Power distribution area line loss anomaly detection method based on consistency fuzzy check

Abstract

The application discloses a transformer area line loss anomaly detection method based on consistent fuzzy verification and belongs to the technical field of electric digital data processing. The method first constructs a reference archive set and a to-be-analyzed archive set with attribute fields and decision fields having a cause-effect relationship according to anomaly detection requirements; subsequently, the advantages or disadvantages between archive records are determined, and upper and lower approximate sets are dynamically constructed based on the selected relationship type, and the difference set is used as a boundary domain representing logical inconsistency; finally, the records in the boundary domain are scored and sorted or compared with correct rules according to application scenarios, and anomaly analysis results are output. The application can identify non-outlier anomalies, overcome the dependence of traditional clustering methods on overall distance, achieve efficient research and judgment only with a small amount of typical samples, reduce data and computing power requirements, and improve the identification accuracy of complex archive anomaly patterns and operation and maintenance troubleshooting efficiency.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of digital data processing, and particularly relates to a transformer area line loss anomaly detection method based on consistent fuzzy verification. BACKGROUND

[0002] In the field of power distribution data analysis, the management line loss judgment of the power transformer area is a key link to ensure the economic operation of the power grid. After the meter is installed or replaced, the key information such as CT / PT configuration, line formula, installation and removal time needs to be accurately recorded in the power transformer area file. Then, the system calculates the theoretical line loss according to the real-time collected data and the configuration information in the file. In actual operation, problems such as incorrect comprehensive ratio, inconsistent wiring mode, chaotic relationship, inconsistent formula, inconsistent meter reading period, and inaccurate installation and removal time record will directly lead to inaccurate calculation of the supply-in and supply-out power, and further cause the management line loss anomaly.

[0003] The traditional system generally lacks the ability to automatically verify when the file is changed, and can only respond passively after the line loss anomaly occurs. Moreover, when the operation and maintenance personnel investigate, due to the lack of effective technical means, they often need to manually compare each field of all suspicious records or rely on unreliable experience, which not only reduces the work efficiency, but also makes it difficult to ensure the success rate of error positioning.

[0004] To improve the efficiency of file anomaly identification, various solutions for analyzing and investigating abnormal file records have been proposed by those skilled in the art, but all have limitations:

[0005] 1. Clustering-based method: Although this method is simple to implement, it has inherent defects in its mechanism of relying on outlier identification: when only a single field in a multi-field file record is incorrect, the overall record may not have a significant outlier distance from normal records, making it difficult to effectively find such "non-outlier anomalies".

[0006] 2. Rule-based method: This method relies on static rules obtained by manual presetting or statistics. When facing the diversity of file structure and the dynamic changes of business scenarios, the coverage and adaptability of the rules are often insufficient, affecting the accuracy of analysis and judgment.

[0007] 3. Supervised learning method based on graph network classification: This method is limited by the current situation of non-uniform power file business structure and the scarcity of labeled samples, and has limited practical application scenarios.

[0008] 4. AI large model-based method: Although this method has improved data processing capability, its deployment cost is high, and in the business background where file data is ambiguous, there is a risk of model illusion, and its reliability cannot be guaranteed.

[0009] In summary, the prior art solution has deficiencies in the accuracy and adaptability of power archive anomaly detection, and cannot meet the actual needs of power transformer area line loss fine management. SUMMARY

[0010] The present application provides a transformer area line loss anomaly detection method based on consistency fuzzy verification, which aims to: first, accurately identify various problem archives including non-outlier anomalies; second, dynamically adapt to the diversity of archive structure and changes in business scenarios without relying on pre-set static rules; third, achieve effective research and judgment of suspected abnormal archives on the premise of only a small number of typical samples without relying on large-scale labeled data and complex models.

