Intelligent checking method and system for power grid data management based on four-dimensional cross validation

Abstract

The invention discloses a power grid data management intelligent checking method and system based on four-dimensional cross validation. The method comprises the following steps: acquiring data from each service system of a power grid and preprocessing to generate a to-be-verified data set; performing value, rule, relation and attribute dimension verification on the to-be-verified data set in sequence; summarizing four-dimensional verification results of value-rule-relationship-attribute and dividing anomaly levels to generate an anomaly diagnosis report; and based on feedback data of an abnormal diagnosis result and an actual rectification result in the report, mining an association rule to expand a rule base, dynamically adjusting the weight of the rule in the rule base, and detecting and eliminating a failure rule. According to the method, full-dimension anomaly recognition can be achieved, complex scenes can be covered, the false alarm rate is effectively reduced, design rules are self-optimized, a rule base can be autonomously evolved, manual intervention does not need to be completely relied on, the rule updating period is greatly shortened, and dynamic changes of power grid services and data can be rapidly responded.

Description

Technical Field

[0001] This invention relates to an intelligent verification method in the field of power systems, and more particularly to an intelligent verification method for power grid data governance based on four-dimensional cross-validation, and also to an intelligent verification system for power grid data governance based on four-dimensional cross-validation. Background Technology

[0002] Data quality is a core prerequisite for realizing the value of power grid data, directly impacting the accuracy of power grid simulation calculations, real-time scheduling, and operation and maintenance decisions. Currently, power grid data quality verification mainly relies on traditional methods based on manual verification and static rule-driven approaches. Traditional methods typically use pre-defined static rules such as "numerical range" and "format specifications" to verify the compliance of power grid data, including equipment parameters and operational data. This method focuses on the compliance of the data's "values" themselves, such as exceeding limits, format errors, and missing fields. Data processing is primarily offline manual analysis, and rule maintenance relies on manual iteration. In practical applications, a typical process is as follows: operation and maintenance personnel export the data to be verified from the database, perform initial screening using pre-defined rule scripts (such as Python scripts), manually review each suspected abnormal data, and generate a rectification report after confirming the anomaly, which is then pushed to the business departments for correction.

[0003] However, traditional methods have three significant drawbacks: First, the verification dimension is singular, limited to compliance checks of the data's "value" itself, failing to effectively identify hidden anomalies at the "relationship" and "attribute" levels, such as topological inconsistencies and business logic conflicts. This results in an anomaly detection rate generally below 75% in complex scenarios. Second, the rule design does not deeply integrate with power grid business scenarios and physical laws, leading to a false alarm rate as high as 15%-18%. For example, temporary voltage fluctuations during fault repair are often misjudged as anomalies, requiring significant manpower for verification. Finally, the maintenance of the rule base relies entirely on manual labor, with the cycle from proposing new business requirements to rule design, coding, and deployment taking several days, severely lagging behind the dynamic changes in power grid business and failing to meet the requirements of real-time and automated data quality control for digital twin power grids. Summary of the Invention

[0004] To address the technical problems of existing power grid data quality verification methods, such as single verification dimension, high false alarm rate, and lagging rule maintenance, this invention provides a smart verification method and system for power grid data governance based on four-dimensional cross-validation.

[0005] This invention is achieved using the following technical solution: an intelligent verification method for power grid data governance based on four-dimensional cross-validation, comprising the following steps: Data is collected from various power grid business systems and preprocessed to generate a dataset to be verified. The following steps are performed on the dataset to be verified in sequence: 1) Verify the data in the "value" dimension based on the designed multiple basic rules; 2) Verify the data in the "rule" dimension based on the designed multiple business rules; 3) Import the data into the constructed topology graph model to detect node / edge attribute anomalies for "relationship" dimension verification; 4) Perform physical law verification and machine learning verification, and then weight and fuse the obtained parameter logic anomalies and data inference anomalies to obtain the "attribute" dimension verification result. Summarize the verification results across four dimensions: "value-rule-relationship-attribute," classify the anomaly levels, and generate an anomaly diagnosis report; Based on the feedback data of the abnormal diagnosis results and actual rectification results in the report, we mine association rules to expand the rule base, dynamically adjust the weight of rules in the rule base, and detect and eliminate invalid rules.

[0006] This invention addresses the issue of a single verification dimension by constructing a four-dimensional feature cross-validation system of "value-rule-relationship-attribute". This expands the verification scope from the traditional single "value" dimension to include a "rule" dimension encompassing business rules, a topological "relationship" dimension based on graph computation, and an "attribute" dimension integrating physical laws and AI. Furthermore, by introducing a hybrid diagnostic model integrating power grid physical laws and machine learning, along with a rule self-optimization mechanism, this invention combines deterministic verification based on power grid physical laws with probabilistic diagnosis based on machine learning algorithms. Using weighted fusion, it fully leverages the high accuracy of physical laws in deterministic judgments and utilizes AI models to identify implicit correlation patterns, significantly reducing misjudgments caused by special business scenarios (such as maintenance and faults), thus solving the problem of high false alarm rates. Finally, by designing a rule self-optimization mechanism, this invention automatically mines new rules from historical data and diagnostic feedback, dynamically adjusts rule weights based on performance indicators, and promptly detects and eliminates invalid rules. This mechanism enables the rule base to evolve autonomously without complete reliance on manual intervention, greatly shortening the rule update cycle and solving the problem of lagging rule maintenance.

[0007] As a further improvement to the above scheme, the intelligent verification method for power grid data governance also includes the following steps: An anomaly diagnosis report is pushed to operations and maintenance personnel / systems via an intelligent verification and management platform. After viewing the anomaly details, the operations and maintenance personnel / systems dispatch a rectification work order to the corresponding business responsibility department through the intelligent verification and management platform. After the business responsibility department completes the rectification and uploads the rectification results to the intelligent verification and management platform, the operations and maintenance personnel / systems verify the rectification results on the intelligent verification and management platform. If the verification is successful, the corresponding anomaly is marked as closed-loop management; otherwise, the work order is returned and rectification is required again. The intelligent verification and management platform records the rectification progress and results and updates the data quality score.

[0008] Furthermore, the intelligent verification method for power grid data governance also includes the following steps: The intelligent verification management platform monitors core indicators including data verification volume, anomaly identification rate and rectification completion rate in real time and displays the core indicators in a visual interface; the intelligent verification management platform periodically and automatically generates data quality verification reports; the data quality verification reports include: (1) a data verification overview including the number of data verification entries, the number of passed entries, and the pass rate of each system; (2) anomaly analysis results including the distribution of anomaly types, the distribution of severity levels, and the distribution of equipment types involved; (3) the rule effectiveness including the ranking of the accuracy rate, coverage rate, and false alarm rate of each rule; and (4) the anomaly rectification situation including the number of rectification work orders dispatched, the number of completed orders, the completion rate, and the average rectification time.

