A time difference risk hedging method and device for insurance fund custody

Abstract

The application discloses a kind of time difference risk hedging method and device of insurance fund custody, it is related to financial risk control technical field, the method includes: creating virtual sub-account for each user and being bound with corresponding policy account one by one to realize fund physical isolation;When consumption incentive enters the custody special account, it is frozen to corresponding virtual sub-account by calling bank freezing interface and generates directional payment instruction;When daily business ends, generate custody fund flow sheet snapshot, before batch deduction of insurance company next day, pre-reconciliation is carried out to custody fund flow sheet snapshot and to-be-deducted list, and according to result, unfreezing transfer or difference processing is executed;Risk reserve pool is established simultaneously, when unilateral account is abnormal, first pay, complete account adjustment to realize error recovery after finding out reason and clearing up.The application realizes fund physical isolation by virtual sub-account binding and fund freezing directional transfer, effectively hedges the supervision and operation risk caused by fund circulation time difference, and improves the stability of fund custody.

Description

Technical Field

[0001] This invention relates to the field of financial risk control technology, and more specifically, to a method and apparatus for hedging the time difference risk of insurance fund custody. Background Technology

[0002] In the scenario of "consumption-based insurance," there is an inherent time difference between fund transfers and premium deductions. After a user completes a transaction, the payment institution completes the settlement on the same day, transferring the funds to a bank custody account. Meanwhile, the insurance company typically initiates batch premium deductions the following day or later, transferring the funds from the custody account to the insurance institution's account. This time lag results in a temporary retention of premium funds awaiting transfer in the bank custody account, creating a short-term fund accumulation. This type of fund retention pattern is prevalent in existing business processes. As the business scales up, the amount of accumulated funds increases accordingly, making it susceptible to being identified as a fund pooling risk during regulatory audits. This places higher demands on the compliance of business operations and the standardization of fund supervision.

[0003] However, existing technologies mostly use a unified account to collect funds without achieving separate account isolation. This can easily lead to the formation of idle funds due to time differences, triggering the risk of fund pool supervision. Furthermore, the lack of a pre-reconciliation mechanism results in a high rate of deduction errors. When unilateral account anomalies occur, there is no means of rapid advance payment to restore the situation, which can easily affect policy performance and fund security.

[0004] No effective solutions have yet been proposed to address the problems in the relevant technologies. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention proposes a method and apparatus for hedging the time difference risk in insurance fund custody. This solves the problems mentioned in the background technology, which states that existing technologies often use a unified account to collect funds, fail to achieve separate accounts, are prone to the formation of idle funds due to time differences, trigger fund pool supervision risks, lack a pre-reconciliation mechanism, have a high deduction error rate, and lack a rapid advance payment recovery method when unilateral account anomalies occur, which can easily affect policy performance and fund security.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] According to one aspect of the present invention, a method for hedging the time difference risk of insurance fund custody is provided, comprising: S1. Based on the bank's custody account, and through the application programming interface of the second-class account or virtual account provided by the bank, a virtual sub-account is created for each user, and the virtual sub-account is bound to the corresponding policy account of the user to achieve physical isolation of funds. S2. When the consumer incentive fund enters the bank custody account, the bank's freezing interface is called to freeze the consumer incentive fund to a virtual sub-account bound to the policy account, and a targeted payment instruction is generated to restrict the consumer incentive fund to be transferred only to the designated insurance company account. S3. After the business ends on the same day, automatically generate a snapshot of the custodian fund flow as the benchmark data for the next day's reconciliation. Before the insurance company initiates batch deductions on the next day, pre-reconcile the snapshot of the custodian fund flow with the insurance company's list of deductions, and perform unfreezing and transfer or difference processing based on the pre-reconciliation results. S4. Establish a risk reserve pool. When a one-sided account anomaly occurs, the risk reserve pool will be used to make advance payments to ensure that the user's policy rights are not affected. After the cause of the anomaly is identified and the settlement is completed, the corresponding account adjustments will be made to achieve automatic error recovery.

[0008] Furthermore, when the consumer incentive fund enters the bank's custodian account, the bank's freezing interface is invoked to freeze the consumer incentive fund in a virtual sub-account linked to the policy account, and a targeted payment instruction is generated to restrict the consumer incentive fund to be transferred only to the designated insurance company account, including: S21. The bank custody account receives the notification of the consumer incentive fund in real time and obtains the user identifier and policy number corresponding to the consumer incentive fund. S22. Match the virtual sub-accounts bound to the policy account based on the user identifier and policy number; S23. Call the bank's freeze interface to freeze the entire amount of the consumption incentive fund to the matching virtual sub-account; S24. Generate a unique and tamper-proof targeted payment instruction based on the policy number, premium amount, and the insurance company's designated collection account; S25. Link the targeted payment instruction with the virtual sub-account and the consumption incentive fund, and restrict the consumption incentive fund to be transferred only to the insurance company's designated account in accordance with the targeted payment instruction.

[0009] Furthermore, after the day's business concludes, a snapshot of the custodian fund flow is automatically generated as the baseline data for the next day's reconciliation. Before the insurance company initiates batch deductions the following day, the snapshot of the custodian fund flow is pre-reconciled with the insurance company's list of funds to be deducted, and the unfreezing and transfer or discrepancy processing is performed based on the pre-reconciliation results, including: S31. After the business is completed on the same day, the fund status of all virtual sub-accounts under the bank custody account is collected through a traversal. S32. Based on the collected results, automatically generate a snapshot of the deposited funds' flow and store the snapshot as the benchmark data for reconciliation the next day. S33. Obtain a list of payments to be deducted from the insurance company before the insurance company initiates the bulk deduction the following day; S34. Compare the stored snapshot of the escrow funds flow with the list of funds to be deducted item by item to complete the pre-reconciliation; S35. If the pre-reconciliation results are consistent, the matching consumption incentive funds will be unfrozen and the funds will be transferred according to the targeted payment instructions. If the pre-reconciliation results are inconsistent, the discrepancy will be handled, and the abnormal funds will be frozen for verification or returned to the original payment method.

[0010] Furthermore, the stored snapshots of escrow fund flows are compared item by item with the list of funds to be deducted to complete the pre-reconciliation, including: S341. Construct a fund reconciliation matching network, initialize all data records in the fund flow snapshot and the list of pending deductions to an unmatched state, and set the number of overlapping matches of all data records to unmatched; S342. Select a central matching record in the reconciliation matching network, calculate the exclusion quality to centrality ratio of each data record, and include data records whose matching distance from the central matching record is less than the preset matching threshold into the same matching range. S343. Mark the center matching records that have been matched as matched, and select the record with the smallest ratio of exclusion quality to centrality from the remaining unmatched records as the center matching record of the next group. S344. Execute the matching record selection and data record matching division operations in the cycle execution center until the cash flow snapshot and all data records in the pending deduction list are matched and compared to complete the pre-reconciliation.

[0011] Furthermore, in the reconciliation matching network, a central matching record is selected, and the exclusion quality to centrality ratio of each data record is calculated. Data records whose matching distance to the central matching record is less than a preset matching threshold are grouped into the same matching range, including: S3421. In the established reconciliation matching network, select the initial central matching record based on the unique identifier of the data record as the benchmark record for the current batch of matching comparison; S3422. Determine the exclusion quality and centrality of each data record in the cash flow snapshot and the list of deductions respectively, and calculate the exclusion quality and centrality ratio of each data record accordingly. S3423. Using the field characteristics of the central matching record as a reference, calculate the matching distance between other data records and the central matching record one by one to quantify the degree of similarity between the data. S3424. Compare the calculated matching distance for each data record with a pre-set matching threshold to determine whether the matching conditions are met. S3425. Data records with a matching distance less than the preset matching threshold are grouped into the same matching range as the current center matching record to complete the matching division of the current batch of data.

