A system and method for optimizing business management processes and controlling risks in small and medium-sized enterprises.

By constructing a dynamic process causal topology graph and local state correction, the problem of inconsistent process states caused by asynchronous data is solved, achieving efficient risk identification and interception, reducing system stagnation and resource consumption, and improving the stability and accuracy of risk interception in the business management processes of SMEs.

CN123117745APending Publication Date: 2026-11-17HENAN UNIV OF URBAN CONSTR
View PDF 1 Cites 0 Cited by

Patent Information

Application Number
CN202611164066.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-03
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing process control systems are prone to timing reversal and state inconsistency when asynchronous data arrives, leading to system stagnation or inaccurate risk identification. Furthermore, existing improvement methods consume computing resources under high concurrency and struggle to balance efficiency and accuracy in risk interception.

Method used

A dynamic process causal topology graph is constructed using an event acquisition and topology parsing module. Data anomalies are identified through a divergence calculation and conflict arbitration module. A multidimensional risk transmission module generates routing interception instructions. Controlled event nodes are isolated to shadow transaction queues through a local state isolation and correction unit. Causal arc weights are attenuated to correct state claim values. An adaptive threshold dead zone smoothing unit is used to adjust the risk interception threshold.

Benefits of technology

It enables accurate identification and isolation of high-risk transactions when asynchronous data arrives, reduces process stagnation, improves the system's anti-interference capability and operational stability, and ensures the continuous operation of normal business.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123117745A_ABST
    Figure CN123117745A_ABST
Patent Text Reader

Abstract

This invention relates to the field of information and communication technology, and discloses a system and method for optimizing business management processes and controlling risks in small and medium-sized enterprises. The system includes: an event acquisition and topology parsing module for receiving asynchronous transaction event vectors, extracting timestamps and transaction values ​​to construct a dynamic process causal topology graph; a divergence calculation and conflict arbitration module for calculating the process state divergence, which is a composite of time-weighted deviation and numerical deviation; a multi-dimensional risk transmission and routing control module for generating routing interception instructions when the divergence exceeds a risk interception threshold; and a local state isolation and correction unit for locking the write permissions of controlled nodes and stripping them to a shadow transaction isolation queue, reverse-compensating state claim values, and attenuating causal arc weights. This invention establishes a topology arbitration mechanism through composite divergence, combined with local shadow isolation and dynamic decoupling, to eliminate cascading deadlocks and improve concurrent throughput efficiency and interception accuracy.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to a system and method for optimizing business management processes and controlling risks in small and medium-sized enterprises, belonging to the field of information and communication technology. Background Technology

[0002] Current conventional process control systems mostly use state machines or preset rule engines that run in a fixed order. For event data such as logistics vouchers, invoice data, and fund flow that are transferred across departments, the system usually requires them to be entered in a preset time order. However, in a multi-source business operation environment, data from different business nodes may be affected by external system interface delays, network transmission fluctuations, or delays in manual review, and may not arrive in a unified time order. When data arrives out of order, it is easy to cause time sequence reversal or incomplete business status records at a certain point in time, making it difficult for process control systems that rely on fixed time sequence to process normally.

[0003] When asynchronously arriving event data causes inconsistencies in the recorded states across different business processes, existing systems typically employ two approaches. One approach is to lock the entire database or pause the main process, awaiting manual verification and data cleanup. This approach easily leads to process stagnation and a large backlog of pending business transactions. The other approach is to continue executing subsequent authorization processes even when business state records are incomplete, potentially resulting in compliance risks such as discrepancies between accounts and actual transactions and overpayments. To mitigate these issues, current improvements typically involve increasing the frequency of global state verification or expanding the scope of distributed transaction locks. However, this approach faces challenges in handling concurrent business processes. When the volume is large, these methods not only consume a lot of computing resources, but also cause continuous blockage of interdependent business links, making it difficult to simultaneously ensure system processing efficiency and risk interception accuracy. For example, Chinese invention patent application with publication number CN120198229B discloses a large model-driven time-series intelligent risk transmission control system and method for capital chains. However, this prior art often relies on the assumption of fixed neighborhood aggregation in multi-hop transmission analysis, ignoring the decay law of capital flow under nonlinear time series and the dynamic evolution of node resistance capabilities. This makes it easy to fail when dealing with hidden and variable capital transfers and lacks a refined blocking mechanism.

[0004] Therefore, how to reduce the reliance on global transaction locks and fixed timing, accurately identify and isolate high-risk transaction value anomalies while allowing for general data delays, and avoid continuous blocking of subsequent business processes after abnormal nodes are paused, has become a technical problem that needs to be solved. Summary of the Invention

[0005] To address the problems in the background art, the technical solution of the present invention is as follows: A business management process optimization and risk control system for small and medium-sized enterprises, comprising:

[0006] The event acquisition and topology parsing module is used to receive asynchronous business transaction event vectors from multiple business nodes, extract timestamps, entity identifiers, process status claims, and transaction values ​​from them, and construct a dynamic process causal topology graph based on entity identifiers and timestamps.

[0007] The divergence calculation and conflict arbitration module is used to obtain the expected process trajectory constraint set and calculate the process state divergence of the dynamic process causal topology graph relative to the expected process trajectory constraint set.

[0008] The multidimensional risk transmission and routing control module is used to generate routing interception instructions when the process state divergence is greater than the risk interception threshold, close the subsequent authorization process that depends on the process state claim value, and generate local state correction vectors based on the dynamic process causal topology graph.

[0009] The local state isolation and correction unit is used to lock the write permissions of controlled event nodes related to the process state claim value to the main process database when a route interception command is received. It then strips the controlled event nodes to the shadow transaction isolation queue, performs reverse compensation on the state claim value of the controlled event nodes according to the local state correction vector, and decays the causal arc weight between the controlled event nodes and downstream dependent nodes according to the difference between the process state divergence and the risk interception threshold, so as to correct the dynamic process causal topology graph.

[0010] Preferably, the local state isolation and correction unit is further used to calculate the residual divergence of the controlled event node after the controlled event node has completed state reverse compensation in the shadow transaction isolation queue, and to ensure that the residual divergence is not greater than the risk interception threshold. At that time, a synthetic compensation arc is constructed to re-stitch the controlled event nodes back into the causal topology of the dynamic process.