[0009] As a further improvement to the above scheme, there are four categories of basic rules, namely, numerical range, format specification, non-empty integrity, and uniqueness; there are three categories of business rules, namely, cross-system consistency, state transition, and data freshness; the "value" dimension verification and the "rule" dimension verification are executed through the rule engine, and the performance is optimized by rule grouping, caching, and multi-threaded parallel mechanism.

[0010] As a further improvement to the above scheme, the "relationship" dimension verification specifically includes: (1) Construct a power grid topology model based on a graph database. The topology model includes nodes that represent power grid equipment or stations and edges that represent connection or affiliation relationships. (2) Detect missing or incorrect connection relationships in the topology model based on the breadth-first search algorithm; detect the logical consistency of topology loops and devices within loops in the topology model based on the depth-first search algorithm; divide the topology model into communities based on the community detection algorithm and detect the voltage level correlation contradictions of nodes within the community.

[0011] As a further improvement to the above scheme, the physical law verification specifically involves constructing a verification function based on the physical laws of the power grid, calculating the parameter deviations of power balance, transformer voltage ratio, or line impedance, and identifying parameter logic anomalies based on the deviation threshold; the machine learning verification specifically involves inferring from the operating data based on a pre-trained machine learning model to identify trend anomalies, correlation anomalies, or outlier anomalies; the verification result of the "attribute" dimension is a confirmed anomaly or a suspected anomaly.

[0012] As a further improvement to the above scheme, the association mining rules specifically include: The historical data and the feedback data are preprocessed, including discretizing continuous features; The Apriori algorithm is used to mine frequent itemsets with support at or above a preset threshold. Candidate rules are generated based on the frequent itemsets, and candidate rules with confidence scores not lower than a first threshold and lift scores greater than a second threshold are selected and included in the rule base.

[0013] As a further improvement to the above scheme, the dynamic adjustment of the weights of rules in the rule base specifically includes: Periodically calculate the accuracy, coverage, and false alarm rate of each rule in the rule base; In response to the increase in accuracy and the decrease in false positive rate, the weight of the corresponding rule is increased; In response to the decrease in accuracy and the increase in false positive rate, the weight of the corresponding rule is reduced.

[0014] As a further improvement to the above solution, the business system is at least one of a GIS system, an EMS system, and a marketing system. The collected data includes equipment parameters, operating data, and topology data. The collected data is then cleaned of duplicate data, filled with missing values, and converted in format to obtain the dataset to be verified.

[0015] This invention also provides an intelligent verification system for power grid data governance based on four-dimensional cross-validation, which is used to implement any of the above-described intelligent verification methods for power grid data governance based on four-dimensional cross-validation. The intelligent verification system for power grid data governance comprises: The four-dimensional feature verification module performs a four-dimensional verification of the dataset to be verified, which includes "value-rule-relationship-attribute". A hybrid diagnostic model is communicatively connected to the four-dimensional feature verification module and is used to summarize the four-dimensional verification results and generate an anomaly diagnostic report. A rule self-optimization engine, which is communicatively connected to the hybrid diagnostic model, is used to perform the association rule mining, the dynamic adjustment of the rule weights, and the detection and elimination of the failed rules; The intelligent verification management platform communicates with the four-dimensional feature verification module, the hybrid diagnostic model, and the rule self-optimization engine, and is used to provide data access, anomaly visualization, anomaly rectification work order dispatch, and report generation functions.

[0016] Compared with existing power grid data quality verification methods, the intelligent verification method and system for power grid data governance based on four-dimensional cross-validation of the present invention has the following beneficial effects: 1. This intelligent verification method for power grid data governance based on four-dimensional cross-validation constructs a four-dimensional feature cross-validation system of "value-rule-relationship-attribute". It expands the verification scope from the traditional single "value" dimension to include the "rule" dimension containing business rules, the topological "relationship" dimension based on graph computing, and the "attribute" dimension that integrates physical laws and AI. This solves the problem of single verification dimension and enables comprehensive and in-depth mining of complex hidden anomalies such as semantic ambiguity, topological contradictions, and business logic conflicts. It can identify anomalies in all dimensions and cover complex scenarios.

[0017] 2. This intelligent verification method for power grid data governance based on four-dimensional cross-validation introduces a hybrid diagnostic model that integrates power grid physical laws and machine learning, along with a rule self-optimization mechanism. It combines deterministic verification based on power grid physical laws with probabilistic diagnosis based on machine learning algorithms, and employs weighted fusion. This fully leverages the high accuracy of physical laws in deterministic judgments and uses AI models to identify implicit correlation patterns, thereby significantly reducing misjudgments caused by special business scenarios (such as maintenance and faults), solving the problem of high false alarm rates, and effectively reducing the false alarm rate.

[0018] 3. The intelligent verification method for power grid data governance based on four-dimensional cross-validation achieves automatic mining of new rules from historical data and diagnostic feedback, dynamic adjustment of rule weights according to performance indicators, and timely detection and elimination of invalid rules through rule self-optimization. This mechanism enables the rule base to evolve autonomously without relying entirely on manual intervention, greatly shortens the rule update cycle, solves the problem of lagging rule maintenance, and can quickly respond to the dynamic changes in power grid business and data.

[0019] 4. This intelligent verification method for power grid data governance based on four-dimensional cross-validation, after testing, shows that by combining physical laws with machine learning to identify business logic conflicts, it can improve the complex anomaly identification rate from the traditional <75% to ≥92.5%, with a 30% improvement in semantic anomaly identification rate, fully covering the multi-dimensional data quality control needs of the digital twin power grid. Traditional methods, due to their lack of integration with power grid business scenarios and physical laws, have a false alarm rate as high as 15%-18%. This intelligent verification method for power grid data governance ensures the accuracy of deterministic anomaly identification through physical laws, optimizes the judgment of implicit anomalies through machine learning algorithms, and dynamically adjusts rule weights based on business scenarios (such as lowering the weight of "power is 0" anomaly judgment during maintenance), reducing the false alarm rate to below 8% and reducing the workload of manual review by more than 40%, significantly improving operation and maintenance efficiency. Traditional rule maintenance relies on manual methods, requiring 10-15 new rules to be added for each new type of equipment, which takes more than 24 hours. This intelligent verification method for power grid data governance automatically mines new rules through the Apriori algorithm, improving rule expansion efficiency by 80%. At the same time, it dynamically adjusts weights and eliminates invalid rules based on feedback loops, shortening the rule update cycle from several days to hours, and can quickly adapt to changes in power grid business (such as new energy grid connection and maintenance process adjustments).

