A Gateway Optimization Method and System Based on Process Mining for Redundancy Removal and Merging of AIAPI Call Chains
By constructing multi-domain call event logs and generating policy evidence debt indicators through process mining, the calling strategy of the intelligent application interface gateway is optimized, which solves the problems of redundant calls and policy deviations in complex business processes, and achieves resource saving and response time improvement.
Patent Information
- Application Number
- CN202611409604.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-09-11
- Publication Date
- 2026-11-17
AI Technical Summary
Existing intelligent application interface gateways struggle to effectively identify and optimize redundant calls in complex business processes, leading to increased resource consumption and response latency. Furthermore, the lack of real-time evaluation of process variations and policy evidence results in policy deviations and risk accumulation.
By using a process mining approach, a multi-domain call event log is constructed, a process knowledge relationship model is generated, and the policy evidence debt index and the proportion of real support are calculated to perform policy risk control, suppress the self-reinforcement of erroneous control policies, and optimize the gateway call control policy set.
It reduces the number of unnecessary smart application interface calls, reduces resource consumption and response waiting time, improves the granularity of gateway governance, and ensures the accuracy of policies and the adaptability of processes.
Smart Images

Figure CN123123155A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of intelligent application interface gateway optimization technology, and more specifically, to a gateway optimization method and system for redundancy removal and merging of AIAPI call chains based on process mining. Background Technology
[0002] In intelligent business processing scenarios, intelligent application interface gateways typically handle functions such as interface request access, call forwarding, authentication control, traffic governance, call logging, and operational monitoring. With the widespread use of intelligent application interfaces in business processes such as order review, risk identification, content generation, price calculation, user profiling, and knowledge retrieval, multiple nodes, branches, and rounds of interface call chains often form within the same business process instance. Different intelligent application interfaces may have dependencies, parallel call relationships, result consumption relationships, and conditional triggering relationships. Some calls exhibit repetitive execution characteristics under the same or similar business conditions, and some call results may be consumed by multiple subsequent business nodes. Therefore, identifying redundant calls in the call chain, merging jointly executable calls, and reusing trusted call results on the gateway side are of practical significance for reducing interface call costs, reducing business response latency, and improving the granularity of gateway governance.
[0003] Existing gateway optimization methods typically focus on caching strategies, rate limiting strategies, routing strategies, circuit breaking strategies, or manually configured interface orchestration rules. In scenarios with relatively stable business processes and clearly defined interface input-output relationships, these methods can achieve good governance results. However, in complex business processes involving intelligent application interfaces, the call chain is often affected by business context, process variations, user states, external data changes, and downstream consumption paths. Simply relying on static rules or cache hit judgments at the single-call level is insufficient to fully reflect the true location and propagation impact of interface calls within the entire business process instance.
[0004] Furthermore, when the gateway performs redundancy removal, merging, or result reuse based on historical call patterns, skipped, merged, or replaced interface calls reduce the amount of actual execution observation data. If subsequent process mining continues to rely on policy-derived call events as reliable evidence, the source of policy evidence may gradually weaken, and the proportion of actual support may decrease without timely identification. For business scenarios with multiple process variations, the same gateway call control policy may remain highly effective in ordinary process paths, but may deviate in process variations such as promotional overlays, anomaly audits, and cross-domain data supplementation. Without a joint evaluation mechanism addressing policy evidence debt, the proportion of actual support, and differences in process variations, policy risks are difficult to mitigate in a timely manner.
[0005] Therefore, this application proposes a gateway optimization method and system for redundancy removal and merging of AIAPI call chains based on process mining. Summary of the Invention
[0006] To address the shortcomings of existing technologies, the present invention aims to provide a gateway optimization method and system for redundancy removal and merging of AIAPI call chains based on process mining.
[0007] To achieve the above objectives, the present invention provides the following technical solution:
[0008] A gateway optimization method based on process mining for AIAPI call chain redundancy removal and merging is applied to intelligent application interface gateways, including the following steps:
[0009] S1: Collect smart application interface call events in the same business process instance on the smart application interface gateway side, and build a multi-domain call event log based on the smart application interface call events;
[0010] S2: Perform process mining based on multi-domain call event logs, generate a process knowledge relationship model, and generate a gateway call control strategy set based on the process knowledge relationship model;
[0011] S3: When the intelligent application interface gateway executes any gateway call control strategy in the gateway call control strategy set, it generates a strategy hit result record and a strategy derivation call event, and calculates the strategy evidence debt index corresponding to the gateway call control strategy based on the strategy hit result record and the multi-domain call event log.
[0012] S4: For any gateway call control strategy in the gateway call control strategy set, calculate the actual support ratio corresponding to the gateway call control strategy based on the correlation between the gateway call control strategy and the multi-domain call event log.
[0013] S5: Determine the risk status value of the gateway's call control strategy based on the strategy evidence debt indicator and the actual support ratio indicator, and perform strategy risk control processing on the gateway's call control strategy based on the risk status value of the two types of indicators.
[0014] S6: Based on the strategy risk control processing, perform closed-loop updates on the process knowledge relationship model and gateway call control strategy set to suppress the self-reinforcement of erroneous gateway call control strategies in the process mining closed loop.
[0015] In one embodiment, the construction of the multi-domain call event log includes: recording real call execution events, policy derivation call events, counterfactual verification call events, and pollution correction call events generated during the history and / or closed-loop update process, and configuring event domain type record field, evidence attribute record field, and result impact record field to construct the multi-domain call event log.
