Insurance industry supervision submitted data verification method and system

Abstract

The invention discloses an insurance industry supervision submitted data verification method and system, and mainly relates to the technical field of intelligent insurance industry supervision. Comprising the following steps: acquiring multi-source submitted data of the insurance industry, and preprocessing the multi-source submitted data to obtain structured data; constructing a supervision rule knowledge graph; using a multi-agent system to perform collaborative verification on the structured data, wherein the multi-agent system comprises a rule verification agent, an anomaly detection agent, a trend comparison agent, a correction suggestion agent and a feedback learning agent; a verification task of the multi-agent system is executed through a distributed computing framework, and a verification result is generated; and based on the verification result, generating a verification report and classifying the verification report, and updating the supervision rule knowledge graph through the feedback learning agent. The method has the beneficial effects that efficient and accurate verification and intelligent self-adaptive closed-loop management of insurance submission data are realized.

Description

Technical Field

[0001] This invention relates to the field of intelligent insurance industry supervision technology, specifically a method and system for verifying insurance industry regulatory reporting data based on big data intelligent agents. Background Technology

[0002] With increasingly stringent regulatory requirements and rapid growth in data volume within the insurance industry, insurance institutions are required to regularly submit massive amounts of business data to regulatory agencies. Currently, the industry commonly employs data verification methods based on predefined rules, typically using SQL scripts or fixed procedures to perform static validation on the submitted data. These methods can, to some extent, identify data issues such as formatting errors or obvious rule violations. However, existing technologies suffer from the following prominent problems: 1. Rigid rule management and poor adaptability: The verification rules are implemented in a hard-coded manner, which makes it difficult to update quickly with frequent changes in regulatory standards, resulting in a lag in rule base maintenance; 2. Low verification efficiency and insufficient processing capacity: Faced with hundreds of millions of policy data, traditional stand-alone scripts or manual verification methods are slow and cannot meet the timeliness requirements for reporting. 3. Limited intelligence level and weak anomaly detection capability: It mainly relies on preset rules for surface consistency verification and lacks the ability to proactively detect potential anomaly patterns (such as hidden risks and trend drift).

[0003] The reasons for the above problems are as follows: Technical architecture limitations: The system lacks a dynamically scalable rule representation and storage mechanism (such as a knowledge graph), causing rule changes to rely on manual code modification; Data processing capacity bottleneck: The distributed computing framework is not fully utilized to process large-scale data, and the execution efficiency of verification tasks is limited by single-point resources; Lack of closed-loop learning mechanism: The verification system has failed to form an effective closed-loop learning mechanism with regulatory feedback and historical data, making it impossible to achieve self-optimization of rules and continuous evolution of anomaly detection models.

[0004] Therefore, there is an urgent need for a data verification method and system for insurance industry regulatory reporting based on big data intelligent agents to solve the above problems. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for verifying insurance industry regulatory reporting data based on big data intelligent agents, which realizes efficient and accurate verification of insurance reporting data and intelligent adaptive closed-loop management.

[0006] To achieve the above objectives, the present invention employs the following technical solution: On the one hand, this invention provides a method for verifying data submitted by insurance industry regulators, comprising the following steps: Step S1: Obtain multi-source reporting data from the insurance industry and preprocess the multi-source reporting data to obtain structured data; Step S2: Construct a regulatory rule knowledge graph, which includes multiple nodes and edges. The nodes represent regulatory rules, data items, and anomaly types, and the edges represent the relationships between nodes. Step S3: Collaboratively verify the structured data using a multi-agent system, wherein the multi-agent system includes a rule verification agent, an anomaly detection agent, a trend comparison agent, a correction suggestion agent, and a feedback learning agent; Step S4: Execute the verification task of the multi-agent system through a distributed computing framework and generate verification results; Step S5: Based on the verification results, generate a verification report and classify it, and at the same time update the regulatory rule knowledge graph through the feedback learning agent.

[0007] Preferably, in step S1, the preprocessing of the multi-source reported data includes: Data is cleaned using ETL tools and real-time stream processing platforms, and a unified data standard model is established based on regulatory reporting standard templates. At the same time, a data lineage tracing mechanism is used to record the source and change process of fields. The real-time stream processing platform uses Kafka or Flink technology, the data standard model is a structured data table, and the data lineage tracing mechanism achieves field-level traceability through metadata management.