[0012] Furthermore, using the field features of the central matching record as a reference, the matching distance between other data records and the central matching record is calculated one by one to quantify the similarity between the data, including: S34231. Extract the core alignment features from the center matching record and other data records to be compared, and convert the core alignment features of the center matching record and other data records to be compared into standardized data that can be used for numerical calculation. S34232. Calculate the difference for each of the standardized data of the same type in the center matching record and other data records to be compared, and take the absolute value of the calculated difference to obtain the difference value of the corresponding single-class feature. S34233. Based on the preset weight values ​​of each core comparison feature, the difference values ​​of each single-class feature are weighted and the weighted result is squared to obtain the weighted difference square value of the corresponding single-class feature. S34234. Sum the weighted squared differences of all single-class features, and take the square root of the sum to obtain the matching distance between the center matching record and other data records to be compared, so as to quantify the similarity between the data.

[0013] Furthermore, a risk reserve pool is established. When a unilateral accounting anomaly occurs, the risk reserve pool is automatically used to make an advance payment to ensure that the policyholder's rights and interests are not affected. After the cause of the anomaly is identified and the liquidation is completed, corresponding accounting adjustments are made to achieve automatic error recovery, including: S41. Based on the historical fund clearing scale and the frequency of daily anomalies, calculate and allocate the corresponding amount of risk reserve fund to complete the initial establishment and special account storage of the risk reserve fund pool. S42. Monitor fund transfers and reconciliation results in real time. When a one-sided account anomaly is detected, lock the corresponding accounting information and trigger the reserve activation instruction. S43. Based on the policy performance amount corresponding to the anomaly, the corresponding amount shall be drawn from the risk reserve pool for advance payment to maintain the normal effectiveness of the relevant policy rights; S44. After the advance payment is completed, initiate an anomaly check, trace the cause of the one-sided account, and record the anomaly type, the amount involved, and related data information simultaneously; S45. After the cause of the anomaly is identified and all parties have settled their accounts, the risk reserve pool shall be replenished or adjusted according to the settlement results in order to complete the automatic write-off and recovery of the erroneous accounts.

[0014] Furthermore, after the advance payment is completed, an anomaly investigation is initiated to trace the cause of the unilateral transaction and simultaneously record the anomaly type, the amount involved, and related data information, including: S441. From the obtained abnormal feature terms related to unilateral accounts, select the core abnormal feature terms with the highest weight as the priority verification targets, and at the same time randomly select some terms from the remaining abnormal feature terms as supplementary verification targets. S442. Conduct a comprehensive review of the abnormal characteristics corresponding to priority verification objects and supplementary verification objects, expand the scope of abnormal related information, and improve the traceability clues for single-sided account abnormalities. S443. Classify and sort the priority verification objects, supplementary verification objects and their corresponding abnormal feature information to form structured abnormal feature groups; S444. Check whether the total number of individuals with abnormal characteristics that have been identified exceeds the preset maximum number of individuals to be checked. If it exceeds the maximum number, proceed with the screening process. If it does not exceed the maximum number, continue to supplement and improve the abnormal characteristic information. S445. Sort each abnormal feature group according to the verification priority, and select the abnormal feature individuals with the highest verification priority in each group as key verification objects. If the number of abnormal feature groups reaches the preset limit, start screening again from the first abnormal feature group until the number of key verification objects selected meets the preset requirements. S446. Check if the current iteration count of the anomaly check has reached the maximum iteration count. If it has, terminate the anomaly check, complete the tracing of the cause of the single-sided account anomaly and the recording of relevant information. If it has not reached the maximum iteration count, return to the anomaly feature expansion process and continue to improve the check process until all anomaly tracing and information recording are completed.

[0015] Furthermore, the priority verification targets, supplementary verification targets, and corresponding abnormal feature information are classified and sorted to form structured abnormal feature groups, including: S4431. Based on the preset niche classification rules, extract the feature dimensions of priority verification objects and supplementary verification objects, and establish anomaly feature classification standards; S4432. According to the classification criteria, classify and divide various abnormal feature information according to the corresponding dimensions, and distinguish the feature sets corresponding to different abnormal manifestations. S4433. According to the degree of abnormal correlation and the importance of verification, sort the various abnormal features in the feature set internally, determine the verification order of features within the same category, and integrate the sorted abnormal features in a structured way to form a structured abnormal feature group.

[0016] According to another aspect of the present invention, a time difference risk hedging device for insurance fund custody is also provided, the device comprising: The virtual sub-account binding module is used to create a virtual sub-account for each user based on the bank's custody account and through the application programming interface of the bank's Class II account or virtual account, and bind the virtual sub-account to the user's corresponding policy account to achieve physical isolation of funds. The incentive fund freeze control module is used to call the bank's freeze interface when the consumer incentive fund enters the bank's custody account, freeze the consumer incentive fund to a virtual sub-account bound to the policy account, and generate a targeted payment instruction to restrict the consumer incentive fund to be transferred only to the designated insurance company account; The transaction snapshot and pre-reconciliation module is used to automatically generate a transaction snapshot of the escrow funds after the end of the day's business, which serves as the benchmark data for reconciliation the next day. Before the insurance company initiates batch deductions the next day, the transaction snapshot of the escrow funds is pre-reconciled with the insurance company's list of funds to be deducted, and the unfreezing and transfer or difference processing is performed based on the pre-reconciliation results. The risk reserve advance payment module is used to establish a risk reserve pool. When a one-sided account anomaly occurs, the risk reserve is automatically used to make an advance payment to ensure that the user's policy rights are not affected. After the cause of the anomaly is identified and the settlement is completed, the corresponding account adjustments are made to achieve automatic error recovery.

[0017] The beneficial effects of this invention are as follows: 1. This invention achieves physical isolation of funds by binding virtual sub-accounts and freezing funds for targeted transfer, avoiding the risk of funds pooling in special accounts; the pre-reconciliation mechanism automatically identifies and handles one-sided account anomalies, verifies fund consistency in advance, reduces deduction discrepancies, lowers manual intervention costs, protects policyholder rights, effectively hedges the regulatory and operational risks caused by time differences in fund transfers, and improves the stability of fund custody.

[0018] 2. This invention constructs a reconciliation matching network and adopts a central matching iterative approach to automatically match and classify fund flow and deduction list data, reducing manual intervention and improving data processing efficiency. By calculating exclusion quality, centrality ratio, and weighted Euclidean matching distance, it achieves quantitative measurement of differences in multi-dimensional data features, improving data matching accuracy and reducing false and missed matching rates. Through standardized numerical transformation and feature weighting operations, it achieves objective calculation of data similarity, improves the stability of reconciliation results, and provides reliable technical support for fund transfer timing control and anomaly identification.

[0019] 3. This invention determines the verification targets by combining weighted screening and supplementary selection, and classifies and sorts abnormal features according to niche classification rules, which can quickly focus on core abnormal information; by iteratively expanding feature clues and dynamically controlling the number of verifications and iterations, it realizes the automation and standardization of abnormal verification, and improves the efficiency of locating the cause of single-sided accounts; the structured feature groups can clearly sort out the abnormal relationship, reduce human judgment error, realize accurate tracing of abnormal causes, and quickly complete the recording of abnormal information, further improving the stability of fund custody and the efficiency of risk disposal. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a flowchart of a time difference risk hedging method for insurance fund custody according to an embodiment of the present invention; Figure 2 This is a schematic diagram of a time difference risk hedging device for insurance fund custody according to an embodiment of the present invention.