[0016] In one embodiment, the process knowledge relationship model includes process variant information, call structure information, downstream consumption information, and multi-domain event support information; the gateway call control strategy set includes call merging execution strategy, result substitution and reuse strategy, and call skip execution strategy; each gateway call control strategy has strategy attribute information, which includes strategy identification information, strategy applicability information, strategy evaluation information, and strategy basis verification information; wherein, the strategy evaluation information includes strategy confidence level index, and the strategy applicability information includes strategy effective scope information.
[0017] In one embodiment, calculating the policy evidence debt index includes: when any gateway invokes a control policy that causes the smart application interface call represented by the corresponding smart application interface call event to be merged, reused, or skipped, generating a policy hit result record, a policy derived call event, and reuse result object information in the reuse scenario; obtaining the basic evidence debt index, the counterfactual verification compensation amount, and the policy evidence debt vector, and calculating the policy evidence debt index based on the basic evidence debt index and the counterfactual verification compensation amount.
[0018] In one embodiment, a policy evidence debt vector is constructed based on policy hit result records, multi-domain call event logs, policy evaluation information, and policy basis verification information. The policy evidence debt vector is normalized and weighted according to the policy evaluation information to obtain the basic evidence debt index. The counterfactual verification compensation amount is calculated based on the evidence attribute record field of the counterfactual verification call event corresponding to the gateway call control policy.
[0019] In one embodiment, the strategy evidence debt vector includes the original unobserved debt component, the critical path risk debt component, the downstream propagation risk debt component, the alternative reuse extrapolation debt component, the strategy timeliness debt component, the verification coverage gap debt component, and the lineage contamination pending debt component.
[0020] In one embodiment, the gateway invocation control strategy has a corresponding set of strategy evidence units; the true support ratio indicator includes the true support ratio of the strategy and the true support ratio of the variant; wherein, the true support ratio of the strategy is the true support ratio of the gateway invocation control strategy in the set of strategy evidence units, and the true support ratio of the variant is the true support ratio of the gateway invocation control strategy on the process variant corresponding to the process variant information.
[0021] In one embodiment, the policy risk control processing includes at least one of the following: increasing the shadow execution ratio parameter, generating counterfactual verification call events through shadow execution or comparative execution, increasing verification coverage, reducing the policy confidence index, narrowing the policy effective scope information, downgrading the gateway call control policy, suspending the execution of the gateway call control policy, or generating a pollution rollback trigger record.
[0022] In one embodiment, when the policy risk control process includes generating a pollution rollback trigger record, S6 includes: constructing a policy causal relationship graph based on the gateway call control policy, policy hit result record, policy derivation call event, and counterfactual verification call event; in response to the pollution rollback trigger record, performing pollution rollback processing based on the policy causal relationship graph, writing the processing result as a pollution correction call event into the multi-domain call event log, and updating the process knowledge relationship model and the gateway call control policy set accordingly.
[0023] In one embodiment, a gateway optimization system for redundancy removal and merging of AIAPI call chains based on process mining includes:
[0024] The data collection module is used to collect intelligent application interface call events in the same business process instance and build multi-domain call event logs.
[0025] The mining and generation module is used to mine processes based on multi-domain call event logs and generate process knowledge relationship models and gateway call control strategy sets.
[0026] The strategy execution module is used to execute any gateway call control strategy in the gateway call control strategy set, and generate strategy hit result records and strategy derivation call events;
[0027] The debt calculation module is used to calculate the policy evidence debt indicators corresponding to the gateway call control policy based on the policy hit result record and multi-domain call event log;
[0028] The support calculation module is used to calculate the actual support ratio based on the correlation between the gateway call control strategy and the multi-domain call event log;
[0029] The risk control module is used to determine the risk status values of two types of indicators based on the strategy evidence debt indicator and the actual support ratio indicator, and to perform strategy risk control processing.
[0030] The closed-loop update module is used to process the update process knowledge relationship model and gateway call control strategy set according to the risk control strategy, and to suppress the self-reinforcement of erroneous gateway call control strategies.
[0031] Compared with the prior art, the present invention has the following beneficial effects:
[0032] This invention generates a process knowledge relationship model based on multi-domain call event logs, and forms a gateway call control strategy set based on the process knowledge relationship model. This enables the gateway call control strategy to be generated by combining business process instances, process paths, interface dependencies and call event associations, avoiding call optimization based solely on single call caching or static configuration. This helps reduce the number of unnecessary smart application interface calls and reduces the call resource consumption and business response waiting time caused by repeated requests.
[0033] During the execution of any gateway call control strategy, a strategy hit result record and a strategy derivation call event are generated. The strategy evidence debt index and the actual support ratio index are calculated by combining the multi-domain call event log. Based on the risk status values of the two types of indicators, strategy risk control is carried out. Then, the process knowledge relationship model and the gateway call control strategy set are updated in a closed loop. This can identify situations where evidence support is insufficient, actual support is declining, or strategy derivation events interfere with process mining results, thereby reducing the risk of continuous accumulation and self-reinforcement of erroneous gateway call control strategies in the closed-loop optimization process. Attached Figure Description
[0034] Figure 1 A flowchart illustrating the gateway optimization method for deduplication and merging of AIAPI call chains based on process mining, provided in an embodiment of the present invention.
