Multi-source ammeter data intelligent fusion and abnormity identification method

Abstract

The invention provides a multi-source ammeter data intelligent fusion and abnormity identification method. According to the method, original electric parameter data of a plurality of intelligent electric meters are collected under a unified time reference, a cross-equipment measurement data set indexed according to electric meter numbers is constructed, and electric parameter classification and associated information generation are completed in combination with electric wiring information and functional parameters; an electric parameter change curve is extracted through a continuous time window, a value change consistency index is calculated, a response offset feature vector is further introduced for joint judgment, and accurate recognition of an abnormal electric meter and an abnormal time period is achieved; for the intelligent electric meter which is identified as abnormal, constructing an estimated value through data of other electric meters, and calculating residual errors between the estimated value and an original value and between the estimated value and an adjacent estimated value according to a main conductive parameter so as to realize dual residual error verification; judging an anomaly type according to the consistency index and the residual result, and outputting an anomaly recognition result; according to the method, the anomaly detection precision in a multi-table collaboration scene is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of electricity meters, and in particular, to a method for intelligent fusion and anomaly recognition of multi-source electricity meter data. Background Art

[0002] In the prior art, for the smart electricity meters widely deployed in the distribution system, anomaly detection methods based on power balance, meter comparison, or single-meter event analysis have been widely applied. Some technical solutions calculate the deviation based on the energy difference between the total meter and the user sub-meter in the same area, or identify anomalies by analyzing the voltage and current change trends of a single electricity meter; other solutions classify specific anomaly types using fixed rules or simple threshold settings. These methods have achieved certain effects in practical applications and have been integrated into the distribution automation system.

[0003] However, the prior art generally has the problem of insufficient data fusion ability when dealing with heterogeneous electricity meter data, lacks a systematic modeling of the electrical parameter logical relationship between multi-source measurement data, and is difficult to effectively extract the consistent evolution characteristics across electricity meters and time windows. At the same time, anomaly judgment mostly relies on single features or rough rules, and it is difficult to meet the metering anomaly detection requirements in areas with dynamic load changes or complex topological structures, and it is easy to have false positives or false negatives, affecting the system diagnosis accuracy and response efficiency.

[0004] In view of this, there is an urgent need to provide a data fusion and recognition method that can be oriented to multi-source heterogeneous smart electricity meters and has the ability of spatio-temporal collaborative analysis and anomaly information reconstruction mechanism. Summary of the Invention

[0005] The present application provides a method for intelligent fusion and anomaly recognition of multi-source electricity meter data to improve the diagnosis accuracy and traceability of hidden faults in the distribution system.

[0006] The present application provides a method for intelligent fusion and anomaly recognition of multi-source electricity meter data, including: Under a unified time reference, collecting original electrical parameter data from multiple heterogeneous smart electricity meters, and organizing the original electrical parameter data into a cross-device measurement data set indexed by the electricity meter number; According to the electrical wiring information and functional parameters of each smart electricity meter, classifying and marking each item of the original electrical parameter data in the cross-device measurement data set to generate the association information between electrical parameters; Segmenting the cross-device measurement data set according to continuous time windows, extracting the electrical parameter change curves of each smart electricity meter, and calculating the numerical change consistency index within each time window based on the association information to obtain a consistency index sequence; Calculate the response offset eigenvector between the curves of the electrical parameter changes, and make a combined judgment in combination with the consistency index to identify the abnormal smart meter numbers and their corresponding continuous time windows; For the smart meter numbers identified as abnormal, call the original electrical parameter data of other smart meters to generate estimated electrical parameter values, and calculate the first residual between the estimated value and the original value according to the dominant electrical parameter, as well as the second residual between the estimated value and the adjacent estimated value; According to the consistency index, the first residual and the second residual, determine whether there are wiring errors, reverse current connections, measurement abnormalities or other abnormal types, and output an abnormal recognition result record including the abnormal type, smart meter number and continuous time window.

[0007] The beneficial effects of the technical solution provided by this application include: (1) By constructing a cross-device measurement data set and electrical parameter correlation information, the data of single-phase, three-phase and two-way metering smart meters are effectively integrated, and the scope of application and accuracy of data consistency analysis are improved. (2) The introduction of the electrical parameter change consistency index and the response offset eigenvector can identify abnormalities from two dimensions of time series change and topological relationship, and significantly improve the recognition accuracy of hidden faults such as wiring errors and reverse current connections. (3) By calculating the first residual between the estimated value and the original value and the second residual between the estimated values at the same time, the misjudgment risk caused by a single error index is reduced, and more reliable abnormality confirmation is achieved. (4) It can accurately identify the time section and associated meter numbers where the abnormality occurs, and provide efficient and traceable data support for fault tracing, status assessment and operation and maintenance scheduling of the distribution system. Description of the Drawings

[0008] Figure 1 is a flowchart of a method for intelligent fusion and abnormality recognition of multi-source meter data provided by the first embodiment of this application. Detailed Embodiment

[0009] Many specific details are set forth in the following description in order to provide a thorough understanding of this application. However, this application can be implemented in many other ways different from those described herein, and those skilled in the art can make similar generalizations without departing from the connotation of this application. Therefore, this application is not limited by the specific implementations disclosed below.

[0010] The first embodiment of this application provides a method for intelligent fusion and abnormality recognition of multi-source meter data. Please refer to Figure 1 , which is a schematic diagram of the first embodiment of this application. The following will be described in detail Figure 1 a method for intelligent fusion and abnormality recognition of multi-source meter data provided by the first embodiment of this application.

[0011] Step S101: Collect original electrical parameter data from multiple heterogeneous smart meters under a unified time reference. The heterogeneous smart meters include single-phase smart meters, three-phase smart meters, and composite smart meters supporting two-way metering. The original electrical parameter data includes instantaneous voltage, current, active power, reactive power, electrical energy, and power factor, and organize the original electrical parameter data into a cross-device measurement data set indexed by device number.

[0012] When implementing Step S101, it is necessary to synchronously collect original electrical parameter data from multiple heterogeneous smart meters under a unified time reference. The so-called unified time reference means that the data collection time points of all meters must be aligned during the entire data collection period, that is, data is obtained at the same time stamp to ensure the comparability of data between different meters. This step should rely on a high-precision clock synchronization mechanism, such as achieving clock alignment of each meter in the system through the IEEE 1588 Precision Time Protocol (PTP) or the Network Time Protocol (NTP), ensuring that the sampling start time and interval of all devices are consistent.

[0013] The multiple heterogeneous smart meters include, but are not limited to, three types: The first type is a single-phase smart meter, which is suitable for ordinary residences or light-load terminals and collects voltage, current, and power information in a single-phase circuit; the second type is a three-phase smart meter, which is suitable for commercial, industrial loads, or polyphase power supply scenarios and can simultaneously collect all electrical parameters of phases A, B, and C; the third type is a composite smart meter supporting two-way metering, which can measure both power consumption and reverse power supply and is suitable for distributed power grid connection environments to capture two-way power flow information.

[0014] The collected original electrical parameter data should at least include the following: instantaneous voltage, instantaneous current, active power, reactive power, electrical energy, and power factor. Specifically, instantaneous voltage and current can obtain the effective value of each phase at the current moment through high-frequency sampling; active power and reactive power are calculated through the instantaneous phase relationship between voltage and current; electrical energy is the integral of power over unit time; the power factor represents the cosine value between the current phase and the voltage phase, reflecting the nature of the load. Each of the above electrical parameters needs to be marked with structured fields such as the corresponding meter number, phase (such as phase A / B / C), measurement channel (such as incoming / outgoing line), sampling time stamp, etc., to ensure the accurate traceability and subsequent processing of the data.

[0015] All the collected electrical parameter data needs to be sorted according to the unique identification code of the device and organized into a cross-device measurement data set indexed by device number. Each device number should uniquely correspond to an intelligent electricity meter, for example, distinguished by its device ID, installation location code, or electricity meter communication address. The structural form of this cross-device measurement data set can adopt a two-dimensional or three-dimensional matrix. Each row of the matrix represents a time point, and the columns represent different electrical parameter fields or combinations of electricity meter numbers, so that at any time point, the electrical parameters corresponding to all electricity meters can be quickly retrieved to achieve efficient comparison of cross-device data.

[0016] In specific engineering deployments, an edge computing device or a concentrator device can be configured as a data aggregation node to periodically poll the measurement channels of all electricity meters and cache the raw data. This cached data is uploaded to the master station system after preliminary verification and participates in subsequent fusion analysis and processing. To ensure data integrity and timeliness, an automatic re-sampling strategy for data loss should also be set, and abnormal marks should be made for communication anomalies, error codes, or resume data transfer to ensure the formation of a raw measurement data set with clear structure, consistent time, and complete content.

[0017] By implementing step S101 in the above manner, it can provide a unified, accurate, and traceable multi-source heterogeneous electricity meter raw measurement basis for subsequent data fusion, classification marking, consistency analysis, and anomaly identification. The accuracy and synchronization of this step directly determine the reliability of the entire system's fusion and identification capabilities.

