A method, system, and medium for reliable data flow assessment and dynamic control.
By standardizing the processing of data transfer requests, calculating the reliability of evidence, and generating dynamic weights, the uncertainties and risks caused by missing or conflicting evidence in data transfer are resolved. This enables the automatic transformation from credible assessment to execution control, improving the accuracy and adaptability of the assessment.
Patent Information
- Application Number
- CN202611582407.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-10-13
- Publication Date
- 2026-11-10
AI Technical Summary
Existing technologies struggle to effectively address the uncertainties and risks arising from missing, expired, and conflicting evidence during data transfer, as well as the difficulty in directly translating assessment results into actionable control strategies.
By receiving data flow requests, parsing and standardizing the request context, obtaining multi-source evidence, calculating the reliability of the evidence and mapping it to intuitionistic fuzzy numbers, aggregating it with dynamic weights, generating a structured control strategy vector, and compiling it into instructions executable by the data flow execution gateway, updating the subject's trust status and historical behavioral risks.
It enables explicit handling of uncertain risks, improves the accuracy and adaptability of credible assessments, ensures that assessment results can be directly transformed into executable control strategies, reduces manual configuration errors and strategy conflicts, and adapts to dynamic adjustments in different workflow scenarios.
Smart Images

Figure CN123069443A_ABST
Abstract
Description
Technical Field
[0001] This application belongs to the field of trusted data flow technology, and more specifically, relates to a data flow trusted assessment and dynamic control method, system and medium. Background Technology
[0002] In the processes of government data sharing, enterprise data collaboration, authorized operation of public data, data trading, and cross-institutional joint computing, data transfer requests typically involve multiple types of information, including data objects, requesting entities, purposes of use, scope of authorization, target environment, transmission paths, and historical behavior. Data providers not only need to determine whether the requesting entity has the necessary formal permissions, but also whether the requested fields are the minimum required, whether the purpose aligns with the authorization, whether the target environment provides sufficient protection, and whether the post-transfer behavior is auditable.
[0003] Existing access control lists, role-based access control, and fixed rule methods typically determine requests as either allowed or denied directly. When evidence is missing, expired, or conflicting, these methods often rely on default values or manual processing, failing to differentiate between evidence supporting permission, evidence opposing permission, and evidence that cannot be determined temporarily. While ordinary continuous risk scoring can generate scores, it usually compresses unknown states into a single score, failing to retain the degree of uncertainty.
[0004] Intuitive fuzzy sets describe three states—support, opposition, and uncertainty—using membership, non-membership, and hesitation, making them suitable for handling incomplete information in data flow. However, simply providing subjective intuitive fuzzy numbers for a few indicators and then weighting and aggregating them still leads to the following problems: there are no reproducible rules for transforming original evidence into intuitive fuzzy numbers; there is no quantification method when multiple evidence sources conflict; weights cannot change with data security level, flow scenario, and degree of evidence uncertainty; evaluation results cannot be directly converted into control rules that can be executed by database agents or interface gateways; and abnormal events after the flow cannot update the evaluation status of subsequent requests.
[0005] Furthermore, there is an engineering gap between the generation and execution of control policies. Risk levels are usually output as textual conclusions, but the execution gateway still requires manual configuration of field filtering, desensitization, encryption, watermarking, access periods, call counts, and auditing rules, which can easily lead to policy translation errors, rule conflicts, configuration delays, and mismatches between requests and policies.
[0006] Therefore, a complete computer implementation method is needed. First, multi-source evidence is standardized into computable indicator evidence records. Then, the reliability of the evidence and the intuition fuzzy number are calculated. Subsequently, dynamic weight aggregation and hierarchical decision-making are performed, and the decision is converted into a structured control strategy. After verification, it is atomically distributed to the data flow execution gateway. Finally, the trust status of the subject and historical behavioral risks are updated using audit events. Summary of the Invention
[0007] This invention provides a data flow reliability assessment and dynamic control method, system and medium, which aims to solve the technical problems that existing technologies have difficulty in characterizing the uncertain risks of missing, expired and conflicting evidence, and that assessment results are difficult to directly convert into executable control strategies.
[0008] In a first aspect, the present invention provides a method for data flow reliability assessment and dynamic control, comprising the following steps: S1. Receive a data transfer request, parse and normalize the data transfer request to obtain a request context; the request context includes at least a data identifier, request subject, transfer purpose, target node, operation type and authorization credential; S2. Based on the request context, obtain corresponding evidence data or evidence unavailability status from multiple preset evidence sources. For each preset evaluation index, normalize and fuse the multi-source evidence corresponding to the evaluation index. Calculate the evidence reliability of the evaluation index based on evidence completeness, evidence timeliness, source credibility, and evidence conflict degree, and generate indicator evidence records corresponding to each evaluation index. S3. For each evaluation indicator, based on the normalized support value and the reliability of the evidence, the evaluation indicator is mapped to an intuitionistic fuzzy number; the intuitionistic fuzzy number includes membership degree, non-membership degree and hesitation degree, where membership degree represents the degree of support of existing evidence for allowing data flow, non-membership degree represents the degree of opposition of existing evidence to allowing data flow, and hesitation degree represents the degree of uncertainty caused by insufficient evidence. S4. Obtain the basic weights of each evaluation indicator, and calculate the dynamic weights of each evaluation indicator by combining the correction coefficients corresponding to data security level, flow scenario, hesitation degree and abnormal risk; perform aggregation operation on the intuitive fuzzy numbers of all evaluation indicators based on the dynamic weights to obtain the overall credibility score and deterministic score. S5. According to the preset priority decision rules, combined with the credibility score, certainty score, key indicator veto threshold and mandatory control rules, determine a unique flow decision level and generate a control strategy vector corresponding to the flow decision level; the control strategy vector includes at least data range parameters, data transformation parameters, channel security parameters, usage constraint parameters and audit parameters; S6. The control strategy vector is compiled into a strategy instruction that can be directly executed by the data flow execution gateway, and sent to the data flow execution gateway. The data flow execution gateway performs token verification, field filtering, data transformation, access metering and abnormal blocking operations on the data flow process. S7. Obtain the audit events generated by the data flow execution gateway, and update the trust status and historical behavior risk indicators of the corresponding request subject according to the audit events for trust assessment of subsequent data flow requests.
[0009] As one possible implementation of the present invention, the generation of indicator evidence records corresponding to each evaluation indicator includes the following steps: For each evaluation indicator, the corresponding original evidence tuples are collected from the preset evidence source set, and the original evidence tuples are deduplicated, validated, and marked with status. The original evidence values are normalized according to the direction of the evaluation indicators to obtain single-source support values. The indicator direction is used to characterize the relationship between the original value of the indicator and the normalized support value that supports the flow of data, and is divided into positive, negative, and moderate types. Among them, the higher the original value of the positive evaluation indicator, the higher the corresponding normalized support value; the higher the original risk value of the negative evaluation indicator, the lower the corresponding normalized support value; the moderate type indicator has a preset optimal value range and allowable value boundary. When the original value is within the optimal value range, the normalized support value is the maximum value; when the original value deviates from the optimal value range but does not exceed the allowable value boundary, the normalized support value is the minimum value. The completeness of evidence is calculated based on the validity of all required evidence fields of the evaluation indicators; the timeliness of evidence is calculated based on the generation time and time-lapse rules of each piece of evidence; the credibility of the source is calculated based on the credibility attributes of each evidence source; and the degree of evidence conflict is calculated based on the degree of difference between multiple single-source support values. Multiply the values of evidence completeness, evidence timeliness, and source credibility together, and then multiply by the difference between the value 1 and the degree of evidence conflict to obtain the evidence reliability of the evaluation index. The support values from each single source are weighted and fused according to their source weights to obtain the multi-source fused support value of the evaluation index. The multi-source fused support value, evidence completeness, evidence timeliness, source credibility, evidence conflict degree, and evidence reliability are stored in the indicator evidence record corresponding to the evaluation index.
