An industrial metering point abnormal data identification and repair method
By constructing a measurement point association graph and a machine learning model, and combining the topological relationship identification and repair of abnormal data in the industrial metrology system, the problem of insufficient accuracy in abnormal data identification and insufficient confidence in repair in the existing technology is solved, and more efficient industrial metrology data processing is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- CHANGCHUN UNIV OF TECH
- Filing Date
- 2026-05-13
- Publication Date
- 2026-07-21
AI Technical Summary
Existing technologies struggle to effectively utilize the correlation between measurement points in industrial metering systems, resulting in low accuracy in identifying abnormal data, difficulty in distinguishing between real and pseudo-anomalies, and a lack of confidence in the repair results, which affects the accuracy of energy consumption accounting and prediction models.
A measurement point association graph is constructed. Combining the relationships between branches at the same level, adjacent branches with high correlation, and branches between superior and subordinate branches, anomaly scores and types are generated using a machine learning model. The judgment path is output through a rule interpretation model, and data repair is performed by combining historical trends and estimated values of associated measurement points.
It improves the accuracy of anomaly identification and the reliability of repair, can distinguish different types of anomalies, provides a traceable repair path, and ensures data quality.
Smart Images
Figure CN122174197B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the fields of industrial energy data processing, metering data anomaly identification and machine learning technology, and specifically relates to a method for identifying and repairing abnormal data at industrial metering points. Background Technology
[0002] Energy management systems in industrial parks, factories, workshops, and production lines typically include metering points for various energy sources such as electricity, water, gas, heat, and steam. These metering points are arranged hierarchically, including master meters, sub-meters, branch meters, and equipment meters, and generate continuous metering data according to a fixed data collection cycle.
[0003] In actual operation, due to factors such as metering device failure, communication link interruption, data acquisition gateway malfunction, instrument drift, delays in manual data entry, and sudden changes in production load, missing values, abrupt values, bad values, and short-term spikes are prone to appear in the metering point data. If energy consumption statistics, energy balance analysis, or predictive modeling are directly based on the above-mentioned abnormal data, it is easy to lead to inaccurate energy consumption accounting, abnormal amplification of branch losses, and bias in the training of subsequent predictive models.
[0004] In existing technologies, fixed thresholds, manual empirical rules, or single time series smoothing methods are commonly used to handle anomalies in measurement data. For example, outliers are identified by setting maximum value thresholds, minimum value thresholds, or thresholds for changes between adjacent time points, and then repaired using mean interpolation, linear interpolation, or previous value padding. While these methods are simple to implement, they mainly rely on the data from individual measurement points and are difficult to utilize the inherent hierarchical branch relationships and correlations between measurement points at the same level within industrial measurement systems.
[0005] Furthermore, anomalies in industrial metering data are not always erroneous. For example, when a production line starts up, stops, or switches loads, multiple related metering points may change simultaneously; this change reflects actual operating conditions. However, when only the target metering point experiences a sudden change, but its upstream metering points, downstream branch metering points, and highly correlated measuring points do not show corresponding changes, this sudden change is more likely to be a bad point or a data acquisition anomaly. Existing methods struggle to distinguish between genuine anomalies, pseudo-anomalies, and bad point anomalies, easily leading to incorrect corrections of actual load changes or the retention of erroneous data in subsequent business processes.
[0006] Furthermore, existing anomaly handling methods typically only output a single judgment result of "abnormal" or "normal," lacking anomaly type, judgment basis, rule path, and repair confidence. This is not conducive to verification by operation and maintenance personnel, nor to the formation of a traceable data quality management process in industrial energy management systems.
[0007] Therefore, it is necessary to provide a method for identifying and repairing abnormal data of industrial metering points that can combine the time-series characteristics of the metering points themselves, the characteristics of highly correlated metering points at the same level, and the balance relationship between upper and lower branches, so as to improve the accuracy of anomaly identification, the ability to distinguish anomaly types, and the reliability of data repair. Summary of the Invention
[0008] To address the problems in existing technologies, such as reliance on single-point thresholds for anomaly identification, difficulty in utilizing measurement point correlations, difficulty in distinguishing between real and pseudo-anomalies, and lack of confidence in repair results, this invention provides a method for identifying and repairing abnormal data from industrial measurement points.