[0020] 5. This intelligent verification method for power grid data governance, based on four-dimensional cross-validation, enables real-time and automated verification, supporting dynamic management and control. Traditional methods primarily rely on offline batch processing, with verification time exceeding one hour for a single batch of 100,000 data entries, and requiring manual intervention for anomaly responses. This new method optimizes verification efficiency, reducing verification time for a single batch of 100,000 data entries to ≤3 minutes, and anomaly response time to the second level. Combined with an intelligent verification management platform, it achieves full-process automation of "data access-verification-anomaly handling-rectification closed loop," meeting the real-time data quality control needs of the digital twin power grid. This intelligent verification method for power grid data governance integrates data access, verification, visualization analysis, and report generation functions. Maintenance personnel can complete anomaly viewing, work order dispatching, and rectification acceptance on a single interface, improving data quality control efficiency by over 50%. It also supports multi-departmental data sharing and collaboration, promoting the standardization of power grid data governance.

[0021] 6. The beneficial effects of the intelligent verification system for power grid data governance based on four-dimensional cross-validation are the same as those of the aforementioned intelligent verification method for power grid data governance, and will not be elaborated here. Attached Figure Description

[0022] Figure 1 This is a flowchart of the intelligent verification method for power grid data governance based on four-dimensional cross-validation in Embodiment 1 of the present invention. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0024] Existing power grid data quality verification methods are based on manual verification and static rule-driven approaches. The specific technical solutions are as follows: Data verification relies on predefined single rules: static rules such as "numerical range" and "format specifications" are manually preset to verify the compliance of power grid data (such as equipment parameters and operational data). For example, "the rated current range of a 220kV transformer is set to 500-2000A", and anything exceeding this range is considered abnormal; the "equipment ID format is standardized to 'voltage level-equipment type-number'", and any mismatch in the format is marked as abnormal.

[0025] Verification dimensions are limited to the "value" level: traditional methods only focus on the compliance of the data "value" itself, such as numerical limits, format errors, and missing fields, and cannot identify hidden anomalies at the "relationship" and "attribute" levels. For example, they cannot detect topological inconsistencies such as "a 10kV line connecting to a 220kV substation", nor can they discover attribute logic conflicts such as "the rated capacity of the transformer does not match the voltage level".

[0026] Data processing primarily relies on offline manual analysis: collected power grid data must first be stored in a database, then manually extracted and verified offline using Excel or simple analysis tools. For example, maintenance personnel periodically export equipment status data from the GIS and EMS systems, manually compare their consistency, and only enter issues such as "inconsistent operating status" into the rectification log.

[0027] Rule maintenance relies on manual iteration: When grid operations change (such as adding new equipment types or adjusting operating standards), new rules need to be redesigned, coded, and deployed manually. For example, when a new requirement for "grid connection data verification of new energy power plants" is added, rules such as "inverter power factor range rules" and "photovoltaic module temperature threshold rules" need to be added manually. The whole process relies on professional technicians and is time-consuming and costly.

[0028] In practical applications, the typical process for existing technologies is as follows: maintenance personnel export data to be verified from the database, perform initial screening using preset rule scripts (such as Python scripts), and manually review each suspected abnormal data. After confirming the abnormality, a rectification report is generated and the business department is urged to make corrections. This process is highly dependent on manual intervention and cannot cover hidden anomalies in complex scenarios, making it difficult to meet the data quality control needs of large-scale, highly dynamic power grids.

[0029] Therefore, existing power grid data quality verification methods have the following drawbacks.

[0030] First, the verification dimension is too single, resulting in a weak ability to identify hidden anomalies: Traditional methods only cover the "value" dimension for verification, and cannot associate multi-dimensional features such as "rules-relationships-attributes". This leads to a general recognition rate of less than 75% for complex anomalies such as semantic ambiguity (e.g., the expression "110kV Chengdong Substation #1 Main Transformer" is inconsistent with "Chengdong 110kV Substation No.1 Main Transformer"), topological contradictions (e.g., incorrect equipment ownership relationship), and business logic conflicts.

[0031] Second, the false alarm rate is high, and the cost of manual verification is high: the rule design does not incorporate the power grid business scenario and physical laws, and it is easy to misjudge data that conforms to special business logic as abnormal. For example, normal data such as "temporary voltage fluctuations during fault repair" and "equipment power of 0 during maintenance" are judged as abnormal, with a false alarm rate as high as 15%-18%, requiring a lot of manpower for verification and occupying operation and maintenance resources.

[0032] Third, rule maintenance is lagging behind and cannot adapt to dynamic changes in business: rule updates rely entirely on manual intervention, and adding a new type of equipment requires 10-15 new rules, which takes an average of more than 24 hours; when business processes are adjusted (such as changes in equipment maintenance processes), the rule iteration cycle can take as long as 3-5 working days, resulting in the verification rules being out of touch with actual business and unable to cover new abnormal patterns.

[0033] Fourth, the verification efficiency is low and cannot support real-time control: Traditional methods mainly rely on offline batch processing, and the verification time for a single batch of 100,000 data records exceeds 1 hour. Moreover, abnormal response requires manual intervention, and the cycle from data collection to abnormal rectification can last for several days, which cannot meet the "real-time and automated" control requirements of the digital twin power grid for data quality.

[0034] To overcome the above shortcomings, the present invention provides the following embodiments, which can break through the limitations of single-dimensional verification, construct a four-dimensional feature cross-verification system of "value-rule-relationship-attribute", integrate the physical laws of power grids and machine learning algorithms, reduce the false alarm rate of anomalies, and establish a self-optimization mechanism for diagnostic rules, realize the automatic mining of new rules from historical data, dynamically adjust the rule weights based on diagnostic feedback, significantly shorten the rule update cycle, and optimize the verification process and algorithm efficiency.

[0035] Example 1 Please see Figure 1 This embodiment provides an intelligent verification method for power grid data governance based on four-dimensional cross-validation. This verification method deeply integrates rule engines, graph computing, power grid physical laws, and machine learning techniques to construct an intelligent verification system of "full-dimensional diagnosis and dynamic optimization," achieving automated and accurate identification of power grid data quality anomalies. In this embodiment, the intelligent verification method for power grid data governance includes the following steps.