[0010] As one possible implementation of the present invention, mapping the evaluation index to an intuitionistic fuzzy number includes the following steps: The product of the evidence reliability and the normalized support value is used as the membership degree, the product of the evidence reliability and the normalized opposition value is used as the non-membership degree, and the difference between the value 1 and the evidence reliability is used as the hesitation degree; where the normalized opposition value is the difference between the value 1 and the normalized support value. The sum of membership degree, non-membership degree, and hesitation degree is 1, and the mean of the three is within a closed interval between 0 and 1.
[0011] As one possible implementation of the present invention, the calculation process of the dynamic weight includes the following steps: Obtain the basic weights of each evaluation indicator; the basic weights are preset by safety experts, or obtained by weighting subjective weights and objective weights according to preset combination coefficients; wherein the subjective weights are determined according to pairwise comparison rules or regulatory requirements, and the objective weights are determined according to the dispersion or information entropy of historical evaluation samples; The unnormalized dynamic weight of each evaluation indicator is obtained by multiplying its base weight by the corresponding flow scenario correction coefficient, data security level correction coefficient, hesitation correction coefficient, and abnormal risk correction coefficient. The flow scenario correction coefficient is obtained by matching a preset value with the flow scenario type described in the current request; the data security level correction coefficient is obtained by matching a preset value with the data security level corresponding to the current request; the hesitation correction coefficient is obtained by adding the product of the value 1 and the preset non-negative adjustment coefficient and the hesitation of the corresponding evaluation index, and its value increases as the hesitation of the corresponding evaluation index increases; the abnormal risk correction coefficient is calculated based on the historical behavioral risk of the requesting subject and the preset event risk amplification coefficient. Divide the unnormalized dynamic weight of each evaluation indicator by the sum of the unnormalized dynamic weights of all evaluation indicators to obtain the final normalized dynamic weight of each evaluation indicator. The sum of the normalized dynamic weights of all evaluation indicators is 1.
[0012] As a possible implementation of the present invention, the aggregation operation on the intuitionistic fuzzy number includes the following steps: The evaluation indicators are divided into multiple dimensions according to data object, request subject, usage authorization, execution environment, flow path and audit history. Within each dimension, the dynamic weights of all evaluation indicators within that dimension are re-normalized. Based on the normalized dynamic weights within the dimension, the intuitive fuzzy numbers of each evaluation indicator within the dimension are aggregated at the first level of intuitive fuzziness to obtain the dimension intuitive fuzzy number of the corresponding dimension. Dimension weights are assigned to all dimensions, and the weights of all dimensions are normalized. The intuitionistic fuzzy weighted average operator is used to perform a second-level aggregation of the intuitionistic fuzzy numbers of all dimensions to obtain the comprehensive intuitionistic fuzzy number. The confidence score is calculated based on the difference between the membership degree and non-membership degree of the comprehensive intuitionistic fuzzy number, and the deterministic score is calculated based on the sum of the membership degree and non-membership degree of the comprehensive intuitionistic fuzzy number.
[0013] As one possible implementation of the present invention, the preset priority decision rules are arranged in descending order of priority as follows: First priority: If a mandatory prohibition rule is met or the non-membership degree of a key indicator reaches the key indicator veto threshold, the circulation will be prohibited or the circulation will be manually reviewed. Second priority: If the original data must not leave the domain but controlled computation is allowed, it is determined to be a sandbox computation flow; Third priority: If the certainty score is lower than the preset certainty threshold and the non-membership degree of the key indicator reaches the key indicator rejection threshold, or if key evidence is missing or conflicting, it is determined to be manually reviewed and processed. Fourth priority: If the credibility score is not lower than the preset allowable threshold, the certainty score is not lower than the preset certainty threshold, and the mandatory protection rule is not hit, then the circulation is allowed. Fifth priority: If a request can be released under controlled conditions with additional protection measures, including at least a trust score that is not lower than the preset prohibition threshold and lower than the preset allowance threshold, or if it hits the mandatory protection rule, it is determined to be an enhanced protection flow. Sixth priority: If the credibility score is lower than the preset prohibition threshold, it is determined that circulation is prohibited; When making decisions, the rules are judged in order of priority from top to bottom. Once the highest priority rule is hit, the corresponding unique decision level is output and the lower priority rules are no longer executed.
[0014] As one possible implementation of the present invention, the allow threshold, prohibit threshold, and deterministic threshold are all adaptive thresholds; The allowable threshold is obtained based on the following steps: Based on the preset basic allowable threshold, the allowable threshold is obtained by adding the trust threshold increment corresponding to the data security level of the current request and the trust threshold increment corresponding to the operation type of the current request; The prohibition threshold is obtained by directly using a preset basic prohibition threshold, or by adding the trusted threshold increment corresponding to the data security level of the current request and the trusted threshold increment corresponding to the operation type of the current request to the preset basic prohibition threshold. The deterministic threshold is obtained based on the following steps: Based on the preset basic deterministic threshold, a deterministic threshold increment corresponding to the data security level of the current request and a deterministic threshold increment corresponding to the operation type of the current request are superimposed to obtain the deterministic threshold.
[0015] As a possible implementation of the present invention, when generating the control strategy vector, if there are conflicts between the control measures generated by multiple rules, they are merged in order of protection level from high to low; control measures with higher protection level are applied first, wherein prohibition has higher priority than permission, shorter validity period has higher priority than longer validity period, smaller field set has higher priority than larger field set, and lower call limit has higher priority than higher call limit; conflicts that cannot be compared according to protection level are transferred to manual review.
[0016] As one possible implementation of the present invention, the step of compiling the control policy vector into policy instructions includes: The control policy vector is converted into a policy intermediate representation that is independent of the gateway type. The policy intermediate representation includes a unique policy identifier, a request identifier, a request subject, a data identifier, a flow purpose, a policy effect, an effective condition, an execution obligation, an effective time, and a time limit. Based on the type of data flow execution gateway, the intermediate representation of the strategy is compiled into specific strategy instructions that can be executed by the corresponding data flow execution gateway. The specific strategy instructions include at least one of SQL query rewriting rules, field filtering rules, data desensitization parameters, access token declarations, access metering rules, and audit templates. The compiled policy instructions are statically validated. If the validation passes, the policy is sent to the data flow execution gateway in an atomic switching manner to take effect. If the validation fails, the previous version of the policy is retained and the exception is recorded. The static validation includes whether the field exists, whether the filter expression can be parsed, whether the algorithm identifier is supported, whether the validity period is valid, whether the dependent services are available, and whether there are irresolvable conflicts between the rules.
[0017] As one possible implementation of the present invention, the execution steps of the data flow execution gateway include: The access token is bound to a unique policy identifier, request identifier, request subject, data identifier, transfer purpose, operation type, validity period and random number, and the access token is protected for integrity. Before data transfer, the validity and matching of the access token are verified. During the transfer, field filtering, data transformation, access counting, rate limiting and exception blocking are performed. After the transfer is completed, the summary information of the actual transferred data and the execution results are recorded. If the security services that the policy depends on are unavailable, the flow request should be blocked according to the failure shutdown principle, or downgraded to a policy with higher protection strength if the configuration allows it. It must not be downgraded to unprotected access.
[0018] As one possible implementation of the present invention, the step of updating the trust status and historical behavior risk indicators of the corresponding requesting subject according to the audit event includes: The collected audit events are classified and processed. Based on the event type, severity level, impact data level, duration, and handling result of the audit event, a scalar behavioral risk quantity with a value in the closed interval of 0 to 1 is calculated. Obtain the subject's trustworthiness status before the current assessment cycle update and the preset historical retention coefficient. Use the historical retention coefficient as the weight to weight the subject's trustworthiness status before the update. Use the difference between 1 and the historical retention coefficient as the weight to weight the difference between 1 and the behavioral risk amount. Sum the results of the two weightings and limit them to a closed interval between 0 and 1 to obtain the subject's trustworthiness status after the next assessment cycle update. Multiply the preset non-negative event risk amplification coefficient by the behavioral risk quantity, and then add the product to the value 1 to obtain the abnormal risk correction coefficient corresponding to the historical behavioral risk indicator. Furthermore, the maximum change in the subject's trust status in a single period, the event deduplication window, and the number of consecutive compliance steps required to restore the normal state are set, limiting the change in the subject's trust status within a single evaluation period to not exceed the maximum change in a single period.
