A method and system for dynamic workflow reconfiguration based on multi-source architecture change awareness

Abstract

This invention relates to the field of workflow refactoring management technology, specifically disclosing a method and system for dynamic workflow refactoring based on multi-source architecture change perception. The method includes: firstly, sensing and fusing multi-source heterogeneous data through an intelligent agent workflow system; dynamically adjusting data processing strategies based on fusion process parameters to improve the consistency and availability of multi-source data; then, monitoring the fused data to capture multi-source architecture change signals and obtaining an adaptive adjustment parameter set for multi-source architecture changes; finally, parsing the parameter set and iteratively adjusting the architecture change process to achieve a closed loop for dynamic workflow refactoring, shortening process refactoring time. This method constructs a full-link system of "data fusion - architecture perception - dynamic refactoring," enhancing the intelligent agent workflow system's response and adaptability to multi-source environmental changes, improving process execution efficiency, reducing process blockage risks, and providing technical support for the dynamic management of complex businesses.

Description

Technical Field

[0001] This invention relates to the field of workflow refactoring management technology, specifically to a method and system for dynamic workflow refactoring based on multi-source architecture change awareness. Background Technology

[0002] With the deepening of digital transformation, business processes in enterprises and industrial scenarios are becoming increasingly complex. Intelligent agent workflow systems need to process heterogeneous data from multiple sources such as sensors, databases, and cloud services, and respond quickly to architectural changes to maintain efficient operation. Existing technologies have conducted numerous studies on dynamic workflow reconstruction based on awareness of multi-source architectural changes: at the data fusion level, methods such as early fusion, recursive fusion, and dynamic weight adjustment, combined with edge computing and GPU acceleration technologies, are used to achieve real-time integration and preprocessing of multi-source data; in terms of dynamic workflow adjustment, based on event-driven architecture and the BPMN 2.0 standard, using the publish-subscribe pattern and stream processing engine, combined with compiled, interpreted intelligent agents and multi-agent federated learning technologies, real-time monitoring and adaptive optimization of processes are achieved.

[0003] For example, the invention patent with publication number CN118134410A discloses a flexible workflow management method based on the ESR meta-model, which includes a workflow meta-model of events, states, and rules. This meta-model can effectively describe various workflow requirements, especially the relationship between processes and business functions, as well as dynamic changes in the workflow environment. Based on the ESR meta-model, a flexible workflow management method is constructed, which includes process and activity management, business element mapping management, rule base management, workflow scheduling management, and workflow refactoring management.

[0004] For example, the invention patent with publication number CN118967055A discloses a workflow management method, apparatus, device, and computer-readable storage medium, which is applied in the field of workflow management. The method includes: dynamically configuring approval rules and form permissions for each process node based on the Activity workflow engine to obtain a created workflow; binding the created workflow with the corresponding target form; and when a user initiates a target form, executing the workflow processing work corresponding to the target form based on the Activity workflow engine.

[0005] However, in the process of implementing the embodiments of this application, it was found that the above-mentioned technology has at least the following technical problems: In actual business scenarios, it is difficult to trigger the update of process rules in real time based on changes in management elements such as departmental adjustments and job changes, which leads to the inability to update workflows and task allocations in a timely manner, exacerbating the risk of process blockage and affecting team efficiency and project progress. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides a method and system for dynamic workflow reconfiguration based on multi-source architecture change awareness, which can effectively solve the problems mentioned in the background technology.

[0007] To achieve the above objectives, the present invention provides the following technical solution: The first aspect of the present invention provides a method for dynamic workflow reconstruction based on multi-source architecture change perception, comprising: Step 1, perceiving multi-source heterogeneous data through an intelligent agent workflow system, fusing the multi-source heterogeneous data, analyzing the parameters of the multi-source heterogeneous data fusion process, and determining whether to adjust the multi-source heterogeneous data fusion process of the intelligent agent workflow system; Step 2, monitoring the fused multi-source heterogeneous data, perceiving multi-source architecture changes through the intelligent agent workflow system, and obtaining an adaptive adjustment parameter set for the multi-source architecture changes; Step 3, parsing the adaptive adjustment parameter set for the multi-source architecture changes through the intelligent agent workflow system, and determining whether to cyclically adjust the multi-source architecture change process of the intelligent agent workflow system, thereby achieving dynamic workflow reconstruction.

[0008] The second aspect of this invention provides a workflow dynamic reconstruction system based on multi-source architecture change perception, comprising: a multi-source data fusion perception and adjustment module, used to perceive multi-source heterogeneous data through an intelligent agent workflow system, fuse the multi-source heterogeneous data, analyze the parameters of the multi-source heterogeneous data fusion process, and thereby determine whether to adjust the multi-source heterogeneous data fusion process of the intelligent agent workflow system; an architecture change monitoring and parameter acquisition module, used to monitor the fused multi-source heterogeneous data, the intelligent agent workflow system perceives multi-source architecture changes, and acquires an adaptive adjustment parameter set for the multi-source architecture changes; and an architecture change parsing and workflow reconstruction module, used by the intelligent agent workflow system to parse the adaptive adjustment parameter set for the multi-source architecture changes, and determine whether to cyclically adjust the multi-source architecture change process of the intelligent agent workflow system, thereby realizing dynamic workflow reconstruction.

[0009] Compared with the prior art, the embodiments of the present invention have at least the following advantages or beneficial effects:

[0010] (1) This invention provides a workflow dynamic reconstruction method and system based on multi-source architecture change perception, and constructs a full-link adaptive system of "data fusion-architecture perception-dynamic reconstruction". It realizes the intelligent agent workflow system's accurate response and efficient adaptation to changes in multi-source heterogeneous environment, which can improve process execution efficiency and task allocation rationality, significantly reduce the risk of process blockage caused by changes in data source or architecture, and provide enterprise workflow systems with full closed-loop dynamic management technology support from change perception, intelligent decision-making to process reconstruction and execution when dealing with complex business scenarios such as organizational structure adjustment and personnel changes. It ensures the continuity and efficiency of workflow under changes in multi-source heterogeneous environment, and helps enterprise workflow systems achieve digital operation and maintenance and agile iteration.

[0011] (2) By sensing and fusing multi-source heterogeneous data through the intelligent agent workflow system, and analyzing the parameters of the fusion process, the data processing strategy can be dynamically optimized. This step effectively solves the fusion problem caused by the differences in format, semantics and timing of multi-source data. By adjusting the parameters, the accuracy of data fusion is improved, and workflow execution errors caused by data quality problems are reduced. This provides standardized and consistent data input support for subsequent processes such as task scheduling and permission allocation in the enterprise workflow system.

[0012] (3) Monitor the fused data in real time and perceive changes in the multi-source architecture, accurately obtain the adaptive adjustment parameter set, so that the system can quickly capture signals when the architecture changes (such as organizational structure adjustment), ensure the system's sensitivity and response speed to environmental changes, and avoid task interruption or resource waste caused by architecture changes.

[0013] (4) The adaptive adjustment parameter set is parsed and the multi-source architecture change process of the workflow is cyclically adjusted to realize the dynamic reconstruction closed loop of the workflow. This mechanism can automatically adjust the task allocation rules, permission system and process logic according to the architecture change, so that the system can still maintain stable operation when facing frequent architecture changes. This step can shorten the process reconstruction time and significantly improve the continuous adaptability and business continuity of the system. Attached Figure Description

[0014] The present invention will be further described with reference to the accompanying drawings, but the embodiments in the drawings do not constitute any limitation on the present invention. For those skilled in the art, other drawings can be obtained based on the following drawings without creative effort.

[0015] Figure 1 This is a schematic diagram of the method steps of the present invention.

[0016] Figure 2 This is a schematic diagram of the system module connections of the present invention.

[0017] Figure 3 This is a schematic diagram of the data flow of the present invention.

[0018] Figure 4 This is a schematic diagram of the intelligent agent workflow system architecture of the present invention.

[0019] Figure 5 This is a schematic diagram of the workflow steps for sensing and adjusting according to the present invention. Detailed Implementation

[0020] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0021] Reference Figure 1 As shown, the first aspect of the present invention provides a method for dynamic workflow reconstruction based on multi-source architecture change perception, comprising: Step 1, perceiving multi-source heterogeneous data through an intelligent agent workflow system, fusing the multi-source heterogeneous data, analyzing the parameters of the multi-source heterogeneous data fusion process, and determining whether to adjust the multi-source heterogeneous data fusion process of the intelligent agent workflow system; Step 2, monitoring the fused multi-source heterogeneous data, perceiving multi-source architecture changes through the intelligent agent workflow system, and obtaining an adaptive adjustment parameter set for the multi-source architecture changes; Step 3, parsing the adaptive adjustment parameter set for the multi-source architecture changes through the intelligent agent workflow system, and determining whether to cyclically adjust the multi-source architecture change process of the intelligent agent workflow system, thereby realizing dynamic workflow reconstruction.