[0011] The technical solution of the present application is as follows:

[0012] A transformer area line loss anomaly detection method based on consistency fuzzy verification, comprising the following steps:

[0013] Step S1. Construct a reference archive set T and a to-be-analyzed archive set U according to the anomaly detection requirements;

[0014] The reference archive set T contains more than one reference archive record, the to-be-analyzed archive set U contains more than one to-be-analyzed archive record, the reference archive record and the to-be-analyzed archive record have the same field structure, the field structure contains a plurality of attribute fields and a decision field, there is a causal relationship between the attribute field and the decision field, the attribute field is "cause" and the decision field is "effect";

[0015] Step S2. Perform archive consistency fuzzy research and judgment on the reference archive set T and the to-be-analyzed archive set U to obtain a consistent boundary domain, the boundary domain is a subset of the to-be-analyzed archive set U, used to represent the uncertain state of the two archive records in the attribute field and the decision field logical relationship;

[0016] Step S3. Based on the to-be-analyzed archive record in the boundary domain obtained in step S2, perform anomaly detection analysis and output the analysis result.

[0017] As a further improvement of the above-mentioned transformer area line loss anomaly detection method based on consistency fuzzy verification, the consistency fuzzy research and judgment in step S2 comprises:

[0018] Step S21. Traverse all to-be-analyzed archive records, and for each to-be-analyzed archive record, execute the following steps S22-S23;

[0019] Step S22. Perform fuzzy relationship judgment on the current to-be-analyzed archive record and all reference archive records, if each fuzzy relationship judgment meets the above approximate judgment condition, add the current to-be-analyzed archive record to the above approximate set;

[0020] Step S23. Perform fuzzy relationship judgment between the current to-be-analyzed archive record and all reference archive records respectively, and if each fuzzy relationship judgment meets the lower approximate judgment condition, add the current to-be-analyzed archive record to the lower approximate set;

[0021] Step S24. After the traversal is completed, obtain the upper approximate set and the lower approximate set, and take the difference set of the upper approximate set and the lower approximate set as the boundary domain.

[0022] As a further improvement of the above-mentioned transformer area line loss anomaly detection method based on consistency fuzzy checking, for any one to-be-analyzed archive record and any one reference archive record , the fuzzy relationship judgment between the two is performed in the following manner: calculate the fuzzy relationship degrees between the same field values in the to-be-analyzed archive record and the reference archive record respectively, and then determine whether the current fuzzy relationship judgment meets the upper approximate judgment condition or the lower approximate judgment condition according to the relationship between the fuzzy relationship degrees of each field and the preset threshold in the upper approximate judgment condition or the lower approximate judgment condition.

[0023] As a further improvement of the above-mentioned transformer area line loss anomaly detection method based on consistency fuzzy checking: the fuzzy relationship judgment includes two types of superior relationship judgment and inferior relationship judgment; the types of fuzzy relationship judgment performed in steps S22 and S23 must be the same.

[0024] As a further improvement of the above-mentioned transformer area line loss anomaly detection method based on consistency fuzzy checking, for any one to-be-analyzed archive record and any one reference archive record :

[0025] The manner of performing superior relationship judgment is as follows:

[0026] ;

[0027] In the above formula, the superior relationship judgment result is established only when the right side condition is met. , , are the set of text attribute fields and the set of numerical attribute fields in the archive record respectively; when , and represent the vector of attribute field in the to-be-analyzed archive record and the vector of attribute field in the reference archive record respectively; when , and These represent the files to be analyzed. Attribute fields Numerical values ​​and reference files Attribute fields The value; express and Fuzzy similarity, express and fuzzy order membership degree; and These represent the files to be analyzed. Values ​​and reference files for decision fields The value of the decision field in the middle. express and fuzzy order membership degree; , , These are the preset attribute fuzzy similarity advantage threshold, attribute fuzzy order membership advantage threshold, and decision fuzzy order membership advantage threshold in the current upper approximation judgment or lower approximation judgment, respectively.

[0028] The method for determining a disadvantageous relationship is as follows:

[0029] ;

[0030] In the above equation, the condition on the right-hand side is satisfied if and only if the condition on the right-hand side is satisfied. The assessment of a disadvantageous relationship is valid. , , These are the preset thresholds for attribute fuzzy similarity disadvantage, attribute fuzzy order membership disadvantage, and decision fuzzy order membership disadvantage in the current upper or lower approximation judgment.