[0019] As one possible implementation of the present invention, the normalization processing of the data transfer request includes: The request context is normalized and serialized, and a request digest is calculated. The unique request identifier, request digest, and receipt time are stored in the request registration table. If the same unique request identifier corresponds to different request digests, it is determined that the request content is abnormal. If the same request subject, data identifier, transfer destination and random number appear repeatedly within the valid window, the original decision or decision duplicate request is reused according to the idempotency rule to prevent request replay.
[0020] As one possible implementation of the present invention, the method further includes decision tracing: The decision evidence record is formed by storing the following together: the set of indicator evidence records for this request, the intuitive fuzzy number of each evaluation indicator, the dynamic weight, the credibility score, the deterministic score, the hit mandatory and veto rules, the final control strategy vector, the compiled decision representation, the summary of related audit events, and the configuration version used in this decision. A completeness summary is calculated for the decision evidence record, and the completeness summary is associated with the unique identifier of this request to prove the correspondence between evaluation inputs, decision results, execution strategies and feedback events, forming a traceable decision chain.
[0021] Secondly, the present invention provides a data flow reliability assessment and dynamic control system, including a processor and a memory; the memory stores computer-executable instructions, and when the processor executes the computer-executable instructions in the memory, it implements a data flow reliability assessment and dynamic control method as described in the present invention.
[0022] Thirdly, the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements a data flow reliability assessment and dynamic control method as described in the present invention.
[0023] The beneficial effects of this invention include: 1. This invention quantifies the completeness, timeliness, source credibility, and conflict of multi-source evidence into evidence reliability. It then maps the original indicators to an intuitive fuzzy number containing membership, non-membership, and hesitation through evidence reliability. This transforms uncertainties such as missing evidence, unreliable sources, and multi-source conflicts into calculable hesitation, explicitly preserving unknown risks. Unlike traditional methods that compress unknown states into a single risk and easily misjudge insufficient evidence as low or high risk, this invention accurately distinguishes between three states: evidence supporting permission for transfer, evidence opposing permission for transfer, and insufficient evidence making a judgment impossible. This improves the accuracy and rationality of credibility assessment under incomplete information.
[0024] 2. This invention employs a dynamic weight calculation method that combines basic weights with four types of correction coefficients: circulation scenario, data security level, hesitation level, and anomaly risk. This method can automatically adjust indicator weights based on differences in circulation scenarios, the level of data sensitivity, the degree of evidence uncertainty, and the risk of historical behavior by the subject. This solves the problem that fixed weights cannot adapt to different circulation scenarios such as government sharing, authorized operation, data trading, joint computing, and cross-domain exchange. Simultaneously, it adopts a two-level hierarchical aggregation method to avoid unreasonable weight amplification in a single dimension due to an excessive number of indicators, ensuring the weight balance of multi-dimensional evaluation and significantly improving the adaptability and scientific rigor of credible evaluation in different application scenarios.
[0025] 3. This invention transforms the trust assessment results into a structured control policy vector that includes data range, data transformation, channel security, usage constraints, and audit requirements. This vector is then converted into field filtering, de-identification encryption, access rate limiting, and audit logging instructions that can be directly executed by the data flow execution gateway through a policy compilation module. Combined with static verification and atomic release mechanisms, this achieves an automated control link from trust assessment to gateway execution. It solves the problems of traditional solutions where risk assessment only outputs textual conclusions, requires manual configuration of gateway rules which easily leads to policy translation errors, rule conflicts, and configuration lags behind requested policy mismatches. This significantly improves the accuracy and efficiency of control policy execution.
[0026] 4. This invention calculates behavioral risk through audit events generated by the data flow execution gateway, uses an exponential smoothing mechanism to update the trust status and abnormal risk correction coefficient of the requesting subject, and applies the updated results to the dynamic weight calculation and trust assessment of subsequent flow requests. At the same time, it sets anti-oscillation mechanisms such as the maximum change in a single period and the event deduplication window, so that the trust assessment model can be continuously iterated and adjusted according to the actual flow behavior of the subject, thus solving the problem that static assessment cannot adapt to the dynamic changes in subject behavior. Attached Figure Description
[0027] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0028] Figure 1 This is a flowchart of the data flow reliability assessment and dynamic control method provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the data flow reliability assessment and dynamic control system structure provided in an embodiment of the present invention.
[0029] Figure 3 The flowchart for audit feedback and parameter update provided in the embodiments of the present invention. Detailed Implementation
[0030] To make the technical problems, technical solutions, and beneficial effects to be solved by this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and are not intended to limit the scope of this application.
[0031] Example 1 See appendix Figure 1 and Figure 2 As shown, this embodiment provides a data flow trust assessment and dynamic control method, which runs on a data flow management platform. The data flow management platform is communicatively connected to the data source node, data demand node, identity authorization system, data catalog system, security assessment system, data flow execution gateway, and audit module.
[0032] The data flow management platform can be deployed independently or as a trusted decision service for a data sharing and exchange platform or a data trading platform; the data flow execution gateway can be deployed as an API gateway plugin, a database agent, a file exchange agent, a cross-domain exchange gateway, or a computing task agent.
[0033] Each functional module within the platform uses a unique request identifier (rid) as its primary key. Modules can exchange structured records through synchronization interfaces, message queues, or shared data. All calculation processes retain original evidence, intermediate results, configuration versions, and hit rules to ensure end-to-end traceability. The functional modules within the platform include: request parsing module, evidence collection module, indicator calculation module, intuitionistic fuzzy mapping module, dynamic weighting module, trust assessment module, decision and strategy generation module, strategy compilation module, and audit feedback module.
[0034] In this embodiment, the core data architecture is designed as shown in Table 1 below: Table 1 Core Data Architecture Design ; Referring to Table 1 above, the request context R describes a data flow request to be evaluated. Here, `rid` is a unique request identifier used to associate the evidence, evaluation results, strategies, and audit events generated by this request; `did` represents the data object identifier; `uid` represents the request subject identifier; `purpose` is the flow purpose normalized by the purpose directory; `target` is the target node for receiving data, performing calculations, or saving results; `op` is the operation type such as query, download, sharing, API call, statistical analysis, federated computation, model training, or file exchange; `auth` is the citation identifier or summary of authorization credentials, approval records, or contract authorization; `term` is the authorization or access period; `output` is the output format such as original details, anonymized details, statistical results, and model parameters; and `scene` is the flow scenario such as government sharing, authorized operation, data transaction, federated computation, or cross-domain exchange. The request context may also include a random number `nonce` and a request time `request_time` to prevent request replay.
[0035] Evidence tuple Indicators The corresponding k-th original evidence, where i is the evaluation index number and k is the evidence source number within that index, taking values k=1, 2, …, , Evaluation indicators The current number of available evidence sources; `source` is the identifier of the evidence source system or responsible entity; `source_index` is assigned by the platform after verifying, deduplicating, and sorting available evidence for the same indicator, and its value is equal to the evidence source sequence number `k`; `value` is the original value returned by the evidence source interface; `timestamp` is the time the evidence was generated or collected; `valid_to` is the expiration time of the evidence's validity; `version` is the version of the evidence or source interface; `signature` is the digital signature, message authentication code, or digest verification value, used to verify the integrity of the evidence; `status` is the evidence status, which can be valid, expired, revoked, conflicting, or unavailable. When multiple sources provide different values for the same indicator, each evidence tuple is retained separately, and later evidence does not directly overwrite earlier evidence, so as to facilitate subsequent calculation of source credibility and evidence conflict degree.