[0009] The technical solution adopted by this invention to solve the above-mentioned technical problems is as follows: acquiring time-series data of multiple metering points, topological hierarchical relationships, and anomaly labeling information in an industrial energy metering system, and constructing an anomaly identification feature system; constructing a metering point association diagram based on the relationships between metering points at the same level, the relationships between adjacent highly correlated metering points, and the relationships between upper and lower level branches; generating anomaly scores, anomaly types, and anomaly confidence levels using an anomaly identification model; extracting anomaly determination paths using a rule interpretation model; and generating repair data using an anomaly repair model combined with the historical trends of target metering points, estimated values of associated metering points, and hierarchical balance constraints.
[0010] This invention provides a method for identifying and repairing abnormal data of industrial measurement points based on measurement point association graphs and machine learning, including data acquisition, data preprocessing, feature construction, measurement point association graph construction, anomaly identification, rule interpretation, relationship stability analysis, current data evaluation, and anomaly repair output.
[0011] The method includes the following steps:
[0012] Step 1: Data Acquisition and Sample Set Construction
[0013] Historical metering data, metering point topology data, equipment operating status data, and anomaly labeling information from multiple metering points in an industrial energy metering system are acquired to construct a metering point sample set. The historical metering data includes the metering values of the target metering point and its associated metering points at multiple acquisition times. The metering point topology data includes the hierarchical relationships between the master table and sub-tables, the relationships between branches at the same level, and the spatial relationships between adjacent production areas or adjacent equipment. The anomaly labeling information can be generated from manual inspection records, metering device fault records, data supplementation records, energy consumption balance verification records, and simulated anomaly samples.
[0014] Step 2, Data Preprocessing:
[0015] The historical measurement data undergoes time alignment, missing segment location, initial screening of abnormal spikes, unit consistency, and normalization to generate standardized measurement sequences and data quality markers. Specifically, time alignment maps data from different measurement points to a unified acquisition time; missing segment location identifies continuous and discrete missing segments; and initial screening of abnormal spikes identifies candidate outliers that may be jump values or bad values.
[0016] Step 3: Construction of an anomaly identification feature system:
[0017] Based on the standardized measurement sequence, measurement point topological relationship data, and data quality markers, an anomaly identification feature system is constructed. This anomaly identification feature system includes at least time-varying features, jump / fluctuation features, hierarchical balance features, correlation consistency features, and data quality features.
[0018] Step 4: Constructing the measurement point association diagram:
[0019] A measurement point association graph is constructed based on the relationships between measurement points at the same level, between adjacent highly correlated measurement points, and between measurement points on upper and lower branches. Nodes in the graph represent measurement points, and edges represent the relationships between them. These edges may include hierarchical edges, correlated edges, and spatial adjacency edges.
[0020] Step 5: Anomaly detection model processing:
[0021] The aforementioned anomaly identification feature system and measurement point correlation diagram are input into the anomaly identification model to generate the anomaly score, anomaly type, and anomaly confidence level of the target measurement point at the target time. The anomaly types include missing anomalies, jump anomalies, bad point anomalies, true anomalies, and false anomalies.
[0022] Step 6: Training the rule interpretation model:
[0023] Based on the anomaly score, anomaly type, and anomaly confidence level output by the anomaly recognition model, a rule interpretation model is trained to extract rule paths characterizing the anomaly determination process. These rule paths include key features involved in anomaly determination, feature threshold ranges, and the order in which features are determined.
[0024] Step 7: Stability analysis of candidate associations:
[0025] Resampling stability analysis is performed on candidate associations among multiple measurement points. The frequency of occurrence, directional consistency and contribution of candidate associations in multiple resampling are statistically analyzed. Candidate associations that meet the preset stability conditions are screened and the measurement point association diagram is updated.
[0026] Step 8: Current Data Evaluation
[0027] Obtain the current measurement data of the measurement point to be identified, and generate the current evaluation features based on the current measurement data, the updated measurement point association diagram, and the anomaly identification feature system.
[0028] Step 9, Anomaly Identification and Data Repair Output:
[0029] The current evaluation features are input into the anomaly identification model and the rule interpretation model, and combined with the anomaly repair model to output the anomaly identification result, rule path, repair data, and repair confidence of the measurement point to be identified. When the anomaly type is a pseudo-anomaly, the current measurement data is retained and a pseudo-anomaly label is output; when the anomaly type is a missing anomaly, a jump anomaly, or a bad pixel anomaly, the repair data is output to replace the current measurement data; when the anomaly type is a true anomaly, the current measurement data is retained and a true anomaly label is output.