[0022] Specifically, the parameters of the multi-source heterogeneous data fusion process are analyzed. The specific analysis process includes the event correlation accuracy factor, conflict resolution success rate factor, and information gain factor.

[0023] The aforementioned event association accuracy factor represents the relative level of event association accuracy and event definition accuracy of multi-source heterogeneous data within the fusion monitoring period, and is taken as the ratio of the two; the aforementioned conflict resolution success rate factor represents the relative level of conflict resolution success rate and event definition conflict resolution success rate of multi-source heterogeneous data within the fusion monitoring period, and is taken as the ratio of the two; the aforementioned information gain factor represents the relative level of information gain and definition information gain of multi-source heterogeneous data within the fusion monitoring period, and is taken as the ratio of the two.

[0024] Influence coefficients are introduced from the reconstructed database to quantify the impact of event association accuracy factor, conflict resolution success rate factor, and information gain factor on the multi-source heterogeneous data fusion coefficient. These coefficients are then coupled to obtain the multi-source heterogeneous data fusion coefficient. The multi-source heterogeneous data fusion coefficient represents the degree of fusion of data from different data sources with different structures and formats. The specific evaluation method is as follows:

[0025] ;

[0026] ;

[0027] ;

[0028] ;

[0029] In the formula, PYTN is the fusion coefficient of multi-source heterogeneous data, NML is the event association accuracy factor, KLO is the event association accuracy of multi-source heterogeneous data within the fusion monitoring period, FGT is the pre-defined event association accuracy in the reconstruction database, GYU is the conflict resolution success rate factor, YRK is the conflict resolution success rate of multi-source heterogeneous data within the fusion monitoring period, WEU is the pre-defined conflict resolution success rate in the reconstruction database, UYV is the information gain factor, VGF is the information gain of multi-source heterogeneous data within the fusion monitoring period, OIU is the pre-defined information gain in the reconstruction database, τ1 is the influence coefficient corresponding to the pre-defined event association accuracy factor in the reconstruction database, τ2 is the influence coefficient corresponding to the pre-defined conflict resolution success rate factor in the reconstruction database, and τ3 is the influence coefficient corresponding to the pre-defined information gain factor in the reconstruction database.

[0030] The aforementioned event association accuracy is a core indicator for measuring the correctness of cross-data source event matching during the fusion of multi-source heterogeneous data. It is achieved by standardizing and extracting features from multi-source data, using machine learning algorithms (such as deep learning algorithms) to achieve event association, and calculating the event association accuracy based on the percentage of correctly associated event logs. The aforementioned conflict resolution success rate is an indicator for measuring the system's ability to handle data contradictions or inconsistencies during the fusion of multi-source heterogeneous data. It is achieved by detecting numerical, temporal, or semantic conflicts in multi-source data, using machine learning to resolve them, and calculating the conflict resolution success rate based on the percentage of successfully resolved conflicts out of the total number of conflicts. The aforementioned information gain is an indicator for measuring the degree to which newly added information improves the system's decision-making or cognition during the data fusion process. It is obtained by calculating the absolute value of the conditional entropy after introducing new information from the original entropy of the target variable.

[0031] The above definition of event association accuracy rate represents the lower limit of the specified range of event association accuracy rate; the above definition of conflict resolution success rate represents the lower limit of the specified range of conflict resolution success rate; the above definition of information gain represents the lower limit of the specified range of information gain.

[0032] The influence coefficient corresponding to the aforementioned event association accuracy factor refers to the degree of influence of the unit value of the event association accuracy factor on the fusion coefficient of multi-source heterogeneous data; the influence coefficient corresponding to the aforementioned conflict resolution success rate factor refers to the degree of influence of the unit value of the conflict resolution success rate factor on the fusion coefficient of multi-source heterogeneous data; the influence coefficient corresponding to the aforementioned information gain factor refers to the degree of influence of the unit value of the information gain factor on the fusion coefficient of multi-source heterogeneous data. The reconstruction database stores the mapping relationship between the event association accuracy factor and its corresponding influence coefficient, the mapping relationship between the conflict resolution success rate factor and its corresponding influence coefficient, and the mapping relationship between the information gain factor and its corresponding influence coefficient. For example, when the event association accuracy factor, the conflict resolution success rate factor, and the information gain factor are imported into the reconstruction database, the reconstruction database will generate the influence coefficients corresponding to the event association accuracy factor, the conflict resolution success rate factor, and the information gain factor based on preset mapping rules, and the numerical range of each coefficient is strictly controlled between 0 and 1.

[0033] The more accurate the event association (such as correctly matching the same entity event in multi-source data), the more precise the object of conflict detection, reducing "pseudo-conflicts" caused by association errors, thereby improving the success rate of conflict resolution; high information gain data sources (such as those containing timestamps) can provide more discriminative information, helping the algorithm to more accurately associate cross-source events; high information gain data sources can provide more basis for conflict resolution (such as credibility labels), making resolution strategies (such as machine learning) more efficient.

[0034] A higher event association accuracy factor indicates a higher actual event association accuracy relative to the threshold value, meaning higher matching accuracy for similar events in multi-source data. This provides a more reliable foundation for data fusion and directly promotes an increase in the fusion coefficient. A higher conflict resolution success rate factor indicates a greater actual conflict resolution success rate exceeding the threshold value, improving the efficiency of data contradiction handling, enhancing consistency, reducing noise interference during the fusion process, and positively impacting the fusion coefficient. A higher information gain factor represents a higher actual information gain relative to the threshold value, meaning a larger amount of effective information provided by the data source. This more accurately supports event association and conflict resolution, improving the overall effectiveness of the fusion system from a data value perspective.

[0035] In one specific embodiment, the intelligent agent workflow system constructs a dynamic and adaptive data processing mechanism through real-time perception and deep fusion of multi-source heterogeneous data. This mechanism relies on precise analysis of core parameters such as event association accuracy, conflict resolution success rate, and information gain during the fusion process to intelligently identify potential problems in data format conversion, semantic mapping, and time alignment. By dynamically adjusting the conflict tolerance threshold and calibration time window threshold strategy, it significantly improves the accuracy and reliability of data fusion, effectively avoiding risks such as process blockage and task mismatch caused by data quality defects, and laying a solid data foundation for the stable operation and efficient execution of subsequent workflows.

[0036] Specifically, the determination of whether to adjust the multi-source heterogeneous data fusion process of the agent workflow system involves the following steps: comparing the multi-source heterogeneous data fusion coefficient with the preset multi-source heterogeneous data fusion threshold in the reconstruction database; if the multi-source heterogeneous data fusion coefficient is greater than or equal to the multi-source heterogeneous data fusion threshold, it is determined that the multi-source heterogeneous data fusion process of the agent workflow system will not be adjusted, and the adaptive coefficient correction coefficient of the agent workflow system will be matched from the reconstruction database based on the multi-source heterogeneous data fusion coefficient, thereby correcting the adaptive coefficient of the agent workflow system, thereby improving the collaborative efficiency of the agent workflow system and enhancing environmental adaptability; at the same time, the adaptive threshold of the agent workflow system will be matched based on the multi-source heterogeneous data fusion coefficient, enabling the agent workflow system to dynamically adjust the triggering conditions of the optimization strategy according to the data quality, thereby improving resource allocation efficiency and decision reliability.

[0037] The aforementioned multi-source heterogeneous data fusion threshold represents the lower limit of the allowed multi-source heterogeneous data fusion coefficient within a specified range in the reconstruction database. The aforementioned matching of the agent workflow system's adaptive coefficient correction coefficient from the reconstruction database based on the multi-source heterogeneous data fusion coefficient refers to pre-building a fusion coefficient interval-correction coefficient mapping table in the reconstruction database, and directly matching the corresponding correction coefficient according to the interval where the real-time multi-source heterogeneous data fusion coefficient is located. A higher multi-source heterogeneous data fusion coefficient indicates better data fusion quality, allowing the system to withstand greater adaptive adjustments (such as more aggressive process reconstruction), requiring an increase in the agent workflow system's adaptive coefficient. The coefficient is adjusted to enhance the adjustment power of the adaptive coefficient. The adaptive threshold of the agent workflow system is matched based on the fusion coefficient of multi-source heterogeneous data. The specific matching process is as follows: extract the correspondence between historical fusion coefficients and adaptive thresholds from the reconstructed database (as shown in the table), and match the adaptive threshold of the agent workflow system corresponding to the current multi-source heterogeneous data fusion coefficient through interpolation. The higher the multi-source heterogeneous data fusion coefficient, the better the data fusion quality, and the more complex organizational changes the system can withstand (such as high-frequency departmental adjustments). It is necessary to increase the adaptive threshold of the agent workflow system to trigger a more stringent adaptive strategy and avoid over-adjustment.