[0036] Indicator Configuration It is created by the configuration management module before processing specific requests, and its effective version is identified by config_version; where id and name are the indicator code and Chinese name, respectively; orientation specifies whether the indicator is positive, negative, or moderate; type specifies whether the original value is numeric, enumerated, boolean, or text. and To normalize the lower and upper bounds; The base weights are: critical (indicator of criticality), missing_rule (handling rule for missing, expired, withdrawn, or unavailable evidence), and source_set (set of allowed values for evaluation metrics). The system provides a set of source identifiers for the evidence, each element corresponding to a system or responsible entity identifier in the evidence source registration information; `source_weight_rule` is a set of source weight rules, each rule item including at least the source identifier, source weight, evidence version priority, source legal validity coefficient, and rule version, for evaluation indicators. The source weight of the corresponding k-th original evidence can be denoted as: The base source weight is a non-negative number; sources without configured source weight rules are handled according to source credibility or average weight rules; required_field_set is the evaluation metric. The set of evidence fields that must be checked, where each element of the set must contain at least the field name, field importance, and validity determination rules. Both `source_set` and `required_field_set` are configuration fields and are not evidence records generated in this request.
[0037] Indicator Evidence Record Used to store evaluation indicators The standardized results before entering trusted computing; where id is the indicator code; orientation is the indicator direction, which can be positive, negative or moderate; type is the indicator data type, which can be numeric, enumerated, boolean or text. This represents the number of available sources of evidence for this indicator; Calculate the number of required evidence fields for this indicator; and These are the normalized lower and upper bounds, respectively; These are the normalized support values; For the sake of completeness of evidence; To consider the timeliness of evidence; To assess the credibility of the sources; For the degree of conflict of multiple sources of evidence; For the reason , , and The reliability of evidence jointly determined.
[0038] Intuitive Fuzzy Numbers By membership degree Non-membership degree and hesitation composition; This indicates the degree to which existing evidence supports allowing the data transfer. This indicates the degree of opposition to allowing the data transfer based on existing evidence. This indicates the degree to which a definitive judgment cannot be made due to missing, expired, uncertain, or conflicting evidence. All three factors are located in the range [0, 1] and satisfy the following condition: + + =1; With the corresponding It is necessary to maintain the connection with the original evidence tuples to support decision interpretation and subsequent review.
[0039] The control strategy vector P is used to convert the trusted assessment results into executable control requirements; where D is the flow decision level; F is the data range parameter, used to limit allowed fields, masked fields, record filtering conditions, maximum return volume, and statistical granularity; V is the data transformation parameter, used to specify de-identification, generalization, perturbation, encryption, hashing, tokenization, or watermarking; N is the channel and execution environment security parameter, used to specify two-way authentication, transmission encryption, interface signature, trusted execution environment, or cross-domain verification; Y is the usage constraint parameter, used to limit validity period, access count, concurrency, interface rate, download and forwarding permissions, and usage binding; and G is the audit parameter, used to specify log fields, sampling rate, alarm level, and retention period.
[0040] The policy intermediate representation (IR) is generated by the policy compilation module based on P and R. Here, policy_id is the policy identifier generated by the policy compilation module; rid, uid, did, and purpose are copied from R and used to bind the policy to a specific request, subject, data object, and purpose; effect is derived from the decision level D in P, representing the policy effect such as allow, restrict, review, or prohibit; conditions are compiled from F, N, Y, and the pre-hit rules, representing the conditions that must be met for the policy to take effect; obligations are compiled from V, G, and other mandatory protection measures, representing the actions that must be executed synchronously when allowing access; valid_from is the policy release or approval effective time; valid_to is the evidence validity deadline, whose value can be the earliest of the request deadline, authorization deadline, and policy restriction deadline; version is generated by the policy release module and used to distinguish between candidate versions and the effective version with the same policy_id.
[0041] Audit event log This is used to record the actual behavior generated at the t-th evaluation cycle or execution point; where eid is the unique identifier of the audit event; rid is the unique identifier of the request; policy_id is the policy identifier associated with the audit event; event_type is the event type such as normal fulfillment, field unauthorization, usage deviation, illegal download, token reuse, or policy execution failure; severity is the event severity level; time is the event occurrence time; result is the result such as gateway permission, restriction, blocking, or execution failure; and digest is the event summary formed by information such as the actual returned fields, number of records, data summary, and hit rules. This indicates a structured event log, which will be discussed later. The two terms, representing the behavioral risk calculated from the event, have different meanings.
[0042] The Decision Evidence Record (DR) is used to form a traceable decision chain. Specifically, `indicator_records` stores the set of indicator evidence records used in this request; `ifs_values` stores the intuitionistic fuzzy numbers for each indicator; `dynamic_weights` stores the dynamic weights; `S` and `H` store the confidence score and deterministic score, respectively; `rule_hits` stores the hit mandatory and veto rules; `P` stores the final control policy vector; `policy_id` stores the compiled policy identifier; `audit_digest` stores the summary of associated audit events; and `config_version` stores the configuration versions of the indicators, weights, thresholds, and rules used in this decision. The underlined names above are all computer structured record field names and do not represent mathematical subscripts. The platform calculates an integrity summary for the DR and associates it with `rid` to demonstrate the correspondence between evaluation inputs, decision results, execution policies, and feedback events.
[0043] The above objects form the following source chain: provided by the configuration management module. The request parsing module generates R; the evidence collection module uses... .source_set generation The indicator calculation module is based on .required_field_set and generate The intuitive fuzzy mapping module is composed of generate The decision and strategy generation module generates P; the strategy compilation module generates IR from R and P; the execution gateway executes and generates based on IR. DR summarizes and solidifies the above inputs, calculation results, strategies, and audit correlations. One R corresponds to n evaluation indicators; one Define one Allowed sources and required fields; a One And can be associated indivual Each examine Each module must not transmit only the final score while discarding the original evidence, evidence quality parameters, configuration version, or hit rules.
[0044] In this embodiment, the data flow reliability assessment and dynamic control method includes the following steps: S1. After receiving a data transfer request, the data transfer management platform verifies the field type, length, character set, and enumeration value, unifies the time to a preset time zone, maps the subject identifier and data identifier to unique identifiers within the platform, maps the transfer purpose to the usage directory code, and maps the operation type to one of the following: query, download, share, interface call, statistical analysis, joint calculation, modeling training, or file exchange.
[0045] The data flow management platform normalizes and serializes the request context according to a preset field order, and calculates the request digest. and put rid, The receiving time is written into the request registration table; if the same RID corresponds to different... If the same uid, did, purpose, and random number appear repeatedly within the valid window, the original decision is reused or the repeated request is rejected according to the idempotency rule to avoid repeated access and replay.
[0046] S2. The evidence collection module reads the current config_version. The `.source_set` function parses the connection address, interface protocol, authentication method, and source trustworthiness configuration from the evidence source registration information based on the source identifier in the set. It then calls systems such as data directory, identity authentication, role permissions, contract approval, business matters, security assessment, network domain management, interface logs, watermark tracing, and violation handling. For evaluation metrics... Only when the actual source identifier `source` belongs to `source_set` and passes the availability check will an evidence tuple be generated. After deduplication and sorting, the index k is written to source_index, and the number of available tuples is written to... .evidence_count.
[0047] Preprocessing includes: deduplication by source identifier, business primary key, and version number; verification of evidence signature or digest; determination of validity based on valid_to; marking revoked evidence as unavailable; retaining contradictory valid evidence without directly overwriting it; recording connection failures, timeouts, and null values; for critical evidence sources, a maximum waiting time can be set, after which evaluation will continue according to the missing evidence rules or be transferred to manual review; configuration for each indicator. The source_set includes one or more preset evidence source identifiers.