[0030] Compared with the prior art, the present invention has the following beneficial effects:
[0031] First, by constructing a measurement point association diagram, this invention incorporates measurement points at the same level, highly correlated measurement points, and measurement points of upper and lower branch lines into the anomaly identification process, which can avoid misjudgment caused by relying solely on single-point thresholds.
[0032] Second, this invention comprehensively judges the source of anomalies by using time change characteristics, jump fluctuation characteristics, hierarchical balance characteristics, and correlation consistency characteristics, and can distinguish between missing anomalies, jump anomalies, bad point anomalies, real anomalies, and pseudo-anomalies.
[0033] Third, this invention outputs anomaly determination paths through a rule interpretation model, making the anomaly identification results traceable and facilitating review by operations and maintenance personnel and subsequent data governance.
[0034] Fourth, this invention generates repair data by combining historical trends, related measurement point estimation, and upper and lower branch balance constraints through an anomaly repair model, which can improve the reliability of industrial measurement data repair. Attached Figure Description
[0035] Figure 1 This is a schematic diagram of the overall architecture of the industrial metering point abnormal data identification and repair system of the present invention.
[0036] Figure 2 This is a schematic diagram of the process for identifying and repairing abnormal data at industrial metering points according to the present invention.
[0037] Figure 3 This is a schematic diagram of the construction of the measurement point association diagram of the present invention.
[0038] Figure 4 This is a schematic diagram of the processing flow of the anomaly identification model and anomaly repair model of the present invention.
[0039] Figure 5This is a schematic diagram of the anomaly determination path output by the rule interpretation model of this invention.
[0040] Figure 6 This is a schematic diagram illustrating the logic for distinguishing between real anomalies, pseudo-anomalies, and bad pixel anomalies in this invention. Detailed Implementation
[0041] The present invention will now be described in detail with reference to the accompanying drawings. It should be understood that the following embodiments are for illustrative purposes only and are not intended to limit the scope of protection of the present invention.
[0042] like Figure 1 As shown, the system in this embodiment includes a data acquisition unit 101, a preprocessing unit 102, a feature construction unit 103, a measurement point association graph construction unit 104, an anomaly identification unit 105, a rule interpretation unit 106, a relationship stability analysis unit 107, an anomaly repair unit 108, and a result output unit 109.
[0043] The data acquisition unit 101 acquires historical metering data, metering point topology data, equipment operating status data, and anomaly labeling information from multiple metering points in the industrial energy metering system. The preprocessing unit 102 generates standardized metering sequences and data quality markers. The feature construction unit 103 constructs an anomaly identification feature system. The metering point association graph construction unit 104 constructs a metering point association graph. The anomaly identification unit 105 outputs anomaly scores, anomaly types, and anomaly confidence levels. The rule interpretation unit 106 outputs rule paths. The relationship stability analysis unit 107 updates the metering point association graph. The anomaly repair unit 108 generates repair data and repair confidence levels. The result output unit 109 outputs anomaly identification results, rule paths, repair data, and repair confidence levels.
[0044] like Figure 2 As shown, the method flow of the present invention includes: obtaining a sample set of measurement points, performing data preprocessing, constructing an anomaly identification feature system, constructing a measurement point association graph, performing anomaly identification, extracting rule paths, performing candidate association stability analysis, generating current evaluation features, and outputting anomaly identification results and repair data.
[0045] In one embodiment, the industrial energy metering system includes multiple metering points, each generating metering data according to a fixed data collection cycle. The metering points can be electricity meters, water meters, gas meters, heat meters, steam flow meters, or other industrial energy metering devices.
[0046] In one embodiment, the original measurement sequence of multiple measurement points is represented as follows:
[0047] (1)
[0048] in, Indicates the first The original measurement sequence of the measurement points; Indicates the metering point number; Indicates the first Each metering point at the time of data collection The measured value; Indicates the first Each data collection moment; Indicates the data collection time sequence number; Indicates the total number of data collection moments; This represents the transpose symbol.
[0049] In one embodiment, the original measurement sequence is normalized to obtain standardized measurement values:
[0050] (2)
[0051] in, Indicates the first Each metering point at the time of data collection Standardized measurement values; Indicates the first Each metering point at the time of data collection The original measurement value; Indicates the first The average value of each measurement point within a historical time window; Indicates the first The standard deviation of measurement points within a historical time window; This represents a positive constant used to avoid a denominator of zero; Indicates the metering point number; Indicates the first Each data collection moment; Indicates the sequence number of the data collection time.