[0038] If the multi-source heterogeneous data fusion coefficient is less than the multi-source heterogeneous data fusion threshold, it is determined that the multi-source heterogeneous data fusion process of the intelligent agent workflow system should be adjusted. The specific adjustment process is as follows: based on the multi-source heterogeneous data fusion coefficient and the multi-source heterogeneous data fusion threshold, the multi-source heterogeneous data fusion deviation value is obtained. Based on the multi-source heterogeneous data fusion deviation value, the time alignment window length increase coefficient is matched to increase the time alignment window length of multi-source data acquisition and improve the cross-source data time synchronization accuracy.

[0039] The aforementioned method of obtaining the multi-source heterogeneous data fusion deviation value refers to subtracting the multi-source heterogeneous data fusion coefficient from the multi-source heterogeneous data fusion threshold, dividing the result by the multi-source heterogeneous data fusion threshold, and finally obtaining the multi-source heterogeneous data fusion deviation value. The specific matching process for matching the time alignment window length amplification factor based on the multi-source heterogeneous data fusion deviation value is as follows: The time alignment window length amplification factor corresponding to each multi-source heterogeneous data fusion deviation value interval is stored in the reconstruction database. The obtained multi-source heterogeneous data fusion deviation value is input into the reconstruction database, which then matches the corresponding multi-source heterogeneous data fusion deviation value interval. The time alignment window length amplification factor corresponding to this interval is the required amplification factor. Multiplying the obtained time alignment window length amplification factor by the original time alignment window length yields the time alignment window length to be adjusted. A time alignment window length amplification factor greater than 1 indicates that the time alignment window length for multi-source data acquisition needs to be increased by a certain factor.

[0040] After adjusting the multi-source heterogeneous data fusion process of the intelligent agent workflow system, the secondary fusion coefficient of the multi-source heterogeneous data is obtained, and it is determined whether the multi-source heterogeneous data fusion process of the intelligent agent workflow system needs to be adjusted a second time. The aforementioned secondary fusion coefficient of the multi-source heterogeneous data represents the multi-source heterogeneous data fusion coefficient that is re-obtained after adjusting the multi-source heterogeneous data fusion process of the intelligent agent workflow system.

[0041] Furthermore, it is determined whether to perform a secondary adjustment to the multi-source heterogeneous data fusion process of the agent workflow system. Specifically, the determination process involves comparing the secondary fusion coefficient of the multi-source heterogeneous data with the multi-source heterogeneous data fusion threshold. If the secondary fusion coefficient is greater than or equal to the multi-source heterogeneous data fusion threshold, it is determined that no secondary adjustment will be performed on the multi-source heterogeneous data fusion process of the agent workflow system. Based on the secondary fusion coefficient, the adaptive coefficient correction coefficient of the agent workflow system is matched from the reconstruction database, thereby correcting the adaptive coefficient of the agent workflow system. The matching process is the same as described above. The adaptive coefficient and correction coefficient of the intelligent agent workflow system are matched from the reconstructed database based on the multi-source heterogeneous data fusion coefficient. Simultaneously, the adaptive threshold of the intelligent agent workflow system is matched based on the secondary fusion coefficient of the multi-source heterogeneous data, and the matching process is consistent with the above-mentioned matching of the adaptive threshold of the intelligent agent workflow system based on the multi-source heterogeneous data fusion coefficient. If the secondary fusion coefficient of the multi-source heterogeneous data is less than the multi-source heterogeneous data fusion threshold, it is determined that the multi-source heterogeneous data fusion process of the intelligent agent workflow system will be adjusted a second time. The specific adjustment process is as follows: based on the secondary fusion coefficient of the multi-source heterogeneous data and the multi-source heterogeneous data fusion threshold, obtain the multi-source heterogeneous data... The secondary fusion deviation value of heterogeneous data is matched with a time alignment window length amplification factor, thereby further increasing the time alignment window length of multi-source data acquisition. This effectively compensates for fusion misalignment caused by differences in data time series, enabling precise alignment of time series data from different sources over a longer time dimension and improving the temporal consistency of cross-source data. Simultaneously, a conflict tolerance threshold reduction factor is matched with the secondary fusion deviation value of heterogeneous data, thereby reducing the conflict tolerance threshold of data values. This significantly enhances the sensitivity of conflict detection by lowering the tolerance for data value conflicts, enabling timely detection. It also intercepts potential data quality issues such as inconsistent formats and semantic ambiguity, preventing low-quality data from flowing into subsequent workflow stages. The secondary adjustment mechanism, through a dual strategy of "dynamic expansion of the time window + enhanced conflict detection accuracy," not only addresses the challenges of fusion of multi-source data at the temporal and semantic levels, but also forms a closed loop of "detection-feedback-optimization" through adaptive parameter adjustment. This allows the data fusion process to continuously evolve with changes in business scenarios, ultimately providing higher-quality data input for the intelligent agent workflow system, effectively reducing the risk of process interruptions caused by data inconsistency, and improving the stability and decision accuracy of workflow execution.

[0042] The aforementioned method of obtaining the secondary fusion deviation value of multi-source heterogeneous data refers to subtracting the secondary fusion coefficient of multi-source heterogeneous data from the multi-source heterogeneous data fusion threshold, dividing the result by the multi-source heterogeneous data fusion threshold, and finally obtaining the secondary fusion deviation value of multi-source heterogeneous data. The aforementioned matching of the time alignment window length double-increase coefficient based on the secondary fusion deviation value of multi-source heterogeneous data involves the following matching process: The database stores the time alignment window length double-increase coefficients corresponding to each interval of the secondary fusion deviation value of multi-source heterogeneous data. The obtained secondary fusion deviation values ​​of multi-source heterogeneous data are input into the database, and the database can match the corresponding interval of the secondary fusion deviation value of multi-source heterogeneous data. The time alignment window length double-increase coefficient corresponding to this interval is the required double-increase coefficient. Multiplying the obtained time alignment window length double-increase coefficient by the original time alignment window length yields the time alignment window length to be adjusted. A time alignment window length quadratic increase factor greater than 1 indicates that the time alignment window length of multi-source data acquisition needs to be increased by a certain factor. It's important to note that this quadratic increase factor is greater than the time alignment window length increase factor. The above-mentioned matching of conflict tolerance threshold reduction factor based on the secondary fusion deviation value of multi-source heterogeneous data involves the following matching process: The conflict tolerance threshold reduction factor corresponding to each multi-source heterogeneous secondary data fusion deviation value interval is stored in the reconstruction database. The obtained multi-source heterogeneous secondary data fusion deviation values ​​are input into the reconstruction database, which then matches the corresponding multi-source heterogeneous secondary data fusion deviation value interval. The conflict tolerance threshold reduction factor corresponding to this interval is the required reduction factor. Multiplying the obtained conflict tolerance threshold reduction factor by the original conflict tolerance threshold yields the required adjustment to the conflict tolerance threshold. A conflict tolerance threshold reduction factor less than 1 indicates that the conflict tolerance threshold of the data value needs to be reduced by a certain percentage.

[0043] After the second adjustment, the tertiary fusion coefficient of the multi-source heterogeneous data is obtained and compared with the multi-source heterogeneous data fusion threshold. If the tertiary fusion coefficient of the multi-source heterogeneous data is greater than or equal to the multi-source heterogeneous data fusion threshold, no warning is required. At the same time, based on the tertiary fusion coefficient of the multi-source heterogeneous data, the adaptive coefficient correction coefficient of the intelligent agent workflow system is matched from the reconstruction database to correct the adaptive coefficient of the intelligent agent workflow system. The matching process is the same as the above-mentioned matching of the adaptive coefficient correction coefficient of the intelligent agent workflow system from the reconstruction database based on the tertiary fusion coefficient of the multi-source heterogeneous data. The adaptive threshold of the intelligent agent workflow system is matched based on the tertiary fusion coefficient of the multi-source heterogeneous data, and the matching process is the same as the above-mentioned matching of the adaptive threshold of the intelligent agent workflow system based on the tertiary fusion coefficient of the multi-source heterogeneous data. If the tertiary fusion coefficient of the multi-source heterogeneous data is less than the multi-source heterogeneous data fusion threshold, the intelligent agent workflow system is monitored in real time, and the abnormality of the intelligent agent workflow system is traced.

[0044] It should be explained that the above-mentioned three-stage fusion coefficient of multi-source heterogeneous data represents the fusion coefficient of multi-source heterogeneous data obtained again after complete two-stage adjustment.

[0045] Reference Figure 3 As shown in the data flow diagram of this invention, the HR database generates change logs, which are then transferred to the stream processing engine via the event bus. The stream processing engine divides the data into historical archives (stored in the data lake for offline training) and real-time features (input into the decision knowledge base). The offline training results of the data lake and the content of the decision knowledge base enter the federated learning layer, and after model updates, they are applied to the agent cluster. The agent cluster outputs control commands to the business system API, and after the business system API executes them, it feeds back the results to the feedback analysis. The feedback analysis uses a reinforcement learning mechanism to send optimization information back to the agent cluster, thereby realizing the complete process of data-driven agent adjustment of workflow.

[0046] Figure 3 In the middle, the federated learning layer refers to cross-departmental model training that protects privacy; feedback analysis refers to continuous optimization to achieve an instruction accuracy of 95%+; and the data lake refers to storing 180 days of historical change records for backtracking.