[0031] As a further improvement to the above-mentioned method for detecting abnormal line loss in transformer substations based on consistency fuzzy verification, and Fuzzy similarity The calculation method is as follows:

[0032] .

[0033] As a further improvement to the above-mentioned method for detecting abnormal line loss in transformer substations based on consistency fuzzy verification, for any two values... and Fuzzy order membership degree The calculation method is as follows:

[0034] ;

[0035] In the above formula, is a parameter for controlling fuzzy order relation.

[0036] As a further improvement of the above-mentioned transformer area line loss anomaly detection method based on consistency fuzzy checking, if the current anomaly detection requirement is to investigate the current archive record in the system, then:

[0037] In step S1, the reference archive set T is composed of typical correct archive records screened by manual selection, and the archive set U to be analyzed is composed of archive records to be investigated at present;

[0038] In step S3, the anomaly score of each archive record to be analyzed in the boundary domain is calculated, and the archive records with high ranking are marked as potential anomaly archives according to the anomaly score.

[0039] As a further improvement of the above-mentioned transformer area line loss anomaly detection method based on consistency fuzzy checking, for any archive record to be analyzed in the boundary domain , the calculation method of the anomaly score is:

[0040] ;

[0041] In the above formula, , are the set of text type attribute fields and the set of numerical type attribute fields in the archive record, respectively; when , and represent the vector of attribute field in the archive record to be analyzed and the vector of attribute field in the archive record to be analyzed , respectively; when , and represent the numerical value of attribute field in the archive record to be analyzed and the numerical value of attribute field in the archive record to be analyzed , respectively; represents the fuzzy similarity of and , represents the fuzzy order membership degree of and .

[0042] As a further improvement of the above-mentioned transformer area line loss anomaly detection method based on consistency fuzzy checking, if the current anomaly detection requirement is that the system has already appeared anomaly, and according to the anomaly, the archive record in the system which may have a correlation with the anomaly has been locked, then:

[0043] In step S1, the reference file set T is composed of the above-mentioned file records possibly associated with abnormalities, and the file set U to be analyzed is composed of all file records in the system;

[0044] In step S3, the file records to be analyzed in the boundary domain are file records that are correct but deviate from the reference file set T, the file records in the boundary domain are compared and analyzed with the above-mentioned file records possibly associated with abnormalities as reference, and whether the above-mentioned file records possibly associated with abnormalities have abnormalities is investigated based on the rules embodied by the file records in the boundary domain.

[0045] Compared with the prior art, the present application has the following beneficial effects:

[0046] 1. The present application can effectively identify non-outlier abnormalities that cannot be found by traditional clustering methods by constructing a boundary domain through file consistency fuzzy judgment. Since a fuzzy relationship judgment mechanism is adopted between attribute fields and decision fields, even if only individual fields in the file records have logical deviations, as long as they fall into the uncertain state represented by the boundary domain in the overall logical relationship with the reference file, they can be screened out. This method breaks through the dependence of traditional outlier detection on overall distance difference and is more suitable for the actual business scenario of line loss abnormalities caused by single field errors in power files.

[0047] 2. Compared with methods that rely on static rules, the present application uses typical samples to construct a reference set and dynamically adapts to the diversity of file structure and business changes through fuzzy relationship thresholds. Since there is no need to preset fixed rules, when the system faces new CT / PT configurations, line formula changes and other scenarios, it only needs to update the reference file set to maintain the effectiveness of the analysis, avoiding misjudgment or omission due to insufficient rule coverage, and improving the adaptability of the system in a dynamic business environment.

[0048] 3. The present application only needs a small amount of typical samples as prior knowledge, generates a boundary domain through the difference set operation of the upper and lower approximate sets, and does not need to rely on large-scale labeled data or complex model training. This feature effectively overcomes the limitations of supervised learning methods and large model applications due to the scarcity of labeled samples and the non-uniform structure of power file business, while ensuring the reliability of anomaly detection and significantly reducing the cost of data preparation and model deployment.