[0048] In this embodiment, the original evidence values are normalized according to the direction of the evaluation index, that is, the original evidence values are converted into single-source support values in the interval [0,1]. ,The indicators are categorized into three types: positive, negative, and moderate, as exemplified in Table 2 below: Table 2 Design of Evaluation Index System ; The aforementioned positive or negative directions refer to the relationship between the original value of the indicator and the normalized value that supports its transfer. A higher original value for a positive indicator typically corresponds to a higher normalized value; conversely, a higher original risk value for a negative indicator corresponds to a lower normalized support value. The direction of the indicator does not change the meaning of the original evidence. All direction conversion processes preserve the orientation and mapping version to avoid misinterpreting a higher risk value as a higher support value. Each indicator configuration is represented as follows: In this embodiment, for positive indicators, the allowable boundary is... and Normalized support values For inverse numerical indicators ;when If the indicator configuration has no valid range, the platform will mark the indicator as having an abnormal configuration and stop using that configuration version. To obtain evidence tuples The original value obtained from the value field and parsed by data type; in this embodiment Indicates when When less than 0, take 0; when Take 1 if the value is greater than 1, otherwise take 1. .
[0049] For a moderate index, let the optimal range be... ,in For the optimal lower bound, As the optimal upper bound, the allowable boundary is and .when lie in hour ;when hour ;when hour Other cases This method can be used for metrics such as call frequency, data volume, or retention period, which are not necessarily better the larger they are.
[0050] For enumerated indicators, the enumerated values are mapped to [0, 1] through a versioned mapping table; for Boolean indicators, the value is 1 if the condition is met and 0 if not; for textual evidence, support values can be generated through keyword rules, semantic similarity models, or manual review, but the rule version, model version, or review record must be saved at the same time; when the model output exceeds [0, 1], clip is executed.
[0051] In this embodiment, the integrity of evidence Used to represent indicators Are all required fields complete and valid? The required fields are determined by the ratio of the number of valid required fields to the total number of required fields. When the importance of each required field is different, the following formula is used for calculation: ; in, Evaluation indicators The The importance of each required field Evaluation indicators Number of required evidence fields This is the validity indicator for the field; when the field is valid... =1, when the field is missing, expired, revoked, or validation fails. =0.
[0052] Time limit of evidence Used to represent indicators Does the evidence presented have a time limit? The value can be calculated using the following formula: ; in, As an indicator The number of available sources of evidence, For evidence tuples Source weight, ; The time limit for single-source evidence is calculated using the following formula: ; in, For current time and indicators The non-negative time difference between the timestamps recorded by the k-th original evidence. As an indicator The time-degradation parameter, >0 and with Use the same time unit; fully credible evidence must be maintained before `valid_to`, or it can be set to 1 within the validity period and decay exponentially after expiration; for critical evidence such as authorization certificates or security assessments, it can also be set to... Set as all key evidence The minimum value in the middle is used to prevent expired key evidence from being compensated by other new evidence.
[0053] set up For evidence tuples The source credibility is determined based on the source authentication method, the data responsible party, historical availability, signature verification results, and anti-tampering capabilities, with a value range of [0, 1]. The credibility of the combined sources can be calculated using the following formula: ; After each source of evidence completes the index value normalization (i.e., the normalization operation based on index direction described earlier), a single-source support value is obtained. When evaluation indicators Number of available sources of evidence When the value is ≥1 and the sum of the weights of each source is greater than 0, the indicator... The multi-source fusion support value is calculated according to the following formula: ; when At that time, the degree of conflict of evidence can be determined according to... Calculate, where the maximum and minimum values are both in Therefore, it is selected from all single-source support values; The closer the conclusions from different sources, the better. The smaller.
[0054] when =0 or When, it indicates the indicator Without available evidence, the multi-source weighted division method is not performed; instead, the order is directly given. And trigger the evidence missing flag, and in the intuitive fuzzy mapping stage, it causes ;when At that time, there are no comparable multi-source conflicts, making However, the reliability of a single piece of evidence is still determined based on its completeness, timeliness, and the credibility of its source.
[0055] In obtaining , , and After that, indicators The reliability of evidence is calculated according to the following formula: ; , and The higher The higher; The higher The lower the indicator; Key metrics identified by critical and If the value is 0 or below the critical reliability threshold, it simultaneously triggers supplementary materials, manual review, or rules to prohibit automatic release; thus, missing, expired, and conflicting information reduces... It enters the subsequent hesitation level, instead of being directly filled with a high-risk or low-risk value.
[0056] S3. For each evaluation indicator Based on its multi-source fusion support value and the reliability of evidence Constructing intuitive fuzzy numbers Membership degree This indicates the degree of reliable evidence supporting the permissibility of the transfer; non-membership degree. This indicates the degree of credible evidence opposing the permission to transfer; degree of hesitation. This indicates that the evidence is insufficient to support a definitive judgment; therefore ;when hour ;when hour .
[0057] In this embodiment, to prevent a single extreme indicator from being smoothed out, a rejection threshold can be set for key reverse indicators. When its non-membership degree Not less than When a mandatory prohibition rule is met, the decision module should at least restrict the request to manual review, sandbox calculation, or prohibit flow, rather than directly allowing it due to other high indicators.
[0058] S4. For the i-th indicator among n evaluation indicators Basic weight It can be configured by security experts or by subjective weighting. and objective weight Combining to obtain ,in Subjective weights can be determined by pairwise comparisons or regulatory requirements; objective weights can be determined by the dispersion or information entropy of historical samples; all basic weights are positive numbers.
[0059] index The unnormalized dynamic weights are ,in Correction coefficients for circulation scenarios. This is a correction factor for the data security level. This is a correction factor for hesitation. This is an abnormal risk correction coefficient. The non-negative adjustment coefficient is used; the normalized dynamic weights are calculated according to... Calculate, where j is the index summation subscript, therefore each All are non-negative and .
[0060] In this embodiment, to avoid an unreasonablely high weight due to a large number of indicators in a certain dimension, a two-level hierarchical aggregation method is adopted, as follows: First, aggregate the data within each dimension: data object, request subject, usage authorization, execution environment, workflow path, and audit history. Then, perform a second-level aggregation on the results of each dimension. At each level, only the weights of the objects participating in the aggregation at that level are re-normalized to prevent a certain dimension from receiving a higher weight simply because it contains a larger number of metrics. The following example uses the aggregation of n metrics at once. The same calculation method is used at each level during hierarchical aggregation, as detailed below: When using the intuitionistic fuzzy weighted average operator, the overall membership degree is: The overall non-membership degree is The overall hesitation level is , The above refers to the normalized dynamic weights. To avoid numerical underflow caused by multiplying a large number of indicators, an equivalent logarithmic form can be used: , When the logarithmic independent variable is 0, the minimum positive number is preset. replace.
[0061] Credible score The value range is [-1, 1]; Deterministic score The value range is [0, 1]; S represents the net difference between support and opposition, and H represents the degree to which the current evidence can form a judgment; the two are used together to avoid high-hesitation requests being automatically allowed based solely on S.
[0062] S5. Let Based on the allowed threshold, Based on the prohibition threshold, The basic deterministic threshold is defined; SL represents the current data security level, and OT represents the current operation type; the allowable threshold used for the current request is... According to Determined, current deterministic threshold According to Confirmed, among which and These are the trusted threshold increment and the deterministic threshold increment corresponding to the data security level, respectively. and These represent the corresponding increments for the operation type; the current prohibited threshold. Desirable Alternatively, it can be modified according to the security level and operation type using the same modification method as the basic allowable threshold.
[0063] In this embodiment, based on the aforementioned threshold, decisions are made sequentially from highest to lowest priority. Once the highest priority rule is matched, a unique flow decision level is output, and lower priority decisions are no longer executed. The specific decision rules are shown in Table 3 below: Table 3 Priority Decision Rule Design ; Based on Table 3 above, we can see that: First priority: If a mandatory prohibition rule is hit, or the non-membership degree of a key indicator reaches the key indicator veto threshold, then the circulation is prohibited or the circulation is manually reviewed. Second priority: If the original data must not leave the domain but controlled computation is allowed, it is determined to be a sandbox computation flow; Third priority: If the certainty score is lower than the certainty threshold, the non-membership degree of the key indicator reaches the veto threshold, or the key evidence is missing or conflicting, it will be judged as manual review and processing. Fourth priority: If the credibility score is not lower than the allowable threshold, the certainty score is not lower than the certainty threshold, and the mandatory protection rule is not hit, then the transfer is allowed. Fifth priority: If a request can be released under controlled conditions with additional protection measures, including at least a trust score that is not lower than the prohibition threshold and lower than the allowance threshold, or a mandatory protection rule is met, then it is determined to be an enhanced protection flow. Sixth priority: If the credibility score is lower than the prohibition threshold, it is determined that circulation is prohibited.