[0008] Preferably, in step S2, constructing the regulatory rule knowledge graph includes: The node types are defined as regulatory caliber nodes, data item nodes, rule nodes, and anomaly type nodes, and the relationships between nodes are represented by triples, where each triple includes a subject, a relation, and an object. The regulatory rule knowledge graph supports dynamic updates and version management, and performs logical reasoning through a rule reasoning engine. The specific form of the triple is as follows: .

[0009] Preferably, in step S3, the agents in the multi-agent system communicate and work collaboratively through a message bus; The rule verification agent is used to automatically match rules based on the knowledge graph and generate verification expressions. The verification expressions are converted into SQL or logical expressions for field consistency verification. The anomaly detection agent is used to identify potential anomalies using a machine learning model, which includes the Isolation Forest algorithm and the Autoencoder algorithm. The anomaly score is calculated using the following formula: For the Isolation Forest algorithm, anomaly scoring Defined as: ; in, For data samples, The path length of the sample in the isolated tree. For the expected path length, This is a correction term for the average path length of the tree. The number of samples; For autoencoder algorithms, reconstruction error Defined as: ; in, For input data, To reconstruct the output, when If the threshold is exceeded, it is considered abnormal; The trend comparison agent is used to detect abnormal fluctuations by comparing historical reported data and to calculate volatility using time series analysis methods. The correction suggestion agent is used to generate correction suggestions based on rule-based reasoning and historical correction records; The feedback learning agent is used to summarize regulatory verification feedback and adjust rule weights and model parameters.

[0010] Preferably, in step S4, performing the verification task through the distributed computing framework includes: translating the rules into a distributed query task, calling the distributed model inference engine to perform batch anomaly detection, and comparing the verification results of the previous period to perform trend analysis. The distributed query task is executed using Spark SQL or Presto, the distributed model inference engine uses TensorFlow Serving, and the trend analysis calculates the percentage deviation of data changes using a sliding window.

[0011] On the other hand, the present invention also provides an insurance industry regulatory reporting data verification system for implementing the above-described insurance industry regulatory reporting data verification method, including: The data access and preprocessing module is used to acquire multi-source reporting data from the insurance industry and preprocess the multi-source reporting data to form structured data that meets regulatory reporting standards. The rule knowledge graph construction module is used to construct a regulatory rule knowledge graph, which includes multiple nodes and edges. The nodes represent regulatory rules, data items, and anomaly types, and the edges represent the relationships between nodes. A multi-agent verification module is used to collaboratively verify the structured data using a multi-agent system, wherein the multi-agent system includes a rule verification agent, an anomaly detection agent, a trend comparison agent, a correction suggestion agent, and a feedback learning agent. The distributed verification execution module is used to execute the verification task of the multi-agent system through a distributed computing framework and generate verification results. The results summary and feedback module is used to generate and classify verification reports based on the verification results, and at the same time update the regulatory rule knowledge graph through the feedback learning agent to form a closed-loop feedback mechanism.

[0012] Preferably, the data access and preprocessing module is specifically used to: perform data cleaning through ETL tools and a real-time stream processing platform, establish a unified data standard model based on regulatory reporting standard templates, and use a data lineage tracing mechanism to record the source and change process of fields.

[0013] Preferably, the rule knowledge graph construction module is specifically used to: define node types as regulatory caliber nodes, data item nodes, rule nodes, and anomaly type nodes, and represent the relationships between nodes in the form of triples, wherein the triples include subject, relation, and object.

[0014] Preferably, in the multi-agent verification module, the agents in the multi-agent system communicate and work collaboratively through a message bus; The rule verification agent is used to automatically match rules based on the knowledge graph and generate verification expressions. The verification expressions are converted into SQL or logical expressions for field consistency verification. The anomaly detection agent is used to identify potential anomalies using machine learning models, including the isolated forest algorithm and the autoencoder algorithm. The trend comparison agent is used to detect abnormal fluctuations by comparing historical reported data and to calculate volatility using time series analysis methods. The correction suggestion agent is used to generate correction suggestions based on rule-based reasoning and historical correction records; The feedback learning agent is used to summarize regulatory verification feedback and adjust rule weights and model parameters.