[0036] Step 1, Data Access and Preprocessing: Data is collected from various business systems of the power grid and preprocessed to generate a dataset to be verified. In this embodiment, the specific steps are as follows: (1) The intelligent verification management platform collects data from various business systems of the power grid (the business system is at least one of the GIS system, EMS system, and marketing system) through standardized interfaces (such as API, direct database connection). The collected data may include equipment parameters, operating data, and topology data. (2) The platform preprocesses the collected data: cleans up duplicate data, fills in missing values ​​(mean for numerical data and mode for categorical data), converts the format (unified date / equipment ID format), and generates a "dataset to be verified". (3) The dataset to be verified is pushed to the four-dimensional feature verification module through the data layer interface.

[0037] Step 2, Four-dimensional feature cross-validation: The following steps are performed sequentially on the dataset to be validated: (1) Validate the data in the "value" dimension based on the designed multi-class basic rules; (2) Validate the data in the "rule" dimension based on the designed multi-class business rules; (3) Import the data into the constructed topology graph model to detect node / edge attribute anomalies for "relationship" dimension validation; (4) Perform physical law validation and machine learning validation, and weight and fuse the obtained parameter logic anomalies and data inference anomalies to obtain the "attribute" dimension validation result. In this way, this step breaks through the limitations of traditional single-dimensional validation through the "value-rule-relationship-attribute" four-dimensional feature cross-validation framework, and realizes the full-dimensional recognition of complex anomalies of semantic, topological, and business logic types.

[0038] In this embodiment, there are four categories of basic rules: numerical range, format specification, non-empty integrity, and uniqueness. There are three categories of business rules: cross-system consistency, state transition, and data freshness. Value-level and rule-level validations are performed by a rule engine, and performance is optimized using rule grouping, caching, and multi-threaded parallel mechanisms.

[0039] During the "value-rule" dimension validation, the JSON format rule file is first loaded through the rule engine, parsed into a DRL script, and then the data to be validated is encapsulated into a POJO object and input into the rule engine. The Rete algorithm matches the "value-rule" dimension rules, and finally outputs abnormal results in the "value-rule" dimension (such as numerical limits exceeding limits or cross-system data inconsistency) and stores them in the abnormal database.

[0040] When performing "relationship" dimension verification, the "relationship" dimension verification can specifically include: (1) constructing a power grid topology model based on a graph database. The topology model includes nodes representing power grid equipment or stations and edges representing connection or affiliation relationships; (2) detecting missing or incorrect connection relationships in the topology model based on a breadth-first search algorithm; detecting topology loops and logical consistency of equipment within loops in the topology model based on a depth-first search algorithm; dividing the topology model into communities based on a community detection algorithm and detecting voltage level correlation contradictions among nodes within the community. Finally, outputting the "relationship" dimension anomaly results, associating the topology graph node IDs, and pushing them to the hybrid diagnostic model. If a graph database is not used here, but a relational database (such as MySQL) is used to store the topology data, and simple correlation queries are implemented through SQL statements, the SQL query can only identify simple connection relationship errors and cannot detect hidden anomalies such as "voltage level correlation contradictions" and "topology logic errors". The topology anomaly identification rate will drop from ≥96% in this scheme to <60%.

[0041] During the "attribute" dimension verification, the parameters / operating data of the device to be verified are first input into the "attribute" dimension verification submodule. Then, physical law verification and machine learning verification are performed. Finally, a weighted voting method is used to fuse the results of physical laws and machine learning, outputting "confirmed anomaly / suspected anomaly," which is then pushed to the hybrid diagnostic model. Specifically, physical law verification involves constructing a verification function based on power grid physical laws, calculating parameter deviations for power balance, transformer voltage ratio, or line impedance, and identifying logical anomalies based on deviation thresholds. Machine learning verification involves inferring from operating data using a pre-trained machine learning model (random forest, GAM, isolated forest) to identify trend anomalies, correlation anomalies, or outlier anomalies; the "attribute" dimension verification result is either a confirmed anomaly or a suspected anomaly. This fusion of power grid physical laws (power balance, transformer voltage ratio) and machine learning algorithms (random forest, GAM, isolated forest) to construct a hybrid diagnostic model reduces false alarm rates and improves the ability to identify latent anomalies.

[0042] It should be noted that if the four-dimensional fusion verification of "value-rule-relationship-attribute" is not adopted, but instead the four dimensions are broken down into independent modules, with each module outputting results separately without cross-validation, complex anomaly identification cannot be achieved, and the performance is far inferior to that of this embodiment. This difference is demonstrated by comparing indicators such as anomaly identification rate and false alarm rate. If physical laws are completely abandoned and only a single machine learning algorithm is used, the accuracy of deterministic anomaly identification (such as numerical exceeding limits) will decrease.

[0043] Step 3: Hybrid Diagnosis and Anomaly Output: Summarize the verification results across the four dimensions of "value-rule-relationship-attribute" and classify the anomalies into levels, generating an anomaly diagnosis report. Specifically, the hybrid diagnosis model summarizes the verification results across these four dimensions and classifies the anomalies into "severe / moderate / minor" levels (e.g., "voltage level correlation contradiction" is a severe anomaly, "format error" is a moderate anomaly). The model generates an "anomaly diagnosis report," which includes the anomaly ID, data ID, anomaly type, anomaly description, affected equipment scope, and suggested corrective measures. The anomaly diagnosis report is pushed to the intelligent verification management platform, which updates the anomaly list and visualization interface.

[0044] Step 4, Rule Self-Optimization: Based on the feedback data of abnormal diagnostic results and actual rectification results in the report (such as TP true positive, FP false positive, FN false negative), a "rule feedback dataset" is formed (the rule feedback dataset is synchronized to the rule self-optimization engine every week). The association rules are mined to expand the rule base, the weight of the rules in the rule base is dynamically adjusted, and invalid rules are detected and eliminated.

[0045] In this embodiment, the process of mining association rules may specifically include: preprocessing historical and feedback data, including discretizing continuous features; mining frequent itemsets with support no less than a preset threshold (which can be set to 5%) using the Apriori algorithm; generating candidate rules based on frequent itemsets, and selecting candidate rules with confidence no less than a first threshold (which can be set to 80%) and lift greater than a second threshold (which can be set to 1.2) to be included in the rule base. Finally, the candidate rules are manually reviewed (review rate ≤ 20%) to remove rules with contradictory business logic (such as "equipment status = decommissioned → normal operating data"), and valid rules are included in the rule base, thus completing the automatic rule expansion.