[0022] In the picture: 1. Virtual sub-account binding module; 2. Incentive bonus freeze control module; 3. Transaction snapshot and pre-reconciliation module; 4. Risk reserve fund advance payment module. Detailed Implementation

[0023] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0024] In the description of this invention, unless otherwise stated, "a plurality of" means two or more. Furthermore, the terms "first," "second," "third," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0025] According to an embodiment of the present invention, a method and apparatus for hedging the time difference risk of insurance fund custody are provided.

[0026] The present invention will now be further described in conjunction with the accompanying drawings and specific embodiments, such as... Figure 1 As shown, the time difference risk hedging method for insurance fund custody according to an embodiment of the present invention includes: S1. Based on the bank's custody account, and through the application programming interface (API) of the second-class account or virtual account provided by the bank, a virtual sub-account is created for each user, and the virtual sub-account is bound to the corresponding policy account of the user to achieve physical isolation of funds. S2. When the consumer incentive fund enters the bank custody account, the bank's freezing interface is called to freeze the consumer incentive fund to a virtual sub-account bound to the policy account, and a targeted payment instruction is generated to restrict the consumer incentive fund to be transferred only to the designated insurance company account. Specifically, during the freeze period, funds cannot be used for payments, withdrawals, or other purposes; they can only be transferred to the designated insurance company account. The targeted payment instruction contains the target insurance company's account information, premium amount, and policy number, and the instruction cannot be altered once generated.

[0027] S3. After the business ends on the same day, automatically generate a snapshot of the custodian fund flow as the benchmark data for the next day's reconciliation. Before the insurance company initiates batch deductions on the next day, pre-reconcile the snapshot of the custodian fund flow with the insurance company's list of deductions, and perform unfreezing and transfer or difference processing based on the pre-reconciliation results. Specifically, at the end of T+0 (i.e., the same day), a snapshot of the escrow funds is generated, including the virtual sub-account balance, the total frozen funds, and details of the funds to be insured; at the beginning of T+1 (i.e., the next day), before the insurance company initiates batch deductions, a pre-reconciliation is performed: the insurance company's list of deductions to be made is compared with the escrow snapshot; if the amounts match, the funds are unfrozen and transferred; if the amounts do not match (e.g., due to user cancellation resulting in fund refund), a difference handling process is triggered: the difference remains frozen until the reason is determined; or it is refunded to the user's payment account via the original payment method.

[0028] S4. Establish a risk reserve fund (funded by the platform's own funds). When a one-sided account anomaly occurs (the bank has deducted the payment but the insurance has not been credited, or vice versa), the risk reserve fund will be used to make advance payments to ensure that the user's policy rights are not affected. After the cause of the anomaly is identified and the settlement is completed, the corresponding account adjustments will be made to achieve automatic error recovery.

[0029] Specifically, the reserve provision ratio is dynamically calculated based on the historical daily peak fund amount as follows: Reserve balance = Max(Daily peak fund amount × 10%, 1 million yuan).

[0030] In this optional embodiment, when the consumer incentive fund enters the bank's custodian account, the bank's freezing interface is invoked to freeze the consumer incentive fund in a virtual sub-account bound to the policy account, and a targeted payment instruction is generated to restrict the consumer incentive fund to be transferred only to a designated insurance company account, including: S21. The bank custody account receives the notification of the consumer incentive fund in real time and obtains the user identifier and policy number corresponding to the consumer incentive fund. S22. Match the virtual sub-accounts bound to the policy account based on the user identifier and policy number; S23. Call the bank's freeze interface to freeze the entire amount of the consumption incentive fund to the matching virtual sub-account; S24. Generate a unique and tamper-proof targeted payment instruction based on the policy number, premium amount, and the insurance company's designated collection account; S25. Link the targeted payment instruction with the virtual sub-account and the consumption incentive fund, and restrict the consumption incentive fund to be transferred only to the insurance company's designated account in accordance with the targeted payment instruction.

[0031] In this optional embodiment, after the end of the day's business, a snapshot of the escrow fund flow is automatically generated as the baseline data for reconciliation the next day. Before the insurance company initiates batch deductions the next day, the snapshot of the escrow fund flow is pre-reconciled with the insurance company's list of funds to be deducted, and the unfreezing and transfer or discrepancy processing is performed based on the pre-reconciliation results, including: S31. After the business is completed on the same day, the fund status of all virtual sub-accounts under the bank custody account is collected through a traversal. S32. Based on the collected results, automatically generate a snapshot of the deposited funds' flow and store the snapshot as the benchmark data for reconciliation the next day. S33. Obtain a list of payments to be deducted from the insurance company before the insurance company initiates the bulk deduction the following day; S34. Compare the stored snapshot of the escrow funds flow with the list of funds to be deducted item by item to complete the pre-reconciliation; S35. If the pre-reconciliation results are consistent, the matching consumption incentive funds will be unfrozen and the funds will be transferred according to the targeted payment instructions. If the pre-reconciliation results are inconsistent, the discrepancy will be handled, and the abnormal funds will be frozen for verification or returned to the original payment method.

[0032] In this optional embodiment, the pre-reconciliation process involves comparing the stored snapshots of escrow fund flows with the list of funds to be deducted item by item. S341. Construct a fund reconciliation matching network, initialize all data records in the fund flow snapshot and the list of pending deductions to an unmatched state, and set the number of overlapping matches of all data records to unmatched; S342. Select a central matching record in the reconciliation matching network, calculate the exclusion quality to centrality ratio of each data record, and include data records whose matching distance from the central matching record is less than the preset matching threshold into the same matching range. S343. Mark the center matching records that have been matched as matched, and select the record with the smallest ratio of exclusion quality to centrality from the remaining unmatched records as the center matching record of the next group. S344. Execute the matching record selection and data record matching division operations in the cycle execution center until the cash flow snapshot and all data records in the pending deduction list are matched and compared to complete the pre-reconciliation.

[0033] Specifically, a fund reconciliation matching network is constructed. All data records in the transaction snapshot and the pending deduction list, including user identity information, policy details, fund amounts, and transfer accounts, are uniformly initialized and set to an unmatched state. Simultaneously, the number of overlapping matches for each record is marked as unmatched to avoid duplicate data comparisons. In the matching network, a record with a complete policy number and clear fund information is selected as the initial central matching record. The exclusion quality to centrality ratio of all other records is calculated one by one. Then, using this central record as a benchmark, the matching distance between each record to be compared and the central record is calculated. This matching distance is compared to a preset matching threshold (ranging from 0.2 to 0.6, dynamically adjusted according to business scenarios). Records with a value less than this threshold are grouped into the same matching group to achieve accurate matching. After completing a group matching, the central record is marked as matched. From the remaining unmatched records, the record with the smallest exclusion quality to centrality ratio is selected as the new central matching record. The above calculation, comparison, and grouping operations are repeated, iterating until all transaction snapshots and pending deduction list records are fully matched. Based on the matching results, the system automatically determines the consistency of fund transfers. For those with correct matches, the system unfreezes and transfers the funds. For those with discrepancies, the system initiates a discrepancy handling process to ensure that insurance fund transfers are compliant and accurate, meeting the fund management needs in time-difference scenarios.

[0034] Specifically, this invention uses the overlapping box covering algorithm to compare the cash flow snapshot with the list of deductions item by item and achieve pre-reconciliation. The overlapping box covering algorithm is a heuristic matching algorithm based on multi-dimensional feature space geometric measurement. It follows the operation rules of selecting the center node, calculating the feature distance, dividing the box range, and iterative full-domain coverage. Through feature space mapping, matching distance measurement, threshold boundary constraints, and cyclic optimization matching mechanism, it achieves accurate matching and consistency verification of the full amount of cash data.