[0015] Preferably, the distributed verification execution module is specifically used to: translate rules into distributed query tasks, call the distributed model inference engine to perform batch anomaly detection, and compare the verification results of the previous period to perform trend analysis.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. It solves the problem of rigid rule management, and realizes the flexibility and evolvability of rules: By constructing a regulatory rule knowledge graph, previously fixed SQL scripts or hard-coded rules are transformed into a structured, reasonable knowledge network. This enables rules to be visualized, managed, and dynamically updated in the form of triples, allowing for rapid response to changes in regulatory interpretations. Simultaneously, a closed-loop feedback mechanism is introduced, forming a feedback learning agent. This allows the system to automatically optimize rule weights and model parameters based on verification results and regulatory feedback, achieving "self-learning" and "self-evolution" of the rule system and fundamentally overcoming the shortcomings of existing technologies in terms of lagging rule updates.

[0017] 2. Significantly improved the efficiency of verification and processing of massive amounts of data and the system's throughput capacity: This invention employs a technical architecture combining multi-agent collaboration and a distributed computing framework. On one hand, multiple agents work in parallel through a message bus, achieving dynamic division of labor and load balancing for verification tasks. On the other hand, by translating verification tasks into distributed query tasks such as Spark SQL and Presto, and utilizing TensorFlow Serving for distributed model inference, the computational advantages of the big data platform are fully leveraged. This enables efficient and rapid verification of hundreds of millions of insurance policy data, effectively solving the bottleneck of insufficient processing power of traditional single-machine scripts.

[0018] 3. Enhanced the intelligence of the verification process, achieving a leap from "rule verification" to "intelligent insight": This invention does not simply apply rules, but rather constructs a comprehensive intelligent verification system integrating rule verification, anomaly detection, trend comparison, and correction suggestions. In particular, by employing machine learning algorithms such as Isolation Forest and Autoencoders through an anomaly detection agent, the system can proactively discover potential, hidden anomaly patterns and risks in the data, rather than just surface errors. The trend comparison agent effectively identifies abnormal fluctuations in the data through time series analysis. This transforms the verification process from passively meeting compliance requirements to proactively monitoring data quality and providing risk warnings.

[0019] 4. A complete, traceable verification and audit loop has been established, improving management transparency and reliability: From data lineage tracing during data access to traceable logs of the verification process, and finally to visual reports and API pushes, this invention achieves end-to-end monitoring from the data source to the verification result. This not only facilitates internal auditing and problem localization but also provides a clear and complete chain of evidence for regulatory inspections, effectively solving the pain point of "inefficient feedback links" in traditional methods. Attached Figure Description

[0020] Figure 1 This is a flowchart of the method of the present invention; Figure 2This is a schematic diagram of the system structure of the present invention. Detailed Implementation

[0021] The present invention will be further illustrated below with reference to specific embodiments. It should be understood that these embodiments are for illustrative purposes only and are not intended to limit the scope of the invention. Furthermore, it should be understood that after reading the teachings of this invention, those skilled in the art can make various alterations or modifications to the invention, and these equivalent forms also fall within the scope defined in this application.

[0022] In this invention, terms such as "upper," "lower," "left," "right," "front," "back," "vertical," "horizontal," "side," and "bottom" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. These terms are used only to facilitate the description of the structural relationships of the various components or elements of this invention and do not specifically refer to any component or element in this invention. They should not be construed as limiting the invention.

[0023] Example: like Figure 1 As shown in the figure, this embodiment provides a method for verifying data submitted by the insurance industry regulators, including the following steps: Step S1: Obtain multi-source reporting data from the insurance industry and preprocess the multi-source reporting data to obtain structured data; Step S2: Construct a regulatory rule knowledge graph. The regulatory rule knowledge graph includes multiple nodes and edges. Nodes represent regulatory rules, data items, and anomaly types, and the edges represent the relationships between nodes. Step S3: Use a multi-agent system to collaboratively verify structured data, where the multi-agent system includes a rule verification agent, an anomaly detection agent, a trend comparison agent, a correction suggestion agent, and a feedback learning agent; Step S4: Execute the verification task of the multi-agent system through a distributed computing framework and generate the verification results; Step S5: Based on the verification results, generate and classify the verification report, and update the regulatory rule knowledge graph through feedback learning agent.