[0049] 4. By introducing fuzzy similarity calculation of text type fields and fuzzy order membership evaluation of numerical type fields, the present application realizes fine comparison of multi-dimensional features of file records. This comprehensive analysis mechanism that combines text semantics and numerical logic can more accurately capture the internal consistency relationship between fields and improve the recognition accuracy of complex file abnormal patterns.

[0050] 5. The method can be flexibly configured according to different investigation needs. Whether it is for the active screening of the current system archives or for the retrospective analysis of the associated archives of the abnormal occurrence, it can be achieved by adjusting the set construction strategy to achieve targeted research and judgment, and is flexible and adaptable to various operation and maintenance scenarios, and has high practical value.

[0051] 6. The method realizes the quantitative evaluation and ordering of potential abnormal archives by designing an abnormal score calculation model based on the similarity and sequence relationship between records in the boundary domain. This mechanism provides clear investigation priorities for operation and maintenance personnel, effectively reduces the scope of manual verification, and improves the efficiency and accuracy of abnormal positioning. DETAILED DESCRIPTION

[0052] The technical solutions of the present application will be described in detail below. Obviously, the described embodiments are only part of the embodiments of the present application, not all.

[0053] First, the common fields in the power district archives and their meanings are described. The power district archives record various types of information related to line loss calculation, including meter number, comprehensive multiplier (CT / PT configuration), connection direction, parent (such as user, district, substation, line), formula, metering period, installation and removal time, meter clock, etc. Among them, the meter number is used as the primary key to uniquely identify the record; the comprehensive multiplier error will cause the electric quantity calculation to deviate; the connection direction error will affect the use of forward / reverse active power; the parent archive error will cause the supply-in and supply-out power mismatch; the formula error will also cause the line loss supply-in and supply-out mismatch; the power consumption anomaly (such as missing or zero) will directly cause the line loss anomaly; the time error in the meter replacement record may cause the replacement period power deviation; the meter clock deviation will cause the calculation period to be misaligned. Among these fields, any archive information that affects power calculation or line loss calculation belongs to the scope of concern of the present method.

[0054] For subsequent analysis, all records related to detection are first combined into a wide table according to business association. The wide table refers to a table in which multiple fields from different data tables are integrated according to business logic, and the fields include all attribute fields and decision fields that need to participate in analysis. The attribute fields are generally multiple, and the decision fields are usually one. The attribute fields are the aggregation of archive fields, such as comprehensive rate, belonging to the superior, and installation and removal table time, etc.; the decision field preferentially selects the power, installation time, etc. numerical field, which is usually the result data obtained by archive association. There should be a causal relationship between the attribute field and the decision field, that is, the attribute field as the "cause" and the decision field as the "effect". For example, the earlier the installation time and the larger the rate, the larger the historical cumulative power usually means; the power jump decrease will lead to line loss anomaly, at this time the power can be used as the decision field, and the fields such as installation time and rate which affect the power are used as the attribute field. In actual analysis, the fields participating in analysis can be flexibly allocated, for example, 12 fields are selected in a certain analysis, of which 10 are attribute fields and 2 are decision fields, that is, the analysis wide table can be constructed in the form of 10+1 or 11+1.

[0055] The fields in the wide table are preprocessed. Numerical fields (including time type fields converted to time stamps) need to be standardized to eliminate dimensional influence and ensure that the data is on the same scale. Text fields need to be vectorized to convert them into vector representations that can be used for similarity calculation.

[0056] The implementation steps of the transformer area line loss anomaly detection method based on consistency fuzzy verification include:

[0057] Step S1. Construct a reference archive set T and a to-be-analyzed archive set U according to the anomaly detection requirements.

[0058] The reference archive set T contains more than one reference archive record, and the to-be-analyzed archive set U contains more than one to-be-analyzed archive record. The reference archive record and the to-be-analyzed archive record have the same field structure, and the field structure contains a plurality of attribute fields and a decision field. There is a causal relationship between the attribute field and the decision field, the attribute field is the "cause", and the decision field is the "effect".