[0018] Step S102: Classify and mark the raw electrical parameter data in the cross-device measurement data set according to the electrical wiring information and functional parameters of each intelligent electricity meter to generate the association information between electrical parameters.

[0019] When implementing step S102, it is necessary to further combine the electrical wiring information and functional parameters corresponding to each intelligent electricity meter for the cross-device measurement data set that has been uniformly collected and archived by device number in step S101, and conduct systematic classification and marking on the collected raw electrical parameter data to achieve clear distinction and semantic understanding of the data meaning, source, and context relationship during subsequent analysis.

[0020] The so-called electrical wiring information refers to the physical wiring method, access type, connected load type, and phase relationship of each smart meter in the power system. For example, a three-phase smart meter may adopt a three-phase four-wire wiring method, and its corresponding electrical parameters should include the voltage, current, active power, and reactive power of phase A, phase B, and phase C. A single-phase smart meter may only measure the voltage and current between L and N. For a composite smart meter that supports bidirectional metering, it is necessary to clarify its forward and reverse power metering channels, and whether it has harmonic analysis, power factor correction, or independent three-phase metering capabilities. The above wiring information is usually entered into the system during the installation of the meter, or can be confirmed by the master station system through an automatic recognition program.

[0021] Functional parameters include the technical specifications of the meter, measurement accuracy level, sampling frequency, data communication method, types of electrical parameters supported, measurement range, etc., which are used to judge the trust level and applicable range of a certain electrical parameter. For example, when analyzing the change trend of reactive power within a certain period, if a certain type of meter only supports active power measurement or its reactive power accuracy is insufficient, the system should automatically block its reactive power data from participating in subsequent analysis to avoid introducing deviations.

[0022] Specifically, for the set of original electrical parameter data corresponding to each smart meter number, the system needs to establish a structured tag system, label its original data fields according to information such as electrical parameter type (such as voltage, current, power, etc.), phase (such as phase A, phase B, phase C, or single-phase), measurement direction (forward, reverse), data nature (instantaneous value, average value, integral value, etc.), measurement point location (incoming line side, outgoing line side), sampling frequency, metering accuracy level, etc., and record meta-information such as the associated wiring method, device role (such as main meter, branch meter, relay meter), and its affiliated regional topology number.

[0023] Finally, based on the above classification tags and label management mechanism, the system will generate an electrical parameter association information table, which establishes a functional relationship mapping between electrical parameters across meters, channels, phases, and directions. For example, it will indicate that the power factor of a three-phase load should be jointly derived from the active and reactive powers of phase A / B / C, or point out logical relationships such as the missing phase C current data of a meter can be compensated by referring to the phase matching data of the adjacent meter.

[0024] The successful implementation of this step provides a solid data structure and semantic basis for subsequent operations such as time series consistency analysis, extraction of response differences between multiple meters, and residual verification, ensuring that the system can achieve accurate and controllable fusion analysis and anomaly recognition in a complex data environment.

[0025] The following provides a specific example: Suppose there are three smart meters in the power distribution system, numbered M1001, M1002, and M1003 respectively. Among them, M1001 is a three-phase four-wire meter installed at the main incoming line of a commercial building, M1002 is a single-phase smart meter installed on a certain branch of this building, and M1003 is a two-way metering meter connected to the distributed photovoltaic grid-connected interface. After the system executes step S101, the original electrical parameter data of each meter at a certain unified time point has been collected, including content such as voltage, current, active power, reactive power, electrical energy, and power factor.

[0026] In step S102, the system first retrieves the pre-configured wiring information and function parameters in the meter management platform. For example, it is identified that M1001 is a three-phase meter, supporting A / B / C three-phase voltage and current sampling, with a metering accuracy of 0.5S level; M1002 is a single-phase meter, only having the ability to sample L-N voltage and current, with a sampling frequency of 15 minutes; M1003 is a composite meter supporting forward and reverse metering, whose active power and reactive power can both distinguish the directionality and support high-frequency data transmission at the 1-second level. Subsequently, based on the preset marking template, the system annotates each of the above data fields item by item and classifies them into structured measurement semantic tags.

[0027] For example, for the field "I_B" of M1001, the system marks it as: "M1001, three-phase four-wire, phase B current, instantaneous value, forward channel, main trunk node, sampling frequency 1Hz, accuracy 0.5S", while for the field "P_reverse" of M1003, it is marked as: "M1003, single-phase two-way, reverse active power, average value, distributed interface, sampling frequency 1Hz, accuracy 1.0 level".

[0028] In the finally generated electrical parameter correlation information table of the system, the A / B / C phase power data of M1001 is marked as "three-phase synthesis source", the data of M1002 is marked as "branch reference point", the forward and reverse power of M1003 is marked as "boundary flow point", and it is clear that there is a "space-time comparison relationship" with the upstream node M1001, thus supporting data mapping and abnormal linkage analysis between subsequent multi-source meters.

[0029] Step S103: Segment the cross-device measurement data set according to continuous time windows, extract the electrical parameter change curves of each smart meter, and calculate the numerical change consistency index within each time window based on the correlation information to obtain a consistency index sequence.

[0030] When implementing step S103, it is necessary to perform sequence segmentation processing on the cross-device measurement data set constructed in step S101 based on the structured electrical parameter tags generated in step S102 and their associated information in the time dimension, so as to extract the change curves reflecting the evolution behavior of the electrical parameters of each smart meter, and on this basis, quantify the co-variation characteristics of the electrical parameters between different meters within the same time window.

[0031] First of all, the system should set a time window parameter with a fixed length. For example, taking 15 minutes, 1 minute or 10 seconds as the unit, the measurement data set is divided into multiple non-overlapping or sliding overlapping time periods according to the sampling frequency of the smart meter. Each time window contains a number of consecutive sampling points to ensure that each meter has a complete record of the original electrical parameters during this time period. If there is data missing for a certain meter in a certain window, it can be filled in by time proximity interpolation, mean filling or structured missing marking to ensure that the data set structure for analysis is consistent and the dimensions are unified.

[0032] Next, within each time window, the system extracts the time series data of the target electrical parameters according to the classified and marked fields. For example, it extracts the active power curve of phase A, the total reactive power curve, the power factor curve, etc. of all meters. These curve data are not mainly based on single-point values, but on the sequence trajectories formed within the window, and the change trajectories of each electrical parameter of each device within the current time period are constructed. The system should perform normalization processing on these trajectories so that the values of different types of meters (such as single-phase and three-phase) can be compared under the same dimension.

[0033] In order to evaluate the operation consistency of multiple smart meters within the same time window, it is necessary to calculate the numerical change consistency index between devices based on the correlation information between electrical parameters. This index is not a simple difference or average error, but reflects the degree of coordination between multiple electrical parameter trajectories. The calculation method can be implemented in various ways such as sliding correlation coefficient, dynamic time warping (DTW) distance, average slope deviation, cross standard deviation, etc. For example, if three meters are connected in the same branch and have highly correlated load change characteristics, then the phase A current curves of them should have a synchronous rising or falling trend; if one of the meters suddenly shows abnormal fluctuations or no response, its consistency index will decrease significantly.

[0034] The system can construct a consistency score vector for each group of electrical parameters, including the correlation measurement values between each pair of meters within the current window, and then perform weighted fusion on the consistency scores of all electrical parameters to generate a numerical index representing the overall consistency level of this time window, that is, the consistency index. This index can be a real number or a grading level, used to mark the possibility of abnormal behavior occurring.

[0035] This consistency index sequence will ultimately form a continuous time trajectory over time, serving as an important basis for identifying abnormal trends and diagnosing the starting points of anomalies. The subsequent step S104 will further combine the response offset characteristics based on this sequence for anomaly judgment. Therefore, step S103 not only realizes the time structuring and sequence abstraction of measurement data, but also lays the core foundation for behavior modeling and quantitative evaluation in the entire intelligent identification process.

[0036] To facilitate understanding of the specific implementation of step S103, an example is provided below for illustration. Suppose three smart meters are deployed in a certain power distribution area, numbered M2001, M2002, and M2003 respectively. Among them, M2001 is the main meter of the power distribution area, and M2002 and M2003 are the sub-meters of two branch users respectively. All three support data reporting at the minute level, and the collected electrical parameters include fields such as three-phase current, active power, and power factor. The system sets the time window length to 15 minutes and divides all data into continuous time window segments in a non-overlapping segmentation manner.

[0037] Within the time window from 10:00 to 10:15 on May 1, 2025, the system extracts the A-phase active power curves of the three meters respectively, and confirms that all three meters are three-phase four-wire systems and belong to the main and branch nodes of the same power supply branch based on the label information provided in step S102. Next, the system normalizes each power curve so that its value range within this window is normalized to the interval [0, 1], and uses the sliding correlation coefficient method to calculate the correlation degrees of power change trends between M2001 and M2002, M2001 and M2003, and M2002 and M2003 respectively. Suppose the obtained correlation coefficients are 0.93, 0.89, and 0.15 respectively, and the system immediately judges that there is a significant deviation between M2002 and M2003.