[0035] In this invention, the overlapping box covering algorithm adopts an iterative center matching architecture. It uses fund flow and pending deduction data records as search units, excludes quality and centrality ratio as filtering criteria, and uses matching distance threshold as boundary constraint. Through four stages, namely, constructing a reconciliation matching network, selecting center matching records, box range division, and iterative global matching, it achieves fully automatic and accurate pre-reconciliation. The specific usage process and operation details are as follows: Based on the five core comparison features of the fund reconciliation scenario, namely user information, policy number, amount, transfer account, and account identifier, the number of center matching records in a single batch is set to 1. At the same time, the maximum number of iterations is preset to 50, and the preset matching threshold is 0.2 to 0.6 (used to distinguish between valid matches and abnormal data). The initial overlapping matching count is uniformly set to unmatched state.

[0036] Based on the full set of raw data from the cash flow snapshot and the list of deductions to be processed, the matching status and number of overlapping matches of all data records are initialized, corresponding to four major data types: cash flow identifier, amount value, account information, and policy association. At the same time, the matching judgment rules are determined (i.e., the correspondence between matching distance and matching result is clarified; matching distance less than the threshold is a valid match, and matching distance greater than or equal to the threshold is abnormal data to be reviewed). The initialized status parameters and judgment rules are entered into the algorithm to complete all the initial configurations for iterative matching, and the reconciliation matching network of the overlapping box covering algorithm is constructed to ensure that there are clear execution standards for subsequent iterations.

[0037] The overlapping box covering algorithm calls the preset feature quantization rules to perform feature standardization processing on all data records after initialization. Combining the topological relationship of each record in the matching network, the algorithm obtains the exclusion quality and centrality ratio of each data record through the built-in exclusion quality and centrality calculation logic. This ratio is used as the core basis for selecting the center record, completing the first round of feature quantization and providing quantitative support for subsequent matching and partitioning.

[0038] The overlapping box covering algorithm automatically selects the initial center matching record as the central benchmark of the box coverage range based on the ratio of exclusion quality to centrality of each data record. The feature weight of the center record is set to the benchmark value of 1.0, making it the sole reference standard in the matching distance calculation. At the same time, it automatically compares the feature similarity between the current center record and the remaining unmatched records. If the newly selected record has a smaller ratio and higher feature integrity, it is adopted as the center node of the next stage, and the matching coverage range is expanded synchronously. This achieves adaptive optimization of the matching benchmark and promotes the algorithm towards full coverage.

[0039] The overlapping box covering algorithm uses multi-dimensional feature difference calculation combined with weighted distance operation to update the matching distance value between each record and the center record. During the calculation process, it automatically filters out valid matching records with matching distance less than a preset threshold, removes invalid records with abnormal matching or missing features, simplifies the comparison process, reduces interference, and optimizes the overall calculation process of pre-reconciliation. For example, after the first round of matching, it divides the data into 3 valid matching groups, marks 2 abnormal records, and completes the matching division of the current batch of data.

[0040] The overlapping box covering algorithm counts the current iteration matching count in real time. After each round of center selection and range division, it automatically accumulates the iteration step size. When the step size reaches the preset iteration cycle node, the algorithm automatically clears the temporary matching mark. At the same time, it updates the center record selection strategy according to the current iteration matching accuracy (for example, prioritizing records with high feature completeness as centers and delaying records with high dispersion to participate in matching), and synchronously adjusts the matching judgment direction to ensure the rationality of subsequent comparison operations, avoid efficiency loss caused by repeated calculations, and keep the algorithm maintaining a stable matching orientation.

[0041] The overlapping box covering algorithm continuously counts the current iteration count. After each round of center selection, distance calculation, and box partitioning, the iteration count is incremented by 1. The current iteration count is compared with the preset maximum iteration count. If the maximum limit has been reached, the algorithm terminates the iteration and automatically outputs the final pre-reconciliation result, clearly identifying matching data, discrepancies, and abnormal records. If the maximum limit has not been reached, the algorithm automatically returns to the center matching record selection stage and continues to perform feature quantization, distance calculation, and range partitioning until the iteration termination condition is met. The overlapping box covering algorithm's computation process is fully specified. Input data comes directly from cash flow snapshots and pending deduction lists. The raw data has undergone field standardization, numerical normalization, and duplicate data removal. Data sources cover bank custody systems, insurance company business systems, and user insurance records. All data undergoes format validation, anomaly removal, account anonymization, and field alignment to ensure stable, standardized, and traceable algorithm input.

[0042] It adopts an evaluation mechanism centered on feature distance measurement, quality selection for exclusion, and full-domain coverage matching. With effective matching and grouping, abnormal data marking, and full record comparison as optimization objectives, it can achieve accurate matching without the need for traditional loss functions. The matching logic is completely consistent with the business objectives of fund reconciliation. It has the characteristics of high accuracy, fast speed, and strong stability, and is highly consistent with the matching rules of multi-dimensional fund data features.

[0043] The key parameters of the overlapping box covering algorithm are clearly defined and have sufficient selection criteria. The preset matching threshold is 0.2~0.6, the maximum number of iterations is 50, and the initial matching state is uniformly unmatched. All parameters are determined in combination with the accuracy requirements of fund reconciliation, data volume and computing efficiency. They have clear adjustment ranges (for example, the maximum number of iterations can be adjusted between 40 and 60, and the matching threshold can be adjusted between 0.1 and 0.7) and optimization strategies (dynamically fine-tuning the threshold based on the real-time matching success rate).

[0044] The input data for the overlapping box covering algorithm includes standardized cash flow snapshots, a list of pending deductions, matching thresholds, maximum number of iterations, initial state parameters, and feature distance calculation rules. All inputs are directly derived from the fund custody and business reconciliation process, making them highly relevant to the application scenario, reusable, and verifiable. The algorithm output data includes the matching status of each data record, the sequence of center-matched records, the combined matching results, the full data matching report, and the final pre-reconciliation conclusion. The output and input form a complete logical closed loop through feature quantization, center selection, distance calculation, box partitioning, and iteration termination.

[0045] In this optional embodiment, a central matching record is selected in the reconciliation matching network, the exclusion quality to centrality ratio of each data record is calculated, and data records whose matching distance from the central matching record is less than a preset matching threshold are grouped into the same matching range, including: S3421. In the established reconciliation matching network, select the initial central matching record based on the unique identifier of the data record as the benchmark record for the current batch of matching comparison; S3422. Determine the exclusion quality and centrality of each data record in the cash flow snapshot and the list of deductions respectively, and calculate the exclusion quality and centrality ratio of each data record accordingly. S3423. Using the field characteristics of the central matching record as a reference, calculate the matching distance between other data records and the central matching record one by one to quantify the degree of similarity between the data. S3424. Compare the calculated matching distance for each data record with a pre-set matching threshold to determine whether the matching conditions are met. S3425. Data records with a matching distance less than the preset matching threshold are grouped into the same matching range as the current center matching record to complete the matching division of the current batch of data.