[0052] In one embodiment, the difference between adjacent time points is expressed as:
[0053] (3)
[0054] in, Indicates the first Each metering point at the time of data collection The difference between adjacent time points; Indicates the first Each metering point at the time of data collection Standardized measurement values; Indicates the first Each metering point at the time of data collection Standardized measurement values; Indicates the metering point number; Indicates the first Each data collection moment; Indicates the first Each data collection moment; Indicates the sequence number of the data collection time.
[0055] In one embodiment, the correlation coefficient between the target measurement point and the candidate associated measurement point is expressed as:
[0056] (4)
[0057] in, Indicates the first The measurement point and the first The correlation coefficient between the measurement points; and These represent the numbers of different measurement points; Indicates the first Each metering point at the time of data collection The measured value; Indicates the first Each metering point at the time of data collection The measured value; Indicates the first The average measurement value of each measurement point within the statistical time window; Indicates the first The average measurement value of each measurement point within the statistical time window; Indicates the first Each data collection moment; Indicates the data collection time sequence number; This represents the total number of data collection moments within the statistical time window.
[0058] In one embodiment, if the first Each metering point is a higher-level metering point, and the set of its lower-level branch metering points is... The hierarchical balance residual is then expressed as:
[0059] (5)
[0060] in, Indicates the first Each superior metering point at the time of data collection Hierarchical balance residuals; Indicates the number of the superior metering point; Indicates the first Each superior metering point at the time of data collection The measured value; Indicates the first The set of lower-level branch metering points corresponding to each higher-level metering point; Indicates the metering point number of the lower-level branch road; Indicates the first Each lower-level branch metering point at the time of data collection The measured value; Indicates the first Each data collection moment; Indicates the sequence number of the data collection time.
[0061] like Figure 3 As shown, the measurement point association graph includes a node set and an edge set. Each node in the node set corresponds to a measurement point, and each edge in the edge set corresponds to the association relationship between two measurement points. Association relationships include hierarchical branch relationships, same-level relationships, and high-relevance relationships.
[0062] In one embodiment, the measurement point association diagram is represented as follows:
[0063] (6)
[0064] in, This represents the correlation diagram of measurement points; Represents a set of nodes; Represents the set of edges; Represents the adjacency matrix; a set of nodes. The nodes in the table represent metering points in an industrial energy metering system; the edge set... The edges in the adjacency matrix represent the relationships between measurement points; The elements in the table are used to indicate whether there is a correlation or the strength of the correlation between two measurement points.
[0065] In one embodiment, the elements in the adjacency matrix are represented as:
[0066] (7)
[0067] in, Representing the adjacency matrix The Middle Line number Column elements; and These represent the numbers of different measurement points; Indicates the first The measurement point and the first Candidate association edges between measurement points; Represents the set of edges in the graph of the measurement point association; when When, it indicates that there is at least one of the following relationships between the two measurement points: a hierarchical relationship, a peer relationship, or a high correlation relationship; when When the condition is met, it indicates that there is no correlation between the two measurement points that meets the preset conditions.
[0068] In one embodiment, the anomaly identification model includes a temporal feature encoding network, a graph association feature aggregation network, and an anomaly classification network. The temporal feature encoding network is used to extract the historical trend of the target measurement point, the graph association feature aggregation network is used to extract the consistency features between the target measurement point and its adjacent associated measurement points, and the anomaly classification network is used to output the anomaly score, anomaly type, and anomaly confidence level.
[0069] In one embodiment, the anomaly score is represented as:
[0070] (8)
[0071] in, Indicates the first Each metering point at the time of data collection Abnormal scores; Indicates the first Each metering point at the time of data collection Reconstruction deviation score; Indicates the first Each metering point at the time of data collection Prediction bias score; Indicates the first Each metering point at the time of data collection Hierarchical balance deviation score; Indicates the first Each metering point at the time of data collection Data quality score; , , and These represent the weighting coefficients for the reconstruction bias score, prediction bias score, hierarchical balance bias score, and data quality score, respectively. Indicates the metering point number; Indicates the first Each data collection moment; Indicates the sequence number of the data collection time.
[0072] like Figure 4 As shown, after the anomaly identification model outputs an anomaly score, it further outputs the anomaly type. If the target measurement point has missing data while the related measurement point data is complete, it is judged as a missing data anomaly; if the target measurement point experiences a short-term sudden increase or decrease, and the residual of the balance between the upper and lower branches increases significantly, it is judged as a jump anomaly or a bad point anomaly; if the target measurement point, highly correlated measurement points at the same level, and measurement points of the upper and lower branches change synchronously and the hierarchical balance relationship is not disrupted, it is judged as a real anomaly or a false anomaly.