[0035] Figure 2 A schematic diagram illustrating the generation of the process knowledge relationship model and gateway call control strategy set provided in this embodiment of the invention;
[0036] Figure 3 This is a schematic diagram of the policy attribute information of the gateway invocation control policy provided in an embodiment of the present invention;
[0037] Figure 4 This is a schematic diagram illustrating the relationship between the strategy causal relationship diagram and the contamination rollback process provided in an embodiment of the present invention.
[0038] Figure 5 The diagram below illustrates the structure of a gateway optimization system for AIAPI call chain redundancy removal and merging based on process mining, as provided in this embodiment of the invention. Detailed Implementation
[0039] Example 1: Refer to Figures 1 to 4 A gateway optimization method based on process mining for AIAPI call chain redundancy removal and merging is applied to intelligent application interface gateways, including the following steps:
[0040] S1: Collect smart application interface call events in the same business process instance on the smart application interface gateway side, and construct a multi-domain call event log based on the smart application interface call events; establish event collection boundaries on the smart application interface gateway side according to the business process instance dimension, include smart application interface call events occurring in the same business process instance into the same log sequence, and write the following records for each smart application interface call event: business process instance identifier, smart application interface identifier, call request summary, call response summary, call start time, call end time, call status, upstream trigger source, and downstream consumer object;
[0041] In this log sequence, the actual intelligent application interface calls completed during the historical operation are written into real call execution events; the merged execution, replacement reuse, or skipped execution results formed after the gateway call control strategy is hit are written into policy derived call events; the verification results generated by shadow execution or comparison execution are written into counterfactual verification call events; and the correction results generated by pollution rollback processing are written into pollution correction call events.
[0042] Each type of event is configured with an event domain type record field, an evidence attribute record field, and a result impact record field. The event domain type record field records whether the event belongs to a real call execution event, a policy-derived call event, a counterfactual verification call event, or a pollution correction call event. The evidence attribute record field records the credibility of the event source, the verification coverage, whether it participates in policy basis verification, and the corresponding gateway call control policy. The result impact record field records the impact of the event on the call result, downstream consumption result, and the execution status of the business process instance.
[0043] Statistics Window The multi-domain call event log within the application consists of intelligent application interface call events. Determined by the business process instance batch, strategy version cycle, or strategy risk control cycle. Any event At least record the business process instance identifier and the intelligent application interface identifier. Event domain types Call input signature Response Summary The event domain type includes: invocation time, upstream trigger source, downstream consumer object, process variant identifier, associated strategy identifier, evidence attribute record, and result impact record. include , , and ,in Indicates the actual execution event. This indicates a strategy-derived invocation event. This indicates a counterfactual verification call event. This indicates a pollution correction call event. Counterfactual verification call events generated by shadow execution or comparative execution are not used as external responses of the current business process instance and do not rewrite the business state of the current business process instance; this event is associated with the corresponding policy hit result record, the verified policy derived call event, and the associated real call execution event, and participates in policy evidence debt indicators, real support ratio indicators, and pollution rollback processing.
[0044] For example, in an order review business process instance, the intelligent application interface gateway sequentially collects intelligent application interface call events from the identity verification interface, inventory query interface, price calculation interface, and risk control review interface. When the inventory query interface is actually executed in the historical records, an actual call execution event is written. When the result of the price calculation interface is reused by the result reuse strategy, a strategy derivation call event is written. When the gateway re-calls the price calculation interface through shadow execution for comparison, a counterfactual verification call event is written. When the comparison finds that the reused result is inconsistent with the actual call result and triggers pollution rollback processing, a pollution correction call event is written. This constructs a multi-domain call event log containing multiple event sources, multiple evidence attributes, and multiple result impacts.
[0045] S2: Based on the multi-domain call event log, perform process mining to generate a process knowledge relationship model, and generate a gateway call control strategy set based on the process knowledge relationship model; based on the real call execution events, policy-derived call events, counterfactual verification call events, and pollution correction call events already collected in the multi-domain call event log, perform process mining according to the event occurrence order, call dependency relationship, result consumption relationship, and event domain type record fields of the same business process instance, identify the repeated intelligent application interface call paths, parallel call segments, result reuse segments, and low contribution call segments between different business process instances, and generate a process knowledge relationship model accordingly;
[0046] In the process knowledge relationship model, process variant information records the differences in call paths formed by different business process instances, call structure information records the dependencies, parallel relationships, and conditional branch relationships between intelligent application interfaces, downstream consumption information records the situation where the call result of a certain intelligent application interface is consumed by other intelligent application interfaces or business process nodes, and multi-domain event support information records the support of real call execution events, strategy-derived call events, counterfactual verification call events, and pollution correction call events for the same call relationship;
[0047] Based on this, a gateway call control strategy set is generated according to the process knowledge relationship model. The gateway call control strategy set includes call merging execution strategy, result substitution and reuse strategy, and call skip execution strategy. Each gateway call control strategy is written with strategy attribute information, which includes strategy identification information, strategy applicability information, strategy evaluation information, and strategy basis verification information. Among them, the strategy identification information records the unique identifier and strategy type of the gateway call control strategy, the strategy applicability information records the effective scope of the strategy, the strategy evaluation information records the strategy confidence level index, and the strategy basis verification information records the corresponding multi-domain event support source.