[0047] Specifically, the adaptive adjustment parameter set for multi-source architecture changes is obtained. The specific analysis process is as follows: the adaptive adjustment parameter set for multi-source architecture changes includes the permission inheritance index, workflow reorganization index, collaborative optimization gain, and drift error rate.

[0048] The aforementioned permission inheritance index refers to the ratio of the average latency of permission synchronization in the agent workflow system to the average latency of defined permission synchronization, subtracted from 1, to measure the real-time performance of permission inheritance. The aforementioned workflow reorganization index refers to the product of the ratio of the downtime of the agent workflow system to the downtime of the defined process, multiplied by the workflow reorganization accuracy of the agent workflow system, and then subtracted from 1. The aforementioned collaborative optimization gain refers to the degree of deviation between the baseline cycle and the current cycle of the agent workflow system. The aforementioned drift error rate refers to the ratio between the number of organization-process inconsistency events in the agent workflow system and the total number of change events in the agent workflow system.

[0049] Influence coefficients are introduced from the reconstructed database to quantify the impact of permission inheritance index, workflow reorganization index, collaborative optimization gain, and drift error rate on the adaptive coefficient of the agent workflow system. These influence levels are coupled to obtain the adaptive coefficient of the agent workflow system. The adaptive coefficient of the agent workflow system represents a core indicator for quantifying the dynamic adaptability of the agent workflow system to organizational structure changes and its cross-entity collaborative efficiency. The corrected adaptive coefficient of the agent workflow system is obtained and labeled as the adaptive correction index of the agent workflow system. The specific evaluation method is as follows:

[0050] ;

[0051] ;

[0052] ;

[0053] ;

[0054] ;

[0055] ;

[0056] In the formula, OAW is the adaptive coefficient of the agent workflow system, OAW_M is the adaptive correction index of the agent workflow system, ξ is the adaptive coefficient correction coefficient of the agent workflow system matched in the reconstruction database, Pi is the permission inheritance index of the agent workflow system, BPG is the average delay of permission synchronization of the agent workflow system, Tp is the preset average delay of permission synchronization in the reconstruction database, Wr is the workflow reorganization index of the agent workflow system, REW is the process downtime of the agent workflow system, Tw is the preset process downtime in the reconstruction database, Ra is the workflow reorganization accuracy of the agent workflow system, and C o represents the collaborative optimization gain of the agent workflow system, NPL represents the baseline cycle of the agent workflow system, NMD represents the current cycle of the agent workflow system, Ed represents the drift error rate of the agent workflow system, FLP represents the number of organization-process inconsistency events of the agent workflow system, DSW represents the total number of change events of the agent workflow system, α represents the influence coefficient corresponding to the preset permission inheritance index in the refactoring database, β represents the influence coefficient corresponding to the preset workflow reorganization index in the refactoring database, γ represents the influence coefficient corresponding to the preset collaborative optimization gain in the refactoring database, and λ represents the influence coefficient corresponding to the preset drift error rate in the refactoring database.

[0057] It should be explained that the above-mentioned average latency of permission synchronization refers to the average time taken for a permission change instruction to be transmitted from the triggering source to the target system and take effect in a permission management scenario. This is calculated by recording the initiation timestamp at the permission change triggering end and the effective timestamp at the receiving end, calculating the latency for each single change, and averaging the latency over multiple changes. The above-mentioned process downtime refers to the cumulative time during workflow execution when the process is stalled due to various anomalies (such as data blocking). This is calculated by extracting the start and end timestamps of the process stall from the intelligent agent workflow system logs, calculating the duration for each stall, and then summing them up. The above-mentioned workflow reorganization accuracy refers to the accuracy of the reorganized process after the intelligent agent workflow system has adjusted the process structure (such as optimizing task allocation) and executes correctly according to the expected logic. The proportion is obtained by using production logs to statistically analyze the percentage of times the reorganized process was executed correctly as expected, based on preset process execution rules; the above-mentioned baseline period refers to the standard reference time for process execution in the intelligent agent workflow system, and the baseline period can be obtained from the preset value in the process design configuration; the above-mentioned current period refers to the actual execution time of the workflow process, obtained by real-time monitoring of the process instance time; the above-mentioned number of organization-process inconsistency events refers to the number of events in the intelligent agent workflow system where the organizational structure (such as departments) does not match the process definition logic, obtained by comparing the organizational entities referenced by the process with the number of currently valid organizational structures; the above-mentioned total number of change events refers to the sum of all types of change operations in the intelligent agent workflow system, obtained by statistically analyzing all change operation records in the system audit log.

[0058] The above definition of average delay for permission synchronization represents the upper limit of the average delay for permission synchronization within the specified range. The above definition of process downtime represents the upper limit of the process downtime within the specified range.

[0059] The influence coefficients corresponding to the aforementioned permission inheritance index represent the degree of influence of a unit value of the permission inheritance index on the adaptive coefficient of the agent workflow system; the influence coefficients corresponding to the aforementioned workflow reorganization index represent the degree of influence of a unit value of the workflow reorganization index on the adaptive coefficient of the agent workflow system; the influence coefficients corresponding to the aforementioned collaborative optimization gain represent the degree of influence of a unit value of the collaborative optimization gain on the adaptive coefficient of the agent workflow system; and the influence coefficients corresponding to the aforementioned drift error rate represent the degree of influence of a unit value of the drift error rate on the adaptive coefficient of the agent workflow system. The mapping relationships between the permission inheritance index and its corresponding influence coefficient, the workflow reorganization index and its corresponding influence coefficient, the collaborative optimization gain and its corresponding influence coefficient, and the drift error rate and its corresponding influence coefficient are stored in the reconstruction database. For example, importing the permission inheritance index, workflow reorganization index, collaborative optimization gain, and drift error rate into the reconstruction database will generate the corresponding influence coefficients for the permission inheritance index, workflow reorganization index, collaborative optimization gain, and drift error rate based on preset mapping rules, with the values ​​of all coefficients strictly controlled between 0 and 1.

[0060] A higher permission inheritance index means that permissions for new nodes take effect quickly during workflow refactoring, reducing downtime caused by missing permissions and thus improving workflow refactoring efficiency. A higher permission inheritance index also ensures clear and seamless permissions in cross-departmental collaboration, smooth task handover, shorter process cycles, and a higher likelihood of positive and larger collaborative optimization gains (better collaborative optimization effect). A higher drift error rate means that during workflow refactoring, numerous conflicts arising from "mismatches between organizational structure and process logic" need to be addressed, leading to lower refactoring accuracy, increased downtime, reduced collaborative optimization gains, and poor stability. A higher workflow refactoring index results in a more rational process structure (e.g., streamlined nodes), smoother task execution, shorter cycles, and more significant positive collaborative optimization gains.

[0061] A higher permission inheritance index indicates faster and more accurate permission synchronization, a healthier system permission governance status, and a better adaptive coefficient for the intelligent agent workflow system. A higher workflow reorganization index indicates better process optimization quality, higher process reorganization efficiency, less downtime, and a positive reinforcement of the adaptive coefficient for the intelligent agent workflow system (such as accelerating the promotion of new processes). When the collaborative optimization gain is positive and its value increases, it reflects improved collaborative efficiency, and the adaptive coefficient of the intelligent agent workflow system will be positively amplified. When the collaborative optimization gain is negative or shows a downward trend, it indicates reduced collaborative efficiency, and the system needs to adjust its collaborative strategy, and the adaptive coefficient of the intelligent agent workflow system will be negatively corrected. A higher drift error rate indicates more organizational-process conflicts, and the system judges "poor organizational adaptability," causing the adaptive coefficient of the intelligent agent workflow system to fluctuate negatively.

[0062] In a specific example embodiment, the values ​​of the permission inheritance index, workflow reorganization index, and drift error rate range from 0 to 1, while the value of the collaborative optimization gain ranges from -1 to 1. The influence coefficients corresponding to the permission inheritance index, workflow reorganization index, collaborative optimization gain, and drift error rate are all set to 1 by default. The default average latency for defining permission synchronization is 10 seconds. The default process downtime is 60 seconds. If the process downtime is greater than 60 seconds, the workflow reorganization index is recorded as 0. It is recommended to adjust the influence coefficient based on different business priorities. For example, if the business priority is high security and compliance, it is recommended to increase λ to strengthen the drift penalty.

[0063] In one specific embodiment, the fused data is monitored in real time across all dimensions. Relying on the intelligent perception module, it can keenly capture subtle changes in the multi-source architecture. With the help of a precise algorithm model, it can quickly deduce and obtain a set of adjustment parameters adapted to the current scenario. This allows the system to capture the signal the moment an architectural change (such as an organizational restructuring) occurs. A highly efficient closed loop is formed from data collection to strategy response, ensuring that the system always maintains a keen perception of environmental changes and a millisecond-level rapid response speed. This fundamentally prevents task interruptions and resource waste caused by architectural changes, ensuring the continuous and stable operation of business processes.