[0046] Specifically, within the established reconciliation matching network, the unique identifiers of data records—such as unique policy numbers, user real-name registration information, and virtual sub-account binding identifiers—are used as the filtering criteria. Records with complete fields, no missing information, and no abnormal markers are automatically selected as the initial central matching record. This record serves as the sole benchmark record for the current batch of matching, ensuring the accuracy and reliability of the comparison benchmark. The core parameters of each data record in the cash flow snapshot and the insurance company's pending deduction list are automatically extracted. The exclusion quality (based on data completeness and field standardization) and centrality (based on the degree of correlation between the record and other data) of each record are determined. The ratio of exclusion quality to centrality for each record is obtained through numerical division. The exclusion quality is automatically quantified based on technical characteristics such as field completeness, format standardization, and information uniqueness of the data record; the centrality is automatically calculated based on the record's correlation density and feature overlap with other records in the reconciliation matching network. The exclusion quality value of the same record is used as the numerator, and the centrality value is used as the denominator. The corresponding ratio is obtained by arithmetic division, providing a quantitative basis for subsequent screening. Using the core field features of the selected initial central matching record, including user information, policy details, fund amount, and transfer account, as a reference standard, the matching distance between each of the other data records to be compared and the central matching record is calculated. This distance quantifies the similarity between the two data records; the smaller the distance, the higher the similarity. A preset matching threshold (ranging from 0.2 to 0.6, dynamically adjustable according to business scale and data accuracy requirements; the default value is 0.4 in typical business scenarios) is called. The calculated matching distance of each record to be compared is automatically compared with the preset threshold to accurately determine whether each record meets the matching conditions. For data records whose matching distance is less than the preset matching threshold and are determined to meet the matching conditions, they are automatically grouped into the same matching range as the current initial central matching record, completing the data matching division for the current batch.

[0047] In this optional embodiment, using the field features of the central matching record as a reference, the matching distance between other data records and the central matching record is calculated one by one to quantify the similarity between the data, including: S34231. Extract the core alignment features from the center matching record and other data records to be compared, and convert the core alignment features of the center matching record and other data records to be compared into standardized data that can be used for numerical calculation. S34232. Calculate the difference for each of the standardized data of the same type in the center matching record and other data records to be compared, and take the absolute value of the calculated difference to obtain the difference value of the corresponding single-class feature. S34233. Based on the preset weight values ​​of each core comparison feature, the difference values ​​of each single-class feature are weighted and the weighted result is squared to obtain the weighted difference square value of the corresponding single-class feature. S34234. Sum the weighted squared differences of all single-class features, and take the square root of the sum to obtain the matching distance between the center matching record and other data records to be compared, so as to quantify the similarity between the data.

[0048] Specifically, the core comparison features of the central matching record and the data record to be compared are extracted, including policy number, user virtual sub-account, incentive amount, and transfer account information. Various textual and coded features are uniformly converted into standardized numerical data, enabling computational processing of different types of features. The difference between each standardized data of the same type in the two sets of records is calculated, and the absolute value is taken to obtain the single-class feature difference value. Based on preset feature weight values ​​(e.g., amount weight 0.4, policy number weight 0.3, account information weight 0.3, summed to 1), each difference value is weighted. The weighted result is squared to obtain the weighted difference square value. The weighted difference square values ​​of all features are summed, and the square root of the sum is taken to obtain the matching distance between the two records, thus quantifying the degree of data similarity.

[0049] In this optional embodiment, a risk reserve pool is established. When a unilateral accounting anomaly occurs, the risk reserve pool is automatically used to make an advance payment to ensure that the user's policy rights are not affected. After the cause of the anomaly is identified and the liquidation is completed, corresponding accounting adjustments are made to achieve automatic error recovery, including: S41. Based on the historical fund clearing scale and the frequency of daily anomalies, calculate and allocate the corresponding amount of risk reserve fund to complete the initial establishment and special account storage of the risk reserve fund pool. S42. Monitor fund transfers and reconciliation results in real time. When a one-sided account anomaly is detected, lock the corresponding accounting information and trigger the reserve activation instruction. S43. Based on the policy performance amount corresponding to the anomaly, the corresponding amount shall be drawn from the risk reserve pool for advance payment to maintain the normal effectiveness of the relevant policy rights; S44. After the advance payment is completed, initiate an anomaly check, trace the cause of the one-sided account, and record the anomaly type, the amount involved, and related data information simultaneously; S45. After the cause of the anomaly is identified and all parties have settled their accounts, the risk reserve pool shall be replenished or adjusted according to the settlement results in order to complete the automatic write-off and recovery of the erroneous accounts.

[0050] In this optional embodiment, after the advance payment is completed, an anomaly check is initiated to trace the cause of the one-sided account, and the anomaly type, the amount involved, and related data information are recorded simultaneously, including: S441. From the obtained abnormal feature terms related to unilateral accounts, select the core abnormal feature terms with the highest weight as the priority verification targets, and at the same time randomly select some terms from the remaining abnormal feature terms as supplementary verification targets. S442. Conduct a comprehensive review of the abnormal characteristics corresponding to priority verification objects and supplementary verification objects, expand the scope of abnormal related information, and improve the traceability clues for single-sided account abnormalities. S443. Classify and sort the priority verification objects, supplementary verification objects and their corresponding abnormal feature information to form structured abnormal feature groups; S444. Check whether the total number of individuals with abnormal characteristics that have been identified exceeds the preset maximum number of individuals to be checked. If it exceeds the maximum number, proceed with the screening process. If it does not exceed the maximum number, continue to supplement and improve the abnormal characteristic information. S445. Sort each abnormal feature group according to the verification priority, and select the abnormal feature individuals with the highest verification priority in each group as key verification objects. If the number of abnormal feature groups reaches the preset limit, start screening again from the first abnormal feature group until the number of key verification objects selected meets the preset requirements. S446. Check if the current iteration count of the anomaly check has reached the maximum iteration count. If it has, terminate the anomaly check, complete the tracing of the cause of the single-sided account anomaly and the recording of relevant information. If it has not reached the maximum iteration count, return to the anomaly feature expansion process and continue to improve the check process until all anomaly tracing and information recording are completed.

[0051] Specifically, from the acquired abnormal feature terms related to one-sided accounts, the core abnormal feature terms with the highest weight are selected as priority verification targets, with a weight of 60% or more. Supplementary verification targets are randomly selected from the remaining abnormal feature terms to ensure comprehensive verification. The abnormal features corresponding to the priority and supplementary verification targets are comprehensively reviewed, focusing on expanding the scope of abnormal related information, including related user information, policy information, and fund flow information, to improve the tracing clues for one-sided account anomalies. The priority and supplementary verification targets and their corresponding abnormal feature information are sorted by category to form structured abnormal feature groups, such as account-related, amount-related, and system-related abnormal feature groups. It is checked whether the total number of currently identified abnormal feature individuals exceeds the preset maximum verification number. If it does, core abnormal feature individuals are selected, and duplicate and irrelevant features are removed; if it does not exceed the maximum, the abnormal feature information is further supplemented and improved. The abnormal feature groups are sorted according to their verification priority (i.e., core abnormalities > supplementary abnormalities > general abnormalities). The individuals with the highest verification priority within each group are selected as key verification targets. If the number of abnormal feature groups reaches a preset limit, the selection process restarts from the first abnormal feature group until the number of selected key verification targets meets the preset requirement. The system checks whether the current iteration count for abnormal verification has reached the maximum iteration count. If it has, the abnormal verification is terminated, and the cause tracing and related information recording for the single-sided account abnormality are completed. If not, the system returns to the abnormal feature expansion process to continue refining the verification until all abnormal tracing and information recording are completed.

[0052] Specifically, this invention initiates anomaly verification after the funds are advanced using the Invading Weeds algorithm, tracing the cause of the one-sided account and simultaneously recording the anomaly type, the amount involved, and related data. The Invading Weeds algorithm is a heuristic optimization algorithm that simulates the invasion, spread, nutrient competition, survival of the fittest, and community iteration of wild weeds in a complex environment. It follows the natural evolutionary laws of seed initialization, spread growth, competitive selection, advantage preservation, and iterative convergence. Through feature weight selection, random supplementary sampling, association range expansion, and population iterative screening mechanisms, it achieves accurate positioning and efficient tracing of one-sided account anomaly features.