[0011] Preferably, the divergence calculation and conflict arbitration module determines the time weight coefficient based on the number of subsequent dependent nodes of the event node in the process. The more dependent nodes there are, the higher the time weighting coefficient becomes. The larger the value, the greater the sum of the time weight coefficients of each event node participating in the evaluation, which is 1.0.

[0012] Preferably, the local state isolation and correction unit is also used to calculate the arc weight decay factor based on the following formula. : And utilize the arc weight decay factor The original causal arc weights between the controlled event node and the downstream dependent node are decayed and updated. When the updated causal arc weights are lower than the determined dependency decoupling threshold, the dependency of the downstream dependent node on the static state value of the controlled event node is cut off, and the downstream dependent node continues to perform parallel calculations by the temporary state baseline generated based on the local state correction vector.

[0013] Preferably, the multidimensional risk transmission and routing control module includes an adaptive threshold dead zone smoothing unit, used to monitor the process state divergence within a defined historical observation window. The sliding variance and sliding mean; the adaptive threshold dead zone smoothing unit only applies to the divergence in the current process state. When the risk continuously exceeds the dead zone boundary range determined by the moving mean and moving variance, the risk interception threshold is triggered. Numerical fine-tuning, while process state divergence Maintain the risk interception threshold when inside the dead zone boundary interval Constant.

[0014] Preferably, the asynchronous business transaction event vector includes a purchase order generation vector, an electronic invoice receipt vector, a goods warehousing inspection vector, and a fund payment initiation vector. The event acquisition and topology parsing module extracts the basic purchase value from the purchase order generation vector, the invoice claim value from the electronic invoice receipt vector, and the acceptance benchmark value from the goods warehousing inspection vector. It then uses the invoice claim value as the claim value of the transaction value item and the acceptance benchmark value as the benchmark verification value of the transaction value item, and transmits them to the divergence calculation and conflict arbitration module. The event acquisition and topology parsing module collects the business chains of each asynchronously arriving event vector based on the entity identifier and establishes directed causal edges according to the order of the timestamps corresponding to each event vector. When the event acquisition and topology parsing module identifies that the timestamp of a subsequent event vector is earlier than the timestamp of an preceding event vector, it determines that there is a time reversal and assigns the corresponding directed causal edge to an initial causal arc weight.

[0015] Preferably, the local state isolation and correction unit generates an independent shadow transaction calculation thread, and performs reverse reconciliation compensation on the difference between the claimed value of the transaction value item and the benchmark verification value of the transaction value item according to the local state correction vector, and generates a corrected reconciliation and cancellation status.

[0016] Preferably, the local state isolation and correction unit has a residual divergence not greater than the risk interception threshold. At that time, the causal arc weight between the controlled event node and the downstream dependent node is restored to the original causal arc weight, and the reconciliation status is injected into the main process data stream through the synthesis of compensation arc, thereby unlocking the controlled event node's write permission to the main process database.

[0017] Preferably, the multidimensional risk transmission and routing control module is also used to calculate the process state divergence within the historical observation window. The system's historical average value is used to determine the risk interception sensitivity. When the historical average value is consistently lower than the established safety divergence baseline, the width of the dead zone boundary interval is increased to improve the system's resistance to network jitter. When the historical average value is consistently higher than the established safety divergence baseline, the width of the dead zone boundary interval is reduced to improve the risk interception sensitivity.

[0018] A method for optimizing business management processes and controlling risks in small and medium-sized enterprises (SMEs), implemented by an SME business management process optimization and risk control system, includes the following steps:

[0019] Step S1: Receive asynchronous business transaction event vectors from multiple business nodes, extract timestamps, entity identifiers, process status claim values, and transaction values ​​from them, and construct a dynamic process causal topology graph based on entity identifiers and timestamps.

[0020] Step S2: Obtain the expected process trajectory constraint set and calculate the process state divergence of the dynamic process causal topology graph relative to the expected process trajectory constraint set.

[0021] Step S3: When the process state divergence is greater than the risk interception threshold, a route interception instruction is generated to close the subsequent authorization process that depends on the process state claim value, and a local state correction vector is generated based on the dynamic process causal topology graph.

[0022] Step S4: Upon receiving the route interception instruction, lock the write permissions of the controlled event nodes related to the process state claim value to the main process database, strip the controlled event nodes to the shadow transaction isolation queue, perform reverse compensation on the state claim value of the controlled event nodes according to the local state correction vector, and reduce the causal arc weight between the controlled event nodes and downstream dependent nodes according to the difference between the process state divergence and the risk interception threshold, so as to correct the dynamic process causal topology graph.

[0023] Compared with the prior art, the beneficial effects of the present invention are:

[0024] 1. In the optimization of business management processes and risk control for SMEs, the event collection and topology analysis module can aggregate business transaction events from different business nodes with inconsistent arrival times and construct a dynamic process causal topology graph; the divergence calculation and conflict arbitration module comprehensively judges time deviation and transaction value deviation, enabling the system to distinguish between general data delays and actual transaction value anomalies. When the process state divergence exceeds the risk interception threshold, it only closes the subsequent authorization process that depends on the abnormal process state claim value, without affecting the continued operation of other normal business branches, thereby reducing the overall process stagnation caused by local anomalies and improving the timeliness of risk identification and interception.

[0025] 2. The local state isolation and correction unit transfers controlled event nodes to the shadow transaction isolation queue, which can prevent abnormal transaction data from continuing to enter subsequent business links such as payment and performance. By attenuating the causal arc weight between controlled event nodes and downstream dependent nodes, and using a temporary state baseline to drive downstream dependent nodes to continue running when the decoupling conditions are met, it can avoid downstream business from being continuously blocked due to waiting for abnormal nodes to be processed, reduce cascading deadlocks, and ensure that the business process can still maintain continuous operation during abnormal processing.