[0059] Step S2. Archive consistency fuzzy research and judgment is performed on the reference archive set T and the to-be-analyzed archive set U to obtain a consistent boundary domain, which is a subset of the to-be-analyzed archive set U, used to represent the uncertain state of the two archive records in the attribute field and the decision field logical relationship.

[0060] The consistency fuzzy research and judgment in step S2 includes:

[0061] Step S21. Traverse all the records to be analyzed, and for each record to be analyzed, perform the following steps S22-S23.

[0062] Step S22. Perform fuzzy relationship judgment between the current record to be analyzed and each reference record, and if each fuzzy relationship judgment meets the upper approximation judgment condition, add the current record to be analyzed to the upper approximation set.

[0063] Step S23. Perform fuzzy relationship judgment between the current record to be analyzed and each reference record, and if each fuzzy relationship judgment meets the lower approximation judgment condition, add the current record to be analyzed to the lower approximation set.

[0064] For any record to be analyzed and any reference record , the fuzzy relationship judgment between them is performed as follows: the fuzzy relationship degrees between the same field values in the record to be analyzed and the reference record are calculated respectively, and then it is determined whether the current fuzzy relationship judgment meets the upper approximation judgment condition or the lower approximation judgment condition according to the relationship between the fuzzy relationship degrees and the preset threshold in the upper approximation judgment condition or the lower approximation judgment condition.

[0065] It should be noted that the fuzzy relationship judgment includes two types of superior relationship judgment and inferior relationship judgment, and the types of fuzzy relationship judgment performed in steps S22 and S23 must be the same. Whether to select the superior relationship or the inferior relationship depends on the "positive-negative correlation" between the attribute field and the decision field in the business. Positive correlation means that when the value of the attribute field increases, the value of the decision field also tends to increase, in which case the superior relationship is selected, and vice versa.

[0066] Specifically, for any record to be analyzed and any reference record , the superior relationship judgment is performed as follows:

[0067] ;

[0068] In the above formula, the right side condition is met only when the superior relationship judgment result is established; , are the set of text attribute fields and the set of numerical attribute fields in the record, respectively; when , and represent the vector of the attribute field in the record to be analyzed and the reference record, respectively. Attribute fields The vector; when hour, and These represent the files to be analyzed. Attribute fields Numerical values ​​and reference files Attribute fields The value; express and Fuzzy similarity, express and fuzzy order membership degree; and These represent the files to be analyzed. Values ​​and reference files for decision fields The value of the decision field in the middle. express and fuzzy order membership degree; , , These are the preset attribute fuzzy similarity advantage threshold, attribute fuzzy order membership advantage threshold, and decision fuzzy order membership advantage threshold in the current upper approximation judgment or lower approximation judgment, respectively.

[0069] and Fuzzy similarity The calculation method is as follows:

[0070] .

[0071] The range of fuzzy similarity is [0,1], and the closer it is to 1, the more similar it is.

[0072] For any two values and Fuzzy order membership degree The calculation method is as follows:

[0073] ;

[0074] In the above formula, It is a parameter that controls the fuzzy order relation, and is usually taken as 0.9 to 1.0.

[0075] The method for determining a disadvantageous relationship is as follows:

[0076] ;

[0077] In the above equation, the condition on the right-hand side is satisfied if and only if the condition on the right-hand side is satisfied. The assessment of a disadvantageous relationship is valid. 、 、 are respectively the preset attribute fuzzy similarity disadvantage threshold, attribute fuzzy order membership disadvantage threshold and decision fuzzy order membership disadvantage threshold in the current upper approximation judgment or lower approximation judgment.

[0078] The upper approximation set adopts a relatively loose threshold, for example, the advantage relation judgment takes 、 、 , and the disadvantage relation judgment takes 、 、 , indicating that the records between the archives may be similar and consistent; the lower approximation set adopts a relatively strict threshold, for example, the advantage relation judgment takes 、 、 , and the disadvantage relation judgment takes 、 、 , indicating that the records between the archives are similar and consistent. The boundary domain is the difference set of the upper approximation set and the lower approximation set, and the records therein are in an uncertain state and need to be further analyzed.