[0064] Specifically, the determination of whether to cyclically adjust the multi-source architecture change process of the agent workflow system is as follows: the adaptive correction index of the agent workflow system is compared with the preset adaptive threshold of the agent workflow system in the reconstruction database; if the adaptive correction index of the agent workflow system is greater than or equal to the adaptive threshold of the agent workflow system, it is determined that the multi-source architecture change process of the agent workflow system should not be cyclically adjusted; if the adaptive correction index of the agent workflow system is less than the adaptive threshold of the agent workflow system, the abnormality of the agent workflow system is traced, and it is determined that the multi-source architecture change process of the agent workflow system should be cyclically adjusted.

[0065] It should be explained that the aforementioned adaptive threshold of the intelligent agent workflow system refers to the lower limit of the adaptive coefficient of the intelligent agent workflow system extracted from the reconstructed database within a specified range.

[0066] Furthermore, anomaly tracing is performed on the agent workflow system. The specific tracing process is as follows: The perception delay duration is obtained and compared with a preset perception delay threshold in the reconstruction database. If the perception delay duration is greater than the threshold, the agent workflow system anomaly type is marked as a perception layer anomaly; otherwise, it is not marked as a perception layer anomaly. The false reconstruction rate is obtained and compared with a preset false reconstruction rate threshold in the reconstruction database. If the false reconstruction rate is greater than the threshold, the agent workflow system anomaly type is marked as a decision layer anomaly; otherwise, it is not marked as a decision layer anomaly. The abnormal type is marked as a decision-level anomaly; the reconstruction power is obtained and compared with the preset reconstruction power threshold in the reconstruction database. If the reconstruction power is less than the reconstruction power threshold, the abnormal type of the agent workflow system is marked as an execution-level anomaly; otherwise, it is not marked as an execution-level anomaly; the policy optimization gain is obtained and compared with the preset policy optimization gain threshold in the reconstruction database. If the policy optimization gain is less than the policy optimization gain threshold, the abnormal type of the agent workflow system is marked as a learning-level anomaly; otherwise, it is not marked as a learning-level anomaly.

[0067] By dividing anomaly tracing into four dimensions—perception, decision-making, execution, and learning—it can fully cover the key links of the entire intelligent agent workflow system, from "data perception to decision analysis to execution control to strategy evolution." This avoids the omission of learning layer evolution capability assessment in a three-level division or the redundancy of indicators in a five-level division. Furthermore, it can accurately locate the source of anomalies through independent indicators at each level, and construct a closed-loop optimization mechanism of "detection-location-repair-learning" through the linkage of indicators at all four levels. This achieves a balance between systematicity and efficiency while ensuring the accuracy of anomaly location.

[0068] It should be explained that the aforementioned perception delay duration represents the time interval from the actual occurrence of a change in the multi-source architecture (such as the issuance of an organizational structure adjustment instruction) to the system successfully capturing the change signal. This is obtained by recording timestamps at both the architecture change source and the system perception module, and calculating the time difference between the occurrence of the change and the signal reception. The aforementioned false reconstruction rate represents the proportion of times the system erroneously triggers an architecture reconstruction in scenarios where there is no actual architecture change or no reconstruction is required. This is obtained by filtering the number of falsely triggered reconstructions and the total number of reconstructions in the system reconstruction log and calculating the ratio between the two. The aforementioned reconstruction success rate represents the proportion of times the core functions such as task flow and permission allocation run normally after the system performs an architecture reconstruction. This is obtained by automatically verifying the status of core functions after the architecture reconstruction and statistically analyzing the percentage of successful reconstructions. The aforementioned strategy optimization gain represents the improvement in business indicators (such as task processing efficiency) after the system performs adaptive strategy optimization (such as adjusting weight coefficients). This is obtained by comparing business indicators (such as efficiency) before and after strategy optimization and calculating the improvement.

[0069] The aforementioned perception latency threshold represents the lower limit of perception latency within the specified range in the reconstruction database; the aforementioned false reconstruction rate threshold represents the upper limit of false reconstruction rate within the specified range in the reconstruction database; the aforementioned reconstruction power threshold represents the lower limit of reconstruction power within the specified range in the reconstruction database; and the aforementioned policy optimization gain threshold represents the lower limit of policy optimization gain within the specified range in the reconstruction database.

[0070] Specifically, the process of iteratively adjusting the multi-source architecture change process of the intelligent agent workflow system is as follows: If the anomaly type of the intelligent agent workflow system is a perception layer anomaly, then based on the adaptive correction index and adaptive threshold of the intelligent agent workflow system, the adaptive deviation value of the intelligent agent workflow system is obtained. Based on the adaptive deviation value of the intelligent agent workflow system, the event sampling frequency increase coefficient is matched from the reconstruction database, thereby increasing the event sampling frequency of the intelligent agent workflow system, controlling the event capture density, and improving the detection rate. Based on the adaptive deviation value of the intelligent agent workflow system, the data aggregation time window decrease coefficient is matched from the reconstruction database, thereby reducing the data aggregation time window of the intelligent agent workflow system, balancing real-time performance and data integrity, and reducing perception latency. This collaborative adjustment mechanism of "high-frequency sampling + short-window aggregation" can accurately solve the anomaly problem caused by sparse sampling or processing lag in the perception layer, providing timely and accurate data support for workflow reconstruction, and effectively enhancing the system's response efficiency and robustness to multi-source architecture changes.

[0071] The aforementioned method of obtaining the adaptive deviation value of the agent workflow system refers to subtracting the adaptive correction index of the agent workflow system from its adaptive threshold, then dividing the result by the adaptive threshold. The final ratio is the adaptive deviation value of the agent workflow system. The method of matching the event sampling frequency enhancement coefficient from the reconstruction database based on the adaptive deviation value is as follows: The reconstruction database stores the event sampling frequency enhancement coefficients corresponding to each adaptive deviation value interval of the agent workflow system. The obtained adaptive deviation values ​​are input into the reconstruction database, which then matches the corresponding adaptive deviation value interval. The event sampling frequency enhancement coefficient corresponding to this interval is the required enhancement coefficient. Multiplying the obtained event sampling frequency enhancement coefficient by the original event sampling frequency yields the adaptive deviation value of the agent workflow system. The workflow system needs to adjust to a specific event sampling frequency; the event sampling frequency increase factor greater than 1 refers to the numerical value by which the event sampling frequency needs to be increased; the data aggregation window reduction factor matched from the reconstruction database based on the adaptive deviation value of the agent workflow system is as follows: the reconstruction database stores the data aggregation window reduction factor corresponding to the adaptive deviation value interval of each agent workflow system; the obtained adaptive deviation value of the agent workflow system is input into the reconstruction database, and the reconstruction database can match the corresponding adaptive deviation value interval of the agent workflow system. The data aggregation window reduction factor corresponding to this interval is the required reduction factor. The obtained data aggregation window reduction factor is multiplied by the original data aggregation window, and the result is the data aggregation window that the agent workflow system needs to adjust to; the data aggregation window reduction factor less than 1 refers to the proportion by which the event data aggregation window needs to be reduced.

[0072] If the anomaly type of the agent workflow system is a decision-level anomaly, then the consensus threshold reduction coefficient is matched from the reconstruction database based on the adaptive deviation value of the agent workflow system, thereby reducing the consensus threshold of the agent workflow system, controlling the efficiency of distributed decision-making, reducing decision latency, and directly matching the policy exploration rate improvement value from the reconstruction database based on the adaptive deviation value of the agent workflow system, thereby improving the policy exploration rate of the agent workflow system, balancing the use of experience and the discovery of new policies, and improving the reconstruction recall rate.

[0073] The above-mentioned consensus threshold reduction coefficient is obtained by matching the adaptive deviation value of the agent workflow system from the reconstruction database. The specific matching process is as follows: The reconstruction database stores the consensus threshold reduction coefficients corresponding to the adaptive deviation value intervals of each agent workflow system. The obtained adaptive deviation values ​​of the agent workflow systems are input into the reconstruction database, which then matches the corresponding adaptive deviation value intervals. The consensus threshold reduction coefficient corresponding to this interval is the required reduction coefficient. Multiplying the obtained consensus threshold reduction coefficient by the original consensus threshold yields the consensus threshold that the agent workflow system needs to adjust to. The consensus threshold reduction coefficient being less than 1 indicates that the consensus threshold is... The percentage reduction required to achieve the threshold; the above-mentioned policy exploration rate improvement value is directly matched from the reconstruction database based on the adaptive deviation value of the agent workflow system. The specific matching process is as follows: the reconstruction database stores the policy exploration rate improvement value corresponding to the adaptive deviation value range of each agent workflow system. The obtained adaptive deviation value of the agent workflow system is input into the database, and the database can match the corresponding adaptive deviation value range of the agent workflow system. The policy exploration rate improvement value corresponding to this range is the required improvement value. The original policy exploration rate is added to the obtained policy exploration rate improvement value, and the result is the policy exploration rate that needs to be adjusted. The above-mentioned policy exploration rate improvement value being greater than 0 refers to the specific value by which the policy exploration rate needs to be increased.