[0024] like Figure 2 As shown, this embodiment also provides a data verification system for insurance industry regulatory reporting, including: The data access and preprocessing module is used to acquire multi-source reporting data from the insurance industry and preprocess the multi-source reporting data to form structured data that meets regulatory reporting standards. The rule knowledge graph construction module is used to build a regulatory rule knowledge graph. The regulatory rule knowledge graph includes multiple nodes and edges. Nodes represent regulatory rules, data items and anomaly types, and edges represent the relationships between nodes. The multi-agent verification module is used to collaboratively verify structured data using a multi-agent system, which includes a rule verification agent, an anomaly detection agent, a trend comparison agent, a correction suggestion agent, and a feedback learning agent. The distributed verification execution module is used to execute verification tasks of a multi-agent system through a distributed computing framework and generate verification results. The results summary and feedback module is used to generate and classify verification reports based on the verification results, and at the same time update the regulatory rule knowledge graph through feedback learning agents to form a closed-loop feedback mechanism.

[0025] The following embodiment uses Insurance Company A as an example to illustrate the application of the above method and system in detail.

[0026] 1. Data access and preprocessing: Implementation scenario: On the 1st of each month, the reporting cycle begins, and Insurance Company A's policy system, claims system, financial system, and intermediary management system start generating new business data; The specific implementation of the above-mentioned scenarios is as follows: (1) Multi-source data access: The data access and preprocessing module is started, and data is acquired in two ways: Batch access: For data updated on T+1 (such as yesterday's full insurance policies), data is extracted from the databases of various source systems using ETL tools (such as Apache Sqoop / DataX); Real-time access: For critical data that requires real-time verification (such as large claims), the data change stream sent by the business system is received through a real-time stream processing platform (using Apache Kafka); (2) Data cleaning and standardization: After the raw data enters the platform, it first goes through a stream and batch processing engine implemented by Apache Flink: Perform data cleaning rules: filter out null values, correct obvious errors (such as marking future birth dates as abnormal), and standardize data formats (such as standardizing dates to YYYY-MM-DD format). According to the latest version of the "Insurance Business Regulatory Data Standard" issued by the regulatory authorities, heterogeneous data from different systems will be mapped and converted into a unified data standard model (Data Schema). For example, the "premium" field will be uniformly mapped from PREMIUM_AMT in the financial system and Amount in the policy system to premium_amount in the standard model. (3) Data lineage recording: Throughout the preprocessing process, the data lineage tracking mechanism is activated, and the system automatically records the source system, source table, source field, and transformation logic of each target field, for example: The T_ins_report_raw.premium_amount field is recorded from the financial system .GL_Table.PREMIUM_AMT and has been converted by "dividing by 100 (yuan to cent)". This metadata is stored in the graph database Neo4j for later traceability. (4) After preprocessing, a clean and standard structured data table T_ins_report_raw is generated and stored in a distributed data warehouse (such as Apache Hive) to prepare for subsequent verification.

[0027] 2. Construction of rule knowledge graph: Implementation scenario: During the system initialization phase, compliance experts and IT personnel from Insurance Company A collaborated to input regulatory provisions and internal control rules into the system. One day, the regulator issued a new regulation requiring that "for short-term health insurance policies, the cancellation fee shall not exceed 20% of the policy fee." The specific implementation of the above-mentioned scenarios is as follows: (1) In the rule knowledge graph construction module, construct a regulatory rule knowledge graph named KG_reg: Create a pipe diameter node: upper limit for short-term health insurance cancellation fees; Create data item nodes: policy_type (policy type), surrender_charge (surrender fee), total_premium (total premium); Create a rule node: R_new_001; Creating an exception type node: Fee exceeding the limit; Establish relationships (triples): (policy_type, is, short-term health insurance) → This is a condition; (surrender_charge, <=, total_premium*0.2) → This is the core constraint; (R_new_001, hasCondition, and policy_type are short-term health insurance); (R_new_001, hasConstraint, surrender_charge<= total_premium * 0.2); (Violation of R_new_001, leadsTo, fee overstepping limits); (2) Compliance personnel can add the new rule R_new_001 by dragging and configuring the above nodes and relationships through the visual interface. There is no need to write and deploy SQL scripts. The rule reasoning engine built into the knowledge graph can understand these logical relationships. When verifying, the engine will automatically identify all records with "policy type is short-term health insurance" and apply the new handling fee constraint rules to them. (3) The system performs version management of the KG_reg knowledge graph. A new version is generated each time a rule is changed, which facilitates auditing and rollback.