[0046] Thus, this embodiment, based on Apriori association rule mining and feedback loop, designs a rule self-optimization mechanism to achieve automatic discovery of new rules, dynamic adjustment of rule weights, and elimination of invalid rules, reducing manual maintenance costs. If the Apriori algorithm is not used to automatically mine new rules, and instead a semi-automatic rule maintenance is achieved through "manually preset templates + system-automated parameter filling," it cannot cover unknown abnormal patterns, and the rule expansion efficiency is far lower than in this embodiment.

[0047] Dynamically adjusting the weights of rules in the rule base can specifically include: periodically calculating the accuracy (TP / (TP+FP)), coverage (TP / (TP+FN)), and false positive rate (FP / (FP+TN)) of each rule in the rule base; increasing the weight of the corresponding rule in response to an increase in accuracy and a decrease in false positive rate; and decreasing the weight of the corresponding rule in response to a decrease in accuracy and an increase in false positive rate. In this embodiment, the weights are adjusted based on the indicator results: a 5% increase in accuracy and a 3% decrease in false positive rate → a 5%-10% increase in weight; a 5% decrease in accuracy and a 3% increase in false positive rate → a 5%-10% decrease in weight; indicator fluctuations ≤2% → the weight remains unchanged. For special scenarios (such as grid connection of new energy power plants), manual temporary adjustment of rule weights is supported, and the default weights are automatically restored after the scenario ends.

[0048] This embodiment uses a rule failure detection submodule to check the effectiveness of rules weekly. If a rule meets any of the following conditions for three consecutive weeks: "accuracy ≤ 60%", "coverage ≤ 30%", or "false positive rate ≥ 20%", it is marked as a "failed rule". Furthermore, business experts are notified to review the failed rule. If the failure is confirmed to be due to business changes or data model changes, the rule is marked "disabled" and removed from the rule library. Finally, the Apriori association rule mining submodule is triggered to generate alternative rules, ensuring the integrity and effectiveness of the rule library.

[0049] Step 5, Anomaly Handling and Closed-Loop Management: Anomaly diagnostic reports are pushed to maintenance personnel / systems via an intelligent verification management platform (e.g., web interface notifications, SMS reminders). After the maintenance personnel / systems view the anomaly details (e.g., location of the abnormal device, associated data, cause of the anomaly), a rectification work order is dispatched to the corresponding business responsibility department through the intelligent verification management platform. After the business responsibility department completes the rectification and uploads the rectification results to the intelligent verification management platform, the maintenance personnel / systems verify the rectification results on the intelligent verification management platform (e.g., corrected device parameters, topology adjustment records). If the verification is successful, the corresponding anomaly is marked as closed-loop managed; otherwise, the work order is returned and re-rectification is required. The intelligent verification management platform records the rectification progress and results and updates the data quality score. In some other embodiments, step 5 may be omitted.

[0050] Step 6, Data Quality Monitoring and Report Generation: Real-time monitoring of core indicators including data verification volume, anomaly identification rate, and rectification completion rate is conducted through the intelligent verification management platform, and the core indicators are displayed on the visualization interface; the intelligent verification management platform automatically generates data quality verification reports periodically (e.g., daily / weekly / monthly); the data quality verification report includes: (1) a data verification overview including the number of data verification entries, the number of passed entries, and the pass rate for each system; (2) anomaly analysis results including the distribution of anomaly types, severity levels, and the distribution of equipment types involved; (3) the rule effectiveness including the ranking of accuracy, coverage, and false alarm rate for each rule; (4) anomaly rectification status including the number of rectification work orders dispatched, the number of completed work orders, the completion rate, and the average rectification time. The report can be exported in Excel / PDF format for reporting by maintenance personnel and optimization reference by business departments. It should also be noted that in some other embodiments, step 6 may not be set.

[0051] In summary, compared with existing power grid data quality verification methods, the intelligent verification method for power grid data governance based on four-dimensional cross-validation in this embodiment has the following advantages: 1. This intelligent verification method for power grid data governance based on four-dimensional cross-validation constructs a four-dimensional feature cross-validation system of "value-rule-relationship-attribute". It expands the verification scope from the traditional single "value" dimension to include the "rule" dimension containing business rules, the topological "relationship" dimension based on graph computing, and the "attribute" dimension that integrates physical laws and AI. This solves the problem of single verification dimension and enables comprehensive and in-depth mining of complex hidden anomalies such as semantic ambiguity, topological contradictions, and business logic conflicts. It can identify anomalies in all dimensions and cover complex scenarios.

[0052] 2. This intelligent verification method for power grid data governance based on four-dimensional cross-validation introduces a hybrid diagnostic model that integrates power grid physical laws and machine learning, along with a rule self-optimization mechanism. It combines deterministic verification based on power grid physical laws with probabilistic diagnosis based on machine learning algorithms, and employs weighted fusion. This fully leverages the high accuracy of physical laws in deterministic judgments and uses AI models to identify implicit correlation patterns, thereby significantly reducing misjudgments caused by special business scenarios (such as maintenance and faults), solving the problem of high false alarm rates, and effectively reducing the false alarm rate.

[0053] 3. The intelligent verification method for power grid data governance based on four-dimensional cross-validation achieves automatic mining of new rules from historical data and diagnostic feedback, dynamic adjustment of rule weights according to performance indicators, and timely detection and elimination of invalid rules through rule self-optimization. This mechanism enables the rule base to evolve autonomously without relying entirely on manual intervention, greatly shortens the rule update cycle, solves the problem of lagging rule maintenance, and can quickly respond to the dynamic changes in power grid business and data.

[0054] 4. This intelligent verification method for power grid data governance based on four-dimensional cross-validation, after testing, shows that by combining physical laws and machine learning to identify business logic conflicts, it can improve the complex anomaly identification rate from the traditional <75% to ≥92.5%, with a 30% improvement in semantic anomaly identification rate, fully covering the multi-dimensional data quality control needs of the digital twin power grid. Traditional methods, due to their lack of integration with power grid business scenarios and physical laws, have a false alarm rate as high as 15%-18%. This intelligent verification method for power grid data governance ensures the accuracy of deterministic anomaly identification through physical laws (such as a weight of 0.6), optimizes the judgment of implicit anomalies through machine learning algorithms (such as a weight of 0.4), and dynamically adjusts rule weights based on business scenarios (such as lowering the weight of the "power is 0" anomaly judgment during maintenance), reducing the false alarm rate to below 8%, reducing the workload of manual review by more than 40%, and significantly improving operation and maintenance efficiency. Traditional rule maintenance relies on manual methods, requiring 10-15 new rules to be added for each new type of equipment, which takes more than 24 hours. This intelligent verification method for power grid data governance automatically mines new rules through the Apriori algorithm, improving rule expansion efficiency by 80%. At the same time, it dynamically adjusts weights and eliminates invalid rules based on feedback loops, shortening the rule update cycle from several days to hours, and can quickly adapt to changes in power grid business (such as new energy grid connection and maintenance process adjustments).