[0053] In this invention, the Invading Weeds algorithm adopts a population-based iterative anomaly search architecture. It uses unilateral account anomaly features as individual algorithms, feature weights and correlation strengths as the basis for growth competitiveness, and maximum verification count, group upper limit, and maximum iteration count as environmental constraints. Through six stages—seed selection, feature diffusion and expansion, population classification and grouping, quantity constraint screening, dominant individual purification, and iterative convergence termination—it achieves rapid location and complete tracing of the causes of unilateral account anomalies. The specific usage process and operational details are as follows: Based on the four categories of anomaly features in unilateral account anomaly scenarios—fund transfer, account status, transaction sequence, and data matching—the initial number of dominant features is clearly set as the core feature with the highest weight. Simultaneously, the maximum verification count is preset to 12 to 18 (balancing tracing completeness and computational efficiency), the upper limit of anomaly feature groups is 4 to 6 categories, reasonably divided according to anomaly dimensions, the maximum iteration count is 8 to 12 times, and the weight screening threshold is 0.6, used to distinguish core anomaly features from general anomaly features.

[0054] Based on the extracted set of anomalous feature terms related to one-sided accounts, an anomalous feature population for the "Invasion Weeds" algorithm is initialized, corresponding to four major anomaly types: fund amount, account information, interaction time series, and data matching. Simultaneously, a competitiveness judgment rule for anomalous features is determined (i.e., a weight ratio greater than or equal to 0.6 indicates a core advantage seed, 0.3 to 0.6 indicates a common feature, and less than 0.3 indicates a weakly correlated feature). The initialized feature population and judgment rules are then input into the algorithm, completing all initial configurations for iterative search and constructing the initial anomaly search environment for the "Invasion Weeds" algorithm, ensuring clear execution standards for subsequent iterations. The "Invasion Weeds" algorithm calls the preset weight calculation rules to quantify the competitiveness of each anomalous feature term. Combining the correlation strength value between each feature and the one-sided account event, the algorithm's built-in weight evaluation logic yields the weight ratio value corresponding to each anomalous feature. This weight serves as the core basis for seed selection, completing the first round of advantage seed screening and providing precise support for subsequent feature expansion and tracing.

[0055] The invading weed algorithm automatically selects the core feature with the highest weight as the priority verification object (i.e., dominant seed) based on the weight ratio of each abnormal feature. At the same time, it randomly selects some words from the remaining features as supplementary verification objects (i.e., companion seeds), simulating the selective retention and random diffusion mechanism of weed seeds to ensure the coverage and focus of the tracing. Meanwhile, the algorithm automatically expands and sorts out the association information of the priority and supplementary verification objects, enhances the correlation clues between features, simulates the diffusion behavior of weed roots, makes the abnormal tracing clues more complete and coherent, and promotes the algorithm to converge towards the real abnormal root source.

[0056] The invasive weed algorithm classifies priority verification objects, supplementary verification objects, and expanded abnormal features according to the microhabitat classification rules, namely, the standardized technical classification criteria based on the dimensions of abnormal features, the type of cause, and the data source, forming structured abnormal feature groups to simulate the regional distribution of weed populations. Then, the algorithm automatically counts the total number of current feature individuals and compares it with the preset maximum number of verifications (e.g., 12 to 18). If it exceeds the limit, competitive screening is performed to remove weakly correlated and low-weight redundant features. If it does not exceed the limit, it continues to supplement related features to improve the integrity of the population.

[0057] The invading weed algorithm sorts each abnormal feature group according to the verification priority, and selects the dominant features with the highest weight and strongest correlation in each group as the key verification objects, simulating the competitive mechanism of dominant individuals in weed populations. If the number of feature groups reaches the preset limit, such as 4 to 6 types, the filtering starts from the first group again until the number of key verification objects meets the preset requirements, ensuring that the tracing is focused and no key anomalies are missed.

[0058] The "Invasion Weeds" algorithm continuously monitors the current iteration count. After each round of feature selection, expansion, classification, and filtering, the iteration count is incremented by 1 and compared to the preset maximum iteration count (e.g., 8 to 12). If the maximum iteration count is reached, the iteration terminates, and the final anomaly tracing result is output, recording the anomaly type, involved amount, related data, and root cause. If the maximum iteration count is not reached, the algorithm returns to the feature expansion stage to continue iterative mining until convergence. The input data for the "Invasion Weeds" algorithm comes from single-sided account anomaly feature sets, transaction snapshots, deduction lists, and system logs. The data has been standardized, deduplicated, and anonymized, and its source is traceable. The "Invasion Weeds" algorithm optimizes by retaining high-weight features, eliminating weak features, and converging anomaly root causes. It achieves adaptive optimization without a loss function, highly matching the localization patterns of complex anomalies in single-sided accounts.

[0059] The key parameters of the "Invading Weeds" algorithm are clearly defined and well-founded. The maximum number of checks is 12 to 18, the maximum number of groups is 4 to 6, the maximum number of iterations is 8 to 12, and the weighting threshold is 0.6. These parameters are determined based on the complexity of single-sided account checks, positioning accuracy, and response efficiency, and possess a reasonable adjustment range and dynamic optimization strategy. The algorithm's inputs include anomaly feature sets, weighting rules, and various constraint parameters. Its outputs include key check targets, anomaly groups, tracing process records, and the final single-sided account cause conclusion, forming a complete logical closed loop.

[0060] In this optional embodiment, the priority verification objects, supplementary verification objects, and corresponding abnormal feature information are classified and sorted to form structured abnormal feature groups, including: S4431. Based on the preset niche classification rules, extract the feature dimensions of priority verification objects and supplementary verification objects, and establish anomaly feature classification standards; S4432. According to the classification criteria, classify and divide various abnormal feature information according to the corresponding dimensions, and distinguish the feature sets corresponding to different abnormal manifestations. S4433. According to the degree of abnormal correlation and the importance of verification, sort the various abnormal features in the feature set internally, determine the verification order of features within the same category, and integrate the sorted abnormal features in a structured way to form a structured abnormal feature group.

[0061] Specifically, based on pre-defined niche classification rules, abnormal features are extracted along with dimensions such as funds, accounts, and interactions to establish standardized abnormal feature classification standards. Following these standards, various abnormal features, such as matching failures, inconsistent amounts, and abnormal accounts, are categorized according to their corresponding dimensions, forming feature sets corresponding to different abnormal behaviors. Each feature set is then sorted according to its correlation and importance for verification, prioritizing features with high correlation and importance. The priority difference between features within the same category is set to 0.1 to 0.3. Finally, the sorted abnormal features are structurally integrated to form hierarchical and ordered structured abnormal feature groups, providing a standardized data foundation for subsequent accurate and efficient abnormal verification.

[0062] According to another embodiment of the invention, such as Figure 2 As shown, a time difference risk hedging device for insurance fund custody is also provided, the device comprising: Virtual sub-account binding module 1 is used to create virtual sub-accounts for each user based on the bank's custody account and through the application programming interface of the second-class account or virtual account provided by the bank, and bind the virtual sub-accounts to the corresponding policy accounts of the users to achieve physical isolation of funds. The incentive fund freeze control module 2 is used to call the bank's freeze interface when the consumer incentive fund enters the bank's custody account, freeze the consumer incentive fund to a virtual sub-account bound to the policy account, and generate a targeted payment instruction to restrict the consumer incentive fund to be transferred only to the designated insurance company account; The transaction snapshot and pre-reconciliation module 3 is used to automatically generate a transaction snapshot of the custodian funds after the end of the business day, which serves as the benchmark data for reconciliation the next day. Before the insurance company initiates batch deductions the next day, the transaction snapshot of the custodian funds is pre-reconciled with the insurance company's list of deductions to be processed, and the unfreezing and transfer or difference processing is performed based on the pre-reconciliation results. The Risk Reserve Advance Payment Module 4 is used to establish a risk reserve pool. When a one-sided account anomaly occurs, the risk reserve will be used to make an advance payment to ensure that the user's policy rights are not affected. After the cause of the anomaly is identified and the settlement is completed, the corresponding account adjustments will be made to achieve automatic error recovery.