[0038] Subsequently, the system synthesizes the three groups of correlation values to construct a consistency scoring vector of [0.93, 0.89, 0.15], and obtains the consistency index value of the current time window as 0.656 through weighted average, which is lower than the set normal threshold (such as 0.80). The system records this index value and writes it into the consistency index sequence for subsequent steps to identify the abnormal occurrence points and infer the abnormal evolution trends.

[0039] Step S104: Calculate the response offset feature vectors between the change curves of the electrical parameters, and perform a joint judgment in combination with the consistency index to identify the abnormal smart meter numbers and their corresponding continuous time windows.

[0040] During the implementation of step S104, the system needs to further analyze the response differences between the trajectories of electrical parameter changes based on the obtained consistency index sequence. By introducing the response offset feature vector, it captures the micro-timing mismatch, hysteretic change, or sudden offset in the behavior of the electricity meters, thereby enhancing the sensitivity and accuracy of anomaly recognition.

[0041] Specifically, the meaning of response offset refers to the time difference in the synchronous changes of the measured values of different electricity meters when facing events such as system load changes, power supply disturbances, or operation state switches. In theory, if multiple smart electricity meters are located at consecutive positions on the same substation topology path, their active power or current should respond approximately synchronously to external disturbances. However, in actual scenarios, due to communication delays, device response time differences, or wiring differences, the change trajectories of some electricity meters may have significant lags or leads. If this time offset does not match their physical topology relationship, it may reflect potential problems such as abnormal wiring, clock drift, or data errors.

[0042] When calculating the response offset feature vector, the system first selects a certain key electrical parameter as the analysis object, such as active power or phase A current, and extracts the change curves corresponding to all electricity meters within each time window. On this basis, the system precisely calculates the response time difference between each electricity meter by matching the change trends and inflection point positions of the curves, using the sliding window delay comparison method or the cross-correlation peak positioning method. This time difference is expressed in milliseconds or the number of sampling points, and constitutes a response offset feature vector indexed by the electricity meter number.

[0043] For example, within a 15-minute time window, if the main electricity meter M3001 in the substation and two branch electricity meters M3002 and M3003 all show a load increase, but the response of M3002 lags behind M3001 by 2 sampling points, while the response of M3003 is 1 sampling point ahead, then the response offset feature vector can be expressed as {M3001:0, M3002:+2, M3003:-1}. If M3003 is physically located downstream of M3001, its "ahead response" behavior will be regarded as an abnormal offset, thus triggering further review by the system.

[0044] Subsequently, the system jointly judges the obtained response offset feature vector and the consistency index sequence calculated in step S103. On the one hand, if a certain electricity meter shows a large response offset and a low consistency index within a certain time window, it indicates that the behavior of this electricity meter is significantly inconsistent with that of other electricity meters, and the possibility of anomaly is significantly increased; on the other hand, even if the consistency index is high, but the response offset value seriously does not match the expected topology response logic, it can also indicate the existence of structural anomalies. The system can set double-condition judgment rules, such as "the response offset exceeds ±2 sampling points and the consistency index is lower than 0.6" to trigger a preliminary anomaly mark.

[0045] Through this determination mechanism that combines response behavior with change consistency, the system can identify problems that are difficult to detect by traditional energy difference methods, such as incorrect wiring order, abnormal mapping of meter addresses, sampling time-base drift, and local power supply disturbances. The final output of the abnormal judgment result will include the number of the smart meter, the number of the continuous time window when the abnormality occurs, and the corresponding response offset value range, providing a data basis for subsequent residual verification and abnormal classification.

[0046] The design of this step ensures that the abnormal identification process not only focuses on the differences between values, but also comprehensively considers the dynamic behavior of electrical parameters in the time dimension, making the abnormal judgment have higher time-series sensitivity and topological interpretability.

[0047] To facilitate the understanding of this embodiment, an example is provided below. Suppose a distribution transformer area contains three smart meters numbered M3001, M3002, and M3003. Among them, M3001 is the main meter of the main line, and M3002 and M3003 are the user sub-meters of two end branches respectively. The three form a typical topological structure of main-branch-branch in terms of physical wiring. In a certain continuous time window, the system detects that the active power of phase A of all three meters shows a significant upward trend.

[0048] The system first extracts the rising inflection point positions of each meter based on the electrical parameter change curves within this time window. Through the cross-correlation peak localization method, the system finds that the power increase of M3001 appears at time point T+4, M3002 appears at T+6, and M3003 appears at T+2. Taking M3001 as the reference, the response of M3002 lags by 2 sampling points, and the response of M3003 advances by 2 sampling points, thus forming a response offset feature vector of {M3001: 0, M3002: +2, M3003: –2}.

[0049] Then the system consults the wiring topology of this transformer area and confirms that the physical position of M3003 is after M3001, belonging to a typical downstream node. It should logically lag or be synchronous with M3001 in response, but here an early response occurs, violating the expected logic. At the same time, the system also finds that the change consistency index between M3003 and other meters within this time window is only 0.41, far lower than the average value of normal transformer areas.

[0050] According to the preset dual-condition judgment logic (response offset exceeds ±2 sampling points and consistency index is lower than 0.6), the system determines that M3003 has an abnormal response behavior within this time window. After further tracking its historical data, it is confirmed that this meter just completed a replacement operation a few days ago, and it may be due to incorrect construction wiring that causes the topological configuration to not match the sampling response. Based on this, the system automatically marks the meter number and time window, generates an abnormal record, and pushes it to the main station operation and maintenance system for triggering on-site verification tasks.

[0051] Further, calculating the response offset eigenvector between the electrical parameter change curves includes: In the cross-device measurement data set segmented by continuous time windows, for each continuous time window, extract the electrical parameter change curves of all smart meters under the marked dominant electrical parameter dimension, and identify the main response inflection point sequence in each electrical parameter change curve based on the correlation information between electrical parameters; Calculate the response delay of each smart meter in the current continuous time window according to the relative position difference between the main response inflection point sequence corresponding to each smart meter number and the main response inflection point sequence of the reference smart meter, where the reference smart meter is the smart meter with the highest numerical change consistency index in the current continuous time window; Combine the electrical wiring information corresponding to each smart meter in the correlation information to construct a topological path directed distance matrix, and convert the response delay of each smart meter in the current continuous time window into a topological offset distance based on the constructed topological path directed distance matrix; Jointly encode the response delay of each smart meter in the current continuous time window with its topological offset distance to generate a response offset eigenvector indexed by the smart meter number.

[0052] First, it is necessary to clarify the essential meaning of the "response offset eigenvector": This eigenvector is an ordered vector used to characterize the response time difference of multiple smart meters to electrical parameter change events in the same continuous time window, and is generated by combining its physical wiring structure for subsequent identification of abnormal response behaviors. The construction process of this vector requires a complete process of combining topological structure, inflection point recognition and time series analysis based on the completed cross-device measurement data set and electrical parameter semantic marking.

[0053] First, in the step of extracting the electrical parameter change curve in the continuous time window, the field marked as the "dominant electrical parameter" should be used as the standard. For example, active power is usually selected when detecting measurement anomalies, and current direction or power factor is preferred when judging reverse connection. For each smart meter, extract the sampling value sequence of the dominant electrical parameter in the current time window to form the electrical parameter change curve of the meter in this window. After the curve is constructed, based on the correlation information between electrical parameters (such as voltage-current correspondence, power calculation path, three-phase symmetry constraint, etc.), extract the main response inflection point sequence of the meter through the first derivative extreme value detection or trend turning point detection algorithm. The time index position of each main response inflection point is used to characterize the dynamic response position of the meter to load changes.

[0054] Next, the system needs to determine the reference smart meter under this window as the benchmark for response alignment. This reference meter should not be pre-specified, but should be dynamically selected by calculating the consistency index of the numerical changes of all meters within the current continuous time window. The consistency index can be evaluated using methods such as the Pearson correlation coefficient, dynamic time warping (DTW) similarity, or trend angle deviation, and the meter with the highest index is used as the reference smart meter. Based on this, the system compares the difference in the time axis between the main response inflection point sequences of each meter and the inflection point sequence of the reference meter, calculates the response delay of each meter, usually in the unit of sampling points or milliseconds, and a positive value indicates lag, while a negative value indicates advance.

[0055] Then, the system calls the association information table generated in step S102, extracts the upstream and downstream relationships and line path numbers of each smart meter in the physical wiring structure, and constructs a topological path directed distance matrix in combination with electrical wiring information (such as incoming / outgoing line direction, phase, wiring method). Each element in this matrix represents the path length or path level difference from the reference meter to the target meter, which is used to quantify the transfer order in the physical topology. For example, if M7011 is the reference meter, its path to M7012 is 1 level, and its path to M7013 is 2 levels, then the path values in the matrix are 1 and 2 in sequence. This path value is used to judge whether the response delay meets the physical transfer logic.