[0074] If the anomaly type of the agent workflow system is an execution layer anomaly, then the resource buffer pool capacity increase coefficient is matched from the reconstruction database based on the agent workflow system's adaptive deviation value, thereby increasing the resource buffer pool capacity of the agent workflow system, reserving resources to cope with sudden demands, and improving the reconstruction success rate. The state snapshot interval reduction coefficient is matched from the reconstruction database based on the agent workflow system's adaptive deviation value, thereby reducing the state snapshot interval of the agent workflow system and balancing the state saving overhead and recovery speed.

[0075] The above-mentioned resource buffer pool capacity increase coefficient is matched from the reconstruction database based on the adaptive deviation value of the agent workflow system. The specific matching process is as follows: The reconstruction database stores the resource buffer pool capacity increase coefficients corresponding to the adaptive deviation value intervals of each agent workflow system. The obtained adaptive deviation values ​​of the agent workflow system are input into the reconstruction database, which then matches the corresponding adaptive deviation value intervals. The resource buffer pool capacity increase coefficient corresponding to this interval is the required increase coefficient. Multiplying the obtained resource buffer pool capacity increase coefficient by the original resource buffer pool capacity yields the adjusted resource buffer pool capacity of the agent workflow system. A resource buffer pool capacity increase coefficient greater than 1 indicates that the resource buffer pool capacity needs to be increased. The numerical value of the multiple; the above-mentioned state snapshot interval reduction coefficient is matched from the reconstruction database based on the adaptive deviation value of the intelligent agent workflow system. The specific matching process is as follows: the reconstruction database stores the state snapshot interval reduction coefficient corresponding to the adaptive deviation value range of each intelligent agent workflow system. The obtained adaptive deviation value of the intelligent agent workflow system is input into the reconstruction database, and the reconstruction database can match the corresponding adaptive deviation value range of the intelligent agent workflow system. The state snapshot interval reduction coefficient corresponding to this range is the required reduction coefficient. The obtained state snapshot interval reduction coefficient is multiplied by the original state snapshot interval, and the result is the state snapshot interval that the intelligent agent workflow system needs to adjust to. The above-mentioned state snapshot interval reduction coefficient is less than 1, which means that the state snapshot interval needs to be reduced by a certain percentage.

[0076] If the anomaly type of the agent workflow system is a learning layer anomaly, then based on the adaptive deviation value of the agent workflow system, a knowledge sharing frequency improvement coefficient is matched from the reconstruction database to increase the knowledge sharing frequency of the agent workflow system, control the weight of historical experience in training, and improve the handling rate of unknown scenarios. Based on the adaptive deviation value of the agent workflow system, an experience replay sampling rate improvement coefficient is matched from the reconstruction database to increase the experience replay sampling rate of the agent workflow system, control the weight of historical experience in training, and improve the policy optimization gain.

[0077] The above-mentioned method of matching the knowledge sharing frequency improvement coefficient from the reconstruction database based on the adaptive deviation value of the intelligent agent workflow system is as follows: The reconstruction database stores the knowledge sharing frequency improvement coefficients corresponding to the adaptive deviation value intervals of each intelligent agent workflow system. The obtained adaptive deviation values ​​of the intelligent agent workflow system are input into the reconstruction database, which then matches the corresponding adaptive deviation value intervals. The knowledge sharing frequency improvement coefficient corresponding to this interval is the required improvement coefficient. Multiplying the obtained knowledge sharing frequency improvement coefficient by the original knowledge sharing frequency yields the knowledge sharing frequency that the intelligent agent workflow system needs to adjust to. A knowledge sharing frequency improvement coefficient greater than 1 indicates that the knowledge sharing frequency needs to be increased by a multiple. The adaptive deviation value of the agent workflow system is matched with the empirical replay sampling rate improvement coefficient from the reconstruction database. The specific matching process is as follows: The reconstruction database stores the empirical replay sampling rate improvement coefficients corresponding to the adaptive deviation value intervals of each agent workflow system. The obtained adaptive deviation values ​​of the agent workflow system are input into the reconstruction database, and the reconstruction database can match the corresponding adaptive deviation value intervals of the agent workflow system. The empirical replay sampling rate improvement coefficient corresponding to this interval is the required improvement coefficient. The obtained empirical replay sampling rate improvement coefficient is multiplied by the original empirical replay sampling rate, and the result is the empirical replay sampling rate that the agent workflow system needs to be adjusted to. The above empirical replay sampling rate improvement coefficient is greater than 1, which means that the empirical replay sampling rate needs to be increased by a factor of 1.

[0078] If no anomaly type exists in the intelligent agent workflow system, an early warning is issued for the multi-source architecture change process of the intelligent agent workflow system. The number of cyclic adjustments for each anomaly type of the intelligent agent workflow system is obtained to determine whether an early warning should be issued for each anomaly type. The number of cyclic adjustments for each anomaly type of the intelligent agent workflow system represents the cumulative number of times the intelligent agent workflow system triggers the "adjustment-detection-readjustment" cycle when handling different types of anomalies. This is achieved by embedding log points in the anomaly handling process to record each adjustment cycle event, storing the data in the anomaly handling log table, and then using structured query language to aggregate and statistically calculate the total number of cycles by anomaly type. Early warning is issued based on the number of cyclic adjustments for each anomaly type. High-frequency adjustments can identify potential risks: analyzing the adjustment frequency distribution can locate potential bottlenecks in architecture changes, and combining historical data modeling can predict risk probabilities, triggering parameter optimization in advance (such as adjusting the aggregation time window), realizing a shift from passive response to proactive prevention, reducing the risk of sudden failures, and improving the long-term stability and adaptability of the system.

[0079] In one specific embodiment, by deeply analyzing the adaptive adjustment parameter set, a cyclic adjustment mechanism for multi-source architecture changes is constructed, forming a closed-loop dynamic reconfiguration system for workflow. This mechanism can intelligently adjust task allocation rules, permission system, and process logic according to architecture changes, so that the system can always maintain a stable operating state during high-frequency architecture iterations. This process significantly reduces the time spent on process reconfiguration, significantly enhances the system's continuous adaptive capability and business continuity assurance, and achieves seamless coordination between architecture changes and workflow execution.

[0080] Reference Figure 4 As shown in the schematic diagram of the intelligent agent workflow system architecture of the present invention, the HR system / organizational data source pushes change events, which are processed into standardized events by the event perception engine. The standardized events are transformed into organizational topology relationships by the knowledge graph builder, and at the same time, the prediction model library generates optimization suggestions based on log data. The organizational topology relationships and optimization suggestions are jointly input into the intelligent decision center. The intelligent decision center outputs control instructions to the workflow executor (and provides feedback on operation information after execution), and also outputs permission adjustment instructions to the ABE encryption module to generate dynamic keys. Finally, the dynamic keys are applied to the business system (CRM / OA), and the relevant data of the business system is also fed back for subsequent analysis and optimization, thereby realizing the dynamic adaptation and control of the business system after organizational structure changes.

[0081] Figure 4 In this context, the event-aware engine refers to monitoring changes in the HR system via Webhook; the knowledge graph builder refers to maintaining a three-dimensional relationship network of "personnel-department-project" in real time; the ABE encryption module refers to implementing dynamic access control for attributes; and the prediction model library refers to containing seven types of machine learning models, including process bottleneck prediction.

[0082] Specifically, the process for determining whether to issue warnings for each anomaly type is as follows: If the number of cyclic adjustments for a certain anomaly type in the intelligent agent workflow system equals the defined number of cyclic adjustments, then the adjustment parameters for that anomaly type in the intelligent agent workflow system are obtained and defined for adjustment. After adjustment, it is determined whether the anomaly type still exists in the intelligent agent workflow system. If it still exists, a warning is issued for that anomaly type in the intelligent agent workflow system. If it does not exist, the working status of the intelligent agent workflow system is continuously monitored. At the same time, the threshold for multi-source heterogeneous data fusion is increased based on the final adaptive coefficient of the intelligent agent workflow system. By improving the data fusion screening criteria (such as data consistency requirements), anomalies caused by data quality issues are reduced. If the number of cyclic adjustments for a certain anomaly type in the intelligent agent workflow system is less than the defined number of cyclic adjustments, the working status of the intelligent agent workflow system is continuously monitored.

[0083] The aforementioned definition of the number of cyclic adjustments refers to the upper limit of the allowed number of cyclic adjustments within the specified range in the reconstructed database; the aforementioned definition of adjustment refers to adjusting the adjustment parameters of the intelligent agent workflow system to the upper limit of the specified range for that anomaly type; the aforementioned adjustment parameters include knowledge sharing frequency, resource buffer pool capacity, etc.; the aforementioned provision of early warning prompts for the intelligent agent workflow system to that anomaly type refers to sending early warning information for that anomaly type to the operation and maintenance platform of the intelligent agent workflow system; the aforementioned improvement of the multi-source heterogeneous data fusion threshold based on the final intelligent agent workflow system adaptive coefficient refers to the fact that the larger the adaptive coefficient of the intelligent agent workflow system, the greater the increase in the multi-source heterogeneous data fusion threshold, thereby reducing the risk of anomaly recurrence by dynamically strengthening the data fusion standard.