[0063] To facilitate understanding of the above technical solutions of the present invention, the following provides a detailed description of how the present invention hedges against the time difference risk in the actual process of insurance fund custody.

[0064] I. Virtual sub-accounts are physically separated from funds.

[0065] An online insurance platform offers a "consumption-based insurance" service, integrating with a commercial bank's custody system. Through the bank's API interface for Class II accounts, a dedicated virtual sub-account is created for each insured user, achieving physical segregation of funds. When user A completes a purchase through the platform, a virtual sub-account (6214XXXX0001) is automatically opened for them, uniquely linked to their corresponding policy number POL2026042200179. Similarly, when user B completes a purchase, their corresponding virtual sub-account (6214XXXX0002) is linked to policy number POL2026042200180. All virtual sub-accounts belong to the platform's dedicated bank custody account. Funds between these accounts are independent and physically segregated; changes in user A's and user B's funds do not affect each other, effectively preventing the risk of misappropriation or misuse of funds and ensuring the traceability of every insurance purchase.

[0066] II. Freezing and Targeted Payment of Consumption Incentive Funds.

[0067] User A receives a consumption incentive of 200 yuan after making a purchase. Once this fund enters the platform's bank custody account, the bank's freeze interface is immediately invoked to freeze the entire 200 yuan in User A's linked virtual sub-account 6214XXXX0001. During the freeze period, the funds cannot be withdrawn, misappropriated, or transferred to non-designated accounts. Simultaneously, an immutable targeted payment instruction is generated, clearly indicating: the target insurance company account is 95500XXXX123456, the transfer amount is 200 yuan, and the associated policy number is POL2026042200179. This instruction is permanently stored and cannot be modified after generation. User B receives a consumption incentive of 150 yuan, which is simultaneously frozen in their virtual sub-account 6214XXXX0002. The targeted payment instruction is associated with policy number POL2026042200180, ensuring that the funds are used only for the corresponding insurance policy.

[0068] III. End-of-day snapshot and T+1 pre-reconciliation.

[0069] 1) T+0 end-of-day cash snapshot (unit: yuan), as shown in Table 1: Table 1. T+0 End-of-Day Funds Snapshot

[0070] 2) T+1 Day Pre-reconciliation and Fund Processing: The following day, the insurance company uploaded a batch list of pending deductions, which included deductions of 200 yuan for policy POL2026042200179 and 150 yuan for policy POL2026042200180, for a total deduction of 350 yuan. The list was compared with the end-of-day snapshot, confirming a perfect match. The system automatically executed the unfreezing instruction, transferring the two funds to the target insurance company's account, thus completing the insurance application process.

[0071] If a discrepancy occurs: For example, if user A cancels an order, the deduction amount for that policy in the pending deduction list becomes 0 yuan, which does not match the 200 yuan frozen amount in the snapshot, the discrepancy processing will be triggered immediately, and the 200 yuan frozen funds will be returned to user A's payment account via the original payment method, while the snapshot data will be updated.

[0072] IV. Risk reserve advance payment and unilateral accounting treatment.

[0073] Based on historical data from the past 6 months, the platform calculated that the peak daily fund transfer was 12 million yuan. In accordance with the rules, risk reserves were set aside: Reserve balance = Max (12 million yuan × 10%, 1 million yuan) = 1.2 million yuan. The platform allocated 1.2 million yuan from its own funds to deposit into the reserve account.

[0074] One day, the bank deducted 200 yuan from user A's insurance premium from the platform's escrow account, but the insurance company's interface failed to record the transaction due to timeout, resulting in a one-sided account. The reserve fund was immediately activated, allocating 200 yuan from the 1.2 million yuan reserve to the insurance company's account to ensure user A's policy took effect immediately. Subsequent investigation revealed an interface malfunction. After clearing, the 200 yuan was returned to the reserve pool from the escrow account, restoring the reserve balance to 1.2 million yuan. The anomaly record was updated simultaneously, and the accounting adjustments were completed.

[0075] 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, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for hedging the time difference risk in insurance fund custody, characterized in that, include: S1. Based on the bank's custody account, and through the application programming interface of the second-class account or virtual account provided by the bank, a virtual sub-account is created for each user, and the virtual sub-account is bound to the corresponding policy account of the user to achieve physical isolation of funds. S2. When the consumer incentive fund enters the bank custody account, the bank's freezing interface is called to freeze the consumer incentive fund to a virtual sub-account bound to the policy account, and a targeted payment instruction is generated to restrict the consumer incentive fund to be transferred only to the designated insurance company account. S3. After the business ends on the same day, automatically generate a snapshot of the custodian fund flow as the benchmark data for the next day's reconciliation. Before the insurance company initiates batch deductions on the next day, pre-reconcile the snapshot of the custodian fund flow with the insurance company's list of deductions, and perform unfreezing and transfer or difference processing based on the pre-reconciliation results. S4. Establish a risk reserve pool. When a one-sided account anomaly occurs, the risk reserve pool will be used to make advance payments to ensure that the user's policy rights are not affected. After the cause of the anomaly is identified and the settlement is completed, the corresponding account adjustments will be made to achieve automatic error recovery.

2. The method for hedging the time difference risk of insurance fund custody according to claim 1, characterized in that, When the consumer incentive fund enters the bank's escrow account, the bank's freezing interface is invoked to freeze the consumer incentive fund in a virtual sub-account linked to the policy account, and a targeted payment instruction is generated to restrict the consumer incentive fund to be transferred only to the designated insurance company account. S21. The bank custody account receives the notification of the consumer incentive fund in real time and obtains the user identifier and policy number corresponding to the consumer incentive fund. S22. Match the virtual sub-accounts bound to the policy account based on the user identifier and policy number; S23. Call the bank's freeze interface to freeze the entire amount of the consumption incentive fund to the matching virtual sub-account; S24. Generate a unique and tamper-proof targeted payment instruction based on the policy number, premium amount, and the insurance company's designated collection account; S25. Link the targeted payment instruction with the virtual sub-account and the consumption incentive fund, and restrict the consumption incentive fund to be transferred only to the insurance company's designated account in accordance with the targeted payment instruction.

3. The method for hedging the time difference risk of insurance fund custody according to claim 1, characterized in that, The process of automatically generating a snapshot of the escrow fund flow after the end of the day's business, serving as the baseline data for reconciliation the following day, and pre-reconciling the escrow fund flow snapshot with the insurance company's list of funds to be deducted before the insurance company initiates batch deductions the following day, and performing unfreezing and transfer or discrepancy processing based on the pre-reconciliation results includes: S31. After the business is completed on the same day, the fund status of all virtual sub-accounts under the bank custody account is collected through a traversal. S32. Based on the collected results, automatically generate a snapshot of the deposited funds' flow and store the snapshot as the benchmark data for reconciliation the next day. S33. Obtain a list of payments to be deducted from the insurance company before the insurance company initiates the bulk deduction the following day; S34. Compare the stored snapshot of the escrow funds flow with the list of funds to be deducted item by item to complete the pre-reconciliation; S35. If the pre-reconciliation results are consistent, the matching consumption incentive funds will be unfrozen and the funds will be transferred according to the targeted payment instructions. If the pre-reconciliation results are inconsistent, the discrepancy will be handled, and the abnormal funds will be frozen for verification or returned to the original payment method.