[0048] Call input signature From the summary function Generate, denoted as
[0049]
[0050] in, This is the intelligent application interface version. For model version, This is a prompt template version. To retrieve corpus versions or external data versions, To call the parameter summary, For business context summary, For permission domains, For business time window, A summarization function that produces the same summary result while maintaining the same input. Let... In relation to the event The set of historical real call execution events with the same call input signature is calculated as follows:
[0051]
[0052] set up In relation to the event The set of counterfactual verification call events with the same call input signature is calculated as follows:
[0053]
[0054] in, For historical real call execution events or counterfactual verification call events, For the event The call input signature, For the event The event domain type. Let... For the event The call response summary For historical real-time execution events Counterfact check call event The difference in call response summaries This is the threshold for result equivalence. When... and All are not empty, and any , All meet In the case of a certain condition, the result equivalence rule is deemed to be satisfied; otherwise, the result equivalence rule is deemed not to be satisfied. Recorded in the policy basis verification information, configured according to the interface output type, business fault tolerance boundary and the sensitivity of downstream consumer objects. The calculation criteria are synchronously recorded in the policy basis verification information, and the comparison is performed according to the interface output type, selecting structured field comparison, numerical error normalization comparison, text response summary comparison, or a combination thereof. Let... This serves as an identifier for interface side effects. This indicates that the smart application interface call does not have side effects such as writing state, deducting fees, external notifications, or other changes to external state. This indicates the existence of side effects; assuming This serves as an idempotency identifier for the interface. This indicates that repeated executions with the same call input signature yield consistent results. This indicates that the idempotency requirement is not met. When , If the result equivalence rule is met, the security prerequisite for policy application is deemed to be met; if the security prerequisite is not met, the smart application interface gateway will not execute the call merging execution strategy, the result substitution and reuse strategy, or the call skip execution strategy for the smart application interface call.
[0055] For example, in an order review business process instance, the identity verification interface, inventory query interface, price calculation interface, and risk control review interface form stable call paths in a large number of historical instances. Process mining identifies that the price calculation interface has highly consistent call results under the same product, the same membership level, and the same promotion conditions, and this result is mainly consumed by the risk control review interface. Based on this, the process knowledge relationship model generates records containing process variant information, call structure information between the price calculation interface and the risk control review interface, downstream consumption information of the price calculation result, and multi-domain event support information. Furthermore, it generates a result replacement and reuse strategy, and writes the strategy confidence level index, strategy effective scope information, and strategy basis verification information.
[0056] S3: When the intelligent application interface gateway executes any gateway call control strategy in the gateway call control strategy set, it generates a strategy hit result record and a strategy derivation call event, and calculates the strategy evidence debt index corresponding to the gateway call control strategy based on the strategy hit result record and the multi-domain call event log.
[0057] When the intelligent application interface gateway executes any gateway call control strategy in the gateway call control strategy set, it matches the gateway call control strategy with the intelligent application interface call event in the current business process instance. If the matching result falls within the policy effective range information and meets the call conditions limited by the policy basis verification information, it records the policy identification information, hit time, hit business process instance identification, controlled intelligent application interface identification, control type and control result, and generates a policy hit result record.
[0058] When the controlled smart application interface call is merged, reused, or skipped, the actual gateway execution result is written into the policy derivation call event. In the case of reuse, the reuse result object information is also recorded. The reuse result object information includes the source of the reused result, the time of the reused result generation, the reuse conditions, the reuse result summary, and the downstream consumer object.
[0059] For example, in an order review business process instance, the price calculation interface hits the strategy because the product identifier, membership level, and promotion conditions are consistent with the historical records. The smart application interface gateway generates a strategy hit result record for this strategy and writes the reused price result and its source into the strategy derivation call event.
[0060] Based on the strategy hit result record, multi-domain call event log, strategy evaluation information and strategy basis verification information, a strategy evidence debt vector is constructed. The strategy evidence debt vector includes the original unobserved debt component, the critical path risk debt component, the downstream propagation risk debt component, the alternative reuse extrapolation debt component, the strategy timeliness debt component, the verification coverage gap debt component, and the lineage contamination pending debt component.
[0061] Among them, the original unobserved debt component is determined based on the missing real call execution events after being merged, replaced, reused, or skipped; the critical path risk debt component is determined based on the criticality of the smart application interface call in the business process instance; the downstream propagation risk debt component is determined based on downstream consumption information; the replacement reuse extrapolation debt component is determined based on the degree of deviation between the reuse result object information and the current call conditions; the strategy timeliness debt component is determined based on the interval between the strategy generation time and the current hit time; the verification coverage gap debt component is determined based on the coverage ratio of the counterfactual verification call events; and the bloodline contamination pending debt component is determined based on the correlation between the strategy derived call events and the contamination correction call events.
[0062] Based on the strategy confidence index in the strategy evaluation information, the above components are normalized and weighted to obtain the basic evidence debt index. The counterfactual verification compensation amount is calculated based on the evidence attribute record field of the counterfactual verification call event corresponding to the gateway call control strategy. The strategy evidence debt index is obtained by subtracting the counterfactual verification compensation amount from the basic evidence debt index, and the result is not less than zero.