[0084] In one specific embodiment, the present invention provides a workflow dynamic reconstruction method based on multi-source architecture change awareness, establishing a full-link adaptive mechanism of "data fusion-architecture awareness-dynamic reconstruction". This solution enables the intelligent agent workflow system to accurately respond to and efficiently adapt to multi-source heterogeneous environments through multi-dimensional data fusion and real-time perception of architecture changes, effectively improving process execution efficiency and task allocation rationality. At the same time, the dynamic reconstruction mechanism significantly reduces the risk of process blockage caused by data source or architecture changes, providing reliable technical support for the dynamic management of complex business scenarios in enterprise digital transformation.

[0085] Reference Figure 5 As shown in the flowchart of the workflow perception and adjustment method of the present invention, when an HR system change event occurs, the first step is to enter the change type detection stage. If it is determined to be a departmental structure adjustment, topology recalculation and workflow DAG reconstruction will be carried out sequentially. If it is determined to be a personnel change, the impact scope analysis will be performed first, and then further differentiated processing based on whether it is a key position. If it is a key position, a three-level response mechanism will be triggered, a handover report will be generated simultaneously, and a security audit will be triggered. If it is not a key position, automatic task migration and minimum permission allocation will be executed. After completing the above corresponding branch processes, the multi-system synchronous execution stage will be entered. After the execution is completed, completion confirmation will be carried out. Finally, log archiving and model training will be carried out to achieve full-process control and subsequent optimization closed loop of HR system change events.

[0086] Figure 5 In this context, Level 3 response refers to triggering additional security audits for positions such as CTO / CFO; DAG reconstruction refers to maintaining process acyclicity based on graph algorithms; and permission minimization refers to adhering to the NIST zero-trust architecture principles.

[0087] In a specific example implementation, when a change in project manager is detected, the system triggers an automated collaboration mechanism using an intelligent agent workflow: First, it captures manager change events (such as changes in AD account status or job ID updates) in real time via the HR system / organizational structure API to achieve change detection. Then, based on knowledge graph analysis of the projects, approval processes, and sensitive documents involved in the manager's work, it generates a handover list to complete the impact assessment, achieving real-time perception and dynamic response (topology awareness). Next, it recommends the most suitable successor through a skill graph (such as Git commit records and past project experience). If no successor is specified, it automatically nominates a temporary manager for candidate matching based on "reporting priority + workload balancing," while automatically inheriting the original manager's project permissions (such as Jira administrator and Confluence space permissions). Attribute-based encryption (ABE) is used to dynamically adjust access rights to sensitive documents, enabling intelligent task reassignment (policy migration). In-progress tasks (such as approval delays or unclosed pull requests) are automatically associated with the new responsible party, and historical operation records are retained. Differential analysis is used to mark high-risk incomplete items (such as milestones nearing deadlines) and push specific reminders to synchronize task status. Communication records (emails / Slack) and document editing history of the original responsible party are extracted to generate a handover summary (NLP summary + key node timeline). If critical information is missing (such as password not being transferred), the process is frozen until manual confirmation, thus achieving context-aware handover assurance (anomaly blocking). This fully realizes the automatic transfer of tasks and documents to the new responsible party and simultaneously notifies relevant team members.

[0088] Reference Figure 2 As shown, the second aspect of the present invention provides a workflow dynamic reconstruction system based on multi-source architecture change perception, including: a multi-source data fusion perception and adjustment module, an architecture change monitoring and parameter acquisition module, an architecture change parsing and workflow reconstruction module, and a reconstruction database.

[0089] The multi-source data fusion perception and adjustment module is connected to the architecture change monitoring and parameter acquisition module, and the architecture change monitoring and parameter acquisition module is connected to the architecture change parsing and workflow reconstruction module. All three are connected to the reconstruction database. The reconstruction database is used to store various parameters involved in the workflow dynamic reconstruction system based on multi-source architecture change perception.

[0090] The multi-source data fusion sensing and adjustment module is used to sense multi-source heterogeneous data through the intelligent agent workflow system, fuse the multi-source heterogeneous data, analyze the parameters of the multi-source heterogeneous data fusion process, and thus determine whether to adjust the multi-source heterogeneous data fusion process of the intelligent agent workflow system. The architecture change monitoring and parameter acquisition module is used to monitor the fused multi-source heterogeneous data, the intelligent agent workflow system senses multi-source architecture changes, and acquires the adaptive adjustment parameter set of the multi-source architecture changes. The architecture change parsing and workflow reconstruction module is used by the intelligent agent workflow system to parse the adaptive adjustment parameter set of multi-source architecture changes, and determine whether to cyclically adjust the multi-source architecture change process of the intelligent agent workflow system, thereby realizing the dynamic reconstruction of the workflow.

[0091] The above description is merely an example and illustration of the structure of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the structure of the invention or exceed the scope defined by the present invention, and all such modifications and additions should fall within the protection scope of the present invention.

Claims

1. A workflow dynamic reconstruction method based on multi-source architecture change awareness, characterized in that, include: Step 1: The intelligent agent workflow system senses multi-source heterogeneous data and fuses it. The parameters of the multi-source heterogeneous data fusion process are analyzed to determine whether the multi-source heterogeneous data fusion process of the intelligent agent workflow system should be adjusted. The multi-source heterogeneous data fusion process parameters include event association accuracy factor, conflict resolution success rate factor, and information gain factor. An influence coefficient is introduced from the reconstructed database to quantify the degree of influence of the event association accuracy factor, conflict resolution success rate factor, and information gain factor on the multi-source heterogeneous data fusion coefficient. The influence degrees are coupled to obtain the multi-source heterogeneous data fusion coefficient. The multi-source heterogeneous data fusion coefficient represents the degree of fusion of data from different data sources with different structures and formats. Step 2: Monitor the fused multi-source heterogeneous data. The intelligent agent workflow system senses changes in the multi-source architecture and obtains an adaptive adjustment parameter set for these changes. This set includes a permission inheritance index, a workflow reorganization index, a collaborative optimization gain, and a drift error rate. Influence coefficients are introduced from the reconstructed database to quantify the impact of the permission inheritance index, workflow reorganization index, collaborative optimization gain, and drift error rate on the adaptive coefficient of the intelligent agent workflow system. These influence levels are coupled to obtain the adaptive coefficient of the intelligent agent workflow system. The adaptive coefficient of the intelligent agent workflow system represents a core indicator for quantifying and evaluating the dynamic adaptability of the intelligent agent workflow system to organizational structure changes and its cross-entity collaborative efficiency. The corrected adaptive coefficient of the intelligent agent workflow system is obtained and labeled as the adaptive correction index of the intelligent agent workflow system. Step 3: The intelligent agent workflow system analyzes the adaptive adjustment parameter set of the multi-source architecture change and determines whether to cyclically adjust the multi-source architecture change process of the intelligent agent workflow system, thereby realizing the dynamic reconstruction of the workflow.

2. The workflow dynamic reconstruction method based on multi-source architecture change awareness as described in claim 1, characterized in that: The specific determination process for whether to adjust the multi-source heterogeneous data fusion process of the intelligent agent workflow system is as follows: The fusion coefficients of multi-source heterogeneous data are compared with the preset multi-source heterogeneous data fusion thresholds in the reconstructed database; If the multi-source heterogeneous data fusion coefficient is greater than or equal to the multi-source heterogeneous data fusion threshold, it is determined that the multi-source heterogeneous data fusion process of the agent workflow system will not be adjusted. Based on the multi-source heterogeneous data fusion coefficient, the adaptive coefficient correction coefficient of the agent workflow system is matched from the reconstruction database to correct the adaptive coefficient of the agent workflow system. At the same time, the adaptive threshold of the agent workflow system is matched based on the multi-source heterogeneous data fusion coefficient. If the multi-source heterogeneous data fusion coefficient is less than the multi-source heterogeneous data fusion threshold, it is determined that the multi-source heterogeneous data fusion process of the intelligent agent workflow system should be adjusted. The specific adjustment process is as follows: based on the multi-source heterogeneous data fusion coefficient and the multi-source heterogeneous data fusion threshold, the multi-source heterogeneous data fusion deviation value is obtained, and the time alignment window length is increased by matching the multi-source heterogeneous data fusion deviation value with the time alignment window length increase coefficient, thereby increasing the time alignment window length of multi-source data acquisition. After adjusting the multi-source heterogeneous data fusion process of the intelligent agent workflow system, obtain the secondary fusion coefficient of the multi-source heterogeneous data, and determine whether to perform a secondary adjustment on the multi-source heterogeneous data fusion process of the intelligent agent workflow system.