[0073] In one embodiment, the rule interpretation model employs a decision tree, rule list, or gradient boosting tree interpretation model. The rule interpretation model uses the anomaly type output by the anomaly detection model as the learning objective and features from the anomaly detection feature system as input to extract anomaly determination paths.
[0074] like Figure 5 As shown, the rule path may include: first, determining whether the hierarchical balance residual exceeds the first threshold; if it exceeds the first threshold, then determining whether the correlation deviation between the target measurement point and the highly correlated measurement point exceeds the second threshold; if the correlation deviation exceeds the second threshold, then determining whether the sliding window deviation exceeds the third threshold; if the sliding window deviation exceeds the third threshold and the data quality label is abnormal, then outputting bad point abnormality or jump abnormality.
[0075] In one embodiment, the stability index of candidate associations is expressed as:
[0076] (9)
[0077] in, Indicates the first The measurement point and the first The frequency of occurrence of candidate associations between measurement points; and These represent the numbers of different measurement points; Indicates the number of resampling attempts; Indicates the first Secondary resampling; Indicates the first In the second resampling The measurement point and the first An indicator variable indicating whether a candidate association exists between measurement points. The value is 1 when a candidate association exists and 0 when no candidate association exists.
[0078] In one embodiment, if the frequency of occurrence of a candidate association is not less than a frequency threshold, the directional consistency is not less than a directional consistency threshold, and the contribution is not less than a contribution threshold, then the candidate association is written into the measurement point association graph. This method can reduce the impact of accidental associations on anomaly identification results.
[0079] In one embodiment, the anomaly repair model generates repair data based on the historical trend prediction of the target metering point, the estimated value of highly correlated metering points at the same level, and the estimated value of the balance constraint of the upper and lower level branches.
[0080] In one embodiment, the repair data is represented as:
[0081] (10)
[0082] in, Indicates the first Each metering point at the time of data collection Repair data; This indicates the weight corresponding to the historical trend prediction value; Indicates based on the first The estimated measurement values obtained from the historical trend prediction of each measurement point; This indicates the weights corresponding to the estimated values of highly relevant econometric points at the same level. This indicates the first value estimated based on highly correlated econometric points at the same level. Estimated measurement values for each measurement point; This represents the weights corresponding to the estimated values of the balance constraints between upper and lower level branches; This represents the first branch obtained based on the balance constraint estimation of the upper and lower branches. Estimated measurement values for each measurement point; Indicates the metering point number; Indicates the first Each data collection moment; Indicates the sequence number of the data collection time.
[0083] In one embodiment, the repair confidence is represented as:
[0084] (11)
[0085] in, Indicates the first Each metering point at the time of data collection Repair confidence; Indicates the first Each metering point at the time of data collection Repair data; Indicates the first Each metering point at the time of data collection The reference estimate is generated from at least one of historical trend predictions, associated measurement point estimates, or upper and lower branch balance constraint estimates. This represents a positive constant used to avoid a denominator of zero; Indicates the metering point number; Indicates the first Each data collection moment; Indicates the sequence number of the data collection time.
[0086] like Figure 6As shown, this invention distinguishes between real anomalies, pseudo-anomalies, and bad pixel anomalies. If the change in the value of a target meter changes synchronously with its superior meter, subordinate branch meter, and highly correlated meter, and the hierarchical balance residual does not exceed a preset range, it is determined to be a real or pseudo-anomaly, and replacement repair is not directly performed. If the target meter shows an abnormal change on its own, but the associated meter does not change accordingly, and the hierarchical balance residual exceeds a preset range, it is determined to be a bad pixel anomaly, and repair data replacement is performed.
[0087] In one embodiment, the output includes anomaly identification results, anomaly type, anomaly score, anomaly confidence level, rule path, repair data, repair confidence level, and processing suggestions. The processing suggestions include retaining the current data, replacing it with repair data, initiating manual review, or triggering a metering device inspection.
[0088] This invention is not limited to the embodiments described above. Various modifications and improvements can be made by those skilled in the art without departing from the principles of this invention, and these modifications and improvements should also be considered within the scope of protection of this invention.