[0063] Assume the gateway call control strategy is as follows: Statistics window Internal strategy The number of hits is ,Strategy The corresponding counterfactual verification call event count is ,Strategy The number of events with consistent results in the corresponding counterfactual verification call events is: .set up It is the denominator protection constant, and Denominator protection function Calculate using the following formula:
[0064]
[0065] in, The denominator to be processed To determine the function that takes the larger value. Let... For strategy The Each debt component, , to These correspond to the original unobserved debt component, the critical path risk debt component, the downstream propagation risk debt component, the alternative reuse extrapolation debt component, the strategy timeliness debt component, the verification coverage gap debt component, and the lineage contamination pending debt component, respectively. Each component is processed within the normalized boundary recorded in the strategy-based verification information. Internal values. Let... For strategy The Individual debt component weights, Recorded from strategy evaluation information, ,and Basic evidence of debt indicators Calculate using the following formula:
[0066]
[0067] Let the counterfact check coverage be . The counterfactual consistency rate is The quality of counterfactual evidence is The compensation coefficient is ,and , , , All belong to , Determined by evidence attribute records and outcome impact records. Counterfactual verification coverage. Calculate using the following formula:
[0068]
[0069] Counterfactual consistency rate Calculate using the following formula:
[0070]
[0071] Counterfact check compensation Calculate using the following formula:
[0072]
[0073] Strategy Evidence Debt Indicators Calculate using the following formula:
[0074]
[0075] When the counterfactual verification call count is zero, press Implement denominator protection; when there are inconsistent results in the counterfactual verification call event, the inconsistent record is not included. It then proceeds to calculate the actual support ratio indicator and the risk status value of the dual-category indicator.
[0076] S4: For any gateway call control strategy in the gateway call control strategy set, calculate the actual support ratio corresponding to the gateway call control strategy based on the correlation between the gateway call control strategy and the multi-domain call event log; based on the strategy identification information, strategy application information, strategy basis verification information of the gateway call control strategy, and the event domain type record field in the multi-domain call event log, extract the actual call execution events, strategy derived call events, counterfactual verification call events, and pollution correction call events that are associated with the gateway call control strategy, and merge them according to the same business process instance identifier, intelligent application interface identifier, call input conditions, call result consistency, downstream consumer object, and process variant attribution relationship to form a set of strategy evidence units corresponding to the gateway call control strategy;
[0077] Each strategy evidence unit records the evidence source, evidence domain type, invocation conditions, hit strategy, actual execution result, strategy derivation result, counterfactual verification result, pollution correction status, and process variant identifier. Records that are only supported by strategy derivation invocation events but lack confirmation from actual invocation execution events or counterfactual verification invocation events are marked as non-true supporting evidence. Records that have the same results as actual invocation execution events and counterfactual verification invocation events and are not negated by pollution correction invocation events are marked as true supporting evidence.
[0078] The true support ratio is calculated based on the policy evidence unit set. Let the policy be... The set of strategy evidence units is Any strategy evidence unit It is obtained by merging the same business process instance identifier, smart application interface identifier, call input signature, process variant identifier, and associated strategy identifier. Let... For evidence unit The weighting of the calculation , The impact of the results on records and the sensitivity of downstream consumers are determined; assuming To truly support the indicated quantity, When the evidence unit When the modified call event is not negated by contamination, and there is a real call execution event or a counterfactual verification call event with a result consistent with the real call execution event, When the evidence unit When supported only by policy-derived call events, negated by counterfactual check call events with inconsistent results, or negated by contaminated correction call events. Evidence units whose pollution correction call events have been negated do not enter the weight of true supporting evidence, but are retained in the total weight of the policy evidence unit set. We will continue using the aforementioned denominator protection constant. The actual support percentage of the strategy. Calculate using the following formula:
[0079]
[0080] set up As a process variant, for The process variant is The set of evidence units. The percentage of variants that truly support the claim. Calculate using the following formula:
[0081]
[0082] set up For strategy There exists a set of process variations for the strategy evidence unit. For process variants The variant weights, and The weighted average of the percentage of variants that actually support the variant is denoted as... Calculate using the following formula:
[0083]
[0084] The degree of dispersion of the variant is denoted as Calculate using the following formula:
[0085]
[0086] when When empty, strategy Record the flag to be verified, and do not classify it as a low-risk state; when a certain process variant This process variant does not participate when no strategy evidence unit is available. The calculation, and its missing cases are entered into the verification coverage gap debt component.
[0087] For example, in an order review business process instance, when the result substitution reuse strategy is applied to the price calculation interface, if 70 out of 100 strategy evidence units have consistent results from both real call execution events and counterfactual verification call events, 10 are negated by polluted correction call events, and 20 only have strategy derivation call events, then the policy real support ratio of this gateway call control strategy is calculated based on the 70 real support evidences relative to the range of valid evidence. If the real support ratio is higher in the standard order process variant but lower in the promotional overlay order process variant, then the variant real support ratio reflects the support difference of this gateway call control strategy on different process variants.