[0079] Step S24. After the traversal is completed, the upper approximation set and the lower approximation set are obtained, and the difference set of the upper approximation set and the lower approximation set is taken as the boundary domain.

[0080] Step S3. Abnormal detection analysis is performed based on the to-be-analyzed archive records in the boundary domain obtained in step S2, and an analysis result is output.

[0081] The first scenario is active screening for the current archives of the system. At this time, the reference archive set T in step S1 is composed of typical correct archive records manually screened, and the to-be-analyzed archive set U is composed of the current to-be-investigated archive records. After the boundary domain is obtained in step S3, the abnormal score of each to-be-analyzed archive record x in the boundary domain is calculated:

[0082] ;

[0083] wherein, is the boundary domain, and are respectively the text type and numerical type attribute field sets. The score reflects the inconsistency degree of the to-be-analyzed record and other records in the attribute field in the boundary domain, and the higher the score, the greater the abnormal possibility. Finally, the archive records are sorted according to the abnormal score, and the archive records at the top of the sorting are marked as potential abnormal archives for manual intensive investigation.

[0084] The second scenario is the backtracking analysis of the associated archives for the line loss abnormality that has occurred. At this time, the reference archive set T in step S1 is composed of the archive records that are possibly associated with the abnormality and are automatically locked by the system, and the archive set U to be analyzed is composed of all archive records in the system. After the boundary domain is obtained in step S3, the archive records to be analyzed in the boundary domain are the records that are correct but deviate from the reference archive set T. These records are used as the reference benchmark to compare and analyze the abnormality associated archive records, and the rules embodied by the correct records in the boundary domain (for example, the metering ratio in a certain area is usually 0.1, and the abnormality ratio is 0.01) are used to investigate whether the abnormality associated archive records have problems.

[0085] It should be noted that, for those skilled in the art, it is obvious that the present application is not limited to the details of the above exemplary embodiments, and the present application can be implemented in other specific forms without departing from the spirit or essential characteristics of the present application. The scope of the present application is defined by the claims rather than the above description.