[0064] In this embodiment, the platform records each metric. corresponding , , , And the triggering rules. The contribution of an indicator can be calculated using the leave-one-out method: keeping the remaining n-1 indicators unchanged, delete indicator x. i The remaining weights are then renormalized to obtain the reliable score after removing the indicator. ; then according to Calculation indicators The contribution of the score; according to Sort by size from largest to smallest; select a preset number of top-ranked indicators to generate key supporting factors, key opposing factors, and key uncertainties; the interpretation results are used for manual review and auditing, and do not directly replace decision rules; for decisions triggered by mandatory rules, the interpretation prioritizes displaying the rule identifier, rule version, and triggering evidence.
[0065] Then, based on the flow decision level and request context, construct the control strategy vector P=<D, F, V, N, Y, G> When multiple rules generate conflicting control measures, the policy generation module merges them according to protection level order. Prohibition is higher than permission; prohibited downloads are higher than allowed downloads; shorter validity periods are higher than longer validity periods; smaller field sets are higher than larger field sets; and lower call limits are higher than higher call limits. For conflicts that cannot be compared by level, they are transferred to manual review, and the conflicting rules are retained.
[0066] In the control strategy vector, decision level D is the only decision output from the priority decision rule table (Table 3); controlled release that meets regular verification is allowed for the flow; enhanced protection flows must be configured with at least one of the following: field reduction, data transformation, encrypted channel, access limit, download prohibition, or enhanced auditing; manually reviewed flows will not generate automatic release instructions before the review is passed; sandbox computing flows are only allowed to participate in calculations within a specified controlled computing environment and output audited results; gateway blocking instructions are prohibited for flows; different decision levels are not output in parallel.
[0067] S6. The policy compilation module first converts the control policy vector P into an intermediate policy representation IR independent of the specific execution gateway. The IR includes at least policy_id, rid, uid, did, purpose, effect, conditions, obligations, valid_from, valid_to, and version. Subsequently, the adapter compiles the IR into SQL query rewriting rules, data views, API gateway filters, de-identified service parameters, encrypted or watermarked call parameters, access token declarations, counter keys, and audit templates. Static validation includes checking the existence of fields, whether filter expressions can be parsed, whether algorithm identifiers are supported, whether the validity period is valid, whether dependent services are available, and whether there are irresolvable conflicts between rules. After static validation passes, the policy instructions are written to the candidate version. After the execution gateway completes loading and self-checking, it atomically switches the candidate version to the effective version. In case of loading failure, the previous version is retained and the reason for failure is recorded. In case of emergency reversal, the token is invalidated based on policy_id or uid, and the rule is rolled back to the prohibited rule.
[0068] Furthermore, in this embodiment, the gateway binds policy_id, rid, uid, did, purpose, op, valid_to, and a random number to the access token and performs integrity protection; the token cannot be reused by other subjects, data objects, or purposes; before the flow, the gateway verifies the token, subject status, policy version, and enforcement rules; during the flow, it performs query rewriting, field filtering, data transformation, counting, rate limiting, and exception blocking; after the flow, it records the actual returned fields, record quantity, data summary, policy hit, and execution result; if the desensitization, encryption, watermarking, or auditing services that the policy depends on are unavailable, the gateway blocks the request according to the failure closure principle, or, if the configuration allows, degrades to a policy with higher protection strength that does not depend on the service, and must not degrade to unprotected access.
[0069] S7. See also Figure 3 As shown, the types of audit events include normal performance, minor anomalies, frequency anomalies, field unauthorization, deviation from intended use, illegal downloads, token reuse, policy execution failures, and watermark tracking hits. This represents the structured audit event records collected in the t-th evaluation period, where t is the evaluation period number; the event classification unit is based on... The scalar behavioral risk is calculated based on the event type, severity level, impact data level, duration, and handling outcome. , .
[0070] Let u be the request subject sequence number. Let u be the trusted state of subject u before the update in period t. For the updated trusted state, To preserve the historical coefficient, The subject's trustworthiness is determined according to... Updated (0, 1); the anomaly correction coefficient for historical behavioral risk indicators is adjusted according to... Update, in which Anomaly correction coefficients are used for historical behavioral risk indicators. This is a non-negative event risk amplification factor.
[0071] To avoid state oscillations caused by frequent events, a maximum change Δmax can be set for a single period, so that... The limit is Δmax; the cooldown time, event deduplication window, and number of consecutive compliance attempts required for recovery can also be set; even if an entity in an isolated state has a high trust score, it cannot automatically enter the allowed flow.
[0072] Furthermore, to facilitate understanding of the specific technical solutions of this embodiment, the following example is provided to fully demonstrate the entire process of this embodiment from indicator mapping, dynamic weight aggregation to credible score calculation, decision level determination, and control strategy generation; A government department requests statistical analysis of highly sensitive population data and the output of statistical results, with an authorization period of 30 days. The platform constructs 9 indicators, λ=0.5; and integrates support values. and the reliability of evidence As shown in Figure 4 below: Table 4 Indicator Data ; Based on the basic weights and correction coefficients, the dynamic weights are obtained as follows: 0.1852, 0.0912, 0.1184, 0.1542, 0.1262, 0.1130, 0.0701, 0.0803, and 0.0614, which sum to 1. Substituting these values into the intuitionistic fuzzy weighted average operator, the comprehensive intuitionistic fuzzy number A = (0.6507, 0.2188, 0.1304) is calculated, with a confidence score S = 0.4319 and a deterministic score H = 0.8696.
[0073] set up The request enters the enhanced protection flow; the policy vector requires that only statistical fields be opened, direct identification fields be blocked, align identification fields be generalized, encrypted channels be used, task-level watermarks be added, the period be limited to 30 days, downloading of original details be prohibited, interface rate be limited, and enhanced auditing be enabled.
[0074] In response to the aforementioned enhanced protection flow decision, the policy compilation module generates a policy identifier POL-2026-001, compiling the control policy vector into specific instructions executable by the data flow execution gateway, including: Query rewriting rule: Find the intersection of the requested fields and the minimum necessary field set; Field filtering rules: Delete directly identifying fields such as ID card number and mobile phone number; Data transformation rules: Convert age to age range, and retain address down to the district / county level; Token rules: Bind the request subject identifier, data identifier, transfer purpose, and 30-day validity period; Metering rules: Set a limit on the number of calls per minute; Audit rules: Record the actual returned fields and the number of records.
[0075] After the gateway self-test passes, the policy is published atomically. When a request attempts to add an unauthorized field or use the token for other circulation purposes, the gateway blocks the request before the query reaches the data source node and generates an audit event containing the request identifier, policy identifier, unauthorized field, and execution result, thus completing the closed loop of control and audit.
[0076] Furthermore, to demonstrate the impact of the degree of conflict on the reliability and uncertainty of evidence when conflicting evidence exists from multiple sources, and to verify that this invention can quantify the uncertainty risk caused by inconsistencies in multi-source evidence, avoiding the direct averaging of conflicting evidence into a high-confidence conclusion, the aforementioned environmental safety indicators are used. For example, this indicator corresponds to three sources of evidence, each from a different security verification channel, as follows: Security assessment system: Single-source support value Source weight ; Real-time configuration check: Single-source supported values Source weight ; Manual filing record: Single source support value Source weight .
[0077] Fusion support value Conflict level ;like ,but ,get ;Although High, but Higher It reduces and increases hesitation, therefore this indicator cannot be regarded as a certain safe environment.