[0026] 3. The adaptive threshold dead zone smoothing unit adjusts the dead zone boundary interval based on the changes in process state divergence within the historical observation window, and only fine-tunes the risk interception threshold when the process state divergence continuously exceeds the interval. This can reduce false interceptions caused by network transmission fluctuations, short-term interface delays, or manual review delays, avoid frequent changes in the risk interception threshold, and enable the system to maintain risk interception sensitivity while having good operational stability and anti-interference capabilities. Attached Figure Description

[0027] Figure 1 This is a flowchart of the commercial transaction risk interception and local state compensation method of the present invention;

[0028] Figure 2 This is a diagram of the asynchronous event topology parsing and multi-dimensional routing control architecture of the present invention.

[0029] The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0030] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0031] A business management process optimization and risk control system for small and medium-sized enterprises includes:

[0032] The event acquisition and topology parsing module is used to receive asynchronous business transaction event vectors from multiple business nodes, extract timestamps, entity identifiers, process status claims, and transaction values ​​from them, and construct a dynamic process causal topology graph based on entity identifiers and timestamps.

[0033] The divergence calculation and conflict arbitration module is used to obtain the expected process trajectory constraint set and calculate the process state divergence of the dynamic process causal topology graph relative to the expected process trajectory constraint set.

[0034] The multidimensional risk transmission and routing control module is used to generate routing interception instructions when the process state divergence is greater than the risk interception threshold, close the subsequent authorization process that depends on the process state claim value, and generate local state correction vectors based on the dynamic process causal topology graph.

[0035] The local state isolation and correction unit is used to lock the write permissions of controlled event nodes related to the process state claim value to the main process database when a route interception command is received. It then strips the controlled event nodes to the shadow transaction isolation queue, performs reverse compensation on the state claim value of the controlled event nodes according to the local state correction vector, and decays the causal arc weight between the controlled event nodes and downstream dependent nodes according to the difference between the process state divergence and the risk interception threshold, so as to correct the dynamic process causal topology graph.

[0036] Preferably, the local state isolation and correction unit is further used to calculate the residual divergence of the controlled event node after the controlled event node has completed state reverse compensation in the shadow transaction isolation queue, and to ensure that the residual divergence is not greater than the risk interception threshold. At that time, a synthetic compensation arc is constructed to re-stitch the controlled event nodes back into the causal topology of the dynamic process.

[0037] Preferably, the divergence calculation and conflict arbitration module determines the time weight coefficient based on the number of subsequent dependent nodes of the event node in the process. The more dependent nodes there are, the higher the time weighting coefficient becomes. The larger the value, the greater the sum of the time weight coefficients of each event node participating in the evaluation, which is 1.0.

[0038] Preferably, the local state isolation and correction unit is also used to calculate the arc weight decay factor based on the following formula. : And utilize the arc weight decay factor The original causal arc weights between the controlled event node and the downstream dependent node are decayed and updated. When the updated causal arc weights are lower than the determined dependency decoupling threshold, the dependency of the downstream dependent node on the static state value of the controlled event node is cut off, and the downstream dependent node continues to perform parallel calculations by the temporary state baseline generated based on the local state correction vector.

[0039] Preferably, the multidimensional risk transmission and routing control module includes an adaptive threshold dead zone smoothing unit, used to monitor the process state divergence within a defined historical observation window. The sliding variance and sliding mean; the adaptive threshold dead zone smoothing unit only applies to the divergence in the current process state. When the risk continuously exceeds the dead zone boundary range determined by the moving mean and moving variance, the risk interception threshold is triggered. Numerical fine-tuning, while process state divergence Maintain the risk interception threshold when inside the dead zone boundary interval Constant.

[0040] Preferably, the asynchronous business transaction event vector includes a purchase order generation vector, an electronic invoice receipt vector, a goods warehousing inspection vector, and a fund payment initiation vector. The event acquisition and topology parsing module extracts the basic purchase value from the purchase order generation vector, the invoice claim value from the electronic invoice receipt vector, and the acceptance benchmark value from the goods warehousing inspection vector. It then uses the invoice claim value as the claim value of the transaction value item and the acceptance benchmark value as the benchmark verification value of the transaction value item, and transmits them to the divergence calculation and conflict arbitration module. The event acquisition and topology parsing module collects the business chains of each asynchronously arriving event vector based on the entity identifier and establishes directed causal edges according to the order of the timestamps corresponding to each event vector. When the event acquisition and topology parsing module identifies that the timestamp of a subsequent event vector is earlier than the timestamp of an preceding event vector, it determines that there is a time reversal and assigns the corresponding directed causal edge to an initial causal arc weight.

[0041] Preferably, the local state isolation and correction unit generates an independent shadow transaction calculation thread, and performs reverse reconciliation compensation on the difference between the claimed value of the transaction value item and the benchmark verification value of the transaction value item according to the local state correction vector, and generates a corrected reconciliation and cancellation status.

[0042] Preferably, the local state isolation and correction unit has a residual divergence not greater than the risk interception threshold. At that time, the causal arc weight between the controlled event node and the downstream dependent node is restored to the original causal arc weight, and the reconciliation status is injected into the main process data stream through the synthesis of compensation arc, thereby unlocking the controlled event node's write permission to the main process database.

[0043] Preferably, the multidimensional risk transmission and routing control module is also used to calculate the process state divergence within the historical observation window. The system's historical average value is used to determine the risk interception sensitivity. When the historical average value is consistently lower than the established safety divergence baseline, the width of the dead zone boundary interval is increased to improve the system's resistance to network jitter. When the historical average value is consistently higher than the established safety divergence baseline, the width of the dead zone boundary interval is reduced to improve the risk interception sensitivity.

[0044] A method for optimizing business management processes and controlling risks in small and medium-sized enterprises (SMEs), implemented by an SME business management process optimization and risk control system, includes the following steps:

[0045] Step S1: Receive asynchronous business transaction event vectors from multiple business nodes, extract timestamps, entity identifiers, process status claim values, and transaction values ​​from them, and construct a dynamic process causal topology graph based on entity identifiers and timestamps.

[0046] Step S2: Obtain the expected process trajectory constraint set and calculate the process state divergence of the dynamic process causal topology graph relative to the expected process trajectory constraint set.

[0047] Step S3: When the process state divergence is greater than the risk interception threshold, a route interception instruction is generated to close the subsequent authorization process that depends on the process state claim value, and a local state correction vector is generated based on the dynamic process causal topology graph.