[0088] S5: Determine the dual-category indicator risk status value of the gateway call control strategy based on the strategy evidence debt indicator and the actual support ratio indicator, and perform strategy risk control processing on the gateway call control strategy based on the dual-category indicator risk status value; For any gateway call control strategy in the gateway call control strategy set, incorporate the strategy evidence debt indicator and the actual support ratio indicator into the same risk judgment coordinate for joint evaluation, and determine the dual-category indicator risk status value of the gateway call control strategy according to the high and low range of the strategy evidence debt indicator, the credible range of the strategy actual support ratio, and the dispersion of the variant actual support ratio on different process variants; The dual-category indicator risk status values include low risk status, observation risk status, verification enhancement risk status, execution contraction risk status, and contamination pending rollback risk status, and different risk statuses correspond to different strategy risk control processing content;
[0089] Set a strategy The risk status value of the dual-category indicator .set up As evidence of a low-risk debt threshold, As evidence of a high-risk debt threshold, and Less than ;set up To determine the threshold for the actual support percentage of the strategy, For the actual support percentage threshold of variants, For the threshold of the degree of dispersion of the variant, The threshold for the counterfactual verification inconsistency rate. This represents the undetermined threshold for bloodline contamination. These thresholds are recorded from strategy evaluation information or strategy basis verification information, and all fall within the value range of the corresponding indicator. Let... For strategy The number of times the results are inconsistent in the corresponding counterfactual verification call events. To counterfactual verification inconsistency rate, Using the aforementioned denominator protection constant, and
[0090]
[0091] For strategy There exists a set of process variations for the strategy evidence unit. When When empty, strategy Retain the flag to be verified. Handle according to the enhanced risk status verification. When not empty, the risk status values of the two types of indicators are determined in the following order: contamination rollback risk status, execution contraction risk status, verification enhancement risk status, observation risk status, and low risk status: when Not less than and Not less than hour, Recorded as a pollution risk status pending rollback; when Below or exists satisfy Below ,or Higher than hour, This is recorded as a state of risk of contraction; when Higher than and Not less than hour, This is recorded as a verification enhancement risk state; when Higher than and not higher than ,and Not less than hour, Recorded as the state of observation risk; when Not higher than , Not less than ,all All meet Not less than ,and Not higher than hour, This is recorded as a low-risk state. When the same strategy meets multiple state conditions, the state that appears first in the above order will be used. The strategy verification information records the thresholds, weights, normalization boundaries, and policy effectiveness scope involved in the risk state values of the strategy evidence debt indicator, the proportion of true support indicator, and the dual-category indicator, and configures the rule version. The strategy hit result record retains the rule version used when the hit occurred. When the closed-loop update changes the threshold, weight, normalization boundary, or policy effective range, a new rule version is generated; historical policy hit result records and existing policy evidence units are not rewritten according to the new rule version.
[0092] When the policy evidence debt index is low and the policy true support ratio and variant true support ratio are both in a stable range, maintain the current policy confidence index and policy effective range information for the gateway to call the control policy; when the policy evidence debt index increases but the true support ratio index does not decrease significantly, increase the shadow execution ratio parameter and generate counterfactual verification call events through shadow execution or comparative execution.
[0093] When the percentage of actual support for a strategy decreases, or when the percentage of actual support for a variant corresponding to a certain process variant falls below a preset support threshold, the verification coverage is increased, the strategy confidence index is reduced, and the scope of strategy effectiveness is narrowed. When the strategy evidence debt index remains high and the percentage of actual support decreases across multiple process variants, the gateway call control strategy is downgraded or its execution is suspended. When the counterfactual verification call event shows a persistent deviation between the strategy-derived call event and the actual call execution event, and the pollution correction call event points to the same gateway call control strategy, a pollution rollback trigger record is generated.
[0094] For example, in the order review business process instance, the result substitution reuse strategy of the price calculation interface maintains a high proportion of real support in the normal order process variant, but a deviation in the reuse result occurs in the promotion superimposed order process variant. Based on this, the intelligent application interface gateway adjusts the risk status value of the two types of indicators to the execution contraction risk status, reduces the policy confidence index of the result substitution reuse strategy, limits the policy effective scope information to the normal order process variant, and increases the shadow execution ratio parameter under the promotion superimposed order process variant; if the shadow execution continuously finds that the reused price is inconsistent with the real price, a pollution rollback trigger record is generated.
[0095] S6: Based on the strategy risk control processing, perform a closed-loop update of the process knowledge relationship model and the gateway call control strategy set to suppress the self-reinforcement of erroneous gateway call control strategies in the process mining closed loop. Based on the strategy risk control processing results formed in step S5, perform closed-loop correction on the process variant information, call structure information, downstream consumption information, and multi-domain event support information in the process knowledge relationship model, and simultaneously adjust the strategy confidence level indicators, strategy effective scope information, strategy basis verification information, and strategy execution status in the gateway call control strategy set;
[0096] When the strategy risk control process includes pollution rollback trigger records, a strategy causal relationship graph is established around the triggered gateway call control strategy. The strategy causal relationship graph takes the gateway call control strategy as the root node and the strategy hit result record, strategy derivation call event, counterfactual verification call event, reuse result object information, downstream consumer object and pollution correction object as associated nodes, recording the causal path between strategy hit, result derivation, downstream propagation, verification deviation and pollution confirmation.
[0097] Strategy The corresponding strategy causal relationship diagram is denoted as follows: Its node set is The set of directed edges is . It includes at least the gateway call control policy node, policy hit result record node, policy derivation call event node, counterfactual verification call event node, downstream consumer object node, and pollution correction call event node; Record the relationships between hits, derivations, consumption, verification, and corrections. Correction edges are initiated by pollution correction event nodes pointing to the negated policy derivation event nodes. Let... Indicates in From the node Pointing to node A directed path exists. To correct the pollution, call the event node. To derive event nodes from strategies, and They are nodes and nodes The event domain type of the corresponding event. The set of events invoked by the contaminated policy derivation. Determine by the following formula:
[0098]
[0099] for The policy derivation call event in the associated policy evidence unit is the true support indicator. Set as The pollution correction status is updated to "negated", and the associated policy adds a pollution correction call event identifier based on the verification record.
[0100] In response to the pollution rollback trigger record, the evidence units, process variant support records and policy basis verification records that have a causal relationship with the erroneous policy derivation results in the policy causal lineage diagram are rolled back. The polluted policy derivation call events are removed from the true supporting evidence. The deviation results confirmed by the counterfactual verification call events are written into the pollution correction call events, and the pollution correction call events are included in the multi-domain call event log.