[0056] Finally, the response delay of each smart meter within the current continuous time window is jointly encoded with its corresponding topological path distance. The specific encoding method can adopt a vector splicing structure, that is, the response delay of each meter and the topological offset distance are combined into a binary group in sequence, such as {M7012: [+2, 1], M7013: [–1, 2]}, or it can be further converted into a single offset score, such as the offset residual of the response delay minus the topological path value. This vector structure is indexed by the meter number to form a complete response offset feature vector, which is used as the input basis for subsequent identification of abnormal smart meter numbers and their corresponding continuous time windows.

[0057] For easy understanding, the following provides a specific example to show the complete construction process of the response offset feature vector in the actual distribution transformer area anomaly monitoring scenario, ensuring that those skilled in the art can accurately understand and implement it.

[0058] Suppose there are three smart meters deployed in a certain urban distribution transformer area, numbered M8011, M8012, and M8013 respectively, where M8011 is the main incoming line meter of the transformer area, and M8012 and M8013 are two branch user meters. During the period from 14:00 to 14:15 on September 12, 2025, the system detected a significant change in the load and entered this continuous time window for processing.

[0059] First, the system extracts the electrical parameter change curves of three smart meters in the dimension of the main electrical parameter "active power" from the cross-device measurement data set indexed by device number. The M8011 curve shows a sharp increase in power at 14:03, M8012 shows the same trend around 14:03.5, and M8013 shows a change at 14:02.5. Based on the inflection point detection algorithm, the system identifies the positions of the main response inflection points in each electrical parameter change curve: M8011 is at t = 180 s, M8012 is at t = 183 s, and M8013 is at t = 177 s. Taking the sampling period as 1 second, M8012 lags behind M8011 by 3 sampling points, and M8013 is 3 sampling points ahead, and the response delay amounts are marked as +3 and –3 respectively.

[0060] Subsequently, the system evaluates the numerical change consistency index of the three meters within the current continuous time window. The results are: the consistency index between M8011 and M8012 is 0.91, between M8011 and M8013 is 0.76, and between M8012 and M8013 is 0.68. Since the change trend of M8011 is the most consistent with that of other meters, M8011 is determined as the reference smart meter.

[0061] Next, the system retrieves the electrical wiring information of M8011, M8012, and M8013. From the associated information table, it is known that M8011 is the main incoming line meter, and M8012 and M8013 are the first-level and second-level downstream sub-meters respectively. Based on this, the system constructs a topological path directed distance matrix, where the path from M8011 to M8012 is level 1, and the path to M8013 is level 2. The topological path values are recorded as 1 and 2 respectively.

[0062] The response delay amount of each meter is jointly encoded with its topological path value to generate the following response offset feature vector: M8012: [+3, 1] → offset residual = +2; M8013: [–3, 2] → offset residual = –5; If the offset residual is used to construct a single score value (for example, residual = response delay amount – topological path value), then the complete response offset feature vector can be written as: {  M8012: +2, M8013: –5 } From this vector, it can be seen that M8012 shows a slight response lag, which basically conforms to the expectation of its first-level path; while M8013, although located on a farther second-level branch, responds in advance, and the offset residual of –5 significantly deviates from the topological logic, indicating that there may be wiring errors or address configuration anomalies in this device.

[0063] This vector will then be used as input for subsequent steps to perform anomaly recognition and judgment. In this example, the offset feature not only quantifies the timing characteristics of the response but also combines the response time difference with the physical structure, avoiding misjudgment caused by simple timing comparison, and has clear engineering significance and application value.

[0064] Step S105: For the smart meter numbers identified as anomalies, call the original electrical parameter data of other smart meters to generate estimated electrical parameter values, and calculate the first residual between the estimated value and the original value, as well as the second residual between the estimated value and the neighboring estimated values based on the dominant electrical parameter.

[0065] During the implementation of step S105, the system needs to further call the original electrical parameter data of relevant smart meters based on the anomaly smart meter numbers and their corresponding continuous time windows identified in step S104 to generate estimated electrical parameter values for the anomaly meters, and on this basis, construct a residual vector for error verification. This process is not only a secondary verification of the anomaly determination but also an important basic link for classifying and quantifying the causes of anomalies.

[0066] First of all, the so-called "estimated electrical parameter value" refers to the normal measurement result of the anomaly meter during this period by fusing the original electrical parameter data of other non-anomaly smart meters in the same substation area during the same period. This estimated value is not a simple average but is obtained through weighted combination, interpolation extrapolation, or model fitting considering factors such as topological structure, adjacent physical relationships, load similarity, and electrical parameter correlation. For example, if a branch meter is identified as an anomaly, its upstream main meter and other user meters on the same branch can provide estimated reference values. The system inputs indicators such as its active power, phase A current, or power factor into the reconstruction function to output the estimated electrical parameters of the target meter.

[0067] To ensure that those skilled in the art can clearly implement the process of estimating electrical parameter values, the following provides a feasible specific algorithm example for this process, which is applicable to the meter deployment structure with main and branch relationships in most distribution substations.

[0068] After identifying that the device with the smart meter number M5012 has abnormal behavior in a specific time window, the system needs to estimate the dominant electrical parameter (such as active power) of this meter during this period. For this purpose, the system first selects its upstream main meter M5001 and the lateral user meter M5013 on the same branch from other non-anomaly smart meters with clear physical connection relationships with M5012 in the same substation area as reference meters.

[0069] Next, the system extracts the historical measured active power values of M5001, M5012, and M5013 within the same time period in the past consecutive several days (e.g., the past 30 days) to construct three time series. Based on these time series, the system calculates the Pearson correlation coefficients between M5001 and M5012, and between M5013 and M5012 respectively. Suppose the calculation results are: Corr(M5001, M5012) = 0.82, Corr(M5013, M5012) = 0.64. Based on this, the system determines the estimated contribution weight ratios of each reference meter to M5012, specifically: The weight of M5001 is 0.82 / (0.82 + 0.64) ≈ 0.5625; The weight of M5013 is 0.64 / (0.82 + 0.64) ≈ 0.4375.

[0070] Then, the system extracts the real-time active power data of M5001 and M5013 within the current time window to be estimated, which are 1.80 kW and 1.20 kW respectively. Based on the weighted average method, the system calculates the estimated active power of M5012 as: Estimate(M5012) = 1.80 × 0.5625 + 1.20 × 0.4375 = 1.53 kW.

[0071] This estimated value is the comprehensive prediction value of M5012 in this time window based on the data of physically adjacent meters and historical correlations. If the actual measured value of M5012 in this time period is 2.10 kW, then the first residual is +0.57 kW, and the residual percentage is +37.25%, which is much higher than the system's preset warning threshold of ±10%. Thus, it can be preliminarily determined that this meter has an abnormal deviation.

[0072] Next, it is necessary to clarify the meaning of the "dominant electrical parameter". In different types of anomaly discrimination, the system should select the electrical parameter that is most sensitive and representative to this type of anomaly as the dominant indicator. For wiring errors and reverse current connection, the phase A current or power direction is the most crucial; for metering anomalies, active power or electrical energy is more suitable. Therefore, the system automatically identifies the dominant electrical parameter of the current scenario based on the pre-marked semantic tags of the electrical parameters, and uses this as the main input quantity for residual analysis.

[0073] Subsequently, based on this dominant electrical parameter, the system calculates the first residual between the estimated value and the original measured value, that is, the so-called "prediction error". Its physical meaning is the numerical deviation between the measured behavior of the target meter in the current time window and the result derived from the normal behavior of adjacent meters. If this deviation exceeds the normal statistical range or threshold (such as exceeding ±10% or ±2σ), it indicates that the meter behavior may be affected by abnormal factors.

[0074] Meanwhile, to avoid misjudgment caused by estimation errors in the first residual, the system also introduces a "second residual" as an auxiliary judgment basis. The second residual refers to the deviation between the estimated value and its neighboring estimated values, and is used to test whether the estimated value is consistent with other estimated data within the entire estimation scenario. For example, if the sequences formed by the estimated values of multiple adjacent electricity meters are consistent with each other, and the estimated value of a certain abnormal electricity meter has a large deviation from these sequences, it may indicate that the abnormality of this electricity meter is an independent disturbance rather than a group fluctuation.

[0075] The system finally combines the first residual and the second residual to construct a residual feature vector, and conducts multi-dimensional cross-validation in combination with the response offset and the consistency index. The unit of the residual value can be the same as that of the original data, such as kWh, kW, or A; it can also be normalized into a percentage form for unified determination. The positive and negative values of the residual can also reflect the abnormal directionality, such as the cases of power backflow and current backfeeding.

[0076] In summary, step S105 not only provides a refined abnormal error quantification mechanism, but also realizes a verifiable, interpretable, and classifiable electricity meter abnormality confirmation process through the selection of the main electrical parameters and the construction of double residuals.

[0077] To facilitate the understanding of the specific implementation process of step S105, the following provides a detailed example in combination with an actual scenario. Assume that multiple smart electricity meters are deployed in the substation area of a certain residential community, where M4011 is the main meter on the main line, and M4012 and M4013 are the user sub-meters of two branches. The system identifies that M4013 has abnormal behavior in the time window from 10:00 to 10:15 on July 15, 2025 in step S104, mainly manifested as an early response and a significant inconsistency in the power change trend compared with other electricity meters.