[0048] Step S4: Upon receiving the route interception instruction, lock the write permissions of the controlled event nodes related to the process state claim value to the main process database, strip the controlled event nodes to the shadow transaction isolation queue, perform reverse compensation on the state claim value of the controlled event nodes according to the local state correction vector, and reduce the causal arc weight between the controlled event nodes and downstream dependent nodes according to the difference between the process state divergence and the risk interception threshold, so as to correct the dynamic process causal topology graph.

[0049] Example 1: The current procurement-to-payment management process for SMEs includes cross-departmental collaboration steps such as purchase order generation, electronic invoice receipt, goods warehousing inspection, and fund payment initiation. Delays in external supplier system interfaces, network transmission fluctuations, and lags in manual review can cause asynchronous arrival of business transaction event data generated by various business nodes, leading to time sequence reversals or inconsistent status data. This example deploys an SME business management process optimization and risk control system on a cloud server. The system includes an event acquisition and topology analysis module, a divergence calculation and conflict arbitration module, a multi-dimensional risk transmission and routing control module, and a local state isolation and correction unit. The standard process time window length... For 100s, the numerical risk scaling factor The risk interception threshold is 0.001. The threshold is 2.5000, the dependency decoupling threshold is 0.4000, the event acquisition and topology parsing module is configured with a data receiving interface, the sampling frequency is 100Hz, and the parsing delay is no higher than 1ms; the divergence calculation and conflict arbitration module is carried by a logic operator running on a 2GHz floating-point processor.

[0050] During a procurement transaction, the system received three asynchronous business transaction event vectors from different business nodes. The total number of event nodes participating in the evaluation is [not specified]. The transaction value items involved in this numerical deviation assessment are a set of correspondences between the invoice claimed values ​​and the acceptance and warehousing benchmark values. Therefore, the total number of transaction value items is... Event node 1 corresponds to the purchase order generation event, observation timestamp 10.0s, expected timestamp The time is 10.0s. The basic purchase value extracted from the purchase order generation vector is 50,000.00 yuan. The order claim value and the corresponding benchmark verification value are both 50,000.00 yuan. Event node 2 corresponds to the electronic invoice receipt event. Due to the transmission delay of the external supplier system interface, its observation timestamp is affected. The expected timestamp is 45.0s. The time is 25.0s; the invoice claim value extracted from the electronic invoice receiving vector is used as the transaction value item claim value in the process status claim value of this event node, denoted as... Its value is 52000.00 yuan. Event node 3 corresponds to the goods receiving inspection event, and the observation timestamp is... The expected timestamp is 30.0s. The value is 30.0s, and its claimed value is 50,000.00 yuan; the acceptance benchmark value extracted from the goods receiving inspection vector is used as the benchmark verification value for the transaction value item, denoted as... Its value is 50,000.00 yuan.

[0051] The divergence calculation and conflict arbitration module counts the number of subsequent dependent nodes for each event node in the current management process. It then divides the number of subsequent dependent nodes for each event node by the total number of subsequent dependent nodes for all nodes participating in the evaluation to obtain the corresponding time weight coefficient. The more subsequent dependent nodes, the larger the time weight coefficient. Based on the node dependencies recorded in this business chain, the time weight coefficients for event node 1, event node 2, and event node 3 are determined. , and The values ​​are 0.2, 0.5, and 0.3 respectively, and they satisfy the normalization condition: The event acquisition and topology parsing module, based on entity identifiers, aggregates asynchronously arriving event vectors into the same business chain. Then, it establishes directed causal edges according to the timestamp order of each event vector, forming a dynamic process causal topology graph. The goods receiving inspection event is a subsequent event to the electronic invoice receiving event. Since the observation timestamp of event node 3 is earlier than that of event node 2, it is determined that there is a time sequence reversal between the two, and the initial causal arc weight of the corresponding directed causal edge is set to... When a time inversion occurs, the initial causal arc weight adopts the dimensionless baseline scalar 1.0000, indicating that the original causal dependency between the two event nodes is maintained; when no time inversion occurs, the original connection relationship in the standard topology graph is maintained.

[0052] The divergence calculation and conflict arbitration module reads the expected process trajectory constraint set from the system memory. This set is stored in a standard business process template configuration file, which records the standard time interval, historical statistical average timestamp, and transaction value matching rules for each business node under normal performance conditions. When the module runs, it reads the key-value correspondence in the configuration file as a comparison benchmark. The weighted time deviation of event node 1 is: The weighted time skew of event node 2 is: The weighted time skew of event node 3 is: The sum of the weighted time deviation terms for the three event nodes is 0.1000, and the deviation of the transaction value term is the absolute value of the difference between the invoice claimed value and the acceptance and warehousing benchmark value. The transaction value deviation item is calculated by first squaring the difference between the claimed invoice value and the acceptance and warehousing benchmark value, then taking the square root of the result, which is 2000.00 yuan. After multiplying by the value risk scaling factor, the transaction value deviation item is: The divergence calculation and conflict arbitration module adds the weighted time deviation term to the transaction value deviation term to obtain the divergence of the current process state. : ,in, The total number of event nodes participating in the evaluation. The total number of transaction value items participating in the evaluation. For the first The time weighting coefficient of each event node For the first The observation timestamps of each event node. For the first The expected timestamp of each event node. The standard process time window length, This is a numerical risk scaling factor. For the first The claimed value of each transaction value item. For the first The baseline verification value for each transaction value item, with the numerical risk scaling factor in reciprocal of yuan, is used to convert the transaction value deviation in yuan into a dimensionless value, allowing it to be added to the time deviation item normalized by the standard process time window. The multi-dimensional risk transmission and routing control module will then use the process state divergence... Risk interception threshold Comparison, because The system does not generate a route interception instruction, and the authorization process continues after fund payment.