Claims

1. A method for detecting abnormal line loss in transformer substations based on consistency fuzzy verification, characterized in that, Includes the following steps: Step S1. Construct a reference file set T and a file set U to be analyzed to meet the anomaly detection requirements; The reference archive set T contains one or more reference archive records, and the archive set U to be analyzed contains one or more archive records to be analyzed. The reference archive records and the archive records to be analyzed have the same field structure. The field structure contains several attribute fields and one decision field. There is a causal relationship between the attribute fields and the decision field. The attribute fields are the "cause" and the decision field is the "effect". Step S2. Perform fuzzy analysis on the consistency of the reference archive set T and the archive set U to be analyzed to obtain the consistency boundary region. The boundary region is a subset of the archive set U to be analyzed and is used to characterize the uncertain state of the two archive records in the logical relationship between attribute fields and decision fields. The consistency fuzzy assessment in step S2 includes: Step S21. Traverse all the records to be analyzed, and perform the following steps S22-S23 for each record to be analyzed; Step S22. Perform fuzzy relation judgments on the current file record to be analyzed and all reference file records respectively. If each fuzzy relation judgment meets the above approximation judgment condition, then add the current file record to be analyzed to the above approximation set. Step S23. Perform fuzzy relation judgments on the current file record to be analyzed and all reference file records respectively. If each fuzzy relation judgment meets the next approximation judgment condition, then add the current file record to be analyzed to the next approximation set. For any record to be analyzed and any reference file record The method for determining the fuzzy relationship between the two is as follows: calculate the records to be analyzed separately. and reference archives The degree of fuzzy relationship between the same field values ​​is determined, and then the relationship between the degree of fuzzy relationship of each field and the preset threshold in the corresponding upper or lower approximation judgment condition is used to determine whether the current fuzzy relationship judgment meets the upper or lower approximation judgment condition. The fuzzy relation judgment includes two types: dominant relation judgment and disadvantageous relation judgment; the fuzzy relation judgment types performed in steps S22 and S23 must be the same. For any file to be analyzed and any reference file : The method for determining dominance relationships is as follows: ； In the above equation, the condition on the right-hand side is satisfied if and only if the condition on the right-hand side is satisfied. The assessment of the dominant relationship is correct; , These are the collections of text-type attribute fields and the collection of numeric attribute fields in the archive records, respectively; when hour, and These represent the files to be analyzed. Attribute fields vectors and reference files Attribute fields The vector; when hour, and These represent the files to be analyzed. Attribute fields Numerical values ​​and reference files Attribute fields The value; express and Fuzzy similarity, express and fuzzy order membership degree; and These represent the files to be analyzed. Values ​​and reference files for decision fields The value of the decision field in the middle. express and fuzzy order membership degree; , , These are the preset attribute fuzzy similarity advantage threshold, attribute fuzzy order membership advantage threshold, and decision fuzzy order membership advantage threshold in the current upper approximation judgment or lower approximation judgment, respectively. The method for determining a disadvantageous relationship is as follows: ； In the above equation, the condition on the right-hand side is satisfied if and only if the condition on the right-hand side is satisfied. The assessment of a disadvantageous relationship is valid. , , These are the preset attribute fuzzy similarity disadvantage threshold, attribute fuzzy order membership disadvantage threshold, and decision fuzzy order membership disadvantage threshold in the current upper approximation judgment or lower approximation judgment, respectively. For any two values and Fuzzy order membership degree The calculation method is as follows: ； In the above formula, These are parameters that control fuzzy order relations; Step S24. After the traversal is completed, the upper approximation set and the lower approximation set are obtained, and the difference between the upper approximation set and the lower approximation set is taken as the boundary domain; Step S3. Perform anomaly detection analysis on the archive records to be analyzed in the boundary domain obtained in Step S2, and output the analysis results.

2. The method for detecting abnormal line loss in transformer substations based on consistency fuzzy verification as described in claim 1, characterized in that, and Fuzzy similarity The calculation method is as follows: 。 3. The method for detecting abnormal line loss in transformer substations based on consistency fuzzy verification as described in claim 1, characterized in that, If the current anomaly detection requirement is to check the current archive records in the system, then: In step S1, the reference file set T consists of typical correct file records selected by manual screening, and the file set U to be analyzed consists of the file records currently to be investigated. In step S3, the anomaly score of each record to be analyzed in the boundary domain is calculated, and the records are sorted according to the anomaly score. The records with the highest scores are marked as potential anomaly records.

4. The method for detecting abnormal line loss in transformer substations based on consistency fuzzy verification as described in claim 3, characterized in that, For the boundary domain Any file record to be analyzed in The abnormal score is calculated as follows: ； In the above formula, , These are the collections of text-type attribute fields and the collection of numeric attribute fields in the archive records, respectively; when hour, and These represent the files to be analyzed. Attribute fields Vectors and files to be analyzed Attribute fields The vector; when hour, and These represent the files to be analyzed. Attribute fields The numerical values ​​and the files to be analyzed Attribute fields The value; express and Fuzzy similarity, express and The fuzzy order membership degree.

5. The method for detecting abnormal line loss in transformer substations based on consistency fuzzy verification as described in claim 1, characterized in that, If the current anomaly detection requirement is that an anomaly has occurred in the system, and based on the anomaly, file records in the system that may be associated with the anomaly have been identified, then: In step S1, the reference file set T consists of the file records that may be associated with the anomaly, and the file set U to be analyzed consists of all file records in the system. In step S3, the file records to be analyzed in the boundary domain are correct but deviate from the reference file set T. The file records in the boundary domain are used as a reference benchmark and compared with the file records that may be associated with the anomaly. Based on the patterns shown by the file records in the boundary domain, it is determined whether the file records that may be associated with the anomaly are abnormal.

Images