[0055] 5. This intelligent verification method for power grid data governance, based on four-dimensional cross-validation, enables real-time and automated verification, supporting dynamic management and control. Traditional methods primarily rely on offline batch processing, with verification time exceeding one hour for a single batch of 100,000 data entries, and requiring manual intervention for anomaly responses. This new method optimizes verification efficiency, reducing verification time for a single batch of 100,000 data entries to ≤3 minutes, and anomaly response time to the second level. Combined with an intelligent verification management platform, it achieves full-process automation of "data access-verification-anomaly handling-rectification closed loop," meeting the real-time data quality control needs of the digital twin power grid. Traditional methods require multi-system collaboration (such as Excel data analysis and manual work order dispatch), resulting in fragmented management. This intelligent verification method for power grid data governance integrates data access, verification, visualization analysis, and report generation functions. Maintenance personnel can complete anomaly viewing, work order dispatch, and rectification acceptance on a single interface, improving data quality control efficiency by over 50%. It also supports multi-departmental data sharing and collaboration, promoting the standardization of power grid data governance.

[0056] Example 2 This embodiment provides an intelligent verification system for power grid data governance based on four-dimensional cross-validation. This system applies the intelligent verification method for power grid data governance based on four-dimensional cross-validation described in Embodiment 1. Specifically, the system includes a four-dimensional feature verification module, a hybrid diagnostic model, a rule self-optimization engine, and an intelligent verification management platform. These modules work collaboratively through standardized interfaces.

[0057] The system adopts a “layered deployment and collaborative linkage” architecture, which is divided into the following layers from top to bottom: (1) Application layer: intelligent verification management platform, which provides functions such as data access, anomaly visualization, rectification tracking, and report generation; (2) Core capability layer: which includes three core components: four-dimensional feature verification module, hybrid diagnostic model, and rule self-optimization engine; (3) Data layer: which integrates power grid equipment parameter library, operation database, topology relation library, and historical anomaly library to provide data support for upper-layer modules; (4) Infrastructure layer: which includes computing resources (CPU / GPU server), storage resources (relational database, graph database) and network resources to ensure stable operation of the system.

[0058] The four-dimensional feature verification module is used to perform four-dimensional verification of the dataset to be verified, namely "value-rule-relationship-attribute". This module is the core of data quality verification. Through cross-validation of the four dimensions of "value-rule-relationship-attribute", it achieves full-dimensional anomaly identification and specifically includes the following sub-modules.

[0059] (1) Value-Rule Dimension Validation Submodule Hardware / Software Dependencies: Developed based on the Drools rule engine, deployed on a CPU server, and supports importing rules in JSON format.

[0060] "Value" dimension validation: This involves four basic rules: numerical range, format specifications, non-empty integrity, and uniqueness. For example, rules such as "10kV line voltage deviation must not exceed ±7%", "date format is 'YYYY-MM-DDHH:MM:SS'", and "equipment ID is unique across the entire power grid" are used to match data with these rules via a rule engine, outputting "value" dimension anomalies.

[0061] "Rule" dimension verification: Design three categories of business rules: cross-system consistency, state transition, and data freshness. For example, "GIS and EMS system equipment status is consistent", "Equipment status only allows 'running → maintenance → running' transition", and "Real-time data acquisition time deviation from current time ≤ 5 seconds". The Rete algorithm is used to quickly match rules and identify business compliance anomalies.

[0062] Performance optimization: Adopting a "rule grouping + Redis caching + multi-threaded parallelism" mechanism, rules are loaded in groups according to data type (device parameters, running data), high-frequency rules are cached in Redis, and an 8-core CPU supports 8 threads for parallel verification. The verification time for a single batch of 100,000 data records is ≤3 minutes.

[0063] (2) Topological association verification submodule of "relationship" dimension Hardware / Software Dependencies: The power grid topology model is built based on the Neo4j graph database, and BFS, DFS, and Louvain graph computation algorithm libraries are integrated.

[0064] Topology graph model construction: Define a “node (equipment / site) - edge (connection / ownership relationship)” model. Node attributes include “equipment ID, rated voltage, and operating status”, and edge attributes include “relationship type and voltage level constraint”. It supports batch import of 100,000 nodes + 200,000 edges (import time ≤ 1 hour).

[0065] Topology Anomaly Identification: A "substation equipment association tree" is generated based on the BFS algorithm and compared with the GIS equipment list to identify anomalies such as "missing / incorrect connection relationships"; topology loops are detected based on the DFS algorithm to determine "consistency of equipment voltage levels within the loop" and identify anomalies such as "topology logic errors"; topology communities are divided based on the Louvain algorithm, and the voltage levels of nodes within the community are checked to identify anomalies such as "voltage level association contradictions"; Visualization: The topology map is visualized using ECharts, with anomaly nodes / edges marked with different colors (e.g., orange for missing connections, red for logic errors), and "one-click penetration query" of anomaly details is supported.

[0066] (3) Business logic validation submodule for “attribute” dimension Hardware / software dependencies: Integrated power grid physics law calculation library and machine learning framework (TensorFlow / PyTorch), supporting GPU-accelerated inference.

[0067] Physical law verification: Based on physical formulas such as power balance, transformer voltage ratio, and line impedance, verification functions are designed. For example, "Substation input active power = output active power + station losses (deviation ≤ 5%)" and "Transformer actual voltage ratio deviation from nameplate voltage ratio ≤ 2%", the system automatically extracts data, substitutes it into the formula for calculation, and identifies abnormal parameter logic.