[0053] When the invoice claim value of event node 2 in the same business chain If the amount is changed to 58,000.00 yuan, while other input parameters and timestamps remain unchanged, the weighted time deviation remains 0.1000. At this point, the transaction value deviation is: The transaction value deviation item is: Updated process state divergence for: The multi-dimensional risk transmission and routing control module will distribute the process state divergence. Risk interception threshold Comparison, because This module generates a route interception instruction, closes the payment authorization interface that depends on the process status claim value of event node 2, and generates a local state correction vector based on the dynamic process causal topology graph. The three correction components of the local state correction vector are the identification information of the controlled event node, the time deviation between the observed timestamp and the expected timestamp, and the numerical difference between the invoice claim value and the acceptance and storage benchmark value. The vector also carries the forward input relationship, backward dependency node relationship and unique data handle of the controlled event node in the dynamic process causal topology graph, forming a directional correction instruction for the shadow transaction isolation queue to perform state reverse compensation. After receiving the route interception instruction, the local state isolation and correction unit locks the write permission of the controlled event node 2 to the main process database, strips the node to the shadow transaction isolation queue, and stops the transmission of abnormal process status claim values ​​to the fund payment link.

[0054] The local state isolation and correction unit extracts the original causal arc weights between controlled event node 2 and downstream dependent node 3. The arc weight decay factor is calculated based on the difference between the process state divergence and the risk interception threshold. : Calculated After substituting, we get: in, For arc weight decay factor, For process state divergence, As the risk interception threshold, the causal arc weight between controlled event node 2 and its downstream dependent node 3 is updated as follows: The decay of the causal arc weight does not directly control the database lock, but serves as a logical arbitration condition for the routing scheduler. When the updated causal arc weight is lower than the dependency decoupling threshold, the routing scheduler switches the strong dependency flag between the controlled event node and the downstream dependent node to a decoupling state, stops the downstream dependent node from synchronously blocking queries on the real-time process status claim value of the controlled event node, and allows it to read the verified baseline data snapshot. Since the updated causal arc weight of 0.3464 is lower than the dependency decoupling threshold of 0.4000, the system cuts off the dependency of downstream dependent node 3 on the invoice claim value of 58,000.00 yuan in controlled event node 2, and generates a temporary state baseline with the 50,000.00 yuan acceptance and warehousing baseline value as its content based on the local state correction vector. The system provides this temporary state baseline to the calculation thread of the downstream dependent node through a read-only data interface. The temporary state baseline is saved as a read-only snapshot in shared memory without write locks, and only provides the values ​​required for calculation to the downstream dependent node without modifying the original records in the main process database, allowing the downstream dependent node to continue parallel calculation during the isolation of the controlled event node.

[0055] Example 2: In the shadow transaction isolation queue, the local state isolation and correction unit generates an independent shadow transaction calculation thread. Based on the local state correction vector, it performs reverse state compensation on the claimed value of the transaction value item in the process state claim value of controlled event node 2. The shadow transaction calculation thread performs reverse reconciliation compensation according to the difference between the invoice claimed value of 58,000.00 yuan and the acceptance and warehousing benchmark value of 50,000.00 yuan, correcting the claimed value of the transaction value item to 50,000.00 yuan, and generating a reconciliation and cancellation status. After completing the state reverse compensation, the local state isolation and correction unit recalculates the residual divergence of controlled event node 2. At this point, the transaction value deviation term is: The residual divergence is: The synthetic compensation arc is formed by creating a new directed edge in the data structure of the dynamic process causal topology graph, from the shadow transaction isolation queue to the downstream dependent nodes of the controlled event node. The weight of this directed edge records the reconciliation state difference after state reverse compensation. The outgoing edges of the controlled event node are then redirected to the synthetic compensation arc in the graph adjacency list to restore the connectivity of the dynamic process causal topology graph. This is done due to residual divergence. Local state isolation and correction unit constructs synthetic compensation arc The reconciliation and cancellation status is injected into the main process data flow, and the controlled event node 2 is re-stitched into the dynamic process causal topology graph, restoring the original causal arc weights. And unlock the node's write permissions to the main process database.

[0056] When unlocking write permissions and injecting reconciliation status, the system performs concurrent control transaction comparison. For intermediate data generated by downstream dependent nodes during isolation based on the temporary state baseline, the system performs incremental reconciliation verification according to the final reconciliation status carried by the synthetic compensation arc. Both the temporary state baseline and the final reconciliation status use the 50,000.00 yuan acceptance and warehousing benchmark value. Therefore, the intermediate data generated by downstream dependent nodes can be directly merged into the main process database without performing data rollback.

[0057] The multidimensional risk transmission and routing control module is equipped with an adaptive threshold dead zone smoothing unit, which operates within a historical observation window containing 50 sampling periods. Internally, calculate the divergence of the process state. sliding variance and moving average And construct the dead zone boundary interval The adaptive threshold dead zone smoothing unit sets an upper boundary continuous breakthrough counter, a lower boundary continuous breakthrough counter, a preset number of consecutive judgments, and a preset fine-tuning step size. Each time a new process state divergence is obtained, if the value is higher than the upper boundary of the dead zone boundary interval, the upper boundary continuous breakthrough counter is incremented by one, and the lower boundary continuous breakthrough counter is reset to zero; if the value is lower than the lower boundary of the dead zone boundary interval, the lower boundary continuous breakthrough counter is incremented by one, and the upper boundary continuous breakthrough counter is reset to zero; if the value is within the dead zone boundary interval, both counters are reset to zero, and the risk interception threshold remains unchanged. When the upper boundary continuous breakthrough counter reaches... When the preset number of consecutive judgments is reached, the risk interception threshold is increased according to the preset fine-tuning step size; when the lower boundary consecutive breakthrough counter reaches the preset number of consecutive judgments, the risk interception threshold is decreased according to the preset fine-tuning step size. The preset number of consecutive judgments and the preset fine-tuning step size are calibrated offline based on network transmission fluctuations and manual review delay data during normal business operation. Through the above processing, the controlled event node completes state reverse compensation in the shadow transaction isolation queue. The downstream dependent nodes that are not isolated continue to calculate based on the temporary state baseline. After the corrected controlled event node meets the risk interception threshold, it reconnects to the dynamic process causal topology graph.