[0078] When executing step S105, the system first retrieves the original electrical parameter data corresponding to M4013 in this time window from M4011 and M4012, mainly including fields such as the active power of phase A, the active electrical energy, and the power factor. The system combines the previous electrical parameter semantic tags to determine that the main electrical parameter of this abnormal event should be "active power" because this parameter has strong response characteristics to wiring errors, reverse current connection, and measurement inaccuracy.

[0079] Subsequently, the system confirms that there is an upstream-downstream power supply relationship between M4011 and M4013 through topological path comparison. M4012 and M4013 are physically in the same branch but not directly in series. Based on this structure, the system constructs an estimation model and uses the weighted summation method to reconstruct the estimated active power of M4013. The weights are set according to topological distance and historical correlation. For example, the weight of M4011 is 0.7 and the weight of M4012 is 0.3. After calculation, the estimated active power of M4013 within this time window is 1.35 kW, while the actual measured value is 1.89 kW, with a difference of 0.54 kW, accounting for 40% of the estimated value, significantly exceeding the system-set error threshold of 10%, thus forming the first residual.

[0080] Furthermore, the system also calculates that the estimated values of M4013 reconstructed by the same method for M4011 and M4012 during this time period are 1.32 kW and 1.37 kW respectively. The average deviation between these two values and the current estimated value of 1.35 kW is less than 2%, indicating that the estimated value itself is stable and reliable among neighboring devices. However, there is a 40% deviation between this estimated value and the original data of M4013, indicating that the anomaly is more likely to originate from the M4013 itself rather than fluctuations in the estimation source, thus forming the second residual, further verifying the establishment of the independence anomaly of M4013.

[0081] Finally, the system records the first residual (+0.54 kW) and the second residual (highly consistent but conflicting with the original value) together as the residual feature vector, and inputs them together with the consistency index (0.42) and response offset (–3 sampling points) formed in the previous steps into the anomaly classifier to confirm that there is a high-confidence abnormal behavior in the wiring direction of the electricity meter for M4013. The system automatically generates an anomaly identification record containing the device number, anomaly type, time window, and comparison table of estimated value and original value for subsequent on-site operation and maintenance.

[0082] Furthermore, calculating the first residual between the estimated value and the original value, and the second residual between the estimated value and the neighboring estimated values according to the leading electrical parameters includes: Within the current continuous time window, based on the generated estimated electrical parameter values, extract the original electrical parameter data of the leading electrical parameters corresponding to the identified abnormal smart meter number, and calculate the difference between its original electrical parameter data and the estimated electrical parameter values. By dividing this difference by the estimated electrical parameter values, obtain the normalized residual value of this smart meter number within this continuous time window, which is used as the basis for constructing the first residual. Call the topological path directed distance matrix, screen multiple smart meter numbers that are topologically adjacent to the smart meter number identified as abnormal from the cross-device measurement data set, obtain the estimated electrical parameter values generated based on the main electrical parameters within this continuous time window, and perform weighted combination according to the topological path directed distance to construct the adjacent estimated value set within this continuous time window; Perform weighted difference calculation on the estimated electrical parameter value of the smart meter number identified as abnormal within this continuous time window and the adjacent estimated value set. The weighting coefficient is determined by the inverse of the topological path directed distance, and in combination with the sampling stability index of the main electrical parameter, generate the second residual of this smart meter number within this continuous time window, which is used to characterize the deviation degree of this estimated electrical parameter value relative to the adjacent estimated value set.

[0083] In the method of the present invention, the "first residual" and the "second residual" are not repeated calculations of single numerical errors, but serve two logical dimensions: one is used to measure the relative deviation between the current behavior of a single meter and the estimated expectation, and the other is used to evaluate the credibility of the current estimated value itself, so as to avoid misjudging as abnormal due to unstable estimation sources, thereby establishing a robust and two-way cross-verification abnormal identification mechanism. The following explains the residual construction process.

[0084] First, the system needs to enter the continuous time window delimited by the previous steps, assumed to be the period from 16:00 to 16:15 on October 12, 2025. Within this time window, the system has identified a group of smart meter numbers marked as "possibly abnormal" based on the abnormal identification mechanism, such as the number M9012. This number already has the following data structure support: the marked information of its main electrical parameter (such as "active power" or "phase B current"), the original electrical parameter sampling sequence within its continuous time window, the connection path and directed distance derived from the topological graph structure, and the estimated electrical parameter value constructed by the system by fusing the data of multiple adjacent meters.

[0085] Before calculating the first residual, it is first necessary to ensure that the acquisition of the estimated electrical parameter value is engineering reasonable. This value is obtained by calling other non-abnormal smart meters that are topologically adjacent in step S105, through the association information table and the topological path directed distance matrix, and combining a weighting algorithm (such as Pearson correlation weighting, path inverse weighting, or historical load similarity weighting). Based on the obtained estimated electrical parameter value, the system then extracts the original electrical parameter data sequence of M9012 within the current continuous time window in the dimension of the main electrical parameter, such as its active power measurement values once per minute: {3.2 kW, 3.4 kW, 3.6 kW, 3.5 kW}.

[0086] The system uses the estimated electrical parameter value sequence as a reference and calculates the difference between the original measurement value and the estimated value point by point. For example, if the estimated values are: {2.9 kW, 3.1 kW, 3.2 kW, 3.3 kW}, then the differences are: {+0.3, +0.3, +0.4, +0.2}. To avoid distortion of the absolute error at different magnitudes (e.g., the judgment criteria for an error of 0.5 for 10 kW and an error of 0.5 for 1 kW are the same), the system divides each difference by the estimated value itself to form a standardized proportional error sequence, that is: The first residual sequence ={(3.2–2.9) / 2.9 = 10.3%, (3.4–3.1) / 3.1 = 9.7%, (3.6–3.2) / 3.2 = 12.5%, (3.5–3.3) / 3.3 = 6.1%}.

[0087] The system can further take the mean or maximum value of the residual sequence as the first residual of M9012 within this time window. For example, select the maximum standardized residual value of 12.5% as the first residual. This value is used to measure whether the behavior of this electricity meter significantly deviates from the normal range inferred by the system and is one of the basic dimensions for anomaly identification.

[0088] After completing the first residual, the system enters the process of constructing the second residual. The purpose of this process is to determine whether the estimated electrical parameter values have reasonable internal consistency, that is, whether the estimated values themselves are credible. If the estimated values themselves fluctuate due to large fluctuations or low correlations in the data of neighboring electricity meters, even if the first residual is large, it cannot be directly determined that the device is abnormal.

[0089] The system first filters out the neighboring smart electricity meter numbers that are "path levels less than or equal to 2" in the topological path directed distance matrix from the cross-device measurement data set for M9012. Suppose they are M9009 and M9010, and the path distances are 1 and 2 respectively. Within the current continuous time window, the system obtains the estimated electrical parameter values of M9009 and M9010 in the dominant electrical parameter dimension, which are: M9009: {3.0, 3.2, 3.4, 3.3}; M9010: {2.8, 3.0, 3.2, 3.2}.

[0090] The system performs weighted fusion on these neighboring estimated sequences, and the weights are determined by the inverse of the path distance. For example, if the path is 1, the weight is 1.0, and if the path is 2, the weight is 0.5. The neighboring estimated value at each moment can be expressed as a weighted average. For example: Neighboring estimated value (point 1) = (3.0×1 + 2.8×0.5) / (1+0.5) = 2.93 Neighboring estimated value (point 2) = (3.2×1 + 3.0×0.5) / (1+0.5) = 3.13…… Calculate the difference between the estimated electrical parameter values of M9012 and each point in this adjacent estimated value sequence to obtain a deviation sequence. For example, if the estimated values of M9012 are {2.9, 3.1, 3.2, 3.3}, then the differences are as follows: {2.9 – 2.93 = –0.03, 3.1 – 3.13 = –0.03, 3.2 – 3.27 = –0.07, 3.3 – 3.18 = +0.12}.

[0091] The system calculates the mean square error or mean absolute error for this difference sequence. For example, take the mean absolute error as the second residual.

[0092] To improve the accuracy of credibility judgment, the system also needs to combine the sampling stability of the main electrical parameter within the current continuous time window, that is, the standard deviation or the change rate within the sliding window. If the estimated value itself fluctuates violently, even if the deviation is small, it is not considered stable. Assume that the system sets the sliding change rate not to exceed 5%, then it can be further confirmed whether the estimated value is stable. If it meets the requirements, the current second residual is valid.

[0093] Finally, the system takes the first residual (12.5%) and the second residual (for example, 0.04) as the residual output results of the smart meter number M9012 identified as abnormal within the current continuous time window, and records them in a structured manner for subsequent abnormal classification judgment.