3. The workflow dynamic reconstruction method based on multi-source architecture change awareness according to claim 2, characterized in that: The determination of whether to perform secondary adjustments to the multi-source heterogeneous data fusion process of the intelligent agent workflow system is as follows: The secondary fusion coefficient of multi-source heterogeneous data is compared with the fusion threshold of multi-source heterogeneous data; If the secondary fusion coefficient of multi-source heterogeneous data is greater than or equal to the multi-source heterogeneous data fusion threshold, it is determined that no secondary adjustment will be made to the multi-source heterogeneous data fusion process of the intelligent agent workflow system. Based on the secondary fusion coefficient of multi-source heterogeneous data, the adaptive coefficient correction coefficient of the intelligent agent workflow system is matched from the reconstruction database, thereby correcting the adaptive coefficient of the intelligent agent workflow system. At the same time, the adaptive threshold of the intelligent agent workflow system is matched based on the secondary fusion coefficient of multi-source heterogeneous data. If the secondary fusion coefficient of multi-source heterogeneous data is less than the multi-source heterogeneous data fusion threshold, then it is determined that the multi-source heterogeneous data fusion process of the intelligent agent workflow system should be adjusted a second time. The specific adjustment process is as follows: based on the secondary fusion coefficient and the multi-source heterogeneous data fusion threshold, the secondary fusion deviation value of multi-source heterogeneous data is obtained. Based on the secondary fusion deviation value of multi-source heterogeneous data, a time alignment window length increase coefficient is matched to further increase the time alignment window length of multi-source data acquisition. At the same time, based on the secondary fusion deviation value of multi-source heterogeneous data, a conflict tolerance threshold reduction coefficient is matched to reduce the conflict tolerance threshold of the data value. After completing the second adjustment, the third fusion coefficient of the multi-source heterogeneous data is obtained and compared with the fusion threshold of the multi-source heterogeneous data. If the three-dimensional fusion coefficient of multi-source heterogeneous data is greater than or equal to the multi-source heterogeneous data fusion threshold, no warning is required. At the same time, based on the three-dimensional fusion coefficient of multi-source heterogeneous data, the adaptive coefficient correction coefficient of the intelligent agent workflow system is matched from the reconstructed database, thereby correcting the adaptive coefficient of the intelligent agent workflow system. Based on the three-dimensional fusion coefficient of multi-source heterogeneous data, the adaptive threshold of the intelligent agent workflow system is matched. If the fusion coefficient of multi-source heterogeneous data is less than the fusion threshold of multi-source heterogeneous data, the intelligent agent workflow system is monitored in real time, and the abnormality of the intelligent agent workflow system is traced.

4. The workflow dynamic reconstruction method based on multi-source architecture change awareness according to claim 1, characterized in that: The specific process for determining whether to cyclically adjust the multi-source architecture change process of the intelligent agent workflow system is as follows: The adaptive correction index of the agent workflow system is compared with the preset adaptive threshold of the agent workflow system in the reconstruction database. If the adaptive correction index of the agent workflow system is greater than or equal to the adaptive threshold of the agent workflow system, it is determined that the multi-source architecture change process of the agent workflow system will not be cyclically adjusted. If the adaptive correction index of the agent workflow system is less than the adaptive threshold of the agent workflow system, then the abnormality of the agent workflow system is traced, and it is determined that the multi-source architecture change process of the agent workflow system should be cyclically adjusted.

5. The workflow dynamic reconstruction method based on multi-source architecture change awareness according to claim 4, characterized in that: The specific process for tracing the anomalies in the intelligent agent workflow system is as follows: The perception delay duration is obtained and compared with the preset perception delay duration threshold in the reconstruction database. If the perception delay duration is greater than the perception delay duration threshold, the abnormal type of the agent workflow system is marked as a perception layer abnormality; otherwise, the abnormal type of the agent workflow system is not marked as a perception layer abnormality. Obtain the error reconstruction rate and compare it with the preset error reconstruction rate threshold in the reconstruction database. If the error reconstruction rate is greater than the error reconstruction rate threshold, mark the intelligent agent workflow system anomaly type as decision layer anomaly; otherwise, do not mark the intelligent agent workflow system anomaly type as decision layer anomaly. Obtain the reconstruction power and compare it with the preset reconstruction power threshold in the reconstruction database. If the reconstruction power is less than the reconstruction power threshold, mark the agent workflow system exception type as an execution layer exception; otherwise, do not mark the agent workflow system exception type as an execution layer exception. Obtain the policy optimization gain and compare it with the preset policy optimization gain threshold in the reconstruction database. If the policy optimization gain is less than the policy optimization gain threshold, mark the agent workflow system anomaly type as a learning layer anomaly; otherwise, do not mark the agent workflow system anomaly type as a learning layer anomaly.

6. The workflow dynamic reconfiguration method based on multi-source architecture change awareness according to claim 4, characterized in that: The specific analysis process for iteratively adjusting the multi-source architecture change process of the intelligent agent workflow system is as follows: If the anomaly type of the agent workflow system is a perception layer anomaly, then based on the agent workflow system adaptive correction index and the agent workflow system adaptive threshold, the agent workflow system adaptive deviation value is obtained. Based on the agent workflow system adaptive deviation value, the event sampling frequency increase coefficient is matched from the reconstruction database, thereby increasing the event sampling frequency of the agent workflow system. Based on the agent workflow system adaptive deviation value, the data aggregation window decrease coefficient is matched from the reconstruction database, thereby decreasing the data aggregation window of the agent workflow system. If the anomaly type of the agent workflow system is a decision-level anomaly, then the consensus threshold reduction coefficient is matched from the reconstruction database based on the adaptive deviation value of the agent workflow system, thereby reducing the consensus threshold of the agent workflow system. The policy exploration rate improvement value is directly matched from the reconstruction database based on the adaptive deviation value of the agent workflow system, thereby improving the policy exploration rate of the agent workflow system. If the exception type of the agent workflow system is an execution layer exception, then the resource buffer pool capacity increase coefficient is matched from the reconstruction database based on the adaptive deviation value of the agent workflow system, thereby increasing the resource buffer pool capacity of the agent workflow system. Then, the state snapshot interval reduction coefficient is matched from the reconstruction database based on the adaptive deviation value of the agent workflow system, thereby reducing the state snapshot interval of the agent workflow system. If the anomaly type of the agent workflow system is a learning layer anomaly, then the knowledge sharing frequency improvement coefficient is matched from the reconstruction database based on the adaptive deviation value of the agent workflow system, thereby improving the knowledge sharing frequency of the agent workflow system. The experience replay sampling rate improvement coefficient is matched from the reconstruction database based on the adaptive deviation value of the agent workflow system, thereby improving the experience replay sampling rate of the agent workflow system. If there are no abnormal types in the agent workflow system, then an early warning will be issued for the multi-source architecture change process of the agent workflow system; Obtain the number of cycles for each anomaly type in the intelligent agent workflow system, and determine whether to issue an early warning for each anomaly type.

7. The workflow dynamic reconfiguration method based on multi-source architecture change awareness according to claim 6, characterized in that: The specific process for determining whether to issue an alert for each of the relevant anomaly types is as follows: If the number of cyclic adjustments for a certain type of anomaly in the intelligent agent workflow system is equal to the number of defined cyclic adjustments, then the adjustment parameters for that type of anomaly in the intelligent agent workflow system are obtained and defined adjustments are made. After the adjustment is completed, it is determined whether the anomaly type of the intelligent agent workflow system still exists. If it still exists, then an early warning is issued for that type of anomaly in the intelligent agent workflow system. If it does not exist, then the working status of the intelligent agent workflow system is continuously monitored, and the threshold for multi-source heterogeneous data fusion is increased based on the final adaptive coefficient of the intelligent agent workflow system. If the number of cyclic adjustments for a certain type of anomaly in the agent workflow system is less than the defined number of cyclic adjustments, the working status of the agent workflow system will be continuously monitored.

8. A system applying the workflow dynamic reconfiguration method based on multi-source architecture change awareness as described in any one of claims 1 to 7, characterized in that: include: The multi-source data fusion sensing and adjustment module is used to sense multi-source heterogeneous data through the intelligent agent workflow system, fuse the multi-source heterogeneous data, analyze the parameters of the multi-source heterogeneous data fusion process, and thus determine whether to adjust the multi-source heterogeneous data fusion process of the intelligent agent workflow system. The architecture change monitoring and parameter acquisition module is used to monitor the fused multi-source heterogeneous data. The intelligent agent workflow system perceives the multi-source architecture changes and obtains the adaptive adjustment parameter set of the multi-source architecture changes. The architecture change analysis and workflow reconstruction module is used by the intelligent agent workflow system to analyze the adaptive adjustment parameter set of multi-source architecture changes and determine whether to cyclically adjust the multi-source architecture change process of the intelligent agent workflow system, thereby realizing the dynamic reconstruction of the workflow.

Images