[0078] Based on the above, although the support value after multi-source fusion is high, the reliability of the evidence is significantly reduced and the degree of hesitation is greatly increased due to the large differences and high degree of conflict in the conclusions of multi-source evidence. Therefore, this indicator cannot be used as a definite basis for security, which is consistent with the representation logic of intuitionistic fuzzy sets for uncertain information.
[0079] If the environmental safety assessment expires, the environmental safety indicators will be affected. If it decreases from 0.85 to 0.35, then The value increased from 0.15 to 0.65. Keeping other inputs constant, the result was recalculated. If highly sensitive data If configured as 0.85, the request will be transferred from the enhanced protection process to the manual review process.
[0080] If the authorization certificate is completely missing, since the consistency of authorization is a key indicator, the platform will not use other high-scoring indicators to compensate, but will directly output supplementary authorization materials or prohibit circulation based on the key evidence missing rule.
[0081] Example 2 This embodiment provides a data flow trust assessment and dynamic control system, including a processor and a memory. The memory stores computer-executable instructions, and the processor is connected to the memory via a bus. When the processor executes the computer-executable instructions stored in the memory, it implements the data flow trust assessment and dynamic control method described in Embodiment 1 above. The system can be deployed as an independent trust decision server, or it can be integrated into a data sharing and exchange platform, a data trading platform, or an API gateway device. The processor can be a CPU, a dedicated chip, or an embedded processor, and the memory includes random access memory, read-only memory, disk storage media, optical storage media, etc.
[0082] Example 3 This embodiment provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it implements the data flow reliability assessment and dynamic control method described in the above embodiment. The computer-readable storage medium includes, but is not limited to, read-only memory (ROM), random access memory (RAM), magnetic disk, optical disk, flash memory, solid-state drive, and other media capable of storing computer programs, and can be used for the distribution, installation, and operation of software products.
[0083] The above are merely preferred embodiments of this application and are not intended to limit this application. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the protection scope of this application.
Claims
1. A method for reliable data flow assessment and dynamic control, characterized in that, Includes the following steps: S1. Receive a data transfer request, parse and normalize the data transfer request to obtain a request context; the request context includes at least a data identifier, request subject, transfer purpose, target node, operation type and authorization credential; S2. Based on the request context, obtain corresponding evidence data or evidence unavailability status from multiple preset evidence sources. For each preset evaluation index, normalize and fuse the multi-source evidence corresponding to the evaluation index. Calculate the evidence reliability of the evaluation index based on evidence completeness, evidence timeliness, source credibility, and evidence conflict degree, and generate indicator evidence records corresponding to each evaluation index. S3. For each evaluation indicator, based on the normalized support value and the reliability of the evidence, the evaluation indicator is mapped to an intuitionistic fuzzy number; the intuitionistic fuzzy number includes membership degree, non-membership degree and hesitation degree, where membership degree represents the degree of support of existing evidence for allowing data flow, non-membership degree represents the degree of opposition of existing evidence to allowing data flow, and hesitation degree represents the degree of uncertainty caused by insufficient evidence. S4. Obtain the basic weights of each evaluation indicator, and calculate the dynamic weights of each evaluation indicator by combining the correction coefficients corresponding to data security level, flow scenario, hesitation degree and abnormal risk; perform aggregation operation on the intuitive fuzzy numbers of all evaluation indicators based on the dynamic weights to obtain the overall credibility score and deterministic score. S5. According to the preset priority decision rules, combined with the credibility score, certainty score, key indicator veto threshold and mandatory control rules, determine a unique flow decision level and generate a control strategy vector corresponding to the flow decision level; the control strategy vector includes at least data range parameters, data transformation parameters, channel security parameters, usage constraint parameters and audit parameters; S6. The control strategy vector is compiled into a strategy instruction that can be directly executed by the data flow execution gateway, and sent to the data flow execution gateway. The data flow execution gateway performs token verification, field filtering, data transformation, access metering and abnormal blocking operations on the data flow process. S7. Obtain the audit events generated by the data flow execution gateway, and update the trust status and historical behavior risk indicators of the corresponding request subject according to the audit events for trust assessment of subsequent data flow requests.
2. The data flow reliability assessment and dynamic control method according to claim 1, characterized in that, The process of generating the indicator evidence records corresponding to each evaluation indicator includes the following steps: For each evaluation indicator, the corresponding original evidence tuples are collected from the preset evidence source set, and the original evidence tuples are deduplicated, validated, and marked with status. The original evidence values are normalized according to the direction of the evaluation indicators to obtain single-source support values. The direction of the indicators is used to characterize the relationship between the original values of the indicators and the normalized support values that support the flow of data, and is divided into positive, negative and moderate types. The higher the original value of the positive evaluation indicator, the higher the corresponding normalized support value. The higher the original risk value of the negative evaluation indicator, the lower the corresponding normalized support value. The moderate index has a preset optimal value range and allowable value boundary. When the original value is within the optimal value range, the normalized support value takes the maximum value. When the original value deviates from the optimal value range but does not exceed the allowable value boundary, the normalized support value takes the minimum value. The completeness of evidence is calculated based on the validity of all required evidence fields of the evaluation indicators; the timeliness of evidence is calculated based on the generation time and time-lapse rules of each piece of evidence. The source credibility is calculated based on the credibility attributes of each source of evidence; the degree of evidence conflict is calculated based on the degree of difference in the support values of multiple single sources. Multiply the values of evidence completeness, evidence timeliness, and source credibility together, and then multiply by the difference between the value 1 and the degree of evidence conflict to obtain the evidence reliability of the evaluation index. The support values from each single source are weighted and fused according to their source weights to obtain the multi-source fused support value of the evaluation index. The multi-source fused support value, evidence completeness, evidence timeliness, source credibility, evidence conflict degree, and evidence reliability are stored in the indicator evidence record corresponding to the evaluation index.
3. The data flow reliability assessment and dynamic control method according to claim 1, characterized in that, Mapping the evaluation index to an intuitive fuzzy number includes the following steps: The product of the evidence reliability and the normalized support value is used as the membership degree, the product of the evidence reliability and the normalized opposition value is used as the non-membership degree, and the difference between the value 1 and the evidence reliability is used as the hesitation degree; where the normalized opposition value is the difference between the value 1 and the normalized support value. The sum of membership degree, non-membership degree, and hesitation degree is 1, and the mean of the three is within a closed interval between 0 and 1.
4. The data flow reliability assessment and dynamic control method according to claim 1, characterized in that, The calculation process of the dynamic weights includes the following steps: Obtain the basic weights of each evaluation indicator; the basic weights are preset by safety experts, or obtained by weighting subjective weights and objective weights according to preset combination coefficients; wherein the subjective weights are determined according to pairwise comparison rules or regulatory requirements, and the objective weights are determined according to the dispersion or information entropy of historical evaluation samples; The unnormalized dynamic weight of each evaluation indicator is obtained by multiplying its base weight by the corresponding flow scenario correction coefficient, data security level correction coefficient, hesitation correction coefficient, and abnormal risk correction coefficient. The flow scenario correction coefficient is obtained by matching a preset value with the flow scenario type described in the current request; the data security level correction coefficient is obtained by matching a preset value with the data security level corresponding to the current request; the hesitation correction coefficient is obtained by adding the product of the value 1 and the preset non-negative adjustment coefficient and the hesitation of the corresponding evaluation index, and its value increases as the hesitation of the corresponding evaluation index increases; the abnormal risk correction coefficient is calculated based on the historical behavioral risk of the requesting subject and the preset event risk amplification coefficient. Divide the unnormalized dynamic weight of each evaluation indicator by the sum of the unnormalized dynamic weights of all evaluation indicators to obtain the final normalized dynamic weight of each evaluation indicator. The sum of the normalized dynamic weights of all evaluation indicators is 1.