[0094] This implementation path is not only completely closed-loop in terms of data dependence, ensuring that the first residual depends on the estimated value and the original electrical parameter data, and the second residual depends on the adjacent estimated value and the topological structure, but also through the standardized error and stability control mechanism, it avoids the misjudgment risk of "high error means abnormal" in the prior art. Its residual model is not a static numerical comparison, but a power grid intelligent discrimination logic that integrates physical structure, data correlation, and dynamic sampling behavior.

[0095] Furthermore, for the smart meter number identified as abnormal, calling the original electrical parameter data of other smart meters to generate estimated electrical parameter values includes: Within the current continuous time window, select from the cross-device measurement data set multiple smart meter numbers that have a finite path distance with the smart meter number identified as abnormal in the topological path directed distance matrix, and extract the original electrical parameter data of these smart meter numbers in the dimension of the main electrical parameter to form a topological adjacent electrical parameter data set; Based on the Pearson correlation coefficients between each topological adjacent smart meter number and the smart meter number identified as abnormal for the corresponding main electrical parameter in multiple historical continuous time windows, calculate the correlation weight vector, which is used to determine the linear combination contribution coefficients of each adjacent meter to the estimated electrical parameter value; Combining the electrical wiring information with the topological path directed distance matrix, determine the path directionality and level difference of each adjacent smart meter number relative to the smart meter number identified as abnormal, and construct a topological ratio correction vector for adjusting the linear combination contribution coefficient according to the path direction, where the upstream node adjustment coefficient is less than 1 and the downstream node adjustment coefficient is greater than 1; Perform an element-wise product fusion of the correlation weight vector and the topological ratio correction vector to obtain a weighted correction coefficient vector, and use this weighted correction coefficient vector as a weighting factor to perform a weighted summation operation on the topological adjacent electrical parameter data set to generate the estimated electrical parameter value of the smart meter number identified as abnormal within this current continuous time window.

[0096] In a smart distribution network, due to factors such as wiring errors, reverse current connection, local disturbances, meter inaccuracy, or communication anomalies, the original electrical parameter data of some smart meters may experience mutations, abnormal offsets, or temporary failures. To improve the system's ability to identify these abnormal electrical parameters and fault tolerance, it is necessary to construct an estimated electrical parameter generation mechanism with physical constraints, historical relevance, and topological interpretability, especially in an operating environment lacking manual verification or unable to rely on external device measurement redundancy.

[0097] The present invention realizes a robust, dynamic, non-line loss type electrical parameter estimation method by fusing the temporal data correlation, physical wiring relationship, and topological structure path characteristics among multiple smart meters. This method is specifically used when a certain smart meter is identified as abnormal. By using the original electrical parameter data of its adjacent smart meters, combined with historical behavior similarity and path structure difference, the estimated electrical parameter value of this meter at the current time period is reconstructed, providing input support for subsequent residual analysis, anomaly classification, and diagnosis.

[0098] In actual deployment, the system has first completed the identification of the abnormal smart meter number according to the previous processing flow and located the continuous time window where it is located. Assume that the current processing object is the smart meter numbered M4056, and the time window is from 17:30 to 17:45 on November 3, 2025. At this time, it is necessary to construct the estimated electrical parameter value of the dominant electrical parameter of M4056 in this time window. For this purpose, the system needs to screen out several smart meter numbers from the cross-device measurement data set that have an effective physical proximity relationship with M4056 in the topological path directed distance matrix.

[0099] The so-called "topological path directed distance matrix" is a structural matrix predefined according to electrical wiring information, load connection diagrams, and communication configuration information, reflecting the connection paths and level differences between each smart meter in the distribution area. Each element in the matrix It represents the number of node hops or the length of the electrical path required to reach smart meter number j starting from smart meter number i along the distribution network structure. For example, if M4056 is a certain main node, its neighboring numbers may include M4052, M4054, M4060, etc., and the path distances are 1, 2, 1 respectively.

[0100] The system obtains all smart meter numbers from this matrix whose topological distance from M4056 in this time window is less than or equal to the threshold (e.g., ≤2). Suppose 4 numbers are screened out: M4052, M4053, M4054, M4060. Then, the original electrical parameter data of these 4 meters in the dimension of the main electrical parameter is extracted from the cross-device measurement data set. For example, if the main electrical parameter is "active power", then the original sampling values per minute between 17:30 and 17:45 are extracted to form a data set composed of multiple sampling sequences corresponding to the numbers, that is, the topological neighboring electrical parameter data set.

[0101] However, simply taking the average of the original data of these neighboring meters will ignore the historical behavior characteristics and structural location differences, and it is difficult to reflect the true response relationship of the system. Therefore, the present invention further introduces historical data correlation as an estimation weight control factor. Outside the current window, the system calls the original electrical parameter sequences of these neighboring meters and the target meter M4056 in multiple consecutive time windows in the past day, past week, or defined period in the dimension of the main electrical parameter, and calculates the Pearson correlation coefficient through point-by-point matching.

[0102] The Pearson correlation coefficient r reflects the linear correlation degree of two time series, and its value range is [–1, +1]. Here, to avoid directional interference, the absolute value is taken and normalized to between [0,1] to form a correlation weight vector. Suppose the correlation coefficients of M4052, M4053, M4054, M4060 with M4056 are 0.83, 0.51, 0.93, 0.70 respectively. After normalization, it can be set as {0.30, 0.18, 0.34, 0.26} as the contribution weights of each neighboring meter in the linear combination.

[0103] Next, to further improve the response accuracy of the estimated electrical parameter value to the actual physical path, the system needs to combine the wiring method and the topological path directed distance matrix to construct a proportional correction mechanism. The goal of this mechanism is to introduce directional discrimination: the measured value of the upstream node should be appropriately weakened when conducted to the target meter, and the response value of the downstream node should be proportionally compensated considering the load distribution or aggregation effect.

[0104] Based on the actual topological information, the system can establish a set of path direction discrimination rules. For example, if in the path from node i to node j, j is downstream of i, it is denoted as a "forward path", otherwise it is a "reverse path"; the greater the path distance, the more obvious the signal attenuation. Define a proportional adjustment coefficient based on the path distance, for example: Path distance = 1, upstream correction coefficient = 0.95, downstream correction coefficient = 1.05; Path distance = 2, upstream correction coefficient = 0.90, downstream correction coefficient = 1.10; In this way, the system can perform an element-by-element multiplication of the topological ratio correction vector and the previously obtained correlation weight vector to obtain the final weighted correction coefficient vector. For example, M4052 is an upstream node, path 1; M4053 is upstream, path 2; M4054 is a parallel node; M4060 is downstream, path 1. Their correction vectors are {0.95, 0.90, 1.00, 1.05} respectively, then the weighted correction coefficient vector is {0.285, 0.162, 0.340, 0.273}.

[0105] In the generation of the final estimated electrical parameter value, the system uses this weighted correction coefficient vector as a weighting factor to perform weighted summation on the dominant electrical parameter values of the four adjacent electric meters at each time point. For example, at 17:32, the sampling values of each meter are: M4052: 3.5 kW, M4053: 3.1 kW, M4054: 3.6 kW, M4060: 3.4 kW.

[0106] Then the estimated electrical parameter value of M4056 at 17:32 is: 3.5×0.285 + 3.1×0.162 + 3.6×0.340 + 3.4×0.273 = 0.9975 + 0.5022 + 1.224 + 0.9282 = 3.6519 kW; The system can generate a complete sequence of estimated electrical parameters of M4056 within the entire continuous time window in this way and write it into the numbered index structure for subsequent residual construction steps to call.

[0107] It should be noted that this estimation method is essentially different from the "global average" method widely used in the prior art. In the prior art, often directly estimate the value of each electric meter by subtracting the sum of the branch meters from the main meter data, or perform unified averaging on all smart meter data, ignoring the actual electrical positions between the electric meters, the differences in load response characteristics, and the historical cooperation relationship. And this method not only considers the adjacent relationship, but also uses historical cooperation as the weight and topological path directionality as the proportional correction, making the estimated value highly personalized and having the ability of physical interpretation.

[0108] In addition, this method is also easy to deploy at the implementation level. The calculation of the correlation coefficient can be preprocessed in the historical data in a sliding window manner. The topological path matrix and electrical wiring information can be generated by a power system topology modeling tool and adjusted in real time as the system is updated. Each weighted operation and ratio correction can also be efficiently implemented in the database layer or edge computing nodes, supporting the real-time processing ability for large-scale distribution areas.

[0109] To sum up, this estimation method provides an electric parameter reconstruction mechanism with a clear structure, traceable logic, and strong dynamic response ability, which is particularly suitable for the abnormal identification task of multi-source collaborative intelligent meters at the distribution area level. Through this method, the identification error caused by sampling loss, sensor drift, or data communication interruption can be effectively reduced. At the same time, it provides input data with credibility and physical consistency for the construction process of the first residual and the second residual, thus significantly improving the robustness and accuracy of abnormal identification.

[0110] Step S106: According to the consistency index, the first residual, and the second residual, determine whether there are wiring errors, reverse current connection, metering anomalies, or other abnormal types, and output an abnormal identification result record including the abnormal type, intelligent meter number, and continuous time window.