[0101] When updating the process knowledge relationship model in a closed loop, update the event configuration weights in the multi-domain call event log. .set up To update the event weights, Preserve weight coefficients for strategy-derived event calls. To verify counterfactual events, call the event weight coefficient, and ;set up To verify the consistency of the event call record in the counterfactual verification, To indicate agreement, This indicates inconsistency. The value is written according to the aforementioned result equivalence rules, and is taken when the result equivalence rules are satisfied. If the result equivalence rule is not satisfied, take Event weights are updated according to the following data caliber: when hour, Pick ;when and hour, Pick ;when and hour, Pick ;when ,or and ,or hour, Pick The process knowledge relationship model is based on... Recalculate process variant information, call structure information, downstream consumption information, and multi-domain event support information; based on the gateway call control strategy set. , , , and Update the policy confidence index, policy effective scope information, policy basis verification information, and policy execution status. Pollution correction invocation events are not included in the numerator of the actual policy support percentage.
[0102] For example, in an order review business process instance, the price calculation interface's result replacement reuse strategy generates an incorrect reuse price due to promotional orders. This incorrect price is consumed by the risk control review interface and affects the order's risk level. The intelligent application interface gateway establishes a policy causal relationship graph based on the policy hit result record, reuse result object information, and counterfactual verification call events. It locates the contaminated promotional order process variant and its corresponding downstream consumption path, marks the relevant policy-derived call events as invalid supporting evidence, writes them into the contamination correction call events, and updates the process knowledge relationship model and gateway call control strategy set to suppress the self-reinforcement of the incorrect gateway call control strategy in the process mining closed loop.
[0103] Example 2: Refer to Figure 5 A gateway optimization system for AIAPI call chain redundancy removal and merging based on process mining, including:
[0104] The data collection module is used to collect smart application interface call events from the same business process instance and build a multi-domain call event log. Deployed on the smart application interface gateway side, it receives smart application interface call events from the same business process instance, and collects the call time, call object, call result and event source to form the data foundation of the multi-domain call event log.
[0105] The mining and generation module is used to perform process mining based on multi-domain call event logs, generate a process knowledge relationship model and a set of gateway call control strategies; it takes the event records in the multi-domain call event logs, performs process mining on the intelligent application interface call relationships in the same business process instance, forms a process knowledge relationship model, and generates a set of gateway call control strategies based on the process knowledge relationship model.
[0106] The strategy execution module is used to execute any gateway call control strategy in the gateway call control strategy set, generate a strategy hit result record and a strategy derivation call event; during the operation of the smart application interface gateway, it matches the gateway call control strategies in the gateway call control strategy set, forms a strategy hit result record for the hit gateway call control strategy, and synchronously generates a strategy derivation call event.
[0107] The debt calculation module is used to calculate the policy evidence debt index corresponding to the gateway call control policy based on the policy hit result record and the multi-domain call event log; combined with the policy hit result record and the multi-domain call event log, it analyzes the sufficiency of evidence generated by the gateway call control policy, the verification coverage and the scope of call impact, and obtains the policy evidence debt index corresponding to the gateway call control policy.
[0108] The support calculation module is used to calculate the true support ratio indicator based on the correlation between the gateway call control strategy and the multi-domain call event log. Based on the correlation between the gateway call control strategy and the multi-domain call event log, the module organizes the set of policy evidence units corresponding to the gateway call control strategy and calculates the true support ratio indicator, which includes the true support ratio of the policy and the true support ratio of the variant.
[0109] The risk control module is used to determine the risk status values of two types of indicators based on the strategy evidence debt indicator and the actual support ratio indicator, and to perform strategy risk control processing. It combines the strategy evidence debt indicator and the actual support ratio indicator to determine the risk status values of the two types of indicators for the gateway to call the control strategy, and executes strategy risk control processing based on the risk status values of the two types of indicators, including verification enhancement, confidence level adjustment, effective scope adjustment, strategy downgrade, strategy suspension, or pollution rollback trigger record generation.
[0110] The closed-loop update module is used to update the process knowledge relationship model and gateway call control strategy set according to the policy risk control processing, and to suppress the self-reinforcement of erroneous gateway call control strategies. It receives the results of the policy risk control processing, performs a closed-loop update of the process knowledge relationship model and gateway call control strategy set; in the case of generating a pollution rollback trigger record, it also performs pollution rollback processing in conjunction with the policy causal lineage graph, and writes the processing result as a pollution correction call event to the multi-domain call event log.
[0111] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.
Claims
1. A gateway optimization method for AIAPI call chain redundancy removal and merging based on process mining, characterized in that, For use in smart application interface gateways, the following steps are included: S1: Collect smart application interface call events in the same business process instance on the smart application interface gateway side, and build a multi-domain call event log based on the smart application interface call events; S2: Perform process mining based on multi-domain call event logs, generate a process knowledge relationship model, and generate a gateway call control strategy set based on the process knowledge relationship model; S3: When the intelligent application interface gateway executes any gateway call control strategy in the gateway call control strategy set, it generates a strategy hit result record and a strategy derivation call event, and calculates the strategy evidence debt index corresponding to the gateway call control strategy based on the strategy hit result record and the multi-domain call event log. S4: For any gateway call control strategy in the gateway call control strategy set, calculate the actual support ratio corresponding to the gateway call control strategy based on the correlation between the gateway call control strategy and the multi-domain call event log. S5: Determine the risk status value of the gateway's call control strategy based on the strategy evidence debt indicator and the actual support ratio indicator, and perform strategy risk control processing on the gateway's call control strategy based on the risk status value of the two types of indicators. S6: Based on the strategy risk control processing, perform closed-loop updates on the process knowledge relationship model and gateway call control strategy set to suppress the self-reinforcement of erroneous gateway call control strategies in the process mining closed loop.