[0068] Machine Learning Validation: Random Forest Model: Trained based on historical normal data (3 months of load / voltage data), extracting time series features (hourly mean, weekly trend) and statistical features (standard deviation) to predict abnormal trends in operating data, with a recognition rate ≥92.3%; Generalized Additive Model (GAM): Constructing a "current-voltage-power factor-power" correlation model, calculating the relative error between actual power and predicted power (error >10% is considered abnormal), with a recognition rate ≥91.5%; Isolated Forest Model: Unsupervised detection of outliers in equipment parameters (e.g., abnormally high transformer rated capacity), with an anomaly score ≥0.7, and a recognition rate ≥90.8%; Result Fusion: Using a weighted voting method (physical law weight 0.6, machine learning weight 0.4), a total score ≥1.0 is considered "confirmed abnormality", and a score 0.5 ≤ score <1.0 is considered "suspected abnormality" (requires manual review, review rate ≤10%).

[0069] The hybrid diagnostic model communicates with the four-dimensional feature verification module and is used to summarize the four-dimensional verification results and generate anomaly diagnostic reports. The hybrid diagnostic model integrates "deterministic diagnosis by the rule engine" and "probabilistic diagnosis by machine learning" to form a complementary diagnostic mechanism. Workflow: First, the rule engine prioritizes handling deterministic anomalies in the "value-rule" dimension (such as format errors and numerical limits), with an accuracy rate ≥99.2%. Then, for implicit anomalies that the rule engine cannot identify (such as topological contradictions and attribute association errors), it triggers verification in the "relationship-attribute" dimension. Finally, the hybrid diagnostic model summarizes the four-dimensional verification results, generates a diagnostic report containing "anomaly ID, type, level, and scope of impact," and pushes it to the intelligent verification management platform.

[0070] The rule self-optimization engine communicates with the hybrid diagnostic model and is used to perform association rule mining, dynamic adjustment of rule weights, and detection and elimination of invalid rules. This engine realizes "automatic mining-dynamic adjustment-invalidation" of rules, solving the problem of lagging rule maintenance in traditional systems, and specifically includes the following sub-modules.

[0071] (1) Apriori Association Rule Mining Submodule This submodule performs the following: (a) Data preprocessing: collecting historical data (≥100,000 records, covering 12 types of equipment and 30 types of anomalies), filling missing values ​​(mean / mode), discretizing continuous features (e.g., current is divided into "<500A" and "500-1000A"), and cleaning noisy data; (b) Frequent itemset mining: generating 1-itemsets (e.g., "current > 2000A") and candidate k-itemsets (e.g., "current > 2000A ∧ voltage < 100kV") using the Apriori algorithm, and filtering frequent itemsets with support ≥ 5%; (c) Rule generation and filtering: generating "antecedent → consequent" rules based on frequent itemsets (e.g., "current > 2000A ∧ voltage < 100kV → power anomaly"), filtering rules with confidence ≥ 80% and lift > 1.2, and including rules in the rule base after manual review ≤ 20%, thus improving rule expansion efficiency by 80%.

[0072] (2) Rule weight dynamic adjustment submodule First, it supports an evaluation indicator system, defining three categories of indicators: accuracy (the proportion of actual anomalies among rule-determined anomalies), coverage (the proportion of actual anomalies identified by rules), and false positive rate (the proportion of actual normal anomalies among rule-determined anomalies). Second, it can provide feedback on data collection: automatically recording "rule determination result - actual rectification result" (e.g., TP true positive, FP false positive), and manually supplementing missed data (FN false negative). Finally, it has weight adjustment logic: indicators are calculated weekly; a 5% increase in accuracy and a 3% decrease in false positive rate → weight increase of 5%-10%; a 5% decrease in accuracy and a 3% increase in false positive rate → weight decrease of 5%-10%; indicator fluctuation ≤2% → weight unchanged; it supports temporary weight adjustments for special scenarios (e.g., new energy grid connection), automatically restoring weights after the scenario ends.

[0073] (3) Rule Failure Detection Submodule If a rule meets any of the following conditions for three consecutive weeks: "accuracy ≤ 60%", "coverage ≤ 30%", or "false positive rate ≥ 20%", it is determined to be an "invalid rule". Business experts are notified to review it. After confirmation of invalidity, it is marked as "disabled" and removed. At the same time, the Apriori algorithm is triggered to generate alternative rules to ensure the "health" of the rule base. The overall diagnostic accuracy of the rule base is improved by ≥ 12.3%.

[0074] The intelligent verification management platform communicates with the four-dimensional feature verification module, hybrid diagnostic model, and rule self-optimization engine, providing functions such as data access, anomaly visualization, anomaly rectification work order dispatch, and report generation. The intelligent verification management platform adopts a cloud-native architecture and containerized deployment (Docker / Kubernetes), supporting public cloud / private cloud / hybrid cloud deployments. If traditional physical machine deployment is used instead of cloud-native containerized deployment, or if multi-system data access is not supported and verification is only performed on a single business system (such as only the GIS system), single-system verification cannot achieve cross-system data consistency verification (such as comparing the status of GIS and EMS equipment), failing to meet the needs of large-scale power grid data governance.

[0075] The intelligent verification management platform supports: (a) Data access: Supports data access from systems such as GIS, EMS, and marketing, and is compatible with CSV, JSON, and direct database connection methods; (b) Anomaly management: Provides functions such as viewing anomaly lists, detail penetration, dispatching rectification work orders, and acceptance of rectification results; (c) Visual analysis: Displays the distribution of anomaly types (such as "connection relationship anomalies account for 40%"), equipment data quality scores (by "authenticity-accuracy-consistency-completeness"), and ranking of rule effects; (d) Report generation: Automatically generates daily / weekly / monthly reports, including indicators such as data verification volume, anomaly identification rate, and rectification completion rate, and supports Excel / PDF export.

[0076] It should be noted that if the intelligent verification management platform is split into independent systems (such as "data access system", "verification system" and "abnormal work order system"), and each system connects through APIs, claiming "distributed architecture advantages", it will lead to data delays (such as abnormal data needing to be manually imported into the work order system), data inconsistencies (such as data updates not being synchronized between systems), and a decrease in management efficiency of more than 40%.

[0077] Example 3 This embodiment provides a computer terminal, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps of the intelligent verification method for power grid data governance based on four-dimensional cross-validation described in Embodiment 1.

[0078] The method in Example 1 can be applied in software form, such as by designing it as a standalone program and installing it on a computer terminal, which can be a computer, smartphone, control system, or other IoT device. Alternatively, the method in Example 1 can be designed as an embedded program and installed on a computer terminal, such as on a microcontroller.

[0079] Example 4 This embodiment provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, it implements the steps of the intelligent verification method for power grid data governance based on four-dimensional cross-validation in Embodiment 1.

[0080] When applying the method of Example 1, it can be applied in the form of software, such as by designing it as a program that can run independently on a computer-readable storage medium. The computer-readable storage medium can be a USB flash drive, designed as a USB security token, and the program can be designed to start the entire method through an external trigger.