[0111] During the implementation of step S106, the system needs to comprehensively utilize the three core analysis results of the consistency index, the first residual, and the second residual obtained in the previous stage for multi-dimensional cross-verification and logical attribution, so as to clarify whether the abnormality actually exists, further judge its type and nature, and finally generate a structured abnormal identification result output.

[0112] The system needs to set multi-dimensional judgment rules in this step. For example, normalize the above three indicators to the same dimension, use a weighted scoring function for aggregation, or construct a three-dimensional threshold model. For example, when the consistency index is less than 0.6, the first residual is greater than ±15%, and the second residual is less than ±5%, it is automatically judged as a high-confidence abnormality. Each judgment threshold should be dynamically adjusted in combination with the actual load characteristics of the distribution area, the measurement accuracy level of the equipment, and the operation and maintenance experience.

[0113] After identifying the abnormal behavior, the system also needs to further perform abnormal type identification operations in combination with the response offset feature vector, device type, electrical parameter category, and topological information. For example: If the current direction is reversed, the power symbol is reversed, and the first residual is prominent while the consistency index is normal, it is very likely that the current transformer wiring is reversed; If the consistency index of a certain phase among the three phases of A, B, and C suddenly drops, the power change amplitude is extremely high, and the time lag is significant, it may be a single-phase wiring error or phase sequence disorder; If all indicators deviate, the estimated value has a large difference from the adjacent table, and even the stability of the estimation itself is poor, then it may be due to equipment configuration errors, electricity meter address mapping errors, or communication channel chaos; If all indicators only fluctuate slightly within a specific time window, it can be judged as a short-term disturbance or transient load jump, which belongs to negligible events and can avoid false alarms.

[0114] Finally, the system formats and outputs the recognition results, including the abnormal type (such as reverse connection, phase error, misalignment), electricity meter number, start and end time windows, dominant electrical parameter type, actual measurement value, estimated value, residual value, and confidence level, etc., and writes them into the abnormal recognition result record table. This table will be synchronously pushed to the master station system or the abnormal data alarm platform for further processing by the dispatching system, operation and maintenance personnel, or data analysis module.

[0115] To facilitate the understanding of the application process of step S106, a specific example is provided below.

[0116] For example, multiple smart electricity meters are deployed in an industrial substation area. Among them, M6011 is the main incoming line meter, and M6012 and M6013 are the branch meters for parallel branch users. Within the time window from 14:30 to 14:45 on August 10, 2025, the system identified abnormal behavior of M6012 in step S104. The A-phase current change curve of this electricity meter within this window is completely out of sync with other electricity meters, with a response offset of -3 sampling points. At the same time, the consistency index is 0.35, which is much lower than the normal threshold of 0.8 set by the system.

[0117] After entering step S105, the system calls the original electrical parameter data of M6011 and M6013 to estimate the dominant electrical parameter (A-phase active power) of M6012 during this period. The estimated value is 4.10 kW, while the actual measurement value of M6012 is 5.02 kW. Thus, the first residual is calculated as +0.92 kW, accounting for 22.4% of the estimated value. This residual is much higher than the deviation threshold of ±15% set by the system, constituting a significant deviation. In addition, the system further detects the average deviation between this estimated value and the estimated sequences of M6011 and M6013, which is only ±0.03 kW, indicating that the estimated value itself has high stability. Therefore, the second residual is a low value, excluding the influence of modeling errors.

[0118] After entering step S106, the system makes a cross-judgment based on the aforementioned results. First, the consistency index is extremely low (0.35), the first residual significantly exceeds the limit (+22.4%), and the second residual is extremely small, meeting the condition of "high-confidence anomaly". Subsequently, the system further analyzes the active power sign of the current electricity meter and finds that its power direction changes from positive to negative and the current direction also reverses. The system confirms in the topology diagram that the load connected to this electricity meter is a single-phase motor, and theoretically there should be no reverse power phenomenon, ruling out the possibility of load feedback. Therefore, based on the preset rules and combining the characteristic patterns of abnormal power direction and reverse current, the system determines that this event is a reverse connection of the current transformer.

[0119] The system formats and outputs the above recognition results, including: the abnormal type is "reverse current connection", the intelligent electricity meter number is M6012, the abnormal start and end time window is "2025-05-10 14:30–14:45", the main electrical parameter is "A-phase active power", the actual value is 5.02 kW, the estimated value is 4.10 kW, the first residual is +0.92 kW, the residual percentage is +22.4%, the consistency index is 0.35, the response offset is –3 sampling points, and the confidence level is "high". This result is written into the abnormal recognition result table and uploaded to the master station system for the subsequent dispatching system to issue maintenance instructions.

[0120] It can be seen from this example that the comprehensive judgment logic and classification recognition mechanism described in step S106 can be directly embedded in the existing distribution automation platform to support local abnormal self-diagnosis or centralized alarm linkage mechanism.

[0121] Furthermore, determining whether there is a wiring error, reverse current connection, metering anomaly or other abnormal types according to the consistency index, the first residual and the second residual includes: Extract the consistency index, the first residual and the second residual corresponding to the value change of each identified intelligent electricity meter number in the current continuous time window, and construct an abnormal quantization feature vector of this intelligent electricity meter number in this continuous time window. This abnormal quantization feature vector includes three standardized index values, which are used to comprehensively measure the behavior deviation degree, estimation stability and group consistency; Based on the response offset feature vector, extract the response delay amount and topological offset distance corresponding to this intelligent electricity meter number in the current continuous time window, and combine the symbol change trend of the main electrical parameter in this continuous time window to construct a response behavior feature label of this intelligent electricity meter number in this continuous time window, which is used to reflect its time consistency state relative to the physical topology transfer logic; Taking the abnormal quantization feature vector and the response behavior feature label as input items, substituting them into a preset set of abnormal type discrimination rules, where the set of abnormal type discrimination rules is structurally defined by the system through the combined mapping relationship between the variation law of the main electrical parameters, electrical wiring information, response offset direction, and error distribution characteristics, and is used to achieve attribution matching for wiring errors, reverse current connection, metering abnormalities, or other abnormal types; Taking the matching result of the set of abnormal type discrimination rules as the judgment output to determine whether there are wiring errors, reverse current connection, metering abnormalities, or other abnormal types for the smart meter number within the current continuous time window.

[0122] In actual engineering implementation, first, the smart meter numbers identified as "abnormal" need to be used as the processing objects, relying on the processing structure established in the previous steps, especially the following three key quantities generated within each continuous time window: First, the numerical change consistency index, which reflects the operation coordination of the target meter relative to other smart meters, that is, whether it shows a similar change trend to other meters; Second, the first residual, which is the standardized deviation between the original measured value and the estimated electrical parameter value of the main electrical parameter of the target meter within this time window, characterizing the abnormality degree of its measurement behavior relative to the predicted behavior; Third, the second residual, which is used to evaluate whether the estimated electrical parameter value of this meter is credible, that is, whether there is a deviation from the estimated values of adjacent meters on the topological path. If the second residual is very large, the current estimated value may have an unstable source or the surrounding load fluctuates violently.

[0123] The system combines the above three indicators into an abnormal quantization feature vector in a unified format. Taking the smart meter number M8032 as an example, assume the current continuous time window is from 14:00 to 14:15 on October 25, 2025. Within this window, the system extracts its consistency index as 0.41, the first residual as 18.2%, and the second residual as 4.7%. The system takes these three values as the three-dimensional feature vector {0.41, 0.182, 0.047}, and binds the smart meter number M8032 and the time window ID to form a structured record for subsequent attribution use.

[0124] Next, the system further retrieves the response offset feature vector, which is constructed through inflection point extraction, response delay calculation, and topological path mapping. Taking M8032 as an example, its main electrical parameter is the B-phase current. Within the current time window, the system has calculated its response delay as -2 sampling points, indicating an early response, the topological offset distance as 1, and this electricity meter should be physically located on the downstream path. At this time, the system further analyzes the sign change trend of the main electrical parameter. For example, the direction of its B-phase current changes from positive to negative, and at the same time, the active power changes from positive to negative, indicating an obvious sign of power reversal. So the system constructs a response behavior feature label for this electricity meter as: "Early response + Power direction reversal + Negative topological offset". This label is used to deduce whether its time response conforms to the actual electrical structure transfer logic.

[0125] After completing the construction of the above two features, the system inputs the abnormal quantization feature vector and the response behavior feature label into the abnormal type discrimination rule set. This rule set is a decision mapping matrix summarized through a large number of sample trainings and simulation models. The core bases include: If the consistency index is low, the first residual is greater than the set threshold (such as 15%), the second residual is small (such as less than 5%), and the response delay is less than 0 (early response), the power direction reverses, and the topological offset is negative, then it is judged that the current transformer is reversely connected; If the consistency index is extremely low (such as <0.2), and only one of the three phase main electrical parameters deviates significantly, and the response lags severely, then it can be judged as single-phase wiring error or phase misconnection by combining the wiring method analysis; If the consistency index is medium (0.4 - 0.7), the first residual exceeds 20%, but the second residual fluctuates greatly (higher than 10%), and at the same time, the estimated value of the main electrical parameter itself changes violently (large sliding standard deviation), then it is determined that the estimation path is unstable or the abnormality is caused by adjacent disturbances, and it cannot be directly classified as an abnormality of the equipment body; If all three indicators are near the critical value, but intermittent deviation behaviors occur for a long time, and the response behavior is normal at the same time, then it is initially judged that the main electrical parameter measurement is inaccurate, and it needs to be confirmed in combination with long-term behaviors.

