal difference analysis and statistics on each candidate cluster in the candidate ghosting cluster set to obtain the difference variance for cluster subdivision, and count the transmission path number corresponding to all data in a single candidate cluster to calculate the path distribution entropy. If the difference variance does not exceed the preset variance threshold and the path distribution entropy does not exceed the preset entropy threshold, it is determined to be a replication ghost structure; otherwise, it is determined to be a variation normal structure, and a ghost structure identification matrix is ​​constructed. Data types are distinguished based on the ghost structure identifier matrix. A differentiated weight fusion algorithm is used to obtain a pure deduplicated sequence. During the fusion process, ghost structure features are obtained and a ghost feature set is constructed.

4. The intelligent diagnostic method for metering faults in electricity meters according to claim 3, characterized in that, Based on the initial annotation matrix, similarity clustering analysis is performed on data within the same temporal neighborhood to obtain a set of candidate ghost clusters for ghost structure difference analysis, including: Divide a fixed-width sliding time window based on the correction timestamp, calculate the similarity between measurement data within the sliding time window, and obtain a similarity matrix for identifying time-series redundant retransmission data; The similarity matrix is ​​weighted and corrected by retrieving potential retransmission data from the initial annotation matrix, and the path attribute vector is used as a constraint to construct the clustering priority. Based on clustering priority, density clustering algorithm is used to group and aggregate the weighted similarity matrix to obtain candidate clusters and generate a set of candidate ghost clusters.

5. The intelligent diagnostic method for metering faults in electricity meters according to claim 4, characterized in that, Based on the ghosting feature set, the structural characteristics of metrological behavior under different parameter states are analyzed, and combined with historical data, anomaly window sets are identified, including: Based on the pure deduplication sequence as the time reference, combined with the remote parameter transmission record of the electricity meter, the parameter state sequence is obtained by transforming discrete parameter events into continuous state trajectory curves. The parameter state transitions within the parameter state sequence are identified to divide the sequence into several state intervals. Statistical features of each state interval are calculated to analyze the structural features of the measurement behavior under different parameter states, so as to construct a state segmentation feature matrix. Based on the state segmentation feature matrix, the time relationship between parameter changes and measurement response is analyzed to identify state misalignment phenomena, obtain the state misalignment vector, and fuse it with the ghost feature set to obtain a composite feature set.

6. The intelligent diagnostic method for metering faults in electricity meters according to claim 5, characterized in that, Based on the ghosting feature set, the structural characteristics of metrological behavior under different parameter states are analyzed, and combined with historical data, abnormal window sets are identified, including: A conditional probability model is constructed based on historical data, and the current composite feature set is input into the conditional probability model to obtain the measurement anomaly probability at the corresponding time and generate a continuous anomaly probability sequence. The average abnormal probability of each sliding time window within the continuous abnormal probability sequence is used. When the average abnormal probability exceeds a preset probability threshold, the corresponding window is marked as an abnormal interval. After filtering and merging, an abnormal window set is formed. Extract the state misalignment vector and ghosting feature set corresponding to all time intervals within the abnormal window set, and perform collaborative relationship analysis on the composite features within each interval to calculate the collaborative coefficient.

7. The intelligent diagnostic method for metering faults in an electricity meter according to claim 6, characterized in that, The process of classifying fault categories using an abnormal window set includes: Extract the state misalignment vector and ghosting feature set corresponding to all time intervals within the abnormal window set, and obtain the ghosting intensity and misalignment intensity of each interval within the abnormal window set; Based on the average intensity of ghosting features, the average intensity of state misalignment features, and the coordination coefficient, the intervals within the abnormal window set are classified to generate a fault label set.

8. The intelligent diagnostic method for metering faults in an electricity meter according to claim 7, characterized in that, The path propagation sequence generation process includes: Extract the data values, transmission paths, and time information of each interval within the fault label set from the path structure set to obtain a ternary subset. Based on the transmission path number, classify the data within the ternary subset according to the same transmission path to obtain a data subset. Using the device identifiers of the corresponding transmission paths within the data subset as graph nodes and the transmission flow direction of the data subset from the previous node to the next node as edges, a topology propagation graph is constructed and a path propagation sequence is generated.

9. The intelligent diagnostic method for metering faults in an electricity meter according to claim 8, characterized in that, Based on the level of involvement of each propagation path, the corresponding fault labels of the devices are identified, including: Based on the path propagation sequence, the degree of participation of each propagation path in all intervals within the fault label set is analyzed to obtain the path influence weight; The set of paths corresponding to the largest path influence weight is selected as the critical path set, and the set of nodes with the largest path influence weight in each critical path in the critical path set is selected as the critical node set. Based on the fault label set, the fault labels corresponding to each key node in the key node set are identified to obtain the location set.

10. An intelligent diagnostic system for electricity meter metering faults, used to implement the intelligent diagnostic method for electricity meter metering faults as described in any one of claims 1 to 9, characterized in that, include: The data reconstruction module is used to acquire metering data streams from the electricity meter, communication logs from the concentrator, and receiving records from the main station. After data preprocessing, a path structure set is obtained. Through the path structure set, potential retransmission data is identified, and an initial annotation matrix is ​​constructed. The structure recognition module is used to identify the data structure type based on the initial annotation matrix through similarity clustering analysis, and obtain the ghosting feature set. The anomaly diagnosis module is used to analyze the structural characteristics of metrological behavior under different parameter states based on the ghosting feature set, and to identify the set of abnormal windows by combining historical data. The positioning module is used to classify fault categories by abnormal window set, generate path propagation sequence by combining data structure type, and lock the fault label of corresponding device according to the degree of participation of each propagation path, so as to complete the intelligent diagnosis of faults in electricity metering.

Images