5. The data flow reliability assessment and dynamic control method according to claim 1, characterized in that, The aggregation operation on the intuitionistic fuzzy numbers includes the following steps: The evaluation indicators are divided into multiple dimensions according to data object, request subject, usage authorization, execution environment, flow path and audit history. Within each dimension, the dynamic weights of all evaluation indicators within that dimension are re-normalized. Based on the normalized dynamic weights within the dimension, the intuitive fuzzy numbers of each evaluation indicator within the dimension are aggregated at the first level of intuitive fuzziness to obtain the dimension intuitive fuzzy number of the corresponding dimension. Dimension weights are assigned to all dimensions, and the weights of all dimensions are normalized. The intuitionistic fuzzy weighted average operator is used to perform a second-level aggregation of the intuitionistic fuzzy numbers of all dimensions to obtain the comprehensive intuitionistic fuzzy number. The confidence score is calculated based on the difference between the membership degree and the non-membership degree of the comprehensive intuitionistic fuzzy number, and the deterministic score is calculated based on the sum of the membership degree and the non-membership degree of the comprehensive intuitionistic fuzzy number.
6. The data flow reliability assessment and dynamic control method according to claim 1, characterized in that, The preset priority decision rules, in descending order of priority, are as follows: First priority: If a mandatory prohibition rule is met or the non-membership degree of a key indicator reaches the key indicator veto threshold, the circulation will be prohibited or the circulation will be manually reviewed. Second priority: If the original data must not leave the domain but controlled computation is allowed, it is determined to be a sandbox computation flow; Third priority: If the certainty score is lower than the preset certainty threshold and the non-membership degree of the key indicator reaches the key indicator rejection threshold, or if key evidence is missing or conflicting, it is determined to be manually reviewed and processed. Fourth priority: If the credibility score is not lower than the preset allowable threshold, the certainty score is not lower than the preset certainty threshold, and the mandatory protection rule is not hit, then the circulation is allowed. Fifth priority: If a request can be released under controlled conditions with additional protection measures, including at least a trust score that is not lower than the preset prohibition threshold and lower than the preset allowance threshold, or if it hits the mandatory protection rule, it is determined to be an enhanced protection flow. Sixth priority: If the credibility score is lower than the preset prohibition threshold, it is determined that circulation is prohibited; When making decisions, the rules are judged in order of priority from top to bottom. Once the highest priority rule is hit, the corresponding unique decision level is output and the lower priority rules are no longer executed.
7. The data flow reliability assessment and dynamic control method according to claim 6, characterized in that, The allowed threshold, prohibited threshold, and deterministic threshold are all adaptive thresholds; The allowable threshold is obtained based on the following steps: Based on the preset basic allowable threshold, the allowable threshold is obtained by adding the trust threshold increment corresponding to the data security level of the current request and the trust threshold increment corresponding to the operation type of the current request; The prohibition threshold is obtained by directly using a preset basic prohibition threshold, or by adding the trusted threshold increment corresponding to the data security level of the current request and the trusted threshold increment corresponding to the operation type of the current request to the preset basic prohibition threshold. The deterministic threshold is obtained based on the following steps: Based on the preset basic deterministic threshold, a deterministic threshold increment corresponding to the data security level of the current request and a deterministic threshold increment corresponding to the operation type of the current request are superimposed to obtain the deterministic threshold.
8. The data flow reliability assessment and dynamic control method according to claim 1, characterized in that, When generating the control strategy vector, if there is a conflict between the control measures generated by multiple rules, they are merged in order of protection level from high to low; control measures with higher protection level are applied first, with prohibition having higher priority than permission, shorter validity period having higher priority than longer validity period, smaller field set having higher priority than larger field set, and lower call limit having higher priority than higher call limit. Conflicts that cannot be compared according to protection level will be transferred to manual review.
9. The data flow reliability assessment and dynamic control method according to claim 1, characterized in that, The steps of compiling the control policy vector into policy instructions include: The control policy vector is converted into a policy intermediate representation that is independent of the gateway type. The policy intermediate representation includes a unique policy identifier, a request identifier, a request subject, a data identifier, a flow purpose, a policy effect, an effective condition, an execution obligation, an effective time, and a time limit. Based on the type of data flow execution gateway, the intermediate representation of the strategy is compiled into specific strategy instructions that can be executed by the corresponding data flow execution gateway. The specific strategy instructions include at least one of SQL query rewriting rules, field filtering rules, data desensitization parameters, access token declarations, access metering rules, and audit templates. The compiled policy instructions are statically validated. If the validation passes, the policy is sent to the data flow execution gateway in an atomic switching manner to take effect. If the validation fails, the previous version of the policy is retained and the exception is recorded. The static validation includes whether the field exists, whether the filter expression can be parsed, whether the algorithm identifier is supported, whether the validity period is valid, whether the dependent services are available, and whether there are irresolvable conflicts between the rules.
10. The data flow reliability assessment and dynamic control method according to claim 1, characterized in that, The execution steps of the data flow execution gateway include: The access token is bound to a unique policy identifier, request identifier, request subject, data identifier, transfer purpose, operation type, validity period and random number, and the access token is protected for integrity. Before data transfer, the validity and matching of the access token are verified. During the transfer, field filtering, data transformation, access counting, rate limiting and exception blocking are performed. After the transfer is completed, the summary information of the actual transferred data and the execution results are recorded. If the security services that the policy depends on are unavailable, the flow request should be blocked according to the failure shutdown principle, or downgraded to a policy with higher protection strength if the configuration allows it. It must not be downgraded to unprotected access.
11. The data flow reliability assessment and dynamic control method according to claim 1, characterized in that, The steps for updating the trust status and historical behavioral risk indicators of the corresponding requesting entity based on the audit event include: The collected audit events are classified and processed. Based on the event type, severity level, impact data level, duration, and handling result of the audit event, a scalar behavioral risk quantity with a value in the closed interval of 0 to 1 is calculated. Obtain the subject's trustworthiness status before the current assessment cycle update and the preset historical retention coefficient. Use the historical retention coefficient as the weight to weight the subject's trustworthiness status before the update. Use the difference between 1 and the historical retention coefficient as the weight to weight the difference between 1 and the behavioral risk amount. Sum the results of the two weightings and limit them to a closed interval between 0 and 1 to obtain the subject's trustworthiness status after the next assessment cycle update. Multiply the preset non-negative event risk amplification coefficient by the behavioral risk amount, and then add the product to the value 1 to obtain the abnormal risk correction coefficient corresponding to the historical behavioral risk index. Furthermore, the maximum change in the subject's trust status in a single period, the event deduplication window, and the number of consecutive compliance steps required to restore the normal state are set, limiting the change in the subject's trust status within a single evaluation period to not exceed the maximum change in a single period.
12. The data flow reliability assessment and dynamic control method according to claim 1, characterized in that, When standardizing the data transfer request, the following steps are included: The request context is normalized and serialized, and a request digest is calculated. The unique request identifier, request digest, and receipt time are stored in the request registration table. If the same unique request identifier corresponds to different request digests, it is determined that the request content is abnormal. If the same request subject, data identifier, transfer destination and random number appear repeatedly within the valid window, the original decision or decision duplicate request is reused according to the idempotency rule to prevent request replay.
13. The data flow reliability assessment and dynamic control method according to claim 1, characterized in that, The method also includes decision tracing: The decision evidence record is formed by storing the following together: the set of indicator evidence records for this request, the intuitive fuzzy number of each evaluation indicator, the dynamic weight, the credibility score, the deterministic score, the hit mandatory and veto rules, the final control strategy vector, the compiled decision representation, the summary of related audit events, and the configuration version used in this decision. A completeness summary is calculated for the decision evidence record, and the completeness summary is associated with the unique identifier of this request to prove the correspondence between evaluation inputs, decision results, execution strategies and feedback events, forming a traceable decision chain.
14. A data flow reliability assessment and dynamic control system, characterized in that, It includes a processor and a memory; the memory stores computer-executable instructions, and when the processor executes the computer-executable instructions in the memory, it implements a data flow reliability assessment and dynamic control method as described in any one of claims 1 to 13.
15. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which, when executed by a processor, implements a data flow reliability assessment and dynamic control method as described in any one of claims 1 to 13.