[0126] The rule set is not a single-threshold judgment, but a logical decision set based on index combinations. The system can use a multi-dimensional Boolean rule chain, a fuzzy membership function method, or a small interpretable rule network model for matching. Each rule finally outputs an abnormal type label, limited to the "wiring error", "current reverse connection", "measurement abnormality", or "other abnormal types".

[0127] Based on the discrimination result, the system writes the label "reverse current connection" in the structured recognition record of M8032, and packs and binds its main electrical parameter as "phase B current" and the response behavior as "early response + reverse power direction" to provide a complete traceable evidence chain for subsequent result output and operation and maintenance processing.

[0128] Through the above method, this embodiment not only constructs a complete closed-loop structure of "abnormality quantification - response behavior - attribution matching" from the data, but more importantly, breaks through the limitation of traditional electric meter abnormality recognition methods that only rely on numerical deviation or fixed rule judgment, introduces three dimensions of topological structure, response behavior directionality and estimation reliability, and realizes truly interpretable and attributable abnormality classification and recognition.

[0129] The second embodiment of the application provides an electronic device, and the electronic device includes: A processor; A memory for storing a program, and when the program is read and executed by the processor, it executes a method for intelligent fusion and abnormality recognition of multi-source electric meter data provided in the first embodiment of the present application.

[0130] The third embodiment of the application provides a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, it executes a method for intelligent fusion and abnormality recognition of multi-source electric meter data provided in the first embodiment of the present application.

[0131] Although the present application is disclosed above with preferred embodiments, it is not used to limit the present application. Any person skilled in the art can make possible changes and modifications without departing from the spirit and scope of the present application. Therefore, the protection scope of the present application should be subject to the scope defined by the claims of the present application.

Claims

1. A method for intelligent fusion and anomaly recognition of multi-source electricity meter data, characterized in that, Including: Collecting original electrical parameter data from multiple heterogeneous smart meters under a unified time reference, and organizing the original electrical parameter data into a cross-device measurement data set indexed by smart meter numbers; Classifying and marking each item of original electrical parameter data in the cross-device measurement data set according to the electrical wiring information and functional parameters of each smart meter to generate correlation information between electrical parameters; Segmenting the cross-device measurement data set by continuous time windows, extracting the electrical parameter change curves of each smart meter, and calculating the numerical change consistency index within each time window based on the correlation information to obtain a consistency index sequence; Calculating the response offset feature vectors between the electrical parameter change curves, and making a joint judgment in combination with the consistency index to identify the abnormal smart meter numbers and their corresponding continuous time windows; For the identified abnormal smart meter numbers, calling the original electrical parameter data of other smart meters to generate estimated electrical parameter values, and calculating the first residual between the estimated value and the original value, and the second residual between the estimated value and the adjacent estimated value based on the dominant electrical parameter; Determining whether there are wiring errors, reverse current connections, metering anomalies or other abnormal types according to the consistency index, the first residual and the second residual, and outputting an abnormal identification result record including the abnormal type, smart meter number and continuous time window; 2. The method for intelligent fusion and anomaly recognition of multi-source electricity meter data according to claim 1, wherein The calculating the response offset feature vectors between the electrical parameter change curves includes: In the cross-device measurement data set segmented by continuous time windows, for each continuous time window, extracting the electrical parameter change curves of all smart meters in the marked dominant electrical parameter dimension, and identifying the main response inflection point sequences in each electrical parameter change curve based on the correlation information between electrical parameters; Calculating the response delay amount of each smart meter in the current continuous time window according to the relative position difference between the main response inflection point sequence corresponding to each smart meter number and the main response inflection point sequence of the reference smart meter, where the reference smart meter is the smart meter with the highest numerical change consistency index in the current continuous time window; Combining the electrical wiring information corresponding to each smart meter in the correlation information to construct a topological path directed distance matrix, and converting the response delay amount of each smart meter in the current continuous time window into a topological offset distance based on the constructed topological path directed distance matrix; Jointly encoding the response delay amount of each smart meter in the current continuous time window with its topological offset distance to generate a response offset feature vector indexed by smart meter numbers; 3. The method for intelligent fusion and anomaly recognition of multi-source electricity meter data according to claim 2, characterized in that, The calculating the first residual between the estimated value and the original value, and the second residual between the estimated value and the adjacent estimated value based on the dominant electrical parameter includes: In the current continuous time window, based on the generated estimated electrical parameter values, extracting the original electrical parameter data of the dominant electrical parameter corresponding to the identified abnormal smart meter number, and calculating the difference between its original electrical parameter data and the estimated electrical parameter value. By dividing the difference by the estimated electrical parameter value, the normalized residual value of the smart meter number in the continuous time window is obtained, which is used as the construction basis for the first residual; Call the topological path directed distance matrix, screen multiple smart meter numbers that are topologically adjacent to the smart meter number identified as abnormal from the cross-device measurement data set, obtain the estimated electrical parameter values generated based on the main electrical parameters within this continuous time window, and perform weighted combination according to the topological path directed distance to construct a set of adjacent estimated values within this continuous time window; Perform weighted difference calculation on the estimated electrical parameter value of the smart meter number identified as abnormal within this continuous time window and the set of adjacent estimated values. The weighting coefficient is determined by the inverse of the topological path directed distance, and combined with the sampling stability index of the main electrical parameter, generate the second residual of this smart meter number within this continuous time window, which is used to characterize the deviation degree of the estimated electrical parameter value relative to the set of adjacent estimated values.

4. The method for intelligent fusion and anomaly recognition of multi-source electric meter data according to claim 3, wherein Based on the consistency index, the first residual, and the second residual, determine whether there are wiring errors, reverse current connection, measurement anomalies, or other anomaly types, including: Extract the numerical change consistency index, the first residual, and the second residual corresponding to each smart meter number identified as abnormal within the current continuous time window, and construct an abnormal quantization feature vector of this smart meter number within this continuous time window. This abnormal quantization feature vector includes three standardized index values, which are used to comprehensively measure the behavior deviation degree, estimation stability, and group consistency; Based on the response offset feature vector, extract the response delay amount and topological offset distance corresponding to this smart meter number within the current continuous time window, and combine the sign change trend of the main electrical parameter within this continuous time window to construct a response behavior feature label of this smart meter number within this continuous time window, which is used to reflect its time consistency state relative to the physical topology transfer logic; Take the abnormal quantization feature vector and the response behavior feature label as input items, and substitute them into a preset set of abnormal type discrimination rules. The set of abnormal type discrimination rules is structurally defined by the system through the combined mapping relationship between the variation law of the main electrical parameter, electrical wiring information, response offset direction, and error distribution characteristics, and is used to achieve attribution matching of wiring errors, reverse current connection, measurement anomalies, or other anomaly types; Take the matching result of the set of abnormal type discrimination rules as the judgment output to determine whether there are wiring errors, reverse current connection, measurement anomalies, or other anomaly types for this smart meter number within the current continuous time window.

5. The method for intelligent fusion and anomaly recognition of multi-source electric meter data according to claim 4, wherein, For the smart meter number identified as abnormal, call the original electrical parameter data of other smart meters to generate estimated electrical parameter values, including: Within the current continuous time window, select multiple smart meter numbers with a finite path distance from the cross-device measurement data set to the smart meter number identified as abnormal in the topological path directed distance matrix, and extract the original electrical parameter data of these smart meter numbers in the dimension of the main electrical parameter to form a set of topologically adjacent electrical parameter data; Calculate a correlation weight vector based on the Pearson correlation coefficient between the smart meter numbers based on topological proximity and the smart meter numbers identified as abnormal for the corresponding leading electrical parameters in multiple historical consecutive time windows, for determining the linear combination contribution coefficients of each neighboring meter to the estimated electrical parameter value; Combine the electrical wiring information and the topological path directed distance matrix to determine the path directionality and level difference of each neighboring smart meter number relative to the smart meter number identified as abnormal, and construct a topological ratio correction vector for adjusting the linear combination contribution coefficient according to the path direction, where the adjustment coefficient of the upstream node is less than 1 and the adjustment coefficient of the downstream node is greater than 1; Perform an element-wise product fusion of the correlation weight vector and the topological ratio correction vector to obtain a weighted correction coefficient vector, and use this weighted correction coefficient vector as a weighting factor to perform a weighted summation operation on the topological neighboring electrical parameter data set to generate the estimated electrical parameter value of the smart meter number identified as abnormal in this current consecutive time window.

Images