[0058] Example 3: This example builds a simulation test platform on a cloud server to process multi-source asynchronous commercial transaction event vectors from the procurement to payment management process of SMEs. The test platform uses a multi-core x86 architecture CPU with a main frequency of 2.4GHz, 64GB of memory, and Linux operating system. The system control program and logic operators are written and compiled in C++. The test data is obtained by network packet capture and system log extraction, and contains 1000 sets of asynchronous commercial transaction event vectors. To reproduce the event disorder caused by external system interface delay, manual review lag, and network jitter, Gaussian distributed random delay noise with a mean of 0s and a variance of 5.0s² is superimposed on the timestamps of the event vectors. At the same time, a 5% time-series inversion interference event is added, and abnormal deviations of 100.00 yuan to 10000.00 yuan are randomly injected into the transaction value items.

[0059] During the experiment, the standard process time window length Set to 100.0s, numerical risk scaling factor Set to 0.001, risk interception threshold The baseline value is set to 2.5000, the dependency decoupling threshold is set to 0.4000, and the historical observation window is set to... The experiment was set to 50 sampling periods, with an experimental group and four control groups. The experimental group used the complete system, including dynamic process causal topology graph construction, process state divergence calculation, multi-dimensional risk transmission and routing control, shadow transaction isolation, and causal arc weight decay. Control group 1 only retained deterministic sequence verification and global database locking mechanism, without performing process state divergence calculation or setting shadow transaction isolation. Control group 2 retained process state divergence calculation and routing interception, but removed causal arc weight decay and topology decoupling. Control group 3 set the risk interception threshold. Setting it to 0.5000 is below the lower limit of its reasonable value range; control group 4 sets the risk interception threshold. The value is set to 8.0000, which is higher than the upper limit of its reasonable range.

[0060] Each group ran in parallel using the same dataset and interference conditions, recording the average processing latency, process stall rate, high-risk numerical anomaly interception rate, and false interception rate for the entire process. The process stall rate was the percentage of processes that triggered global locking or deadlock suspension, and the false interception rate was the percentage of processes whose normal network latency was judged as high-risk numerical anomalies.

[0061] In control group 1, the deterministic sequential check triggered a global database lock when asynchronous delay and timing reversal occurred. Its average processing delay was 1420.5ms and the process stall rate was 38.6%. Since the process state divergence was not calculated in this group, the time deviation and the amount abnormality were not combined for judgment. The high-risk value abnormality interception rate was 62.4% and the false interception rate was 35.2%.

[0062] Control group 2 generated routing interception instructions based on process state divergence. The interception rate of high-risk numerical anomalies was 99.1%, the false interception rate was 4.8%, and the average processing latency of the entire process was 850.2ms. This group did not reduce the causal arc weight between controlled event nodes and downstream dependent nodes. After the controlled event nodes were stripped, the downstream dependent nodes still waited for their process state claim values, and the process stall rate was 18.4%.

[0063] After setting the risk interception threshold to 0.5000 in control group 3, the process state dispersion caused by network transmission fluctuations and normal review delays exceeded the threshold. The high-risk value abnormal interception rate was 99.8%, the false interception rate was 28.6%, the frequently generated routing interception commands caused the process stall rate to reach 29.2%, and the average processing delay of the entire process was 1105.8ms.

[0064] After setting the risk interception threshold to 8.0000 in control group 4, the normal latency triggered fewer routing interceptions, the false interception rate was 0.2%, the process stall rate was 0.8%, and the average processing latency for the entire process was 310.4ms; however, the process status divergence corresponding to some abnormal amounts did not exceed the threshold, the interception rate of high-risk abnormal values ​​dropped to 41.5%, and the corresponding abnormal transaction values ​​continued to enter the subsequent authorization process.

[0065] The test group set the risk interception threshold. With a value set to 2.5000 and process state divergence calculation, route interception, shadow transaction isolation, and causal arc weight decay enabled simultaneously, the high-risk numerical anomaly interception rate in the test group was 99.4%, the false interception rate was 0.6%, the average processing latency for the entire process was 325.6ms, and the process stall rate was 1.2%. Process state divergence calculation caused normal latency and monetary anomalies to fall into different judgment intervals. After the controlled event node enters the shadow transaction isolation queue, the downstream dependent nodes continue to calculate based on the temporary state baseline, reducing process stalls caused by dependency waiting.

[0066] In the experimental group, the claimed ticket value for event node 2 was further analyzed. Benchmark verification value of transaction value item The deviations between the values ​​were tested using a gradient increment method, with the weighted time deviation term kept at 0.1000. When the numerical deviation was 0.00, the numerical deviation term was 0.0000. (Process state divergence...) The initial value was 0.1000; after the numerical deviation increased to 2000.00 yuan, the numerical deviation term increased to 2.0000, and the process state divergence... Increased to 2.1000, the divergence of the states of both processes mentioned above is below the risk interception threshold. The multi-dimensional risk transmission and routing control module does not generate routing interception instructions, and the subsequent authorization process remains connected.

[0067] When the numerical deviation continues to increase to 5000.00 yuan, the numerical deviation term is 5.0000, and the process state divergence... The value is 5.1000; when the numerical deviation increases to 8000.00 yuan, the numerical deviation term is 8.0000, and the process state divergence is... When the process state divergence continuously exceeds the risk interception threshold of 8.1000, the adaptive threshold dead zone smoothing unit triggers the threshold adjustment judgment, the multidimensional risk transmission and routing control module generates a routing interception instruction, determines event node 2 as a controlled event node, and closes the subsequent authorization process that depends on the process state claim value of this node.

[0068] process state divergence At that time, the local state isolation and correction unit calculates the arc weight decay factor. : Local state isolation and correction unit utilizes arc weight decay factor The weights of the original causal arcs between controlled event node 2 and its downstream dependent nodes are determined. Attenuation to Since the weight of the decayed causal arc is lower than the dependency decoupling threshold of 0.4000, the system cuts off the dependency of downstream dependent nodes on the static state value of the controlled event node, and generates a temporary state baseline based on the local state correction vector. This temporary state baseline uses a transaction value item benchmark verification value of 50,000.00 yuan, and provides it to the downstream dependent nodes through a read-only interface so that they can continue to perform parallel calculations. The processing latency of the downstream dependent nodes is maintained at 15.2ms.