[0028] 3. Multi-agent verification system and distributed execution: Implementation scenario: After the standardized data T_ins_report_raw and the dynamically updated rule graph KG_reg are ready, the system starts the multi-agent verification process; The specific implementation of the above-mentioned scenarios is as follows: (1) Each agent in the multi-agent verification module receives task instructions through a message queue (RabbitMQ) and starts working in parallel. They share a distributed cache (such as Redis) to store intermediate states and results. (2) Rule-verification agent workflow: Task allocation: Rule-Agent retrieves all active rules from the knowledge graph and, based on the partitioning of the data table, breaks down the verification task into multiple sub-tasks; Distributed execution: It translates rule R_new_001 into a Spark SQL statement: “SELECT...FROM T_ins_report_raw WHERE policy_type = 'Short-term health insurance' AND surrender_charge>total_premium * 0.2”; This SQL task was submitted to the Apache Spark cluster for distributed execution, where hundreds of millions of data points were scanned and computed in parallel to quickly identify all policy records that violated this rule. The verification results were marked as "non-compliant". (3) Anomaly Detection Agent Workflow: Model inference: Rule-Agent primarily handles "whitelist" rules, while Anomaly-Agent is responsible for discovering "unknown anomalies." It invokes a pre-trained Isolation Forest model; Algorithm Application: This model runs in a distributed mode on TensorFlow Serving, performing joint analysis on multiple numerical features such as "policy premium," "sum insured," and "insured's age." For each data point The model calculates its anomaly score. ; Insurance company A's sales staff set a threshold of 0.65. When a certain policy... When a policy is identified as abnormal, for example, if the system finds a batch of policies from the same agent with an abnormally high "sum insured / premium" ratio, even though the documents for each policy comply with the rules, the overall pattern is abnormal and may indicate a risk of fraud. Such records are marked as "abnormal - pending review". (4) Trend-Agent Workflow: Historical data comparison: Trend-Agent extracts "premium income" data for the same period and region over the past 12 months from the historical verification result database; Volatility Calculation: Using time series analysis, the month-on-month and year-on-year change rates of the current monthly premium were calculated. It was found that "the premium of a certain branch in East China decreased by 40% month-on-month," a volatility far exceeding the historical normal range. Trend-Agent will issue a "trend anomaly" alert, indicating that there may be a data reporting omission or a sudden business situation. (5) Results summary: The verification results of all agents (rule non-compliance, AI abnormality, trend abnormality) are collected in a unified manner and written into the T_ins_report_check_result result table in JSON format.