4. The method for hedging the time difference risk of insurance fund custody according to claim 3, characterized in that, The step of comparing the stored snapshot of the escrow funds flow with the list of funds to be deducted item by item to complete the pre-reconciliation includes: S341. Construct a fund reconciliation matching network, initialize all data records in the fund flow snapshot and the list of pending deductions to an unmatched state, and set the number of overlapping matches of all data records to unmatched; S342. Select a central matching record in the reconciliation matching network, calculate the exclusion quality to centrality ratio of each data record, and include data records whose matching distance from the central matching record is less than the preset matching threshold into the same matching range. S343. Mark the center matching records that have been matched as matched, and select the record with the smallest ratio of exclusion quality to centrality from the remaining unmatched records as the center matching record of the next group. S344. Execute the matching record selection and data record matching division operations in the cycle execution center until the cash flow snapshot and all data records in the pending deduction list are matched and compared to complete the pre-reconciliation.

5. A method for hedging the time difference risk in insurance fund custody according to claim 4, characterized in that, The step of selecting a central matching record in the reconciliation matching network, calculating the exclusion quality to centrality ratio of each data record, and grouping data records whose matching distance from the central matching record is less than a preset matching threshold into the same matching range includes: S3421. In the established reconciliation matching network, select the initial central matching record based on the unique identifier of the data record as the benchmark record for the current batch of matching comparison; S3422. Determine the exclusion quality and centrality of each data record in the cash flow snapshot and the list of deductions respectively, and calculate the exclusion quality and centrality ratio of each data record accordingly. S3423. Using the field characteristics of the central matching record as a reference, calculate the matching distance between other data records and the central matching record one by one to quantify the degree of similarity between the data. S3424. Compare the calculated matching distance for each data record with a pre-set matching threshold to determine whether the matching conditions are met. S3425. Data records with a matching distance less than the preset matching threshold are grouped into the same matching range as the current center matching record to complete the matching division of the current batch of data.

6. The method for hedging the time difference risk of insurance fund custody according to claim 5, characterized in that, The step of calculating the matching distance between other data records and the central matching record, using the field features of the central matching record as a reference, to quantify the similarity between the data includes: S34231. Extract the core alignment features from the center matching record and other data records to be compared, and convert the core alignment features of the center matching record and other data records to be compared into standardized data that can be used for numerical calculation. S34232. Calculate the difference for each of the standardized data of the same type in the center matching record and other data records to be compared, and take the absolute value of the calculated difference to obtain the difference value of the corresponding single-class feature. S34233. Based on the preset weight values ​​of each core comparison feature, the difference values ​​of each single-class feature are weighted and the weighted result is squared to obtain the weighted difference square value of the corresponding single-class feature. S34234. Sum the weighted squared differences of all single-class features, and take the square root of the sum to obtain the matching distance between the center matching record and other data records to be compared, so as to quantify the similarity between the data.

7. The method for hedging the time difference risk of insurance fund custody according to claim 1, characterized in that, The establishment of a risk reserve pool, which automatically uses the risk reserve to make advance payments in the event of a unilateral account anomaly, ensures that the policyholder's rights are not affected. After the cause of the anomaly is identified and the liquidation is completed, corresponding account adjustments will be made to achieve automatic error recovery. S41. Based on the historical fund clearing scale and the frequency of daily anomalies, calculate and allocate the corresponding amount of risk reserve fund to complete the initial establishment and special account storage of the risk reserve fund pool. S42. Monitor fund transfers and reconciliation results in real time. When a one-sided account anomaly is detected, lock the corresponding accounting information and trigger the reserve activation instruction. S43. Based on the policy performance amount corresponding to the anomaly, the corresponding amount shall be drawn from the risk reserve pool for advance payment to maintain the normal effectiveness of the relevant policy rights; S44. After the advance payment is completed, initiate an anomaly check, trace the cause of the one-sided account, and record the anomaly type, the amount involved, and related data information simultaneously; S45. After the cause of the anomaly is identified and all parties have settled their accounts, the risk reserve pool shall be replenished or adjusted according to the settlement results in order to complete the automatic write-off and recovery of the erroneous accounts.

8. A method for hedging the time difference risk in insurance fund custody according to claim 7, characterized in that, After the advance payment is completed, an anomaly check is initiated to trace the cause of the one-sided account, and the anomaly type, the amount involved, and related data information are recorded simultaneously, including: S441. From the obtained abnormal feature terms related to unilateral accounts, select the core abnormal feature terms with the highest weight as the priority verification targets, and at the same time randomly select some terms from the remaining abnormal feature terms as supplementary verification targets. S442. Conduct a comprehensive review of the abnormal characteristics corresponding to priority verification objects and supplementary verification objects, expand the scope of abnormal related information, and improve the traceability clues for single-sided account abnormalities. S443. Classify and sort the priority verification objects, supplementary verification objects and their corresponding abnormal feature information to form structured abnormal feature groups; S444. Check whether the total number of individuals with abnormal characteristics that have been identified exceeds the preset maximum number of individuals to be checked. If it exceeds the maximum number, proceed with the screening process. If it does not exceed the maximum number, continue to supplement and improve the abnormal characteristic information. S445. Sort each abnormal feature group according to the verification priority, and select the abnormal feature individuals with the highest verification priority in each group as key verification objects. If the number of abnormal feature groups reaches the preset limit, start screening again from the first abnormal feature group until the number of key verification objects selected meets the preset requirements. S446. Check if the current iteration count of the anomaly check has reached the maximum iteration count. If it has, terminate the anomaly check, complete the tracing of the cause of the single-sided account anomaly and the recording of relevant information. If it has not reached the maximum iteration count, return to the anomaly feature expansion process and continue to improve the check process until all anomaly tracing and information recording are completed.

9. A method for hedging the time difference risk in insurance fund custody according to claim 8, characterized in that, The process of classifying and sorting priority verification targets, supplementary verification targets, and corresponding abnormal feature information to form structured abnormal feature groups includes: S4431. Based on the preset niche classification rules, extract the feature dimensions of priority verification objects and supplementary verification objects, and establish anomaly feature classification standards; S4432. According to the classification criteria, classify and divide various abnormal feature information according to the corresponding dimensions, and distinguish the feature sets corresponding to different abnormal manifestations. S4433. According to the degree of abnormal correlation and the importance of verification, sort the various abnormal features in the feature set internally, determine the verification order of features within the same category, and integrate the sorted abnormal features in a structured way to form a structured abnormal feature group.

10. A time difference risk hedging device for insurance fund custody, used to implement the time difference risk hedging method for insurance fund custody as described in any one of claims 1-9, characterized in that, The device includes: The virtual sub-account binding module is used to create a virtual sub-account for each user based on the bank's custody account and through the application programming interface of the bank's Class II account or virtual account, and bind the virtual sub-account to the user's corresponding policy account to achieve physical isolation of funds. The incentive fund freeze control module is used to call the bank's freeze interface when the consumer incentive fund enters the bank's custody account, freeze the consumer incentive fund to a virtual sub-account bound to the policy account, and generate a targeted payment instruction to restrict the consumer incentive fund to be transferred only to the designated insurance company account; The transaction snapshot and pre-reconciliation module is used to automatically generate a transaction snapshot of the escrow funds after the end of the day's business, which serves as the benchmark data for reconciliation the next day. Before the insurance company initiates batch deductions the next day, the transaction snapshot of the escrow funds is pre-reconciled with the insurance company's list of funds to be deducted, and the unfreezing and transfer or difference processing is performed based on the pre-reconciliation results. The risk reserve advance payment module is used to establish a risk reserve pool. When a one-sided account anomaly occurs, the risk reserve is automatically used to make an advance payment to ensure that the user's policy rights are not affected. After the cause of the anomaly is identified and the settlement is completed, the corresponding account adjustments are made to achieve automatic error recovery.

Images