[0069] In the shadow transaction isolation queue, the local state isolation and correction unit performs reverse state compensation on the process state claim value of controlled event node 2. For the transaction value items, it performs reverse reconciliation compensation based on the difference between the invoice claim value and the benchmark verification value of the transaction value item, and generates a reconciliation and cancellation status. After completing the state reverse compensation, the residual divergence of controlled event node 2 is recalculated as follows: The local state isolation and correction unit constructs a synthetic compensation arc accordingly. The controlled event nodes are re-stitched into the dynamic process causal topology graph, restoring the causal arc weights. The node's write permissions to the main process database are unlocked, thus isolating and correcting the amount anomaly within the controlled event node. Downstream dependent nodes continue to run based on the temporary state baseline during the isolation period, and the corrected reconciliation and cancellation status is then reconnected to the main process data flow.

[0070] Example 4: This example combines Figures 1 to 2 This paper describes a system and method for optimizing business management processes and controlling risks in small and medium-sized enterprises (SMEs). Figure 1 As shown, asynchronous business transaction event vectors from multiple business nodes are received. Timestamps, entity identifiers, process status claims, and transaction values ​​are extracted from these vectors. A dynamic process causal topology graph is constructed based on the entity identifier and timestamp to execute step S1. Subsequently, the expected process trajectory constraint set is obtained, and the process status divergence of the dynamic process causal topology graph relative to the expected process trajectory constraint set is calculated to advance to step S2. Then, when the process status divergence is greater than the risk interception threshold, a route interception instruction is generated to close the subsequent authorization process that depends on the process status claim. A local state correction vector is generated based on the dynamic process causal topology graph to flow to step S3. Finally, when the route interception instruction is received, the write permissions of the controlled event nodes related to the process status claim to the main process database are locked. The controlled event nodes are stripped to the shadow transaction isolation queue. The status claim of the controlled event nodes is reverse-compensated according to the local state correction vector. The causal arc weight between the controlled event nodes and downstream dependent nodes is decayed according to the difference between the process status divergence and the risk interception threshold to correct the dynamic process causal topology graph and complete step S4.

[0071] like Figure 2 As shown, the asynchronous business transaction event vector carries a timestamp, entity identifier, process status claim value, and transaction value, and transmits its data content unidirectionally to the event acquisition and topology parsing module. This module receives the transaction event vector and extracts the timestamp and transaction value to construct a causal topology graph. The topology parsing result is then directed to the divergence calculation and conflict arbitration module. This module obtains the expected trajectory set and calculates the process status divergence to determine the time weight coefficient. Subsequently, the status evaluation result is transmitted to the multidimensional risk transmission and routing control module. At the same time, the adaptive threshold dead zone smoothing unit monitors the sliding variance and mean and triggers the interception threshold fine-tuning, inputting its dynamic control parameters to the multidimensional risk transmission and routing control module. This module integrates the above inputs to generate a routing interception command and closes the subsequent authorization process. Finally, the interception flow signal is transmitted to the local state isolation and correction unit, which drives the local state isolation and correction unit to lock the database write permission and strip the controlled node to the shadow isolation queue, thereby attenuating the causal arc weight and compensating the status claim value in reverse.

[0072] Example 5: Before the system is put into operation, the multidimensional risk transmission and routing control module determines the initial moving mean and initial moving variance of the process state divergence within the historical observation window through offline calibration. The calibration data is taken from continuous historical baseline operation phases. The historical event vector data stream corresponding to each historical business transaction cycle, where The data processing unit calculates the data sequentially according to the time order of each historical business transaction cycle. Process status divergence corresponding to each historical business transaction cycle and with Each sampling period is used as the length of the historical observation window, and the initial moving average is calculated window by window. and initial sliding variance .

[0073] Initial moving average Calculate according to the following formula: ,in, The initial moving average within the historical observation window. This represents the total number of sampling periods included in the historical observation window, with a value of 50. For the first The process state divergence of each historical business transaction cycle is a dimensionless value.

[0074] Initial sliding variance Calculate according to the following formula: ,in, As the initial moving average within the historical observation window, for a historical event vector data stream formed by 1000 historical business transaction cycles, the data processing unit processes the data window by window using 50 sampling periods as one historical observation window, following the same calculation method, and then calculates the initial moving average. and initial sliding variance Write to the memory register to allow the multidimensional risk transmission and routing control module to construct the initial dead zone boundary interval.

[0075] In this embodiment, the initial moving average benchmark from the historical baseline operation phase is used as the safety divergence baseline. After the system enters the online operation state, the multidimensional risk transmission and routing control module continues to use a historical observation window of 50 sampling periods to continuously collect the current process state divergence. Update the moving average and sliding variance The current process state divergence is compared with the dead zone boundary interval. Within a continuous historical observation window, when the historical mean is consistently lower than the safe divergence baseline, the width of the dead zone boundary interval is increased; when the historical mean is consistently higher than the safe divergence baseline, the width of the dead zone boundary interval is decreased.

[0076] When the current process state divergence continuously exceeds the dead zone boundary, the multi-dimensional risk transmission and routing control module triggers the risk interception threshold. Step-by-step fine-tuning; when the current process state divergence is within the dead zone boundary interval, the risk interception threshold... It remains unchanged.

[0077] The risk interception threshold of 2.5000, the dependency decoupling threshold of 0.4000, and the numerical risk scaling factor of 0.001 were all obtained through historical data calibration. Before system deployment, simulation calculations were performed using 1000 sets of historical event vectors. When the risk interception threshold was below 2.0000, the proportion of regular network latency being judged as high-risk anomalies increased; when the risk interception threshold was above 3.0000, the number of uninterrupted high-risk numerical anomalies increased. Therefore, the risk interception threshold was set to 2.5000, the dependency decoupling threshold of 0.4000 was determined from the simulation results of the same set of historical event vectors, and the numerical risk scaling factor of 0.001 was determined based on the ratio of the maximum tolerance error of a single transaction to the length of the standard process time window.