[0029] 4. Results summary and closed-loop feedback: Implementation scenario: After the verification is completed, Xiao Wang, the data governance officer in the business department, logs into the system to view the verification report and handle any abnormal data; The specific implementation of the above-mentioned scenarios is as follows: (1) Result Classification and Report Generation: The Result Summary and Feedback module classifies the results: Compliant data: 95% is permitted to be submitted; Minor Abnormal Data: For example, if a branch office's premium decline is detected through trend comparison, the system will generate a prompt and suggest that the business department verify the information. Non-compliant data (serious): such as policies that violate the new regulation R_new_001, and suspected fraudulent policies discovered by AI; (2) Correction suggestion: The Repair-Agent should intervene. For the violation record R_new_001, Repair-Agent automatically generated a suggestion based on the knowledge graph and historical correction records: "According to the regulatory rule 'maximum fee for surrendering short-term health insurance policies,' the surrender fee of XXX yuan for this policy has exceeded 20% (YYY yuan) of the total premium. It is recommended to adjust it to no more than ZZZ yuan." For errors involving missing fields, it might suggest "completing the missing occupation type field from the customer information table based on client_id"; (3) Feedback-Agent achieves closed-loop: Employees of Insurance Company A distributed the correction suggestions generated by the system to the relevant branches for verification and correction. The corrected data was then re-entered into the verification process. A week later, the regulatory agency provided feedback on the results of the submitted spot check, and Feedback-Agent entered this feedback information (such as "a certain type of rule was emphasized") into the system; Learning and Optimization: Rule weight adjustment: Feedback-Agent has increased the weight of rules that are emphasized by regulators in the knowledge graph to ensure that they are executed first in the next verification. Model parameter tuning: For the AI ​​model, it uses the confirmed "true anomalies" and "false anomalies" samples to start incremental training of the model, fine-tuning the parameters to make the next anomaly detection more accurate; Thus, the complete closed loop of "rule knowledge graph ← feedback optimization ← agent learning ← verification results ← data reporting" is realized.

[0030] 5. Intelligent visualization and audit trail: Implementation scenario: Company A's management and compliance department need to have a clear understanding of the overall quality of the submitted data, and at the same time prepare for regulatory inspections; The specific implementation of the above-mentioned scenarios is as follows: The system provides a visual dashboard that displays key metrics such as the monthly verification pass rate, pie charts showing the distribution of various anomalies, and anomaly trend curves. For any record marked as abnormal, auditors can drill down through the interface to view its complete data lineage (which system it came from, when it was accessed, and what transformation it underwent), the specific rules that triggered it (the content and version of the rules), the verification execution logs (which agent executed it on which computing node), and the subsequent correction process. This forms an immutable and complete audit trail chain. Ultimately, the system automatically pushes compliant data packets and verification report summaries to the regulatory agency's reporting and receiving system via the configured API interface.

[0031] The above is a detailed description of the preferred embodiments of the present invention. However, the present invention is not limited to the embodiments described. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the present invention. All such equivalent modifications or substitutions are included within the scope defined by the claims of this application.

Claims

1. An insurance industry regulatory submission data verification method, characterized in that, The method comprises the following steps: Step S1: obtaining multi-source reporting data of the insurance industry and preprocessing the multi-source reporting data to obtain structured data; Step S2: constructing a supervision rule knowledge graph, the supervision rule knowledge graph comprising a plurality of nodes and edges, the nodes representing supervision rules, data items and abnormal types, and the edges representing relationships between the nodes; Step S3: using a multi-agent system to cooperatively verify the structured data, wherein the multi-agent system comprises a rule verification agent, an anomaly detection agent, a trend comparison agent, a correction suggestion agent and a feedback learning agent; Step S4: executing the verification tasks of the multi-agent system through a distributed computing framework to generate verification results; Step S5: based on the verification results, generating a verification report and classifying it, and updating the supervision rule knowledge graph through the feedback learning agent.

2. The method for verifying the submission data of insurance industry supervision according to claim 1, characterized in that, In step S1, preprocessing the multi-source reporting data comprises: performing data cleaning through an ETL tool and a real-time stream processing platform, and establishing a unified data standard model based on a supervision reporting standard template, while using a data bloodline tracking mechanism to record field sources and change processes; wherein the real-time stream processing platform uses Kafka or Flink technology, the data standard model is a structured data table, and the data bloodline tracking mechanism realizes field-level tracing through metadata management.

3. The method of claim 1, wherein the method further comprises: In step S2, constructing the supervision rule knowledge graph comprises: defining node types as supervision caliber nodes, data item nodes, rule nodes and abnormal type nodes, and representing relationships between the nodes in the form of triples, wherein the triples comprise subjects, relationships and objects; the supervision rule knowledge graph supports dynamic updating and version management, and performs logical reasoning through a rule reasoning engine; The specific form of the triplets is .