[0081] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A smart verification method for power grid data governance based on four-dimensional cross-validation, characterized in that, It includes the following steps: Data is collected from various power grid business systems and preprocessed to generate a dataset to be verified. The following steps are performed sequentially on the dataset to be verified: Validating the data in the "value" dimension based on the designed multi-class basic rules; Based on the design of multiple business rules, the data is validated in the "rule" dimension; the data is imported into the constructed topology graph model to detect node / edge attribute anomalies for validation in the "relationship" dimension; physical law validation and machine learning validation are performed, and the parameter logic anomalies and data reasoning anomalies obtained respectively are weighted and fused to obtain the "attribute" dimension validation results; Summarize the verification results across four dimensions: "value-rule-relationship-attribute" and classify the anomaly levels to generate an anomaly diagnosis report; Based on the feedback data of the abnormal diagnosis results and actual rectification results in the report, we mine association rules to expand the rule base, dynamically adjust the weight of rules in the rule base, and detect and eliminate invalid rules.

2. The intelligent verification method for power grid data governance based on four-dimensional cross-validation as described in claim 1, characterized in that, The intelligent verification method for power grid data governance also includes the following steps: An anomaly diagnosis report is pushed to operations and maintenance personnel / systems via an intelligent verification and management platform. After viewing the anomaly details, the operations and maintenance personnel / systems dispatch a rectification work order to the corresponding business responsibility department through the intelligent verification and management platform. After the business responsibility department completes the rectification and uploads the rectification results to the intelligent verification and management platform, the operations and maintenance personnel / systems verify the rectification results on the intelligent verification and management platform. If the verification is successful, the corresponding anomaly is marked as closed-loop management; otherwise, the work order is returned and rectification is required again. The intelligent verification and management platform records the rectification progress and results and updates the data quality score.

3. The intelligent verification method for power grid data governance based on four-dimensional cross-validation as described in claim 2, characterized in that, The intelligent verification method for power grid data governance also includes the following steps: The intelligent verification management platform monitors core indicators including data verification volume, anomaly identification rate and rectification completion rate in real time and displays the core indicators in a visual interface; the intelligent verification management platform periodically and automatically generates data quality verification reports; the data quality verification reports include: (1) a data verification overview including the number of data verification entries, the number of passed entries, and the pass rate of each system; (2) anomaly analysis results including the distribution of anomaly types, the distribution of severity levels, and the distribution of equipment types involved; (3) the rule effectiveness including the ranking of the accuracy rate, coverage rate, and false alarm rate of each rule; and (4) the anomaly rectification situation including the number of rectification work orders dispatched, the number of completed orders, the completion rate, and the average rectification time.

4. The intelligent verification method for power grid data governance based on four-dimensional cross-validation as described in claim 1, characterized in that, There are four categories of basic rules, namely, numerical range, format specification, non-empty integrity, and uniqueness; there are three categories of business rules, namely, cross-system consistency, state transition, and data freshness; the "value" dimension verification and the "rule" dimension verification are executed through the rule engine, and the performance is optimized by rule grouping, caching, and multi-threaded parallel mechanism.

5. The intelligent verification method for power grid data governance based on four-dimensional cross-validation as described in claim 1, characterized in that, The "relationship" dimension validation specifically includes: (1) Construct a power grid topology model based on a graph database. The topology model includes nodes that represent power grid equipment or stations and edges that represent connection or affiliation relationships. (2) Detect missing or incorrect connection relationships in the topology model based on the breadth-first search algorithm; detect the logical consistency of topology loops and devices within loops in the topology model based on the depth-first search algorithm; divide the topology model into communities based on the community detection algorithm and detect the voltage level correlation contradictions of nodes within the community.

6. The intelligent verification method for power grid data governance based on four-dimensional cross-validation as described in claim 1, characterized in that, The physical law verification specifically involves constructing a verification function based on the physical laws of the power grid, calculating parameter deviations of power balance, transformer voltage ratio, or line impedance, and identifying parameter logic anomalies based on deviation thresholds. The machine learning verification specifically involves inferring from operational data based on a pre-trained machine learning model to identify trend anomalies, correlation anomalies, or outlier anomalies. The verification result of the "attribute" dimension is a confirmed anomaly or a suspected anomaly.

7. The intelligent verification method for power grid data governance based on four-dimensional cross-validation as described in claim 1, characterized in that, The association mining rules specifically include: The historical data and the feedback data are preprocessed, including discretizing continuous features; The Apriori algorithm is used to mine frequent itemsets with support at or above a preset threshold. Candidate rules are generated based on the frequent itemsets, and candidate rules with confidence scores not lower than a first threshold and lift scores greater than a second threshold are selected and included in the rule base.

8. The intelligent verification method for power grid data governance based on four-dimensional cross-validation as described in claim 1, characterized in that, The specific weights of rules in the dynamically adjusted rule base include: Periodically calculate the accuracy, coverage, and false alarm rate of each rule in the rule base; In response to the increase in accuracy and the decrease in false positive rate, the weight of the corresponding rule is increased; In response to the decrease in accuracy and the increase in false positive rate, the weight of the corresponding rule is reduced.

9. The intelligent verification method for power grid data governance based on four-dimensional cross-validation as described in claim 1, characterized in that, The business system is at least one of a GIS system, an EMS system, and a marketing system. The collected data includes equipment parameters, operating data, and topology data. The collected data is then cleaned of duplicate data, missing values ​​are filled, and the format is converted to obtain the dataset to be verified.

10. The intelligent verification system for power grid data governance based on four-dimensional cross-validation as described in claim 1, used to implement the intelligent verification method for power grid data governance based on four-dimensional cross-validation as described in any one of claims 1-9, characterized in that, The intelligent verification system for power grid data governance includes: The four-dimensional feature verification module performs a four-dimensional verification of the dataset to be verified, which includes "value-rule-relationship-attribute". A hybrid diagnostic model is communicatively connected to the four-dimensional feature verification module and is used to summarize the four-dimensional verification results and generate an anomaly diagnostic report. A rule self-optimization engine, which is communicatively connected to the hybrid diagnostic model, is used to perform the association rule mining, the dynamic adjustment of the rule weights, and the detection and elimination of the failed rules; The intelligent verification management platform communicates with the four-dimensional feature verification module, the hybrid diagnostic model, and the rule self-optimization engine, and is used to provide data access, anomaly visualization, anomaly rectification work order dispatch, and report generation functions.

Images