Claims
1. A method for identifying and repairing abnormal data at industrial metering points, characterized in that, Performed by an electronic device, including the following steps: S1. Obtain historical metering data, metering point topology data, equipment operating status data, and anomaly labeling information from multiple metering points in the industrial energy metering system, and construct a metering point sample set. S2. Perform time alignment, missing segment location, abnormal peak screening, unit consistency and normalization processing on the historical measurement data to generate standardized measurement sequences and data quality markers. S3. Based on the standardized measurement sequence, measurement point topological relationship data, and data quality markers, construct an anomaly identification feature system; S4. Construct a measurement point association diagram based on the relationships between measurement points at the same level, the relationships between adjacent highly correlated measurement points, and the relationships between measurement points on upper and lower branch roads; S5. Input the anomaly identification feature system and measurement point association diagram into the anomaly identification model to generate anomaly score, anomaly type and anomaly confidence. S6. Based on the anomaly score, anomaly type, and anomaly confidence, train a rule interpretation model and extract rule paths; S7. Perform resampling stability analysis on the candidate associations, screen the candidate associations that meet the preset stability conditions, and update the measurement point association diagram; S8. Based on the current measurement data of the measurement point to be identified, the updated measurement point association diagram, and the anomaly identification feature system, generate the current evaluation features; S9. Input the current evaluation features into the anomaly identification model and the rule interpretation model, and combine them with the anomaly repair model to output the anomaly identification result, rule path, repair data and repair confidence.
2. The method according to claim 1, characterized in that, In step S1, the anomaly labeling information is generated from at least one of the following: historical manual inspection records, metering device fault records, data supplement records, energy consumption balance verification records, and simulated anomaly samples generated based on normal metering sequences.
3. The method according to claim 1, characterized in that, In step S2, the missing segment localization includes identifying continuous missing segments and discrete missing segments; the abnormal peak screening includes identifying candidate peak points based on at least one of the following methods: adjacent time difference score, sliding window median deviation, and upper and lower branch balance residual.
4. The method according to claim 1, characterized in that, In step S3, the anomaly identification feature system includes time change features, jump fluctuation features, hierarchical balance features, correlation consistency features, and data quality features; The time variation characteristics include periodic mean, periodic variance, sliding window trend value, and difference between adjacent time points; The jump fluctuation characteristics include instantaneous rate of change, sliding window deviation, peak duration, and fall amplitude; The hierarchical balance characteristics include balance residual, balance residual change rate, and branch contribution ratio; The correlation consistency characteristics include correlation coefficient, synchronous change ratio, and correlation deviation. The data quality characteristics include the missing value ratio, the duplicate value ratio, the duration of zero values, and the sampling interval deviation.
5. The method according to claim 4, characterized in that, The balance residual is characterized by the difference between the sum of the measurement values of the upper-level metering point and the corresponding lower-level branch metering points; the synchronous change ratio is characterized by the proportion of the number of times when the target metering point and the associated metering point change in the same direction within the same time window to the length of the time window. The correlation deviation is characterized by the difference between the actual measurement value of the target measurement point and the estimated measurement value predicted based on the correlation measurement point.
6. The method according to claim 1, characterized in that, In step S5, the anomaly identification model includes a temporal feature encoding network, a graph association feature aggregation network, and an anomaly classification network; the anomaly types include missing anomalies, jump anomalies, bad pixel anomalies, real anomalies, and pseudo-anomalies.
7. The method according to claim 1, characterized in that, In step S6, the rule path includes: using hierarchical balance residuals as the first-level judgment feature, correlation deviation as the second-level judgment feature, sliding window deviation as the third-level judgment feature, and combining data quality features to determine the anomaly type.
8. The method according to claim 1, characterized in that, In step S7, the preset stability conditions include an occurrence frequency threshold, a direction consistency threshold, and a contribution threshold; when the occurrence frequency of a candidate association is not less than the occurrence frequency threshold, the direction consistency is not less than the direction consistency threshold, and the contribution is not less than the contribution threshold, the candidate association is written into the measurement point association diagram.
9. The method according to claim 1, characterized in that, In step S9, the anomaly repair model generates repair data based on the historical trend prediction value of the target metering point, the estimated value of highly correlated metering points at the same level, the estimated value of the balance constraint of the upper and lower branches, and the anomaly type. When the anomaly type is a pseudo-anomaly or a real anomaly, the current metering data is retained and the corresponding anomaly label is output. When the anomaly type is a missing anomaly, a jump anomaly, or a bad point anomaly, the repair data is output to replace the current metering data.