[0078] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0079] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A business management process optimization and risk control system for small and medium-sized enterprises, characterized in that, include: The event acquisition and topology parsing module is used to receive asynchronous business transaction event vectors from multiple business nodes, extract timestamps, entity identifiers, process status claims, and transaction values ​​from them, and construct a dynamic process causal topology graph based on entity identifiers and timestamps. The divergence calculation and conflict arbitration module is used to obtain the expected process trajectory constraint set and calculate the process state divergence of the dynamic process causal topology graph relative to the expected process trajectory constraint set. The multidimensional risk transmission and routing control module is used to generate routing interception instructions when the process state divergence is greater than the risk interception threshold, close the subsequent authorization process that depends on the process state claim value, and generate local state correction vectors based on the dynamic process causal topology graph. The local state isolation and correction unit is used to lock the write permissions of controlled event nodes related to the process state claim value to the main process database when a route interception command is received. It then strips the controlled event nodes to the shadow transaction isolation queue, performs reverse compensation on the state claim value of the controlled event nodes according to the local state correction vector, and decays the causal arc weight between the controlled event nodes and downstream dependent nodes according to the difference between the process state divergence and the risk interception threshold, so as to correct the dynamic process causal topology graph.

2. The SME business management process optimization and risk control system according to claim 1, characterized in that, The local state isolation and correction unit is also used to calculate the residual divergence of the controlled event node after the controlled event node has completed state reversal compensation in the shadow transaction isolation queue, and to ensure that the residual divergence is not greater than the risk interception threshold. At that time, a synthetic compensation arc is constructed to re-stitch the controlled event nodes back into the dynamic process causal topology graph.

3. The SME business management process optimization and risk control system according to claim 1, characterized in that, The divergence calculation and conflict arbitration module determines the time weighting coefficient based on the number of subsequent dependent nodes of an event node in the process. The more dependent nodes there are, the higher the time weighting coefficient becomes. The larger the value, the greater the sum of the time weight coefficients of each event node participating in the evaluation, which is 1.

0.

4. The SME business management process optimization and risk control system according to claim 1, characterized in that, The local state isolation and correction unit is also used to calculate the arc weight decay factor based on the following formula. : And utilize the arc weight decay factor The original causal arc weights between the controlled event node and its downstream dependent nodes are decayed and updated. When the updated causal arc weight is lower than the determined dependency decoupling threshold, the dependency of downstream dependent nodes on the static state value of the controlled event node is cut off, and the downstream dependent nodes continue to perform parallel computation by the temporary state baseline generated based on the local state correction vector.

5. The SME business management process optimization and risk control system according to claim 1, characterized in that, The multidimensional risk transmission and routing control module includes an adaptive threshold dead zone smoothing unit, used to monitor the process state divergence within a defined historical observation window. The moving variance and moving mean; The adaptive threshold dead zone smoothing unit only applies to the divergence of the current process state. When the risk continuously exceeds the dead zone boundary range determined by the moving mean and moving variance, the risk interception threshold is triggered. Numerical fine-tuning, while process state divergence Maintain the risk interception threshold when inside the dead zone boundary interval Constant.

6. The SME business management process optimization and risk control system according to claim 1, characterized in that, The asynchronous business transaction event vector includes a purchase order generation vector, an electronic invoice receipt vector, a goods receiving inspection vector, and a payment initiation vector. The event acquisition and topology parsing module extracts the basic purchase value from the purchase order generation vector, the invoice claim value from the electronic invoice receipt vector, and the goods receiving inspection benchmark value from the goods receiving inspection vector. It then uses the invoice claim value as the claim value of the transaction value item and the goods receiving benchmark value as the benchmark verification value of the transaction value item, and transmits them to the divergence calculation and conflict arbitration module. The event acquisition and topology parsing module aggregates the business chains of each asynchronously arriving event vector based on the entity identifier and establishes directed causal edges according to the chronological order of the timestamps corresponding to each event vector. When the event acquisition and topology parsing module identifies that the timestamp of a subsequent event vector is earlier than that of an preceding event vector, it determines that there is a time reversal and assigns initial causal arc weights to the corresponding directed causal edges.

7. The SME business management process optimization and risk control system according to claim 4, characterized in that, The local state isolation and correction unit generates an independent shadow transaction calculation thread, and performs reverse reconciliation compensation for the difference between the claimed value of the transaction value item and the benchmark verification value of the transaction value item according to the local state correction vector, generating a corrected reconciliation and cancellation status.

8. The SME business management process optimization and risk control system according to claim 7, characterized in that, Local state isolation and correction units are used when the residual divergence is no greater than the risk interception threshold. At that time, the causal arc weight between the controlled event node and the downstream dependent node is restored to the original causal arc weight, and the reconciliation status is injected into the main process data stream through the synthesis of compensation arc, thereby unlocking the controlled event node's write permission to the main process database.

9. The SME business management process optimization and risk control system according to claim 5, characterized in that, The multidimensional risk transmission and routing control module is also used to calculate the process state divergence within the historical observation window. The system's historical average value is used to determine the risk interception sensitivity. When the historical average value is consistently lower than the established safety divergence baseline, the width of the dead zone boundary interval is increased to improve the system's resistance to network jitter. When the historical average value is consistently higher than the established safety divergence baseline, the width of the dead zone boundary interval is reduced to improve the risk interception sensitivity.

10. A method for optimizing business management processes and controlling risks in small and medium-sized enterprises (SMEs), implemented by the SME business management process optimization and risk control system described in claim 1, characterized in that, Includes the following steps: Step S1: Receive asynchronous business transaction event vectors from multiple business nodes, extract timestamps, entity identifiers, process status claim values, and transaction values ​​from them, and construct a dynamic process causal topology graph based on entity identifiers and timestamps. Step S2: Obtain the expected process trajectory constraint set and calculate the process state divergence of the dynamic process causal topology graph relative to the expected process trajectory constraint set. Step S3: When the process state divergence is greater than the risk interception threshold, a route interception instruction is generated to close the subsequent authorization process that depends on the process state claim value, and a local state correction vector is generated based on the dynamic process causal topology graph. Step S4: Upon receiving the route interception instruction, lock the write permissions of the controlled event nodes related to the process state claim value to the main process database, strip the controlled event nodes to the shadow transaction isolation queue, perform reverse compensation on the state claim value of the controlled event nodes according to the local state correction vector, and reduce the causal arc weight between the controlled event nodes and downstream dependent nodes according to the difference between the process state divergence and the risk interception threshold, so as to correct the dynamic process causal topology graph.

Citation Information

Patent Citations

  • Large model driven time series capital chain intelligent risk transmission control system and method

    CN120198229B