2. The gateway optimization method for AIAPI call chain redundancy removal and merging based on process mining as described in claim 1, characterized in that, The construction of the multi-domain call event log includes: recording real call execution events, policy-derived call events, counterfactual verification call events, and pollution correction call events generated during the history and / or closed-loop update process, and configuring event domain type record fields, evidence attribute record fields, and result impact record fields to construct the multi-domain call event log.
3. The gateway optimization method for AIAPI call chain redundancy removal and merging based on process mining according to claim 2, characterized in that, The process knowledge relationship model includes process variant information, call structure information, downstream consumption information, and multi-domain event support information; The gateway call control strategy set includes call merging and execution strategy, result substitution and reuse strategy, and call skipping execution strategy; Each gateway invokes a control policy with policy attribute information, which includes policy identification information, policy applicability information, policy evaluation information, and policy basis verification information. Among these, the policy evaluation information includes a policy confidence level index, and the policy applicability information includes the policy effective scope information.
4. The gateway optimization method for AIAPI call chain redundancy removal and merging based on process mining according to claim 3, characterized in that, The calculation of the policy evidence debt index includes: when any gateway calls the control policy and causes the smart application interface call represented by the corresponding smart application interface call event to be merged, replaced, reused, or skipped, generating policy hit result records, policy derived call events, and reuse result object information in the replacement and reuse scenario; obtaining the basic evidence debt index, counterfactual verification compensation amount, and policy evidence debt vector, and calculating the policy evidence debt index based on the basic evidence debt index and the counterfactual verification compensation amount.
5. The gateway optimization method for AIAPI call chain redundancy removal and merging based on process mining according to claim 4, characterized in that, A policy evidence debt vector is constructed based on policy hit result records, multi-domain call event logs, policy evaluation information, and policy basis verification information. The policy evidence debt vector is normalized and weighted according to the policy evaluation information to obtain the basic evidence debt index. The counterfactual verification compensation amount is calculated based on the evidence attribute record fields of the counterfactual verification call event corresponding to the gateway call control policy.
6. The gateway optimization method for AIAPI call chain redundancy removal and merging execution based on process mining according to claim 5, characterized in that, The strategy evidence debt vector includes the original unobserved debt component, the critical path risk debt component, the downstream propagation risk debt component, the alternative reuse extrapolation debt component, the strategy timeliness debt component, the verification coverage gap debt component, and the lineage contamination pending debt component.
7. The gateway optimization method for AIAPI call chain redundancy removal and merging based on process mining according to claim 1, characterized in that, The gateway call control strategy has a corresponding set of strategy evidence units; the true support ratio indicator includes the true support ratio of the strategy and the true support ratio of the variant; whereby the true support ratio of the strategy is the true support ratio of the gateway call control strategy in the set of strategy evidence units, and the true support ratio of the variant is the true support ratio of the gateway call control strategy on the process variant corresponding to the process variant information.
8. The gateway optimization method for AIAPI call chain redundancy removal and merging execution based on process mining according to claim 7, characterized in that, The strategy risk control measures include at least one of the following: increasing the shadow execution ratio parameter, generating counterfactual verification call events through shadow execution or comparative execution, increasing verification coverage, reducing the strategy confidence index, narrowing the scope of policy effectiveness, downgrading the gateway call control strategy, suspending the execution of the gateway call control strategy, or generating a pollution rollback trigger record.
9. The gateway optimization method for AIAPI call chain redundancy removal and merging based on process mining according to claim 8, characterized in that, When the strategy risk control process includes generating pollution rollback trigger records, S6 includes: constructing a strategy causal relationship graph based on gateway call control strategy, strategy hit result records, strategy derived call events, and counterfactual verification call events; responding to the pollution rollback trigger records, performing pollution rollback processing based on the strategy causal relationship graph, writing the processing result as a pollution correction call event to the multi-domain call event log, and updating the process knowledge relationship model and gateway call control strategy set accordingly.
10. A gateway optimization system for AIAPI call chain redundancy removal and merging execution based on process mining, applied to the gateway optimization method for AIAPI call chain redundancy removal and merging execution based on process mining as described in any one of claims 1-9, characterized in that, include: The data collection module is used to collect intelligent application interface call events in the same business process instance and build multi-domain call event logs. The mining and generation module is used to mine processes based on multi-domain call event logs and generate process knowledge relationship models and gateway call control strategy sets. The strategy execution module is used to execute any gateway call control strategy in the gateway call control strategy set, and generate strategy hit result records and strategy derivation call events; The debt calculation module is used to calculate the policy evidence debt indicators corresponding to the gateway call control policy based on the policy hit result record and multi-domain call event log; The support calculation module is used to calculate the actual support ratio based on the correlation between the gateway call control strategy and the multi-domain call event log; The risk control module is used to determine the risk status values of two types of indicators based on the strategy evidence debt indicator and the actual support ratio indicator, and to perform strategy risk control processing. The closed-loop update module is used to process the update process knowledge relationship model and gateway call control strategy set according to the risk control strategy, and to suppress the self-reinforcement of erroneous gateway call control strategies.