4. The method of claim 1, wherein, In step S3, the agents in the multi-agent system communicate and work cooperatively through a message bus; wherein the rule verification agent is used to automatically match rules based on the knowledge graph and generate verification expressions, which are converted into SQL or logical expressions for field consistency verification; the anomaly detection agent is used to identify potential anomalies using a machine learning model, which comprises an isolation forest algorithm and a self-encoder algorithm, and the anomaly score calculation formula is as follows: For Isolation Forest algorithm, anomaly score is defined as: ； wherein, is the data sample, is the path length of the sample in the isolated tree, is the expectation of the path length, is the average path length correction term for the tree, is the number of samples; For the autoencoder algorithm, the reconstruction error is defined as: ； wherein, is the input data, is the reconstructed output, when is determined to be abnormal when it exceeds a threshold value; the trend comparison agent is used to compare historical reporting data to detect abnormal fluctuations, and uses a time series analysis method to calculate volatility; the correction suggestion agent is used to generate correction suggestions based on rule reasoning and historical correction records; the feedback learning agent is used to aggregate supervision verification feedback and adjust rule weights and model parameters.

5. The method of claim 1, wherein, In step S4, executing the verification tasks through a distributed computing framework comprises: translating rules into distributed query tasks, calling a distributed model reasoning engine for batch anomaly detection, and comparing previous verification results for trend analysis; The distributed query task is executed by Spark SQL or Presto, the distributed model inference engine adopts TensorFlow Serving, and the trend analysis is performed by calculating the percentage deviation of data changes through a sliding window.

6. An insurance industry regulatory reporting data verification system for implementing an insurance industry regulatory reporting data verification method as claimed in any one of claims 1 to 5, characterized in that, The method comprises the following steps: a data access and preprocessing module configured to obtain multi-source reporting data of the insurance industry and preprocess the multi-source reporting data to form structured data conforming to a regulatory reporting standard; a rule knowledge graph construction module configured to construct a regulatory rule knowledge graph, the regulatory rule knowledge graph comprising a plurality of nodes and edges, the nodes representing regulatory rules, data items and abnormal types, and the edges representing relationships between the nodes; a multi-agent verification module configured to perform collaborative verification on the structured data by using a multi-agent system, wherein the multi-agent system comprises a rule verification agent, an anomaly detection agent, a trend comparison agent, a correction suggestion agent and a feedback learning agent; a distributed verification execution module configured to execute verification tasks of the multi-agent system through a distributed computing framework to generate verification results; a result summarization and feedback module configured to generate a verification report based on the verification results and classify the verification report, and update the regulatory rule knowledge graph through the feedback learning agent to form a closed-loop feedback mechanism.

7. The insurance industry regulatory submission data verification system of claim 6, wherein, The data access and preprocessing module is specifically configured to perform data cleaning through an ETL tool and a real-time stream processing platform, establish a unified data standard model based on a regulatory reporting standard template, and record field sources and change processes using a data bloodline tracking mechanism.

8. The insurance industry regulatory submission data verification system of claim 6, wherein, The rule knowledge graph construction module is specifically configured to define node types as regulatory scope nodes, data item nodes, rule nodes and abnormal type nodes, and represent relationships between the nodes in the form of triples, wherein the triples comprise a subject, a relationship and an object.

9. The insurance industry regulatory submission data verification system of claim 6, wherein, In the multi-agent verification module, the agents in the multi-agent system communicate and work collaboratively through a message bus. The rule verification agent is configured to automatically match rules based on the knowledge graph and generate a verification expression, and the verification expression is converted into a SQL or logical expression for field consistency verification. The anomaly detection agent is configured to identify potential anomalies using a machine learning model, and the machine learning model comprises an isolation forest algorithm and a self-encoder algorithm. The trend comparison agent is configured to compare historical reporting data to detect abnormal fluctuations, and uses a time series analysis method to calculate a volatility rate. The correction suggestion agent is configured to generate correction suggestions based on rule reasoning and historical correction records. The feedback learning agent is configured to summarize regulatory verification feedback and adjust rule weights and model parameters.

10. The insurance industry regulatory submission data verification system of claim 6, wherein, The distributed verification execution module is specifically configured to translate rules into distributed query tasks, call a distributed model inference engine to perform batch anomaly detection, and compare previous verification results to perform trend analysis.

Images